diff options
| author | Nicolai Hähnle <nicolai.haehnle@amd.com> | 2016-09-09 10:08:11 +0200 |
|---|---|---|
| committer | Nicolai Hähnle <nicolai.haehnle@amd.com> | 2016-10-10 10:36:42 +0200 |
| commit | 77c81164bc1cd9ec98b32c40753f590791450434 (patch) | |
| tree | 10f8468ceb6de6cbf56cba29a457a7098085a1e9 /src/gallium/drivers/radeonsi/si_shader.c | |
| parent | 014bd4acb8b130fb31ec00f6125b8a91881bebdf (diff) | |
| download | external_mesa3d-77c81164bc1cd9ec98b32c40753f590791450434.zip external_mesa3d-77c81164bc1cd9ec98b32c40753f590791450434.tar.gz external_mesa3d-77c81164bc1cd9ec98b32c40753f590791450434.tar.bz2 | |
radeonsi: support ARB_compute_variable_group_size
Not sure if it's possible to avoid programming the block size twice (once for
the userdata and once for the dispatch).
Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Diffstat (limited to 'src/gallium/drivers/radeonsi/si_shader.c')
| -rw-r--r-- | src/gallium/drivers/radeonsi/si_shader.c | 44 |
1 files changed, 30 insertions, 14 deletions
diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c index ff51c8b..49d4121 100644 --- a/src/gallium/drivers/radeonsi/si_shader.c +++ b/src/gallium/drivers/radeonsi/si_shader.c @@ -1770,16 +1770,21 @@ static void declare_system_value( LLVMValueRef values[3]; unsigned i; unsigned *properties = ctx->shader->selector->info.properties; - unsigned sizes[3] = { - properties[TGSI_PROPERTY_CS_FIXED_BLOCK_WIDTH], - properties[TGSI_PROPERTY_CS_FIXED_BLOCK_HEIGHT], - properties[TGSI_PROPERTY_CS_FIXED_BLOCK_DEPTH] - }; - for (i = 0; i < 3; ++i) - values[i] = lp_build_const_int32(gallivm, sizes[i]); + if (properties[TGSI_PROPERTY_CS_FIXED_BLOCK_WIDTH] != 0) { + unsigned sizes[3] = { + properties[TGSI_PROPERTY_CS_FIXED_BLOCK_WIDTH], + properties[TGSI_PROPERTY_CS_FIXED_BLOCK_HEIGHT], + properties[TGSI_PROPERTY_CS_FIXED_BLOCK_DEPTH] + }; + + for (i = 0; i < 3; ++i) + values[i] = lp_build_const_int32(gallivm, sizes[i]); - value = lp_build_gather_values(gallivm, values, 3); + value = lp_build_gather_values(gallivm, values, 3); + } else { + value = LLVMGetParam(radeon_bld->main_fn, SI_PARAM_BLOCK_SIZE); + } break; } @@ -5680,6 +5685,7 @@ static void create_function(struct si_shader_context *ctx) case PIPE_SHADER_COMPUTE: params[SI_PARAM_GRID_SIZE] = v3i32; + params[SI_PARAM_BLOCK_SIZE] = v3i32; params[SI_PARAM_BLOCK_ID] = v3i32; last_sgpr = SI_PARAM_BLOCK_ID; @@ -5716,7 +5722,12 @@ static void create_function(struct si_shader_context *ctx) properties[TGSI_PROPERTY_CS_FIXED_BLOCK_HEIGHT] * properties[TGSI_PROPERTY_CS_FIXED_BLOCK_DEPTH]; - assert(max_work_group_size); + if (!max_work_group_size) { + /* This is a variable group size compute shader, + * compile it for the maximum possible group size. + */ + max_work_group_size = SI_MAX_VARIABLE_THREADS_PER_BLOCK; + } radeon_llvm_add_attribute(ctx->radeon_bld.main_fn, "amdgpu-max-work-group-size", @@ -6653,11 +6664,16 @@ int si_compile_tgsi_shader(struct si_screen *sscreen, unsigned max_vgprs = 256; unsigned max_sgprs = sscreen->b.chip_class >= VI ? 800 : 512; unsigned max_sgprs_per_wave = 128; - unsigned min_waves_per_cu = - DIV_ROUND_UP(props[TGSI_PROPERTY_CS_FIXED_BLOCK_WIDTH] * - props[TGSI_PROPERTY_CS_FIXED_BLOCK_HEIGHT] * - props[TGSI_PROPERTY_CS_FIXED_BLOCK_DEPTH], - wave_size); + unsigned max_block_threads; + + if (props[TGSI_PROPERTY_CS_FIXED_BLOCK_WIDTH]) + max_block_threads = props[TGSI_PROPERTY_CS_FIXED_BLOCK_WIDTH] * + props[TGSI_PROPERTY_CS_FIXED_BLOCK_HEIGHT] * + props[TGSI_PROPERTY_CS_FIXED_BLOCK_DEPTH]; + else + max_block_threads = SI_MAX_VARIABLE_THREADS_PER_BLOCK; + + unsigned min_waves_per_cu = DIV_ROUND_UP(max_block_threads, wave_size); unsigned min_waves_per_simd = DIV_ROUND_UP(min_waves_per_cu, 4); max_vgprs = max_vgprs / min_waves_per_simd; |