OpenHarmony/third_party_mesa3d
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

ac: unify denorm setting enforcement

Browse Source 
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4196>
This commit is contained in:
Marek Olšák
2020-01-23 15:52:01 -05:00
committed by Marge Bot
parent e4959add2f
commit 56cc10bd27
3 changed files with 14 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
src/amd/common/ac_binary.c
View File

@@ -58,11 +58,13 @@ void ac_parse_shader_binary_config(const char *data, size_t nbytes,
conf->num_vgprs = MAX2(conf->num_vgprs, (G_00B028_VGPRS(value) + 1) * 4); conf->num_vgprs = MAX2(conf->num_vgprs, (G_00B028_VGPRS(value) + 1) * 4);
conf->num_sgprs = MAX2(conf->num_sgprs, (G_00B028_SGPRS(value) + 1) * 8); conf->num_sgprs = MAX2(conf->num_sgprs, (G_00B028_SGPRS(value) + 1) * 8);
/* TODO: LLVM doesn't set FLOAT_MODE for non-compute shaders */
conf->float_mode = G_00B028_FLOAT_MODE(value); conf->float_mode = G_00B028_FLOAT_MODE(value);
conf->rsrc1 = value; conf->rsrc1 = value;
break; break;
case R_00B02C_SPI_SHADER_PGM_RSRC2_PS: case R_00B02C_SPI_SHADER_PGM_RSRC2_PS:
conf->lds_size = MAX2(conf->lds_size, G_00B02C_EXTRA_LDS_SIZE(value)); conf->lds_size = MAX2(conf->lds_size, G_00B02C_EXTRA_LDS_SIZE(value));
/* TODO: LLVM doesn't set SHARED_VGPR_CNT for all shader types */
conf->num_shared_vgprs = G_00B02C_SHARED_VGPR_CNT(value); conf->num_shared_vgprs = G_00B02C_SHARED_VGPR_CNT(value);
conf->rsrc2 = value; conf->rsrc2 = value;
break; break;
@@ -124,4 +126,15 @@ void ac_parse_shader_binary_config(const char *data, size_t nbytes,
/* sgprs spills aren't spilling */ /* sgprs spills aren't spilling */
conf->scratch_bytes_per_wave = G_00B860_WAVESIZE(scratch_size) * 256 * 4; conf->scratch_bytes_per_wave = G_00B860_WAVESIZE(scratch_size) * 256 * 4;
} }
/* Enable 64-bit and 16-bit denormals, because there is no performance
* cost.
*
* Don't enable denormals for 32-bit floats, because:
* - denormals disable output modifiers
* - denormals break v_mad_f32
* - GFX6 & GFX7 would be very slow
*/
conf->float_mode &= ~V_00B028_FP_ALL_DENORMS;
conf->float_mode |= V_00B028_FP_64_DENORMS;
} }

14
src/amd/vulkan/radv_shader.c
View File

@@ -966,20 +966,6 @@ radv_shader_variant_create(struct radv_device *device,
return NULL; return NULL;
} }
/* Enable 64-bit and 16-bit denormals, because there is no performance
* cost.
*
* If denormals are enabled, all floating-point output modifiers are
* ignored.
*
* Don't enable denormals for 32-bit floats, because:
* - Floating-point output modifiers would be ignored by the hw.
* - Some opcodes don't support denormals, such as v_mad_f32. We would
* have to stop using those.
* - GFX6 & GFX7 would be very slow.
*/
config.float_mode |= V_00B028_FP_64_DENORMS;
if (rtld_binary.lds_size > 0) { if (rtld_binary.lds_size > 0) {
unsigned alloc_granularity = device->physical_device->rad_info.chip_class >= GFX7 ? 512 : 256; unsigned alloc_granularity = device->physical_device->rad_info.chip_class >= GFX7 ? 512 : 256;
config.lds_size = align(rtld_binary.lds_size, alloc_granularity) / alloc_granularity; config.lds_size = align(rtld_binary.lds_size, alloc_granularity) / alloc_granularity;

19
src/gallium/drivers/radeonsi/si_shader_llvm.c
View File

@@ -130,24 +130,7 @@ bool si_compile_llvm(struct si_screen *sscreen,
bool ok = ac_rtld_read_config(&rtld, conf); bool ok = ac_rtld_read_config(&rtld, conf);
ac_rtld_close(&rtld); ac_rtld_close(&rtld);
if (!ok) return ok;
return false;
/* Enable 64-bit and 16-bit denormals, because there is no performance
* cost.
*
* If denormals are enabled, all floating-point output modifiers are
* ignored.
*
* Don't enable denormals for 32-bit floats, because:
* - Floating-point output modifiers would be ignored by the hw.
* - Some opcodes don't support denormals, such as v_mad_f32. We would
* have to stop using those.
* - GFX6 & GFX7 would be very slow.
*/
conf->float_mode |= V_00B028_FP_64_DENORMS;
return true;
} }
void si_llvm_context_init(struct si_shader_context *ctx, void si_llvm_context_init(struct si_shader_context *ctx,