From 903e9952fbce28677945325fd3ac6f01b3e560ce Mon Sep 17 00:00:00 2001
From: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Date: Thu, 22 Feb 2018 17:12:42 +0000
Subject: [PATCH] i965: add performance query support on CNL

v2: Add brw_oa_cnl.xml to EXTRA_DIST (Emil)

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
---
 src/mesa/drivers/dri/i965/Makefile.sources    |     3 +-
 src/mesa/drivers/dri/i965/brw_oa_cnl.xml      | 10410 ++++++++++++++++
 .../drivers/dri/i965/brw_performance_query.c  |     2 +
 src/mesa/drivers/dri/i965/meson.build         |     1 +
 4 files changed, 10415 insertions(+), 1 deletion(-)
 create mode 100644 src/mesa/drivers/dri/i965/brw_oa_cnl.xml

diff --git a/src/mesa/drivers/dri/i965/Makefile.sources b/src/mesa/drivers/dri/i965/Makefile.sources
index ad45cd84692..3479ceb9d16 100644
--- a/src/mesa/drivers/dri/i965/Makefile.sources
+++ b/src/mesa/drivers/dri/i965/Makefile.sources
@@ -171,4 +171,5 @@ i965_oa_xml_FILES = \
 	brw_oa_kblgt3.xml \
 	brw_oa_glk.xml \
 	brw_oa_cflgt2.xml \
-	brw_oa_cflgt3.xml
+	brw_oa_cflgt3.xml \
+	brw_oa_cnl.xml
diff --git a/src/mesa/drivers/dri/i965/brw_oa_cnl.xml b/src/mesa/drivers/dri/i965/brw_oa_cnl.xml
new file mode 100644
index 00000000000..3a2f8bc250b
--- /dev/null
+++ b/src/mesa/drivers/dri/i965/brw_oa_cnl.xml
@@ -0,0 +1,10410 @@
+<?xml version="1.0"?>
+<metrics version="1508419706" merge_md5="">
+  <set name="Render Metrics Basic Gen10"
+       mdapi_supported_apis="OGL OCL IO BB"
+       underscore_name="render_basic"
+       hw_config_guid="ee9f2b17-a8b3-44c3-bf0c-2608b9665e23"
+       chipset="CNL"
+       symbol_name="RenderBasic"
+       >
+    <counter name="GPU Core Clocks"
+             description="The total number of GPU core clocks elapsed during the measurement."
+             data_type="uint64"
+             equation="GPU_CLOCK 0 READ"
+             underscore_name="gpu_core_clocks"
+             units="cycles"
+             symbol_name="GpuCoreClocks"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="EU Active"
+             description="The percentage of time in which the Execution Units were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 7 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_active"
+             units="percent"
+             symbol_name="EuActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="L3 Misses"
+             description="The total number of L3 misses."
+             data_type="uint64"
+             equation="C 2 READ C 3 READ UADD"
+             underscore_name="l3_misses"
+             units="messages"
+             symbol_name="L3Misses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/TAG"
+             />
+    <counter name="GTI L3 Throughput"
+             description="The total number of GPU memory bytes transferred between L3 caches and GTI."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL"
+             equation="$L3Misses 64 UMUL"
+             underscore_name="gti_l3_throughput"
+             units="bytes"
+             symbol_name="GtiL3Throughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_group="GTI/L3"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EU Both FPU Pipes Active"
+             description="The percentage of time in which both EU FPU pipelines were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 9 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_fpu_both_active"
+             units="percent"
+             symbol_name="EuFpuBothActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Sampler Cache Misses"
+             description="The total number of sampler cache misses in all LODs in all sampler units."
+             data_type="uint64"
+             equation="$SliceMask 1 AND B 4 READ UMUL $SliceMask 1 &gt;&gt; 1 AND B 5 READ UMUL UADD 8 UMUL"
+             underscore_name="sampler_l1_misses"
+             units="messages"
+             symbol_name="SamplerL1Misses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Cache"
+             />
+    <counter name="VS Send Pipe Active"
+             description="The percentage of time in which EU send pipeline was actively processing a vertex shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 12 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="vs_send_active"
+             units="percent"
+             symbol_name="VsSendActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS FPU1 Pipe Active"
+             description="The percentage of time in which EU FPU1 pipeline was actively processing a vertex shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 11 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="vs_fpu1_active"
+             units="percent"
+             symbol_name="VsFpu1Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GS Threads Dispatched"
+             description="The total number of geometry shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 5 READ"
+             underscore_name="gs_threads"
+             units="threads"
+             symbol_name="GsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Geometry Shader"
+             />
+    <counter name="L3 Sampler Throughput"
+             description="The total number of GPU memory bytes transferred between samplers and L3 caches."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSlicesTotalCount UMUL"
+             equation="64 $SamplerL1Misses UMUL"
+             underscore_name="l3_sampler_throughput"
+             units="bytes"
+             symbol_name="L3SamplerThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_group="L3/Sampler"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Early Hi-Depth Test Fails"
+             description="The total number of pixels dropped on early hierarchical depth test."
+             data_type="uint64"
+             equation="A 22 READ 4 UMUL"
+             underscore_name="hi_depth_test_fails"
+             units="pixels"
+             symbol_name="HiDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Hi-Depth Test"
+             />
+    <counter name="FS Both FPU Active"
+             description="The percentage of time in which fragment shaders were processed actively on the both FPUs."
+             data_type="float"
+             max_equation="100"
+             equation="A 18 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_eu_both_fpu_active"
+             units="percent"
+             symbol_name="PsEuBothFpuActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Frame Batch Draw"
+             mdapi_group="3D Pipe/Fragment Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS Threads Dispatched"
+             description="The total number of vertex shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 1 READ"
+             underscore_name="vs_threads"
+             units="threads"
+             symbol_name="VsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Vertex Shader"
+             />
+    <counter name="FS Threads Dispatched"
+             description="The total number of fragment shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 6 READ"
+             underscore_name="ps_threads"
+             units="threads"
+             symbol_name="PsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Fragment Shader"
+             />
+    <counter name="Sampler00 Busy"
+             description="The percentage of time in which Slice0 Sampler0 has been processing EU requests."
+             data_type="float"
+             max_equation="100"
+             equation="B 0 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="sampler00_busy"
+             units="percent"
+             symbol_name="Sampler00Busy"
+             availability="$SubsliceMask 1 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Batch Frame Draw"
+             mdapi_group="Sampler"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="Sampler10 Busy"
+             description="The percentage of time in which Slice1 Sampler0 has been processing EU requests."
+             data_type="float"
+             max_equation="100"
+             equation="B 1 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="sampler10_busy"
+             units="percent"
+             symbol_name="Sampler10Busy"
+             availability="$SubsliceMask 8 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Batch Frame Draw"
+             mdapi_group="Sampler"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="Samplers Busy"
+             description="The percentage of time in which samplers have been processing EU requests."
+             data_type="float"
+             max_equation="100"
+             equation="$SubsliceMask 1 AND $Sampler00Busy FMUL $SubsliceMask 3 &gt;&gt; 1 AND $Sampler10Busy FMUL FADD $SubsliceMask 1 AND $SubsliceMask 3 &gt;&gt; 1 AND FADD FDIV"
+             underscore_name="samplers_busy"
+             units="percent"
+             symbol_name="SamplersBusy"
+             availability="$SubsliceMask 9 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="Sampler"
+             mdapi_hw_unit_type="slice"
+             />
+    <counter name="GTI Fixed Pipe Throughput"
+             description="The total number of GPU memory bytes transferred between 3D Pipeline (Command Dispatch, Input Assembly and Stream Output) and GTI."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL"
+             equation="C 4 READ C 5 READ UADD 64 UMUL"
+             underscore_name="gti_vf_throughput"
+             units="bytes"
+             symbol_name="GtiVfThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_group="GTI/3D Pipe"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Shader Barrier Messages"
+             description="The total number of shader barrier messages."
+             data_type="uint64"
+             equation="A 35 READ"
+             underscore_name="shader_barriers"
+             units="messages"
+             symbol_name="ShaderBarriers"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Barrier"
+             />
+    <counter name="Sampler Texels"
+             description="The total number of texels seen on input (with 2x2 accuracy) in all sampler units."
+             data_type="uint64"
+             equation="A 28 READ 4 UMUL"
+             underscore_name="sampler_texels"
+             units="texels"
+             symbol_name="SamplerTexels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Input"
+             />
+    <counter name="Pixels Failing Tests"
+             description="The total number of pixels dropped on post-FS alpha, stencil, or depth tests."
+             data_type="uint64"
+             equation="A 25 READ 4 UMUL"
+             underscore_name="pixels_failing_post_ps_tests"
+             units="pixels"
+             symbol_name="PixelsFailingPostPsTests"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="GPU Time Elapsed"
+             description="Time elapsed on the GPU during the measurement."
+             data_type="uint64"
+             equation="GPU_TIME 0 READ 1000000000 UMUL $GpuTimestampFrequency UDIV"
+             underscore_name="gpu_time"
+             units="ns"
+             symbol_name="GpuTime"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="AVG GPU Core Frequency"
+             description="Average GPU Core Frequency in the measurement."
+             data_type="uint64"
+             max_equation="$GpuMaxFrequency"
+             equation="$GpuCoreClocks 1000000000 UMUL $GpuTime UDIV"
+             underscore_name="avg_gpu_core_frequency"
+             units="hz"
+             symbol_name="AvgGpuCoreFrequency"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Sampler Texels Misses"
+             description="The total number of texels lookups (with 2x2 accuracy) that missed L1 sampler cache."
+             data_type="uint64"
+             equation="A 29 READ 4 UMUL"
+             underscore_name="sampler_texel_misses"
+             units="texels"
+             symbol_name="SamplerTexelMisses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Batch Frame Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Cache"
+             />
+    <counter name="CS Threads Dispatched"
+             description="The total number of compute shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 4 READ"
+             underscore_name="cs_threads"
+             units="threads"
+             symbol_name="CsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Compute Shader"
+             />
+    <counter name="Shader Memory Accesses"
+             description="The total number of shader memory accesses to L3."
+             data_type="uint64"
+             equation="A 32 READ"
+             underscore_name="shader_memory_accesses"
+             units="messages"
+             symbol_name="ShaderMemoryAccesses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="L3 Lookup Accesses w/o IC"
+             description="The total number of L3 cache lookup accesses w/o IC."
+             data_type="uint64"
+             equation="$SamplerL1Misses $ShaderMemoryAccesses UADD"
+             underscore_name="l3_lookups"
+             units="messages"
+             symbol_name="L3Lookups"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/TAG"
+             />
+    <counter name="Sampler00 Bottleneck"
+             low_watermark="5"
+             description="The percentage of time in which Slice0 Sampler0 has been slowing down the pipe when processing EU requests."
+             data_type="float"
+             high_watermark="15"
+             equation="B 2 READ 100 UMUL $GpuCoreClocks FDIV"
+             max_equation="100"
+             underscore_name="sampler00_bottleneck"
+             units="percent"
+             symbol_name="Sampler00Bottleneck"
+             availability="$SubsliceMask 1 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Indicate Draw"
+             mdapi_group="Sampler"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="SLM Bytes Read"
+             description="The total number of GPU memory bytes read from shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ 64 UMUL"
+             underscore_name="slm_bytes_read"
+             units="bytes"
+             symbol_name="SlmBytesRead"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GTI Read Throughput"
+             description="The total number of GPU memory bytes read from GTI."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL"
+             equation="64  C 6 READ UMUL"
+             underscore_name="gti_read_throughput"
+             units="bytes"
+             symbol_name="GtiReadThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_group="GTI"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="PS FPU1 Pipe Active"
+             description="The percentage of time in which EU FPU1 pipeline was actively processing a pixel shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 16 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_fpu1_active"
+             units="percent"
+             symbol_name="PsFpu1Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pixel Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="PS Send Pipeline Active"
+             description="The percentage of time in which EU send pipeline was actively processing a pixel shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 17 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_send_active"
+             units="percent"
+             symbol_name="PsSendActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pixel Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Rasterized Pixels"
+             description="The total number of rasterized pixels."
+             data_type="uint64"
+             equation="A 21 READ 4 UMUL"
+             underscore_name="rasterized_pixels"
+             units="pixels"
+             symbol_name="RasterizedPixels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer"
+             />
+    <counter name="GPU Busy"
+             description="The percentage of time in which the GPU has been processing GPU commands."
+             data_type="float"
+             max_equation="100"
+             equation="A 0 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="gpu_busy"
+             units="percent"
+             symbol_name="GpuBusy"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GTI Depth Throughput"
+             description="The total number of GPU memory bytes transferred between depth caches and GTI."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL"
+             equation="128  B 6 READ UMUL"
+             underscore_name="gti_depth_throughput"
+             units="bytes"
+             symbol_name="GtiDepthThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_group="GTI/Depth Cache"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS FPU0 Pipe Active"
+             description="The percentage of time in which EU FPU0 pipeline was actively processing a vertex shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 10 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="vs_fpu0_active"
+             units="percent"
+             symbol_name="VsFpu0Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="PS FPU0 Pipe Active"
+             description="The percentage of time in which EU FPU0 pipeline was actively processing a pixel shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 15 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_fpu0_active"
+             units="percent"
+             symbol_name="PsFpu0Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pixel Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="DS Threads Dispatched"
+             description="The total number of domain shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 3 READ"
+             underscore_name="ds_threads"
+             units="threads"
+             symbol_name="DsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Domain Shader"
+             />
+    <counter name="Samples Written"
+             description="The total number of samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 26 READ 4 UMUL"
+             underscore_name="samples_written"
+             units="pixels"
+             symbol_name="SamplesWritten"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="EU Stall"
+             description="The percentage of time in which the Execution Units were stalled."
+             data_type="float"
+             max_equation="100"
+             equation="A 8 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_stall"
+             units="percent"
+             symbol_name="EuStall"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Samples Blended"
+             description="The total number of blended samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 27 READ 4 UMUL"
+             underscore_name="samples_blended"
+             units="pixels"
+             symbol_name="SamplesBlended"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="Early Depth Test Fails"
+             description="The total number of pixels dropped on early depth test."
+             data_type="uint64"
+             equation="A 23 READ 4 UMUL"
+             underscore_name="early_depth_test_fails"
+             units="pixels"
+             symbol_name="EarlyDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Early Depth Test"
+             />
+    <counter name="Sampler10 Bottleneck"
+             low_watermark="5"
+             description="The percentage of time in which Slice1 Sampler0 has been slowing down the pipe when processing EU requests."
+             data_type="float"
+             high_watermark="15"
+             equation="B 3 READ 100 UMUL $GpuCoreClocks FDIV"
+             max_equation="100"
+             underscore_name="sampler10_bottleneck"
+             units="percent"
+             symbol_name="Sampler10Bottleneck"
+             availability="$SubsliceMask 8 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Indicate Draw"
+             mdapi_group="Sampler"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="Samplers Bottleneck"
+             low_watermark="5"
+             description="The percentage of time in which samplers have been slowing down the pipe when processing EU requests."
+             data_type="float"
+             high_watermark="15"
+             equation="$SubsliceMask 1 AND $Sampler00Bottleneck FMUL $SubsliceMask 3 &gt;&gt; 1 AND $Sampler10Bottleneck FMUL FADD $SubsliceMask 1 AND $SubsliceMask 3 &gt;&gt; 1 AND FADD FDIV"
+             max_equation="100"
+             underscore_name="sampler_bottleneck"
+             units="percent"
+             symbol_name="SamplerBottleneck"
+             availability="$SubsliceMask 9 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Indicate System Frame Batch Draw"
+             mdapi_group="Sampler"
+             mdapi_hw_unit_type="slice"
+             />
+    <counter name="GTI RCC Throughput"
+             description="The total number of GPU memory bytes transferred between render color caches and GTI."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL"
+             equation="C 0 READ C 1 READ UADD 128 UMUL"
+             underscore_name="gti_rcc_throughput"
+             units="bytes"
+             symbol_name="GtiRccThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_group="GTI/Color Cache"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="HS Threads Dispatched"
+             description="The total number of hull shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 2 READ"
+             underscore_name="hs_threads"
+             units="threads"
+             symbol_name="HsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Hull Shader"
+             />
+    <counter name="GTI Write Throughput"
+             description="The total number of GPU memory bytes written to GTI."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL"
+             equation="64  C 7 READ UMUL"
+             underscore_name="gti_write_throughput"
+             units="bytes"
+             symbol_name="GtiWriteThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_group="GTI"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="L3 Shader Throughput"
+             description="The total number of GPU memory bytes transferred between shaders and L3 caches w/o URB."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ A 31 READ UADD A 32 READ UADD 64 UMUL"
+             underscore_name="l3_shader_throughput"
+             units="bytes"
+             symbol_name="L3ShaderThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_group="L3/Data Port"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Samples Killed in FS"
+             description="The total number of samples or pixels dropped in fragment shaders."
+             data_type="uint64"
+             equation="A 24 READ 4 UMUL"
+             underscore_name="samples_killed_in_ps"
+             units="pixels"
+             symbol_name="SamplesKilledInPs"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Fragment Shader"
+             />
+    <counter name="SLM Bytes Written"
+             description="The total number of GPU memory bytes written into shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 31 READ 64 UMUL"
+             underscore_name="slm_bytes_written"
+             units="bytes"
+             symbol_name="SlmBytesWritten"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Shader Atomic Memory Accesses"
+             description="The total number of shader atomic memory accesses."
+             data_type="uint64"
+             equation="A 34 READ"
+             underscore_name="shader_atomics"
+             units="messages"
+             symbol_name="ShaderAtomics"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port/Atomics"
+             />
+    <register_config type="NOA">
+        <register type="NOA" address="0x00000D04" value="0x00000200" />
+        <register type="NOA" address="0x00009884" value="0x00000000" />
+        <register type="NOA" address="0x00009888" value="0x146C0008" />
+        <register type="NOA" address="0x00009888" value="0x18110014" />
+        <register type="NOA" address="0x00009888" value="0x14EC0008" />
+        <register type="NOA" address="0x00009888" value="0x18910014" />
+        <register type="NOA" address="0x00009888" value="0x0A6C0054" />
+        <register type="NOA" address="0x00009888" value="0x106C0000" />
+        <register type="NOA" address="0x00009888" value="0x1A6C0444" />
+        <register type="NOA" address="0x00009888" value="0x02738000" />
+        <register type="NOA" address="0x00009888" value="0x0473E000" />
+        <register type="NOA" address="0x00009888" value="0x04742000" />
+        <register type="NOA" address="0x00009888" value="0x0C742000" />
+        <register type="NOA" address="0x00009888" value="0x04752000" />
+        <register type="NOA" address="0x00009888" value="0x0C752000" />
+        <register type="NOA" address="0x00009888" value="0x0C760800" />
+        <register type="NOA" address="0x00009888" value="0x0E760800" />
+        <register type="NOA" address="0x00009888" value="0x0077C000" />
+        <register type="NOA" address="0x00009888" value="0x0E778000" />
+        <register type="NOA" address="0x00009888" value="0x1477C000" />
+        <register type="NOA" address="0x00009888" value="0x167702A0" />
+        <register type="NOA" address="0x00009888" value="0x00194000" />
+        <register type="NOA" address="0x00009888" value="0x1A190020" />
+        <register type="NOA" address="0x00009888" value="0x00110060" />
+        <register type="NOA" address="0x00009888" value="0x08110061" />
+        <register type="NOA" address="0x00009888" value="0x1E110000" />
+        <register type="NOA" address="0x00009888" value="0x10110000" />
+        <register type="NOA" address="0x00009888" value="0x18144000" />
+        <register type="NOA" address="0x00009888" value="0x1C140400" />
+        <register type="NOA" address="0x00009888" value="0x0AEC5400" />
+        <register type="NOA" address="0x00009888" value="0x10EC0000" />
+        <register type="NOA" address="0x00009888" value="0x1AEC0000" />
+        <register type="NOA" address="0x00009888" value="0x04F38000" />
+        <register type="NOA" address="0x00009888" value="0x0AF48000" />
+        <register type="NOA" address="0x00009888" value="0x0CF48000" />
+        <register type="NOA" address="0x00009888" value="0x0AF58000" />
+        <register type="NOA" address="0x00009888" value="0x0CF58000" />
+        <register type="NOA" address="0x00009888" value="0x0EF62200" />
+        <register type="NOA" address="0x00009888" value="0x0EF7C000" />
+        <register type="NOA" address="0x00009888" value="0x16F70230" />
+        <register type="NOA" address="0x00009888" value="0x0E994000" />
+        <register type="NOA" address="0x00009888" value="0x1A990080" />
+        <register type="NOA" address="0x00009888" value="0x06916000" />
+        <register type="NOA" address="0x00009888" value="0x08916100" />
+        <register type="NOA" address="0x00009888" value="0x1E910000" />
+        <register type="NOA" address="0x00009888" value="0x1C941100" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x0B086F4F" />
+        <register type="NOA" address="0x00009888" value="0x0D080001" />
+        <register type="NOA" address="0x00009888" value="0x1D073800" />
+        <register type="NOA" address="0x00009888" value="0x1F070140" />
+        <register type="NOA" address="0x00009888" value="0x09081E93" />
+        <register type="NOA" address="0x00009888" value="0x01083160" />
+        <register type="NOA" address="0x00009888" value="0x030801A6" />
+        <register type="NOA" address="0x00009888" value="0x11080000" />
+        <register type="NOA" address="0x00009888" value="0x0F080000" />
+        <register type="NOA" address="0x00009888" value="0x0D095000" />
+        <register type="NOA" address="0x00009888" value="0x0F095000" />
+        <register type="NOA" address="0x00009888" value="0x01094000" />
+        <register type="NOA" address="0x00009888" value="0x03095000" />
+        <register type="NOA" address="0x00009888" value="0x05091000" />
+        <register type="NOA" address="0x00009888" value="0x13040154" />
+        <register type="NOA" address="0x00009888" value="0x09044000" />
+        <register type="NOA" address="0x00009888" value="0x0B045000" />
+        <register type="NOA" address="0x00009888" value="0x0D049000" />
+        <register type="NOA" address="0x00009888" value="0x0F042000" />
+        <register type="NOA" address="0x00009888" value="0x0507E700" />
+        <register type="NOA" address="0x00009888" value="0x070700D7" />
+        <register type="NOA" address="0x00009888" value="0x25070000" />
+        <register type="NOA" address="0x00009888" value="0x1B070000" />
+        <register type="NOA" address="0x00009888" value="0x1100C000" />
+        <register type="NOA" address="0x00009888" value="0x1F00C000" />
+        <register type="NOA" address="0x00009888" value="0x2100C000" />
+        <register type="NOA" address="0x00009888" value="0x2300C000" />
+        <register type="NOA" address="0x00009888" value="0x2500C000" />
+        <register type="NOA" address="0x00009888" value="0x2700C000" />
+        <register type="NOA" address="0x00009888" value="0x29004000" />
+        <register type="NOA" address="0x00009888" value="0x2B004000" />
+        <register type="NOA" address="0x00009888" value="0x2D004000" />
+        <register type="NOA" address="0x00009888" value="0x2F004000" />
+        <register type="NOA" address="0x00009888" value="0x13004000" />
+        <register type="NOA" address="0x00009888" value="0x15004000" />
+        <register type="NOA" address="0x00009888" value="0x17004000" />
+        <register type="NOA" address="0x00009888" value="0x19004000" />
+        <register type="NOA" address="0x00009888" value="0x1B004000" />
+        <register type="NOA" address="0x00009888" value="0x1D004000" />
+        <register type="NOA" address="0x00009888" value="0x47003331" />
+        <register type="NOA" address="0x00009888" value="0x390018C1" />
+        <register type="NOA" address="0x00009888" value="0x49000333" />
+        <register type="NOA" address="0x00009888" value="0x3D000407" />
+        <register type="NOA" address="0x00009888" value="0x4B001101" />
+        <register type="NOA" address="0x00009888" value="0x3F000120" />
+        <register type="NOA" address="0x00009888" value="0x4D003333" />
+        <register type="NOA" address="0x00009888" value="0x410018C6" />
+        <register type="NOA" address="0x00009888" value="0x35000000" />
+        <register type="NOA" address="0x00009888" value="0x31000000" />
+        <register type="NOA" address="0x00009888" value="0x43000006" />
+        <register type="NOA" address="0x00009888" value="0x3B001CC6" />
+    </register_config>
+    <register_config type="OA">
+        <register type="OA" address="0x00002740" value="0x00000000" />
+        <register type="OA" address="0x00002710" value="0x00000000" />
+        <register type="OA" address="0x00002714" value="0x00800000" />
+        <register type="OA" address="0x00002720" value="0x00000000" />
+        <register type="OA" address="0x00002724" value="0x40800000" />
+        <register type="OA" address="0x000027A0" value="0x00000202" />
+        <register type="OA" address="0x000027A4" value="0x0000FF3F" />
+    </register_config>
+    <register_config type="FLEX">
+        <register type="FLEX" address="0x0000E458" value="0x00005004" />
+        <register type="FLEX" address="0x0000E558" value="0x00010003" />
+        <register type="FLEX" address="0x0000E658" value="0x00012011" />
+        <register type="FLEX" address="0x0000E45C" value="0x00051050" />
+        <register type="FLEX" address="0x0000E55C" value="0x00053052" />
+    </register_config>
+  </set>
+
+  <set name="Compute Metrics Basic Gen10"
+       mdapi_supported_apis="OGL OCL IO BB"
+       underscore_name="compute_basic"
+       hw_config_guid="eed14f91-5f03-4e3a-8281-ac38030ac573"
+       chipset="CNL"
+       symbol_name="ComputeBasic"
+       >
+    <counter name="GPU Core Clocks"
+             description="The total number of GPU core clocks elapsed during the measurement."
+             data_type="uint64"
+             equation="GPU_CLOCK 0 READ"
+             underscore_name="gpu_core_clocks"
+             units="cycles"
+             symbol_name="GpuCoreClocks"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="EU Active"
+             description="The percentage of time in which the Execution Units were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 7 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_active"
+             units="percent"
+             symbol_name="EuActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Untyped Bytes Read"
+             description="The total number of typed memory bytes read via Data Port."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSlicesTotalCount UMUL"
+             equation="$EuSubslicesTotalCount $SubsliceMask 1 AND C 2 READ FMUL $SubsliceMask 3 &gt;&gt; 1 AND C 3 READ FMUL FADD FMUL $SubsliceMask 1 AND $SubsliceMask 3 &gt;&gt; 1 AND FADD FDIV 64 UMUL"
+             underscore_name="untyped_bytes_read"
+             units="bytes"
+             symbol_name="UntypedBytesRead"
+             availability="$SubsliceMask 9 AND "
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EU Both FPU Pipes Active"
+             description="The percentage of time in which both EU FPU pipelines were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 9 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_fpu_both_active"
+             units="percent"
+             symbol_name="EuFpuBothActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Typed Bytes Written"
+             description="The total number of untyped memory bytes written via Data Port."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSlicesTotalCount UMUL"
+             equation="$SubsliceMask 1 AND B 5 READ UMUL $SubsliceMask 1 &gt;&gt; 1 AND B 6 READ UMUL UADD $SubsliceMask 2 &gt;&gt; 1 AND B 7 READ UMUL UADD $SubsliceMask 3 &gt;&gt; 1 AND C 0 READ UMUL UADD $SubsliceMask 4 &gt;&gt; 1 AND C 1 READ UMUL UADD 64 UMUL"
+             underscore_name="typed_bytes_written"
+             units="bytes"
+             symbol_name="TypedBytesWritten"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EU AVG IPC Rate"
+             description="The average rate of IPC calculated for 2 FPU pipelines."
+             data_type="float"
+             max_equation="2"
+             equation="A 9 READ A 10 READ A 11 READ FADD A 9 READ FSUB FDIV 1 FADD"
+             underscore_name="eu_avg_ipc_rate"
+             units="number"
+             symbol_name="EuAvgIpcRate"
+             semantic_type="ratio"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GS Threads Dispatched"
+             description="The total number of geometry shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 5 READ"
+             underscore_name="gs_threads"
+             units="threads"
+             symbol_name="GsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Geometry Shader"
+             />
+    <counter name="Early Hi-Depth Test Fails"
+             description="The total number of pixels dropped on early hierarchical depth test."
+             data_type="uint64"
+             equation="A 22 READ 4 UMUL"
+             underscore_name="hi_depth_test_fails"
+             units="pixels"
+             symbol_name="HiDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Hi-Depth Test"
+             />
+    <counter name="VS Threads Dispatched"
+             description="The total number of vertex shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 1 READ"
+             underscore_name="vs_threads"
+             units="threads"
+             symbol_name="VsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Vertex Shader"
+             />
+    <counter name="FS Threads Dispatched"
+             description="The total number of fragment shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 6 READ"
+             underscore_name="ps_threads"
+             units="threads"
+             symbol_name="PsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Fragment Shader"
+             />
+    <counter name="EU FPU1 Pipe Active"
+             description="The percentage of time in which EU FPU1 pipeline was actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 11 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="fpu1_active"
+             units="percent"
+             symbol_name="Fpu1Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EU FPU0 Pipe Active"
+             description="The percentage of time in which EU FPU0 pipeline was actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 10 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="fpu0_active"
+             units="percent"
+             symbol_name="Fpu0Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Shader Barrier Messages"
+             description="The total number of shader barrier messages."
+             data_type="uint64"
+             equation="A 35 READ"
+             underscore_name="shader_barriers"
+             units="messages"
+             symbol_name="ShaderBarriers"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Barrier"
+             />
+    <counter name="Sampler Texels"
+             description="The total number of texels seen on input (with 2x2 accuracy) in all sampler units."
+             data_type="uint64"
+             equation="A 28 READ 4 UMUL"
+             underscore_name="sampler_texels"
+             units="texels"
+             symbol_name="SamplerTexels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Input"
+             />
+    <counter name="Pixels Failing Tests"
+             description="The total number of pixels dropped on post-FS alpha, stencil, or depth tests."
+             data_type="uint64"
+             equation="A 25 READ 4 UMUL"
+             underscore_name="pixels_failing_post_ps_tests"
+             units="pixels"
+             symbol_name="PixelsFailingPostPsTests"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="GPU Time Elapsed"
+             description="Time elapsed on the GPU during the measurement."
+             data_type="uint64"
+             equation="GPU_TIME 0 READ 1000000000 UMUL $GpuTimestampFrequency UDIV"
+             underscore_name="gpu_time"
+             units="ns"
+             symbol_name="GpuTime"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="AVG GPU Core Frequency"
+             description="Average GPU Core Frequency in the measurement."
+             data_type="uint64"
+             max_equation="$GpuMaxFrequency"
+             equation="$GpuCoreClocks 1000000000 UMUL $GpuTime UDIV"
+             underscore_name="avg_gpu_core_frequency"
+             units="hz"
+             symbol_name="AvgGpuCoreFrequency"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Sampler Texels Misses"
+             description="The total number of texels lookups (with 2x2 accuracy) that missed L1 sampler cache."
+             data_type="uint64"
+             equation="A 29 READ 4 UMUL"
+             underscore_name="sampler_texel_misses"
+             units="texels"
+             symbol_name="SamplerTexelMisses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Batch Frame Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Cache"
+             />
+    <counter name="CS Threads Dispatched"
+             description="The total number of compute shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 4 READ"
+             underscore_name="cs_threads"
+             units="threads"
+             symbol_name="CsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Compute Shader"
+             />
+    <counter name="SLM Bytes Read"
+             description="The total number of GPU memory bytes read from shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ 64 UMUL"
+             underscore_name="slm_bytes_read"
+             units="bytes"
+             symbol_name="SlmBytesRead"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GTI Read Throughput"
+             description="The total number of GPU memory bytes read from GTI."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL"
+             equation="64  C 6 READ UMUL"
+             underscore_name="gti_read_throughput"
+             units="bytes"
+             symbol_name="GtiReadThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_group="GTI"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Untyped Writes"
+             description="The total number of untyped memory bytes written via Data Port."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSlicesTotalCount UMUL"
+             equation="$EuSubslicesTotalCount $SubsliceMask 1 AND C 4 READ FMUL $SubsliceMask 3 &gt;&gt; 1 AND C 5 READ FMUL FADD FMUL $SubsliceMask 1 AND $SubsliceMask 3 &gt;&gt; 1 AND FADD FDIV 64 UMUL"
+             underscore_name="untyped_bytes_written"
+             units="bytes"
+             symbol_name="UntypedBytesWritten"
+             availability="$SubsliceMask 9 AND"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GPU Busy"
+             description="The percentage of time in which the GPU has been processing GPU commands."
+             data_type="float"
+             max_equation="100"
+             equation="A 0 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="gpu_busy"
+             units="percent"
+             symbol_name="GpuBusy"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Rasterized Pixels"
+             description="The total number of rasterized pixels."
+             data_type="uint64"
+             equation="A 21 READ 4 UMUL"
+             underscore_name="rasterized_pixels"
+             units="pixels"
+             symbol_name="RasterizedPixels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer"
+             />
+    <counter name="Typed Bytes Read"
+             description="The total number of typed memory bytes read via Data Port."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSlicesTotalCount UMUL"
+             equation="$SubsliceMask 1 AND B 0 READ UMUL $SubsliceMask 1 &gt;&gt; 1 AND B 1 READ UMUL UADD $SubsliceMask 2 &gt;&gt; 1 AND B 2 READ UMUL UADD $SubsliceMask 3 &gt;&gt; 1 AND B 3 READ UMUL UADD $SubsliceMask 4 &gt;&gt; 1 AND B 4 READ UMUL UADD 64 UMUL"
+             underscore_name="typed_bytes_read"
+             units="bytes"
+             symbol_name="TypedBytesRead"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="DS Threads Dispatched"
+             description="The total number of domain shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 3 READ"
+             underscore_name="ds_threads"
+             units="threads"
+             symbol_name="DsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Domain Shader"
+             />
+    <counter name="EU Thread Occupancy"
+             description="The percentage of time in which hardware threads occupied EUs."
+             data_type="float"
+             max_equation="100"
+             equation="8 A 13 READ FMUL $EuThreadsCount FDIV $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_thread_occupancy"
+             units="percent"
+             symbol_name="EuThreadOccupancy"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EU Stall"
+             description="The percentage of time in which the Execution Units were stalled."
+             data_type="float"
+             max_equation="100"
+             equation="A 8 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_stall"
+             units="percent"
+             symbol_name="EuStall"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Samples Blended"
+             description="The total number of blended samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 27 READ 4 UMUL"
+             underscore_name="samples_blended"
+             units="pixels"
+             symbol_name="SamplesBlended"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="Early Depth Test Fails"
+             description="The total number of pixels dropped on early depth test."
+             data_type="uint64"
+             equation="A 23 READ 4 UMUL"
+             underscore_name="early_depth_test_fails"
+             units="pixels"
+             symbol_name="EarlyDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Early Depth Test"
+             />
+    <counter name="Shader Memory Accesses"
+             description="The total number of shader memory accesses to L3."
+             data_type="uint64"
+             equation="A 32 READ"
+             underscore_name="shader_memory_accesses"
+             units="messages"
+             symbol_name="ShaderMemoryAccesses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="HS Threads Dispatched"
+             description="The total number of hull shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 2 READ"
+             underscore_name="hs_threads"
+             units="threads"
+             symbol_name="HsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Hull Shader"
+             />
+    <counter name="GTI Write Throughput"
+             description="The total number of GPU memory bytes written to GTI."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL"
+             equation="64  C 7 READ UMUL"
+             underscore_name="gti_write_throughput"
+             units="bytes"
+             symbol_name="GtiWriteThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_group="GTI"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="L3 Shader Throughput"
+             description="The total number of GPU memory bytes transferred between shaders and L3 caches w/o URB."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ A 31 READ UADD A 32 READ UADD 64 UMUL"
+             underscore_name="l3_shader_throughput"
+             units="bytes"
+             symbol_name="L3ShaderThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_group="L3/Data Port"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Samples Killed in FS"
+             description="The total number of samples or pixels dropped in fragment shaders."
+             data_type="uint64"
+             equation="A 24 READ 4 UMUL"
+             underscore_name="samples_killed_in_ps"
+             units="pixels"
+             symbol_name="SamplesKilledInPs"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Fragment Shader"
+             />
+    <counter name="SLM Bytes Written"
+             description="The total number of GPU memory bytes written into shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 31 READ 64 UMUL"
+             underscore_name="slm_bytes_written"
+             units="bytes"
+             symbol_name="SlmBytesWritten"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Samples Written"
+             description="The total number of samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 26 READ 4 UMUL"
+             underscore_name="samples_written"
+             units="pixels"
+             symbol_name="SamplesWritten"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="Shader Atomic Memory Accesses"
+             description="The total number of shader atomic memory accesses."
+             data_type="uint64"
+             equation="A 34 READ"
+             underscore_name="shader_atomics"
+             units="messages"
+             symbol_name="ShaderAtomics"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port/Atomics"
+             />
+    <counter name="EU Send Pipe Active"
+             description="The percentage of time in which EU send pipeline was actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 12 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_send_active"
+             units="percent"
+             symbol_name="EuSendActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes"
+             mdapi_hw_unit_type="gpu"
+             />
+    <register_config type="NOA">
+        <register type="NOA" address="0x00000D04" value="0x00000200" />
+        <register type="NOA" address="0x00009884" value="0x00000000" />
+        <register type="NOA" address="0x00009888" value="0x121A0000" />
+        <register type="NOA" address="0x00009888" value="0x101A0007" />
+        <register type="NOA" address="0x00009888" value="0x123A0000" />
+        <register type="NOA" address="0x00009888" value="0x129A0000" />
+        <register type="NOA" address="0x00009888" value="0x109A0007" />
+        <register type="NOA" address="0x00009888" value="0x12BA0000" />
+        <register type="NOA" address="0x00009888" value="0x125A0000" />
+        <register type="NOA" address="0x00009888" value="0x1A6C0140" />
+        <register type="NOA" address="0x00009888" value="0x1C6C0005" />
+        <register type="NOA" address="0x00009888" value="0x186C0880" />
+        <register type="NOA" address="0x00009888" value="0x04736000" />
+        <register type="NOA" address="0x00009888" value="0x0673C000" />
+        <register type="NOA" address="0x00009888" value="0x00734000" />
+        <register type="NOA" address="0x00009888" value="0x02731000" />
+        <register type="NOA" address="0x00009888" value="0x0474A000" />
+        <register type="NOA" address="0x00009888" value="0x0E748000" />
+        <register type="NOA" address="0x00009888" value="0x06748000" />
+        <register type="NOA" address="0x00009888" value="0x0475A000" />
+        <register type="NOA" address="0x00009888" value="0x0E758000" />
+        <register type="NOA" address="0x00009888" value="0x06758000" />
+        <register type="NOA" address="0x00009888" value="0x0C762800" />
+        <register type="NOA" address="0x00009888" value="0x10760002" />
+        <register type="NOA" address="0x00009888" value="0x0E760002" />
+        <register type="NOA" address="0x00009888" value="0x0077C000" />
+        <register type="NOA" address="0x00009888" value="0x0E774000" />
+        <register type="NOA" address="0x00009888" value="0x14774000" />
+        <register type="NOA" address="0x00009888" value="0x167797A0" />
+        <register type="NOA" address="0x00009888" value="0x18770002" />
+        <register type="NOA" address="0x00009888" value="0x0277C000" />
+        <register type="NOA" address="0x00009888" value="0x04778000" />
+        <register type="NOA" address="0x00009888" value="0x0677C000" />
+        <register type="NOA" address="0x00009888" value="0x08778000" />
+        <register type="NOA" address="0x00009888" value="0x0A7B0A00" />
+        <register type="NOA" address="0x00009888" value="0x0C7B0028" />
+        <register type="NOA" address="0x00009888" value="0x0E7C8000" />
+        <register type="NOA" address="0x00009888" value="0x147C0C00" />
+        <register type="NOA" address="0x00009888" value="0x167C0038" />
+        <register type="NOA" address="0x00009888" value="0x1C7D2000" />
+        <register type="NOA" address="0x00009888" value="0x1E7D0080" />
+        <register type="NOA" address="0x00009888" value="0x0019C000" />
+        <register type="NOA" address="0x00009888" value="0x1A191800" />
+        <register type="NOA" address="0x00009888" value="0x0219C000" />
+        <register type="NOA" address="0x00009888" value="0x0619C000" />
+        <register type="NOA" address="0x00009888" value="0x001A2E52" />
+        <register type="NOA" address="0x00009888" value="0x0A1A2880" />
+        <register type="NOA" address="0x00009888" value="0x021A2D80" />
+        <register type="NOA" address="0x00009888" value="0x0E39C000" />
+        <register type="NOA" address="0x00009888" value="0x1A396000" />
+        <register type="NOA" address="0x00009888" value="0x063A2900" />
+        <register type="NOA" address="0x00009888" value="0x0C3A0051" />
+        <register type="NOA" address="0x00009888" value="0x003A0000" />
+        <register type="NOA" address="0x00009888" value="0x0CF48000" />
+        <register type="NOA" address="0x00009888" value="0x12F40400" />
+        <register type="NOA" address="0x00009888" value="0x06F42000" />
+        <register type="NOA" address="0x00009888" value="0x08F42000" />
+        <register type="NOA" address="0x00009888" value="0x0CF58000" />
+        <register type="NOA" address="0x00009888" value="0x14F50020" />
+        <register type="NOA" address="0x00009888" value="0x06F52000" />
+        <register type="NOA" address="0x00009888" value="0x08F52000" />
+        <register type="NOA" address="0x00009888" value="0x0EF62008" />
+        <register type="NOA" address="0x00009888" value="0x10F60080" />
+        <register type="NOA" address="0x00009888" value="0x0CF68000" />
+        <register type="NOA" address="0x00009888" value="0x16F7C070" />
+        <register type="NOA" address="0x00009888" value="0x18F70001" />
+        <register type="NOA" address="0x00009888" value="0x04F7C000" />
+        <register type="NOA" address="0x00009888" value="0x08F7C000" />
+        <register type="NOA" address="0x00009888" value="0x0AFB8000" />
+        <register type="NOA" address="0x00009888" value="0x0CFB0200" />
+        <register type="NOA" address="0x00009888" value="0x14FC8000" />
+        <register type="NOA" address="0x00009888" value="0x16FC0200" />
+        <register type="NOA" address="0x00009888" value="0x1EFD2008" />
+        <register type="NOA" address="0x00009888" value="0x1A990180" />
+        <register type="NOA" address="0x00009888" value="0x1C99000C" />
+        <register type="NOA" address="0x00009888" value="0x0499C000" />
+        <register type="NOA" address="0x00009888" value="0x0899C000" />
+        <register type="NOA" address="0x00009888" value="0x089A2900" />
+        <register type="NOA" address="0x00009888" value="0x0E9A0051" />
+        <register type="NOA" address="0x00009888" value="0x029A005C" />
+        <register type="NOA" address="0x00009888" value="0x049A005B" />
+        <register type="NOA" address="0x00009888" value="0x009A0000" />
+        <register type="NOA" address="0x00009888" value="0x1AB90600" />
+        <register type="NOA" address="0x00009888" value="0x1CB90030" />
+        <register type="NOA" address="0x00009888" value="0x0ABA0052" />
+        <register type="NOA" address="0x00009888" value="0x0EBA2880" />
+        <register type="NOA" address="0x00009888" value="0x00BA0000" />
+        <register type="NOA" address="0x00009888" value="0x1A590060" />
+        <register type="NOA" address="0x00009888" value="0x1C590003" />
+        <register type="NOA" address="0x00009888" value="0x085A0052" />
+        <register type="NOA" address="0x00009888" value="0x0C5A2880" />
+        <register type="NOA" address="0x00009888" value="0x005A0000" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x1D073800" />
+        <register type="NOA" address="0x00009888" value="0x1F070140" />
+        <register type="NOA" address="0x00009888" value="0x0D048000" />
+        <register type="NOA" address="0x00009888" value="0x0F042000" />
+        <register type="NOA" address="0x00009888" value="0x0507E700" />
+        <register type="NOA" address="0x00009888" value="0x070700D7" />
+        <register type="NOA" address="0x00009888" value="0x25070000" />
+        <register type="NOA" address="0x00009888" value="0x1B070000" />
+        <register type="NOA" address="0x00009888" value="0x1100C000" />
+        <register type="NOA" address="0x00009888" value="0x1F00C000" />
+        <register type="NOA" address="0x00009888" value="0x2100C000" />
+        <register type="NOA" address="0x00009888" value="0x2300C000" />
+        <register type="NOA" address="0x00009888" value="0x2500C000" />
+        <register type="NOA" address="0x00009888" value="0x2700C000" />
+        <register type="NOA" address="0x00009888" value="0x2900C000" />
+        <register type="NOA" address="0x00009888" value="0x2B00C000" />
+        <register type="NOA" address="0x00009888" value="0x2D00C000" />
+        <register type="NOA" address="0x00009888" value="0x2F00C000" />
+        <register type="NOA" address="0x00009888" value="0x1300C000" />
+        <register type="NOA" address="0x00009888" value="0x1500C000" />
+        <register type="NOA" address="0x00009888" value="0x1700C000" />
+        <register type="NOA" address="0x00009888" value="0x1900C000" />
+        <register type="NOA" address="0x00009888" value="0x1B004000" />
+        <register type="NOA" address="0x00009888" value="0x1D004000" />
+        <register type="NOA" address="0x00009888" value="0x47001011" />
+        <register type="NOA" address="0x00009888" value="0x39000021" />
+        <register type="NOA" address="0x00009888" value="0x49001330" />
+        <register type="NOA" address="0x00009888" value="0x3D001026" />
+        <register type="NOA" address="0x00009888" value="0x4B001000" />
+        <register type="NOA" address="0x00009888" value="0x3F000400" />
+        <register type="NOA" address="0x00009888" value="0x4D000001" />
+        <register type="NOA" address="0x00009888" value="0x41000081" />
+        <register type="NOA" address="0x00009888" value="0x35000000" />
+        <register type="NOA" address="0x00009888" value="0x31000000" />
+        <register type="NOA" address="0x00009888" value="0x43000000" />
+        <register type="NOA" address="0x00009888" value="0x3B001801" />
+    </register_config>
+    <register_config type="OA">
+        <register type="OA" address="0x00002740" value="0x00000000" />
+        <register type="OA" address="0x00002710" value="0x00000000" />
+        <register type="OA" address="0x00002714" value="0x00800000" />
+        <register type="OA" address="0x00002720" value="0x00000000" />
+        <register type="OA" address="0x00002724" value="0x00800000" />
+    </register_config>
+    <register_config type="FLEX">
+        <register type="FLEX" address="0x0000E458" value="0x00005004" />
+        <register type="FLEX" address="0x0000E558" value="0x00000003" />
+        <register type="FLEX" address="0x0000E658" value="0x00002001" />
+        <register type="FLEX" address="0x0000E758" value="0x00000008" />
+    </register_config>
+  </set>
+
+  <set name="Render Metrics for 3D Pipeline Profile Gen10"
+       mdapi_supported_apis="OGL OCL IO BB"
+       underscore_name="render_pipe_profile"
+       hw_config_guid="0666ccac-f5aa-4dc2-aa12-c6e058f383f3"
+       chipset="CNL"
+       symbol_name="RenderPipeProfile"
+       >
+    <counter name="GPU Core Clocks"
+             description="The total number of GPU core clocks elapsed during the measurement."
+             data_type="uint64"
+             equation="GPU_CLOCK 0 READ"
+             underscore_name="gpu_core_clocks"
+             units="cycles"
+             symbol_name="GpuCoreClocks"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="EU Active"
+             description="The percentage of time in which the Execution Units were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 7 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_active"
+             units="percent"
+             symbol_name="EuActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS Bottleneck"
+             low_watermark="10"
+             description="The percentage of time in which vertex shader pipeline stage was slowing down the 3D pipeline."
+             data_type="float"
+             high_watermark="30"
+             equation="C 3 READ 100 UMUL $GpuCoreClocks FDIV"
+             max_equation="100"
+             underscore_name="vs_bottleneck"
+             units="percent"
+             symbol_name="VsBottleneck"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Indicate Draw"
+             mdapi_group="3D Pipe/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Hi-Depth Bottleneck"
+             low_watermark="5"
+             description="The percentage of time in which early hierarchical depth test pipeline stage was slowing down the 3D pipeline."
+             data_type="float"
+             high_watermark="15"
+             equation="B 3 READ 100 UMUL $GpuCoreClocks FDIV"
+             max_equation="100"
+             underscore_name="hi_depth_bottleneck"
+             units="percent"
+             symbol_name="HiDepthBottleneck"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Indicate Draw"
+             mdapi_group="3D Pipe/Rasterizer/Hi-Depth Test"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GS Bottleneck"
+             low_watermark="5"
+             description="The percentage of time in which geometry shader pipeline stage was slowing down the 3D pipeline."
+             data_type="float"
+             high_watermark="15"
+             equation="C 0 READ 100 UMUL $GpuCoreClocks FDIV"
+             max_equation="100"
+             underscore_name="gs_bottleneck"
+             units="percent"
+             symbol_name="GsBottleneck"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Indicate Draw"
+             mdapi_group="3D Pipe/Geometry Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GS Threads Dispatched"
+             description="The total number of geometry shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 5 READ"
+             underscore_name="gs_threads"
+             units="threads"
+             symbol_name="GsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Geometry Shader"
+             />
+    <counter name="Early Hi-Depth Test Fails"
+             description="The total number of pixels dropped on early hierarchical depth test."
+             data_type="uint64"
+             equation="A 22 READ 4 UMUL"
+             underscore_name="hi_depth_test_fails"
+             units="pixels"
+             symbol_name="HiDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Hi-Depth Test"
+             />
+    <counter name="VS Threads Dispatched"
+             description="The total number of vertex shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 1 READ"
+             underscore_name="vs_threads"
+             units="threads"
+             symbol_name="VsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Vertex Shader"
+             />
+    <counter name="FS Threads Dispatched"
+             description="The total number of fragment shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 6 READ"
+             underscore_name="ps_threads"
+             units="threads"
+             symbol_name="PsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Fragment Shader"
+             />
+    <counter name="BC Bottleneck"
+             low_watermark="5"
+             description="The percentage of time in which barycentric coordinates calculation pipeline stage was slowing down the 3D pipeline."
+             data_type="float"
+             high_watermark="15"
+             equation="B 2 READ 100 UMUL $GpuCoreClocks FDIV"
+             max_equation="100"
+             underscore_name="bc_bottleneck"
+             units="percent"
+             symbol_name="BcBottleneck"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Indicate Draw"
+             mdapi_group="3D Pipe/Rasterizer/Barycentric Calc"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="HS Stall"
+             description="The percentage of time in which hull stall pipeline stage was stalled."
+             data_type="float"
+             max_equation="100"
+             equation="B 1 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="hs_stall"
+             units="percent"
+             symbol_name="HsStall"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Correlate Draw"
+             mdapi_group="3D Pipe/Hull Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Shader Barrier Messages"
+             description="The total number of shader barrier messages."
+             data_type="uint64"
+             equation="A 35 READ"
+             underscore_name="shader_barriers"
+             units="messages"
+             symbol_name="ShaderBarriers"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Barrier"
+             />
+    <counter name="Sampler Texels"
+             description="The total number of texels seen on input (with 2x2 accuracy) in all sampler units."
+             data_type="uint64"
+             equation="A 28 READ 4 UMUL"
+             underscore_name="sampler_texels"
+             units="texels"
+             symbol_name="SamplerTexels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Input"
+             />
+    <counter name="Pixels Failing Tests"
+             description="The total number of pixels dropped on post-FS alpha, stencil, or depth tests."
+             data_type="uint64"
+             equation="A 25 READ 4 UMUL"
+             underscore_name="pixels_failing_post_ps_tests"
+             units="pixels"
+             symbol_name="PixelsFailingPostPsTests"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="GPU Time Elapsed"
+             description="Time elapsed on the GPU during the measurement."
+             data_type="uint64"
+             equation="GPU_TIME 0 READ 1000000000 UMUL $GpuTimestampFrequency UDIV"
+             underscore_name="gpu_time"
+             units="ns"
+             symbol_name="GpuTime"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="AVG GPU Core Frequency"
+             description="Average GPU Core Frequency in the measurement."
+             data_type="uint64"
+             max_equation="$GpuMaxFrequency"
+             equation="$GpuCoreClocks 1000000000 UMUL $GpuTime UDIV"
+             underscore_name="avg_gpu_core_frequency"
+             units="hz"
+             symbol_name="AvgGpuCoreFrequency"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Sampler Texels Misses"
+             description="The total number of texels lookups (with 2x2 accuracy) that missed L1 sampler cache."
+             data_type="uint64"
+             equation="A 29 READ 4 UMUL"
+             underscore_name="sampler_texel_misses"
+             units="texels"
+             symbol_name="SamplerTexelMisses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Batch Frame Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Cache"
+             />
+    <counter name="CS Threads Dispatched"
+             description="The total number of compute shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 4 READ"
+             underscore_name="cs_threads"
+             units="threads"
+             symbol_name="CsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Compute Shader"
+             />
+    <counter name="VF Bottleneck"
+             low_watermark="5"
+             description="The percentage of time in which vertex fetch pipeline stage was slowing down the 3D pipeline."
+             data_type="float"
+             high_watermark="15"
+             equation="B 4 READ 100 UMUL $GpuCoreClocks FDIV"
+             max_equation="100"
+             underscore_name="vf_bottleneck"
+             units="percent"
+             symbol_name="VfBottleneck"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Indicate Draw"
+             mdapi_group="3D Pipe/Input Assembler"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="SLM Bytes Read"
+             description="The total number of GPU memory bytes read from shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ 64 UMUL"
+             underscore_name="slm_bytes_read"
+             units="bytes"
+             symbol_name="SlmBytesRead"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Strip-Fans Bottleneck"
+             low_watermark="5"
+             description="The percentage of time in which strip-fans pipeline stage was slowing down the 3D pipeline."
+             data_type="float"
+             high_watermark="10"
+             equation="B 6 READ 100 UMUL $GpuCoreClocks FDIV"
+             max_equation="100"
+             underscore_name="sf_bottleneck"
+             units="percent"
+             symbol_name="SfBottleneck"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Indicate Draw"
+             mdapi_group="3D Pipe/Rasterizer/Strip-Fans"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="SF Stall"
+             description="The percentage of time in which strip-fans pipeline stage was stalled."
+             data_type="float"
+             max_equation="100"
+             equation="C 4 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="sf_stall"
+             units="percent"
+             symbol_name="SfStall"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Correlate Draw"
+             mdapi_group="3D Pipe/Rasterizer/Strip-Fans"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GPU Busy"
+             description="The percentage of time in which the GPU has been processing GPU commands."
+             data_type="float"
+             max_equation="100"
+             equation="A 0 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="gpu_busy"
+             units="percent"
+             symbol_name="GpuBusy"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="HS Bottleneck"
+             low_watermark="3"
+             description="The percentage of time in which hull shader pipeline stage was slowing down the 3D pipeline."
+             data_type="float"
+             high_watermark="9"
+             equation="C 2 READ 100 UMUL $GpuCoreClocks FDIV"
+             max_equation="100"
+             underscore_name="hs_bottleneck"
+             units="percent"
+             symbol_name="HsBottleneck"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Indicate Draw"
+             mdapi_group="3D Pipe/Hull Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="CL Stall"
+             description="The percentage of time in which clipper pipeline stage was stalled."
+             data_type="float"
+             max_equation="100"
+             equation="C 5 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="cl_stall"
+             units="percent"
+             symbol_name="ClStall"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Correlate Draw"
+             mdapi_group="3D Pipe/Clipper"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="SO Bottleneck"
+             low_watermark="5"
+             description="The percentage of time in which stream output pipeline stage was slowing down the 3D pipeline."
+             data_type="float"
+             high_watermark="15"
+             equation="C 7 READ 100 UMUL $GpuCoreClocks FDIV"
+             max_equation="100"
+             underscore_name="so_bottleneck"
+             units="percent"
+             symbol_name="SoBottleneck"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Indicate Draw"
+             mdapi_group="3D Pipe/Stream Output"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Rasterized Pixels"
+             description="The total number of rasterized pixels."
+             data_type="uint64"
+             equation="A 21 READ 4 UMUL"
+             underscore_name="rasterized_pixels"
+             units="pixels"
+             symbol_name="RasterizedPixels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer"
+             />
+    <counter name="DS Threads Dispatched"
+             description="The total number of domain shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 3 READ"
+             underscore_name="ds_threads"
+             units="threads"
+             symbol_name="DsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Domain Shader"
+             />
+    <counter name="Samples Written"
+             description="The total number of samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 26 READ 4 UMUL"
+             underscore_name="samples_written"
+             units="pixels"
+             symbol_name="SamplesWritten"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="DS Bottleneck"
+             low_watermark="5"
+             description="The percentage of time in which domain shader pipeline stage was slowing down the 3D pipeline."
+             data_type="float"
+             high_watermark="15"
+             equation="C 1 READ 100 UMUL $GpuCoreClocks FDIV"
+             max_equation="100"
+             underscore_name="ds_bottleneck"
+             units="percent"
+             symbol_name="DsBottleneck"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Indicate Draw"
+             mdapi_group="3D Pipe/Domain Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EU Stall"
+             description="The percentage of time in which the Execution Units were stalled."
+             data_type="float"
+             max_equation="100"
+             equation="A 8 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_stall"
+             units="percent"
+             symbol_name="EuStall"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Clipper Bottleneck"
+             low_watermark="10"
+             description="The percentage of time in which clipper pipeline stage was slowing down the 3D pipeline."
+             data_type="float"
+             high_watermark="30"
+             equation="B 7 READ 100 UMUL $GpuCoreClocks FDIV"
+             max_equation="100"
+             underscore_name="cl_bottleneck"
+             units="percent"
+             symbol_name="ClBottleneck"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Indicate Draw"
+             mdapi_group="3D Pipe/Clipper"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="DS Stall"
+             description="The percentage of time in which domain shader pipeline stage was stalled."
+             data_type="float"
+             max_equation="100"
+             equation="B 0 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ds_stall"
+             units="percent"
+             symbol_name="DsStall"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Correlate Draw"
+             mdapi_group="3D Pipe/Domain Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Early Depth Bottleneck"
+             low_watermark="10"
+             description="The percentage of time in which early depth test pipeline stage was slowing down the 3D pipeline."
+             data_type="float"
+             high_watermark="30"
+             equation="B 5 READ 100 UMUL $GpuCoreClocks FDIV"
+             max_equation="100"
+             underscore_name="early_depth_bottleneck"
+             units="percent"
+             symbol_name="EarlyDepthBottleneck"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Indicate Draw"
+             mdapi_group="3D Pipe/Rasterizer/Early Depth Test"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Samples Blended"
+             description="The total number of blended samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 27 READ 4 UMUL"
+             underscore_name="samples_blended"
+             units="pixels"
+             symbol_name="SamplesBlended"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="Early Depth Test Fails"
+             description="The total number of pixels dropped on early depth test."
+             data_type="uint64"
+             equation="A 23 READ 4 UMUL"
+             underscore_name="early_depth_test_fails"
+             units="pixels"
+             symbol_name="EarlyDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Early Depth Test"
+             />
+    <counter name="Shader Memory Accesses"
+             description="The total number of shader memory accesses to L3."
+             data_type="uint64"
+             equation="A 32 READ"
+             underscore_name="shader_memory_accesses"
+             units="messages"
+             symbol_name="ShaderMemoryAccesses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="HS Threads Dispatched"
+             description="The total number of hull shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 2 READ"
+             underscore_name="hs_threads"
+             units="threads"
+             symbol_name="HsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Hull Shader"
+             />
+    <counter name="L3 Shader Throughput"
+             description="The total number of GPU memory bytes transferred between shaders and L3 caches w/o URB."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ A 31 READ UADD A 32 READ UADD 64 UMUL"
+             underscore_name="l3_shader_throughput"
+             units="bytes"
+             symbol_name="L3ShaderThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_group="L3/Data Port"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Samples Killed in FS"
+             description="The total number of samples or pixels dropped in fragment shaders."
+             data_type="uint64"
+             equation="A 24 READ 4 UMUL"
+             underscore_name="samples_killed_in_ps"
+             units="pixels"
+             symbol_name="SamplesKilledInPs"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Fragment Shader"
+             />
+    <counter name="SLM Bytes Written"
+             description="The total number of GPU memory bytes written into shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 31 READ 64 UMUL"
+             underscore_name="slm_bytes_written"
+             units="bytes"
+             symbol_name="SlmBytesWritten"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="SO Stall"
+             description="The percentage of time in which stream-output pipeline stage was stalled."
+             data_type="float"
+             max_equation="100"
+             equation="C 6 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="so_stall"
+             units="percent"
+             symbol_name="SoStall"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Correlate Draw"
+             mdapi_group="3D Pipe/Stream Output"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Shader Atomic Memory Accesses"
+             description="The total number of shader atomic memory accesses."
+             data_type="uint64"
+             equation="A 34 READ"
+             underscore_name="shader_atomics"
+             units="messages"
+             symbol_name="ShaderAtomics"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port/Atomics"
+             />
+    <register_config type="NOA">
+        <register type="NOA" address="0x00000D04" value="0x00000200" />
+        <register type="NOA" address="0x00009884" value="0x00000000" />
+        <register type="NOA" address="0x00009888" value="0x0A7003E0" />
+        <register type="NOA" address="0x00009888" value="0x0A710000" />
+        <register type="NOA" address="0x00009888" value="0x1017001E" />
+        <register type="NOA" address="0x00009888" value="0x06700015" />
+        <register type="NOA" address="0x00009888" value="0x00700000" />
+        <register type="NOA" address="0x00009888" value="0x06710800" />
+        <register type="NOA" address="0x00009888" value="0x00710000" />
+        <register type="NOA" address="0x00009888" value="0x0C746000" />
+        <register type="NOA" address="0x00009888" value="0x0C75A000" />
+        <register type="NOA" address="0x00009888" value="0x0E754000" />
+        <register type="NOA" address="0x00009888" value="0x0E762800" />
+        <register type="NOA" address="0x00009888" value="0x10760002" />
+        <register type="NOA" address="0x00009888" value="0x1477C000" />
+        <register type="NOA" address="0x00009888" value="0x16770330" />
+        <register type="NOA" address="0x00009888" value="0x1A190040" />
+        <register type="NOA" address="0x00009888" value="0x08170001" />
+        <register type="NOA" address="0x00009888" value="0x1C170000" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x01160031" />
+        <register type="NOA" address="0x00009888" value="0x11177C00" />
+        <register type="NOA" address="0x00009888" value="0x11187C00" />
+        <register type="NOA" address="0x00009888" value="0x05190000" />
+        <register type="NOA" address="0x00009888" value="0x110E7800" />
+        <register type="NOA" address="0x00009888" value="0x130F0000" />
+        <register type="NOA" address="0x00009888" value="0x15120000" />
+        <register type="NOA" address="0x00009888" value="0x11130019" />
+        <register type="NOA" address="0x00009888" value="0x0F160C80" />
+        <register type="NOA" address="0x00009888" value="0x0B160000" />
+        <register type="NOA" address="0x00009888" value="0x07162000" />
+        <register type="NOA" address="0x00009888" value="0x03168000" />
+        <register type="NOA" address="0x00009888" value="0x071705C0" />
+        <register type="NOA" address="0x00009888" value="0x011704C0" />
+        <register type="NOA" address="0x00009888" value="0x05170000" />
+        <register type="NOA" address="0x00009888" value="0x0B180015" />
+        <register type="NOA" address="0x00009888" value="0x01185000" />
+        <register type="NOA" address="0x00009888" value="0x07185000" />
+        <register type="NOA" address="0x00009888" value="0x0D185000" />
+        <register type="NOA" address="0x00009888" value="0x0F185000" />
+        <register type="NOA" address="0x00009888" value="0x03186000" />
+        <register type="NOA" address="0x00009888" value="0x05185000" />
+        <register type="NOA" address="0x00009888" value="0x03190022" />
+        <register type="NOA" address="0x00009888" value="0x0F190000" />
+        <register type="NOA" address="0x00009888" value="0x09190000" />
+        <register type="NOA" address="0x00009888" value="0x0D0E8023" />
+        <register type="NOA" address="0x00009888" value="0x030E1100" />
+        <register type="NOA" address="0x00009888" value="0x050E4000" />
+        <register type="NOA" address="0x00009888" value="0x0D0F0180" />
+        <register type="NOA" address="0x00009888" value="0x050F0001" />
+        <register type="NOA" address="0x00009888" value="0x010F0000" />
+        <register type="NOA" address="0x00009888" value="0x1B100054" />
+        <register type="NOA" address="0x00009888" value="0x07104000" />
+        <register type="NOA" address="0x00009888" value="0x09104000" />
+        <register type="NOA" address="0x00009888" value="0x0B104000" />
+        <register type="NOA" address="0x00009888" value="0x0D104000" />
+        <register type="NOA" address="0x00009888" value="0x1B114000" />
+        <register type="NOA" address="0x00009888" value="0x1D110006" />
+        <register type="NOA" address="0x00009888" value="0x07118000" />
+        <register type="NOA" address="0x00009888" value="0x09118000" />
+        <register type="NOA" address="0x00009888" value="0x0B114000" />
+        <register type="NOA" address="0x00009888" value="0x0D114000" />
+        <register type="NOA" address="0x00009888" value="0x05127100" />
+        <register type="NOA" address="0x00009888" value="0x07120073" />
+        <register type="NOA" address="0x00009888" value="0x21120001" />
+        <register type="NOA" address="0x00009888" value="0x1F120000" />
+        <register type="NOA" address="0x00009888" value="0x0F130002" />
+        <register type="NOA" address="0x00009888" value="0x01130000" />
+        <register type="NOA" address="0x00009888" value="0x0B144000" />
+        <register type="NOA" address="0x00009888" value="0x0F144025" />
+        <register type="NOA" address="0x00009888" value="0x11140025" />
+        <register type="NOA" address="0x00009888" value="0x0D145200" />
+        <register type="NOA" address="0x00009888" value="0x11158000" />
+        <register type="NOA" address="0x00009888" value="0x1F158080" />
+        <register type="NOA" address="0x00009888" value="0x13158000" />
+        <register type="NOA" address="0x00009888" value="0x0109A000" />
+        <register type="NOA" address="0x00009888" value="0x0709A000" />
+        <register type="NOA" address="0x00009888" value="0x0B092000" />
+        <register type="NOA" address="0x00009888" value="0x0D09A000" />
+        <register type="NOA" address="0x00009888" value="0x0F09A000" />
+        <register type="NOA" address="0x00009888" value="0x0309A000" />
+        <register type="NOA" address="0x00009888" value="0x0509A000" />
+        <register type="NOA" address="0x00009888" value="0x010BC000" />
+        <register type="NOA" address="0x00009888" value="0x070BC000" />
+        <register type="NOA" address="0x00009888" value="0x0B0B4000" />
+        <register type="NOA" address="0x00009888" value="0x0D0BC000" />
+        <register type="NOA" address="0x00009888" value="0x0F0BC000" />
+        <register type="NOA" address="0x00009888" value="0x030BC000" />
+        <register type="NOA" address="0x00009888" value="0x050BC000" />
+        <register type="NOA" address="0x00009888" value="0x09045000" />
+        <register type="NOA" address="0x00009888" value="0x0F045000" />
+        <register type="NOA" address="0x00009888" value="0x11044000" />
+        <register type="NOA" address="0x00009888" value="0x13040154" />
+        <register type="NOA" address="0x00009888" value="0x0B045000" />
+        <register type="NOA" address="0x00009888" value="0x0D045000" />
+        <register type="NOA" address="0x00009888" value="0x11004000" />
+        <register type="NOA" address="0x00009888" value="0x1F004000" />
+        <register type="NOA" address="0x00009888" value="0x2100C000" />
+        <register type="NOA" address="0x00009888" value="0x2300C000" />
+        <register type="NOA" address="0x00009888" value="0x25004000" />
+        <register type="NOA" address="0x00009888" value="0x2700C000" />
+        <register type="NOA" address="0x00009888" value="0x29004000" />
+        <register type="NOA" address="0x00009888" value="0x2B004000" />
+        <register type="NOA" address="0x00009888" value="0x2D004000" />
+        <register type="NOA" address="0x00009888" value="0x2F004000" />
+        <register type="NOA" address="0x00009888" value="0x13004000" />
+        <register type="NOA" address="0x00009888" value="0x15004000" />
+        <register type="NOA" address="0x00009888" value="0x17004000" />
+        <register type="NOA" address="0x00009888" value="0x19004000" />
+        <register type="NOA" address="0x00009888" value="0x1B004000" />
+        <register type="NOA" address="0x00009888" value="0x1D004000" />
+        <register type="NOA" address="0x00009888" value="0x47001111" />
+        <register type="NOA" address="0x00009888" value="0x39001022" />
+        <register type="NOA" address="0x00009888" value="0x49001111" />
+        <register type="NOA" address="0x00009888" value="0x3D000021" />
+        <register type="NOA" address="0x00009888" value="0x4B000100" />
+        <register type="NOA" address="0x00009888" value="0x3F0018A5" />
+        <register type="NOA" address="0x00009888" value="0x4D001111" />
+        <register type="NOA" address="0x00009888" value="0x41000001" />
+        <register type="NOA" address="0x00009888" value="0x35000000" />
+        <register type="NOA" address="0x00009888" value="0x31000000" />
+        <register type="NOA" address="0x00009888" value="0x43000002" />
+        <register type="NOA" address="0x00009888" value="0x3B000401" />
+    </register_config>
+    <register_config type="OA">
+        <register type="OA" address="0x00002740" value="0x00000000" />
+        <register type="OA" address="0x00002710" value="0x00000000" />
+        <register type="OA" address="0x00002714" value="0x00800000" />
+        <register type="OA" address="0x00002720" value="0x00000000" />
+        <register type="OA" address="0x00002724" value="0x00800000" />
+    </register_config>
+    <register_config type="FLEX">
+        <register type="FLEX" address="0x0000E458" value="0x00005004" />
+    </register_config>
+  </set>
+
+  <set name="Memory Reads Distribution Gen10"
+       mdapi_supported_apis="OGL OCL IO BB"
+       underscore_name="memory_reads"
+       hw_config_guid="3500dcfd-837b-4ec8-b5cf-fe58b966263f"
+       chipset="CNL"
+       symbol_name="MemoryReads"
+       >
+    <counter name="GPU Core Clocks"
+             description="The total number of GPU core clocks elapsed during the measurement."
+             data_type="uint64"
+             equation="GPU_CLOCK 0 READ"
+             underscore_name="gpu_core_clocks"
+             units="cycles"
+             symbol_name="GpuCoreClocks"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="EU Active"
+             description="The percentage of time in which the Execution Units were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 7 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_active"
+             units="percent"
+             symbol_name="EuActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GtiL3Bank0Reads"
+             description="The total number of GTI memory reads from L3 Bank 0 (L3 Cache misses)."
+             data_type="uint64"
+             equation="C 1 READ"
+             underscore_name="gti_l3_bank0_reads"
+             units="messages"
+             symbol_name="GtiL3Bank0Reads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI/L3"
+             />
+    <counter name="GS Threads Dispatched"
+             description="The total number of geometry shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 5 READ"
+             underscore_name="gs_threads"
+             units="threads"
+             symbol_name="GsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Geometry Shader"
+             />
+    <counter name="GtiRingAccesses"
+             description="The total number of all accesses from GTI to the ring."
+             data_type="uint64"
+             equation="C 7 READ"
+             underscore_name="gti_ring_accesses"
+             units="messages"
+             symbol_name="GtiRingAccesses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI"
+             />
+    <counter name="Early Hi-Depth Test Fails"
+             description="The total number of pixels dropped on early hierarchical depth test."
+             data_type="uint64"
+             equation="A 22 READ 4 UMUL"
+             underscore_name="hi_depth_test_fails"
+             units="pixels"
+             symbol_name="HiDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Hi-Depth Test"
+             />
+    <counter name="VS Threads Dispatched"
+             description="The total number of vertex shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 1 READ"
+             underscore_name="vs_threads"
+             units="threads"
+             symbol_name="VsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Vertex Shader"
+             />
+    <counter name="FS Threads Dispatched"
+             description="The total number of fragment shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 6 READ"
+             underscore_name="ps_threads"
+             units="threads"
+             symbol_name="PsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Fragment Shader"
+             />
+    <counter name="Shader Barrier Messages"
+             description="The total number of shader barrier messages."
+             data_type="uint64"
+             equation="A 35 READ"
+             underscore_name="shader_barriers"
+             units="messages"
+             symbol_name="ShaderBarriers"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Barrier"
+             />
+    <counter name="GtiRsMemoryReads"
+             description="The total number of GTI memory reads from Resource Streamer."
+             data_type="uint64"
+             equation="B 2 READ"
+             underscore_name="gti_rs_memory_reads"
+             units="messages"
+             symbol_name="GtiRsMemoryReads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI/3D Pipe/Resource Streamer"
+             />
+    <counter name="Sampler Texels"
+             description="The total number of texels seen on input (with 2x2 accuracy) in all sampler units."
+             data_type="uint64"
+             equation="A 28 READ 4 UMUL"
+             underscore_name="sampler_texels"
+             units="texels"
+             symbol_name="SamplerTexels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Input"
+             />
+    <counter name="Pixels Failing Tests"
+             description="The total number of pixels dropped on post-FS alpha, stencil, or depth tests."
+             data_type="uint64"
+             equation="A 25 READ 4 UMUL"
+             underscore_name="pixels_failing_post_ps_tests"
+             units="pixels"
+             symbol_name="PixelsFailingPostPsTests"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="GtiHizMemoryReads"
+             description="The total number of GTI memory reads from Hierarchical Depth Cache (Hi-Depth Cache misses)."
+             data_type="uint64"
+             equation="B 5 READ"
+             underscore_name="gti_hiz_memory_reads"
+             units="messages"
+             symbol_name="GtiHizMemoryReads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI/Depth Cache"
+             />
+    <counter name="GPU Time Elapsed"
+             description="Time elapsed on the GPU during the measurement."
+             data_type="uint64"
+             equation="GPU_TIME 0 READ 1000000000 UMUL $GpuTimestampFrequency UDIV"
+             underscore_name="gpu_time"
+             units="ns"
+             symbol_name="GpuTime"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="AVG GPU Core Frequency"
+             description="Average GPU Core Frequency in the measurement."
+             data_type="uint64"
+             max_equation="$GpuMaxFrequency"
+             equation="$GpuCoreClocks 1000000000 UMUL $GpuTime UDIV"
+             underscore_name="avg_gpu_core_frequency"
+             units="hz"
+             symbol_name="AvgGpuCoreFrequency"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Sampler Texels Misses"
+             description="The total number of texels lookups (with 2x2 accuracy) that missed L1 sampler cache."
+             data_type="uint64"
+             equation="A 29 READ 4 UMUL"
+             underscore_name="sampler_texel_misses"
+             units="texels"
+             symbol_name="SamplerTexelMisses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Batch Frame Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Cache"
+             />
+    <counter name="GtiRccMemoryReads"
+             description="The total number of GTI memory reads from Render Color Cache (Render Color Cache misses)."
+             data_type="uint64"
+             equation="B 3 READ"
+             underscore_name="gti_rcc_memory_reads"
+             units="messages"
+             symbol_name="GtiRccMemoryReads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI/Color Cache"
+             />
+    <counter name="CS Threads Dispatched"
+             description="The total number of compute shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 4 READ"
+             underscore_name="cs_threads"
+             units="threads"
+             symbol_name="CsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Compute Shader"
+             />
+    <counter name="SLM Bytes Read"
+             description="The total number of GPU memory bytes read from shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ 64 UMUL"
+             underscore_name="slm_bytes_read"
+             units="bytes"
+             symbol_name="SlmBytesRead"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GtiL3Bank1Reads"
+             description="The total number of GTI memory reads from L3 Bank 1 (L3 Cache misses)."
+             data_type="uint64"
+             equation="C 2 READ"
+             underscore_name="gti_l3_bank1_reads"
+             units="messages"
+             symbol_name="GtiL3Bank1Reads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="subslice"
+             mdapi_group="GTI/L3"
+             />
+    <counter name="GPU Busy"
+             description="The percentage of time in which the GPU has been processing GPU commands."
+             data_type="float"
+             max_equation="100"
+             equation="A 0 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="gpu_busy"
+             units="percent"
+             symbol_name="GpuBusy"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GtiCmdStreamerMemoryReads"
+             description="The total number of GTI memory reads from Command Streamer."
+             data_type="uint64"
+             equation="B 0 READ"
+             underscore_name="gti_cmd_streamer_memory_reads"
+             units="messages"
+             symbol_name="GtiCmdStreamerMemoryReads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI/3D Pipe/Command Streamer"
+             />
+    <counter name="GtiL3Bank2Reads"
+             description="The total number of GTI memory reads from L3 Bank 2 (L3 Cache misses)."
+             data_type="uint64"
+             equation="C 3 READ"
+             underscore_name="gti_l3_bank2_reads"
+             units="messages"
+             symbol_name="GtiL3Bank2Reads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="subslice"
+             mdapi_group="GTI/L3"
+             />
+    <counter name="GtiMemoryReads"
+             description="The total number of GTI memory reads."
+             data_type="uint64"
+             equation="C 0 READ"
+             underscore_name="gti_memory_reads"
+             units="messages"
+             symbol_name="GtiMemoryReads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI"
+             />
+    <counter name="Rasterized Pixels"
+             description="The total number of rasterized pixels."
+             data_type="uint64"
+             equation="A 21 READ 4 UMUL"
+             underscore_name="rasterized_pixels"
+             units="pixels"
+             symbol_name="RasterizedPixels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer"
+             />
+    <counter name="GtiRczMemoryReads"
+             description="The total number of GTI memory reads from Render Depth Cache (Render Depth Cache misses)."
+             data_type="uint64"
+             equation="B 7 READ"
+             underscore_name="gti_rcz_memory_reads"
+             units="messages"
+             symbol_name="GtiRczMemoryReads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI/Depth Cache"
+             />
+    <counter name="DS Threads Dispatched"
+             description="The total number of domain shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 3 READ"
+             underscore_name="ds_threads"
+             units="threads"
+             symbol_name="DsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Domain Shader"
+             />
+    <counter name="Samples Written"
+             description="The total number of samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 26 READ 4 UMUL"
+             underscore_name="samples_written"
+             units="pixels"
+             symbol_name="SamplesWritten"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="EU Stall"
+             description="The percentage of time in which the Execution Units were stalled."
+             data_type="float"
+             max_equation="100"
+             equation="A 8 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_stall"
+             units="percent"
+             symbol_name="EuStall"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Samples Blended"
+             description="The total number of blended samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 27 READ 4 UMUL"
+             underscore_name="samples_blended"
+             units="pixels"
+             symbol_name="SamplesBlended"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="Early Depth Test Fails"
+             description="The total number of pixels dropped on early depth test."
+             data_type="uint64"
+             equation="A 23 READ 4 UMUL"
+             underscore_name="early_depth_test_fails"
+             units="pixels"
+             symbol_name="EarlyDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Early Depth Test"
+             />
+    <counter name="GtiMscMemoryReads"
+             description="The total number of GTI memory reads from Multisampling Color Cache (Multisampling Color Cache misses)."
+             data_type="uint64"
+             equation="B 4 READ"
+             underscore_name="gti_msc_memory_reads"
+             units="messages"
+             symbol_name="GtiMscMemoryReads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI/Color Cache"
+             />
+    <counter name="GtiVfMemoryReads"
+             description="The total number of GTI memory reads from Vertex Fetch."
+             data_type="uint64"
+             equation="B 1 READ"
+             underscore_name="gti_vf_memory_reads"
+             units="messages"
+             symbol_name="GtiVfMemoryReads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI/3D Pipe/Vertex Fetch"
+             />
+    <counter name="GtiStcMemoryReads"
+             description="The total number of GTI memory reads from Stencil Cache (Stencil Cache misses)."
+             data_type="uint64"
+             equation="B 6 READ"
+             underscore_name="gti_stc_memory_reads"
+             units="messages"
+             symbol_name="GtiStcMemoryReads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI/Depth Cache"
+             />
+    <counter name="Shader Memory Accesses"
+             description="The total number of shader memory accesses to L3."
+             data_type="uint64"
+             equation="A 32 READ"
+             underscore_name="shader_memory_accesses"
+             units="messages"
+             symbol_name="ShaderMemoryAccesses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="HS Threads Dispatched"
+             description="The total number of hull shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 2 READ"
+             underscore_name="hs_threads"
+             units="threads"
+             symbol_name="HsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Hull Shader"
+             />
+    <counter name="Samples Killed in FS"
+             description="The total number of samples or pixels dropped in fragment shaders."
+             data_type="uint64"
+             equation="A 24 READ 4 UMUL"
+             underscore_name="samples_killed_in_ps"
+             units="pixels"
+             symbol_name="SamplesKilledInPs"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Fragment Shader"
+             />
+    <counter name="SLM Bytes Written"
+             description="The total number of GPU memory bytes written into shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 31 READ 64 UMUL"
+             underscore_name="slm_bytes_written"
+             units="bytes"
+             symbol_name="SlmBytesWritten"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GtiL3Reads"
+             description="The total number of GTI memory reads from L3 (L3 Cache misses)."
+             data_type="uint64"
+             equation="$GtiL3Bank0Reads $GtiL3Bank1Reads UADD $GtiL3Bank2Reads UADD"
+             underscore_name="gti_l3_reads"
+             units="messages"
+             symbol_name="GtiL3Reads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI/L3"
+             />
+    <counter name="Shader Atomic Memory Accesses"
+             description="The total number of shader atomic memory accesses."
+             data_type="uint64"
+             equation="A 34 READ"
+             underscore_name="shader_atomics"
+             units="messages"
+             symbol_name="ShaderAtomics"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port/Atomics"
+             />
+    <register_config type="NOA">
+        <register type="NOA" address="0x00000D04" value="0x00000200" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x17066860" />
+        <register type="NOA" address="0x00009888" value="0x15070017" />
+        <register type="NOA" address="0x00009888" value="0x1F072920" />
+        <register type="NOA" address="0x00009888" value="0x1B038000" />
+        <register type="NOA" address="0x00009888" value="0x0D038000" />
+        <register type="NOA" address="0x00009888" value="0x0F034000" />
+        <register type="NOA" address="0x00009888" value="0x0F063180" />
+        <register type="NOA" address="0x00009888" value="0x01063100" />
+        <register type="NOA" address="0x00009888" value="0x0306006B" />
+        <register type="NOA" address="0x00009888" value="0x05060000" />
+        <register type="NOA" address="0x00009888" value="0x11080100" />
+        <register type="NOA" address="0x00009888" value="0x0D085000" />
+        <register type="NOA" address="0x00009888" value="0x0F094000" />
+        <register type="NOA" address="0x00009888" value="0x01094000" />
+        <register type="NOA" address="0x00009888" value="0x03091000" />
+        <register type="NOA" address="0x00009888" value="0x09046000" />
+        <register type="NOA" address="0x00009888" value="0x0F04A000" />
+        <register type="NOA" address="0x00009888" value="0x1104A800" />
+        <register type="NOA" address="0x00009888" value="0x130401AA" />
+        <register type="NOA" address="0x00009888" value="0x0B041000" />
+        <register type="NOA" address="0x00009888" value="0x010700E0" />
+        <register type="NOA" address="0x00009888" value="0x0707E153" />
+        <register type="NOA" address="0x00009888" value="0x0907E3E2" />
+        <register type="NOA" address="0x00009888" value="0x0B07E5E4" />
+        <register type="NOA" address="0x00009888" value="0x0D07E7E6" />
+        <register type="NOA" address="0x00009888" value="0x0F0700EF" />
+        <register type="NOA" address="0x00009888" value="0x25070000" />
+        <register type="NOA" address="0x00009888" value="0x11070000" />
+        <register type="NOA" address="0x00009888" value="0x23070000" />
+        <register type="NOA" address="0x00009888" value="0x1D070000" />
+        <register type="NOA" address="0x00009888" value="0x11004000" />
+        <register type="NOA" address="0x00009888" value="0x1F004000" />
+        <register type="NOA" address="0x00009888" value="0x21004000" />
+        <register type="NOA" address="0x00009888" value="0x23004000" />
+        <register type="NOA" address="0x00009888" value="0x25004000" />
+        <register type="NOA" address="0x00009888" value="0x27004000" />
+        <register type="NOA" address="0x00009888" value="0x29004000" />
+        <register type="NOA" address="0x00009888" value="0x2B004000" />
+        <register type="NOA" address="0x00009888" value="0x2D004000" />
+        <register type="NOA" address="0x00009888" value="0x2F004000" />
+        <register type="NOA" address="0x00009888" value="0x13004000" />
+        <register type="NOA" address="0x00009888" value="0x15004000" />
+        <register type="NOA" address="0x00009888" value="0x1D004000" />
+        <register type="NOA" address="0x00009888" value="0x47000000" />
+        <register type="NOA" address="0x00009888" value="0x39000003" />
+        <register type="NOA" address="0x00009888" value="0x49000000" />
+        <register type="NOA" address="0x00009888" value="0x3D000C63" />
+        <register type="NOA" address="0x00009888" value="0x4B000000" />
+        <register type="NOA" address="0x00009888" value="0x3F000C63" />
+        <register type="NOA" address="0x00009888" value="0x4D000000" />
+        <register type="NOA" address="0x00009888" value="0x41000C63" />
+        <register type="NOA" address="0x00009888" value="0x35000000" />
+        <register type="NOA" address="0x00009888" value="0x31000000" />
+        <register type="NOA" address="0x00009888" value="0x43000000" />
+    </register_config>
+    <register_config type="OA">
+        <register type="OA" address="0x00002740" value="0x00000000" />
+        <register type="OA" address="0x00002710" value="0x00000000" />
+        <register type="OA" address="0x00002714" value="0xF0800000" />
+        <register type="OA" address="0x00002720" value="0x00000000" />
+        <register type="OA" address="0x00002724" value="0xF0800000" />
+        <register type="OA" address="0x00002770" value="0x0000081A" />
+        <register type="OA" address="0x00002774" value="0x0000FE00" />
+        <register type="OA" address="0x00002778" value="0x0000082A" />
+        <register type="OA" address="0x0000277C" value="0x0000FE00" />
+        <register type="OA" address="0x00002780" value="0x00000872" />
+        <register type="OA" address="0x00002784" value="0x0000FE00" />
+        <register type="OA" address="0x00002788" value="0x000008BA" />
+        <register type="OA" address="0x0000278C" value="0x0000FE00" />
+        <register type="OA" address="0x00002790" value="0x0000087A" />
+        <register type="OA" address="0x00002794" value="0x0000FE00" />
+        <register type="OA" address="0x00002798" value="0x000008EA" />
+        <register type="OA" address="0x0000279C" value="0x0000FE00" />
+        <register type="OA" address="0x000027A0" value="0x000008E2" />
+        <register type="OA" address="0x000027A4" value="0x0000FE00" />
+        <register type="OA" address="0x000027A8" value="0x000008F2" />
+        <register type="OA" address="0x000027AC" value="0x0000FE00" />
+        <register type="OA" address="0x0000274C" value="0x86543210" />
+        <register type="OA" address="0x00002744" value="0x00006667" />
+        <register type="OA" address="0x00002748" value="0x86543210" />
+        <register type="OA" address="0x0000275C" value="0x86543210" />
+        <register type="OA" address="0x00002754" value="0x00000065" />
+    </register_config>
+    <register_config type="FLEX">
+        <register type="FLEX" address="0x0000E458" value="0x00005004" />
+    </register_config>
+  </set>
+
+  <set name="Memory Writes Distribution Gen10"
+       mdapi_supported_apis="OGL OCL IO BB"
+       underscore_name="memory_writes"
+       hw_config_guid="63d18ead-3cfc-4972-a640-5e98b688dfee"
+       chipset="CNL"
+       symbol_name="MemoryWrites"
+       >
+    <counter name="GPU Core Clocks"
+             description="The total number of GPU core clocks elapsed during the measurement."
+             data_type="uint64"
+             equation="GPU_CLOCK 0 READ"
+             underscore_name="gpu_core_clocks"
+             units="cycles"
+             symbol_name="GpuCoreClocks"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="EU Active"
+             description="The percentage of time in which the Execution Units were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 7 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_active"
+             units="percent"
+             symbol_name="EuActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GtiMemoryWrites"
+             description="The total number of GTI memory writes."
+             data_type="uint64"
+             equation="C 0 READ"
+             underscore_name="gti_memory_writes"
+             units="messages"
+             symbol_name="GtiMemoryWrites"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI"
+             />
+    <counter name="GS Threads Dispatched"
+             description="The total number of geometry shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 5 READ"
+             underscore_name="gs_threads"
+             units="threads"
+             symbol_name="GsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Geometry Shader"
+             />
+    <counter name="GtiRingAccesses"
+             description="The total number of all accesses from GTI to the ring."
+             data_type="uint64"
+             equation="C 7 READ"
+             underscore_name="gti_ring_accesses"
+             units="messages"
+             symbol_name="GtiRingAccesses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI"
+             />
+    <counter name="Early Hi-Depth Test Fails"
+             description="The total number of pixels dropped on early hierarchical depth test."
+             data_type="uint64"
+             equation="A 22 READ 4 UMUL"
+             underscore_name="hi_depth_test_fails"
+             units="pixels"
+             symbol_name="HiDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Hi-Depth Test"
+             />
+    <counter name="VS Threads Dispatched"
+             description="The total number of vertex shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 1 READ"
+             underscore_name="vs_threads"
+             units="threads"
+             symbol_name="VsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Vertex Shader"
+             />
+    <counter name="FS Threads Dispatched"
+             description="The total number of fragment shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 6 READ"
+             underscore_name="ps_threads"
+             units="threads"
+             symbol_name="PsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Fragment Shader"
+             />
+    <counter name="GtiMscMemoryWrites"
+             description="The total number of GTI memory writes from Multisampling Color Cache (Multisampling Color Cache invalidations)."
+             data_type="uint64"
+             equation="B 3 READ"
+             underscore_name="gti_msc_memory_writes"
+             units="messages"
+             symbol_name="GtiMscMemoryWrites"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI/Color Cache"
+             />
+    <counter name="Shader Barrier Messages"
+             description="The total number of shader barrier messages."
+             data_type="uint64"
+             equation="A 35 READ"
+             underscore_name="shader_barriers"
+             units="messages"
+             symbol_name="ShaderBarriers"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Barrier"
+             />
+    <counter name="GtiCmdStreamerMemoryWrites"
+             description="The total number of GTI memory writes from Command Streamer."
+             data_type="uint64"
+             equation="B 0 READ"
+             underscore_name="gti_cmd_streamer_memory_writes"
+             units="messages"
+             symbol_name="GtiCmdStreamerMemoryWrites"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI/3D Pipe/Command Streamer"
+             />
+    <counter name="Sampler Texels"
+             description="The total number of texels seen on input (with 2x2 accuracy) in all sampler units."
+             data_type="uint64"
+             equation="A 28 READ 4 UMUL"
+             underscore_name="sampler_texels"
+             units="texels"
+             symbol_name="SamplerTexels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Input"
+             />
+    <counter name="Pixels Failing Tests"
+             description="The total number of pixels dropped on post-FS alpha, stencil, or depth tests."
+             data_type="uint64"
+             equation="A 25 READ 4 UMUL"
+             underscore_name="pixels_failing_post_ps_tests"
+             units="pixels"
+             symbol_name="PixelsFailingPostPsTests"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="GtiL3Bank0Writes"
+             description="The total number of GTI memory writes from L3 Bank 0 (L3 Bank 0 invalidations)."
+             data_type="uint64"
+             equation="C 1 READ"
+             underscore_name="gti_l3_bank0_writes"
+             units="messages"
+             symbol_name="GtiL3Bank0Writes"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI/L3"
+             />
+    <counter name="GtiL3Bank1Writes"
+             description="The total number of GTI memory writes from L3 Bank 1 (L3 Bank 1 invalidations)."
+             data_type="uint64"
+             equation="C 2 READ"
+             underscore_name="gti_l3_bank1_writes"
+             units="messages"
+             symbol_name="GtiL3Bank1Writes"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI/L3"
+             />
+    <counter name="GtiL3Bank2Writes"
+             description="The total number of GTI memory writes from L3 Bank 2 (L3 Bank 2 invalidations)."
+             data_type="uint64"
+             equation="C 3 READ"
+             underscore_name="gti_l3_bank2_writes"
+             units="messages"
+             symbol_name="GtiL3Bank2Writes"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI/L3"
+             />
+    <counter name="GtiL3Writes"
+             description="The total number of GTI memory writes from L3 (L3 invalidations)."
+             data_type="uint64"
+             equation="$GtiL3Bank0Writes $GtiL3Bank1Writes UADD $GtiL3Bank2Writes UADD"
+             underscore_name="gti_l3_writes"
+             units="messages"
+             symbol_name="GtiL3Writes"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI/L3"
+             />
+    <counter name="GPU Time Elapsed"
+             description="Time elapsed on the GPU during the measurement."
+             data_type="uint64"
+             equation="GPU_TIME 0 READ 1000000000 UMUL $GpuTimestampFrequency UDIV"
+             underscore_name="gpu_time"
+             units="ns"
+             symbol_name="GpuTime"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="AVG GPU Core Frequency"
+             description="Average GPU Core Frequency in the measurement."
+             data_type="uint64"
+             max_equation="$GpuMaxFrequency"
+             equation="$GpuCoreClocks 1000000000 UMUL $GpuTime UDIV"
+             underscore_name="avg_gpu_core_frequency"
+             units="hz"
+             symbol_name="AvgGpuCoreFrequency"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Sampler Texels Misses"
+             description="The total number of texels lookups (with 2x2 accuracy) that missed L1 sampler cache."
+             data_type="uint64"
+             equation="A 29 READ 4 UMUL"
+             underscore_name="sampler_texel_misses"
+             units="texels"
+             symbol_name="SamplerTexelMisses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Batch Frame Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Cache"
+             />
+    <counter name="CS Threads Dispatched"
+             description="The total number of compute shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 4 READ"
+             underscore_name="cs_threads"
+             units="threads"
+             symbol_name="CsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Compute Shader"
+             />
+    <counter name="SLM Bytes Read"
+             description="The total number of GPU memory bytes read from shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ 64 UMUL"
+             underscore_name="slm_bytes_read"
+             units="bytes"
+             symbol_name="SlmBytesRead"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GtiRccMemoryWrites"
+             description="The total number of GTI memory writes from Render Color Cache (Render Color Cache invalidations)."
+             data_type="uint64"
+             equation="B 2 READ"
+             underscore_name="gti_rcc_memory_writes"
+             units="messages"
+             symbol_name="GtiRccMemoryWrites"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI/Color Cache"
+             />
+    <counter name="GtiSoMemoryWrites"
+             description="The total number of GTI memory writes from Stream Output."
+             data_type="uint64"
+             equation="B 1 READ"
+             underscore_name="gti_so_memory_writes"
+             units="messages"
+             symbol_name="GtiSoMemoryWrites"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI/3D Pipe/Stream Output"
+             />
+    <counter name="GPU Busy"
+             description="The percentage of time in which the GPU has been processing GPU commands."
+             data_type="float"
+             max_equation="100"
+             equation="A 0 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="gpu_busy"
+             units="percent"
+             symbol_name="GpuBusy"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GtiStcMemoryWrites"
+             description="The total number of GTI memory writes from Stencil Cache."
+             data_type="uint64"
+             equation="B 5 READ"
+             underscore_name="gti_stc_memory_writes"
+             units="messages"
+             symbol_name="GtiStcMemoryWrites"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI/Depth Cache"
+             />
+    <counter name="Rasterized Pixels"
+             description="The total number of rasterized pixels."
+             data_type="uint64"
+             equation="A 21 READ 4 UMUL"
+             underscore_name="rasterized_pixels"
+             units="pixels"
+             symbol_name="RasterizedPixels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer"
+             />
+    <counter name="DS Threads Dispatched"
+             description="The total number of domain shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 3 READ"
+             underscore_name="ds_threads"
+             units="threads"
+             symbol_name="DsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Domain Shader"
+             />
+    <counter name="Samples Written"
+             description="The total number of samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 26 READ 4 UMUL"
+             underscore_name="samples_written"
+             units="pixels"
+             symbol_name="SamplesWritten"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="EU Stall"
+             description="The percentage of time in which the Execution Units were stalled."
+             data_type="float"
+             max_equation="100"
+             equation="A 8 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_stall"
+             units="percent"
+             symbol_name="EuStall"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Samples Blended"
+             description="The total number of blended samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 27 READ 4 UMUL"
+             underscore_name="samples_blended"
+             units="pixels"
+             symbol_name="SamplesBlended"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="Early Depth Test Fails"
+             description="The total number of pixels dropped on early depth test."
+             data_type="uint64"
+             equation="A 23 READ 4 UMUL"
+             underscore_name="early_depth_test_fails"
+             units="pixels"
+             symbol_name="EarlyDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Early Depth Test"
+             />
+    <counter name="Shader Memory Accesses"
+             description="The total number of shader memory accesses to L3."
+             data_type="uint64"
+             equation="A 32 READ"
+             underscore_name="shader_memory_accesses"
+             units="messages"
+             symbol_name="ShaderMemoryAccesses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="HS Threads Dispatched"
+             description="The total number of hull shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 2 READ"
+             underscore_name="hs_threads"
+             units="threads"
+             symbol_name="HsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Hull Shader"
+             />
+    <counter name="GtiRczMemoryWrites"
+             description="The total number of GTI memory writes from Render Depth Cache."
+             data_type="uint64"
+             equation="B 6 READ"
+             underscore_name="gti_rcz_memory_writes"
+             units="messages"
+             symbol_name="GtiRczMemoryWrites"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI/Depth Cache"
+             />
+    <counter name="L3 Shader Throughput"
+             description="The total number of GPU memory bytes transferred between shaders and L3 caches w/o URB."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ A 31 READ UADD A 32 READ UADD 64 UMUL"
+             underscore_name="l3_shader_throughput"
+             units="bytes"
+             symbol_name="L3ShaderThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_group="L3/Data Port"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Samples Killed in FS"
+             description="The total number of samples or pixels dropped in fragment shaders."
+             data_type="uint64"
+             equation="A 24 READ 4 UMUL"
+             underscore_name="samples_killed_in_ps"
+             units="pixels"
+             symbol_name="SamplesKilledInPs"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Fragment Shader"
+             />
+    <counter name="SLM Bytes Written"
+             description="The total number of GPU memory bytes written into shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 31 READ 64 UMUL"
+             underscore_name="slm_bytes_written"
+             units="bytes"
+             symbol_name="SlmBytesWritten"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GtiHizMemoryWrites"
+             description="The total number of GTI memory writes from Hierarchical Depth Cache."
+             data_type="uint64"
+             equation="B 4 READ"
+             underscore_name="gti_hiz_memory_writes"
+             units="messages"
+             symbol_name="GtiHizMemoryWrites"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GTI/Depth Cache"
+             />
+    <counter name="Shader Atomic Memory Accesses"
+             description="The total number of shader atomic memory accesses."
+             data_type="uint64"
+             equation="A 34 READ"
+             underscore_name="shader_atomics"
+             units="messages"
+             symbol_name="ShaderAtomics"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port/Atomics"
+             />
+    <register_config type="NOA">
+        <register type="NOA" address="0x00000D04" value="0x00000200" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x17066860" />
+        <register type="NOA" address="0x00009888" value="0x15070017" />
+        <register type="NOA" address="0x00009888" value="0x1D073C00" />
+        <register type="NOA" address="0x00009888" value="0x1F07000E" />
+        <register type="NOA" address="0x00009888" value="0x1B038000" />
+        <register type="NOA" address="0x00009888" value="0x0D038000" />
+        <register type="NOA" address="0x00009888" value="0x0F034000" />
+        <register type="NOA" address="0x00009888" value="0x0F063180" />
+        <register type="NOA" address="0x00009888" value="0x01063100" />
+        <register type="NOA" address="0x00009888" value="0x0306006B" />
+        <register type="NOA" address="0x00009888" value="0x05060000" />
+        <register type="NOA" address="0x00009888" value="0x11080100" />
+        <register type="NOA" address="0x00009888" value="0x0D085000" />
+        <register type="NOA" address="0x00009888" value="0x0F094000" />
+        <register type="NOA" address="0x00009888" value="0x01094000" />
+        <register type="NOA" address="0x00009888" value="0x03091000" />
+        <register type="NOA" address="0x00009888" value="0x09046000" />
+        <register type="NOA" address="0x00009888" value="0x0F04A000" />
+        <register type="NOA" address="0x00009888" value="0x1104A800" />
+        <register type="NOA" address="0x00009888" value="0x130401AA" />
+        <register type="NOA" address="0x00009888" value="0x0B041000" />
+        <register type="NOA" address="0x00009888" value="0x010700D0" />
+        <register type="NOA" address="0x00009888" value="0x0707D153" />
+        <register type="NOA" address="0x00009888" value="0x0907D3D2" />
+        <register type="NOA" address="0x00009888" value="0x0B07D5D4" />
+        <register type="NOA" address="0x00009888" value="0x0D07D7D6" />
+        <register type="NOA" address="0x00009888" value="0x0F0700DF" />
+        <register type="NOA" address="0x00009888" value="0x25070000" />
+        <register type="NOA" address="0x00009888" value="0x11070000" />
+        <register type="NOA" address="0x00009888" value="0x23070000" />
+        <register type="NOA" address="0x00009888" value="0x11004000" />
+        <register type="NOA" address="0x00009888" value="0x1F004000" />
+        <register type="NOA" address="0x00009888" value="0x21004000" />
+        <register type="NOA" address="0x00009888" value="0x23004000" />
+        <register type="NOA" address="0x00009888" value="0x25004000" />
+        <register type="NOA" address="0x00009888" value="0x27004000" />
+        <register type="NOA" address="0x00009888" value="0x29004000" />
+        <register type="NOA" address="0x00009888" value="0x2B004000" />
+        <register type="NOA" address="0x00009888" value="0x2D004000" />
+        <register type="NOA" address="0x00009888" value="0x2F004000" />
+        <register type="NOA" address="0x00009888" value="0x13004000" />
+        <register type="NOA" address="0x00009888" value="0x15004000" />
+        <register type="NOA" address="0x00009888" value="0x1D004000" />
+        <register type="NOA" address="0x00009888" value="0x47000000" />
+        <register type="NOA" address="0x00009888" value="0x39000003" />
+        <register type="NOA" address="0x00009888" value="0x49000000" />
+        <register type="NOA" address="0x00009888" value="0x3D000C63" />
+        <register type="NOA" address="0x00009888" value="0x4B000000" />
+        <register type="NOA" address="0x00009888" value="0x3F000C63" />
+        <register type="NOA" address="0x00009888" value="0x4D000000" />
+        <register type="NOA" address="0x00009888" value="0x41000C63" />
+        <register type="NOA" address="0x00009888" value="0x35000000" />
+        <register type="NOA" address="0x00009888" value="0x31000000" />
+        <register type="NOA" address="0x00009888" value="0x43000000" />
+    </register_config>
+    <register_config type="OA">
+        <register type="OA" address="0x00002740" value="0x00000000" />
+        <register type="OA" address="0x00002710" value="0x00000000" />
+        <register type="OA" address="0x00002714" value="0xF0800000" />
+        <register type="OA" address="0x00002720" value="0x00000000" />
+        <register type="OA" address="0x00002724" value="0x70800000" />
+        <register type="OA" address="0x00002770" value="0x0000081A" />
+        <register type="OA" address="0x00002774" value="0x0000FE00" />
+        <register type="OA" address="0x00002778" value="0x00000822" />
+        <register type="OA" address="0x0000277C" value="0x0000FE00" />
+        <register type="OA" address="0x00002780" value="0x000008BA" />
+        <register type="OA" address="0x00002784" value="0x0000FE00" />
+        <register type="OA" address="0x00002788" value="0x0000087A" />
+        <register type="OA" address="0x0000278C" value="0x0000FE00" />
+        <register type="OA" address="0x00002790" value="0x000008EA" />
+        <register type="OA" address="0x00002794" value="0x0000FE00" />
+        <register type="OA" address="0x00002798" value="0x000008E2" />
+        <register type="OA" address="0x0000279C" value="0x0000FE00" />
+        <register type="OA" address="0x000027A0" value="0x000008F2" />
+        <register type="OA" address="0x000027A4" value="0x0000FE00" />
+        <register type="OA" address="0x0000274C" value="0x65432108" />
+        <register type="OA" address="0x00002744" value="0x0000CCCE" />
+        <register type="OA" address="0x00002748" value="0x65432108" />
+        <register type="OA" address="0x0000275C" value="0x65432108" />
+        <register type="OA" address="0x00002754" value="0x000000CA" />
+    </register_config>
+    <register_config type="FLEX">
+        <register type="FLEX" address="0x0000E458" value="0x00005004" />
+    </register_config>
+  </set>
+
+  <set name="Compute Metrics Extended Gen10"
+       mdapi_supported_apis="OGL OCL IO BB"
+       underscore_name="compute_extended"
+       hw_config_guid="1b4d1280-6bbd-4d57-90c6-a03ed6059688"
+       chipset="CNL"
+       symbol_name="ComputeExtended"
+       >
+    <counter name="GPU Core Clocks"
+             description="The total number of GPU core clocks elapsed during the measurement."
+             data_type="uint64"
+             equation="GPU_CLOCK 0 READ"
+             underscore_name="gpu_core_clocks"
+             units="cycles"
+             symbol_name="GpuCoreClocks"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="EU Active"
+             description="The percentage of time in which the Execution Units were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 7 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_active"
+             units="percent"
+             symbol_name="EuActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EU Both FPU Pipes Active"
+             description="The percentage of time in which both EU FPU pipelines were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 9 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_fpu_both_active"
+             units="percent"
+             symbol_name="EuFpuBothActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EU AVG IPC Rate"
+             description="The average rate of IPC calculated for 2 FPU pipelines."
+             data_type="float"
+             max_equation="2"
+             equation="A 9 READ A 10 READ A 11 READ FADD A 9 READ FSUB FDIV 1 FADD"
+             underscore_name="eu_avg_ipc_rate"
+             units="number"
+             symbol_name="EuAvgIpcRate"
+             semantic_type="ratio"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Typed Writes 0"
+             description="The subslice 0 typed writes."
+             data_type="uint64"
+             equation="C 4 READ"
+             underscore_name="typed_writes0"
+             units="messages"
+             symbol_name="TypedWrites0"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="subslice"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="EuTypedAtomics0"
+             description="The subslice 0 EU Typed Atomics subslice 0."
+             data_type="uint64"
+             equation="B 7 READ"
+             underscore_name="eu_typed_atomics0"
+             units="messages"
+             symbol_name="EuTypedAtomics0"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="subslice"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="Typed Atomics 0"
+             description="The subslice 0 typed atomics."
+             data_type="uint64"
+             equation="C 5 READ"
+             underscore_name="typed_atomics0"
+             units="messages"
+             symbol_name="TypedAtomics0"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="subslice"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="TypedAtomicsPerCacheLine"
+             description="The ratio of EU typed atomics requests to L3 cache line writes."
+             data_type="float"
+             equation="$EuTypedAtomics0 $TypedAtomics0 FDIV"
+             underscore_name="typed_atomics_per_cache_line"
+             units="eu sends to l3 cache lines"
+             symbol_name="TypedAtomicsPerCacheLine"
+             semantic_type="ratio"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="subslice"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="EuUntypedReads0"
+             description="The subslice 0 EU Untyped Reads subslice 0."
+             data_type="uint64"
+             equation="B 0 READ"
+             underscore_name="eu_untyped_reads0"
+             units="messages"
+             symbol_name="EuUntypedReads0"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="subslice"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="Untyped Writes 0"
+             description="The subslice 0 untyped writes (including SLM writes)."
+             data_type="uint64"
+             equation="C 2 READ"
+             underscore_name="untyped_writes0"
+             units="messages"
+             symbol_name="UntypedWrites0"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="subslice"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="EU FPU1 Pipe Active"
+             description="The percentage of time in which EU FPU1 pipeline was actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 11 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="fpu1_active"
+             units="percent"
+             symbol_name="Fpu1Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EU FPU0 Pipe Active"
+             description="The percentage of time in which EU FPU0 pipeline was actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 10 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="fpu0_active"
+             units="percent"
+             symbol_name="Fpu0Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EuUntypedAtomics0"
+             description="The subslice 0 EU Untyped Atomics subslice 0."
+             data_type="uint64"
+             equation="B 2 READ"
+             underscore_name="eu_untyped_atomics0"
+             units="messages"
+             symbol_name="EuUntypedAtomics0"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="subslice"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="EuUntypedWrites0"
+             description="The subslice 0 EU Untyped Writes subslice 0."
+             data_type="uint64"
+             equation="B 1 READ"
+             underscore_name="eu_untyped_writes0"
+             units="messages"
+             symbol_name="EuUntypedWrites0"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="subslice"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="EuA64UntypedWrites0"
+             description="The subslice 0 EU A64 Untyped Writes subslice 0."
+             data_type="uint64"
+             equation="B 4 READ"
+             underscore_name="eu_a64_untyped_writes0"
+             units="messages"
+             symbol_name="EuA64UntypedWrites0"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="subslice"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="UntypedWritesPerCacheLine"
+             description="The ratio of EU untyped write requests to L3 cache line writes."
+             data_type="float"
+             equation="$EuUntypedWrites0 $EuA64UntypedWrites0 FADD $UntypedWrites0 FDIV"
+             underscore_name="untyped_writes_per_cache_line"
+             units="eu sends to l3 cache lines"
+             symbol_name="UntypedWritesPerCacheLine"
+             semantic_type="ratio"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="subslice"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="Shader Barrier Messages"
+             description="The total number of shader barrier messages."
+             data_type="uint64"
+             equation="A 35 READ"
+             underscore_name="shader_barriers"
+             units="messages"
+             symbol_name="ShaderBarriers"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Barrier"
+             />
+    <counter name="Sampler Texels"
+             description="The total number of texels seen on input (with 2x2 accuracy) in all sampler units."
+             data_type="uint64"
+             equation="A 28 READ 4 UMUL"
+             underscore_name="sampler_texels"
+             units="texels"
+             symbol_name="SamplerTexels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Input"
+             />
+    <counter name="GPU Time Elapsed"
+             description="Time elapsed on the GPU during the measurement."
+             data_type="uint64"
+             equation="GPU_TIME 0 READ 1000000000 UMUL $GpuTimestampFrequency UDIV"
+             underscore_name="gpu_time"
+             units="ns"
+             symbol_name="GpuTime"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="AVG GPU Core Frequency"
+             description="Average GPU Core Frequency in the measurement."
+             data_type="uint64"
+             max_equation="$GpuMaxFrequency"
+             equation="$GpuCoreClocks 1000000000 UMUL $GpuTime UDIV"
+             underscore_name="avg_gpu_core_frequency"
+             units="hz"
+             symbol_name="AvgGpuCoreFrequency"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Sampler Texels Misses"
+             description="The total number of texels lookups (with 2x2 accuracy) that missed L1 sampler cache."
+             data_type="uint64"
+             equation="A 29 READ 4 UMUL"
+             underscore_name="sampler_texel_misses"
+             units="texels"
+             symbol_name="SamplerTexelMisses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Batch Frame Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Cache"
+             />
+    <counter name="CS Threads Dispatched"
+             description="The total number of compute shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 4 READ"
+             underscore_name="cs_threads"
+             units="threads"
+             symbol_name="CsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Compute Shader"
+             />
+    <counter name="SLM Bytes Read"
+             description="The total number of GPU memory bytes read from shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ 64 UMUL"
+             underscore_name="slm_bytes_read"
+             units="bytes"
+             symbol_name="SlmBytesRead"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EuTypedWrites0"
+             description="The subslice 0 EU Typed Writes subslice 0."
+             data_type="uint64"
+             equation="B 6 READ"
+             underscore_name="eu_typed_writes0"
+             units="messages"
+             symbol_name="EuTypedWrites0"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="subslice"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="TypedWritesPerCacheLine"
+             description="The ratio of EU typed write requests to L3 cache line writes."
+             data_type="float"
+             equation="$EuTypedWrites0 $TypedWrites0 FDIV"
+             underscore_name="typed_writes_per_cache_line"
+             units="eu sends to l3 cache lines"
+             symbol_name="TypedWritesPerCacheLine"
+             semantic_type="ratio"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="subslice"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="Typed Reads 0"
+             description="The subslice 0 typed reads."
+             data_type="uint64"
+             equation="C 3 READ"
+             underscore_name="typed_reads0"
+             units="messages"
+             symbol_name="TypedReads0"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="subslice"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="Untyped Reads 0"
+             description="The subslice 0 untyped reads (including SLM reads)."
+             data_type="uint64"
+             equation="C 1 READ"
+             underscore_name="untyped_reads0"
+             units="messages"
+             symbol_name="UntypedReads0"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="subslice"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="EuA64UntypedReads0"
+             description="The subslice 0 EU A64 Untyped Reads subslice 0."
+             data_type="uint64"
+             equation="B 3 READ"
+             underscore_name="eu_a64_untyped_reads0"
+             units="messages"
+             symbol_name="EuA64UntypedReads0"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="subslice"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="EU Thread Occupancy"
+             description="The percentage of time in which hardware threads occupied EUs."
+             data_type="float"
+             max_equation="100"
+             equation="8 A 13 READ FMUL $EuThreadsCount FDIV $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_thread_occupancy"
+             units="percent"
+             symbol_name="EuThreadOccupancy"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EU Stall"
+             description="The percentage of time in which the Execution Units were stalled."
+             data_type="float"
+             max_equation="100"
+             equation="A 8 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_stall"
+             units="percent"
+             symbol_name="EuStall"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EuTypedReads0"
+             description="The subslice 0 EU Typed Reads subslice 0."
+             data_type="uint64"
+             equation="B 5 READ"
+             underscore_name="eu_typed_reads0"
+             units="messages"
+             symbol_name="EuTypedReads0"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="subslice"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="UntypedReadsPerCacheLine"
+             description="The ratio of EU untyped read requests to L3 cache line reads."
+             data_type="float"
+             equation="$EuUntypedReads0 $EuA64UntypedReads0 FADD $UntypedReads0 FDIV"
+             underscore_name="untyped_reads_per_cache_line"
+             units="eu sends to l3 cache lines"
+             symbol_name="UntypedReadsPerCacheLine"
+             semantic_type="ratio"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="subslice"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="Shader Memory Accesses"
+             description="The total number of shader memory accesses to L3."
+             data_type="uint64"
+             equation="A 32 READ"
+             underscore_name="shader_memory_accesses"
+             units="messages"
+             symbol_name="ShaderMemoryAccesses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="TypedReadsPerCacheLine"
+             description="The ratio of EU typed read requests to L3 cache line reads."
+             data_type="float"
+             equation="$EuTypedReads0 $TypedReads0 FDIV"
+             underscore_name="typed_reads_per_cache_line"
+             units="eu sends to l3 cache lines"
+             symbol_name="TypedReadsPerCacheLine"
+             semantic_type="ratio"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="subslice"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="L3 Shader Throughput"
+             description="The total number of GPU memory bytes transferred between shaders and L3 caches w/o URB."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ A 31 READ UADD A 32 READ UADD 64 UMUL"
+             underscore_name="l3_shader_throughput"
+             units="bytes"
+             symbol_name="L3ShaderThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_group="L3/Data Port"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="SLM Bytes Written"
+             description="The total number of GPU memory bytes written into shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 31 READ 64 UMUL"
+             underscore_name="slm_bytes_written"
+             units="bytes"
+             symbol_name="SlmBytesWritten"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Shader Atomic Memory Accesses"
+             description="The total number of shader atomic memory accesses."
+             data_type="uint64"
+             equation="A 34 READ"
+             underscore_name="shader_atomics"
+             units="messages"
+             symbol_name="ShaderAtomics"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port/Atomics"
+             />
+    <counter name="EU Send Pipe Active"
+             description="The percentage of time in which EU send pipeline was actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 12 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_send_active"
+             units="percent"
+             symbol_name="EuSendActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes"
+             mdapi_hw_unit_type="gpu"
+             />
+    <register_config type="NOA"
+                     availability="$SubsliceMask 1 AND"
+                     priority="0"
+                     >
+        <register type="NOA" address="0x00000D04" value="0x00000200" />
+        <register type="NOA" address="0x00009884" value="0x00000000" />
+        <register type="NOA" address="0x00009888" value="0x121A0007" />
+        <register type="NOA" address="0x00009888" value="0x101A0000" />
+        <register type="NOA" address="0x00009888" value="0x121B000B" />
+        <register type="NOA" address="0x00009888" value="0x161B0009" />
+        <register type="NOA" address="0x00009888" value="0x0474A000" />
+        <register type="NOA" address="0x00009888" value="0x0A748000" />
+        <register type="NOA" address="0x00009888" value="0x0C74A000" />
+        <register type="NOA" address="0x00009888" value="0x0E74A000" />
+        <register type="NOA" address="0x00009888" value="0x12741540" />
+        <register type="NOA" address="0x00009888" value="0x0674A000" />
+        <register type="NOA" address="0x00009888" value="0x08742000" />
+        <register type="NOA" address="0x00009888" value="0x0475A000" />
+        <register type="NOA" address="0x00009888" value="0x0A758000" />
+        <register type="NOA" address="0x00009888" value="0x0C75A000" />
+        <register type="NOA" address="0x00009888" value="0x0E75A000" />
+        <register type="NOA" address="0x00009888" value="0x147500AA" />
+        <register type="NOA" address="0x00009888" value="0x0675A000" />
+        <register type="NOA" address="0x00009888" value="0x08752000" />
+        <register type="NOA" address="0x00009888" value="0x0C76A800" />
+        <register type="NOA" address="0x00009888" value="0x0E76AA0A" />
+        <register type="NOA" address="0x00009888" value="0x107602AA" />
+        <register type="NOA" address="0x00009888" value="0x0077C000" />
+        <register type="NOA" address="0x00009888" value="0x0E77C000" />
+        <register type="NOA" address="0x00009888" value="0x1477C000" />
+        <register type="NOA" address="0x00009888" value="0x1677FFF0" />
+        <register type="NOA" address="0x00009888" value="0x18770003" />
+        <register type="NOA" address="0x00009888" value="0x0277C000" />
+        <register type="NOA" address="0x00009888" value="0x0477C000" />
+        <register type="NOA" address="0x00009888" value="0x0677C000" />
+        <register type="NOA" address="0x00009888" value="0x0877C000" />
+        <register type="NOA" address="0x00009888" value="0x00198000" />
+        <register type="NOA" address="0x00009888" value="0x0E198000" />
+        <register type="NOA" address="0x00009888" value="0x1A195540" />
+        <register type="NOA" address="0x00009888" value="0x1C19003A" />
+        <register type="NOA" address="0x00009888" value="0x0219C000" />
+        <register type="NOA" address="0x00009888" value="0x0419C000" />
+        <register type="NOA" address="0x00009888" value="0x0619C000" />
+        <register type="NOA" address="0x00009888" value="0x0819C000" />
+        <register type="NOA" address="0x00009888" value="0x0E1A2A00" />
+        <register type="NOA" address="0x00009888" value="0x001A2980" />
+        <register type="NOA" address="0x00009888" value="0x021A2CDA" />
+        <register type="NOA" address="0x00009888" value="0x041A0058" />
+        <register type="NOA" address="0x00009888" value="0x001B0011" />
+        <register type="NOA" address="0x00009888" value="0x061B0900" />
+        <register type="NOA" address="0x00009888" value="0x081B0A13" />
+        <register type="NOA" address="0x00009888" value="0x0A1B0B15" />
+        <register type="NOA" address="0x00009888" value="0x0C1B2317" />
+        <register type="NOA" address="0x00009888" value="0x0E1B0043" />
+        <register type="NOA" address="0x00009888" value="0x101B0000" />
+        <register type="NOA" address="0x00009888" value="0x10178000" />
+        <register type="NOA" address="0x00009888" value="0x1C1701FE" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x1100C000" />
+        <register type="NOA" address="0x00009888" value="0x1F00C000" />
+        <register type="NOA" address="0x00009888" value="0x2100C000" />
+        <register type="NOA" address="0x00009888" value="0x2300C000" />
+        <register type="NOA" address="0x00009888" value="0x2500C000" />
+        <register type="NOA" address="0x00009888" value="0x2700C000" />
+        <register type="NOA" address="0x00009888" value="0x2900C000" />
+        <register type="NOA" address="0x00009888" value="0x2B00C000" />
+        <register type="NOA" address="0x00009888" value="0x2D00C000" />
+        <register type="NOA" address="0x00009888" value="0x2F00C000" />
+        <register type="NOA" address="0x00009888" value="0x1300C000" />
+        <register type="NOA" address="0x00009888" value="0x1500C000" />
+        <register type="NOA" address="0x00009888" value="0x1700C000" />
+        <register type="NOA" address="0x00009888" value="0x1900C000" />
+        <register type="NOA" address="0x00009888" value="0x47000000" />
+        <register type="NOA" address="0x00009888" value="0x39000420" />
+        <register type="NOA" address="0x00009888" value="0x49000000" />
+        <register type="NOA" address="0x00009888" value="0x3D000000" />
+        <register type="NOA" address="0x00009888" value="0x4B000000" />
+        <register type="NOA" address="0x00009888" value="0x3F000000" />
+        <register type="NOA" address="0x00009888" value="0x4D000000" />
+        <register type="NOA" address="0x00009888" value="0x41000000" />
+        <register type="NOA" address="0x00009888" value="0x35000000" />
+        <register type="NOA" address="0x00009888" value="0x31000000" />
+        <register type="NOA" address="0x00009888" value="0x43000001" />
+        <register type="NOA" address="0x00009888" value="0x3B000021" />
+    </register_config>
+    <register_config type="OA">
+        <register type="OA" address="0x00002740" value="0x00000000" />
+        <register type="OA" address="0x00002710" value="0x00000000" />
+        <register type="OA" address="0x00002714" value="0xF0800000" />
+        <register type="OA" address="0x00002720" value="0x00000000" />
+        <register type="OA" address="0x00002724" value="0xF0800000" />
+        <register type="OA" address="0x00002770" value="0x00000C2A" />
+        <register type="OA" address="0x00002774" value="0x0000FE00" />
+        <register type="OA" address="0x00002778" value="0x00000C6A" />
+        <register type="OA" address="0x0000277C" value="0x0000FE00" />
+        <register type="OA" address="0x00002780" value="0x00000C32" />
+        <register type="OA" address="0x00002784" value="0x0000FE00" />
+        <register type="OA" address="0x00002788" value="0x00000E6A" />
+        <register type="OA" address="0x0000278C" value="0x0000FE00" />
+        <register type="OA" address="0x00002790" value="0x00000E7A" />
+        <register type="OA" address="0x00002794" value="0x0000FE00" />
+        <register type="OA" address="0x00002798" value="0x00000C92" />
+        <register type="OA" address="0x0000279C" value="0x0000FE00" />
+        <register type="OA" address="0x000027A0" value="0x00000CA2" />
+        <register type="OA" address="0x000027A4" value="0x0000FE00" />
+        <register type="OA" address="0x000027A8" value="0x00000C9A" />
+        <register type="OA" address="0x000027AC" value="0x0000FE00" />
+    </register_config>
+    <register_config type="FLEX">
+        <register type="FLEX" address="0x0000E458" value="0x00005004" />
+        <register type="FLEX" address="0x0000E558" value="0x00000003" />
+        <register type="FLEX" address="0x0000E658" value="0x00002001" />
+        <register type="FLEX" address="0x0000E758" value="0x00000008" />
+    </register_config>
+  </set>
+
+  <set name="Compute Metrics L3 Cache Gen10"
+       mdapi_supported_apis="OGL OCL IO BB"
+       underscore_name="compute_l3_cache"
+       hw_config_guid="f3b31b5a-00b3-4fb4-81fd-6446e505534c"
+       chipset="CNL"
+       symbol_name="ComputeL3Cache"
+       >
+    <counter name="GPU Core Clocks"
+             description="The total number of GPU core clocks elapsed during the measurement."
+             data_type="uint64"
+             equation="GPU_CLOCK 0 READ"
+             underscore_name="gpu_core_clocks"
+             units="cycles"
+             symbol_name="GpuCoreClocks"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="EU Active"
+             description="The percentage of time in which the Execution Units were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 7 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_active"
+             units="percent"
+             symbol_name="EuActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="L3 Misses"
+             description="The total number of L3 misses."
+             data_type="uint64"
+             equation="C 2 READ C 3 READ UADD"
+             underscore_name="l3_misses"
+             units="messages"
+             symbol_name="L3Misses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/TAG"
+             />
+    <counter name="GTI L3 Throughput"
+             description="The total number of GPU memory bytes transferred between L3 caches and GTI."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL"
+             equation="$L3Misses 64 UMUL"
+             underscore_name="gti_l3_throughput"
+             units="bytes"
+             symbol_name="GtiL3Throughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_group="GTI/L3"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Slice0 L3 Bank0 Accesses"
+             description="The total number of accesses to Slice0 L3 Bank0."
+             data_type="uint64"
+             equation="B 0 READ 2 UMUL"
+             underscore_name="l3_bank00_accesses"
+             units="messages"
+             symbol_name="L3Bank00Accesses"
+             availability="$SliceMask 1 AND"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Frame Batch Draw"
+             mdapi_hw_unit_type="slice"
+             mdapi_group="L3"
+             />
+    <counter name="Slice0 L3 Bank1 Accesses"
+             description="The total number of accesses to Slice0 L3 Bank1."
+             data_type="uint64"
+             equation="B 1 READ 2 UMUL"
+             underscore_name="l3_bank01_accesses"
+             units="messages"
+             symbol_name="L3Bank01Accesses"
+             availability="$SliceMask 1 AND"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Frame Batch Draw"
+             mdapi_hw_unit_type="slice"
+             mdapi_group="L3"
+             />
+    <counter name="Slice0 L3 Bank2 Accesses"
+             description="The total number of accesses to Slice0 L3 Bank2."
+             data_type="uint64"
+             equation="B 2 READ 2 UMUL"
+             underscore_name="l3_bank02_accesses"
+             units="messages"
+             symbol_name="L3Bank02Accesses"
+             availability="$SliceMask 1 AND"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Frame Batch Draw"
+             mdapi_hw_unit_type="slice"
+             mdapi_group="L3"
+             />
+    <counter name="Slice1 L3 Bank0 Accesses"
+             description="The total number of accesses to Slice1 L3 Bank0."
+             data_type="uint64"
+             equation="B 3 READ 2 UMUL"
+             underscore_name="l3_bank10_accesses"
+             units="messages"
+             symbol_name="L3Bank10Accesses"
+             availability="$SliceMask 2 AND"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Frame Batch Draw"
+             mdapi_hw_unit_type="slice"
+             mdapi_group="L3"
+             />
+    <counter name="Slice1 L3 Bank1 Accesses"
+             description="The total number of accesses to Slice1 L3 Bank1."
+             data_type="uint64"
+             equation="B 4 READ 2 UMUL"
+             underscore_name="l3_bank11_accesses"
+             units="messages"
+             symbol_name="L3Bank11Accesses"
+             availability="$SliceMask 2 AND"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Frame Batch Draw"
+             mdapi_hw_unit_type="slice"
+             mdapi_group="L3"
+             />
+    <counter name="Slice1 L3 Bank2 Accesses"
+             description="The total number of accesses to Slice1 L3 Bank2."
+             data_type="uint64"
+             equation="B 5 READ 2 UMUL"
+             underscore_name="l3_bank12_accesses"
+             units="messages"
+             symbol_name="L3Bank12Accesses"
+             availability="$SliceMask 2 AND"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Frame Batch Draw"
+             mdapi_hw_unit_type="slice"
+             mdapi_group="L3"
+             />
+    <counter name="L3 Accesses"
+             description="The total number of L3 accesses from all entities."
+             data_type="uint64"
+             equation="$L3Bank00Accesses $L3Bank01Accesses UADD $L3Bank02Accesses UADD $L3Bank10Accesses UADD $L3Bank11Accesses UADD $L3Bank12Accesses UADD"
+             underscore_name="l3_accesses"
+             units="messages"
+             symbol_name="L3Accesses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3"
+             />
+    <counter name="EU Both FPU Pipes Active"
+             description="The percentage of time in which both EU FPU pipelines were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 9 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_fpu_both_active"
+             units="percent"
+             symbol_name="EuFpuBothActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="L3 Sampler Throughput"
+             description="The total number of GPU memory bytes transferred between samplers and L3 caches."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSlicesTotalCount UMUL"
+             equation="64  $SliceMask 1 AND C 4 READ UMUL $SliceMask 1 &gt;&gt; 1 AND C 5 READ UMUL UADD 8 UMUL UMUL"
+             underscore_name="l3_sampler_throughput"
+             units="bytes"
+             symbol_name="L3SamplerThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_group="L3/Sampler"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EU AVG IPC Rate"
+             description="The average rate of IPC calculated for 2 FPU pipelines."
+             data_type="float"
+             max_equation="2"
+             equation="A 9 READ A 10 READ A 11 READ FADD A 9 READ FSUB FDIV 1 FADD"
+             underscore_name="eu_avg_ipc_rate"
+             units="number"
+             symbol_name="EuAvgIpcRate"
+             semantic_type="ratio"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EU FPU0 Binary Instruction"
+             description="The percentage of time in which execution units were actively processing binary instructions on FPU0."
+             data_type="float"
+             max_equation="100"
+             equation="A 17 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_binary_fpu0_instruction"
+             units="percent"
+             symbol_name="EuBinaryFpu0Instruction"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes/Instructions"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GS Threads Dispatched"
+             description="The total number of geometry shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 5 READ"
+             underscore_name="gs_threads"
+             units="threads"
+             symbol_name="GsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Geometry Shader"
+             />
+    <counter name="Early Hi-Depth Test Fails"
+             description="The total number of pixels dropped on early hierarchical depth test."
+             data_type="uint64"
+             equation="A 22 READ 4 UMUL"
+             underscore_name="hi_depth_test_fails"
+             units="pixels"
+             symbol_name="HiDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Hi-Depth Test"
+             />
+    <counter name="VS Threads Dispatched"
+             description="The total number of vertex shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 1 READ"
+             underscore_name="vs_threads"
+             units="threads"
+             symbol_name="VsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Vertex Shader"
+             />
+    <counter name="FS Threads Dispatched"
+             description="The total number of fragment shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 6 READ"
+             underscore_name="ps_threads"
+             units="threads"
+             symbol_name="PsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Fragment Shader"
+             />
+    <counter name="EU FPU0 Hybrid Instruction"
+             description="The percentage of time in which execution units were actively processing hybrid instructions on FPU0."
+             data_type="float"
+             max_equation="100"
+             equation="A 13 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_hybrid_fpu0_instruction"
+             units="percent"
+             symbol_name="EuHybridFpu0Instruction"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Frame Batch Draw"
+             mdapi_group="EU Array/Pipes/Instructions"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EU FPU1 Pipe Active"
+             description="The percentage of time in which EU FPU1 pipeline was actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 11 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="fpu1_active"
+             units="percent"
+             symbol_name="Fpu1Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EU FPU0 Pipe Active"
+             description="The percentage of time in which EU FPU0 pipeline was actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 10 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="fpu0_active"
+             units="percent"
+             symbol_name="Fpu0Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Shader Barrier Messages"
+             description="The total number of shader barrier messages."
+             data_type="uint64"
+             equation="A 35 READ"
+             underscore_name="shader_barriers"
+             units="messages"
+             symbol_name="ShaderBarriers"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Barrier"
+             />
+    <counter name="EU FPU0 Move Instruction"
+             description="The percentage of time in which execution units were actively processing move instructions on FPU0."
+             data_type="float"
+             max_equation="100"
+             equation="A 19 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_move_fpu0_instruction"
+             units="percent"
+             symbol_name="EuMoveFpu0Instruction"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes/Instructions"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Sampler Accesses"
+             description="The total number of messages send to samplers."
+             data_type="uint64"
+             equation="A 28 READ"
+             underscore_name="sampler_accesses"
+             units="messages"
+             symbol_name="SamplerAccesses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Batch Frame Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler"
+             />
+    <counter name="Sampler Texels"
+             description="The total number of texels seen on input (with 2x2 accuracy) in all sampler units."
+             data_type="uint64"
+             equation="$SamplerAccesses 4 UMUL"
+             underscore_name="sampler_texels"
+             units="texels"
+             symbol_name="SamplerTexels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Input"
+             />
+    <counter name="Pixels Failing Tests"
+             description="The total number of pixels dropped on post-FS alpha, stencil, or depth tests."
+             data_type="uint64"
+             equation="A 25 READ 4 UMUL"
+             underscore_name="pixels_failing_post_ps_tests"
+             units="pixels"
+             symbol_name="PixelsFailingPostPsTests"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="EU FPU1 Hybrid Instruction"
+             description="The percentage of time in which execution units were actively processing hybrid instructions on FPU1."
+             data_type="float"
+             max_equation="100"
+             equation="A 14 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_hybrid_fpu1_instruction"
+             units="percent"
+             symbol_name="EuHybridFpu1Instruction"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Frame Batch Draw"
+             mdapi_group="EU Array/Pipes/Instructions"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GPU Time Elapsed"
+             description="Time elapsed on the GPU during the measurement."
+             data_type="uint64"
+             equation="GPU_TIME 0 READ 1000000000 UMUL $GpuTimestampFrequency UDIV"
+             underscore_name="gpu_time"
+             units="ns"
+             symbol_name="GpuTime"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="AVG GPU Core Frequency"
+             description="Average GPU Core Frequency in the measurement."
+             data_type="uint64"
+             max_equation="$GpuMaxFrequency"
+             equation="$GpuCoreClocks 1000000000 UMUL $GpuTime UDIV"
+             underscore_name="avg_gpu_core_frequency"
+             units="hz"
+             symbol_name="AvgGpuCoreFrequency"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Sampler Texels Misses"
+             description="The total number of texels lookups (with 2x2 accuracy) that missed L1 sampler cache."
+             data_type="uint64"
+             equation="A 29 READ 4 UMUL"
+             underscore_name="sampler_texel_misses"
+             units="texels"
+             symbol_name="SamplerTexelMisses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Batch Frame Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Cache"
+             />
+    <counter name="CS Threads Dispatched"
+             description="The total number of compute shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 4 READ"
+             underscore_name="cs_threads"
+             units="threads"
+             symbol_name="CsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Compute Shader"
+             />
+    <counter name="L3 Lookup Accesses w/o IC"
+             description="The total number of L3 cache lookup accesses w/o IC."
+             data_type="uint64"
+             equation="$SliceMask 1 AND C 4 READ UMUL $SliceMask 1 &gt;&gt; 1 AND C 5 READ UMUL UADD 8 UMUL A 32 READ UADD"
+             underscore_name="l3_lookups"
+             units="messages"
+             symbol_name="L3Lookups"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/TAG"
+             />
+    <counter name="SLM Bytes Read"
+             description="The total number of GPU memory bytes read from shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ 64 UMUL"
+             underscore_name="slm_bytes_read"
+             units="bytes"
+             symbol_name="SlmBytesRead"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GTI Read Throughput"
+             description="The total number of GPU memory bytes read from GTI."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL"
+             equation="64  C 6 READ UMUL"
+             underscore_name="gti_read_throughput"
+             units="bytes"
+             symbol_name="GtiReadThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_group="GTI"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GPU Busy"
+             description="The percentage of time in which the GPU has been processing GPU commands."
+             data_type="float"
+             max_equation="100"
+             equation="A 0 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="gpu_busy"
+             units="percent"
+             symbol_name="GpuBusy"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EU FPU0 Ternary Instruction"
+             description="The percentage of time in which execution units were actively processing ternary instructions on FPU0."
+             data_type="float"
+             max_equation="100"
+             equation="A 15 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_ternary_fpu0_instruction"
+             units="percent"
+             symbol_name="EuTernaryFpu0Instruction"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes/Instructions"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Shader Atomic Memory Accesses"
+             description="The total number of shader atomic memory accesses."
+             data_type="uint64"
+             equation="A 34 READ"
+             underscore_name="shader_atomics"
+             units="messages"
+             symbol_name="ShaderAtomics"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port/Atomics"
+             />
+    <counter name="Rasterized Pixels"
+             description="The total number of rasterized pixels."
+             data_type="uint64"
+             equation="A 21 READ 4 UMUL"
+             underscore_name="rasterized_pixels"
+             units="pixels"
+             symbol_name="RasterizedPixels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer"
+             />
+    <counter name="DS Threads Dispatched"
+             description="The total number of domain shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 3 READ"
+             underscore_name="ds_threads"
+             units="threads"
+             symbol_name="DsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Domain Shader"
+             />
+    <counter name="Samples Written"
+             description="The total number of samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 26 READ 4 UMUL"
+             underscore_name="samples_written"
+             units="pixels"
+             symbol_name="SamplesWritten"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="EU FPU1 Move Instruction"
+             description="The percentage of time in which execution units were actively processing move instructions on FPU1."
+             data_type="float"
+             max_equation="100"
+             equation="A 20 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_move_fpu1_instruction"
+             units="percent"
+             symbol_name="EuMoveFpu1Instruction"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes/Instructions"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EU Stall"
+             description="The percentage of time in which the Execution Units were stalled."
+             data_type="float"
+             max_equation="100"
+             equation="A 8 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_stall"
+             units="percent"
+             symbol_name="EuStall"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Samples Blended"
+             description="The total number of blended samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 27 READ 4 UMUL"
+             underscore_name="samples_blended"
+             units="pixels"
+             symbol_name="SamplesBlended"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="Early Depth Test Fails"
+             description="The total number of pixels dropped on early depth test."
+             data_type="uint64"
+             equation="A 23 READ 4 UMUL"
+             underscore_name="early_depth_test_fails"
+             units="pixels"
+             symbol_name="EarlyDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Early Depth Test"
+             />
+    <counter name="Shader Memory Accesses"
+             description="The total number of shader memory accesses to L3."
+             data_type="uint64"
+             equation="A 32 READ"
+             underscore_name="shader_memory_accesses"
+             units="messages"
+             symbol_name="ShaderMemoryAccesses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="HS Threads Dispatched"
+             description="The total number of hull shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 2 READ"
+             underscore_name="hs_threads"
+             units="threads"
+             symbol_name="HsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Hull Shader"
+             />
+    <counter name="L3 Total Throughput"
+             description="The total number of GPU memory bytes transferred via L3."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSlicesTotalCount UMUL"
+             equation="$L3Accesses 64 UMUL"
+             underscore_name="l3_total_throughput"
+             units="bytes"
+             symbol_name="L3TotalThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GTI Write Throughput"
+             description="The total number of GPU memory bytes written to GTI."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL"
+             equation="64  C 7 READ UMUL"
+             underscore_name="gti_write_throughput"
+             units="bytes"
+             symbol_name="GtiWriteThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_group="GTI"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="L3 Shader Throughput"
+             description="The total number of GPU memory bytes transferred between shaders and L3 caches w/o URB."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ A 31 READ UADD A 32 READ UADD 64 UMUL"
+             underscore_name="l3_shader_throughput"
+             units="bytes"
+             symbol_name="L3ShaderThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_group="L3/Data Port"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Samples Killed in FS"
+             description="The total number of samples or pixels dropped in fragment shaders."
+             data_type="uint64"
+             equation="A 24 READ 4 UMUL"
+             underscore_name="samples_killed_in_ps"
+             units="pixels"
+             symbol_name="SamplesKilledInPs"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Fragment Shader"
+             />
+    <counter name="SLM Bytes Written"
+             description="The total number of GPU memory bytes written into shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 31 READ 64 UMUL"
+             underscore_name="slm_bytes_written"
+             units="bytes"
+             symbol_name="SlmBytesWritten"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EU FPU1 Binary Instruction"
+             description="The percentage of time in which execution units were actively processing binary instructions on FPU1."
+             data_type="float"
+             max_equation="100"
+             equation="A 18 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_binary_fpu1_instruction"
+             units="percent"
+             symbol_name="EuBinaryFpu1Instruction"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes/Instructions"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EU FPU1 Ternary Instruction"
+             description="The percentage of time in which execution units were actively processing ternary instructions on FPU1."
+             data_type="float"
+             max_equation="100"
+             equation="A 16 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_ternary_fpu1_instruction"
+             units="percent"
+             symbol_name="EuTernaryFpu1Instruction"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes/Instructions"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EU Send Pipe Active"
+             description="The percentage of time in which EU send pipeline was actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 12 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_send_active"
+             units="percent"
+             symbol_name="EuSendActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes"
+             mdapi_hw_unit_type="gpu"
+             />
+    <register_config type="NOA">
+        <register type="NOA" address="0x00000D04" value="0x00000200" />
+        <register type="NOA" address="0x00009884" value="0x00000000" />
+        <register type="NOA" address="0x00009888" value="0x146C0080" />
+        <register type="NOA" address="0x00009888" value="0x14EC0080" />
+        <register type="NOA" address="0x00009888" value="0x006C0050" />
+        <register type="NOA" address="0x00009888" value="0x066C5100" />
+        <register type="NOA" address="0x00009888" value="0x086C0052" />
+        <register type="NOA" address="0x00009888" value="0x026C5C00" />
+        <register type="NOA" address="0x00009888" value="0x106C0000" />
+        <register type="NOA" address="0x00009888" value="0x186C0800" />
+        <register type="NOA" address="0x00009888" value="0x1A6C0540" />
+        <register type="NOA" address="0x00009888" value="0x00739000" />
+        <register type="NOA" address="0x00009888" value="0x02739000" />
+        <register type="NOA" address="0x00009888" value="0x0473F000" />
+        <register type="NOA" address="0x00009888" value="0x00778000" />
+        <register type="NOA" address="0x00009888" value="0x0E778000" />
+        <register type="NOA" address="0x00009888" value="0x14778000" />
+        <register type="NOA" address="0x00009888" value="0x167702A0" />
+        <register type="NOA" address="0x00009888" value="0x06778000" />
+        <register type="NOA" address="0x00009888" value="0x08778000" />
+        <register type="NOA" address="0x00009888" value="0x08EC5000" />
+        <register type="NOA" address="0x00009888" value="0x0AEC5251" />
+        <register type="NOA" address="0x00009888" value="0x04EC005C" />
+        <register type="NOA" address="0x00009888" value="0x10EC0000" />
+        <register type="NOA" address="0x00009888" value="0x1AEC0000" />
+        <register type="NOA" address="0x00009888" value="0x18EC0000" />
+        <register type="NOA" address="0x00009888" value="0x04F3E000" />
+        <register type="NOA" address="0x00009888" value="0x02F31000" />
+        <register type="NOA" address="0x00009888" value="0x16F702A0" />
+        <register type="NOA" address="0x00009888" value="0x08F78000" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x0B080360" />
+        <register type="NOA" address="0x00009888" value="0x1D073800" />
+        <register type="NOA" address="0x00009888" value="0x1F070140" />
+        <register type="NOA" address="0x00009888" value="0x01081060" />
+        <register type="NOA" address="0x00009888" value="0x0D080000" />
+        <register type="NOA" address="0x00009888" value="0x01094000" />
+        <register type="NOA" address="0x00009888" value="0x03091000" />
+        <register type="NOA" address="0x00009888" value="0x09044000" />
+        <register type="NOA" address="0x00009888" value="0x0B041000" />
+        <register type="NOA" address="0x00009888" value="0x0D048000" />
+        <register type="NOA" address="0x00009888" value="0x0F042000" />
+        <register type="NOA" address="0x00009888" value="0x0507E700" />
+        <register type="NOA" address="0x00009888" value="0x070700D7" />
+        <register type="NOA" address="0x00009888" value="0x25070000" />
+        <register type="NOA" address="0x00009888" value="0x1B070000" />
+        <register type="NOA" address="0x00009888" value="0x1100C000" />
+        <register type="NOA" address="0x00009888" value="0x1F00C000" />
+        <register type="NOA" address="0x00009888" value="0x2100C000" />
+        <register type="NOA" address="0x00009888" value="0x2300C000" />
+        <register type="NOA" address="0x00009888" value="0x2500C000" />
+        <register type="NOA" address="0x00009888" value="0x2700C000" />
+        <register type="NOA" address="0x00009888" value="0x13004000" />
+        <register type="NOA" address="0x00009888" value="0x15004000" />
+        <register type="NOA" address="0x00009888" value="0x1700C000" />
+        <register type="NOA" address="0x00009888" value="0x1900C000" />
+        <register type="NOA" address="0x00009888" value="0x1B004000" />
+        <register type="NOA" address="0x00009888" value="0x1D004000" />
+        <register type="NOA" address="0x00009888" value="0x47001111" />
+        <register type="NOA" address="0x00009888" value="0x39002101" />
+        <register type="NOA" address="0x00009888" value="0x49001110" />
+        <register type="NOA" address="0x00009888" value="0x3D000429" />
+        <register type="NOA" address="0x00009888" value="0x4B000001" />
+        <register type="NOA" address="0x00009888" value="0x3F000000" />
+        <register type="NOA" address="0x00009888" value="0x35000000" />
+        <register type="NOA" address="0x00009888" value="0x31000000" />
+        <register type="NOA" address="0x00009888" value="0x3B002401" />
+    </register_config>
+    <register_config type="OA">
+        <register type="OA" address="0x00002740" value="0x00000000" />
+        <register type="OA" address="0x00002710" value="0x00000000" />
+        <register type="OA" address="0x00002714" value="0x00800000" />
+        <register type="OA" address="0x00002720" value="0x00000000" />
+        <register type="OA" address="0x00002724" value="0x00800000" />
+    </register_config>
+    <register_config type="FLEX">
+        <register type="FLEX" address="0x0000E458" value="0x00005004" />
+        <register type="FLEX" address="0x0000E558" value="0x00000003" />
+        <register type="FLEX" address="0x0000E658" value="0x00002001" />
+        <register type="FLEX" address="0x0000E758" value="0x00101100" />
+        <register type="FLEX" address="0x0000E45C" value="0x00201200" />
+        <register type="FLEX" address="0x0000E55C" value="0x00301300" />
+        <register type="FLEX" address="0x0000E65C" value="0x00401400" />
+    </register_config>
+  </set>
+
+  <set name="Metric set HDCAndSF"
+       mdapi_supported_apis="OGL OCL IO BB"
+       underscore_name="hdc_and_sf"
+       hw_config_guid="890094c5-8615-4f69-a8b9-706c5bcbd713"
+       chipset="CNL"
+       symbol_name="HDCAndSF"
+       >
+    <counter name="GPU Core Clocks"
+             description="The total number of GPU core clocks elapsed during the measurement."
+             data_type="uint64"
+             equation="GPU_CLOCK 0 READ"
+             underscore_name="gpu_core_clocks"
+             units="cycles"
+             symbol_name="GpuCoreClocks"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="EU Active"
+             description="The percentage of time in which the Execution Units were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 7 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_active"
+             units="percent"
+             symbol_name="EuActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EU Both FPU Pipes Active"
+             description="The percentage of time in which both EU FPU pipelines were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 9 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_fpu_both_active"
+             units="percent"
+             symbol_name="EuFpuBothActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="SQ is full"
+             description="The percentage of time when SQ is filled above a threshold (usually 48 entries)"
+             data_type="float"
+             max_equation="100"
+             equation="C 7 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="gt_request_queue_full"
+             units="percent"
+             symbol_name="GTRequestQueueFull"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="GTI"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS Send Pipe Active"
+             description="The percentage of time in which EU send pipeline was actively processing a vertex shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 12 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="vs_send_active"
+             units="percent"
+             symbol_name="VsSendActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS FPU1 Pipe Active"
+             description="The percentage of time in which EU FPU1 pipeline was actively processing a vertex shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 11 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="vs_fpu1_active"
+             units="percent"
+             symbol_name="VsFpu1Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GS Threads Dispatched"
+             description="The total number of geometry shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 5 READ"
+             underscore_name="gs_threads"
+             units="threads"
+             symbol_name="GsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Geometry Shader"
+             />
+    <counter name="Early Hi-Depth Test Fails"
+             description="The total number of pixels dropped on early hierarchical depth test."
+             data_type="uint64"
+             equation="A 22 READ 4 UMUL"
+             underscore_name="hi_depth_test_fails"
+             units="pixels"
+             symbol_name="HiDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Hi-Depth Test"
+             />
+    <counter name="FS Both FPU Active"
+             description="The percentage of time in which fragment shaders were processed actively on the both FPUs."
+             data_type="float"
+             max_equation="100"
+             equation="A 18 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_eu_both_fpu_active"
+             units="percent"
+             symbol_name="PsEuBothFpuActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Frame Batch Draw"
+             mdapi_group="3D Pipe/Fragment Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS Threads Dispatched"
+             description="The total number of vertex shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 1 READ"
+             underscore_name="vs_threads"
+             units="threads"
+             symbol_name="VsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Vertex Shader"
+             />
+    <counter name="Polygon Data Ready"
+             description="The percentage of time in which geometry pipeline output is ready"
+             data_type="float"
+             max_equation="100"
+             equation="B 0 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="poly_data_ready"
+             units="percent"
+             symbol_name="PolyDataReady"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/3D Pipe/Strip-Fans"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="FS Threads Dispatched"
+             description="The total number of fragment shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 6 READ"
+             underscore_name="ps_threads"
+             units="threads"
+             symbol_name="PsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Fragment Shader"
+             />
+    <counter name="Shader Barrier Messages"
+             description="The total number of shader barrier messages."
+             data_type="uint64"
+             equation="A 35 READ"
+             underscore_name="shader_barriers"
+             units="messages"
+             symbol_name="ShaderBarriers"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Barrier"
+             />
+    <counter name="Sampler Texels"
+             description="The total number of texels seen on input (with 2x2 accuracy) in all sampler units."
+             data_type="uint64"
+             equation="A 28 READ 4 UMUL"
+             underscore_name="sampler_texels"
+             units="texels"
+             symbol_name="SamplerTexels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Input"
+             />
+    <counter name="Pixels Failing Tests"
+             description="The total number of pixels dropped on post-FS alpha, stencil, or depth tests."
+             data_type="uint64"
+             equation="A 25 READ 4 UMUL"
+             underscore_name="pixels_failing_post_ps_tests"
+             units="pixels"
+             symbol_name="PixelsFailingPostPsTests"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="GPU Time Elapsed"
+             description="Time elapsed on the GPU during the measurement."
+             data_type="uint64"
+             equation="GPU_TIME 0 READ 1000000000 UMUL $GpuTimestampFrequency UDIV"
+             underscore_name="gpu_time"
+             units="ns"
+             symbol_name="GpuTime"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="AVG GPU Core Frequency"
+             description="Average GPU Core Frequency in the measurement."
+             data_type="uint64"
+             max_equation="$GpuMaxFrequency"
+             equation="$GpuCoreClocks 1000000000 UMUL $GpuTime UDIV"
+             underscore_name="avg_gpu_core_frequency"
+             units="hz"
+             symbol_name="AvgGpuCoreFrequency"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Sampler Texels Misses"
+             description="The total number of texels lookups (with 2x2 accuracy) that missed L1 sampler cache."
+             data_type="uint64"
+             equation="A 29 READ 4 UMUL"
+             underscore_name="sampler_texel_misses"
+             units="texels"
+             symbol_name="SamplerTexelMisses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Batch Frame Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Cache"
+             />
+    <counter name="Slice0 Subslice1 Non-sampler Shader Access Stalled On L3"
+             description="Percentage of time when HDC has messages to L3, but it's stalled due to lack of credits (Slice0 Subslice1)"
+             data_type="float"
+             max_equation="100"
+             equation="B 4 READ B 5 READ FSUB 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="non_sampler_shader01_access_stalled_on_l3"
+             units="percent"
+             symbol_name="NonSamplerShader01AccessStalledOnL3"
+             availability="$SubsliceMask 2 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="GPU/Data Port"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="CS Threads Dispatched"
+             description="The total number of compute shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 4 READ"
+             underscore_name="cs_threads"
+             units="threads"
+             symbol_name="CsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Compute Shader"
+             />
+    <counter name="Slice0 Subslice2 Non-sampler Shader Access Stalled On L3"
+             description="Percentage of time when HDC has messages to L3, but it's stalled due to lack of credits (Slice0 Subslice2)"
+             data_type="float"
+             max_equation="100"
+             equation="B 6 READ B 7 READ FSUB 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="non_sampler_shader02_access_stalled_on_l3"
+             units="percent"
+             symbol_name="NonSamplerShader02AccessStalledOnL3"
+             availability="$SubsliceMask 4 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="GPU/Data Port"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="SLM Bytes Read"
+             description="The total number of GPU memory bytes read from shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ 64 UMUL"
+             underscore_name="slm_bytes_read"
+             units="bytes"
+             symbol_name="SlmBytesRead"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="PS FPU1 Pipe Active"
+             description="The percentage of time in which EU FPU1 pipeline was actively processing a pixel shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 16 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_fpu1_active"
+             units="percent"
+             symbol_name="PsFpu1Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pixel Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="PS Send Pipeline Active"
+             description="The percentage of time in which EU send pipeline was actively processing a pixel shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 17 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_send_active"
+             units="percent"
+             symbol_name="PsSendActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pixel Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS FPU0 Pipe Active"
+             description="The percentage of time in which EU FPU0 pipeline was actively processing a vertex shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 10 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="vs_fpu0_active"
+             units="percent"
+             symbol_name="VsFpu0Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GPU Busy"
+             description="The percentage of time in which the GPU has been processing GPU commands."
+             data_type="float"
+             max_equation="100"
+             equation="A 0 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="gpu_busy"
+             units="percent"
+             symbol_name="GpuBusy"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Slice1 Subslice0 Non-sampler Shader Access Stalled On L3"
+             description="Percentage of time when HDC has messages to L3, but it's stalled due to lack of credits (Slice1 Subslice0)"
+             data_type="float"
+             max_equation="100"
+             equation="C 0 READ C 1 READ FSUB 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="non_sampler_shader10_access_stalled_on_l3"
+             units="percent"
+             symbol_name="NonSamplerShader10AccessStalledOnL3"
+             availability="$SubsliceMask 8 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="GPU/Data Port"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="Rasterized Pixels"
+             description="The total number of rasterized pixels."
+             data_type="uint64"
+             equation="A 21 READ 4 UMUL"
+             underscore_name="rasterized_pixels"
+             units="pixels"
+             symbol_name="RasterizedPixels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer"
+             />
+    <counter name="PS FPU0 Pipe Active"
+             description="The percentage of time in which EU FPU0 pipeline was actively processing a pixel shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 15 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_fpu0_active"
+             units="percent"
+             symbol_name="PsFpu0Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pixel Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="DS Threads Dispatched"
+             description="The total number of domain shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 3 READ"
+             underscore_name="ds_threads"
+             units="threads"
+             symbol_name="DsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Domain Shader"
+             />
+    <counter name="Samples Written"
+             description="The total number of samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 26 READ 4 UMUL"
+             underscore_name="samples_written"
+             units="pixels"
+             symbol_name="SamplesWritten"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="EU Stall"
+             description="The percentage of time in which the Execution Units were stalled."
+             data_type="float"
+             max_equation="100"
+             equation="A 8 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_stall"
+             units="percent"
+             symbol_name="EuStall"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Samples Blended"
+             description="The total number of blended samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 27 READ 4 UMUL"
+             underscore_name="samples_blended"
+             units="pixels"
+             symbol_name="SamplesBlended"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="Early Depth Test Fails"
+             description="The total number of pixels dropped on early depth test."
+             data_type="uint64"
+             equation="A 23 READ 4 UMUL"
+             underscore_name="early_depth_test_fails"
+             units="pixels"
+             symbol_name="EarlyDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Early Depth Test"
+             />
+    <counter name="Shader Memory Accesses"
+             description="The total number of shader memory accesses to L3."
+             data_type="uint64"
+             equation="A 32 READ"
+             underscore_name="shader_memory_accesses"
+             units="messages"
+             symbol_name="ShaderMemoryAccesses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="HS Threads Dispatched"
+             description="The total number of hull shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 2 READ"
+             underscore_name="hs_threads"
+             units="threads"
+             symbol_name="HsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Hull Shader"
+             />
+    <counter name="Slice1 Subslice1 Non-sampler Shader Access Stalled On L3"
+             description="Percentage of time when HDC has messages to L3, but it's stalled due to lack of credits (Slice1 Subslice1)"
+             data_type="float"
+             max_equation="100"
+             equation="C 2 READ C 3 READ FSUB 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="non_sampler_shader11_access_stalled_on_l3"
+             units="percent"
+             symbol_name="NonSamplerShader11AccessStalledOnL3"
+             availability="$SubsliceMask 16 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="GPU/Data Port"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="L3 Shader Throughput"
+             description="The total number of GPU memory bytes transferred between shaders and L3 caches w/o URB."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ A 31 READ UADD A 32 READ UADD 64 UMUL"
+             underscore_name="l3_shader_throughput"
+             units="bytes"
+             symbol_name="L3ShaderThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_group="L3/Data Port"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Samples Killed in FS"
+             description="The total number of samples or pixels dropped in fragment shaders."
+             data_type="uint64"
+             equation="A 24 READ 4 UMUL"
+             underscore_name="samples_killed_in_ps"
+             units="pixels"
+             symbol_name="SamplesKilledInPs"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Fragment Shader"
+             />
+    <counter name="Slice0 Subslice0 Non-sampler Shader Access Stalled On L3"
+             description="Percentage of time when HDC has messages to L3, but it's stalled due to lack of credits (Slice0 Subslice0)"
+             data_type="float"
+             max_equation="100"
+             equation="B 2 READ B 3 READ FSUB 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="non_sampler_shader00_access_stalled_on_l3"
+             units="percent"
+             symbol_name="NonSamplerShader00AccessStalledOnL3"
+             availability="$SubsliceMask 1 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="GPU/Data Port"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="SLM Bytes Written"
+             description="The total number of GPU memory bytes written into shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 31 READ 64 UMUL"
+             underscore_name="slm_bytes_written"
+             units="bytes"
+             symbol_name="SlmBytesWritten"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Shader Atomic Memory Accesses"
+             description="The total number of shader atomic memory accesses."
+             data_type="uint64"
+             equation="A 34 READ"
+             underscore_name="shader_atomics"
+             units="messages"
+             symbol_name="ShaderAtomics"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port/Atomics"
+             />
+    <register_config type="NOA">
+        <register type="NOA" address="0x00000D04" value="0x00000200" />
+        <register type="NOA" address="0x00009884" value="0x00000000" />
+        <register type="NOA" address="0x00009888" value="0x121A0011" />
+        <register type="NOA" address="0x00009888" value="0x123A0011" />
+        <register type="NOA" address="0x00009888" value="0x129A0011" />
+        <register type="NOA" address="0x00009888" value="0x12BA0011" />
+        <register type="NOA" address="0x00009888" value="0x125A0011" />
+        <register type="NOA" address="0x00009888" value="0x1C6C0005" />
+        <register type="NOA" address="0x00009888" value="0x186C00A0" />
+        <register type="NOA" address="0x00009888" value="0x0673C000" />
+        <register type="NOA" address="0x00009888" value="0x00736000" />
+        <register type="NOA" address="0x00009888" value="0x0C74A000" />
+        <register type="NOA" address="0x00009888" value="0x0C75A000" />
+        <register type="NOA" address="0x00009888" value="0x0E762800" />
+        <register type="NOA" address="0x00009888" value="0x1477C000" />
+        <register type="NOA" address="0x00009888" value="0x16779570" />
+        <register type="NOA" address="0x00009888" value="0x18770002" />
+        <register type="NOA" address="0x00009888" value="0x02778000" />
+        <register type="NOA" address="0x00009888" value="0x04778000" />
+        <register type="NOA" address="0x00009888" value="0x0A7B8000" />
+        <register type="NOA" address="0x00009888" value="0x0C7B002A" />
+        <register type="NOA" address="0x00009888" value="0x147C8000" />
+        <register type="NOA" address="0x00009888" value="0x167C003E" />
+        <register type="NOA" address="0x00009888" value="0x1E7D0028" />
+        <register type="NOA" address="0x00009888" value="0x1A1901E0" />
+        <register type="NOA" address="0x00009888" value="0x081A2B50" />
+        <register type="NOA" address="0x00009888" value="0x001A0000" />
+        <register type="NOA" address="0x00009888" value="0x1A391E00" />
+        <register type="NOA" address="0x00009888" value="0x0A3A2B50" />
+        <register type="NOA" address="0x00009888" value="0x003A0000" />
+        <register type="NOA" address="0x00009888" value="0x12F41400" />
+        <register type="NOA" address="0x00009888" value="0x14F500A0" />
+        <register type="NOA" address="0x00009888" value="0x10F60280" />
+        <register type="NOA" address="0x00009888" value="0x16F7C000" />
+        <register type="NOA" address="0x00009888" value="0x18F70003" />
+        <register type="NOA" address="0x00009888" value="0x02F74000" />
+        <register type="NOA" address="0x00009888" value="0x04F74000" />
+        <register type="NOA" address="0x00009888" value="0x08FBA000" />
+        <register type="NOA" address="0x00009888" value="0x02FC8000" />
+        <register type="NOA" address="0x00009888" value="0x04FC8000" />
+        <register type="NOA" address="0x00009888" value="0x1CFD000A" />
+        <register type="NOA" address="0x00009888" value="0x1C99003C" />
+        <register type="NOA" address="0x00009888" value="0x0E9A2B50" />
+        <register type="NOA" address="0x00009888" value="0x009A0000" />
+        <register type="NOA" address="0x00009888" value="0x02B9C000" />
+        <register type="NOA" address="0x00009888" value="0x04B9C000" />
+        <register type="NOA" address="0x00009888" value="0x00BA2800" />
+        <register type="NOA" address="0x00009888" value="0x02BA0056" />
+        <register type="NOA" address="0x00009888" value="0x1A596000" />
+        <register type="NOA" address="0x00009888" value="0x1C590003" />
+        <register type="NOA" address="0x00009888" value="0x0C5A2B50" />
+        <register type="NOA" address="0x00009888" value="0x005A0000" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x130E03C0" />
+        <register type="NOA" address="0x00009888" value="0x2100F000" />
+        <register type="NOA" address="0x00009888" value="0x01181000" />
+        <register type="NOA" address="0x00009888" value="0x010E0030" />
+        <register type="NOA" address="0x00009888" value="0x110E0000" />
+        <register type="NOA" address="0x00009888" value="0x01104000" />
+        <register type="NOA" address="0x00009888" value="0x01118000" />
+        <register type="NOA" address="0x00009888" value="0x0B142000" />
+        <register type="NOA" address="0x00009888" value="0x01092000" />
+        <register type="NOA" address="0x00009888" value="0x010B4000" />
+        <register type="NOA" address="0x00009888" value="0x09041000" />
+        <register type="NOA" address="0x00009888" value="0x1D000071" />
+        <register type="NOA" address="0x00009888" value="0x33000000" />
+        <register type="NOA" address="0x00009888" value="0x11004000" />
+        <register type="NOA" address="0x00009888" value="0x2300C000" />
+        <register type="NOA" address="0x00009888" value="0x2500C000" />
+        <register type="NOA" address="0x00009888" value="0x2700C000" />
+        <register type="NOA" address="0x00009888" value="0x2900C000" />
+        <register type="NOA" address="0x00009888" value="0x2B00C000" />
+        <register type="NOA" address="0x00009888" value="0x2D00C000" />
+        <register type="NOA" address="0x00009888" value="0x2F00C000" />
+        <register type="NOA" address="0x00009888" value="0x1300C000" />
+        <register type="NOA" address="0x00009888" value="0x1500C000" />
+        <register type="NOA" address="0x00009888" value="0x47000003" />
+        <register type="NOA" address="0x00009888" value="0x39000000" />
+        <register type="NOA" address="0x00009888" value="0x4B001111" />
+        <register type="NOA" address="0x00009888" value="0x3D000408" />
+        <register type="NOA" address="0x00009888" value="0x3F000421" />
+        <register type="NOA" address="0x00009888" value="0x4D000000" />
+        <register type="NOA" address="0x00009888" value="0x41000084" />
+        <register type="NOA" address="0x00009888" value="0x35000000" />
+        <register type="NOA" address="0x00009888" value="0x31000000" />
+        <register type="NOA" address="0x00009888" value="0x43000000" />
+        <register type="NOA" address="0x00009888" value="0x49000300" />
+    </register_config>
+    <register_config type="OA">
+        <register type="OA" address="0x00002740" value="0x00000000" />
+        <register type="OA" address="0x00002710" value="0x00000000" />
+        <register type="OA" address="0x00002714" value="0x10800000" />
+        <register type="OA" address="0x00002720" value="0x00000000" />
+        <register type="OA" address="0x00002724" value="0x00800000" />
+        <register type="OA" address="0x00002770" value="0x00000002" />
+        <register type="OA" address="0x00002774" value="0x0000FFFE" />
+    </register_config>
+    <register_config type="FLEX">
+        <register type="FLEX" address="0x0000E458" value="0x00005004" />
+        <register type="FLEX" address="0x0000E558" value="0x00010003" />
+        <register type="FLEX" address="0x0000E658" value="0x00012011" />
+        <register type="FLEX" address="0x0000E45C" value="0x00051050" />
+        <register type="FLEX" address="0x0000E55C" value="0x00053052" />
+    </register_config>
+  </set>
+
+  <set name="Metric set L3_1"
+       mdapi_supported_apis="OGL OCL IO BB"
+       underscore_name="l3_1"
+       hw_config_guid="e56f7dd7-ff55-4bae-89a1-a526a2e3b0d6"
+       chipset="CNL"
+       symbol_name="L3_1"
+       >
+    <counter name="GPU Core Clocks"
+             description="The total number of GPU core clocks elapsed during the measurement."
+             data_type="uint64"
+             equation="GPU_CLOCK 0 READ"
+             underscore_name="gpu_core_clocks"
+             units="cycles"
+             symbol_name="GpuCoreClocks"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="EU Active"
+             description="The percentage of time in which the Execution Units were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 7 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_active"
+             units="percent"
+             symbol_name="EuActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Slice0 L3 Bank1 Active"
+             description="The percentage of time in which slice0 L3 bank1 is active"
+             data_type="float"
+             max_equation="100"
+             equation="C 4 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="l30_bank1_active"
+             units="percent"
+             symbol_name="L30Bank1Active"
+             availability="$SliceMask 1 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GTI/L3"
+             mdapi_hw_unit_type="slice"
+             />
+    <counter name="EU Both FPU Pipes Active"
+             description="The percentage of time in which both EU FPU pipelines were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 9 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_fpu_both_active"
+             units="percent"
+             symbol_name="EuFpuBothActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="SQ is full"
+             description="The percentage of time when SQ is filled above a threshold (usually 48 entries)"
+             data_type="float"
+             max_equation="100"
+             equation="C 7 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="gt_request_queue_full"
+             units="percent"
+             symbol_name="GTRequestQueueFull"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="GTI"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS Send Pipe Active"
+             description="The percentage of time in which EU send pipeline was actively processing a vertex shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 12 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="vs_send_active"
+             units="percent"
+             symbol_name="VsSendActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS FPU1 Pipe Active"
+             description="The percentage of time in which EU FPU1 pipeline was actively processing a vertex shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 11 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="vs_fpu1_active"
+             units="percent"
+             symbol_name="VsFpu1Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GS Threads Dispatched"
+             description="The total number of geometry shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 5 READ"
+             underscore_name="gs_threads"
+             units="threads"
+             symbol_name="GsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Geometry Shader"
+             />
+    <counter name="Early Hi-Depth Test Fails"
+             description="The total number of pixels dropped on early hierarchical depth test."
+             data_type="uint64"
+             equation="A 22 READ 4 UMUL"
+             underscore_name="hi_depth_test_fails"
+             units="pixels"
+             symbol_name="HiDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Hi-Depth Test"
+             />
+    <counter name="FS Both FPU Active"
+             description="The percentage of time in which fragment shaders were processed actively on the both FPUs."
+             data_type="float"
+             max_equation="100"
+             equation="A 18 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_eu_both_fpu_active"
+             units="percent"
+             symbol_name="PsEuBothFpuActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Frame Batch Draw"
+             mdapi_group="3D Pipe/Fragment Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS Threads Dispatched"
+             description="The total number of vertex shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 1 READ"
+             underscore_name="vs_threads"
+             units="threads"
+             symbol_name="VsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Vertex Shader"
+             />
+    <counter name="Slice0 L3 Bank1 Stalled"
+             description="The percentage of time in which slice0 L3 bank1 is stalled"
+             data_type="float"
+             max_equation="100"
+             equation="B 1 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="l30_bank1_stalled"
+             units="percent"
+             symbol_name="L30Bank1Stalled"
+             availability="$SliceMask 1 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GTI/L3"
+             mdapi_hw_unit_type="slice"
+             />
+    <counter name="FS Threads Dispatched"
+             description="The total number of fragment shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 6 READ"
+             underscore_name="ps_threads"
+             units="threads"
+             symbol_name="PsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Fragment Shader"
+             />
+    <counter name="Shader Barrier Messages"
+             description="The total number of shader barrier messages."
+             data_type="uint64"
+             equation="A 35 READ"
+             underscore_name="shader_barriers"
+             units="messages"
+             symbol_name="ShaderBarriers"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Barrier"
+             />
+    <counter name="Sampler Texels"
+             description="The total number of texels seen on input (with 2x2 accuracy) in all sampler units."
+             data_type="uint64"
+             equation="A 28 READ 4 UMUL"
+             underscore_name="sampler_texels"
+             units="texels"
+             symbol_name="SamplerTexels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Input"
+             />
+    <counter name="Pixels Failing Tests"
+             description="The total number of pixels dropped on post-FS alpha, stencil, or depth tests."
+             data_type="uint64"
+             equation="A 25 READ 4 UMUL"
+             underscore_name="pixels_failing_post_ps_tests"
+             units="pixels"
+             symbol_name="PixelsFailingPostPsTests"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="GPU Time Elapsed"
+             description="Time elapsed on the GPU during the measurement."
+             data_type="uint64"
+             equation="GPU_TIME 0 READ 1000000000 UMUL $GpuTimestampFrequency UDIV"
+             underscore_name="gpu_time"
+             units="ns"
+             symbol_name="GpuTime"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="AVG GPU Core Frequency"
+             description="Average GPU Core Frequency in the measurement."
+             data_type="uint64"
+             max_equation="$GpuMaxFrequency"
+             equation="$GpuCoreClocks 1000000000 UMUL $GpuTime UDIV"
+             underscore_name="avg_gpu_core_frequency"
+             units="hz"
+             symbol_name="AvgGpuCoreFrequency"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Sampler Texels Misses"
+             description="The total number of texels lookups (with 2x2 accuracy) that missed L1 sampler cache."
+             data_type="uint64"
+             equation="A 29 READ 4 UMUL"
+             underscore_name="sampler_texel_misses"
+             units="texels"
+             symbol_name="SamplerTexelMisses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Batch Frame Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Cache"
+             />
+    <counter name="CS Threads Dispatched"
+             description="The total number of compute shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 4 READ"
+             underscore_name="cs_threads"
+             units="threads"
+             symbol_name="CsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Compute Shader"
+             />
+    <counter name="SLM Bytes Read"
+             description="The total number of GPU memory bytes read from shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ 64 UMUL"
+             underscore_name="slm_bytes_read"
+             units="bytes"
+             symbol_name="SlmBytesRead"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="PS FPU1 Pipe Active"
+             description="The percentage of time in which EU FPU1 pipeline was actively processing a pixel shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 16 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_fpu1_active"
+             units="percent"
+             symbol_name="PsFpu1Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pixel Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="PS Send Pipeline Active"
+             description="The percentage of time in which EU send pipeline was actively processing a pixel shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 17 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_send_active"
+             units="percent"
+             symbol_name="PsSendActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pixel Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS FPU0 Pipe Active"
+             description="The percentage of time in which EU FPU0 pipeline was actively processing a vertex shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 10 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="vs_fpu0_active"
+             units="percent"
+             symbol_name="VsFpu0Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GPU Busy"
+             description="The percentage of time in which the GPU has been processing GPU commands."
+             data_type="float"
+             max_equation="100"
+             equation="A 0 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="gpu_busy"
+             units="percent"
+             symbol_name="GpuBusy"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Slice0 L3 Bank2 Active"
+             description="The percentage of time in which slice0 L3 bank2 is active"
+             data_type="float"
+             max_equation="100"
+             equation="C 5 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="l30_bank2_active"
+             units="percent"
+             symbol_name="L30Bank2Active"
+             availability="$SliceMask 1 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GTI/L3"
+             mdapi_hw_unit_type="slice"
+             />
+    <counter name="Slice0 L3 Bank0 Active"
+             description="The percentage of time in which slice0 L3 bank0 is active"
+             data_type="float"
+             max_equation="100"
+             equation="C 3 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="l30_bank0_active"
+             units="percent"
+             symbol_name="L30Bank0Active"
+             availability="$SliceMask 1 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GTI/L3"
+             mdapi_hw_unit_type="slice"
+             />
+    <counter name="Rasterized Pixels"
+             description="The total number of rasterized pixels."
+             data_type="uint64"
+             equation="A 21 READ 4 UMUL"
+             underscore_name="rasterized_pixels"
+             units="pixels"
+             symbol_name="RasterizedPixels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer"
+             />
+    <counter name="Slice1 L3 Bank2 Stalled"
+             description="The percentage of time in which slice1 L3 bank2 is stalled"
+             data_type="float"
+             max_equation="100"
+             equation="B 2 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="l31_bank2_stalled"
+             units="percent"
+             symbol_name="L31Bank2Stalled"
+             availability="$SliceMask 2 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GTI/L3"
+             mdapi_hw_unit_type="slice"
+             />
+    <counter name="PS FPU0 Pipe Active"
+             description="The percentage of time in which EU FPU0 pipeline was actively processing a pixel shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 15 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_fpu0_active"
+             units="percent"
+             symbol_name="PsFpu0Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pixel Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Slice0 L3 Bank0 Stalled"
+             description="The percentage of time in which slice0 L3 bank0 is stalled"
+             data_type="float"
+             max_equation="100"
+             equation="B 0 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="l30_bank0_stalled"
+             units="percent"
+             symbol_name="L30Bank0Stalled"
+             availability="$SliceMask 1 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GTI/L3"
+             mdapi_hw_unit_type="slice"
+             />
+    <counter name="DS Threads Dispatched"
+             description="The total number of domain shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 3 READ"
+             underscore_name="ds_threads"
+             units="threads"
+             symbol_name="DsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Domain Shader"
+             />
+    <counter name="Samples Written"
+             description="The total number of samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 26 READ 4 UMUL"
+             underscore_name="samples_written"
+             units="pixels"
+             symbol_name="SamplesWritten"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="EU Stall"
+             description="The percentage of time in which the Execution Units were stalled."
+             data_type="float"
+             max_equation="100"
+             equation="A 8 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_stall"
+             units="percent"
+             symbol_name="EuStall"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Samples Blended"
+             description="The total number of blended samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 27 READ 4 UMUL"
+             underscore_name="samples_blended"
+             units="pixels"
+             symbol_name="SamplesBlended"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="Early Depth Test Fails"
+             description="The total number of pixels dropped on early depth test."
+             data_type="uint64"
+             equation="A 23 READ 4 UMUL"
+             underscore_name="early_depth_test_fails"
+             units="pixels"
+             symbol_name="EarlyDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Early Depth Test"
+             />
+    <counter name="Shader Memory Accesses"
+             description="The total number of shader memory accesses to L3."
+             data_type="uint64"
+             equation="A 32 READ"
+             underscore_name="shader_memory_accesses"
+             units="messages"
+             symbol_name="ShaderMemoryAccesses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="HS Threads Dispatched"
+             description="The total number of hull shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 2 READ"
+             underscore_name="hs_threads"
+             units="threads"
+             symbol_name="HsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Hull Shader"
+             />
+    <counter name="L3 Shader Throughput"
+             description="The total number of GPU memory bytes transferred between shaders and L3 caches w/o URB."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ A 31 READ UADD A 32 READ UADD 64 UMUL"
+             underscore_name="l3_shader_throughput"
+             units="bytes"
+             symbol_name="L3ShaderThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_group="L3/Data Port"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Samples Killed in FS"
+             description="The total number of samples or pixels dropped in fragment shaders."
+             data_type="uint64"
+             equation="A 24 READ 4 UMUL"
+             underscore_name="samples_killed_in_ps"
+             units="pixels"
+             symbol_name="SamplesKilledInPs"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Fragment Shader"
+             />
+    <counter name="SLM Bytes Written"
+             description="The total number of GPU memory bytes written into shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 31 READ 64 UMUL"
+             underscore_name="slm_bytes_written"
+             units="bytes"
+             symbol_name="SlmBytesWritten"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Shader Atomic Memory Accesses"
+             description="The total number of shader atomic memory accesses."
+             data_type="uint64"
+             equation="A 34 READ"
+             underscore_name="shader_atomics"
+             units="messages"
+             symbol_name="ShaderAtomics"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port/Atomics"
+             />
+    <register_config type="NOA">
+        <register type="NOA" address="0x00000D04" value="0x00000200" />
+        <register type="NOA" address="0x00009884" value="0x00000000" />
+        <register type="NOA" address="0x00009888" value="0x1661001A" />
+        <register type="NOA" address="0x00009888" value="0x1665001A" />
+        <register type="NOA" address="0x00009888" value="0x1669001A" />
+        <register type="NOA" address="0x00009888" value="0x106C1C00" />
+        <register type="NOA" address="0x00009888" value="0x126C0004" />
+        <register type="NOA" address="0x00009888" value="0x166C001D" />
+        <register type="NOA" address="0x00009888" value="0x10EC0400" />
+        <register type="NOA" address="0x00009888" value="0x16EC0000" />
+        <register type="NOA" address="0x00009888" value="0x02610024" />
+        <register type="NOA" address="0x00009888" value="0x10610000" />
+        <register type="NOA" address="0x00009888" value="0x1E610000" />
+        <register type="NOA" address="0x00009888" value="0x02652400" />
+        <register type="NOA" address="0x00009888" value="0x10650000" />
+        <register type="NOA" address="0x00009888" value="0x1E650000" />
+        <register type="NOA" address="0x00009888" value="0x04690024" />
+        <register type="NOA" address="0x00009888" value="0x10690000" />
+        <register type="NOA" address="0x00009888" value="0x1E690000" />
+        <register type="NOA" address="0x00009888" value="0x006C0020" />
+        <register type="NOA" address="0x00009888" value="0x066C2800" />
+        <register type="NOA" address="0x00009888" value="0x0A6C2820" />
+        <register type="NOA" address="0x00009888" value="0x0E6C3832" />
+        <register type="NOA" address="0x00009888" value="0x186C05A0" />
+        <register type="NOA" address="0x00009888" value="0x1A6C1010" />
+        <register type="NOA" address="0x00009888" value="0x1C6C0000" />
+        <register type="NOA" address="0x00009888" value="0x186D8000" />
+        <register type="NOA" address="0x00009888" value="0x0A6E1000" />
+        <register type="NOA" address="0x00009888" value="0x0073F000" />
+        <register type="NOA" address="0x00009888" value="0x02738000" />
+        <register type="NOA" address="0x00009888" value="0x0473D000" />
+        <register type="NOA" address="0x00009888" value="0x0673D000" />
+        <register type="NOA" address="0x00009888" value="0x00778000" />
+        <register type="NOA" address="0x00009888" value="0x0E778000" />
+        <register type="NOA" address="0x00009888" value="0x14778000" />
+        <register type="NOA" address="0x00009888" value="0x16778A80" />
+        <register type="NOA" address="0x00009888" value="0x18770002" />
+        <register type="NOA" address="0x00009888" value="0x02778000" />
+        <register type="NOA" address="0x00009888" value="0x04778000" />
+        <register type="NOA" address="0x00009888" value="0x06778000" />
+        <register type="NOA" address="0x00009888" value="0x08774000" />
+        <register type="NOA" address="0x00009888" value="0x04784000" />
+        <register type="NOA" address="0x00009888" value="0x0A7B0004" />
+        <register type="NOA" address="0x00009888" value="0x08EC0020" />
+        <register type="NOA" address="0x00009888" value="0x0CEC0020" />
+        <register type="NOA" address="0x00009888" value="0x00EC3600" />
+        <register type="NOA" address="0x00009888" value="0x1AEC0000" />
+        <register type="NOA" address="0x00009888" value="0x18EC0000" />
+        <register type="NOA" address="0x00009888" value="0x04F31000" />
+        <register type="NOA" address="0x00009888" value="0x06F31000" />
+        <register type="NOA" address="0x00009888" value="0x00F32000" />
+        <register type="NOA" address="0x00009888" value="0x14F78000" />
+        <register type="NOA" address="0x00009888" value="0x16F70800" />
+        <register type="NOA" address="0x00009888" value="0x02F78000" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x2100F000" />
+        <register type="NOA" address="0x00009888" value="0x1D000071" />
+        <register type="NOA" address="0x00009888" value="0x33000000" />
+        <register type="NOA" address="0x00009888" value="0x1100C000" />
+        <register type="NOA" address="0x00009888" value="0x1F00C000" />
+        <register type="NOA" address="0x00009888" value="0x2500C000" />
+        <register type="NOA" address="0x00009888" value="0x2700C000" />
+        <register type="NOA" address="0x00009888" value="0x2900C000" />
+        <register type="NOA" address="0x00009888" value="0x2D00C000" />
+        <register type="NOA" address="0x00009888" value="0x2F00C000" />
+        <register type="NOA" address="0x00009888" value="0x1300C000" />
+        <register type="NOA" address="0x00009888" value="0x1500C000" />
+        <register type="NOA" address="0x00009888" value="0x1700C000" />
+        <register type="NOA" address="0x00009888" value="0x1900C000" />
+        <register type="NOA" address="0x00009888" value="0x47000101" />
+        <register type="NOA" address="0x00009888" value="0x39000001" />
+        <register type="NOA" address="0x00009888" value="0x49001300" />
+        <register type="NOA" address="0x00009888" value="0x3D000021" />
+        <register type="NOA" address="0x00009888" value="0x4B001100" />
+        <register type="NOA" address="0x00009888" value="0x3F000420" />
+        <register type="NOA" address="0x00009888" value="0x4D001100" />
+        <register type="NOA" address="0x00009888" value="0x41000400" />
+        <register type="NOA" address="0x00009888" value="0x35000000" />
+        <register type="NOA" address="0x00009888" value="0x31000000" />
+        <register type="NOA" address="0x00009888" value="0x43000001" />
+        <register type="NOA" address="0x00009888" value="0x3B0000C7" />
+    </register_config>
+    <register_config type="OA">
+        <register type="OA" address="0x00002740" value="0x00000000" />
+        <register type="OA" address="0x00002710" value="0x00000000" />
+        <register type="OA" address="0x00002714" value="0x70800000" />
+        <register type="OA" address="0x00002720" value="0x00000000" />
+        <register type="OA" address="0x00002724" value="0x00800000" />
+        <register type="OA" address="0x00002770" value="0x20000801" />
+        <register type="OA" address="0x00002774" value="0x0111FEEE" />
+        <register type="OA" address="0x00002778" value="0x40001001" />
+        <register type="OA" address="0x0000277C" value="0x0222FDDD" />
+        <register type="OA" address="0x00002780" value="0x80002002" />
+        <register type="OA" address="0x00002784" value="0x0444FBBB" />
+    </register_config>
+    <register_config type="FLEX">
+        <register type="FLEX" address="0x0000E458" value="0x00005004" />
+        <register type="FLEX" address="0x0000E558" value="0x00010003" />
+        <register type="FLEX" address="0x0000E658" value="0x00012011" />
+        <register type="FLEX" address="0x0000E45C" value="0x00051050" />
+        <register type="FLEX" address="0x0000E55C" value="0x00053052" />
+    </register_config>
+  </set>
+
+  <set name="Metric set L3_2"
+       mdapi_supported_apis="OGL OCL IO BB"
+       underscore_name="l3_2"
+       hw_config_guid="918834e2-a1a4-4d77-9f3a-0c229f8d803f"
+       chipset="CNL"
+       symbol_name="L3_2"
+       >
+    <counter name="GPU Core Clocks"
+             description="The total number of GPU core clocks elapsed during the measurement."
+             data_type="uint64"
+             equation="GPU_CLOCK 0 READ"
+             underscore_name="gpu_core_clocks"
+             units="cycles"
+             symbol_name="GpuCoreClocks"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="EU Active"
+             description="The percentage of time in which the Execution Units were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 7 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_active"
+             units="percent"
+             symbol_name="EuActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Slice1 L3 Bank1 Stalled"
+             description="The percentage of time in which slice1 L3 bank1 is stalled"
+             data_type="float"
+             max_equation="100"
+             equation="B 1 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="l31_bank1_stalled"
+             units="percent"
+             symbol_name="L31Bank1Stalled"
+             availability="$SliceMask 2 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GTI/L3"
+             mdapi_hw_unit_type="slice"
+             />
+    <counter name="EU Both FPU Pipes Active"
+             description="The percentage of time in which both EU FPU pipelines were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 9 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_fpu_both_active"
+             units="percent"
+             symbol_name="EuFpuBothActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="SQ is full"
+             description="The percentage of time when SQ is filled above a threshold (usually 48 entries)"
+             data_type="float"
+             max_equation="100"
+             equation="C 7 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="gt_request_queue_full"
+             units="percent"
+             symbol_name="GTRequestQueueFull"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="GTI"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS Send Pipe Active"
+             description="The percentage of time in which EU send pipeline was actively processing a vertex shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 12 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="vs_send_active"
+             units="percent"
+             symbol_name="VsSendActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS FPU1 Pipe Active"
+             description="The percentage of time in which EU FPU1 pipeline was actively processing a vertex shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 11 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="vs_fpu1_active"
+             units="percent"
+             symbol_name="VsFpu1Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GS Threads Dispatched"
+             description="The total number of geometry shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 5 READ"
+             underscore_name="gs_threads"
+             units="threads"
+             symbol_name="GsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Geometry Shader"
+             />
+    <counter name="Early Hi-Depth Test Fails"
+             description="The total number of pixels dropped on early hierarchical depth test."
+             data_type="uint64"
+             equation="A 22 READ 4 UMUL"
+             underscore_name="hi_depth_test_fails"
+             units="pixels"
+             symbol_name="HiDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Hi-Depth Test"
+             />
+    <counter name="FS Both FPU Active"
+             description="The percentage of time in which fragment shaders were processed actively on the both FPUs."
+             data_type="float"
+             max_equation="100"
+             equation="A 18 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_eu_both_fpu_active"
+             units="percent"
+             symbol_name="PsEuBothFpuActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Frame Batch Draw"
+             mdapi_group="3D Pipe/Fragment Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS Threads Dispatched"
+             description="The total number of vertex shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 1 READ"
+             underscore_name="vs_threads"
+             units="threads"
+             symbol_name="VsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Vertex Shader"
+             />
+    <counter name="FS Threads Dispatched"
+             description="The total number of fragment shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 6 READ"
+             underscore_name="ps_threads"
+             units="threads"
+             symbol_name="PsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Fragment Shader"
+             />
+    <counter name="Shader Barrier Messages"
+             description="The total number of shader barrier messages."
+             data_type="uint64"
+             equation="A 35 READ"
+             underscore_name="shader_barriers"
+             units="messages"
+             symbol_name="ShaderBarriers"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Barrier"
+             />
+    <counter name="Slice1 L3 Bank0 Stalled"
+             description="The percentage of time in which slice1 L3 bank0 is stalled"
+             data_type="float"
+             max_equation="100"
+             equation="B 0 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="l31_bank0_stalled"
+             units="percent"
+             symbol_name="L31Bank0Stalled"
+             availability="$SliceMask 2 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GTI/L3"
+             mdapi_hw_unit_type="slice"
+             />
+    <counter name="Sampler Texels"
+             description="The total number of texels seen on input (with 2x2 accuracy) in all sampler units."
+             data_type="uint64"
+             equation="A 28 READ 4 UMUL"
+             underscore_name="sampler_texels"
+             units="texels"
+             symbol_name="SamplerTexels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Input"
+             />
+    <counter name="Pixels Failing Tests"
+             description="The total number of pixels dropped on post-FS alpha, stencil, or depth tests."
+             data_type="uint64"
+             equation="A 25 READ 4 UMUL"
+             underscore_name="pixels_failing_post_ps_tests"
+             units="pixels"
+             symbol_name="PixelsFailingPostPsTests"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="GPU Time Elapsed"
+             description="Time elapsed on the GPU during the measurement."
+             data_type="uint64"
+             equation="GPU_TIME 0 READ 1000000000 UMUL $GpuTimestampFrequency UDIV"
+             underscore_name="gpu_time"
+             units="ns"
+             symbol_name="GpuTime"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="AVG GPU Core Frequency"
+             description="Average GPU Core Frequency in the measurement."
+             data_type="uint64"
+             max_equation="$GpuMaxFrequency"
+             equation="$GpuCoreClocks 1000000000 UMUL $GpuTime UDIV"
+             underscore_name="avg_gpu_core_frequency"
+             units="hz"
+             symbol_name="AvgGpuCoreFrequency"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Sampler Texels Misses"
+             description="The total number of texels lookups (with 2x2 accuracy) that missed L1 sampler cache."
+             data_type="uint64"
+             equation="A 29 READ 4 UMUL"
+             underscore_name="sampler_texel_misses"
+             units="texels"
+             symbol_name="SamplerTexelMisses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Batch Frame Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Cache"
+             />
+    <counter name="CS Threads Dispatched"
+             description="The total number of compute shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 4 READ"
+             underscore_name="cs_threads"
+             units="threads"
+             symbol_name="CsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Compute Shader"
+             />
+    <counter name="SLM Bytes Read"
+             description="The total number of GPU memory bytes read from shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ 64 UMUL"
+             underscore_name="slm_bytes_read"
+             units="bytes"
+             symbol_name="SlmBytesRead"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="PS FPU1 Pipe Active"
+             description="The percentage of time in which EU FPU1 pipeline was actively processing a pixel shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 16 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_fpu1_active"
+             units="percent"
+             symbol_name="PsFpu1Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pixel Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Slice0 L3 Bank2 Stalled"
+             description="The percentage of time in which slice0 L3 bank2 is stalled"
+             data_type="float"
+             max_equation="100"
+             equation="B 2 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="l30_bank2_stalled"
+             units="percent"
+             symbol_name="L30Bank2Stalled"
+             availability="$SliceMask 1 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GTI/L3"
+             mdapi_hw_unit_type="slice"
+             />
+    <counter name="PS Send Pipeline Active"
+             description="The percentage of time in which EU send pipeline was actively processing a pixel shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 17 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_send_active"
+             units="percent"
+             symbol_name="PsSendActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pixel Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS FPU0 Pipe Active"
+             description="The percentage of time in which EU FPU0 pipeline was actively processing a vertex shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 10 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="vs_fpu0_active"
+             units="percent"
+             symbol_name="VsFpu0Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GPU Busy"
+             description="The percentage of time in which the GPU has been processing GPU commands."
+             data_type="float"
+             max_equation="100"
+             equation="A 0 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="gpu_busy"
+             units="percent"
+             symbol_name="GpuBusy"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Slice1 L3 Bank2 Active"
+             description="The percentage of time in which slice1 L3 bank2 is active"
+             data_type="float"
+             max_equation="100"
+             equation="C 5 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="l31_bank2_active"
+             units="percent"
+             symbol_name="L31Bank2Active"
+             availability="$SliceMask 2 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GTI/L3"
+             mdapi_hw_unit_type="slice"
+             />
+    <counter name="Rasterized Pixels"
+             description="The total number of rasterized pixels."
+             data_type="uint64"
+             equation="A 21 READ 4 UMUL"
+             underscore_name="rasterized_pixels"
+             units="pixels"
+             symbol_name="RasterizedPixels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer"
+             />
+    <counter name="Slice1 L3 Bank1 Active"
+             description="The percentage of time in which slice1 L3 bank1 is active"
+             data_type="float"
+             max_equation="100"
+             equation="C 4 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="l31_bank1_active"
+             units="percent"
+             symbol_name="L31Bank1Active"
+             availability="$SliceMask 2 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GTI/L3"
+             mdapi_hw_unit_type="slice"
+             />
+    <counter name="PS FPU0 Pipe Active"
+             description="The percentage of time in which EU FPU0 pipeline was actively processing a pixel shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 15 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_fpu0_active"
+             units="percent"
+             symbol_name="PsFpu0Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pixel Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="DS Threads Dispatched"
+             description="The total number of domain shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 3 READ"
+             underscore_name="ds_threads"
+             units="threads"
+             symbol_name="DsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Domain Shader"
+             />
+    <counter name="Samples Written"
+             description="The total number of samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 26 READ 4 UMUL"
+             underscore_name="samples_written"
+             units="pixels"
+             symbol_name="SamplesWritten"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="EU Stall"
+             description="The percentage of time in which the Execution Units were stalled."
+             data_type="float"
+             max_equation="100"
+             equation="A 8 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_stall"
+             units="percent"
+             symbol_name="EuStall"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Slice1 L3 Bank0 Active"
+             description="The percentage of time in which slice1 L3 bank0 is active"
+             data_type="float"
+             max_equation="100"
+             equation="C 3 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="l31_bank0_active"
+             units="percent"
+             symbol_name="L31Bank0Active"
+             availability="$SliceMask 2 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GTI/L3"
+             mdapi_hw_unit_type="slice"
+             />
+    <counter name="Samples Blended"
+             description="The total number of blended samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 27 READ 4 UMUL"
+             underscore_name="samples_blended"
+             units="pixels"
+             symbol_name="SamplesBlended"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="Early Depth Test Fails"
+             description="The total number of pixels dropped on early depth test."
+             data_type="uint64"
+             equation="A 23 READ 4 UMUL"
+             underscore_name="early_depth_test_fails"
+             units="pixels"
+             symbol_name="EarlyDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Early Depth Test"
+             />
+    <counter name="Shader Memory Accesses"
+             description="The total number of shader memory accesses to L3."
+             data_type="uint64"
+             equation="A 32 READ"
+             underscore_name="shader_memory_accesses"
+             units="messages"
+             symbol_name="ShaderMemoryAccesses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="HS Threads Dispatched"
+             description="The total number of hull shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 2 READ"
+             underscore_name="hs_threads"
+             units="threads"
+             symbol_name="HsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Hull Shader"
+             />
+    <counter name="L3 Shader Throughput"
+             description="The total number of GPU memory bytes transferred between shaders and L3 caches w/o URB."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ A 31 READ UADD A 32 READ UADD 64 UMUL"
+             underscore_name="l3_shader_throughput"
+             units="bytes"
+             symbol_name="L3ShaderThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_group="L3/Data Port"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Samples Killed in FS"
+             description="The total number of samples or pixels dropped in fragment shaders."
+             data_type="uint64"
+             equation="A 24 READ 4 UMUL"
+             underscore_name="samples_killed_in_ps"
+             units="pixels"
+             symbol_name="SamplesKilledInPs"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Fragment Shader"
+             />
+    <counter name="SLM Bytes Written"
+             description="The total number of GPU memory bytes written into shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 31 READ 64 UMUL"
+             underscore_name="slm_bytes_written"
+             units="bytes"
+             symbol_name="SlmBytesWritten"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Shader Atomic Memory Accesses"
+             description="The total number of shader atomic memory accesses."
+             data_type="uint64"
+             equation="A 34 READ"
+             underscore_name="shader_atomics"
+             units="messages"
+             symbol_name="ShaderAtomics"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port/Atomics"
+             />
+    <register_config type="NOA">
+        <register type="NOA" address="0x00000D04" value="0x00000200" />
+        <register type="NOA" address="0x00009884" value="0x00000000" />
+        <register type="NOA" address="0x00009888" value="0x106C0400" />
+        <register type="NOA" address="0x00009888" value="0x166C0000" />
+        <register type="NOA" address="0x00009888" value="0x16E1001A" />
+        <register type="NOA" address="0x00009888" value="0x16E5001A" />
+        <register type="NOA" address="0x00009888" value="0x16E9001A" />
+        <register type="NOA" address="0x00009888" value="0x10EC1C00" />
+        <register type="NOA" address="0x00009888" value="0x12EC0004" />
+        <register type="NOA" address="0x00009888" value="0x16EC001D" />
+        <register type="NOA" address="0x00009888" value="0x086C0020" />
+        <register type="NOA" address="0x00009888" value="0x0C6C0020" />
+        <register type="NOA" address="0x00009888" value="0x006C3600" />
+        <register type="NOA" address="0x00009888" value="0x186C0A88" />
+        <register type="NOA" address="0x00009888" value="0x1A6C0504" />
+        <register type="NOA" address="0x00009888" value="0x1C6C0005" />
+        <register type="NOA" address="0x00009888" value="0x0073F000" />
+        <register type="NOA" address="0x00009888" value="0x02739000" />
+        <register type="NOA" address="0x00009888" value="0x0473D000" />
+        <register type="NOA" address="0x00009888" value="0x0673D000" />
+        <register type="NOA" address="0x00009888" value="0x00778000" />
+        <register type="NOA" address="0x00009888" value="0x0E778000" />
+        <register type="NOA" address="0x00009888" value="0x14778000" />
+        <register type="NOA" address="0x00009888" value="0x16778A80" />
+        <register type="NOA" address="0x00009888" value="0x18770002" />
+        <register type="NOA" address="0x00009888" value="0x02778000" />
+        <register type="NOA" address="0x00009888" value="0x04778000" />
+        <register type="NOA" address="0x00009888" value="0x06778000" />
+        <register type="NOA" address="0x00009888" value="0x08778000" />
+        <register type="NOA" address="0x00009888" value="0x02E10024" />
+        <register type="NOA" address="0x00009888" value="0x10E10000" />
+        <register type="NOA" address="0x00009888" value="0x1EE10000" />
+        <register type="NOA" address="0x00009888" value="0x02E52400" />
+        <register type="NOA" address="0x00009888" value="0x10E50000" />
+        <register type="NOA" address="0x00009888" value="0x1EE50000" />
+        <register type="NOA" address="0x00009888" value="0x04E90024" />
+        <register type="NOA" address="0x00009888" value="0x10E90000" />
+        <register type="NOA" address="0x00009888" value="0x1EE90000" />
+        <register type="NOA" address="0x00009888" value="0x00EC0020" />
+        <register type="NOA" address="0x00009888" value="0x06EC2800" />
+        <register type="NOA" address="0x00009888" value="0x0AEC2820" />
+        <register type="NOA" address="0x00009888" value="0x0EEC3832" />
+        <register type="NOA" address="0x00009888" value="0x18EC0580" />
+        <register type="NOA" address="0x00009888" value="0x1AEC0000" />
+        <register type="NOA" address="0x00009888" value="0x1CEC0000" />
+        <register type="NOA" address="0x00009888" value="0x18ED8000" />
+        <register type="NOA" address="0x00009888" value="0x0AEE1000" />
+        <register type="NOA" address="0x00009888" value="0x00F3D000" />
+        <register type="NOA" address="0x00009888" value="0x02F38000" />
+        <register type="NOA" address="0x00009888" value="0x04F3C000" />
+        <register type="NOA" address="0x00009888" value="0x06F3C000" />
+        <register type="NOA" address="0x00009888" value="0x00F78000" />
+        <register type="NOA" address="0x00009888" value="0x0EF78000" />
+        <register type="NOA" address="0x00009888" value="0x16F78280" />
+        <register type="NOA" address="0x00009888" value="0x18F70002" />
+        <register type="NOA" address="0x00009888" value="0x04F78000" />
+        <register type="NOA" address="0x00009888" value="0x06F78000" />
+        <register type="NOA" address="0x00009888" value="0x08F74000" />
+        <register type="NOA" address="0x00009888" value="0x04F84000" />
+        <register type="NOA" address="0x00009888" value="0x0AFB0004" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x2100F000" />
+        <register type="NOA" address="0x00009888" value="0x1D000071" />
+        <register type="NOA" address="0x00009888" value="0x33000000" />
+        <register type="NOA" address="0x00009888" value="0x1100C000" />
+        <register type="NOA" address="0x00009888" value="0x1F00C000" />
+        <register type="NOA" address="0x00009888" value="0x2500C000" />
+        <register type="NOA" address="0x00009888" value="0x2700C000" />
+        <register type="NOA" address="0x00009888" value="0x2900C000" />
+        <register type="NOA" address="0x00009888" value="0x2D00C000" />
+        <register type="NOA" address="0x00009888" value="0x2F00C000" />
+        <register type="NOA" address="0x00009888" value="0x1300C000" />
+        <register type="NOA" address="0x00009888" value="0x1500C000" />
+        <register type="NOA" address="0x00009888" value="0x1700C000" />
+        <register type="NOA" address="0x00009888" value="0x1900C000" />
+        <register type="NOA" address="0x00009888" value="0x47000010" />
+        <register type="NOA" address="0x00009888" value="0x39000883" />
+        <register type="NOA" address="0x00009888" value="0x49000300" />
+        <register type="NOA" address="0x00009888" value="0x3D001064" />
+        <register type="NOA" address="0x00009888" value="0x4B000001" />
+        <register type="NOA" address="0x00009888" value="0x3F000C60" />
+        <register type="NOA" address="0x00009888" value="0x4D000001" />
+        <register type="NOA" address="0x00009888" value="0x41000C04" />
+        <register type="NOA" address="0x00009888" value="0x35000000" />
+        <register type="NOA" address="0x00009888" value="0x31000000" />
+        <register type="NOA" address="0x00009888" value="0x43000003" />
+        <register type="NOA" address="0x00009888" value="0x3B000001" />
+    </register_config>
+    <register_config type="OA">
+        <register type="OA" address="0x00002740" value="0x00000000" />
+        <register type="OA" address="0x00002710" value="0x00000000" />
+        <register type="OA" address="0x00002714" value="0x70800000" />
+        <register type="OA" address="0x00002720" value="0x00000000" />
+        <register type="OA" address="0x00002724" value="0x00800000" />
+        <register type="OA" address="0x00002770" value="0x20000801" />
+        <register type="OA" address="0x00002774" value="0x0111FEEE" />
+        <register type="OA" address="0x00002778" value="0x40001001" />
+        <register type="OA" address="0x0000277C" value="0x0222FDDD" />
+        <register type="OA" address="0x00002780" value="0x80002002" />
+        <register type="OA" address="0x00002784" value="0x0444FBBB" />
+    </register_config>
+    <register_config type="FLEX">
+        <register type="FLEX" address="0x0000E458" value="0x00005004" />
+        <register type="FLEX" address="0x0000E558" value="0x00010003" />
+        <register type="FLEX" address="0x0000E658" value="0x00012011" />
+        <register type="FLEX" address="0x0000E45C" value="0x00051050" />
+        <register type="FLEX" address="0x0000E55C" value="0x00053052" />
+    </register_config>
+  </set>
+
+  <set name="Metric set RasterizerAndPixelBackend"
+       mdapi_supported_apis="OGL OCL IO BB"
+       underscore_name="rasterizer_and_pixel_backend"
+       hw_config_guid="2e95713a-c6d8-47a9-9207-eba3e208be51"
+       chipset="CNL"
+       symbol_name="RasterizerAndPixelBackend"
+       >
+    <counter name="GPU Core Clocks"
+             description="The total number of GPU core clocks elapsed during the measurement."
+             data_type="uint64"
+             equation="GPU_CLOCK 0 READ"
+             underscore_name="gpu_core_clocks"
+             units="cycles"
+             symbol_name="GpuCoreClocks"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="EU Active"
+             description="The percentage of time in which the Execution Units were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 7 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_active"
+             units="percent"
+             symbol_name="EuActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EU Both FPU Pipes Active"
+             description="The percentage of time in which both EU FPU pipelines were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 9 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_fpu_both_active"
+             units="percent"
+             symbol_name="EuFpuBothActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Slice1 Rasterizer Input Available"
+             description="The percentage of time in which slice1 rasterizer input is available"
+             data_type="float"
+             max_equation="100"
+             equation="B 1 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="rasterizer1_input_available"
+             units="percent"
+             symbol_name="Rasterizer1InputAvailable"
+             availability="$SliceMask 2 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Rasterizer"
+             mdapi_hw_unit_type="slice"
+             />
+    <counter name="VS Send Pipe Active"
+             description="The percentage of time in which EU send pipeline was actively processing a vertex shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 12 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="vs_send_active"
+             units="percent"
+             symbol_name="VsSendActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS FPU1 Pipe Active"
+             description="The percentage of time in which EU FPU1 pipeline was actively processing a vertex shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 11 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="vs_fpu1_active"
+             units="percent"
+             symbol_name="VsFpu1Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GS Threads Dispatched"
+             description="The total number of geometry shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 5 READ"
+             underscore_name="gs_threads"
+             units="threads"
+             symbol_name="GsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Geometry Shader"
+             />
+    <counter name="Early Hi-Depth Test Fails"
+             description="The total number of pixels dropped on early hierarchical depth test."
+             data_type="uint64"
+             equation="A 22 READ 4 UMUL"
+             underscore_name="hi_depth_test_fails"
+             units="pixels"
+             symbol_name="HiDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Hi-Depth Test"
+             />
+    <counter name="FS Both FPU Active"
+             description="The percentage of time in which fragment shaders were processed actively on the both FPUs."
+             data_type="float"
+             max_equation="100"
+             equation="A 18 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_eu_both_fpu_active"
+             units="percent"
+             symbol_name="PsEuBothFpuActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Frame Batch Draw"
+             mdapi_group="3D Pipe/Fragment Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS Threads Dispatched"
+             description="The total number of vertex shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 1 READ"
+             underscore_name="vs_threads"
+             units="threads"
+             symbol_name="VsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Vertex Shader"
+             />
+    <counter name="FS Threads Dispatched"
+             description="The total number of fragment shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 6 READ"
+             underscore_name="ps_threads"
+             units="threads"
+             symbol_name="PsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Fragment Shader"
+             />
+    <counter name="Slice0 Pixel Values Ready"
+             description="The percentage of time in which slice0 pixel values are ready"
+             data_type="float"
+             max_equation="100"
+             equation="C 0 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="pixel_values0_ready"
+             units="percent"
+             symbol_name="PixelValues0Ready"
+             availability="$SliceMask 1 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/3D Pipe"
+             mdapi_hw_unit_type="slice"
+             />
+    <counter name="Shader Barrier Messages"
+             description="The total number of shader barrier messages."
+             data_type="uint64"
+             equation="A 35 READ"
+             underscore_name="shader_barriers"
+             units="messages"
+             symbol_name="ShaderBarriers"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Barrier"
+             />
+    <counter name="Sampler Texels"
+             description="The total number of texels seen on input (with 2x2 accuracy) in all sampler units."
+             data_type="uint64"
+             equation="A 28 READ 4 UMUL"
+             underscore_name="sampler_texels"
+             units="texels"
+             symbol_name="SamplerTexels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Input"
+             />
+    <counter name="Pixels Failing Tests"
+             description="The total number of pixels dropped on post-FS alpha, stencil, or depth tests."
+             data_type="uint64"
+             equation="A 25 READ 4 UMUL"
+             underscore_name="pixels_failing_post_ps_tests"
+             units="pixels"
+             symbol_name="PixelsFailingPostPsTests"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="GPU Time Elapsed"
+             description="Time elapsed on the GPU during the measurement."
+             data_type="uint64"
+             equation="GPU_TIME 0 READ 1000000000 UMUL $GpuTimestampFrequency UDIV"
+             underscore_name="gpu_time"
+             units="ns"
+             symbol_name="GpuTime"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="AVG GPU Core Frequency"
+             description="Average GPU Core Frequency in the measurement."
+             data_type="uint64"
+             max_equation="$GpuMaxFrequency"
+             equation="$GpuCoreClocks 1000000000 UMUL $GpuTime UDIV"
+             underscore_name="avg_gpu_core_frequency"
+             units="hz"
+             symbol_name="AvgGpuCoreFrequency"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Sampler Texels Misses"
+             description="The total number of texels lookups (with 2x2 accuracy) that missed L1 sampler cache."
+             data_type="uint64"
+             equation="A 29 READ 4 UMUL"
+             underscore_name="sampler_texel_misses"
+             units="texels"
+             symbol_name="SamplerTexelMisses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Batch Frame Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Cache"
+             />
+    <counter name="CS Threads Dispatched"
+             description="The total number of compute shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 4 READ"
+             underscore_name="cs_threads"
+             units="threads"
+             symbol_name="CsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Compute Shader"
+             />
+    <counter name="Slice0 Rasterizer Input Available"
+             description="The percentage of time in which slice0 rasterizer input is available"
+             data_type="float"
+             max_equation="100"
+             equation="B 0 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="rasterizer0_input_available"
+             units="percent"
+             symbol_name="Rasterizer0InputAvailable"
+             availability="$SliceMask 1 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Rasterizer"
+             mdapi_hw_unit_type="slice"
+             />
+    <counter name="Slice0 Post-EarlyZ Pixel Data Ready"
+             description="The percentage of time in which slice0 post-EarlyZ pixel data is ready (after early Z tests have been applied)"
+             data_type="float"
+             max_equation="100"
+             equation="B 4 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="pixel_data0_ready"
+             units="percent"
+             symbol_name="PixelData0Ready"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Rasterizer/Early Depth Test"
+             mdapi_hw_unit_type="slice"
+             />
+    <counter name="SLM Bytes Read"
+             description="The total number of GPU memory bytes read from shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ 64 UMUL"
+             underscore_name="slm_bytes_read"
+             units="bytes"
+             symbol_name="SlmBytesRead"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="PS FPU1 Pipe Active"
+             description="The percentage of time in which EU FPU1 pipeline was actively processing a pixel shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 16 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_fpu1_active"
+             units="percent"
+             symbol_name="PsFpu1Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pixel Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="PS Send Pipeline Active"
+             description="The percentage of time in which EU send pipeline was actively processing a pixel shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 17 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_send_active"
+             units="percent"
+             symbol_name="PsSendActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pixel Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS FPU0 Pipe Active"
+             description="The percentage of time in which EU FPU0 pipeline was actively processing a vertex shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 10 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="vs_fpu0_active"
+             units="percent"
+             symbol_name="VsFpu0Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GPU Busy"
+             description="The percentage of time in which the GPU has been processing GPU commands."
+             data_type="float"
+             max_equation="100"
+             equation="A 0 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="gpu_busy"
+             units="percent"
+             symbol_name="GpuBusy"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Slice1 Pixel Values Ready"
+             description="The percentage of time in which slice1 pixel values are ready"
+             data_type="float"
+             max_equation="100"
+             equation="C 1 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="pixel_values1_ready"
+             units="percent"
+             symbol_name="PixelValues1Ready"
+             availability="$SliceMask 2 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/3D Pipe"
+             mdapi_hw_unit_type="slice"
+             />
+    <counter name="Slice1 PS Output Available"
+             description="The percentage of time in which slice1 PS output is available"
+             data_type="float"
+             max_equation="100"
+             equation="B 7 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_output1_available"
+             units="percent"
+             symbol_name="PSOutput1Available"
+             availability="$SliceMask 2 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/3D Pipe"
+             mdapi_hw_unit_type="slice"
+             />
+    <counter name="Rasterized Pixels"
+             description="The total number of rasterized pixels."
+             data_type="uint64"
+             equation="A 21 READ 4 UMUL"
+             underscore_name="rasterized_pixels"
+             units="pixels"
+             symbol_name="RasterizedPixels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer"
+             />
+    <counter name="PS FPU0 Pipe Active"
+             description="The percentage of time in which EU FPU0 pipeline was actively processing a pixel shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 15 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_fpu0_active"
+             units="percent"
+             symbol_name="PsFpu0Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pixel Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="DS Threads Dispatched"
+             description="The total number of domain shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 3 READ"
+             underscore_name="ds_threads"
+             units="threads"
+             symbol_name="DsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Domain Shader"
+             />
+    <counter name="Samples Written"
+             description="The total number of samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 26 READ 4 UMUL"
+             underscore_name="samples_written"
+             units="pixels"
+             symbol_name="SamplesWritten"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="EU Stall"
+             description="The percentage of time in which the Execution Units were stalled."
+             data_type="float"
+             max_equation="100"
+             equation="A 8 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_stall"
+             units="percent"
+             symbol_name="EuStall"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Slice0 Rasterizer Output Ready"
+             description="The percentage of time in which slice0 rasterizer output is ready"
+             data_type="float"
+             max_equation="100"
+             equation="B 2 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="rasterizer0_output_ready"
+             units="percent"
+             symbol_name="Rasterizer0OutputReady"
+             availability="$SliceMask 1 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Rasterizer"
+             mdapi_hw_unit_type="slice"
+             />
+    <counter name="Samples Blended"
+             description="The total number of blended samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 27 READ 4 UMUL"
+             underscore_name="samples_blended"
+             units="pixels"
+             symbol_name="SamplesBlended"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="Early Depth Test Fails"
+             description="The total number of pixels dropped on early depth test."
+             data_type="uint64"
+             equation="A 23 READ 4 UMUL"
+             underscore_name="early_depth_test_fails"
+             units="pixels"
+             symbol_name="EarlyDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Early Depth Test"
+             />
+    <counter name="Slice1 Rasterizer Output Ready"
+             description="The percentage of time in which slice1 rasterizer output is ready"
+             data_type="float"
+             max_equation="100"
+             equation="B 3 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="rasterizer1_output_ready"
+             units="percent"
+             symbol_name="Rasterizer1OutputReady"
+             availability="$SliceMask 2 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Rasterizer"
+             mdapi_hw_unit_type="slice"
+             />
+    <counter name="Slice1 Post-EarlyZ Pixel Data Ready"
+             description="The percentage of time in which slice1 post-EarlyZ pixel data is ready (after early Z tests have been applied)"
+             data_type="float"
+             max_equation="100"
+             equation="B 5 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="pixel_data1_ready"
+             units="percent"
+             symbol_name="PixelData1Ready"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Rasterizer/Early Depth Test"
+             mdapi_hw_unit_type="slice"
+             />
+    <counter name="Shader Memory Accesses"
+             description="The total number of shader memory accesses to L3."
+             data_type="uint64"
+             equation="A 32 READ"
+             underscore_name="shader_memory_accesses"
+             units="messages"
+             symbol_name="ShaderMemoryAccesses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="SQ is full"
+             description="The percentage of time when SQ is filled above a threshold (usually 48 entries)"
+             data_type="float"
+             max_equation="100"
+             equation="C 7 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="gt_request_queue_full"
+             units="percent"
+             symbol_name="GTRequestQueueFull"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="GTI"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="HS Threads Dispatched"
+             description="The total number of hull shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 2 READ"
+             underscore_name="hs_threads"
+             units="threads"
+             symbol_name="HsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Hull Shader"
+             />
+    <counter name="L3 Shader Throughput"
+             description="The total number of GPU memory bytes transferred between shaders and L3 caches w/o URB."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ A 31 READ UADD A 32 READ UADD 64 UMUL"
+             underscore_name="l3_shader_throughput"
+             units="bytes"
+             symbol_name="L3ShaderThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_group="L3/Data Port"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Samples Killed in FS"
+             description="The total number of samples or pixels dropped in fragment shaders."
+             data_type="uint64"
+             equation="A 24 READ 4 UMUL"
+             underscore_name="samples_killed_in_ps"
+             units="pixels"
+             symbol_name="SamplesKilledInPs"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Fragment Shader"
+             />
+    <counter name="SLM Bytes Written"
+             description="The total number of GPU memory bytes written into shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 31 READ 64 UMUL"
+             underscore_name="slm_bytes_written"
+             units="bytes"
+             symbol_name="SlmBytesWritten"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Slice0 PS Output Available"
+             description="The percentage of time in which slice0 PS output is available"
+             data_type="float"
+             max_equation="100"
+             equation="B 6 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_output0_available"
+             units="percent"
+             symbol_name="PSOutput0Available"
+             availability="$SliceMask 1 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/3D Pipe"
+             mdapi_hw_unit_type="slice"
+             />
+    <counter name="Shader Atomic Memory Accesses"
+             description="The total number of shader atomic memory accesses."
+             data_type="uint64"
+             equation="A 34 READ"
+             underscore_name="shader_atomics"
+             units="messages"
+             symbol_name="ShaderAtomics"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port/Atomics"
+             />
+    <register_config type="NOA">
+        <register type="NOA" address="0x00000D04" value="0x00000200" />
+        <register type="NOA" address="0x00009884" value="0x00000000" />
+        <register type="NOA" address="0x00009888" value="0x0C710004" />
+        <register type="NOA" address="0x00009888" value="0x10747000" />
+        <register type="NOA" address="0x00009888" value="0x12756800" />
+        <register type="NOA" address="0x00009888" value="0x04795000" />
+        <register type="NOA" address="0x00009888" value="0x127C01C0" />
+        <register type="NOA" address="0x00009888" value="0x0CF10004" />
+        <register type="NOA" address="0x00009888" value="0x10F47000" />
+        <register type="NOA" address="0x00009888" value="0x12F56800" />
+        <register type="NOA" address="0x00009888" value="0x04F95000" />
+        <register type="NOA" address="0x00009888" value="0x12FC01C0" />
+        <register type="NOA" address="0x00009888" value="0x1A6C4044" />
+        <register type="NOA" address="0x00009888" value="0x1C6C0004" />
+        <register type="NOA" address="0x00009888" value="0x186C00A0" />
+        <register type="NOA" address="0x00009888" value="0x00710010" />
+        <register type="NOA" address="0x00009888" value="0x02738000" />
+        <register type="NOA" address="0x00009888" value="0x04732000" />
+        <register type="NOA" address="0x00009888" value="0x0673A000" />
+        <register type="NOA" address="0x00009888" value="0x00736000" />
+        <register type="NOA" address="0x00009888" value="0x08740037" />
+        <register type="NOA" address="0x00009888" value="0x0C740000" />
+        <register type="NOA" address="0x00009888" value="0x0A750F36" />
+        <register type="NOA" address="0x00009888" value="0x10750000" />
+        <register type="NOA" address="0x00009888" value="0x04751000" />
+        <register type="NOA" address="0x00009888" value="0x0C752000" />
+        <register type="NOA" address="0x00009888" value="0x0E750000" />
+        <register type="NOA" address="0x00009888" value="0x0C760800" />
+        <register type="NOA" address="0x00009888" value="0x0E768800" />
+        <register type="NOA" address="0x00009888" value="0x10760002" />
+        <register type="NOA" address="0x00009888" value="0x0077C000" />
+        <register type="NOA" address="0x00009888" value="0x0E778000" />
+        <register type="NOA" address="0x00009888" value="0x1477C000" />
+        <register type="NOA" address="0x00009888" value="0x167767E0" />
+        <register type="NOA" address="0x00009888" value="0x18770002" />
+        <register type="NOA" address="0x00009888" value="0x02778000" />
+        <register type="NOA" address="0x00009888" value="0x04778000" />
+        <register type="NOA" address="0x00009888" value="0x0C790017" />
+        <register type="NOA" address="0x00009888" value="0x08790000" />
+        <register type="NOA" address="0x00009888" value="0x0C7B0088" />
+        <register type="NOA" address="0x00009888" value="0x0E7C0032" />
+        <register type="NOA" address="0x00009888" value="0x107C0000" />
+        <register type="NOA" address="0x00009888" value="0x167C0004" />
+        <register type="NOA" address="0x00009888" value="0x04F10200" />
+        <register type="NOA" address="0x00009888" value="0x00F10000" />
+        <register type="NOA" address="0x00009888" value="0x08F40DC0" />
+        <register type="NOA" address="0x00009888" value="0x0CF40000" />
+        <register type="NOA" address="0x00009888" value="0x00F50D80" />
+        <register type="NOA" address="0x00009888" value="0x02F5003C" />
+        <register type="NOA" address="0x00009888" value="0x10F50000" />
+        <register type="NOA" address="0x00009888" value="0x0AF54000" />
+        <register type="NOA" address="0x00009888" value="0x0CF58000" />
+        <register type="NOA" address="0x00009888" value="0x04F50000" />
+        <register type="NOA" address="0x00009888" value="0x06F50000" />
+        <register type="NOA" address="0x00009888" value="0x0EF62200" />
+        <register type="NOA" address="0x00009888" value="0x0CF6A000" />
+        <register type="NOA" address="0x00009888" value="0x0EF7C000" />
+        <register type="NOA" address="0x00009888" value="0x16F71030" />
+        <register type="NOA" address="0x00009888" value="0x18F70001" />
+        <register type="NOA" address="0x00009888" value="0x02F7C000" />
+        <register type="NOA" address="0x00009888" value="0x04F7C000" />
+        <register type="NOA" address="0x00009888" value="0x0CF905C0" />
+        <register type="NOA" address="0x00009888" value="0x08F90000" />
+        <register type="NOA" address="0x00009888" value="0x0CFB0220" />
+        <register type="NOA" address="0x00009888" value="0x0EFC1900" />
+        <register type="NOA" address="0x00009888" value="0x10FC0000" />
+        <register type="NOA" address="0x00009888" value="0x16FC0010" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x2100F000" />
+        <register type="NOA" address="0x00009888" value="0x1D000071" />
+        <register type="NOA" address="0x00009888" value="0x33000000" />
+        <register type="NOA" address="0x00009888" value="0x1100C000" />
+        <register type="NOA" address="0x00009888" value="0x1F00C000" />
+        <register type="NOA" address="0x00009888" value="0x2300C000" />
+        <register type="NOA" address="0x00009888" value="0x2500C000" />
+        <register type="NOA" address="0x00009888" value="0x2700C000" />
+        <register type="NOA" address="0x00009888" value="0x2900C000" />
+        <register type="NOA" address="0x00009888" value="0x2B00C000" />
+        <register type="NOA" address="0x00009888" value="0x2D00C000" />
+        <register type="NOA" address="0x00009888" value="0x2F00C000" />
+        <register type="NOA" address="0x00009888" value="0x1300C000" />
+        <register type="NOA" address="0x00009888" value="0x1500C000" />
+        <register type="NOA" address="0x00009888" value="0x47000001" />
+        <register type="NOA" address="0x00009888" value="0x39000421" />
+        <register type="NOA" address="0x00009888" value="0x49000300" />
+        <register type="NOA" address="0x00009888" value="0x3D000402" />
+        <register type="NOA" address="0x00009888" value="0x4B001101" />
+        <register type="NOA" address="0x00009888" value="0x3F000840" />
+        <register type="NOA" address="0x00009888" value="0x4D000101" />
+        <register type="NOA" address="0x00009888" value="0x41000801" />
+        <register type="NOA" address="0x00009888" value="0x35000000" />
+        <register type="NOA" address="0x00009888" value="0x31000000" />
+        <register type="NOA" address="0x00009888" value="0x43000001" />
+    </register_config>
+    <register_config type="OA">
+        <register type="OA" address="0x00002740" value="0x00000000" />
+        <register type="OA" address="0x00002710" value="0x00000000" />
+        <register type="OA" address="0x00002714" value="0x30800000" />
+        <register type="OA" address="0x00002720" value="0x00000000" />
+        <register type="OA" address="0x00002724" value="0x30800000" />
+        <register type="OA" address="0x00002770" value="0x00000180" />
+        <register type="OA" address="0x00002774" value="0x0000FFCF" />
+        <register type="OA" address="0x00002778" value="0x00006000" />
+        <register type="OA" address="0x0000277C" value="0x0000F3FF" />
+        <register type="OA" address="0x00002790" value="0x00000002" />
+        <register type="OA" address="0x00002794" value="0x0000FFFE" />
+        <register type="OA" address="0x00002798" value="0x00000002" />
+        <register type="OA" address="0x0000279C" value="0x0000FFFD" />
+    </register_config>
+    <register_config type="FLEX">
+        <register type="FLEX" address="0x0000E458" value="0x00005004" />
+        <register type="FLEX" address="0x0000E558" value="0x00010003" />
+        <register type="FLEX" address="0x0000E658" value="0x00012011" />
+        <register type="FLEX" address="0x0000E45C" value="0x00051050" />
+        <register type="FLEX" address="0x0000E55C" value="0x00053052" />
+    </register_config>
+  </set>
+
+  <set name="Metric set Sampler"
+       mdapi_supported_apis="OGL OCL IO BB"
+       underscore_name="sampler"
+       hw_config_guid="9bcf4ff2-c15f-4bea-8ec8-8db87791942e"
+       chipset="CNL"
+       symbol_name="Sampler"
+       >
+    <counter name="GPU Core Clocks"
+             description="The total number of GPU core clocks elapsed during the measurement."
+             data_type="uint64"
+             equation="GPU_CLOCK 0 READ"
+             underscore_name="gpu_core_clocks"
+             units="cycles"
+             symbol_name="GpuCoreClocks"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="EU Active"
+             description="The percentage of time in which the Execution Units were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 7 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_active"
+             units="percent"
+             symbol_name="EuActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Slice0 Subslice2 Input Available"
+             description="The percentage of time in which slice0 subslice2 sampler input is available"
+             data_type="float"
+             max_equation="100"
+             equation="B 2 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="sampler02_input_available"
+             units="percent"
+             symbol_name="Sampler02InputAvailable"
+             availability="$SubsliceMask 4 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Sampler"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="EU Both FPU Pipes Active"
+             description="The percentage of time in which both EU FPU pipelines were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 9 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_fpu_both_active"
+             units="percent"
+             symbol_name="EuFpuBothActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="SQ is full"
+             description="The percentage of time when SQ is filled above a threshold (usually 48 entries)"
+             data_type="float"
+             max_equation="100"
+             equation="C 2 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="gt_request_queue_full"
+             units="percent"
+             symbol_name="GTRequestQueueFull"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="GTI"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS Send Pipe Active"
+             description="The percentage of time in which EU send pipeline was actively processing a vertex shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 12 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="vs_send_active"
+             units="percent"
+             symbol_name="VsSendActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Slice0 Subslice0 Input Available"
+             description="The percentage of time in which slice0 subslice0 sampler input is available"
+             data_type="float"
+             max_equation="100"
+             equation="B 0 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="sampler00_input_available"
+             units="percent"
+             symbol_name="Sampler00InputAvailable"
+             availability="$SubsliceMask 1 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Sampler"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="VS FPU1 Pipe Active"
+             description="The percentage of time in which EU FPU1 pipeline was actively processing a vertex shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 11 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="vs_fpu1_active"
+             units="percent"
+             symbol_name="VsFpu1Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GS Threads Dispatched"
+             description="The total number of geometry shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 5 READ"
+             underscore_name="gs_threads"
+             units="threads"
+             symbol_name="GsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Geometry Shader"
+             />
+    <counter name="Slice1 Subslice0 Input Available"
+             description="The percentage of time in which slice1 subslice0 sampler input is available"
+             data_type="float"
+             max_equation="100"
+             equation="B 3 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="sampler10_input_available"
+             units="percent"
+             symbol_name="Sampler10InputAvailable"
+             availability="$SubsliceMask 8 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Sampler"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="Early Hi-Depth Test Fails"
+             description="The total number of pixels dropped on early hierarchical depth test."
+             data_type="uint64"
+             equation="A 22 READ 4 UMUL"
+             underscore_name="hi_depth_test_fails"
+             units="pixels"
+             symbol_name="HiDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Hi-Depth Test"
+             />
+    <counter name="FS Both FPU Active"
+             description="The percentage of time in which fragment shaders were processed actively on the both FPUs."
+             data_type="float"
+             max_equation="100"
+             equation="A 18 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_eu_both_fpu_active"
+             units="percent"
+             symbol_name="PsEuBothFpuActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Frame Batch Draw"
+             mdapi_group="3D Pipe/Fragment Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS Threads Dispatched"
+             description="The total number of vertex shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 1 READ"
+             underscore_name="vs_threads"
+             units="threads"
+             symbol_name="VsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Vertex Shader"
+             />
+    <counter name="Slice0 Subslice2 Sampler Output Ready"
+             description="The percentage of time in which slice0 subslice2 sampler output is ready"
+             data_type="float"
+             max_equation="100"
+             equation="C 5 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="sampler02_output_ready"
+             units="percent"
+             symbol_name="Sampler02OutputReady"
+             availability="$SubsliceMask 4 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Sampler"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="FS Threads Dispatched"
+             description="The total number of fragment shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 6 READ"
+             underscore_name="ps_threads"
+             units="threads"
+             symbol_name="PsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Fragment Shader"
+             />
+    <counter name="Shader Barrier Messages"
+             description="The total number of shader barrier messages."
+             data_type="uint64"
+             equation="A 35 READ"
+             underscore_name="shader_barriers"
+             units="messages"
+             symbol_name="ShaderBarriers"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Barrier"
+             />
+    <counter name="Slice0 Subslice1 Input Available"
+             description="The percentage of time in which slice0 subslice1 sampler input is available"
+             data_type="float"
+             max_equation="100"
+             equation="B 1 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="sampler01_input_available"
+             units="percent"
+             symbol_name="Sampler01InputAvailable"
+             availability="$SubsliceMask 2 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Sampler"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="Sampler Texels"
+             description="The total number of texels seen on input (with 2x2 accuracy) in all sampler units."
+             data_type="uint64"
+             equation="A 28 READ 4 UMUL"
+             underscore_name="sampler_texels"
+             units="texels"
+             symbol_name="SamplerTexels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Input"
+             />
+    <counter name="Pixels Failing Tests"
+             description="The total number of pixels dropped on post-FS alpha, stencil, or depth tests."
+             data_type="uint64"
+             equation="A 25 READ 4 UMUL"
+             underscore_name="pixels_failing_post_ps_tests"
+             units="pixels"
+             symbol_name="PixelsFailingPostPsTests"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="GPU Time Elapsed"
+             description="Time elapsed on the GPU during the measurement."
+             data_type="uint64"
+             equation="GPU_TIME 0 READ 1000000000 UMUL $GpuTimestampFrequency UDIV"
+             underscore_name="gpu_time"
+             units="ns"
+             symbol_name="GpuTime"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="AVG GPU Core Frequency"
+             description="Average GPU Core Frequency in the measurement."
+             data_type="uint64"
+             max_equation="$GpuMaxFrequency"
+             equation="$GpuCoreClocks 1000000000 UMUL $GpuTime UDIV"
+             underscore_name="avg_gpu_core_frequency"
+             units="hz"
+             symbol_name="AvgGpuCoreFrequency"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Sampler Texels Misses"
+             description="The total number of texels lookups (with 2x2 accuracy) that missed L1 sampler cache."
+             data_type="uint64"
+             equation="A 29 READ 4 UMUL"
+             underscore_name="sampler_texel_misses"
+             units="texels"
+             symbol_name="SamplerTexelMisses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Batch Frame Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Cache"
+             />
+    <counter name="CS Threads Dispatched"
+             description="The total number of compute shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 4 READ"
+             underscore_name="cs_threads"
+             units="threads"
+             symbol_name="CsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Compute Shader"
+             />
+    <counter name="SLM Bytes Read"
+             description="The total number of GPU memory bytes read from shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ 64 UMUL"
+             underscore_name="slm_bytes_read"
+             units="bytes"
+             symbol_name="SlmBytesRead"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="PS FPU1 Pipe Active"
+             description="The percentage of time in which EU FPU1 pipeline was actively processing a pixel shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 16 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_fpu1_active"
+             units="percent"
+             symbol_name="PsFpu1Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pixel Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="PS Send Pipeline Active"
+             description="The percentage of time in which EU send pipeline was actively processing a pixel shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 17 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_send_active"
+             units="percent"
+             symbol_name="PsSendActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pixel Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS FPU0 Pipe Active"
+             description="The percentage of time in which EU FPU0 pipeline was actively processing a vertex shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 10 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="vs_fpu0_active"
+             units="percent"
+             symbol_name="VsFpu0Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GPU Busy"
+             description="The percentage of time in which the GPU has been processing GPU commands."
+             data_type="float"
+             max_equation="100"
+             equation="A 0 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="gpu_busy"
+             units="percent"
+             symbol_name="GpuBusy"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Slice1 Subslice1 Input Available"
+             description="The percentage of time in which slice1 subslice1 sampler input is available"
+             data_type="float"
+             max_equation="100"
+             equation="B 4 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="sampler11_input_available"
+             units="percent"
+             symbol_name="Sampler11InputAvailable"
+             availability="$SubsliceMask 16 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Sampler"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="Rasterized Pixels"
+             description="The total number of rasterized pixels."
+             data_type="uint64"
+             equation="A 21 READ 4 UMUL"
+             underscore_name="rasterized_pixels"
+             units="pixels"
+             symbol_name="RasterizedPixels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer"
+             />
+    <counter name="PS FPU0 Pipe Active"
+             description="The percentage of time in which EU FPU0 pipeline was actively processing a pixel shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 15 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_fpu0_active"
+             units="percent"
+             symbol_name="PsFpu0Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pixel Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="DS Threads Dispatched"
+             description="The total number of domain shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 3 READ"
+             underscore_name="ds_threads"
+             units="threads"
+             symbol_name="DsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Domain Shader"
+             />
+    <counter name="Samples Written"
+             description="The total number of samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 26 READ 4 UMUL"
+             underscore_name="samples_written"
+             units="pixels"
+             symbol_name="SamplesWritten"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="EU Stall"
+             description="The percentage of time in which the Execution Units were stalled."
+             data_type="float"
+             max_equation="100"
+             equation="A 8 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_stall"
+             units="percent"
+             symbol_name="EuStall"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Samples Blended"
+             description="The total number of blended samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 27 READ 4 UMUL"
+             underscore_name="samples_blended"
+             units="pixels"
+             symbol_name="SamplesBlended"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="Early Depth Test Fails"
+             description="The total number of pixels dropped on early depth test."
+             data_type="uint64"
+             equation="A 23 READ 4 UMUL"
+             underscore_name="early_depth_test_fails"
+             units="pixels"
+             symbol_name="EarlyDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Early Depth Test"
+             />
+    <counter name="Slice0 Subslice0 Sampler Output Ready"
+             description="The percentage of time in which slice0 subslice0 sampler output is ready"
+             data_type="float"
+             max_equation="100"
+             equation="C 7 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="sampler00_output_ready"
+             units="percent"
+             symbol_name="Sampler00OutputReady"
+             availability="$SubsliceMask 1 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Sampler"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="Slice1 Subslice0 Sampler Output Ready"
+             description="The percentage of time in which slice1 subslice0 sampler output is ready"
+             data_type="float"
+             max_equation="100"
+             equation="C 4 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="sampler10_output_ready"
+             units="percent"
+             symbol_name="Sampler10OutputReady"
+             availability="$SubsliceMask 8 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Sampler"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="Slice0 Subslice1 Sampler Output Ready"
+             description="The percentage of time in which slice0 subslice1 sampler output is ready"
+             data_type="float"
+             max_equation="100"
+             equation="C 6 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="sampler01_output_ready"
+             units="percent"
+             symbol_name="Sampler01OutputReady"
+             availability="$SubsliceMask 2 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Sampler"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="Shader Memory Accesses"
+             description="The total number of shader memory accesses to L3."
+             data_type="uint64"
+             equation="A 32 READ"
+             underscore_name="shader_memory_accesses"
+             units="messages"
+             symbol_name="ShaderMemoryAccesses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="Slice1 Subslice1 Sampler Output Ready"
+             description="The percentage of time in which slice1 subslice1 sampler output is ready"
+             data_type="float"
+             max_equation="100"
+             equation="C 3 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="sampler11_output_ready"
+             units="percent"
+             symbol_name="Sampler11OutputReady"
+             availability="$SubsliceMask 16 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Sampler"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="HS Threads Dispatched"
+             description="The total number of hull shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 2 READ"
+             underscore_name="hs_threads"
+             units="threads"
+             symbol_name="HsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Hull Shader"
+             />
+    <counter name="L3 Shader Throughput"
+             description="The total number of GPU memory bytes transferred between shaders and L3 caches w/o URB."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ A 31 READ UADD A 32 READ UADD 64 UMUL"
+             underscore_name="l3_shader_throughput"
+             units="bytes"
+             symbol_name="L3ShaderThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_group="L3/Data Port"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Samples Killed in FS"
+             description="The total number of samples or pixels dropped in fragment shaders."
+             data_type="uint64"
+             equation="A 24 READ 4 UMUL"
+             underscore_name="samples_killed_in_ps"
+             units="pixels"
+             symbol_name="SamplesKilledInPs"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Fragment Shader"
+             />
+    <counter name="SLM Bytes Written"
+             description="The total number of GPU memory bytes written into shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 31 READ 64 UMUL"
+             underscore_name="slm_bytes_written"
+             units="bytes"
+             symbol_name="SlmBytesWritten"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Shader Atomic Memory Accesses"
+             description="The total number of shader atomic memory accesses."
+             data_type="uint64"
+             equation="A 34 READ"
+             underscore_name="shader_atomics"
+             units="messages"
+             symbol_name="ShaderAtomics"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port/Atomics"
+             />
+    <register_config type="NOA">
+        <register type="NOA" address="0x00000D04" value="0x00000200" />
+        <register type="NOA" address="0x00009884" value="0x00000000" />
+        <register type="NOA" address="0x00009888" value="0x14141400" />
+        <register type="NOA" address="0x00009888" value="0x1614000B" />
+        <register type="NOA" address="0x00009888" value="0x14161400" />
+        <register type="NOA" address="0x00009888" value="0x14341400" />
+        <register type="NOA" address="0x00009888" value="0x1634000B" />
+        <register type="NOA" address="0x00009888" value="0x14361400" />
+        <register type="NOA" address="0x00009888" value="0x14941400" />
+        <register type="NOA" address="0x00009888" value="0x1694000B" />
+        <register type="NOA" address="0x00009888" value="0x14961400" />
+        <register type="NOA" address="0x00009888" value="0x14B41400" />
+        <register type="NOA" address="0x00009888" value="0x16B4000B" />
+        <register type="NOA" address="0x00009888" value="0x14B61400" />
+        <register type="NOA" address="0x00009888" value="0x14541400" />
+        <register type="NOA" address="0x00009888" value="0x1654000B" />
+        <register type="NOA" address="0x00009888" value="0x14561400" />
+        <register type="NOA" address="0x00009888" value="0x1A6C5000" />
+        <register type="NOA" address="0x00009888" value="0x1C6C0005" />
+        <register type="NOA" address="0x00009888" value="0x186C0280" />
+        <register type="NOA" address="0x00009888" value="0x0673F000" />
+        <register type="NOA" address="0x00009888" value="0x0073C000" />
+        <register type="NOA" address="0x00009888" value="0x04742000" />
+        <register type="NOA" address="0x00009888" value="0x0A74A000" />
+        <register type="NOA" address="0x00009888" value="0x04752000" />
+        <register type="NOA" address="0x00009888" value="0x0A75A000" />
+        <register type="NOA" address="0x00009888" value="0x0C760800" />
+        <register type="NOA" address="0x00009888" value="0x0E760280" />
+        <register type="NOA" address="0x00009888" value="0x0077C000" />
+        <register type="NOA" address="0x00009888" value="0x0E77C000" />
+        <register type="NOA" address="0x00009888" value="0x14774000" />
+        <register type="NOA" address="0x00009888" value="0x1677A950" />
+        <register type="NOA" address="0x00009888" value="0x18770002" />
+        <register type="NOA" address="0x00009888" value="0x04778000" />
+        <register type="NOA" address="0x00009888" value="0x06778000" />
+        <register type="NOA" address="0x00009888" value="0x08774000" />
+        <register type="NOA" address="0x00009888" value="0x0A774000" />
+        <register type="NOA" address="0x00009888" value="0x0C77C000" />
+        <register type="NOA" address="0x00009888" value="0x0A7BA828" />
+        <register type="NOA" address="0x00009888" value="0x0C7B0002" />
+        <register type="NOA" address="0x00009888" value="0x147CE800" />
+        <register type="NOA" address="0x00009888" value="0x167C0003" />
+        <register type="NOA" address="0x00009888" value="0x087CC000" />
+        <register type="NOA" address="0x00009888" value="0x0A7C8000" />
+        <register type="NOA" address="0x00009888" value="0x1C7D8200" />
+        <register type="NOA" address="0x00009888" value="0x1E7D0002" />
+        <register type="NOA" address="0x00009888" value="0x00194000" />
+        <register type="NOA" address="0x00009888" value="0x0E194000" />
+        <register type="NOA" address="0x00009888" value="0x0C194000" />
+        <register type="NOA" address="0x00009888" value="0x00140050" />
+        <register type="NOA" address="0x00009888" value="0x06145800" />
+        <register type="NOA" address="0x00009888" value="0x10140000" />
+        <register type="NOA" address="0x00009888" value="0x18140000" />
+        <register type="NOA" address="0x00009888" value="0x1C140080" />
+        <register type="NOA" address="0x00009888" value="0x0A150001" />
+        <register type="NOA" address="0x00009888" value="0x06160043" />
+        <register type="NOA" address="0x00009888" value="0x10160000" />
+        <register type="NOA" address="0x00009888" value="0x0C160000" />
+        <register type="NOA" address="0x00009888" value="0x1A3900A0" />
+        <register type="NOA" address="0x00009888" value="0x0A394000" />
+        <register type="NOA" address="0x00009888" value="0x08345850" />
+        <register type="NOA" address="0x00009888" value="0x10340000" />
+        <register type="NOA" address="0x00009888" value="0x1C340020" />
+        <register type="NOA" address="0x00009888" value="0x08358000" />
+        <register type="NOA" address="0x00009888" value="0x04362180" />
+        <register type="NOA" address="0x00009888" value="0x10360000" />
+        <register type="NOA" address="0x00009888" value="0x0A360000" />
+        <register type="NOA" address="0x00009888" value="0x12F40140" />
+        <register type="NOA" address="0x00009888" value="0x06F48000" />
+        <register type="NOA" address="0x00009888" value="0x14F5000A" />
+        <register type="NOA" address="0x00009888" value="0x06F58000" />
+        <register type="NOA" address="0x00009888" value="0x10F60028" />
+        <register type="NOA" address="0x00009888" value="0x0EF60002" />
+        <register type="NOA" address="0x00009888" value="0x16F77C00" />
+        <register type="NOA" address="0x00009888" value="0x18F70001" />
+        <register type="NOA" address="0x00009888" value="0x04F74000" />
+        <register type="NOA" address="0x00009888" value="0x06F7C000" />
+        <register type="NOA" address="0x00009888" value="0x0CFB0280" />
+        <register type="NOA" address="0x00009888" value="0x08FB8000" />
+        <register type="NOA" address="0x00009888" value="0x16FC0280" />
+        <register type="NOA" address="0x00009888" value="0x04FC8000" />
+        <register type="NOA" address="0x00009888" value="0x1EFD2800" />
+        <register type="NOA" address="0x00009888" value="0x1CFD0008" />
+        <register type="NOA" address="0x00009888" value="0x1A992000" />
+        <register type="NOA" address="0x00009888" value="0x1C990001" />
+        <register type="NOA" address="0x00009888" value="0x06994000" />
+        <register type="NOA" address="0x00009888" value="0x0C945850" />
+        <register type="NOA" address="0x00009888" value="0x10940000" />
+        <register type="NOA" address="0x00009888" value="0x1E940000" />
+        <register type="NOA" address="0x00009888" value="0x1C940002" />
+        <register type="NOA" address="0x00009888" value="0x08952000" />
+        <register type="NOA" address="0x00009888" value="0x02962180" />
+        <register type="NOA" address="0x00009888" value="0x10960000" />
+        <register type="NOA" address="0x00009888" value="0x06960000" />
+        <register type="NOA" address="0x00009888" value="0x1CB90014" />
+        <register type="NOA" address="0x00009888" value="0x04B94000" />
+        <register type="NOA" address="0x00009888" value="0x0EB45850" />
+        <register type="NOA" address="0x00009888" value="0x10B40000" />
+        <register type="NOA" address="0x00009888" value="0x1EB40000" />
+        <register type="NOA" address="0x00009888" value="0x1AB48000" />
+        <register type="NOA" address="0x00009888" value="0x08B51000" />
+        <register type="NOA" address="0x00009888" value="0x02B60043" />
+        <register type="NOA" address="0x00009888" value="0x10B60000" />
+        <register type="NOA" address="0x00009888" value="0x04B60000" />
+        <register type="NOA" address="0x00009888" value="0x1A590A00" />
+        <register type="NOA" address="0x00009888" value="0x08594000" />
+        <register type="NOA" address="0x00009888" value="0x0A545850" />
+        <register type="NOA" address="0x00009888" value="0x10540000" />
+        <register type="NOA" address="0x00009888" value="0x1C540008" />
+        <register type="NOA" address="0x00009888" value="0x1E540000" />
+        <register type="NOA" address="0x00009888" value="0x08554000" />
+        <register type="NOA" address="0x00009888" value="0x04560043" />
+        <register type="NOA" address="0x00009888" value="0x10560000" />
+        <register type="NOA" address="0x00009888" value="0x08560000" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x2100F000" />
+        <register type="NOA" address="0x00009888" value="0x13000071" />
+        <register type="NOA" address="0x00009888" value="0x33000000" />
+        <register type="NOA" address="0x00009888" value="0x1100C000" />
+        <register type="NOA" address="0x00009888" value="0x1F00C000" />
+        <register type="NOA" address="0x00009888" value="0x2300C000" />
+        <register type="NOA" address="0x00009888" value="0x2500C000" />
+        <register type="NOA" address="0x00009888" value="0x2700C000" />
+        <register type="NOA" address="0x00009888" value="0x2900C000" />
+        <register type="NOA" address="0x00009888" value="0x2B00C000" />
+        <register type="NOA" address="0x00009888" value="0x2D00C000" />
+        <register type="NOA" address="0x00009888" value="0x2F00C000" />
+        <register type="NOA" address="0x00009888" value="0x1500C000" />
+        <register type="NOA" address="0x00009888" value="0x1700C000" />
+        <register type="NOA" address="0x00009888" value="0x1900C000" />
+        <register type="NOA" address="0x00009888" value="0x1B00C000" />
+        <register type="NOA" address="0x00009888" value="0x1D00C000" />
+        <register type="NOA" address="0x00009888" value="0x47000031" />
+        <register type="NOA" address="0x00009888" value="0x39000143" />
+        <register type="NOA" address="0x00009888" value="0x49001110" />
+        <register type="NOA" address="0x00009888" value="0x3D000C61" />
+        <register type="NOA" address="0x00009888" value="0x4B000011" />
+        <register type="NOA" address="0x00009888" value="0x3F0018C3" />
+        <register type="NOA" address="0x00009888" value="0x4D000000" />
+        <register type="NOA" address="0x00009888" value="0x41000842" />
+        <register type="NOA" address="0x00009888" value="0x35000000" />
+        <register type="NOA" address="0x00009888" value="0x31000000" />
+        <register type="NOA" address="0x00009888" value="0x43000002" />
+        <register type="NOA" address="0x00009888" value="0x3B000480" />
+    </register_config>
+    <register_config type="OA">
+        <register type="OA" address="0x00002740" value="0x00000000" />
+        <register type="OA" address="0x00002710" value="0x00000000" />
+        <register type="OA" address="0x00002714" value="0xF0800000" />
+        <register type="OA" address="0x00002720" value="0x00000000" />
+        <register type="OA" address="0x00002724" value="0x10800000" />
+        <register type="OA" address="0x00002770" value="0x00000018" />
+        <register type="OA" address="0x00002774" value="0x0000FFFC" />
+        <register type="OA" address="0x00002778" value="0x00000060" />
+        <register type="OA" address="0x0000277C" value="0x0000FFF3" />
+        <register type="OA" address="0x00002780" value="0x00000180" />
+        <register type="OA" address="0x00002784" value="0x0000FFCF" />
+        <register type="OA" address="0x00002788" value="0x00000600" />
+        <register type="OA" address="0x0000278C" value="0x0000FF3F" />
+        <register type="OA" address="0x00002790" value="0x00001800" />
+        <register type="OA" address="0x00002794" value="0x0000FCFF" />
+    </register_config>
+    <register_config type="FLEX">
+        <register type="FLEX" address="0x0000E458" value="0x00005004" />
+        <register type="FLEX" address="0x0000E558" value="0x00010003" />
+        <register type="FLEX" address="0x0000E658" value="0x00012011" />
+        <register type="FLEX" address="0x0000E45C" value="0x00051050" />
+        <register type="FLEX" address="0x0000E55C" value="0x00053052" />
+    </register_config>
+  </set>
+
+  <set name="Metric set TDL_1"
+       mdapi_supported_apis="OGL OCL IO BB"
+       underscore_name="tdl_1"
+       hw_config_guid="c46c7bb5-f26d-4634-b491-902394af85b1"
+       chipset="CNL"
+       symbol_name="TDL_1"
+       >
+    <counter name="GPU Core Clocks"
+             description="The total number of GPU core clocks elapsed during the measurement."
+             data_type="uint64"
+             equation="GPU_CLOCK 0 READ"
+             underscore_name="gpu_core_clocks"
+             units="cycles"
+             symbol_name="GpuCoreClocks"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="EU Active"
+             description="The percentage of time in which the Execution Units were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 7 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_active"
+             units="percent"
+             symbol_name="EuActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EU Both FPU Pipes Active"
+             description="The percentage of time in which both EU FPU pipelines were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 9 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_fpu_both_active"
+             units="percent"
+             symbol_name="EuFpuBothActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="SQ is full"
+             description="The percentage of time when SQ is filled above a threshold (usually 48 entries)"
+             data_type="float"
+             max_equation="100"
+             equation="C 7 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="gt_request_queue_full"
+             units="percent"
+             symbol_name="GTRequestQueueFull"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="GTI"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS Send Pipe Active"
+             description="The percentage of time in which EU send pipeline was actively processing a vertex shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 12 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="vs_send_active"
+             units="percent"
+             symbol_name="VsSendActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS FPU1 Pipe Active"
+             description="The percentage of time in which EU FPU1 pipeline was actively processing a vertex shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 11 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="vs_fpu1_active"
+             units="percent"
+             symbol_name="VsFpu1Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GS Threads Dispatched"
+             description="The total number of geometry shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 5 READ"
+             underscore_name="gs_threads"
+             units="threads"
+             symbol_name="GsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Geometry Shader"
+             />
+    <counter name="Early Hi-Depth Test Fails"
+             description="The total number of pixels dropped on early hierarchical depth test."
+             data_type="uint64"
+             equation="A 22 READ 4 UMUL"
+             underscore_name="hi_depth_test_fails"
+             units="pixels"
+             symbol_name="HiDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Hi-Depth Test"
+             />
+    <counter name="FS Both FPU Active"
+             description="The percentage of time in which fragment shaders were processed actively on the both FPUs."
+             data_type="float"
+             max_equation="100"
+             equation="A 18 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_eu_both_fpu_active"
+             units="percent"
+             symbol_name="PsEuBothFpuActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Frame Batch Draw"
+             mdapi_group="3D Pipe/Fragment Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS Threads Dispatched"
+             description="The total number of vertex shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 1 READ"
+             underscore_name="vs_threads"
+             units="threads"
+             symbol_name="VsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Vertex Shader"
+             />
+    <counter name="PS Thread Ready For Dispatch on Slice1 Subslice1"
+             description="The percentage of time in which PS thread is ready for dispatch on slice1 subslice1 thread dispatcher"
+             data_type="float"
+             max_equation="100"
+             equation="B 4 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_thread11_ready_for_dispatch"
+             units="percent"
+             symbol_name="PSThread11ReadyForDispatch"
+             availability="$SubsliceMask 16 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Thread Dispatcher"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="FS Threads Dispatched"
+             description="The total number of fragment shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 6 READ"
+             underscore_name="ps_threads"
+             units="threads"
+             symbol_name="PsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Fragment Shader"
+             />
+    <counter name="NonPS Thread Ready For Dispatch on Slice0 Subslice2"
+             description="The percentage of time in which non-PS thread is ready for dispatch on slice0 subslice2 thread dispatcher"
+             data_type="float"
+             max_equation="100"
+             equation="1  C 4 READ $GpuCoreClocks FDIV FSUB"
+             underscore_name="non_ps_thread02_ready_for_dispatch"
+             units="percent"
+             symbol_name="NonPSThread02ReadyForDispatch"
+             availability="$SubsliceMask 4 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Thread Dispatcher"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="Shader Barrier Messages"
+             description="The total number of shader barrier messages."
+             data_type="uint64"
+             equation="A 35 READ"
+             underscore_name="shader_barriers"
+             units="messages"
+             symbol_name="ShaderBarriers"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Barrier"
+             />
+    <counter name="Sampler Texels"
+             description="The total number of texels seen on input (with 2x2 accuracy) in all sampler units."
+             data_type="uint64"
+             equation="A 28 READ 4 UMUL"
+             underscore_name="sampler_texels"
+             units="texels"
+             symbol_name="SamplerTexels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Input"
+             />
+    <counter name="Pixels Failing Tests"
+             description="The total number of pixels dropped on post-FS alpha, stencil, or depth tests."
+             data_type="uint64"
+             equation="A 25 READ 4 UMUL"
+             underscore_name="pixels_failing_post_ps_tests"
+             units="pixels"
+             symbol_name="PixelsFailingPostPsTests"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="NonPS Thread Ready For Dispatch on Slice1 Subslice0"
+             description="The percentage of time in which non-PS thread is ready for dispatch on slice1 subslice0 thread dispatcher"
+             data_type="float"
+             max_equation="100"
+             equation="1  C 5 READ $GpuCoreClocks FDIV FSUB"
+             underscore_name="non_ps_thread10_ready_for_dispatch"
+             units="percent"
+             symbol_name="NonPSThread10ReadyForDispatch"
+             availability="$SubsliceMask 8 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Thread Dispatcher"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="GPU Time Elapsed"
+             description="Time elapsed on the GPU during the measurement."
+             data_type="uint64"
+             equation="GPU_TIME 0 READ 1000000000 UMUL $GpuTimestampFrequency UDIV"
+             underscore_name="gpu_time"
+             units="ns"
+             symbol_name="GpuTime"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="AVG GPU Core Frequency"
+             description="Average GPU Core Frequency in the measurement."
+             data_type="uint64"
+             max_equation="$GpuMaxFrequency"
+             equation="$GpuCoreClocks 1000000000 UMUL $GpuTime UDIV"
+             underscore_name="avg_gpu_core_frequency"
+             units="hz"
+             symbol_name="AvgGpuCoreFrequency"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Sampler Texels Misses"
+             description="The total number of texels lookups (with 2x2 accuracy) that missed L1 sampler cache."
+             data_type="uint64"
+             equation="A 29 READ 4 UMUL"
+             underscore_name="sampler_texel_misses"
+             units="texels"
+             symbol_name="SamplerTexelMisses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Batch Frame Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Cache"
+             />
+    <counter name="NonPS Thread Ready For Dispatch on Slice0 Subslice0"
+             description="The percentage of time in which non-PS thread is ready for dispatch on slice0 subslice0 thread dispatcher"
+             data_type="float"
+             max_equation="100"
+             equation="1  C 2 READ $GpuCoreClocks FDIV FSUB"
+             underscore_name="non_ps_thread00_ready_for_dispatch"
+             units="percent"
+             symbol_name="NonPSThread00ReadyForDispatch"
+             availability="$SubsliceMask 1 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Thread Dispatcher"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="CS Threads Dispatched"
+             description="The total number of compute shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 4 READ"
+             underscore_name="cs_threads"
+             units="threads"
+             symbol_name="CsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Compute Shader"
+             />
+    <counter name="SLM Bytes Read"
+             description="The total number of GPU memory bytes read from shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ 64 UMUL"
+             underscore_name="slm_bytes_read"
+             units="bytes"
+             symbol_name="SlmBytesRead"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="PS FPU1 Pipe Active"
+             description="The percentage of time in which EU FPU1 pipeline was actively processing a pixel shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 16 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_fpu1_active"
+             units="percent"
+             symbol_name="PsFpu1Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pixel Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="PS Send Pipeline Active"
+             description="The percentage of time in which EU send pipeline was actively processing a pixel shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 17 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_send_active"
+             units="percent"
+             symbol_name="PsSendActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pixel Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS FPU0 Pipe Active"
+             description="The percentage of time in which EU FPU0 pipeline was actively processing a vertex shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 10 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="vs_fpu0_active"
+             units="percent"
+             symbol_name="VsFpu0Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GPU Busy"
+             description="The percentage of time in which the GPU has been processing GPU commands."
+             data_type="float"
+             max_equation="100"
+             equation="A 0 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="gpu_busy"
+             units="percent"
+             symbol_name="GpuBusy"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="PS Thread Ready For Dispatch on Slice0 Subslice0"
+             description="The percentage of time in which PS thread is ready for dispatch on slice0 subslice0 thread dispatcher"
+             data_type="float"
+             max_equation="100"
+             equation="B 0 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_thread00_ready_for_dispatch"
+             units="percent"
+             symbol_name="PSThread00ReadyForDispatch"
+             availability="$SubsliceMask 1 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Thread Dispatcher"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="Rasterized Pixels"
+             description="The total number of rasterized pixels."
+             data_type="uint64"
+             equation="A 21 READ 4 UMUL"
+             underscore_name="rasterized_pixels"
+             units="pixels"
+             symbol_name="RasterizedPixels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer"
+             />
+    <counter name="PS FPU0 Pipe Active"
+             description="The percentage of time in which EU FPU0 pipeline was actively processing a pixel shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 15 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_fpu0_active"
+             units="percent"
+             symbol_name="PsFpu0Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pixel Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="DS Threads Dispatched"
+             description="The total number of domain shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 3 READ"
+             underscore_name="ds_threads"
+             units="threads"
+             symbol_name="DsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Domain Shader"
+             />
+    <counter name="Samples Written"
+             description="The total number of samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 26 READ 4 UMUL"
+             underscore_name="samples_written"
+             units="pixels"
+             symbol_name="SamplesWritten"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="EU Stall"
+             description="The percentage of time in which the Execution Units were stalled."
+             data_type="float"
+             max_equation="100"
+             equation="A 8 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_stall"
+             units="percent"
+             symbol_name="EuStall"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Samples Blended"
+             description="The total number of blended samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 27 READ 4 UMUL"
+             underscore_name="samples_blended"
+             units="pixels"
+             symbol_name="SamplesBlended"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="Early Depth Test Fails"
+             description="The total number of pixels dropped on early depth test."
+             data_type="uint64"
+             equation="A 23 READ 4 UMUL"
+             underscore_name="early_depth_test_fails"
+             units="pixels"
+             symbol_name="EarlyDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Early Depth Test"
+             />
+    <counter name="PS Thread Ready For Dispatch on Slice0 Subslice1"
+             description="The percentage of time in which PS thread is ready for dispatch on slice0 subslice1 thread dispatcher"
+             data_type="float"
+             max_equation="100"
+             equation="B 1 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_thread01_ready_for_dispatch"
+             units="percent"
+             symbol_name="PSThread01ReadyForDispatch"
+             availability="$SubsliceMask 2 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Thread Dispatcher"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="PS Thread Ready For Dispatch on Slice1 Subslice0"
+             description="The percentage of time in which PS thread is ready for dispatch on slice1 subslice0 thread dispatcher"
+             data_type="float"
+             max_equation="100"
+             equation="B 3 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_thread10_ready_for_dispatch"
+             units="percent"
+             symbol_name="PSThread10ReadyForDispatch"
+             availability="$SubsliceMask 8 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Thread Dispatcher"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="NonPS Thread Ready For Dispatch on Slice1 Subslice1"
+             description="The percentage of time in which non-PS thread is ready for dispatch on slice1 subslice1 thread dispatcher"
+             data_type="float"
+             max_equation="100"
+             equation="1  C 6 READ $GpuCoreClocks FDIV FSUB"
+             underscore_name="non_ps_thread11_ready_for_dispatch"
+             units="percent"
+             symbol_name="NonPSThread11ReadyForDispatch"
+             availability="$SubsliceMask 16 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Thread Dispatcher"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="Shader Memory Accesses"
+             description="The total number of shader memory accesses to L3."
+             data_type="uint64"
+             equation="A 32 READ"
+             underscore_name="shader_memory_accesses"
+             units="messages"
+             symbol_name="ShaderMemoryAccesses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="HS Threads Dispatched"
+             description="The total number of hull shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 2 READ"
+             underscore_name="hs_threads"
+             units="threads"
+             symbol_name="HsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Hull Shader"
+             />
+    <counter name="PS Thread Ready For Dispatch on Slice0 Subslice2"
+             description="The percentage of time in which PS thread is ready for dispatch on slice0 subslice2 thread dispatcher"
+             data_type="float"
+             max_equation="100"
+             equation="B 2 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_thread02_ready_for_dispatch"
+             units="percent"
+             symbol_name="PSThread02ReadyForDispatch"
+             availability="$SubsliceMask 4 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Thread Dispatcher"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="L3 Shader Throughput"
+             description="The total number of GPU memory bytes transferred between shaders and L3 caches w/o URB."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ A 31 READ UADD A 32 READ UADD 64 UMUL"
+             underscore_name="l3_shader_throughput"
+             units="bytes"
+             symbol_name="L3ShaderThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_group="L3/Data Port"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Samples Killed in FS"
+             description="The total number of samples or pixels dropped in fragment shaders."
+             data_type="uint64"
+             equation="A 24 READ 4 UMUL"
+             underscore_name="samples_killed_in_ps"
+             units="pixels"
+             symbol_name="SamplesKilledInPs"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Fragment Shader"
+             />
+    <counter name="SLM Bytes Written"
+             description="The total number of GPU memory bytes written into shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 31 READ 64 UMUL"
+             underscore_name="slm_bytes_written"
+             units="bytes"
+             symbol_name="SlmBytesWritten"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Shader Atomic Memory Accesses"
+             description="The total number of shader atomic memory accesses."
+             data_type="uint64"
+             equation="A 34 READ"
+             underscore_name="shader_atomics"
+             units="messages"
+             symbol_name="ShaderAtomics"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port/Atomics"
+             />
+    <counter name="NonPS Thread Ready For Dispatch on Slice0 Subslice1"
+             description="The percentage of time in which non-PS thread is ready for dispatch on slice0 subslice1 thread dispatcher"
+             data_type="float"
+             max_equation="100"
+             equation="1  C 3 READ $GpuCoreClocks FDIV FSUB"
+             underscore_name="non_ps_thread01_ready_for_dispatch"
+             units="percent"
+             symbol_name="NonPSThread01ReadyForDispatch"
+             availability="$SubsliceMask 2 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Thread Dispatcher"
+             mdapi_hw_unit_type="subslice"
+             />
+    <register_config type="NOA">
+        <register type="NOA" address="0x00000D04" value="0x00000200" />
+        <register type="NOA" address="0x00009884" value="0x00000000" />
+        <register type="NOA" address="0x00009888" value="0x12180000" />
+        <register type="NOA" address="0x00009888" value="0x12380000" />
+        <register type="NOA" address="0x00009888" value="0x12980000" />
+        <register type="NOA" address="0x00009888" value="0x12B80000" />
+        <register type="NOA" address="0x00009888" value="0x12580000" />
+        <register type="NOA" address="0x00009888" value="0x1A6C5000" />
+        <register type="NOA" address="0x00009888" value="0x1C6C0005" />
+        <register type="NOA" address="0x00009888" value="0x186C2800" />
+        <register type="NOA" address="0x00009888" value="0x0673F000" />
+        <register type="NOA" address="0x00009888" value="0x02733000" />
+        <register type="NOA" address="0x00009888" value="0x0474A000" />
+        <register type="NOA" address="0x00009888" value="0x0A748000" />
+        <register type="NOA" address="0x00009888" value="0x0475A000" />
+        <register type="NOA" address="0x00009888" value="0x0A758000" />
+        <register type="NOA" address="0x00009888" value="0x0C762800" />
+        <register type="NOA" address="0x00009888" value="0x0E760200" />
+        <register type="NOA" address="0x00009888" value="0x0077C000" />
+        <register type="NOA" address="0x00009888" value="0x0E77C000" />
+        <register type="NOA" address="0x00009888" value="0x14774000" />
+        <register type="NOA" address="0x00009888" value="0x1677A950" />
+        <register type="NOA" address="0x00009888" value="0x18770002" />
+        <register type="NOA" address="0x00009888" value="0x0277C000" />
+        <register type="NOA" address="0x00009888" value="0x04774000" />
+        <register type="NOA" address="0x00009888" value="0x06774000" />
+        <register type="NOA" address="0x00009888" value="0x08778000" />
+        <register type="NOA" address="0x00009888" value="0x0A778000" />
+        <register type="NOA" address="0x00009888" value="0x0A7BA802" />
+        <register type="NOA" address="0x00009888" value="0x0C7B0002" />
+        <register type="NOA" address="0x00009888" value="0x087B8000" />
+        <register type="NOA" address="0x00009888" value="0x147CE800" />
+        <register type="NOA" address="0x00009888" value="0x167C0003" />
+        <register type="NOA" address="0x00009888" value="0x047C8000" />
+        <register type="NOA" address="0x00009888" value="0x067CC000" />
+        <register type="NOA" address="0x00009888" value="0x1C7D8008" />
+        <register type="NOA" address="0x00009888" value="0x1E7D0002" />
+        <register type="NOA" address="0x00009888" value="0x00194000" />
+        <register type="NOA" address="0x00009888" value="0x0E194000" />
+        <register type="NOA" address="0x00009888" value="0x02194000" />
+        <register type="NOA" address="0x00009888" value="0x18148000" />
+        <register type="NOA" address="0x00009888" value="0x1C140200" />
+        <register type="NOA" address="0x00009888" value="0x1A142000" />
+        <register type="NOA" address="0x00009888" value="0x0015C000" />
+        <register type="NOA" address="0x00009888" value="0x0A150002" />
+        <register type="NOA" address="0x00009888" value="0x00168000" />
+        <register type="NOA" address="0x00009888" value="0x0E168000" />
+        <register type="NOA" address="0x00009888" value="0x02168000" />
+        <register type="NOA" address="0x00009888" value="0x0018A1A3" />
+        <register type="NOA" address="0x00009888" value="0x0618A500" />
+        <register type="NOA" address="0x00009888" value="0x10180000" />
+        <register type="NOA" address="0x00009888" value="0x1A3900A0" />
+        <register type="NOA" address="0x00009888" value="0x04394000" />
+        <register type="NOA" address="0x00009888" value="0x1C342800" />
+        <register type="NOA" address="0x00009888" value="0x1A348000" />
+        <register type="NOA" address="0x00009888" value="0x0A35000C" />
+        <register type="NOA" address="0x00009888" value="0x08351000" />
+        <register type="NOA" address="0x00009888" value="0x1A360140" />
+        <register type="NOA" address="0x00009888" value="0x04368000" />
+        <register type="NOA" address="0x00009888" value="0x0838A5A3" />
+        <register type="NOA" address="0x00009888" value="0x023800A1" />
+        <register type="NOA" address="0x00009888" value="0x10380000" />
+        <register type="NOA" address="0x00009888" value="0x12F40140" />
+        <register type="NOA" address="0x00009888" value="0x08F42000" />
+        <register type="NOA" address="0x00009888" value="0x14F5000A" />
+        <register type="NOA" address="0x00009888" value="0x08F52000" />
+        <register type="NOA" address="0x00009888" value="0x10F60028" />
+        <register type="NOA" address="0x00009888" value="0x0EF60008" />
+        <register type="NOA" address="0x00009888" value="0x16F77C00" />
+        <register type="NOA" address="0x00009888" value="0x18F70001" />
+        <register type="NOA" address="0x00009888" value="0x08F7C000" />
+        <register type="NOA" address="0x00009888" value="0x0AF74000" />
+        <register type="NOA" address="0x00009888" value="0x0CFB0280" />
+        <register type="NOA" address="0x00009888" value="0x0AFB0020" />
+        <register type="NOA" address="0x00009888" value="0x16FC0280" />
+        <register type="NOA" address="0x00009888" value="0x0AFC8000" />
+        <register type="NOA" address="0x00009888" value="0x1EFD2800" />
+        <register type="NOA" address="0x00009888" value="0x1CFD0200" />
+        <register type="NOA" address="0x00009888" value="0x1A992000" />
+        <register type="NOA" address="0x00009888" value="0x1C990001" />
+        <register type="NOA" address="0x00009888" value="0x08994000" />
+        <register type="NOA" address="0x00009888" value="0x1E940028" />
+        <register type="NOA" address="0x00009888" value="0x1C940008" />
+        <register type="NOA" address="0x00009888" value="0x0A9500C0" />
+        <register type="NOA" address="0x00009888" value="0x08954000" />
+        <register type="NOA" address="0x00009888" value="0x1A964000" />
+        <register type="NOA" address="0x00009888" value="0x1C960002" />
+        <register type="NOA" address="0x00009888" value="0x08968000" />
+        <register type="NOA" address="0x00009888" value="0x0C98A5A3" />
+        <register type="NOA" address="0x00009888" value="0x049800A1" />
+        <register type="NOA" address="0x00009888" value="0x10980000" />
+        <register type="NOA" address="0x00009888" value="0x1CB90014" />
+        <register type="NOA" address="0x00009888" value="0x0AB94000" />
+        <register type="NOA" address="0x00009888" value="0x1EB40280" />
+        <register type="NOA" address="0x00009888" value="0x1CB40020" />
+        <register type="NOA" address="0x00009888" value="0x0AB50300" />
+        <register type="NOA" address="0x00009888" value="0x08B58000" />
+        <register type="NOA" address="0x00009888" value="0x1CB60028" />
+        <register type="NOA" address="0x00009888" value="0x0AB68000" />
+        <register type="NOA" address="0x00009888" value="0x0EB8A5A3" />
+        <register type="NOA" address="0x00009888" value="0x04B8A100" />
+        <register type="NOA" address="0x00009888" value="0x10B80000" />
+        <register type="NOA" address="0x00009888" value="0x1A590A00" />
+        <register type="NOA" address="0x00009888" value="0x06594000" />
+        <register type="NOA" address="0x00009888" value="0x1C548002" />
+        <register type="NOA" address="0x00009888" value="0x1E540002" />
+        <register type="NOA" address="0x00009888" value="0x0A550030" />
+        <register type="NOA" address="0x00009888" value="0x08552000" />
+        <register type="NOA" address="0x00009888" value="0x1A561400" />
+        <register type="NOA" address="0x00009888" value="0x06568000" />
+        <register type="NOA" address="0x00009888" value="0x0A58A5A3" />
+        <register type="NOA" address="0x00009888" value="0x0258A100" />
+        <register type="NOA" address="0x00009888" value="0x10580000" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x2100F000" />
+        <register type="NOA" address="0x00009888" value="0x1D000071" />
+        <register type="NOA" address="0x00009888" value="0x33000000" />
+        <register type="NOA" address="0x00009888" value="0x1100C000" />
+        <register type="NOA" address="0x00009888" value="0x1F00C000" />
+        <register type="NOA" address="0x00009888" value="0x2300C000" />
+        <register type="NOA" address="0x00009888" value="0x2500C000" />
+        <register type="NOA" address="0x00009888" value="0x2700C000" />
+        <register type="NOA" address="0x00009888" value="0x2900C000" />
+        <register type="NOA" address="0x00009888" value="0x2B00C000" />
+        <register type="NOA" address="0x00009888" value="0x2D00C000" />
+        <register type="NOA" address="0x00009888" value="0x2F00C000" />
+        <register type="NOA" address="0x00009888" value="0x1300C000" />
+        <register type="NOA" address="0x00009888" value="0x1500C000" />
+        <register type="NOA" address="0x00009888" value="0x1700C000" />
+        <register type="NOA" address="0x00009888" value="0x1900C000" />
+        <register type="NOA" address="0x00009888" value="0x1B00C000" />
+        <register type="NOA" address="0x00009888" value="0x47000111" />
+        <register type="NOA" address="0x00009888" value="0x39000421" />
+        <register type="NOA" address="0x00009888" value="0x49001300" />
+        <register type="NOA" address="0x00009888" value="0x3D00042B" />
+        <register type="NOA" address="0x00009888" value="0x4B000011" />
+        <register type="NOA" address="0x00009888" value="0x3F001081" />
+        <register type="NOA" address="0x00009888" value="0x4D000000" />
+        <register type="NOA" address="0x00009888" value="0x41000000" />
+        <register type="NOA" address="0x00009888" value="0x35000000" />
+        <register type="NOA" address="0x00009888" value="0x31000000" />
+        <register type="NOA" address="0x00009888" value="0x43000000" />
+        <register type="NOA" address="0x00009888" value="0x3B000004" />
+    </register_config>
+    <register_config type="OA">
+        <register type="OA" address="0x00002740" value="0x00000000" />
+        <register type="OA" address="0x00002710" value="0x00000000" />
+        <register type="OA" address="0x00002714" value="0xF0800000" />
+        <register type="OA" address="0x00002720" value="0x00000000" />
+        <register type="OA" address="0x00002724" value="0x10800000" />
+        <register type="OA" address="0x00002770" value="0x00000000" />
+        <register type="OA" address="0x00002774" value="0x0000FFFC" />
+        <register type="OA" address="0x00002778" value="0x00000000" />
+        <register type="OA" address="0x0000277C" value="0x0000FFF3" />
+        <register type="OA" address="0x00002780" value="0x00000000" />
+        <register type="OA" address="0x00002784" value="0x0000FFCF" />
+        <register type="OA" address="0x00002788" value="0x00000000" />
+        <register type="OA" address="0x0000278C" value="0x0000FF3F" />
+        <register type="OA" address="0x00002790" value="0x00000000" />
+        <register type="OA" address="0x00002794" value="0x0000FCFF" />
+    </register_config>
+    <register_config type="FLEX">
+        <register type="FLEX" address="0x0000E458" value="0x00005004" />
+        <register type="FLEX" address="0x0000E558" value="0x00010003" />
+        <register type="FLEX" address="0x0000E658" value="0x00012011" />
+        <register type="FLEX" address="0x0000E45C" value="0x00051050" />
+        <register type="FLEX" address="0x0000E55C" value="0x00053052" />
+    </register_config>
+  </set>
+
+  <set name="Metric set TDL_2"
+       mdapi_supported_apis="OGL OCL IO BB"
+       underscore_name="tdl_2"
+       hw_config_guid="aa2d6ab4-99b3-42f8-94d6-4a9aa2baedaf"
+       chipset="CNL"
+       symbol_name="TDL_2"
+       >
+    <counter name="GPU Core Clocks"
+             description="The total number of GPU core clocks elapsed during the measurement."
+             data_type="uint64"
+             equation="GPU_CLOCK 0 READ"
+             underscore_name="gpu_core_clocks"
+             units="cycles"
+             symbol_name="GpuCoreClocks"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="EU Active"
+             description="The percentage of time in which the Execution Units were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 7 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_active"
+             units="percent"
+             symbol_name="EuActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="EU Both FPU Pipes Active"
+             description="The percentage of time in which both EU FPU pipelines were actively processing."
+             data_type="float"
+             max_equation="100"
+             equation="A 9 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_fpu_both_active"
+             units="percent"
+             symbol_name="EuFpuBothActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pipes"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Thread Header Ready on Slice1 Subslice1 Port 1"
+             description="The percentage of time in which thread header is ready on slice1 subslice1 thread dispatcher port 1"
+             data_type="float"
+             max_equation="100"
+             equation="C 7 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="thread_header11_ready_port1"
+             units="percent"
+             symbol_name="ThreadHeader11ReadyPort1"
+             availability="$SubsliceMask 16 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Thread Dispatcher"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="VS Send Pipe Active"
+             description="The percentage of time in which EU send pipeline was actively processing a vertex shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 12 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="vs_send_active"
+             units="percent"
+             symbol_name="VsSendActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS FPU1 Pipe Active"
+             description="The percentage of time in which EU FPU1 pipeline was actively processing a vertex shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 11 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="vs_fpu1_active"
+             units="percent"
+             symbol_name="VsFpu1Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GS Threads Dispatched"
+             description="The total number of geometry shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 5 READ"
+             underscore_name="gs_threads"
+             units="threads"
+             symbol_name="GsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Geometry Shader"
+             />
+    <counter name="Early Hi-Depth Test Fails"
+             description="The total number of pixels dropped on early hierarchical depth test."
+             data_type="uint64"
+             equation="A 22 READ 4 UMUL"
+             underscore_name="hi_depth_test_fails"
+             units="pixels"
+             symbol_name="HiDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Hi-Depth Test"
+             />
+    <counter name="FS Both FPU Active"
+             description="The percentage of time in which fragment shaders were processed actively on the both FPUs."
+             data_type="float"
+             max_equation="100"
+             equation="A 18 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_eu_both_fpu_active"
+             units="percent"
+             symbol_name="PsEuBothFpuActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Frame Batch Draw"
+             mdapi_group="3D Pipe/Fragment Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS Threads Dispatched"
+             description="The total number of vertex shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 1 READ"
+             underscore_name="vs_threads"
+             units="threads"
+             symbol_name="VsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Vertex Shader"
+             />
+    <counter name="FS Threads Dispatched"
+             description="The total number of fragment shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 6 READ"
+             underscore_name="ps_threads"
+             units="threads"
+             symbol_name="PsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Fragment Shader"
+             />
+    <counter name="Shader Barrier Messages"
+             description="The total number of shader barrier messages."
+             data_type="uint64"
+             equation="A 35 READ"
+             underscore_name="shader_barriers"
+             units="messages"
+             symbol_name="ShaderBarriers"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Barrier"
+             />
+    <counter name="Sampler Texels"
+             description="The total number of texels seen on input (with 2x2 accuracy) in all sampler units."
+             data_type="uint64"
+             equation="A 28 READ 4 UMUL"
+             underscore_name="sampler_texels"
+             units="texels"
+             symbol_name="SamplerTexels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Input"
+             />
+    <counter name="Pixels Failing Tests"
+             description="The total number of pixels dropped on post-FS alpha, stencil, or depth tests."
+             data_type="uint64"
+             equation="A 25 READ 4 UMUL"
+             underscore_name="pixels_failing_post_ps_tests"
+             units="pixels"
+             symbol_name="PixelsFailingPostPsTests"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="Thread Header Ready on Slice0 Subslice1 Port 0"
+             description="The percentage of time in which thread header is ready on slice0 subslice1 thread dispatcher port 0"
+             data_type="float"
+             max_equation="100"
+             equation="C 0 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="thread_header01_ready_port0"
+             units="percent"
+             symbol_name="ThreadHeader01ReadyPort0"
+             availability="$SubsliceMask 2 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Thread Dispatcher"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="Thread Header Ready on Slice0 Subslice1 Port 1"
+             description="The percentage of time in which thread header is ready on slice0 subslice1 thread dispatcher port 1"
+             data_type="float"
+             max_equation="100"
+             equation="C 1 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="thread_header01_ready_port1"
+             units="percent"
+             symbol_name="ThreadHeader01ReadyPort1"
+             availability="$SubsliceMask 2 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Thread Dispatcher"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="GPU Time Elapsed"
+             description="Time elapsed on the GPU during the measurement."
+             data_type="uint64"
+             equation="GPU_TIME 0 READ 1000000000 UMUL $GpuTimestampFrequency UDIV"
+             underscore_name="gpu_time"
+             units="ns"
+             symbol_name="GpuTime"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="AVG GPU Core Frequency"
+             description="Average GPU Core Frequency in the measurement."
+             data_type="uint64"
+             max_equation="$GpuMaxFrequency"
+             equation="$GpuCoreClocks 1000000000 UMUL $GpuTime UDIV"
+             underscore_name="avg_gpu_core_frequency"
+             units="hz"
+             symbol_name="AvgGpuCoreFrequency"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Sampler Texels Misses"
+             description="The total number of texels lookups (with 2x2 accuracy) that missed L1 sampler cache."
+             data_type="uint64"
+             equation="A 29 READ 4 UMUL"
+             underscore_name="sampler_texel_misses"
+             units="texels"
+             symbol_name="SamplerTexelMisses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Batch Frame Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="Sampler/Sampler Cache"
+             />
+    <counter name="CS Threads Dispatched"
+             description="The total number of compute shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 4 READ"
+             underscore_name="cs_threads"
+             units="threads"
+             symbol_name="CsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Compute Shader"
+             />
+    <counter name="SQ is full"
+             description="The percentage of time when SQ is filled above a threshold (usually 48 entries)"
+             data_type="float"
+             max_equation="100"
+             equation="B 5 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="gt_request_queue_full"
+             units="percent"
+             symbol_name="GTRequestQueueFull"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="GTI"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="SLM Bytes Read"
+             description="The total number of GPU memory bytes read from shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ 64 UMUL"
+             underscore_name="slm_bytes_read"
+             units="bytes"
+             symbol_name="SlmBytesRead"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Thread Header Ready on Slice1 Subslice1 Port 0"
+             description="The percentage of time in which thread header is ready on slice1 subslice1 thread dispatcher port 0"
+             data_type="float"
+             max_equation="100"
+             equation="C 6 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="thread_header11_ready_port0"
+             units="percent"
+             symbol_name="ThreadHeader11ReadyPort0"
+             availability="$SubsliceMask 16 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Thread Dispatcher"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="PS FPU1 Pipe Active"
+             description="The percentage of time in which EU FPU1 pipeline was actively processing a pixel shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 16 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_fpu1_active"
+             units="percent"
+             symbol_name="PsFpu1Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pixel Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="PS Send Pipeline Active"
+             description="The percentage of time in which EU send pipeline was actively processing a pixel shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 17 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_send_active"
+             units="percent"
+             symbol_name="PsSendActive"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pixel Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="VS FPU0 Pipe Active"
+             description="The percentage of time in which EU FPU0 pipeline was actively processing a vertex shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 10 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="vs_fpu0_active"
+             units="percent"
+             symbol_name="VsFpu0Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Vertex Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="GPU Busy"
+             description="The percentage of time in which the GPU has been processing GPU commands."
+             data_type="float"
+             max_equation="100"
+             equation="A 0 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="gpu_busy"
+             units="percent"
+             symbol_name="GpuBusy"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Thread Header Ready on Slice0 Subslice2 Port 1"
+             description="The percentage of time in which thread header is ready on slice0 subslice2 thread dispatcher port 1"
+             data_type="float"
+             max_equation="100"
+             equation="C 3 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="thread_header02_ready_port1"
+             units="percent"
+             symbol_name="ThreadHeader02ReadyPort1"
+             availability="$SubsliceMask 4 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Thread Dispatcher"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="Thread Header Ready on Slice0 Subslice2 Port 0"
+             description="The percentage of time in which thread header is ready on slice0 subslice2 thread dispatcher port 0"
+             data_type="float"
+             max_equation="100"
+             equation="C 2 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="thread_header02_ready_port0"
+             units="percent"
+             symbol_name="ThreadHeader02ReadyPort0"
+             availability="$SubsliceMask 4 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Thread Dispatcher"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="Rasterized Pixels"
+             description="The total number of rasterized pixels."
+             data_type="uint64"
+             equation="A 21 READ 4 UMUL"
+             underscore_name="rasterized_pixels"
+             units="pixels"
+             symbol_name="RasterizedPixels"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer"
+             />
+    <counter name="PS FPU0 Pipe Active"
+             description="The percentage of time in which EU FPU0 pipeline was actively processing a pixel shader instruction."
+             data_type="float"
+             max_equation="100"
+             equation="A 15 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="ps_fpu0_active"
+             units="percent"
+             symbol_name="PsFpu0Active"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview System Frame Batch Draw"
+             mdapi_group="EU Array/Pixel Shader"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="DS Threads Dispatched"
+             description="The total number of domain shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 3 READ"
+             underscore_name="ds_threads"
+             units="threads"
+             symbol_name="DsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Domain Shader"
+             />
+    <counter name="Samples Written"
+             description="The total number of samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 26 READ 4 UMUL"
+             underscore_name="samples_written"
+             units="pixels"
+             symbol_name="SamplesWritten"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="Thread Header Ready on Slice1 Subslice0 Port 1"
+             description="The percentage of time in which thread header is ready on slice1 subslice0 thread dispatcher port 1"
+             data_type="float"
+             max_equation="100"
+             equation="C 5 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="thread_header10_ready_port1"
+             units="percent"
+             symbol_name="ThreadHeader10ReadyPort1"
+             availability="$SubsliceMask 8 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Thread Dispatcher"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="EU Stall"
+             description="The percentage of time in which the Execution Units were stalled."
+             data_type="float"
+             max_equation="100"
+             equation="A 8 READ $EuCoresTotalCount UDIV 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="eu_stall"
+             units="percent"
+             symbol_name="EuStall"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Overview System Frame Batch Draw"
+             mdapi_group="EU Array"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Samples Blended"
+             description="The total number of blended samples or pixels written to all render targets."
+             data_type="uint64"
+             equation="A 27 READ 4 UMUL"
+             underscore_name="samples_blended"
+             units="pixels"
+             symbol_name="SamplesBlended"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Output Merger"
+             />
+    <counter name="Early Depth Test Fails"
+             description="The total number of pixels dropped on early depth test."
+             data_type="uint64"
+             equation="A 23 READ 4 UMUL"
+             underscore_name="early_depth_test_fails"
+             units="pixels"
+             symbol_name="EarlyDepthTestFails"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Rasterizer/Early Depth Test"
+             />
+    <counter name="Shader Memory Accesses"
+             description="The total number of shader memory accesses to L3."
+             data_type="uint64"
+             equation="A 32 READ"
+             underscore_name="shader_memory_accesses"
+             units="messages"
+             symbol_name="ShaderMemoryAccesses"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port"
+             />
+    <counter name="HS Threads Dispatched"
+             description="The total number of hull shader hardware threads dispatched."
+             data_type="uint64"
+             equation="A 2 READ"
+             underscore_name="hs_threads"
+             units="threads"
+             symbol_name="HsThreads"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="EU Array/Hull Shader"
+             />
+    <counter name="L3 Shader Throughput"
+             description="The total number of GPU memory bytes transferred between shaders and L3 caches w/o URB."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 30 READ A 31 READ UADD A 32 READ UADD 64 UMUL"
+             underscore_name="l3_shader_throughput"
+             units="bytes"
+             symbol_name="L3ShaderThroughput"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier2 Frame Batch Draw"
+             mdapi_group="L3/Data Port"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Samples Killed in FS"
+             description="The total number of samples or pixels dropped in fragment shaders."
+             data_type="uint64"
+             equation="A 24 READ 4 UMUL"
+             underscore_name="samples_killed_in_ps"
+             units="pixels"
+             symbol_name="SamplesKilledInPs"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier4 Overview Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="3D Pipe/Fragment Shader"
+             />
+    <counter name="SLM Bytes Written"
+             description="The total number of GPU memory bytes written into shared local memory."
+             data_type="uint64"
+             max_equation="$GpuCoreClocks 64 UMUL $EuSubslicesTotalCount UMUL"
+             equation="A 31 READ 64 UMUL"
+             underscore_name="slm_bytes_written"
+             units="bytes"
+             symbol_name="SlmBytesWritten"
+             semantic_type="throughput"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_group="L3/Data Port/SLM"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="Thread Header Ready on Slice1 Subslice0 Port 0"
+             description="The percentage of time in which thread header is ready on slice1 subslice0 thread dispatcher port 0"
+             data_type="float"
+             max_equation="100"
+             equation="C 4 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="thread_header10_ready_port0"
+             units="percent"
+             symbol_name="ThreadHeader10ReadyPort0"
+             availability="$SubsliceMask 8 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Thread Dispatcher"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="Thread Header Ready on Slice0 Subslice0 Port 0"
+             description="The percentage of time in which thread header is ready on slice0 subslice0 thread dispatcher port 0"
+             data_type="float"
+             max_equation="100"
+             equation="B 6 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="thread_header00_ready_port0"
+             units="percent"
+             symbol_name="ThreadHeader00ReadyPort0"
+             availability="$SubsliceMask 1 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Thread Dispatcher"
+             mdapi_hw_unit_type="subslice"
+             />
+    <counter name="Shader Atomic Memory Accesses"
+             description="The total number of shader atomic memory accesses."
+             data_type="uint64"
+             equation="A 34 READ"
+             underscore_name="shader_atomics"
+             units="messages"
+             symbol_name="ShaderAtomics"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="L3/Data Port/Atomics"
+             />
+    <counter name="Thread Header Ready on Slice0 Subslice0 Port 1"
+             description="The percentage of time in which thread header is ready on slice0 subslice0 thread dispatcher port 1"
+             data_type="float"
+             max_equation="100"
+             equation="B 7 READ 100 UMUL $GpuCoreClocks FDIV"
+             underscore_name="thread_header00_ready_port1"
+             units="percent"
+             symbol_name="ThreadHeader00ReadyPort1"
+             availability="$SubsliceMask 1 AND"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier3 Overview Frame Batch Draw"
+             mdapi_group="GPU/Thread Dispatcher"
+             mdapi_hw_unit_type="subslice"
+             />
+    <register_config type="NOA">
+        <register type="NOA" address="0x00000D04" value="0x00000200" />
+        <register type="NOA" address="0x00009884" value="0x00000000" />
+        <register type="NOA" address="0x00009888" value="0x12184D60" />
+        <register type="NOA" address="0x00009888" value="0x12384D60" />
+        <register type="NOA" address="0x00009888" value="0x12984D60" />
+        <register type="NOA" address="0x00009888" value="0x12B84D60" />
+        <register type="NOA" address="0x00009888" value="0x12584D60" />
+        <register type="NOA" address="0x00009888" value="0x186C2A00" />
+        <register type="NOA" address="0x00009888" value="0x1A6C0001" />
+        <register type="NOA" address="0x00009888" value="0x00738000" />
+        <register type="NOA" address="0x00009888" value="0x02737000" />
+        <register type="NOA" address="0x00009888" value="0x12740140" />
+        <register type="NOA" address="0x00009888" value="0x1475000A" />
+        <register type="NOA" address="0x00009888" value="0x10760028" />
+        <register type="NOA" address="0x00009888" value="0x16777C00" />
+        <register type="NOA" address="0x00009888" value="0x18770001" />
+        <register type="NOA" address="0x00009888" value="0x02774000" />
+        <register type="NOA" address="0x00009888" value="0x04774000" />
+        <register type="NOA" address="0x00009888" value="0x06778000" />
+        <register type="NOA" address="0x00009888" value="0x08778000" />
+        <register type="NOA" address="0x00009888" value="0x0A778000" />
+        <register type="NOA" address="0x00009888" value="0x0C778000" />
+        <register type="NOA" address="0x00009888" value="0x0C7B0280" />
+        <register type="NOA" address="0x00009888" value="0x087BA000" />
+        <register type="NOA" address="0x00009888" value="0x167C0280" />
+        <register type="NOA" address="0x00009888" value="0x027CC000" />
+        <register type="NOA" address="0x00009888" value="0x047CC000" />
+        <register type="NOA" address="0x00009888" value="0x1E7D2800" />
+        <register type="NOA" address="0x00009888" value="0x1A192000" />
+        <register type="NOA" address="0x00009888" value="0x1C190001" />
+        <register type="NOA" address="0x00009888" value="0x1E140028" />
+        <register type="NOA" address="0x00009888" value="0x0A1500C0" />
+        <register type="NOA" address="0x00009888" value="0x1A164000" />
+        <register type="NOA" address="0x00009888" value="0x1C160002" />
+        <register type="NOA" address="0x00009888" value="0x0C18AFA7" />
+        <register type="NOA" address="0x00009888" value="0x10180000" />
+        <register type="NOA" address="0x00009888" value="0x1C390014" />
+        <register type="NOA" address="0x00009888" value="0x1E340280" />
+        <register type="NOA" address="0x00009888" value="0x0A350300" />
+        <register type="NOA" address="0x00009888" value="0x1C360028" />
+        <register type="NOA" address="0x00009888" value="0x0E38AFA7" />
+        <register type="NOA" address="0x00009888" value="0x10380000" />
+        <register type="NOA" address="0x00009888" value="0x06F48000" />
+        <register type="NOA" address="0x00009888" value="0x08F42000" />
+        <register type="NOA" address="0x00009888" value="0x06F58000" />
+        <register type="NOA" address="0x00009888" value="0x08F52000" />
+        <register type="NOA" address="0x00009888" value="0x0EF6000A" />
+        <register type="NOA" address="0x00009888" value="0x06F7C000" />
+        <register type="NOA" address="0x00009888" value="0x08F7C000" />
+        <register type="NOA" address="0x00009888" value="0x0AF74000" />
+        <register type="NOA" address="0x00009888" value="0x0CF74000" />
+        <register type="NOA" address="0x00009888" value="0x0AFB00A0" />
+        <register type="NOA" address="0x00009888" value="0x0AFC8000" />
+        <register type="NOA" address="0x00009888" value="0x0CFC8000" />
+        <register type="NOA" address="0x00009888" value="0x1CFD0A00" />
+        <register type="NOA" address="0x00009888" value="0x06994000" />
+        <register type="NOA" address="0x00009888" value="0x08994000" />
+        <register type="NOA" address="0x00009888" value="0x1C94000A" />
+        <register type="NOA" address="0x00009888" value="0x08956000" />
+        <register type="NOA" address="0x00009888" value="0x06968000" />
+        <register type="NOA" address="0x00009888" value="0x08968000" />
+        <register type="NOA" address="0x00009888" value="0x0298A700" />
+        <register type="NOA" address="0x00009888" value="0x049800AF" />
+        <register type="NOA" address="0x00009888" value="0x10980000" />
+        <register type="NOA" address="0x00009888" value="0x0AB94000" />
+        <register type="NOA" address="0x00009888" value="0x0CB94000" />
+        <register type="NOA" address="0x00009888" value="0x1CB400A0" />
+        <register type="NOA" address="0x00009888" value="0x08B58000" />
+        <register type="NOA" address="0x00009888" value="0x0AB50001" />
+        <register type="NOA" address="0x00009888" value="0x0AB68000" />
+        <register type="NOA" address="0x00009888" value="0x0CB68000" />
+        <register type="NOA" address="0x00009888" value="0x04B8A700" />
+        <register type="NOA" address="0x00009888" value="0x06B800AF" />
+        <register type="NOA" address="0x00009888" value="0x10B80000" />
+        <register type="NOA" address="0x00009888" value="0x02594000" />
+        <register type="NOA" address="0x00009888" value="0x04594000" />
+        <register type="NOA" address="0x00009888" value="0x1A54A000" />
+        <register type="NOA" address="0x00009888" value="0x00558000" />
+        <register type="NOA" address="0x00009888" value="0x08551000" />
+        <register type="NOA" address="0x00009888" value="0x02568000" />
+        <register type="NOA" address="0x00009888" value="0x04568000" />
+        <register type="NOA" address="0x00009888" value="0x0058A700" />
+        <register type="NOA" address="0x00009888" value="0x025800AF" />
+        <register type="NOA" address="0x00009888" value="0x10580000" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x21003000" />
+        <register type="NOA" address="0x00009888" value="0x27000071" />
+        <register type="NOA" address="0x00009888" value="0x33000000" />
+        <register type="NOA" address="0x00009888" value="0x2900C000" />
+        <register type="NOA" address="0x00009888" value="0x2B00C000" />
+        <register type="NOA" address="0x00009888" value="0x2D00C000" />
+        <register type="NOA" address="0x00009888" value="0x2F00C000" />
+        <register type="NOA" address="0x00009888" value="0x1300C000" />
+        <register type="NOA" address="0x00009888" value="0x1500C000" />
+        <register type="NOA" address="0x00009888" value="0x1700C000" />
+        <register type="NOA" address="0x00009888" value="0x1900C000" />
+        <register type="NOA" address="0x00009888" value="0x1B00C000" />
+        <register type="NOA" address="0x00009888" value="0x1D00C000" />
+        <register type="NOA" address="0x00009888" value="0x4B003000" />
+        <register type="NOA" address="0x00009888" value="0x3F002C00" />
+        <register type="NOA" address="0x00009888" value="0x4D001111" />
+        <register type="NOA" address="0x00009888" value="0x41000421" />
+        <register type="NOA" address="0x00009888" value="0x35000000" />
+        <register type="NOA" address="0x00009888" value="0x31000000" />
+        <register type="NOA" address="0x00009888" value="0x43000001" />
+        <register type="NOA" address="0x00009888" value="0x47000000" />
+        <register type="NOA" address="0x00009888" value="0x39001080" />
+        <register type="NOA" address="0x00009888" value="0x3B000000" />
+        <register type="NOA" address="0x00009888" value="0x49000000" />
+        <register type="NOA" address="0x00009888" value="0x3D000000" />
+    </register_config>
+    <register_config type="OA">
+        <register type="OA" address="0x00002740" value="0x00000000" />
+        <register type="OA" address="0x00002710" value="0x00000000" />
+        <register type="OA" address="0x00002714" value="0x00800000" />
+        <register type="OA" address="0x00002720" value="0x00000000" />
+        <register type="OA" address="0x00002724" value="0x00800000" />
+    </register_config>
+    <register_config type="FLEX">
+        <register type="FLEX" address="0x0000E458" value="0x00005004" />
+        <register type="FLEX" address="0x0000E558" value="0x00010003" />
+        <register type="FLEX" address="0x0000E658" value="0x00012011" />
+        <register type="FLEX" address="0x0000E45C" value="0x00051050" />
+        <register type="FLEX" address="0x0000E55C" value="0x00053052" />
+    </register_config>
+  </set>
+
+  <set name="Metric set TestOa"
+       mdapi_supported_apis="OGL OCL IO BB"
+       underscore_name="test_oa"
+       hw_config_guid="db41edd4-d8e7-4730-ad11-b9a2d6833503"
+       chipset="CNL"
+       symbol_name="TestOa"
+       >
+    <counter name="TestCounter7"
+             description="HW test counter 7. Factor: 0.6666"
+             data_type="uint64"
+             equation="B 7 READ"
+             underscore_name="counter7"
+             units="events"
+             symbol_name="Counter7"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="GPU Time Elapsed"
+             description="Time elapsed on the GPU during the measurement."
+             data_type="uint64"
+             equation="GPU_TIME 0 READ 1000000000 UMUL $GpuTimestampFrequency UDIV"
+             underscore_name="gpu_time"
+             units="ns"
+             symbol_name="GpuTime"
+             semantic_type="duration"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="GPU Core Clocks"
+             description="The total number of GPU core clocks elapsed during the measurement."
+             data_type="uint64"
+             equation="GPU_CLOCK 0 READ"
+             underscore_name="gpu_core_clocks"
+             units="cycles"
+             symbol_name="GpuCoreClocks"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="AVG GPU Core Frequency"
+             description="Average GPU Core Frequency in the measurement."
+             data_type="uint64"
+             max_equation="$GpuMaxFrequency"
+             equation="$GpuCoreClocks 1000000000 UMUL $GpuTime UDIV"
+             underscore_name="avg_gpu_core_frequency"
+             units="hz"
+             symbol_name="AvgGpuCoreFrequency"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Tier1 Overview System Frame Batch Draw"
+             mdapi_group="GPU"
+             mdapi_hw_unit_type="gpu"
+             />
+    <counter name="TestCounter8"
+             description="HW test counter 8. Should be equal to 1."
+             data_type="uint64"
+             equation="C 7 READ"
+             underscore_name="counter8"
+             units="events"
+             symbol_name="Counter8"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="TestCounter4"
+             description="HW test counter 4. Factor: 0.3333"
+             data_type="uint64"
+             equation="B 4 READ"
+             underscore_name="counter4"
+             units="events"
+             symbol_name="Counter4"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="TestCounter5"
+             description="HW test counter 5. Factor: 0.3333"
+             data_type="uint64"
+             equation="B 5 READ"
+             underscore_name="counter5"
+             units="events"
+             symbol_name="Counter5"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="TestCounter6"
+             description="HW test counter 6. Factor: 0.16666"
+             data_type="uint64"
+             equation="B 6 READ"
+             underscore_name="counter6"
+             units="events"
+             symbol_name="Counter6"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="TestCounter3"
+             description="HW test counter 3. Factor: 0.5"
+             data_type="uint64"
+             equation="B 3 READ"
+             underscore_name="counter3"
+             units="events"
+             symbol_name="Counter3"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="TestCounter0"
+             description="HW test counter 0. Factor: 0.0"
+             data_type="uint64"
+             equation="B 0 READ"
+             underscore_name="counter0"
+             units="events"
+             symbol_name="Counter0"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="TestCounter1"
+             description="HW test counter 1. Factor: 1.0"
+             data_type="uint64"
+             equation="B 1 READ"
+             underscore_name="counter1"
+             units="events"
+             symbol_name="Counter1"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <counter name="TestCounter2"
+             description="HW test counter 2. Factor: 1.0"
+             data_type="uint64"
+             equation="B 2 READ"
+             underscore_name="counter2"
+             units="events"
+             symbol_name="Counter2"
+             semantic_type="event"
+             mdapi_supported_apis=""
+             mdapi_usage_flags="Frame Batch Draw"
+             mdapi_hw_unit_type="gpu"
+             mdapi_group="GPU"
+             />
+    <register_config type="NOA">
+        <register type="NOA" address="0x00000D04" value="0x00000200" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x17060000" />
+        <register type="NOA" address="0x00009840" value="0x00000000" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x13034000" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x07060066" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x05060000" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x0F080040" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x07091000" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x0F041000" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x1D004000" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x35000000" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x49000000" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x3D000000" />
+        <register type="NOA" address="0x00009884" value="0x00000007" />
+        <register type="NOA" address="0x00009888" value="0x31000000" />
+    </register_config>
+    <register_config type="OA">
+        <register type="OA" address="0x00002740" value="0x00000000" />
+        <register type="OA" address="0x00002710" value="0x00000000" />
+        <register type="OA" address="0x00002714" value="0xF0800000" />
+        <register type="OA" address="0x00002720" value="0x00000000" />
+        <register type="OA" address="0x00002724" value="0xF0800000" />
+        <register type="OA" address="0x00002770" value="0x00000004" />
+        <register type="OA" address="0x00002774" value="0x0000FFFF" />
+        <register type="OA" address="0x00002778" value="0x00000003" />
+        <register type="OA" address="0x0000277C" value="0x0000FFFF" />
+        <register type="OA" address="0x00002780" value="0x00000007" />
+        <register type="OA" address="0x00002784" value="0x0000FFFF" />
+        <register type="OA" address="0x00002788" value="0x00100002" />
+        <register type="OA" address="0x0000278C" value="0x0000FFF7" />
+        <register type="OA" address="0x00002790" value="0x00100002" />
+        <register type="OA" address="0x00002794" value="0x0000FFCF" />
+        <register type="OA" address="0x00002798" value="0x00100082" />
+        <register type="OA" address="0x0000279C" value="0x0000FFEF" />
+        <register type="OA" address="0x000027A0" value="0x001000C2" />
+        <register type="OA" address="0x000027A4" value="0x0000FFE7" />
+        <register type="OA" address="0x000027A8" value="0x00100001" />
+        <register type="OA" address="0x000027AC" value="0x0000FFE7" />
+    </register_config>
+  </set>
+
+</metrics>
diff --git a/src/mesa/drivers/dri/i965/brw_performance_query.c b/src/mesa/drivers/dri/i965/brw_performance_query.c
index cca74001f19..4144aa3c9b8 100644
--- a/src/mesa/drivers/dri/i965/brw_performance_query.c
+++ b/src/mesa/drivers/dri/i965/brw_performance_query.c
@@ -2142,6 +2142,8 @@ get_register_queries_function(const struct gen_device_info *devinfo)
       if (devinfo->gt == 3)
          return brw_oa_register_queries_cflgt3;
    }
+   if (devinfo->is_cannonlake)
+      return brw_oa_register_queries_cnl;
 
    return NULL;
 }
diff --git a/src/mesa/drivers/dri/i965/meson.build b/src/mesa/drivers/dri/i965/meson.build
index cfa27f31bf8..4d8280df573 100644
--- a/src/mesa/drivers/dri/i965/meson.build
+++ b/src/mesa/drivers/dri/i965/meson.build
@@ -157,6 +157,7 @@ i965_hw_metrics = [
   'kblgt2', 'kblgt3',
   'cflgt2', 'cflgt3',
   'bxt', 'glk',
+  'cnl',
 ]
 
 i965_hw_metrics_xml_files = []
-- 
2.30.2