anv: Record shader compile stats in the pipeline cache

We're going to want these to be available regardless of caching.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

diff --git a/src/intel/vulkan/anv_blorp.c b/src/intel/vulkan/anv_blorp.c
index 5b7981bb49e806aea38ea4d08e21f8f6d1a1cc7e..1ab57fb4e85c6426beb1331a2af7156e274feddf 100644 (file)
--- a/src/intel/vulkan/anv_blorp.c
+++ b/src/intel/vulkan/anv_blorp.c
@@ -74,7 +74,7 @@ upload_blorp_shader(struct blorp_batch *batch,
                                        key, key_size, kernel, kernel_size,
                                        NULL, 0,
                                        prog_data, prog_data_size,
-                                       NULL, &bind_map);
+                                       NULL, 0, NULL, &bind_map);
 
    if (!bin)
       return false;

diff --git a/src/intel/vulkan/anv_pipeline.c b/src/intel/vulkan/anv_pipeline.c
index 5bbd43674957b757b8579feedc0a3a8228124ad8..323a7894e993741ff175a5efa6325b9c7ac16525 100644 (file)
--- a/src/intel/vulkan/anv_pipeline.c
+++ b/src/intel/vulkan/anv_pipeline.c
@@ -524,6 +524,9 @@ struct anv_pipeline_stage {
 
    union brw_any_prog_data prog_data;
 
+   uint32_t num_stats;
+   struct brw_compile_stats stats[3];
+
    VkPipelineCreationFeedbackEXT feedback;
 
    const unsigned *code;
@@ -749,10 +752,12 @@ anv_pipeline_compile_vs(const struct brw_compiler *compiler,
                        vs_stage->nir->info.outputs_written,
                        vs_stage->nir->info.separate_shader);
 
+   vs_stage->num_stats = 1;
    vs_stage->code = brw_compile_vs(compiler, device, mem_ctx,
                                    &vs_stage->key.vs,
                                    &vs_stage->prog_data.vs,
-                                   vs_stage->nir, -1, NULL, NULL);
+                                   vs_stage->nir, -1,
+                                   vs_stage->stats, NULL);
 }
 
 static void
@@ -834,10 +839,12 @@ anv_pipeline_compile_tcs(const struct brw_compiler *compiler,
    tcs_stage->key.tcs.patch_outputs_written =
       tcs_stage->nir->info.patch_outputs_written;
 
+   tcs_stage->num_stats = 1;
    tcs_stage->code = brw_compile_tcs(compiler, device, mem_ctx,
                                      &tcs_stage->key.tcs,
                                      &tcs_stage->prog_data.tcs,
-                                     tcs_stage->nir, -1, NULL, NULL);
+                                     tcs_stage->nir, -1,
+                                     tcs_stage->stats, NULL);
 }
 
 static void
@@ -861,11 +868,13 @@ anv_pipeline_compile_tes(const struct brw_compiler *compiler,
    tes_stage->key.tes.patch_inputs_read =
       tcs_stage->nir->info.patch_outputs_written;
 
+   tes_stage->num_stats = 1;
    tes_stage->code = brw_compile_tes(compiler, device, mem_ctx,
                                      &tes_stage->key.tes,
                                      &tcs_stage->prog_data.tcs.base.vue_map,
                                      &tes_stage->prog_data.tes,
-                                     tes_stage->nir, NULL, -1, NULL, NULL);
+                                     tes_stage->nir, NULL, -1,
+                                     tes_stage->stats, NULL);
 }
 
 static void
@@ -889,10 +898,12 @@ anv_pipeline_compile_gs(const struct brw_compiler *compiler,
                        gs_stage->nir->info.outputs_written,
                        gs_stage->nir->info.separate_shader);
 
+   gs_stage->num_stats = 1;
    gs_stage->code = brw_compile_gs(compiler, device, mem_ctx,
                                    &gs_stage->key.gs,
                                    &gs_stage->prog_data.gs,
-                                   gs_stage->nir, NULL, -1, NULL, NULL);
+                                   gs_stage->nir, NULL, -1,
+                                   gs_stage->stats, NULL);
 }
 
 static void
@@ -1028,7 +1039,12 @@ anv_pipeline_compile_fs(const struct brw_compiler *compiler,
                                    &fs_stage->key.wm,
                                    &fs_stage->prog_data.wm,
                                    fs_stage->nir, NULL, -1, -1, -1,
-                                   true, false, NULL, NULL, NULL);
+                                   true, false, NULL,
+                                   fs_stage->stats, NULL);
+
+   fs_stage->num_stats = (uint32_t)fs_stage->prog_data.wm.dispatch_8 +
+                         (uint32_t)fs_stage->prog_data.wm.dispatch_16 +
+                         (uint32_t)fs_stage->prog_data.wm.dispatch_32;
 
    if (fs_stage->key.wm.nr_color_regions == 0 &&
        !fs_stage->prog_data.wm.has_side_effects &&
@@ -1298,6 +1314,7 @@ anv_pipeline_compile_graphics(struct anv_pipeline *pipeline,
                                   stages[s].nir->constant_data_size,
                                   &stages[s].prog_data.base,
                                   brw_prog_data_size(s),
+                                  stages[s].stats, stages[s].num_stats,
                                   xfb_info, &stages[s].bind_map);
       if (!bin) {
          ralloc_free(stage_ctx);
@@ -1449,9 +1466,10 @@ anv_pipeline_compile_cs(struct anv_pipeline *pipeline,
       NIR_PASS_V(stage.nir, nir_lower_explicit_io,
                  nir_var_mem_shared, nir_address_format_32bit_offset);
 
+      stage.num_stats = 1;
       stage.code = brw_compile_cs(compiler, pipeline->device, mem_ctx,
                                   &stage.key.cs, &stage.prog_data.cs,
-                                  stage.nir, -1, NULL, NULL);
+                                  stage.nir, -1, stage.stats, NULL);
       if (stage.code == NULL) {
          ralloc_free(mem_ctx);
          return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
@@ -1465,6 +1483,7 @@ anv_pipeline_compile_cs(struct anv_pipeline *pipeline,
                                      stage.nir->constant_data_size,
                                      &stage.prog_data.base,
                                      sizeof(stage.prog_data.cs),
+                                     stage.stats, stage.num_stats,
                                      NULL, &stage.bind_map);
       if (!bin) {
          ralloc_free(mem_ctx);

diff --git a/src/intel/vulkan/anv_pipeline_cache.c b/src/intel/vulkan/anv_pipeline_cache.c
index 7b9b1e83678078ec932db6290481f2958552066b..7c9f8a6a8cd0384f1d73bf5e8db67b255226bc7a 100644 (file)
--- a/src/intel/vulkan/anv_pipeline_cache.c
+++ b/src/intel/vulkan/anv_pipeline_cache.c
@@ -37,6 +37,7 @@ anv_shader_bin_create(struct anv_device *device,
                       const void *constant_data, uint32_t constant_data_size,
                       const struct brw_stage_prog_data *prog_data_in,
                       uint32_t prog_data_size, const void *prog_data_param_in,
+                      const struct brw_compile_stats *stats, uint32_t num_stats,
                       const nir_xfb_info *xfb_info_in,
                       const struct anv_pipeline_bind_map *bind_map)
 {
@@ -93,6 +94,10 @@ anv_shader_bin_create(struct anv_device *device,
    shader->prog_data = prog_data;
    shader->prog_data_size = prog_data_size;
 
+   assert(num_stats <= ARRAY_SIZE(shader->stats));
+   typed_memcpy(shader->stats, stats, num_stats);
+   shader->num_stats = num_stats;
+
    if (xfb_info_in) {
       *xfb_info = *xfb_info_in;
       typed_memcpy(xfb_info->outputs, xfb_info_in->outputs,
@@ -143,6 +148,10 @@ anv_shader_bin_write_to_blob(const struct anv_shader_bin *shader,
                     shader->prog_data->nr_params *
                     sizeof(*shader->prog_data->param));
 
+   blob_write_uint32(blob, shader->num_stats);
+   blob_write_bytes(blob, shader->stats,
+                    shader->num_stats * sizeof(shader->stats[0]));
+
    if (shader->xfb_info) {
       uint32_t xfb_info_size =
          nir_xfb_info_size(shader->xfb_info->output_count);
@@ -185,6 +194,10 @@ anv_shader_bin_create_from_blob(struct anv_device *device,
    const void *prog_data_param =
       blob_read_bytes(blob, prog_data->nr_params * sizeof(*prog_data->param));
 
+   uint32_t num_stats = blob_read_uint32(blob);
+   const struct brw_compile_stats *stats =
+      blob_read_bytes(blob, num_stats * sizeof(stats[0]));
+
    const nir_xfb_info *xfb_info = NULL;
    uint32_t xfb_size = blob_read_uint32(blob);
    if (xfb_size)
@@ -208,7 +221,7 @@ anv_shader_bin_create_from_blob(struct anv_device *device,
                                 kernel_data, kernel_size,
                                 constant_data, constant_data_size,
                                 prog_data, prog_data_size, prog_data_param,
-                                xfb_info, &bind_map);
+                                stats, num_stats, xfb_info, &bind_map);
 }
 
 /* Remaining work:
@@ -359,6 +372,8 @@ anv_pipeline_cache_add_shader_locked(struct anv_pipeline_cache *cache,
                                      const struct brw_stage_prog_data *prog_data,
                                      uint32_t prog_data_size,
                                      const void *prog_data_param,
+                                     const struct brw_compile_stats *stats,
+                                     uint32_t num_stats,
                                      const nir_xfb_info *xfb_info,
                                      const struct anv_pipeline_bind_map *bind_map)
 {
@@ -372,7 +387,7 @@ anv_pipeline_cache_add_shader_locked(struct anv_pipeline_cache *cache,
                             kernel_data, kernel_size,
                             constant_data, constant_data_size,
                             prog_data, prog_data_size, prog_data_param,
-                            xfb_info, bind_map);
+                            stats, num_stats, xfb_info, bind_map);
    if (!bin)
       return NULL;
 
@@ -389,6 +404,8 @@ anv_pipeline_cache_upload_kernel(struct anv_pipeline_cache *cache,
                                  uint32_t constant_data_size,
                                  const struct brw_stage_prog_data *prog_data,
                                  uint32_t prog_data_size,
+                                 const struct brw_compile_stats *stats,
+                                 uint32_t num_stats,
                                  const nir_xfb_info *xfb_info,
                                  const struct anv_pipeline_bind_map *bind_map)
 {
@@ -401,6 +418,7 @@ anv_pipeline_cache_upload_kernel(struct anv_pipeline_cache *cache,
                                               constant_data, constant_data_size,
                                               prog_data, prog_data_size,
                                               prog_data->param,
+                                              stats, num_stats,
                                               xfb_info, bind_map);
 
       pthread_mutex_unlock(&cache->mutex);
@@ -417,6 +435,7 @@ anv_pipeline_cache_upload_kernel(struct anv_pipeline_cache *cache,
                                    constant_data, constant_data_size,
                                    prog_data, prog_data_size,
                                    prog_data->param,
+                                   stats, num_stats,
                                    xfb_info, bind_map);
    }
 }
@@ -659,6 +678,8 @@ anv_device_upload_kernel(struct anv_device *device,
                          uint32_t constant_data_size,
                          const struct brw_stage_prog_data *prog_data,
                          uint32_t prog_data_size,
+                         const struct brw_compile_stats *stats,
+                         uint32_t num_stats,
                          const nir_xfb_info *xfb_info,
                          const struct anv_pipeline_bind_map *bind_map)
 {
@@ -668,6 +689,7 @@ anv_device_upload_kernel(struct anv_device *device,
                                              kernel_data, kernel_size,
                                              constant_data, constant_data_size,
                                              prog_data, prog_data_size,
+                                             stats, num_stats,
                                              xfb_info, bind_map);
    } else {
       bin = anv_shader_bin_create(device, key_data, key_size,
@@ -675,6 +697,7 @@ anv_device_upload_kernel(struct anv_device *device,
                                   constant_data, constant_data_size,
                                   prog_data, prog_data_size,
                                   prog_data->param,
+                                  stats, num_stats,
                                   xfb_info, bind_map);
    }
 

diff --git a/src/intel/vulkan/anv_private.h b/src/intel/vulkan/anv_private.h
index b381386a71639903942a5886c753b017fa14c353..6f839db94d7d1610d70cb58fd689362a314c3700 100644 (file)
--- a/src/intel/vulkan/anv_private.h
+++ b/src/intel/vulkan/anv_private.h
@@ -1054,6 +1054,8 @@ anv_pipeline_cache_upload_kernel(struct anv_pipeline_cache *cache,
                                  uint32_t constant_data_size,
                                  const struct brw_stage_prog_data *prog_data,
                                  uint32_t prog_data_size,
+                                 const struct brw_compile_stats *stats,
+                                 uint32_t num_stats,
                                  const struct nir_xfb_info *xfb_info,
                                  const struct anv_pipeline_bind_map *bind_map);
 
@@ -1072,6 +1074,8 @@ anv_device_upload_kernel(struct anv_device *device,
                          uint32_t constant_data_size,
                          const struct brw_stage_prog_data *prog_data,
                          uint32_t prog_data_size,
+                         const struct brw_compile_stats *stats,
+                         uint32_t num_stats,
                          const struct nir_xfb_info *xfb_info,
                          const struct anv_pipeline_bind_map *bind_map);
 
@@ -2753,6 +2757,9 @@ struct anv_shader_bin {
    const struct brw_stage_prog_data *prog_data;
    uint32_t prog_data_size;
 
+   struct brw_compile_stats stats[3];
+   uint32_t num_stats;
+
    struct nir_xfb_info *xfb_info;
 
    struct anv_pipeline_bind_map bind_map;
@@ -2765,6 +2772,7 @@ anv_shader_bin_create(struct anv_device *device,
                       const void *constant_data, uint32_t constant_data_size,
                       const struct brw_stage_prog_data *prog_data,
                       uint32_t prog_data_size, const void *prog_data_param,
+                      const struct brw_compile_stats *stats, uint32_t num_stats,
                       const struct nir_xfb_info *xfb_info,
                       const struct anv_pipeline_bind_map *bind_map);