src/intel/perf/gen_perf.h

   1 /*
   2  * Copyright © 2018 Intel Corporation
   3  *
   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:
  10  *
  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
  13  * Software.
  14  *
  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
  21  * IN THE SOFTWARE.
  22  */
  23
  24 #ifndef GEN_PERF_H
  25 #define GEN_PERF_H
  26
  27 #include <stdio.h>
  28 #include <stdint.h>
  29 #include <string.h>
  30
  31 #if defined(MAJOR_IN_SYSMACROS)
  32 #include <sys/sysmacros.h>
  33 #elif defined(MAJOR_IN_MKDEV)
  34 #include <sys/mkdev.h>
  35 #endif
  36
  37 #include "util/hash_table.h"
  38 #include "compiler/glsl/list.h"
  39 #include "util/ralloc.h"
  40
  41 struct gen_device_info;
  42
  43 struct gen_perf_config;
  44 struct gen_perf_query_info;
  45
  46 enum gen_perf_counter_type {
  47    GEN_PERF_COUNTER_TYPE_EVENT,
  48    GEN_PERF_COUNTER_TYPE_DURATION_NORM,
  49    GEN_PERF_COUNTER_TYPE_DURATION_RAW,
  50    GEN_PERF_COUNTER_TYPE_THROUGHPUT,
  51    GEN_PERF_COUNTER_TYPE_RAW,
  52    GEN_PERF_COUNTER_TYPE_TIMESTAMP,
  53 };
  54
  55 enum gen_perf_counter_data_type {
  56    GEN_PERF_COUNTER_DATA_TYPE_BOOL32,
  57    GEN_PERF_COUNTER_DATA_TYPE_UINT32,
  58    GEN_PERF_COUNTER_DATA_TYPE_UINT64,
  59    GEN_PERF_COUNTER_DATA_TYPE_FLOAT,
  60    GEN_PERF_COUNTER_DATA_TYPE_DOUBLE,
  61 };
  62
  63 struct gen_pipeline_stat {
  64    uint32_t reg;
  65    uint32_t numerator;
  66    uint32_t denominator;
  67 };
  68
  69 /*
  70  * The largest OA formats we can use include:
  71  * For Haswell:
  72  *   1 timestamp, 45 A counters, 8 B counters and 8 C counters.
  73  * For Gen8+
  74  *   1 timestamp, 1 clock, 36 A counters, 8 B counters and 8 C counters
  75  */
  76 #define MAX_OA_REPORT_COUNTERS 62
  77
  78 /*
  79  * When currently allocate only one page for pipeline statistics queries. Here
  80  * we derived the maximum number of counters for that amount.
  81  */
  82 #define STATS_BO_SIZE               4096
  83 #define STATS_BO_END_OFFSET_BYTES   (STATS_BO_SIZE / 2)
  84 #define MAX_STAT_COUNTERS           (STATS_BO_END_OFFSET_BYTES / 8)
  85
  86 #define I915_PERF_OA_SAMPLE_SIZE (8 +   /* drm_i915_perf_record_header */ \
  87                                   256)  /* OA counter report */
  88
  89 struct gen_perf_query_result {
  90    /**
  91     * Storage for the final accumulated OA counters.
  92     */
  93    uint64_t accumulator[MAX_OA_REPORT_COUNTERS];
  94
  95    /**
  96     * Hw ID used by the context on which the query was running.
  97     */
  98    uint32_t hw_id;
  99
 100    /**
 101     * Number of reports accumulated to produce the results.
 102     */
 103    uint32_t reports_accumulated;
 104
 105    /**
 106     * Frequency in the slices of the GT at the begin and end of the
 107     * query.
 108     */
 109    uint64_t slice_frequency[2];
 110
 111    /**
 112     * Frequency in the unslice of the GT at the begin and end of the
 113     * query.
 114     */
 115    uint64_t unslice_frequency[2];
 116 };
 117
 118 struct gen_perf_query_counter {
 119    const char *name;
 120    const char *desc;
 121    enum gen_perf_counter_type type;
 122    enum gen_perf_counter_data_type data_type;
 123    uint64_t raw_max;
 124    size_t offset;
 125
 126    union {
 127       uint64_t (*oa_counter_read_uint64)(struct gen_perf_config *perf,
 128                                          const struct gen_perf_query_info *query,
 129                                          const uint64_t *accumulator);
 130       float (*oa_counter_read_float)(struct gen_perf_config *perf,
 131                                      const struct gen_perf_query_info *query,
 132                                      const uint64_t *accumulator);
 133       struct gen_pipeline_stat pipeline_stat;
 134    };
 135 };
 136
 137 struct gen_perf_query_register_prog {
 138    uint32_t reg;
 139    uint32_t val;
 140 };
 141
 142 /* Register programming for a given query */
 143 struct gen_perf_configuration {
 144    struct gen_perf_query_register_prog *flex_regs;
 145    uint32_t n_flex_regs;
 146
 147    struct gen_perf_query_register_prog *mux_regs;
 148    uint32_t n_mux_regs;
 149
 150    struct gen_perf_query_register_prog *b_counter_regs;
 151    uint32_t n_b_counter_regs;
 152 };
 153
 154 struct gen_perf_query_info {
 155    enum gen_perf_query_type {
 156       GEN_PERF_QUERY_TYPE_OA,
 157       GEN_PERF_QUERY_TYPE_RAW,
 158       GEN_PERF_QUERY_TYPE_PIPELINE,
 159    } kind;
 160    const char *name;
 161    const char *guid;
 162    struct gen_perf_query_counter *counters;
 163    int n_counters;
 164    int max_counters;
 165    size_t data_size;
 166
 167    /* OA specific */
 168    uint64_t oa_metrics_set_id;
 169    int oa_format;
 170
 171    /* For indexing into the accumulator[] ... */
 172    int gpu_time_offset;
 173    int gpu_clock_offset;
 174    int a_offset;
 175    int b_offset;
 176    int c_offset;
 177
 178    struct gen_perf_configuration config;
 179 };
 180
 181 struct gen_perf_config {
 182    struct gen_perf_query_info *queries;
 183    int n_queries;
 184
 185    /* Variables referenced in the XML meta data for OA performance
 186     * counters, e.g in the normalization equations.
 187     *
 188     * All uint64_t for consistent operand types in generated code
 189     */
 190    struct {
 191       uint64_t timestamp_frequency; /** $GpuTimestampFrequency */
 192       uint64_t n_eus;               /** $EuCoresTotalCount */
 193       uint64_t n_eu_slices;         /** $EuSlicesTotalCount */
 194       uint64_t n_eu_sub_slices;     /** $EuSubslicesTotalCount */
 195       uint64_t eu_threads_count;    /** $EuThreadsCount */
 196       uint64_t slice_mask;          /** $SliceMask */
 197       uint64_t subslice_mask;       /** $SubsliceMask */
 198       uint64_t gt_min_freq;         /** $GpuMinFrequency */
 199       uint64_t gt_max_freq;         /** $GpuMaxFrequency */
 200       uint64_t revision;            /** $SkuRevisionId */
 201    } sys_vars;
 202
 203    /* OA metric sets, indexed by GUID, as know by Mesa at build time, to
 204     * cross-reference with the GUIDs of configs advertised by the kernel at
 205     * runtime
 206     */
 207    struct hash_table *oa_metrics_table;
 208
 209    /* Location of the device's sysfs entry. */
 210    char sysfs_dev_dir[256];
 211
 212    struct {
 213       void *(*bo_alloc)(void *bufmgr, const char *name, uint64_t size);
 214       void (*bo_unreference)(void *bo);
 215       void *(*bo_map)(void *ctx, void *bo, unsigned flags);
 216       void (*bo_unmap)(void *bo);
 217       bool (*batch_references)(void *batch, void *bo);
 218       void (*bo_wait_rendering)(void *bo);
 219       int (*bo_busy)(void *bo);
 220       void (*emit_mi_flush)(void *ctx);
 221       void (*emit_mi_report_perf_count)(void *ctx,
 222                                         void *bo,
 223                                         uint32_t offset_in_bytes,
 224                                         uint32_t report_id);
 225       void (*batchbuffer_flush)(void *ctx,
 226                                 const char *file, int line);
 227       void (*capture_frequency_stat_register)(void *ctx, void *bo,
 228                                               uint32_t bo_offset);
 229       void (*store_register_mem64)(void *ctx, void *bo, uint32_t reg, uint32_t offset);
 230
 231    } vtbl;
 232 };
 233
 234 struct gen_perf_query_object;
 235 const struct gen_perf_query_info* gen_perf_query_info(const struct gen_perf_query_object *);
 236
 237 void gen_perf_init_metrics(struct gen_perf_config *perf_cfg,
 238                            const struct gen_device_info *devinfo,
 239                            int drm_fd);
 240
 241 /** Query i915 for a metric id using guid.
 242  */
 243 bool gen_perf_load_metric_id(struct gen_perf_config *perf_cfg,
 244                              const char *guid,
 245                              uint64_t *metric_id);
 246
 247 /** Read the slice/unslice frequency from 2 OA reports and store then into
 248  *  result.
 249  */
 250 void gen_perf_query_result_read_frequencies(struct gen_perf_query_result *result,
 251                                             const struct gen_device_info *devinfo,
 252                                             const uint32_t *start,
 253                                             const uint32_t *end);
 254 /** Accumulate the delta between 2 OA reports into result for a given query.
 255  */
 256 void gen_perf_query_result_accumulate(struct gen_perf_query_result *result,
 257                                       const struct gen_perf_query_info *query,
 258                                       const uint32_t *start,
 259                                       const uint32_t *end);
 260 void gen_perf_query_result_clear(struct gen_perf_query_result *result);
 261
 262 struct gen_perf_context;
 263 struct gen_perf_context *gen_perf_new_context(void *parent);
 264
 265 void gen_perf_init_context(struct gen_perf_context *perf_ctx,
 266                            struct gen_perf_config *perf_cfg,
 267                            void * ctx,  /* driver context (eg, brw_context) */
 268                            void * bufmgr,  /* eg brw_bufmgr */
 269                            const struct gen_device_info *devinfo,
 270                            uint32_t hw_ctx,
 271                            int drm_fd);
 272
 273 struct gen_perf_config *gen_perf_config(struct gen_perf_context *ctx);
 274
 275 int gen_perf_active_queries(struct gen_perf_context *perf_ctx,
 276                             const struct gen_perf_query_info *query);
 277
 278 static inline size_t
 279 gen_perf_query_counter_get_size(const struct gen_perf_query_counter *counter)
 280 {
 281    switch (counter->data_type) {
 282    case GEN_PERF_COUNTER_DATA_TYPE_BOOL32:
 283       return sizeof(uint32_t);
 284    case GEN_PERF_COUNTER_DATA_TYPE_UINT32:
 285       return sizeof(uint32_t);
 286    case GEN_PERF_COUNTER_DATA_TYPE_UINT64:
 287       return sizeof(uint64_t);
 288    case GEN_PERF_COUNTER_DATA_TYPE_FLOAT:
 289       return sizeof(float);
 290    case GEN_PERF_COUNTER_DATA_TYPE_DOUBLE:
 291       return sizeof(double);
 292    default:
 293       unreachable("invalid counter data type");
 294    }
 295 }
 296
 297 static inline struct gen_perf_config *
 298 gen_perf_new(void *ctx)
 299 {
 300    struct gen_perf_config *perf = rzalloc(ctx, struct gen_perf_config);
 301    return perf;
 302 }
 303
 304 struct gen_perf_query_object *
 305 gen_perf_new_query(struct gen_perf_context *, unsigned query_index);
 306
 307
 308 bool gen_perf_begin_query(struct gen_perf_context *perf_ctx,
 309                           struct gen_perf_query_object *query);
 310 void gen_perf_end_query(struct gen_perf_context *perf_ctx,
 311                         struct gen_perf_query_object *query);
 312 void gen_perf_wait_query(struct gen_perf_context *perf_ctx,
 313                          struct gen_perf_query_object *query,
 314                          void *current_batch);
 315 bool gen_perf_is_query_ready(struct gen_perf_context *perf_ctx,
 316                              struct gen_perf_query_object *query,
 317                              void *current_batch);
 318 void gen_perf_delete_query(struct gen_perf_context *perf_ctx,
 319                            struct gen_perf_query_object *query);
 320 void gen_perf_get_query_data(struct gen_perf_context *perf_ctx,
 321                              struct gen_perf_query_object *query,
 322                              int data_size,
 323                              unsigned *data,
 324                              unsigned *bytes_written);
 325
 326 void gen_perf_dump_query_count(struct gen_perf_context *perf_ctx);
 327 void gen_perf_dump_query(struct gen_perf_context *perf_ctx,
 328                          struct gen_perf_query_object *obj,
 329                          void *current_batch);
 330
 331 #endif /* GEN_PERF_H */