intel/perf: move initialization of pipeline statistics metrics to gen_perf

Reviewed-by: Kenneth Graunke <[email protected]>

2 files changed, 219 insertions, 124 deletions

diff --git a/src/intel/perf/gen_perf.c b/src/intel/perf/gen_perf.c
index 5fb6044a859..4e29bc17c0a 100644
--- a/src/intel/perf/gen_perf.c
+++ b/src/intel/perf/gen_perf.c
@@ -61,6 +61,11 @@
 #define  GEN9_RPSTAT0_PREV_GT_FREQ_SHIFT   0
 #define  GEN9_RPSTAT0_PREV_GT_FREQ_MASK    INTEL_MASK(8, 0)
 
+#define GEN6_SO_PRIM_STORAGE_NEEDED     0x2280
+#define GEN7_SO_PRIM_STORAGE_NEEDED(n)  (0x5240 + (n) * 8)
+#define GEN6_SO_NUM_PRIMS_WRITTEN       0x2288
+#define GEN7_SO_NUM_PRIMS_WRITTEN(n)    (0x5200 + (n) * 8)
+
 #define MAP_READ  (1 << 0)
 #define MAP_WRITE (1 << 1)
 
@@ -165,13 +170,32 @@ read_sysfs_drm_device_file_uint64(struct gen_perf_config *perf,
    return read_file_uint64(buf, value);
 }
 
+static inline struct gen_perf_query_info *
+append_query_info(struct gen_perf_config *perf, int max_counters)
+{
+   struct gen_perf_query_info *query;
+
+   perf->queries = reralloc(perf, perf->queries,
+                            struct gen_perf_query_info,
+                            ++perf->n_queries);
+   query = &perf->queries[perf->n_queries - 1];
+   memset(query, 0, sizeof(*query));
+
+   if (max_counters > 0) {
+      query->max_counters = max_counters;
+      query->counters =
+         rzalloc_array(perf, struct gen_perf_query_counter, max_counters);
+   }
+
+   return query;
+}
+
 static void
 register_oa_config(struct gen_perf_config *perf,
                    const struct gen_perf_query_info *query,
                    uint64_t config_id)
 {
-   struct gen_perf_query_info *registred_query =
-      gen_perf_query_append_query_info(perf, 0);
+   struct gen_perf_query_info *registred_query = append_query_info(perf, 0);
 
    *registred_query = *query;
    registred_query->oa_metrics_set_id = config_id;
@@ -395,8 +419,123 @@ get_register_queries_function(const struct gen_device_info *devinfo)
    return NULL;
 }
 
-bool
-gen_perf_load_oa_metrics(struct gen_perf_config *perf, int fd,
+static inline void
+add_stat_reg(struct gen_perf_query_info *query, uint32_t reg,
+             uint32_t numerator, uint32_t denominator,
+             const char *name, const char *description)
+{
+   struct gen_perf_query_counter *counter;
+
+   assert(query->n_counters < query->max_counters);
+
+   counter = &query->counters[query->n_counters];
+   counter->name = name;
+   counter->desc = description;
+   counter->type = GEN_PERF_COUNTER_TYPE_RAW;
+   counter->data_type = GEN_PERF_COUNTER_DATA_TYPE_UINT64;
+   counter->offset = sizeof(uint64_t) * query->n_counters;
+   counter->pipeline_stat.reg = reg;
+   counter->pipeline_stat.numerator = numerator;
+   counter->pipeline_stat.denominator = denominator;
+
+   query->n_counters++;
+}
+
+static inline void
+add_basic_stat_reg(struct gen_perf_query_info *query,
+                                       uint32_t reg, const char *name)
+{
+   add_stat_reg(query, reg, 1, 1, name, name);
+}
+
+static void
+load_pipeline_statistic_metrics(struct gen_perf_config *perf_cfg,
+                                         const struct gen_device_info *devinfo)
+{
+   struct gen_perf_query_info *query =
+      append_query_info(perf_cfg, MAX_STAT_COUNTERS);
+
+   query->kind = GEN_PERF_QUERY_TYPE_PIPELINE;
+   query->name = "Pipeline Statistics Registers";
+
+   add_basic_stat_reg(query, IA_VERTICES_COUNT,
+                                          "N vertices submitted");
+   add_basic_stat_reg(query, IA_PRIMITIVES_COUNT,
+                                          "N primitives submitted");
+   add_basic_stat_reg(query, VS_INVOCATION_COUNT,
+                                          "N vertex shader invocations");
+
+   if (devinfo->gen == 6) {
+      add_stat_reg(query, GEN6_SO_PRIM_STORAGE_NEEDED, 1, 1,
+                   "SO_PRIM_STORAGE_NEEDED",
+                   "N geometry shader stream-out primitives (total)");
+      add_stat_reg(query, GEN6_SO_NUM_PRIMS_WRITTEN, 1, 1,
+                   "SO_NUM_PRIMS_WRITTEN",
+                   "N geometry shader stream-out primitives (written)");
+   } else {
+      add_stat_reg(query, GEN7_SO_PRIM_STORAGE_NEEDED(0), 1, 1,
+                   "SO_PRIM_STORAGE_NEEDED (Stream 0)",
+                   "N stream-out (stream 0) primitives (total)");
+      add_stat_reg(query, GEN7_SO_PRIM_STORAGE_NEEDED(1), 1, 1,
+                   "SO_PRIM_STORAGE_NEEDED (Stream 1)",
+                   "N stream-out (stream 1) primitives (total)");
+      add_stat_reg(query, GEN7_SO_PRIM_STORAGE_NEEDED(2), 1, 1,
+                   "SO_PRIM_STORAGE_NEEDED (Stream 2)",
+                   "N stream-out (stream 2) primitives (total)");
+      add_stat_reg(query, GEN7_SO_PRIM_STORAGE_NEEDED(3), 1, 1,
+                   "SO_PRIM_STORAGE_NEEDED (Stream 3)",
+                   "N stream-out (stream 3) primitives (total)");
+      add_stat_reg(query, GEN7_SO_NUM_PRIMS_WRITTEN(0), 1, 1,
+                   "SO_NUM_PRIMS_WRITTEN (Stream 0)",
+                   "N stream-out (stream 0) primitives (written)");
+      add_stat_reg(query, GEN7_SO_NUM_PRIMS_WRITTEN(1), 1, 1,
+                   "SO_NUM_PRIMS_WRITTEN (Stream 1)",
+                   "N stream-out (stream 1) primitives (written)");
+      add_stat_reg(query, GEN7_SO_NUM_PRIMS_WRITTEN(2), 1, 1,
+                   "SO_NUM_PRIMS_WRITTEN (Stream 2)",
+                   "N stream-out (stream 2) primitives (written)");
+      add_stat_reg(query, GEN7_SO_NUM_PRIMS_WRITTEN(3), 1, 1,
+                   "SO_NUM_PRIMS_WRITTEN (Stream 3)",
+                   "N stream-out (stream 3) primitives (written)");
+   }
+
+   add_basic_stat_reg(query, HS_INVOCATION_COUNT,
+                                          "N TCS shader invocations");
+   add_basic_stat_reg(query, DS_INVOCATION_COUNT,
+                                          "N TES shader invocations");
+
+   add_basic_stat_reg(query, GS_INVOCATION_COUNT,
+                                          "N geometry shader invocations");
+   add_basic_stat_reg(query, GS_PRIMITIVES_COUNT,
+                                          "N geometry shader primitives emitted");
+
+   add_basic_stat_reg(query, CL_INVOCATION_COUNT,
+                                          "N primitives entering clipping");
+   add_basic_stat_reg(query, CL_PRIMITIVES_COUNT,
+                                          "N primitives leaving clipping");
+
+   if (devinfo->is_haswell || devinfo->gen == 8) {
+      add_stat_reg(query, PS_INVOCATION_COUNT, 1, 4,
+                   "N fragment shader invocations",
+                   "N fragment shader invocations");
+   } else {
+      add_basic_stat_reg(query, PS_INVOCATION_COUNT,
+                                             "N fragment shader invocations");
+   }
+
+   add_basic_stat_reg(query, PS_DEPTH_COUNT,
+                                          "N z-pass fragments");
+
+   if (devinfo->gen >= 7) {
+      add_basic_stat_reg(query, CS_INVOCATION_COUNT,
+                                             "N compute shader invocations");
+   }
+
+   query->data_size = sizeof(uint64_t) * query->n_counters;
+}
+
+static bool
+load_oa_metrics(struct gen_perf_config *perf, int fd,
                          const struct gen_device_info *devinfo)
 {
    perf_register_oa_queries_t oa_register = get_register_queries_function(devinfo);
@@ -583,6 +722,62 @@ gen_perf_query_result_clear(struct gen_perf_query_result *result)
 }
 
 static void
+gen_perf_query_register_mdapi_statistic_query(struct gen_perf_config *perf_cfg,
+                                              const struct gen_device_info *devinfo)
+{
+   if (!(devinfo->gen >= 7 && devinfo->gen <= 11))
+      return;
+
+   struct gen_perf_query_info *query =
+      append_query_info(perf_cfg, MAX_STAT_COUNTERS);
+
+   query->kind = GEN_PERF_QUERY_TYPE_PIPELINE;
+   query->name = "Intel_Raw_Pipeline_Statistics_Query";
+
+   /* The order has to match mdapi_pipeline_metrics. */
+   add_basic_stat_reg(query, IA_VERTICES_COUNT,
+                      "N vertices submitted");
+   add_basic_stat_reg(query, IA_PRIMITIVES_COUNT,
+                      "N primitives submitted");
+   add_basic_stat_reg(query, VS_INVOCATION_COUNT,
+                      "N vertex shader invocations");
+   add_basic_stat_reg(query, GS_INVOCATION_COUNT,
+                      "N geometry shader invocations");
+   add_basic_stat_reg(query, GS_PRIMITIVES_COUNT,
+                      "N geometry shader primitives emitted");
+   add_basic_stat_reg(query, CL_INVOCATION_COUNT,
+                      "N primitives entering clipping");
+   add_basic_stat_reg(query, CL_PRIMITIVES_COUNT,
+                      "N primitives leaving clipping");
+   if (devinfo->is_haswell || devinfo->gen == 8) {
+      add_stat_reg(query, PS_INVOCATION_COUNT, 1, 4,
+                   "N fragment shader invocations",
+                   "N fragment shader invocations");
+   } else {
+      add_basic_stat_reg(query, PS_INVOCATION_COUNT,
+                         "N fragment shader invocations");
+   }
+   add_basic_stat_reg(query, HS_INVOCATION_COUNT,
+                      "N TCS shader invocations");
+   add_basic_stat_reg(query, DS_INVOCATION_COUNT,
+                      "N TES shader invocations");
+   if (devinfo->gen >= 7) {
+      add_basic_stat_reg(query, CS_INVOCATION_COUNT,
+                         "N compute shader invocations");
+   }
+
+   if (devinfo->gen >= 10) {
+      /* Reuse existing CS invocation register until we can expose this new
+       * one.
+       */
+      add_basic_stat_reg(query, CS_INVOCATION_COUNT,
+                         "Reserved1");
+   }
+
+   query->data_size = sizeof(uint64_t) * query->n_counters;
+}
+
+static void
 fill_mdapi_perf_query_counter(struct gen_perf_query_info *query,
                               const char *name,
                               uint32_t data_offset,
@@ -618,9 +813,9 @@ fill_mdapi_perf_query_counter(struct gen_perf_query_info *query,
                                  sizeof(struct_name.field_name[0]),     \
                                  GEN_PERF_COUNTER_DATA_TYPE_##type_name)
 
-void
-gen_perf_query_register_mdapi_oa_query(const struct gen_device_info *devinfo,
-                                       struct gen_perf_config *perf)
+static void
+register_mdapi_oa_query(const struct gen_device_info *devinfo,
+                        struct gen_perf_config *perf)
 {
    struct gen_perf_query_info *query = NULL;
 
@@ -632,7 +827,7 @@ gen_perf_query_register_mdapi_oa_query(const struct gen_device_info *devinfo,
 
    switch (devinfo->gen) {
    case 7: {
-      query = gen_perf_query_append_query_info(perf, 1 + 45 + 16 + 7);
+      query = append_query_info(perf, 1 + 45 + 16 + 7);
       query->oa_format = I915_OA_FORMAT_A45_B8_C8;
 
       struct gen7_mdapi_metrics metric_data;
@@ -657,7 +852,7 @@ gen_perf_query_register_mdapi_oa_query(const struct gen_device_info *devinfo,
       break;
    }
    case 8: {
-      query = gen_perf_query_append_query_info(perf, 2 + 36 + 16 + 16);
+      query = append_query_info(perf, 2 + 36 + 16 + 16);
       query->oa_format = I915_OA_FORMAT_A32u40_A4u32_B8_C8;
 
       struct gen8_mdapi_metrics metric_data;
@@ -694,7 +889,7 @@ gen_perf_query_register_mdapi_oa_query(const struct gen_device_info *devinfo,
    case 9:
    case 10:
    case 11: {
-      query = gen_perf_query_append_query_info(perf, 2 + 36 + 16 + 16 + 16 + 2);
+      query = append_query_info(perf, 2 + 36 + 16 + 16 + 16 + 2);
       query->oa_format = I915_OA_FORMAT_A32u40_A4u32_B8_C8;
 
       struct gen9_mdapi_metrics metric_data;
@@ -756,62 +951,6 @@ gen_perf_query_register_mdapi_oa_query(const struct gen_device_info *devinfo,
    }
 }
 
-void
-gen_perf_query_register_mdapi_statistic_query(const struct gen_device_info *devinfo,
-                                              struct gen_perf_config *perf)
-{
-   if (!(devinfo->gen >= 7 && devinfo->gen <= 11))
-      return;
-
-   struct gen_perf_query_info *query =
-      gen_perf_query_append_query_info(perf, MAX_STAT_COUNTERS);
-
-   query->kind = GEN_PERF_QUERY_TYPE_PIPELINE;
-   query->name = "Intel_Raw_Pipeline_Statistics_Query";
-
-   /* The order has to match mdapi_pipeline_metrics. */
-   gen_perf_query_info_add_basic_stat_reg(query, IA_VERTICES_COUNT,
-                                          "N vertices submitted");
-   gen_perf_query_info_add_basic_stat_reg(query, IA_PRIMITIVES_COUNT,
-                                          "N primitives submitted");
-   gen_perf_query_info_add_basic_stat_reg(query, VS_INVOCATION_COUNT,
-                                          "N vertex shader invocations");
-   gen_perf_query_info_add_basic_stat_reg(query, GS_INVOCATION_COUNT,
-                                          "N geometry shader invocations");
-   gen_perf_query_info_add_basic_stat_reg(query, GS_PRIMITIVES_COUNT,
-                                          "N geometry shader primitives emitted");
-   gen_perf_query_info_add_basic_stat_reg(query, CL_INVOCATION_COUNT,
-                                          "N primitives entering clipping");
-   gen_perf_query_info_add_basic_stat_reg(query, CL_PRIMITIVES_COUNT,
-                                          "N primitives leaving clipping");
-   if (devinfo->is_haswell || devinfo->gen == 8) {
-      gen_perf_query_info_add_stat_reg(query, PS_INVOCATION_COUNT, 1, 4,
-                                       "N fragment shader invocations",
-                                       "N fragment shader invocations");
-   } else {
-      gen_perf_query_info_add_basic_stat_reg(query, PS_INVOCATION_COUNT,
-                                             "N fragment shader invocations");
-   }
-   gen_perf_query_info_add_basic_stat_reg(query, HS_INVOCATION_COUNT,
-                                          "N TCS shader invocations");
-   gen_perf_query_info_add_basic_stat_reg(query, DS_INVOCATION_COUNT,
-                                          "N TES shader invocations");
-   if (devinfo->gen >= 7) {
-      gen_perf_query_info_add_basic_stat_reg(query, CS_INVOCATION_COUNT,
-                                             "N compute shader invocations");
-   }
-
-   if (devinfo->gen >= 10) {
-      /* Reuse existing CS invocation register until we can expose this new
-       * one.
-       */
-      gen_perf_query_info_add_basic_stat_reg(query, CS_INVOCATION_COUNT,
-                                             "Reserved1");
-   }
-
-   query->data_size = sizeof(uint64_t) * query->n_counters;
-}
-
 uint64_t
 gen_perf_query_get_metric_id(struct gen_perf_config *perf,
                              const struct gen_perf_query_info *query)
@@ -1012,6 +1151,17 @@ gen_perf_dec_n_users(struct gen_perf_context *perf_ctx)
 }
 
 void
+gen_perf_init_metrics(struct gen_perf_config *perf_cfg,
+                      const struct gen_device_info *devinfo,
+                      int drm_fd)
+{
+   load_pipeline_statistic_metrics(perf_cfg, devinfo);
+   gen_perf_query_register_mdapi_statistic_query(perf_cfg, devinfo);
+   if (load_oa_metrics(perf_cfg, drm_fd, devinfo))
+      register_mdapi_oa_query(devinfo, perf_cfg);
+}
+
+void
 gen_perf_init_context(struct gen_perf_context *perf_ctx,
                       struct gen_perf_config *perf_cfg,
                       void * ctx,  /* driver context (eg, brw_context) */
diff --git a/src/intel/perf/gen_perf.h b/src/intel/perf/gen_perf.h
index f66a088e260..68e2a985bad 100644
--- a/src/intel/perf/gen_perf.h
+++ b/src/intel/perf/gen_perf.h
@@ -505,6 +505,9 @@ struct gen_perf_context {
    int n_query_instances;
 };
 
+void gen_perf_init_metrics(struct gen_perf_config *perf_cfg,
+                           const struct gen_device_info *devinfo,
+                           int drm_fd);
 void gen_perf_init_context(struct gen_perf_context *perf_ctx,
                            struct gen_perf_config *perf_cfg,
                            void * ctx,  /* driver context (eg, brw_context) */
@@ -532,58 +535,6 @@ gen_perf_query_counter_get_size(const struct gen_perf_query_counter *counter)
    }
 }
 
-static inline struct gen_perf_query_info *
-gen_perf_query_append_query_info(struct gen_perf_config *perf, int max_counters)
-{
-   struct gen_perf_query_info *query;
-
-   perf->queries = reralloc(perf, perf->queries,
-                            struct gen_perf_query_info,
-                            ++perf->n_queries);
-   query = &perf->queries[perf->n_queries - 1];
-   memset(query, 0, sizeof(*query));
-
-   if (max_counters > 0) {
-      query->max_counters = max_counters;
-      query->counters =
-         rzalloc_array(perf, struct gen_perf_query_counter, max_counters);
-   }
-
-   return query;
-}
-
-static inline void
-gen_perf_query_info_add_stat_reg(struct gen_perf_query_info *query,
-                                 uint32_t reg,
-                                 uint32_t numerator,
-                                 uint32_t denominator,
-                                 const char *name,
-                                 const char *description)
-{
-   struct gen_perf_query_counter *counter;
-
-   assert(query->n_counters < query->max_counters);
-
-   counter = &query->counters[query->n_counters];
-   counter->name = name;
-   counter->desc = description;
-   counter->type = GEN_PERF_COUNTER_TYPE_RAW;
-   counter->data_type = GEN_PERF_COUNTER_DATA_TYPE_UINT64;
-   counter->offset = sizeof(uint64_t) * query->n_counters;
-   counter->pipeline_stat.reg = reg;
-   counter->pipeline_stat.numerator = numerator;
-   counter->pipeline_stat.denominator = denominator;
-
-   query->n_counters++;
-}
-
-static inline void
-gen_perf_query_info_add_basic_stat_reg(struct gen_perf_query_info *query,
-                                       uint32_t reg, const char *name)
-{
-   gen_perf_query_info_add_stat_reg(query, reg, 1, 1, name, name);
-}
-
 static inline struct gen_perf_config *
 gen_perf_new(void *ctx)
 {
@@ -591,8 +542,6 @@ gen_perf_new(void *ctx)
    return perf;
 }
 
-bool gen_perf_load_oa_metrics(struct gen_perf_config *perf, int fd,
-                              const struct gen_device_info *devinfo);
 bool gen_perf_load_metric_id(struct gen_perf_config *perf, const char *guid,
                              uint64_t *metric_id);
 
@@ -605,10 +554,6 @@ void gen_perf_query_result_accumulate(struct gen_perf_query_result *result,
                                       const uint32_t *start,
                                       const uint32_t *end);
 void gen_perf_query_result_clear(struct gen_perf_query_result *result);
-void gen_perf_query_register_mdapi_statistic_query(const struct gen_device_info *devinfo,
-                                                   struct gen_perf_config *perf);
-void gen_perf_query_register_mdapi_oa_query(const struct gen_device_info *devinfo,
-                                            struct gen_perf_config *perf);
 uint64_t gen_perf_query_get_metric_id(struct gen_perf_config *perf,
                                       const struct gen_perf_query_info *query);
 struct oa_sample_buf * gen_perf_get_free_sample_buf(struct gen_perf_context *perf);