diff options
author | Morph <39850852+Morph1984@users.noreply.github.com> | 2021-09-12 19:53:29 +0200 |
---|---|---|
committer | GitHub <noreply@github.com> | 2021-09-12 19:53:29 +0200 |
commit | 9248442bb2759c071b565a10e959883980ff09d6 (patch) | |
tree | 1058129dc40331f7f688d32467da17214a2dc013 /src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp | |
parent | Merge pull request #6975 from ogniK5377/acc-async-ctx (diff) | |
parent | emit_glsl_warp: Fix shuffle ops for 64-thread warp sizes (diff) | |
download | yuzu-9248442bb2759c071b565a10e959883980ff09d6.tar yuzu-9248442bb2759c071b565a10e959883980ff09d6.tar.gz yuzu-9248442bb2759c071b565a10e959883980ff09d6.tar.bz2 yuzu-9248442bb2759c071b565a10e959883980ff09d6.tar.lz yuzu-9248442bb2759c071b565a10e959883980ff09d6.tar.xz yuzu-9248442bb2759c071b565a10e959883980ff09d6.tar.zst yuzu-9248442bb2759c071b565a10e959883980ff09d6.zip |
Diffstat (limited to 'src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp')
-rw-r--r-- | src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp | 41 |
1 files changed, 35 insertions, 6 deletions
diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp index 78b1e1ba7..cef52c56e 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp +++ b/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp @@ -7,8 +7,13 @@ namespace Shader::Backend::SPIRV { namespace { +Id GetThreadId(EmitContext& ctx) { + return ctx.OpLoad(ctx.U32[1], ctx.subgroup_local_invocation_id); +} + Id WarpExtract(EmitContext& ctx, Id value) { - const Id local_index{ctx.OpLoad(ctx.U32[1], ctx.subgroup_local_invocation_id)}; + const Id thread_id{GetThreadId(ctx)}; + const Id local_index{ctx.OpShiftRightArithmetic(ctx.U32[1], thread_id, ctx.Const(5U))}; return ctx.OpVectorExtractDynamic(ctx.U32[1], value, local_index); } @@ -48,10 +53,17 @@ Id SelectValue(EmitContext& ctx, Id in_range, Id value, Id src_thread_id) { return ctx.OpSelect(ctx.U32[1], in_range, ctx.OpSubgroupReadInvocationKHR(ctx.U32[1], value, src_thread_id), value); } + +Id GetUpperClamp(EmitContext& ctx, Id invocation_id, Id clamp) { + const Id thirty_two{ctx.Const(32u)}; + const Id is_upper_partition{ctx.OpSGreaterThanEqual(ctx.U1, invocation_id, thirty_two)}; + const Id upper_clamp{ctx.OpIAdd(ctx.U32[1], thirty_two, clamp)}; + return ctx.OpSelect(ctx.U32[1], is_upper_partition, upper_clamp, clamp); +} } // Anonymous namespace Id EmitLaneId(EmitContext& ctx) { - const Id id{ctx.OpLoad(ctx.U32[1], ctx.subgroup_local_invocation_id)}; + const Id id{GetThreadId(ctx)}; if (!ctx.profile.warp_size_potentially_larger_than_guest) { return id; } @@ -123,7 +135,15 @@ Id EmitSubgroupGeMask(EmitContext& ctx) { Id EmitShuffleIndex(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp, Id segmentation_mask) { const Id not_seg_mask{ctx.OpNot(ctx.U32[1], segmentation_mask)}; - const Id thread_id{ctx.OpLoad(ctx.U32[1], ctx.subgroup_local_invocation_id)}; + const Id thread_id{GetThreadId(ctx)}; + if (ctx.profile.warp_size_potentially_larger_than_guest) { + const Id thirty_two{ctx.Const(32u)}; + const Id is_upper_partition{ctx.OpSGreaterThanEqual(ctx.U1, thread_id, thirty_two)}; + const Id upper_index{ctx.OpIAdd(ctx.U32[1], thirty_two, index)}; + const Id upper_clamp{ctx.OpIAdd(ctx.U32[1], thirty_two, clamp)}; + index = ctx.OpSelect(ctx.U32[1], is_upper_partition, upper_index, index); + clamp = ctx.OpSelect(ctx.U32[1], is_upper_partition, upper_clamp, clamp); + } const Id min_thread_id{ComputeMinThreadId(ctx, thread_id, segmentation_mask)}; const Id max_thread_id{ComputeMaxThreadId(ctx, min_thread_id, clamp, not_seg_mask)}; @@ -137,7 +157,10 @@ Id EmitShuffleIndex(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id cla Id EmitShuffleUp(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp, Id segmentation_mask) { - const Id thread_id{ctx.OpLoad(ctx.U32[1], ctx.subgroup_local_invocation_id)}; + const Id thread_id{GetThreadId(ctx)}; + if (ctx.profile.warp_size_potentially_larger_than_guest) { + clamp = GetUpperClamp(ctx, thread_id, clamp); + } const Id max_thread_id{GetMaxThreadId(ctx, thread_id, clamp, segmentation_mask)}; const Id src_thread_id{ctx.OpISub(ctx.U32[1], thread_id, index)}; const Id in_range{ctx.OpSGreaterThanEqual(ctx.U1, src_thread_id, max_thread_id)}; @@ -148,7 +171,10 @@ Id EmitShuffleUp(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp, Id EmitShuffleDown(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp, Id segmentation_mask) { - const Id thread_id{ctx.OpLoad(ctx.U32[1], ctx.subgroup_local_invocation_id)}; + const Id thread_id{GetThreadId(ctx)}; + if (ctx.profile.warp_size_potentially_larger_than_guest) { + clamp = GetUpperClamp(ctx, thread_id, clamp); + } const Id max_thread_id{GetMaxThreadId(ctx, thread_id, clamp, segmentation_mask)}; const Id src_thread_id{ctx.OpIAdd(ctx.U32[1], thread_id, index)}; const Id in_range{ctx.OpSLessThanEqual(ctx.U1, src_thread_id, max_thread_id)}; @@ -159,7 +185,10 @@ Id EmitShuffleDown(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clam Id EmitShuffleButterfly(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp, Id segmentation_mask) { - const Id thread_id{ctx.OpLoad(ctx.U32[1], ctx.subgroup_local_invocation_id)}; + const Id thread_id{GetThreadId(ctx)}; + if (ctx.profile.warp_size_potentially_larger_than_guest) { + clamp = GetUpperClamp(ctx, thread_id, clamp); + } const Id max_thread_id{GetMaxThreadId(ctx, thread_id, clamp, segmentation_mask)}; const Id src_thread_id{ctx.OpBitwiseXor(ctx.U32[1], thread_id, index)}; const Id in_range{ctx.OpSLessThanEqual(ctx.U1, src_thread_id, max_thread_id)}; |