From c9ccceff1092049b081db88e53db8335a0a64951 Mon Sep 17 00:00:00 2001 From: =?utf8?q?Marek=20Ol=C5=A1=C3=A1k?= Date: Wed, 6 May 2020 19:29:51 -0400 Subject: [PATCH] radeonsi: test uncached clear/copy buffer performance with compute shaders Acked-by: Pierre-Eric Pelloux-Prayer Part-of: --- src/gallium/drivers/radeonsi/si_test_dma_perf.c | 32 +++++++++++++++---------- 1 file changed, 19 insertions(+), 13 deletions(-) diff --git a/src/gallium/drivers/radeonsi/si_test_dma_perf.c b/src/gallium/drivers/radeonsi/si_test_dma_perf.c index 1613351..1820899 100644 --- a/src/gallium/drivers/radeonsi/si_test_dma_perf.c +++ b/src/gallium/drivers/radeonsi/si_test_dma_perf.c @@ -48,7 +48,7 @@ void si_test_dma_perf(struct si_screen *sscreen) static const unsigned cs_waves_per_sh_list[] = {0, 2, 4, 8, 16}; #define NUM_SHADERS ARRAY_SIZE(cs_dwords_per_thread_list) -#define NUM_METHODS (4 + 2 * NUM_SHADERS * ARRAY_SIZE(cs_waves_per_sh_list)) +#define NUM_METHODS (4 + 3 * NUM_SHADERS * ARRAY_SIZE(cs_waves_per_sh_list)) static const char *method_str[] = { "CP MC ", @@ -103,12 +103,11 @@ void si_test_dma_perf(struct si_screen *sscreen) bool test_sdma = method == 3; bool test_cs = method >= 4; unsigned cs_method = method - 4; - STATIC_ASSERT(L2_STREAM + 1 == L2_LRU); unsigned cs_waves_per_sh = - test_cs ? cs_waves_per_sh_list[cs_method / (2 * NUM_SHADERS)] : 0; - cs_method %= 2 * NUM_SHADERS; + test_cs ? cs_waves_per_sh_list[cs_method / (3 * NUM_SHADERS)] : 0; + cs_method %= 3 * NUM_SHADERS; unsigned cache_policy = - test_cp ? method % 3 : test_cs ? L2_STREAM + (cs_method / NUM_SHADERS) : 0; + test_cp ? method % 3 : test_cs ? (cs_method / NUM_SHADERS) : 0; unsigned cs_dwords_per_thread = test_cs ? cs_dwords_per_thread_list[cs_method % NUM_SHADERS] : 0; @@ -124,6 +123,12 @@ void si_test_dma_perf(struct si_screen *sscreen) continue; } + /* SI_RESOURCE_FLAG_UNCACHED setting RADEON_FLAG_UNCACHED doesn't affect + * chips before gfx9. + */ + if (test_cs && cache_policy && sctx->chip_class < GFX9) + continue; + printf("%s ,", placement_str[placement]); if (test_cs) { printf("CS x%-4u,%3s,", cs_dwords_per_thread, @@ -149,6 +154,7 @@ void si_test_dma_perf(struct si_screen *sscreen) struct pipe_resource *dst, *src; struct pipe_query *q[NUM_RUNS]; unsigned query_type = PIPE_QUERY_TIME_ELAPSED; + unsigned flags = cache_policy == L2_BYPASS ? SI_RESOURCE_FLAG_UNCACHED : 0; if (test_sdma) { if (sctx->chip_class == GFX6) @@ -167,8 +173,8 @@ void si_test_dma_perf(struct si_screen *sscreen) else src_usage = PIPE_USAGE_STREAM; - dst = pipe_buffer_create(screen, 0, dst_usage, size); - src = is_copy ? pipe_buffer_create(screen, 0, src_usage, size) : NULL; + dst = pipe_aligned_buffer_create(screen, flags, dst_usage, size, 256); + src = is_copy ? pipe_aligned_buffer_create(screen, flags, src_usage, size, 256) : NULL; /* Run tests. */ for (unsigned iter = 0; iter < NUM_RUNS; iter++) { @@ -446,17 +452,17 @@ void si_test_dma_perf(struct si_screen *sscreen) printf("return "); assert(best); + const char *cache_policy_str = + best->cache_policy == L2_BYPASS ? "L2_BYPASS" : + best->cache_policy == L2_LRU ? "L2_LRU " : "L2_STREAM"; + if (best->is_cp) { - printf("CP_DMA(%s);\n", - best->cache_policy == L2_BYPASS - ? "L2_BYPASS" - : best->cache_policy == L2_LRU ? "L2_LRU " : "L2_STREAM"); + printf("CP_DMA(%s);\n", cache_policy_str); } if (best->is_sdma) printf("SDMA;\n"); if (best->is_cs) { - printf("COMPUTE(%s, %u, %u);\n", - best->cache_policy == L2_LRU ? "L2_LRU " : "L2_STREAM", + printf("COMPUTE(%s, %u, %u);\n", cache_policy_str, best->dwords_per_thread, best->waves_per_sh); } } -- 2.7.4