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 #include <sys/sysmacros.h>
  32
  33 #include "util/hash_table.h"
  34 #include "util/ralloc.h"
  35
  36 struct gen_device_info;
  37
  38 struct gen_perf;
  39 struct gen_perf_query_info;
  40
  41 enum gen_perf_counter_type {
  42    GEN_PERF_COUNTER_TYPE_EVENT,
  43    GEN_PERF_COUNTER_TYPE_DURATION_NORM,
  44    GEN_PERF_COUNTER_TYPE_DURATION_RAW,
  45    GEN_PERF_COUNTER_TYPE_THROUGHPUT,
  46    GEN_PERF_COUNTER_TYPE_RAW,
  47    GEN_PERF_COUNTER_TYPE_TIMESTAMP,
  48 };
  49
  50 enum gen_perf_counter_data_type {
  51    GEN_PERF_COUNTER_DATA_TYPE_BOOL32,
  52    GEN_PERF_COUNTER_DATA_TYPE_UINT32,
  53    GEN_PERF_COUNTER_DATA_TYPE_UINT64,
  54    GEN_PERF_COUNTER_DATA_TYPE_FLOAT,
  55    GEN_PERF_COUNTER_DATA_TYPE_DOUBLE,
  56 };
  57
  58 struct gen_pipeline_stat {
  59    uint32_t reg;
  60    uint32_t numerator;
  61    uint32_t denominator;
  62 };
  63
  64 /*
  65  * The largest OA formats we can use include:
  66  * For Haswell:
  67  *   1 timestamp, 45 A counters, 8 B counters and 8 C counters.
  68  * For Gen8+
  69  *   1 timestamp, 1 clock, 36 A counters, 8 B counters and 8 C counters
  70  */
  71 #define MAX_OA_REPORT_COUNTERS 62
  72
  73 struct gen_perf_query_result {
  74    /**
  75     * Storage for the final accumulated OA counters.
  76     */
  77    uint64_t accumulator[MAX_OA_REPORT_COUNTERS];
  78
  79    /**
  80     * Hw ID used by the context on which the query was running.
  81     */
  82    uint32_t hw_id;
  83
  84    /**
  85     * Number of reports accumulated to produce the results.
  86     */
  87    uint32_t reports_accumulated;
  88
  89    /**
  90     * Frequency in the slices of the GT at the begin and end of the
  91     * query.
  92     */
  93    uint64_t slice_frequency[2];
  94
  95    /**
  96     * Frequency in the unslice of the GT at the begin and end of the
  97     * query.
  98     */
  99    uint64_t unslice_frequency[2];
 100 };
 101
 102 struct gen_perf_query_counter {
 103    const char *name;
 104    const char *desc;
 105    enum gen_perf_counter_type type;
 106    enum gen_perf_counter_data_type data_type;
 107    uint64_t raw_max;
 108    size_t offset;
 109    size_t size;
 110
 111    union {
 112       uint64_t (*oa_counter_read_uint64)(struct gen_perf *perf,
 113                                          const struct gen_perf_query_info *query,
 114                                          uint64_t *accumulator);
 115       float (*oa_counter_read_float)(struct gen_perf *perf,
 116                                      const struct gen_perf_query_info *query,
 117                                      uint64_t *accumulator);
 118       struct gen_pipeline_stat pipeline_stat;
 119    };
 120 };
 121
 122 struct gen_perf_query_register_prog {
 123    uint32_t reg;
 124    uint32_t val;
 125 };
 126
 127 struct gen_perf_query_info {
 128    enum gen_perf_query_type {
 129       GEN_PERF_QUERY_TYPE_OA,
 130       GEN_PERF_QUERY_TYPE_RAW,
 131       GEN_PERF_QUERY_TYPE_PIPELINE,
 132    } kind;
 133    const char *name;
 134    const char *guid;
 135    struct gen_perf_query_counter *counters;
 136    int n_counters;
 137    int max_counters;
 138    size_t data_size;
 139
 140    /* OA specific */
 141    uint64_t oa_metrics_set_id;
 142    int oa_format;
 143
 144    /* For indexing into the accumulator[] ... */
 145    int gpu_time_offset;
 146    int gpu_clock_offset;
 147    int a_offset;
 148    int b_offset;
 149    int c_offset;
 150
 151    /* Register programming for a given query */
 152    struct gen_perf_query_register_prog *flex_regs;
 153    uint32_t n_flex_regs;
 154
 155    struct gen_perf_query_register_prog *mux_regs;
 156    uint32_t n_mux_regs;
 157
 158    struct gen_perf_query_register_prog *b_counter_regs;
 159    uint32_t n_b_counter_regs;
 160 };
 161
 162 struct gen_perf {
 163    struct gen_perf_query_info *queries;
 164    int n_queries;
 165
 166    /* Variables referenced in the XML meta data for OA performance
 167     * counters, e.g in the normalization equations.
 168     *
 169     * All uint64_t for consistent operand types in generated code
 170     */
 171    struct {
 172       uint64_t timestamp_frequency; /** $GpuTimestampFrequency */
 173       uint64_t n_eus;               /** $EuCoresTotalCount */
 174       uint64_t n_eu_slices;         /** $EuSlicesTotalCount */
 175       uint64_t n_eu_sub_slices;     /** $EuSubslicesTotalCount */
 176       uint64_t eu_threads_count;    /** $EuThreadsCount */
 177       uint64_t slice_mask;          /** $SliceMask */
 178       uint64_t subslice_mask;       /** $SubsliceMask */
 179       uint64_t gt_min_freq;         /** $GpuMinFrequency */
 180       uint64_t gt_max_freq;         /** $GpuMaxFrequency */
 181       uint64_t revision;            /** $SkuRevisionId */
 182    } sys_vars;
 183
 184    /* OA metric sets, indexed by GUID, as know by Mesa at build time, to
 185     * cross-reference with the GUIDs of configs advertised by the kernel at
 186     * runtime
 187     */
 188    struct hash_table *oa_metrics_table;
 189
 190    /* Location of the device's sysfs entry. */
 191    char sysfs_dev_dir[256];
 192
 193    int (*ioctl)(int, unsigned long, void *);
 194 };
 195
 196 static inline struct gen_perf_query_info *
 197 gen_perf_query_append_query_info(struct gen_perf *perf, int max_counters)
 198 {
 199    struct gen_perf_query_info *query;
 200
 201    perf->queries = reralloc(perf, perf->queries,
 202                             struct gen_perf_query_info,
 203                             ++perf->n_queries);
 204    query = &perf->queries[perf->n_queries - 1];
 205    memset(query, 0, sizeof(*query));
 206
 207    if (max_counters > 0) {
 208       query->max_counters = max_counters;
 209       query->counters =
 210          rzalloc_array(perf, struct gen_perf_query_counter, max_counters);
 211    }
 212
 213    return query;
 214 }
 215
 216 static inline void
 217 gen_perf_query_info_add_stat_reg(struct gen_perf_query_info *query,
 218                                  uint32_t reg,
 219                                  uint32_t numerator,
 220                                  uint32_t denominator,
 221                                  const char *name,
 222                                  const char *description)
 223 {
 224    struct gen_perf_query_counter *counter;
 225
 226    assert(query->n_counters < query->max_counters);
 227
 228    counter = &query->counters[query->n_counters];
 229    counter->name = name;
 230    counter->desc = description;
 231    counter->type = GEN_PERF_COUNTER_TYPE_RAW;
 232    counter->data_type = GEN_PERF_COUNTER_DATA_TYPE_UINT64;
 233    counter->size = sizeof(uint64_t);
 234    counter->offset = sizeof(uint64_t) * query->n_counters;
 235    counter->pipeline_stat.reg = reg;
 236    counter->pipeline_stat.numerator = numerator;
 237    counter->pipeline_stat.denominator = denominator;
 238
 239    query->n_counters++;
 240 }
 241
 242 static inline void
 243 gen_perf_query_info_add_basic_stat_reg(struct gen_perf_query_info *query,
 244                                        uint32_t reg, const char *name)
 245 {
 246    gen_perf_query_info_add_stat_reg(query, reg, 1, 1, name, name);
 247 }
 248
 249 static inline struct gen_perf *
 250 gen_perf_new(void *ctx, int (*ioctl_cb)(int, unsigned long, void *))
 251 {
 252    struct gen_perf *perf = rzalloc(ctx, struct gen_perf);
 253
 254    perf->ioctl = ioctl_cb;
 255
 256    return perf;
 257 }
 258
 259 bool gen_perf_load_oa_metrics(struct gen_perf *perf, int fd,
 260                               const struct gen_device_info *devinfo);
 261 bool gen_perf_load_metric_id(struct gen_perf *perf, const char *guid,
 262                              uint64_t *metric_id);
 263
 264 void gen_perf_query_result_read_frequencies(struct gen_perf_query_result *result,
 265                                             const struct gen_device_info *devinfo,
 266                                             const uint32_t *start,
 267                                             const uint32_t *end);
 268 void gen_perf_query_result_accumulate(struct gen_perf_query_result *result,
 269                                       const struct gen_perf_query_info *query,
 270                                       const uint32_t *start,
 271                                       const uint32_t *end);
 272 void gen_perf_query_result_clear(struct gen_perf_query_result *result);
 273
 274
 275 #endif /* GEN_PERF_H */