fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-03-14 06:30:26 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

rusticl: support more subgroup extensions

Browse source 
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37169>
This commit is contained in:
Karol Herbst 2025-09-03 21:57:31 +02:00 • committed by Marge Bot
parent e167bdf4ac
commit ca30514389
3 changed files with 81 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

12
docs/features.txt
View file

@ -889,13 +889,13 @@ Rusticl extensions:
cl_khr_spirv_no_integer_wrap_decoration DONE
cl_khr_spirv_queries DONE
cl_khr_srgb_image_writes not started
cl_khr_subgroup_ballot not started
cl_khr_subgroup_clustered_reduce not started
cl_khr_subgroup_extended_types not started
cl_khr_subgroup_ballot DONE (asahi, iris, llvmpipe, radeonsi, zink)
cl_khr_subgroup_clustered_reduce DONE (asahi, llvmpipe, radeonsi, zink)
cl_khr_subgroup_extended_types DONE (asahi, iris, llvmpipe, radeonsi, zink)
cl_khr_subgroup_named_barrier not started
cl_khr_subgroup_non_uniform_arithmetic not started
cl_khr_subgroup_non_uniform_vote not started
cl_khr_subgroup_rotate not started
cl_khr_subgroup_non_uniform_arithmetic DONE (asahi, llvmpipe, radeonsi, zink)
cl_khr_subgroup_non_uniform_vote DONE (asahi, iris, llvmpipe, radeonsi, zink)
cl_khr_subgroup_rotate DONE (llvmpipe, zink)
cl_khr_subgroup_shuffle DONE (asahi, iris, llvmpipe, radeonsi, zink)
cl_khr_subgroup_shuffle_relative DONE (asahi, iris, llvmpipe, radeonsi, zink)
cl_khr_subgroups in progress

6
docs/relnotes/new_features.txt
View file

@ -23,3 +23,9 @@ VK_EXT_legacy_dithering on panvk
GL_ARB_sample_shading on v3d
VK_KHR_maintenance4 on pvr
VK_ARM_scheduling_controls on panvk
cl_khr_subgroup_ballot on asahi, iris, llvmpipe, radeonsi and zink
cl_khr_subgroup_clustered_reduce on asahi, llvmpipe, radeonsi and zink
cl_khr_subgroup_extended_types on asahi, iris, llvmpipe, radeonsi and zink
cl_khr_subgroup_non_uniform_arithmetic on asahi, llvmpipe, radeonsi and zink
cl_khr_subgroup_non_uniform_vote on asahi, iris, llvmpipe, radeonsi and zink
cl_khr_subgroup_rotate on llvmpipe and zink

69
src/gallium/frontends/rusticl/core/device.rs
View file

@ -744,8 +744,36 @@ impl DeviceBase {
add_cap(SpvCapability::SpvCapabilitySubgroupDispatch);
// requires CL_DEVICE_SUB_GROUP_INDEPENDENT_FORWARD_PROGRESS
//add_ext(1, 0, 0, "cl_khr_subgroups");
add_ext(1, 0, 0, "cl_khr_subgroup_extended_types");
add_feat(1, 0, 0, "__opencl_c_subgroups");
if self.subgroup_ballot_supported() {
add_cap(SpvCapability::SpvCapabilityGroupNonUniformBallot);
add_ext(1, 0, 0, "cl_khr_subgroup_ballot");
}
if self.subgroup_clustered_supported() {
add_cap(SpvCapability::SpvCapabilityGroupNonUniformClustered);
add_ext(1, 0, 0, "cl_khr_subgroup_clustered_reduce");
}
if self.subgroup_non_uniform_arithmetic_supported() {
add_cap(SpvCapability::SpvCapabilityGroupNonUniformArithmetic);
add_ext(1, 0, 0, "cl_khr_subgroup_non_uniform_arithmetic");
}
if self.subgroup_non_uniform_vote_supported() {
add_cap(SpvCapability::SpvCapabilityGroupNonUniform);
add_cap(SpvCapability::SpvCapabilityGroupNonUniformVote);
add_ext(1, 0, 0, "cl_khr_subgroup_non_uniform_vote");
}
if self.subgroup_rotate_supported() {
add_cap(SpvCapability::SpvCapabilityGroupNonUniformRotateKHR);
add_ext(1, 0, 0, "cl_khr_subgroup_rotate");
add_spirv(c"SPV_KHR_subgroup_rotate");
}
if self.subgroup_shuffle_supported() {
add_cap(SpvCapability::SpvCapabilityGroupNonUniformShuffle);
add_ext(1, 0, 0, "cl_khr_subgroup_shuffle");
@ -1183,6 +1211,41 @@ impl DeviceBase {
!= 0
}
pub fn subgroup_ballot_supported(&self) -> bool {
self.subgroups_supported()
&& self.screen().caps().shader_subgroup_supported_features
& PIPE_SHADER_SUBGROUP_FEATURE_BALLOT
!= 0
}
pub fn subgroup_clustered_supported(&self) -> bool {
self.subgroups_supported()
&& self.screen().caps().shader_subgroup_supported_features
& PIPE_SHADER_SUBGROUP_FEATURE_CLUSTERED
!= 0
}
pub fn subgroup_non_uniform_arithmetic_supported(&self) -> bool {
self.subgroups_supported()
&& self.screen().caps().shader_subgroup_supported_features
& PIPE_SHADER_SUBGROUP_FEATURE_ARITHMETIC
!= 0
}
pub fn subgroup_non_uniform_vote_supported(&self) -> bool {
self.subgroups_supported()
&& self.screen().caps().shader_subgroup_supported_features
& PIPE_SHADER_SUBGROUP_FEATURE_VOTE
!= 0
}
pub fn subgroup_rotate_supported(&self) -> bool {
let mask =
PIPE_SHADER_SUBGROUP_FEATURE_ROTATE | PIPE_SHADER_SUBGROUP_FEATURE_ROTATE_CLUSTERED;
self.subgroups_supported()
&& self.screen().caps().shader_subgroup_supported_features & mask == mask
}
pub fn subgroup_shuffle_supported(&self) -> bool {
self.subgroups_supported()
&& self.screen().caps().shader_subgroup_supported_features
@ -1277,6 +1340,12 @@ impl DeviceBase {
intel_subgroups: self.intel_subgroups_supported(),
kernel_clock: self.kernel_clock_supported(),
subgroups: subgroups_supported,
subgroups_ballot: self.subgroup_ballot_supported(),
subgroups_clustered: self.subgroup_clustered_supported(),
subgroups_extended_types: subgroups_supported,
subgroups_non_uniform_arithmetic: self.subgroup_non_uniform_arithmetic_supported(),
subgroups_non_uniform_vote: self.subgroup_non_uniform_vote_supported(),
subgroups_rotate: self.subgroup_rotate_supported(),
subgroups_shuffle: self.subgroup_shuffle_supported(),
subgroups_shuffle_relative: self.subgroup_shuffle_relative_supported(),
..Default::default()