diff options
author | Ameer J <52414509+ameerj@users.noreply.github.com> | 2023-10-22 06:38:30 +0200 |
---|---|---|
committer | Ameer J <52414509+ameerj@users.noreply.github.com> | 2023-10-22 06:45:23 +0200 |
commit | cfe73af6f220734faf32d1871f0da8a6a077dea6 (patch) | |
tree | ce98bce7ccfdc4696d6280c92f65607fe71626df | |
parent | Merge pull request #11786 from v1993/cuda-on-linux (diff) | |
download | yuzu-cfe73af6f220734faf32d1871f0da8a6a077dea6.tar yuzu-cfe73af6f220734faf32d1871f0da8a6a077dea6.tar.gz yuzu-cfe73af6f220734faf32d1871f0da8a6a077dea6.tar.bz2 yuzu-cfe73af6f220734faf32d1871f0da8a6a077dea6.tar.lz yuzu-cfe73af6f220734faf32d1871f0da8a6a077dea6.tar.xz yuzu-cfe73af6f220734faf32d1871f0da8a6a077dea6.tar.zst yuzu-cfe73af6f220734faf32d1871f0da8a6a077dea6.zip |
-rw-r--r-- | src/shader_recompiler/backend/glsl/emit_glsl.cpp | 1 | ||||
-rw-r--r-- | src/shader_recompiler/backend/glsl/emit_glsl_warp.cpp | 12 |
2 files changed, 9 insertions, 4 deletions
diff --git a/src/shader_recompiler/backend/glsl/emit_glsl.cpp b/src/shader_recompiler/backend/glsl/emit_glsl.cpp index d91e04446..66ecfc9f7 100644 --- a/src/shader_recompiler/backend/glsl/emit_glsl.cpp +++ b/src/shader_recompiler/backend/glsl/emit_glsl.cpp @@ -242,6 +242,7 @@ std::string EmitGLSL(const Profile& profile, const RuntimeInfo& runtime_info, IR } if (program.info.uses_subgroup_shuffles) { ctx.header += "bool shfl_in_bounds;"; + ctx.header += "uint shfl_result;"; } ctx.code.insert(0, ctx.header); ctx.code += '}'; diff --git a/src/shader_recompiler/backend/glsl/emit_glsl_warp.cpp b/src/shader_recompiler/backend/glsl/emit_glsl_warp.cpp index 1245c9429..f9be5de1c 100644 --- a/src/shader_recompiler/backend/glsl/emit_glsl_warp.cpp +++ b/src/shader_recompiler/backend/glsl/emit_glsl_warp.cpp @@ -141,7 +141,8 @@ void EmitShuffleIndex(EmitContext& ctx, IR::Inst& inst, std::string_view value, const auto src_thread_id{fmt::format("({})|({})", lhs, min_thread_id)}; ctx.Add("shfl_in_bounds=int({})<=int({});", src_thread_id, max_thread_id); SetInBoundsFlag(ctx, inst); - ctx.AddU32("{}=shfl_in_bounds?readInvocationARB({},{}):{};", inst, value, src_thread_id, value); + ctx.Add("shfl_result=readInvocationARB({},{});", value, src_thread_id); + ctx.AddU32("{}=shfl_in_bounds?shfl_result:{};", inst, value); } void EmitShuffleUp(EmitContext& ctx, IR::Inst& inst, std::string_view value, std::string_view index, @@ -158,7 +159,8 @@ void EmitShuffleUp(EmitContext& ctx, IR::Inst& inst, std::string_view value, std const auto src_thread_id{fmt::format("({}-{})", THREAD_ID, index)}; ctx.Add("shfl_in_bounds=int({})>=int({});", src_thread_id, max_thread_id); SetInBoundsFlag(ctx, inst); - ctx.AddU32("{}=shfl_in_bounds?readInvocationARB({},{}):{};", inst, value, src_thread_id, value); + ctx.Add("shfl_result=readInvocationARB({},{});", value, src_thread_id); + ctx.AddU32("{}=shfl_in_bounds?shfl_result:{};", inst, value); } void EmitShuffleDown(EmitContext& ctx, IR::Inst& inst, std::string_view value, @@ -175,7 +177,8 @@ void EmitShuffleDown(EmitContext& ctx, IR::Inst& inst, std::string_view value, const auto src_thread_id{fmt::format("({}+{})", THREAD_ID, index)}; ctx.Add("shfl_in_bounds=int({})<=int({});", src_thread_id, max_thread_id); SetInBoundsFlag(ctx, inst); - ctx.AddU32("{}=shfl_in_bounds?readInvocationARB({},{}):{};", inst, value, src_thread_id, value); + ctx.Add("shfl_result=readInvocationARB({},{});", value, src_thread_id); + ctx.AddU32("{}=shfl_in_bounds?shfl_result:{};", inst, value); } void EmitShuffleButterfly(EmitContext& ctx, IR::Inst& inst, std::string_view value, @@ -193,7 +196,8 @@ void EmitShuffleButterfly(EmitContext& ctx, IR::Inst& inst, std::string_view val const auto src_thread_id{fmt::format("({}^{})", THREAD_ID, index)}; ctx.Add("shfl_in_bounds=int({})<=int({});", src_thread_id, max_thread_id); SetInBoundsFlag(ctx, inst); - ctx.AddU32("{}=shfl_in_bounds?readInvocationARB({},{}):{};", inst, value, src_thread_id, value); + ctx.Add("shfl_result=readInvocationARB({},{});", value, src_thread_id); + ctx.AddU32("{}=shfl_in_bounds?shfl_result:{};", inst, value); } void EmitFSwizzleAdd(EmitContext& ctx, IR::Inst& inst, std::string_view op_a, std::string_view op_b, |