2 * Copyright 2015 Advanced Micro Devices, Inc.
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
24 * Nicolai Hähnle <nicolai.haehnle@amd.com>
31 #include "pipe/p_defines.h"
32 #include "pipe/p_state.h"
33 #include "util/list.h"
39 struct r600_common_context
;
40 struct r600_common_screen
;
46 R600_QUERY_DRAW_CALLS
= PIPE_QUERY_DRIVER_SPECIFIC
,
47 R600_QUERY_SPILL_DRAW_CALLS
,
48 R600_QUERY_COMPUTE_CALLS
,
49 R600_QUERY_SPILL_COMPUTE_CALLS
,
51 R600_QUERY_CP_DMA_CALLS
,
52 R600_QUERY_NUM_VS_FLUSHES
,
53 R600_QUERY_NUM_PS_FLUSHES
,
54 R600_QUERY_NUM_CS_FLUSHES
,
55 R600_QUERY_NUM_FB_CACHE_FLUSHES
,
56 R600_QUERY_NUM_L2_INVALIDATES
,
57 R600_QUERY_NUM_L2_WRITEBACKS
,
58 R600_QUERY_REQUESTED_VRAM
,
59 R600_QUERY_REQUESTED_GTT
,
60 R600_QUERY_MAPPED_VRAM
,
61 R600_QUERY_MAPPED_GTT
,
62 R600_QUERY_BUFFER_WAIT_TIME
,
63 R600_QUERY_NUM_MAPPED_BUFFERS
,
64 R600_QUERY_NUM_GFX_IBS
,
65 R600_QUERY_NUM_SDMA_IBS
,
66 R600_QUERY_NUM_BYTES_MOVED
,
67 R600_QUERY_NUM_EVICTIONS
,
68 R600_QUERY_VRAM_USAGE
,
70 R600_QUERY_GPU_TEMPERATURE
,
71 R600_QUERY_CURRENT_GPU_SCLK
,
72 R600_QUERY_CURRENT_GPU_MCLK
,
74 R600_QUERY_GPU_SHADERS_BUSY
,
75 R600_QUERY_GPU_TA_BUSY
,
76 R600_QUERY_GPU_GDS_BUSY
,
77 R600_QUERY_GPU_VGT_BUSY
,
78 R600_QUERY_GPU_IA_BUSY
,
79 R600_QUERY_GPU_SX_BUSY
,
80 R600_QUERY_GPU_WD_BUSY
,
81 R600_QUERY_GPU_BCI_BUSY
,
82 R600_QUERY_GPU_SC_BUSY
,
83 R600_QUERY_GPU_PA_BUSY
,
84 R600_QUERY_GPU_DB_BUSY
,
85 R600_QUERY_GPU_CP_BUSY
,
86 R600_QUERY_GPU_CB_BUSY
,
87 R600_QUERY_NUM_COMPILATIONS
,
88 R600_QUERY_NUM_SHADERS_CREATED
,
89 R600_QUERY_BACK_BUFFER_PS_DRAW_RATIO
,
90 R600_QUERY_NUM_SHADER_CACHE_HITS
,
91 R600_QUERY_GPIN_ASIC_ID
,
92 R600_QUERY_GPIN_NUM_SIMD
,
93 R600_QUERY_GPIN_NUM_RB
,
94 R600_QUERY_GPIN_NUM_SPI
,
95 R600_QUERY_GPIN_NUM_SE
,
97 R600_QUERY_FIRST_PERFCOUNTER
= PIPE_QUERY_DRIVER_SPECIFIC
+ 100,
101 R600_QUERY_GROUP_GPIN
= 0,
102 R600_NUM_SW_QUERY_GROUPS
105 struct r600_query_ops
{
106 void (*destroy
)(struct r600_common_context
*, struct r600_query
*);
107 bool (*begin
)(struct r600_common_context
*, struct r600_query
*);
108 bool (*end
)(struct r600_common_context
*, struct r600_query
*);
109 bool (*get_result
)(struct r600_common_context
*,
110 struct r600_query
*, bool wait
,
111 union pipe_query_result
*result
);
112 void (*get_result_resource
)(struct r600_common_context
*,
113 struct r600_query
*, bool wait
,
114 enum pipe_query_value_type result_type
,
116 struct pipe_resource
*resource
,
121 struct r600_query_ops
*ops
;
123 /* The type of query */
128 R600_QUERY_HW_FLAG_NO_START
= (1 << 0),
130 /* whether begin_query doesn't clear the result */
131 R600_QUERY_HW_FLAG_BEGIN_RESUMES
= (1 << 2),
134 struct r600_query_hw_ops
{
135 bool (*prepare_buffer
)(struct r600_common_context
*,
136 struct r600_query_hw
*,
137 struct r600_resource
*);
138 void (*emit_start
)(struct r600_common_context
*,
139 struct r600_query_hw
*,
140 struct r600_resource
*buffer
, uint64_t va
);
141 void (*emit_stop
)(struct r600_common_context
*,
142 struct r600_query_hw
*,
143 struct r600_resource
*buffer
, uint64_t va
);
144 void (*clear_result
)(struct r600_query_hw
*, union pipe_query_result
*);
145 void (*add_result
)(struct r600_common_context
*ctx
,
146 struct r600_query_hw
*, void *buffer
,
147 union pipe_query_result
*result
);
150 struct r600_query_buffer
{
151 /* The buffer where query results are stored. */
152 struct r600_resource
*buf
;
153 /* Offset of the next free result after current query data */
154 unsigned results_end
;
155 /* If a query buffer is full, a new buffer is created and the old one
156 * is put in here. When we calculate the result, we sum up the samples
157 * from all buffers. */
158 struct r600_query_buffer
*previous
;
161 struct r600_query_hw
{
163 struct r600_query_hw_ops
*ops
;
166 /* The query buffer and how many results are in it. */
167 struct r600_query_buffer buffer
;
168 /* Size of the result in memory for both begin_query and end_query,
169 * this can be one or two numbers, or it could even be a size of a structure. */
170 unsigned result_size
;
171 /* The number of dwords for begin_query or end_query. */
172 unsigned num_cs_dw_begin
;
173 unsigned num_cs_dw_end
;
174 /* Linked list of queries */
175 struct list_head list
;
176 /* For transform feedback: which stream the query is for */
180 bool r600_query_hw_init(struct r600_common_context
*rctx
,
181 struct r600_query_hw
*query
);
182 void r600_query_hw_destroy(struct r600_common_context
*rctx
,
183 struct r600_query
*rquery
);
184 bool r600_query_hw_begin(struct r600_common_context
*rctx
,
185 struct r600_query
*rquery
);
186 bool r600_query_hw_end(struct r600_common_context
*rctx
,
187 struct r600_query
*rquery
);
188 bool r600_query_hw_get_result(struct r600_common_context
*rctx
,
189 struct r600_query
*rquery
,
191 union pipe_query_result
*result
);
193 /* Performance counters */
195 /* This block is part of the shader engine */
196 R600_PC_BLOCK_SE
= (1 << 0),
198 /* Expose per-instance groups instead of summing all instances (within
200 R600_PC_BLOCK_INSTANCE_GROUPS
= (1 << 1),
202 /* Expose per-SE groups instead of summing instances across SEs. */
203 R600_PC_BLOCK_SE_GROUPS
= (1 << 2),
206 R600_PC_BLOCK_SHADER
= (1 << 3),
208 /* Non-shader block with perfcounters windowed by shaders. */
209 R600_PC_BLOCK_SHADER_WINDOWED
= (1 << 4),
212 /* Describes a hardware block with performance counters. Multiple instances of
213 * each block, possibly per-SE, may exist on the chip. Depending on the block
214 * and on the user's configuration, we either
215 * (a) expose every instance as a performance counter group,
216 * (b) expose a single performance counter group that reports the sum over all
218 * (c) expose one performance counter group per instance, but summed over all
221 struct r600_perfcounter_block
{
222 const char *basename
;
224 unsigned num_counters
;
225 unsigned num_selectors
;
226 unsigned num_instances
;
230 unsigned group_name_stride
;
232 char *selector_names
;
233 unsigned selector_name_stride
;
238 struct r600_perfcounters
{
241 struct r600_perfcounter_block
*blocks
;
243 unsigned num_start_cs_dwords
;
244 unsigned num_stop_cs_dwords
;
245 unsigned num_instance_cs_dwords
;
246 unsigned num_shaders_cs_dwords
;
248 unsigned num_shader_types
;
249 const char * const *shader_type_suffixes
;
250 const unsigned *shader_type_bits
;
252 void (*get_size
)(struct r600_perfcounter_block
*,
253 unsigned count
, unsigned *selectors
,
254 unsigned *num_select_dw
, unsigned *num_read_dw
);
256 void (*emit_instance
)(struct r600_common_context
*,
257 int se
, int instance
);
258 void (*emit_shaders
)(struct r600_common_context
*, unsigned shaders
);
259 void (*emit_select
)(struct r600_common_context
*,
260 struct r600_perfcounter_block
*,
261 unsigned count
, unsigned *selectors
);
262 void (*emit_start
)(struct r600_common_context
*,
263 struct r600_resource
*buffer
, uint64_t va
);
264 void (*emit_stop
)(struct r600_common_context
*,
265 struct r600_resource
*buffer
, uint64_t va
);
266 void (*emit_read
)(struct r600_common_context
*,
267 struct r600_perfcounter_block
*,
268 unsigned count
, unsigned *selectors
,
269 struct r600_resource
*buffer
, uint64_t va
);
271 void (*cleanup
)(struct r600_common_screen
*);
274 bool separate_instance
;
277 struct pipe_query
*r600_create_batch_query(struct pipe_context
*ctx
,
278 unsigned num_queries
,
279 unsigned *query_types
);
281 int r600_get_perfcounter_info(struct r600_common_screen
*,
283 struct pipe_driver_query_info
*info
);
284 int r600_get_perfcounter_group_info(struct r600_common_screen
*,
286 struct pipe_driver_query_group_info
*info
);
288 bool r600_perfcounters_init(struct r600_perfcounters
*, unsigned num_blocks
);
289 void r600_perfcounters_add_block(struct r600_common_screen
*,
290 struct r600_perfcounters
*,
291 const char *name
, unsigned flags
,
292 unsigned counters
, unsigned selectors
,
293 unsigned instances
, void *data
);
294 void r600_perfcounters_do_destroy(struct r600_perfcounters
*);
295 void r600_query_hw_reset_buffers(struct r600_common_context
*rctx
,
296 struct r600_query_hw
*query
);
298 struct r600_qbo_state
{
300 struct pipe_constant_buffer saved_const0
;
301 struct pipe_shader_buffer saved_ssbo
[3];
304 #endif /* R600_QUERY_H */