radeonsi: print shader cache stats with AMD_DEBUG=cache_stats

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2929>

src/gallium/auxiliary/util/u_live_shader_cache.c

@@ -118,8 +118,10 @@ util_live_shader_cache_get(struct pipe_context *ctx,
    struct util_live_shader *shader = entry ? entry->data : NULL;
 
    /* Increase the refcount. */
-   if (shader)
+   if (shader) {
       pipe_reference(NULL, &shader->reference);
+      cache->hits++;
+   }
    simple_mtx_unlock(&cache->lock);
 
    /* Return if the shader already exists. */
@@ -148,6 +150,7 @@ util_live_shader_cache_get(struct pipe_context *ctx,
    } else {
       _mesa_hash_table_insert(cache->hashtable, shader->sha1, shader);
    }
+   cache->misses++;
    simple_mtx_unlock(&cache->lock);
 
    return shader;

src/gallium/auxiliary/util/u_live_shader_cache.h

@@ -59,6 +59,8 @@ struct util_live_shader_cache {
    void *(*create_shader)(struct pipe_context *,
                           const struct pipe_shader_state *state);
    void (*destroy_shader)(struct pipe_context *, void *);
+
+   unsigned hits, misses;
 };
 
 struct util_live_shader {

src/gallium/drivers/radeonsi/si_pipe.c

@@ -80,6 +80,7 @@ static const struct debug_named_value debug_options[] = {
 	{ "tex", DBG(TEX), "Print texture info" },
 	{ "compute", DBG(COMPUTE), "Print compute info" },
 	{ "vm", DBG(VM), "Print virtual addresses when creating resources" },
+	{ "cache_stats", DBG(CACHE_STATS), "Print shader cache statistics." },
 
 	/* Driver options: */
 	{ "forcedma", DBG(FORCE_SDMA), "Use SDMA for all operations when possible." },
@@ -771,6 +772,18 @@ static void si_destroy_screen(struct pipe_screen* pscreen)
 	if (!sscreen->ws->unref(sscreen->ws))
 		return;
 
+	if (sscreen->debug_flags & DBG(CACHE_STATS)) {
+		printf("live shader cache:   hits = %u, misses = %u\n",
+		       sscreen->live_shader_cache.hits,
+		       sscreen->live_shader_cache.misses);
+		printf("memory shader cache: hits = %u, misses = %u\n",
+		       sscreen->num_memory_shader_cache_hits,
+		       sscreen->num_memory_shader_cache_misses);
+		printf("disk shader cache:   hits = %u, misses = %u\n",
+		       sscreen->num_disk_shader_cache_hits,
+		       sscreen->num_disk_shader_cache_misses);
+	}
+
 	simple_mtx_destroy(&sscreen->aux_context_lock);
 
 	struct u_log_context *aux_log = ((struct si_context *)sscreen->aux_context)->log;

src/gallium/drivers/radeonsi/si_pipe.h

@@ -169,6 +169,7 @@ enum {
 	DBG_TEX,
 	DBG_COMPUTE,
 	DBG_VM,
+	DBG_CACHE_STATS,
 
 	/* Driver options: */
 	DBG_FORCE_SDMA,
@@ -540,7 +541,10 @@ struct si_screen {
 	 * are loading shaders on demand. This is a monotonic counter.
 	 */
 	unsigned			num_shaders_created;
-	unsigned			num_shader_cache_hits;
+	unsigned			num_memory_shader_cache_hits;
+	unsigned			num_memory_shader_cache_misses;
+	unsigned			num_disk_shader_cache_hits;
+	unsigned			num_disk_shader_cache_misses;
 
 	/* GPU load thread. */
 	simple_mtx_t			gpu_load_mutex;

src/gallium/drivers/radeonsi/si_query.c

@@ -251,7 +251,7 @@ static bool si_query_sw_begin(struct si_context *sctx,
 		break;
 	case SI_QUERY_NUM_SHADER_CACHE_HITS:
 		query->begin_result =
-			p_atomic_read(&sctx->screen->num_shader_cache_hits);
+			p_atomic_read(&sctx->screen->num_memory_shader_cache_hits);
 		break;
 	case SI_QUERY_PD_NUM_PRIMS_ACCEPTED:
 		query->begin_result = sctx->compute_num_verts_accepted;
@@ -425,7 +425,7 @@ static bool si_query_sw_end(struct si_context *sctx,
 		break;
 	case SI_QUERY_NUM_SHADER_CACHE_HITS:
 		query->end_result =
-			p_atomic_read(&sctx->screen->num_shader_cache_hits);
+			p_atomic_read(&sctx->screen->num_memory_shader_cache_hits);
 		break;
 	case SI_QUERY_PD_NUM_PRIMS_ACCEPTED:
 		query->end_result = sctx->compute_num_verts_accepted;

src/gallium/drivers/radeonsi/si_state_shaders.c

@@ -246,10 +246,11 @@ bool si_shader_cache_load_shader(struct si_screen *sscreen,
 
 	if (entry) {
 		if (si_load_shader_binary(shader, entry->data)) {
-			p_atomic_inc(&sscreen->num_shader_cache_hits);
+			p_atomic_inc(&sscreen->num_memory_shader_cache_hits);
 			return true;
 		}
 	}
+	p_atomic_inc(&sscreen->num_memory_shader_cache_misses);
 
 	if (!sscreen->disk_shader_cache)
 		return false;
@@ -268,7 +269,7 @@ bool si_shader_cache_load_shader(struct si_screen *sscreen,
 				free(buffer);
 				si_shader_cache_insert_shader(sscreen, ir_sha1_cache_key,
 							      shader, false);
-				p_atomic_inc(&sscreen->num_shader_cache_hits);
+				p_atomic_inc(&sscreen->num_disk_shader_cache_hits);
 				return true;
 			}
 		} else {
@@ -281,6 +282,7 @@ bool si_shader_cache_load_shader(struct si_screen *sscreen,
 	}
 
 	free(buffer);
+	p_atomic_inc(&sscreen->num_disk_shader_cache_misses);
 	return false;
 }