2 * Copyright © 2018 Intel Corporation
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
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
31 #if defined(MAJOR_IN_SYSMACROS)
32 #include <sys/sysmacros.h>
33 #elif defined(MAJOR_IN_MKDEV)
34 #include <sys/mkdev.h>
37 #include "util/hash_table.h"
38 #include "compiler/glsl/list.h"
39 #include "util/ralloc.h"
41 #include "drm-uapi/i915_drm.h"
43 struct gen_device_info
;
45 struct gen_perf_config
;
46 struct gen_perf_query_info
;
48 enum gen_perf_counter_type
{
49 GEN_PERF_COUNTER_TYPE_EVENT
,
50 GEN_PERF_COUNTER_TYPE_DURATION_NORM
,
51 GEN_PERF_COUNTER_TYPE_DURATION_RAW
,
52 GEN_PERF_COUNTER_TYPE_THROUGHPUT
,
53 GEN_PERF_COUNTER_TYPE_RAW
,
54 GEN_PERF_COUNTER_TYPE_TIMESTAMP
,
57 enum gen_perf_counter_data_type
{
58 GEN_PERF_COUNTER_DATA_TYPE_BOOL32
,
59 GEN_PERF_COUNTER_DATA_TYPE_UINT32
,
60 GEN_PERF_COUNTER_DATA_TYPE_UINT64
,
61 GEN_PERF_COUNTER_DATA_TYPE_FLOAT
,
62 GEN_PERF_COUNTER_DATA_TYPE_DOUBLE
,
65 enum gen_perf_counter_units
{
67 GEN_PERF_COUNTER_UNITS_BYTES
,
70 GEN_PERF_COUNTER_UNITS_HZ
,
73 GEN_PERF_COUNTER_UNITS_NS
,
74 GEN_PERF_COUNTER_UNITS_US
,
77 GEN_PERF_COUNTER_UNITS_PIXELS
,
78 GEN_PERF_COUNTER_UNITS_TEXELS
,
79 GEN_PERF_COUNTER_UNITS_THREADS
,
80 GEN_PERF_COUNTER_UNITS_PERCENT
,
83 GEN_PERF_COUNTER_UNITS_MESSAGES
,
84 GEN_PERF_COUNTER_UNITS_NUMBER
,
85 GEN_PERF_COUNTER_UNITS_CYCLES
,
86 GEN_PERF_COUNTER_UNITS_EVENTS
,
87 GEN_PERF_COUNTER_UNITS_UTILIZATION
,
90 GEN_PERF_COUNTER_UNITS_EU_SENDS_TO_L3_CACHE_LINES
,
91 GEN_PERF_COUNTER_UNITS_EU_ATOMIC_REQUESTS_TO_L3_CACHE_LINES
,
92 GEN_PERF_COUNTER_UNITS_EU_REQUESTS_TO_L3_CACHE_LINES
,
93 GEN_PERF_COUNTER_UNITS_EU_BYTES_PER_L3_CACHE_LINE
,
95 GEN_PERF_COUNTER_UNITS_MAX
98 struct gen_pipeline_stat
{
101 uint32_t denominator
;
105 * The largest OA formats we can use include:
107 * 1 timestamp, 45 A counters, 8 B counters and 8 C counters.
109 * 1 timestamp, 1 clock, 36 A counters, 8 B counters and 8 C counters
111 #define MAX_OA_REPORT_COUNTERS 62
114 * When currently allocate only one page for pipeline statistics queries. Here
115 * we derived the maximum number of counters for that amount.
117 #define STATS_BO_SIZE 4096
118 #define STATS_BO_END_OFFSET_BYTES (STATS_BO_SIZE / 2)
119 #define MAX_STAT_COUNTERS (STATS_BO_END_OFFSET_BYTES / 8)
121 #define I915_PERF_OA_SAMPLE_SIZE (8 + /* drm_i915_perf_record_header */ \
122 256) /* OA counter report */
124 struct gen_perf_query_result
{
126 * Storage for the final accumulated OA counters.
128 uint64_t accumulator
[MAX_OA_REPORT_COUNTERS
];
131 * Hw ID used by the context on which the query was running.
136 * Number of reports accumulated to produce the results.
138 uint32_t reports_accumulated
;
141 * Frequency in the slices of the GT at the begin and end of the
144 uint64_t slice_frequency
[2];
147 * Frequency in the unslice of the GT at the begin and end of the
150 uint64_t unslice_frequency
[2];
153 * Timestamp of the query.
155 uint64_t begin_timestamp
;
158 * Whether the query was interrupted by another workload (aka preemption).
163 struct gen_perf_query_counter
{
166 const char *symbol_name
;
167 enum gen_perf_counter_type type
;
168 enum gen_perf_counter_data_type data_type
;
169 enum gen_perf_counter_units units
;
175 uint64_t (*oa_counter_read_uint64
)(struct gen_perf_config
*perf
,
176 const struct gen_perf_query_info
*query
,
177 const uint64_t *accumulator
);
178 float (*oa_counter_read_float
)(struct gen_perf_config
*perf
,
179 const struct gen_perf_query_info
*query
,
180 const uint64_t *accumulator
);
181 struct gen_pipeline_stat pipeline_stat
;
185 struct gen_perf_query_register_prog
{
190 /* Register programming for a given query */
191 struct gen_perf_registers
{
192 struct gen_perf_query_register_prog
*flex_regs
;
193 uint32_t n_flex_regs
;
195 struct gen_perf_query_register_prog
*mux_regs
;
198 struct gen_perf_query_register_prog
*b_counter_regs
;
199 uint32_t n_b_counter_regs
;
202 struct gen_perf_query_info
{
203 enum gen_perf_query_type
{
204 GEN_PERF_QUERY_TYPE_OA
,
205 GEN_PERF_QUERY_TYPE_RAW
,
206 GEN_PERF_QUERY_TYPE_PIPELINE
,
210 struct gen_perf_query_counter
*counters
;
216 uint64_t oa_metrics_set_id
;
219 /* For indexing into the accumulator[] ... */
221 int gpu_clock_offset
;
226 struct gen_perf_registers config
;
229 struct gen_perf_config
{
230 /* Whether i915 has DRM_I915_QUERY_PERF_CONFIG support. */
231 bool i915_query_supported
;
233 /* Version of the i915-perf subsystem, refer to i915_drm.h. */
234 int i915_perf_version
;
236 /* Powergating configuration for the running the query. */
237 struct drm_i915_gem_context_param_sseu sseu
;
239 struct gen_perf_query_info
*queries
;
242 struct gen_perf_query_counter
**counters
;
245 /* Variables referenced in the XML meta data for OA performance
246 * counters, e.g in the normalization equations.
248 * All uint64_t for consistent operand types in generated code
251 uint64_t timestamp_frequency
; /** $GpuTimestampFrequency */
252 uint64_t n_eus
; /** $EuCoresTotalCount */
253 uint64_t n_eu_slices
; /** $EuSlicesTotalCount */
254 uint64_t n_eu_sub_slices
; /** $EuSubslicesTotalCount */
255 uint64_t eu_threads_count
; /** $EuThreadsCount */
256 uint64_t slice_mask
; /** $SliceMask */
257 uint64_t subslice_mask
; /** $SubsliceMask */
258 uint64_t gt_min_freq
; /** $GpuMinFrequency */
259 uint64_t gt_max_freq
; /** $GpuMaxFrequency */
260 uint64_t revision
; /** $SkuRevisionId */
263 /* OA metric sets, indexed by GUID, as know by Mesa at build time, to
264 * cross-reference with the GUIDs of configs advertised by the kernel at
267 struct hash_table
*oa_metrics_table
;
269 /* Location of the device's sysfs entry. */
270 char sysfs_dev_dir
[256];
273 void *(*bo_alloc
)(void *bufmgr
, const char *name
, uint64_t size
);
274 void (*bo_unreference
)(void *bo
);
275 void *(*bo_map
)(void *ctx
, void *bo
, unsigned flags
);
276 void (*bo_unmap
)(void *bo
);
277 bool (*batch_references
)(void *batch
, void *bo
);
278 void (*bo_wait_rendering
)(void *bo
);
279 int (*bo_busy
)(void *bo
);
280 void (*emit_stall_at_pixel_scoreboard
)(void *ctx
);
281 void (*emit_mi_report_perf_count
)(void *ctx
,
283 uint32_t offset_in_bytes
,
285 void (*batchbuffer_flush
)(void *ctx
,
286 const char *file
, int line
);
287 void (*store_register_mem
)(void *ctx
, void *bo
, uint32_t reg
, uint32_t reg_size
, uint32_t offset
);
292 struct gen_perf_counter_pass
{
293 struct gen_perf_query_info
*query
;
294 struct gen_perf_query_counter
*counter
;
298 void gen_perf_init_metrics(struct gen_perf_config
*perf_cfg
,
299 const struct gen_device_info
*devinfo
,
301 bool include_pipeline_statistics
);
303 /** Query i915 for a metric id using guid.
305 bool gen_perf_load_metric_id(struct gen_perf_config
*perf_cfg
,
307 uint64_t *metric_id
);
309 /** Load a configuation's content from i915 using a guid.
311 struct gen_perf_registers
*gen_perf_load_configuration(struct gen_perf_config
*perf_cfg
,
312 int fd
, const char *guid
);
314 /** Store a configuration into i915 using guid and return a new metric id.
316 * If guid is NULL, then a generated one will be provided by hashing the
317 * content of the configuration.
319 uint64_t gen_perf_store_configuration(struct gen_perf_config
*perf_cfg
, int fd
,
320 const struct gen_perf_registers
*config
,
323 /** Read the slice/unslice frequency from 2 OA reports and store then into
326 void gen_perf_query_result_read_frequencies(struct gen_perf_query_result
*result
,
327 const struct gen_device_info
*devinfo
,
328 const uint32_t *start
,
329 const uint32_t *end
);
330 /** Accumulate the delta between 2 OA reports into result for a given query.
332 void gen_perf_query_result_accumulate(struct gen_perf_query_result
*result
,
333 const struct gen_perf_query_info
*query
,
334 const uint32_t *start
,
335 const uint32_t *end
);
336 void gen_perf_query_result_clear(struct gen_perf_query_result
*result
);
339 gen_perf_query_counter_get_size(const struct gen_perf_query_counter
*counter
)
341 switch (counter
->data_type
) {
342 case GEN_PERF_COUNTER_DATA_TYPE_BOOL32
:
343 return sizeof(uint32_t);
344 case GEN_PERF_COUNTER_DATA_TYPE_UINT32
:
345 return sizeof(uint32_t);
346 case GEN_PERF_COUNTER_DATA_TYPE_UINT64
:
347 return sizeof(uint64_t);
348 case GEN_PERF_COUNTER_DATA_TYPE_FLOAT
:
349 return sizeof(float);
350 case GEN_PERF_COUNTER_DATA_TYPE_DOUBLE
:
351 return sizeof(double);
353 unreachable("invalid counter data type");
357 static inline struct gen_perf_config
*
358 gen_perf_new(void *ctx
)
360 struct gen_perf_config
*perf
= rzalloc(ctx
, struct gen_perf_config
);
364 uint32_t gen_perf_get_n_passes(struct gen_perf_config
*perf
,
365 const uint32_t *counter_indices
,
366 uint32_t counter_indices_count
,
367 struct gen_perf_query_info
**pass_queries
);
368 void gen_perf_get_counters_passes(struct gen_perf_config
*perf
,
369 const uint32_t *counter_indices
,
370 uint32_t counter_indices_count
,
371 struct gen_perf_counter_pass
*counter_pass
);
373 #endif /* GEN_PERF_H */