fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-05 13:58:04 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

radv/gfx11: add a RADV_PERFTEST flag to expose bfloat16 cmat

Browse source 
This doesn't pass CTS because of precision issues, but might still be useful.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/14699
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39456>
This commit is contained in:
Georg Lehmann 2026-01-22 13:06:32 +01:00 • committed by Marge Bot
parent 455b692e4f
commit 763b4f1f0a
4 changed files with 20 additions and 4 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
docs/envvars.rst
View file

@ -1559,6 +1559,8 @@ RADV driver environment variables
a comma-separated list of named flags, which do various things:
``bfloat16``
enable bfloat16 cooperative matrix support on GFX11-11.5
``cswave32``
enable wave32 for compute shaders (GFX10+)
``dccmsaa``

1
src/amd/vulkan/radv_debug.h
View file

@ -100,6 +100,7 @@ enum {
RADV_PERFTEST_HIC = 1u << 16,
RADV_PERFTEST_SPARSE = 1u << 17,
RADV_PERFTEST_RT_CPS = 1u << 18,
RADV_PERFTEST_BFLOAT16 = 1u << 19,
};
enum {

1
src/amd/vulkan/radv_instance.c
View file

@ -125,6 +125,7 @@ static const struct debug_control radv_perftest_options[] = {
{"hic", RADV_PERFTEST_HIC},
{"sparse", RADV_PERFTEST_SPARSE},
{"rtcps", RADV_PERFTEST_RT_CPS},
{"bfloat16", RADV_PERFTEST_BFLOAT16},
{NULL, 0},
};

20
src/amd/vulkan/radv_physical_device.c
View file

@ -178,6 +178,18 @@ radv_cooperative_matrix2_nv_enabled(const struct radv_physical_device *pdev)
return instance->drirc.features.cooperative_matrix2_nv;
}
static bool
radv_bfloat16_enabled(const struct radv_physical_device *pdev)
{
const struct radv_instance *instance = radv_physical_device_instance(pdev);
if (pdev->info.gfx_level < GFX11)
return false;
/* GFX11-11.5 has precision issues. */
return (instance->perftest_flags & RADV_PERFTEST_BFLOAT16) || pdev->info.gfx_level >= GFX12;
}
static bool
radv_shader_fp16_enabled(const struct radv_physical_device *pdev)
{
@ -680,7 +692,7 @@ radv_physical_device_get_supported_extensions(const struct radv_physical_device
.KHR_sampler_ycbcr_conversion = true,
.KHR_separate_depth_stencil_layouts = true,
.KHR_shader_atomic_int64 = true,
.KHR_shader_bfloat16 = pdev->info.gfx_level >= GFX12, /* GFX11 has precision issues. */
.KHR_shader_bfloat16 = radv_bfloat16_enabled(pdev),
.KHR_shader_clock = true,
.KHR_shader_draw_parameters = true,
.KHR_shader_expect_assume = true,
@ -1430,7 +1442,7 @@ radv_physical_device_get_features(const struct radv_physical_device *pdev, struc
/* VK_KHR_shader_bfloat16 */
.shaderBFloat16Type = true,
.shaderBFloat16DotProduct = true,
.shaderBFloat16DotProduct = pdev->info.gfx_level >= GFX12,
.shaderBFloat16CooperativeMatrix = radv_cooperative_matrix_enabled(pdev),
/* VK_EXT_zero_initialize_device_memory */
@ -3288,8 +3300,8 @@ fill_array_sizes_structs(const struct radv_physical_device *pdev, struct __vk_ou
prop.a_type = prop.b_type = bfloat ? VK_COMPONENT_TYPE_BFLOAT16_KHR : VK_COMPONENT_TYPE_FLOAT16_KHR;
prop.c_type = prop.r_type = fp32 ? VK_COMPONENT_TYPE_FLOAT32_KHR : prop.a_type;
if (pdev->info.gfx_level < GFX12 && bfloat)
continue; /* BF16 isn't working precisely on GFX11. */
if (!radv_bfloat16_enabled(pdev) && bfloat)
continue;
(*array_size_cb)(base, &prop);
}