Merge pull request #11847 from ameerj/glsl-shfl-fix

emit_glsl_warp: Fix shfl_in_bounds conditional
This commit is contained in:
liamwhite 2023-10-23 10:33:24 -04:00 committed by GitHub
commit da5c49f22d
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 9 additions and 4 deletions

View File

@ -242,6 +242,7 @@ std::string EmitGLSL(const Profile& profile, const RuntimeInfo& runtime_info, IR
} }
if (program.info.uses_subgroup_shuffles) { if (program.info.uses_subgroup_shuffles) {
ctx.header += "bool shfl_in_bounds;"; ctx.header += "bool shfl_in_bounds;";
ctx.header += "uint shfl_result;";
} }
ctx.code.insert(0, ctx.header); ctx.code.insert(0, ctx.header);
ctx.code += '}'; ctx.code += '}';

View File

@ -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)}; const auto src_thread_id{fmt::format("({})|({})", lhs, min_thread_id)};
ctx.Add("shfl_in_bounds=int({})<=int({});", src_thread_id, max_thread_id); ctx.Add("shfl_in_bounds=int({})<=int({});", src_thread_id, max_thread_id);
SetInBoundsFlag(ctx, inst); 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, 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)}; const auto src_thread_id{fmt::format("({}-{})", THREAD_ID, index)};
ctx.Add("shfl_in_bounds=int({})>=int({});", src_thread_id, max_thread_id); ctx.Add("shfl_in_bounds=int({})>=int({});", src_thread_id, max_thread_id);
SetInBoundsFlag(ctx, inst); 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, 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)}; const auto src_thread_id{fmt::format("({}+{})", THREAD_ID, index)};
ctx.Add("shfl_in_bounds=int({})<=int({});", src_thread_id, max_thread_id); ctx.Add("shfl_in_bounds=int({})<=int({});", src_thread_id, max_thread_id);
SetInBoundsFlag(ctx, inst); 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, 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)}; const auto src_thread_id{fmt::format("({}^{})", THREAD_ID, index)};
ctx.Add("shfl_in_bounds=int({})<=int({});", src_thread_id, max_thread_id); ctx.Add("shfl_in_bounds=int({})<=int({});", src_thread_id, max_thread_id);
SetInBoundsFlag(ctx, inst); 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, void EmitFSwizzleAdd(EmitContext& ctx, IR::Inst& inst, std::string_view op_a, std::string_view op_b,