AlexIndustrial/mesa
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

aco: rename max_wave64_per_simd -> max_waves_per_simd

Browse Source 
and update usage. Changes are because the scheduler targets
a different number of waves.

Totals from 195 (0.25% of 79330) affected shaders: (GFX11)

MaxWaves: 3120 -> 3108 (-0.38%)
Instrs: 71202 -> 71070 (-0.19%); split: -0.27%, +0.09%
CodeSize: 383272 -> 382828 (-0.12%); split: -0.21%, +0.10%
VGPRs: 7392 -> 7752 (+4.87%)
Latency: 2280141 -> 2262487 (-0.77%); split: -0.79%, +0.02%
InvThroughput: 4759022 -> 5725442 (+20.31%); split: -0.01%, +20.32%
VClause: 1737 -> 1741 (+0.23%); split: -3.11%, +3.34%
SClause: 2385 -> 2376 (-0.38%); split: -0.80%, +0.42%
Copies: 5257 -> 5274 (+0.32%); split: -0.25%, +0.57%
Branches: 1213 -> 1212 (-0.08%)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26521>
This commit is contained in:
Daniel Schürmann
2023-12-05 16:58:13 +01:00
committed by Marge Bot
parent f5bdc46a57
commit 5ebba87772
3 changed files with 7 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/amd/compiler/aco_ir.cpp
+5 -6
View File
@@ -143,13 +143,13 @@ init_program(Program* program, Stage stage, const struct aco_shader_info* info,
program->dev.scratch_alloc_granule = gfx_level >= GFX11 ? 256 : 1024;
program->dev.max_wave64_per_simd = 10;
program->dev.max_waves_per_simd = 10;
if (program->gfx_level >= GFX10_3)
program->dev.max_wave64_per_simd = 16;
program->dev.max_waves_per_simd = 16;
else if (program->gfx_level == GFX10)
program->dev.max_wave64_per_simd = 20;
program->dev.max_waves_per_simd = 20;
else if (program->family >= CHIP_POLARIS10 && program->family <= CHIP_VEGAM)
program->dev.max_wave64_per_simd = 8;
program->dev.max_waves_per_simd = 8;
program->dev.simd_per_cu = program->gfx_level >= GFX10 ? 2 : 4;
@@ -1353,8 +1353,7 @@ dealloc_vgprs(Program* program)
return false;
/* skip if deallocating VGPRs won't increase occupancy */
uint16_t max_waves = program->dev.max_wave64_per_simd * (64 / program->wave_size);
max_waves = max_suitable_waves(program, max_waves);
uint16_t max_waves = max_suitable_waves(program, program->dev.max_waves_per_simd);
if (program->max_reg_demand.vgpr <= get_addr_vgpr_from_waves(program, max_waves))
return false;
src/amd/compiler/aco_ir.h
+1 -1
View File
@@ -2020,7 +2020,7 @@ struct DeviceInfo {
uint16_t sgpr_alloc_granule;
uint16_t vgpr_alloc_granule;
unsigned scratch_alloc_granule;
unsigned max_wave64_per_simd;
uint16_t max_waves_per_simd;
unsigned simd_per_cu;
bool has_fast_fma32 = false;
bool has_mac_legacy32 = false;
src/amd/compiler/aco_live_var_analysis.cpp
+1 -2
View File
@@ -464,8 +464,7 @@ update_vgpr_sgpr_demand(Program* program, const RegisterDemand new_demand)
get_vgpr_alloc(program, new_demand.vgpr) + program->config->num_shared_vgprs / 2;
program->num_waves =
std::min<uint16_t>(program->num_waves, program->dev.physical_vgprs / vgpr_demand);
uint16_t max_waves = program->dev.max_wave64_per_simd * (64 / program->wave_size);
program->num_waves = std::min(program->num_waves, max_waves);
program->num_waves = std::min(program->num_waves, program->dev.max_waves_per_simd);
/* Adjust for LDS and workgroup multiples and calculate max_reg_demand */
program->num_waves = max_suitable_waves(program, program->num_waves);