Downloafy/suyu-forked
1
0
Fork 1
forked from suyu/suyu
Code Pull requests 1 Activity

emit_glsl_atomic: Implement 32x2 fallback atomic ops

Browse source
This commit is contained in:
ameerj 2022-01-29 15:49:18 -05:00
parent 90a0506d56
commit e394e1ecc4
1 changed files with 55 additions and 9 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

64
src/shader_recompiler/backend/glsl/emit_glsl_atomic.cpp
View file

@ -274,47 +274,93 @@ void EmitStorageAtomicExchange64(EmitContext& ctx, IR::Inst& inst, const IR::Val
void EmitStorageAtomicIAdd32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding, void EmitStorageAtomicIAdd32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) { const IR::Value& offset, std::string_view value) {
throw NotImplementedException("GLSL Instrucion"); LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to non-atomic");
ctx.AddU32x2("{}=uvec2({}_ssbo{}[{}>>2],{}_ssbo{}[({}>>2)+1]);", inst, ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name, binding.U32(),
ctx.var_alloc.Consume(offset));
ctx.Add("{}_ssbo{}[{}>>2]+={}.x;{}_ssbo{}[({}>>2)+1]+={}.y;", ctx.stage_name, binding.U32(),
ctx.var_alloc.Consume(offset), value, ctx.stage_name, binding.U32(),
ctx.var_alloc.Consume(offset), value);
} }
void EmitStorageAtomicSMin32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding, void EmitStorageAtomicSMin32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) { const IR::Value& offset, std::string_view value) {
throw NotImplementedException("GLSL Instrucion"); LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to non-atomic");
ctx.AddU32x2("{}=ivec2({}_ssbo{}[{}>>2],{}_ssbo{}[({}>>2)+1]);", inst, ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name, binding.U32(),
ctx.var_alloc.Consume(offset));
ctx.Add("for(int "
"i=0;i<2;++i){{{}_ssbo{}[({}>>2)+i]=uint(min(int({}_ssbo{}[({}>>2)+i]),int({}[i])));}}",
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), value);
} }
void EmitStorageAtomicUMin32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding, void EmitStorageAtomicUMin32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) { const IR::Value& offset, std::string_view value) {
throw NotImplementedException("GLSL Instrucion"); LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to non-atomic");
ctx.AddU32x2("{}=uvec2({}_ssbo{}[{}>>2],{}_ssbo{}[({}>>2)+1]);", inst, ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name, binding.U32(),
ctx.var_alloc.Consume(offset));
ctx.Add("for(int i=0;i<2;++i){{ "
"{}_ssbo{}[({}>>2)+i]=min({}_ssbo{}[({}>>2)+i],{}[i]);}}",
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), value);
} }
void EmitStorageAtomicSMax32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding, void EmitStorageAtomicSMax32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) { const IR::Value& offset, std::string_view value) {
throw NotImplementedException("GLSL Instrucion"); LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to non-atomic");
ctx.AddU32x2("{}=ivec2({}_ssbo{}[{}>>2],{}_ssbo{}[({}>>2)+1]);", inst, ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name, binding.U32(),
ctx.var_alloc.Consume(offset));
ctx.Add("for(int "
"i=0;i<2;++i){{{}_ssbo{}[({}>>2)+i]=uint(max(int({}_ssbo{}[({}>>2)+i]),int({}[i])));}}",
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), value);
} }
void EmitStorageAtomicUMax32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding, void EmitStorageAtomicUMax32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) { const IR::Value& offset, std::string_view value) {
throw NotImplementedException("GLSL Instrucion"); LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to non-atomic");
ctx.AddU32x2("{}=uvec2({}_ssbo{}[{}>>2],{}_ssbo{}[({}>>2)+1]);", inst, ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name, binding.U32(),
ctx.var_alloc.Consume(offset));
ctx.Add("for(int i=0;i<2;++i){{{}_ssbo{}[({}>>2)+i]=max({}_ssbo{}[({}>>2)+i],{}[i]);}}",
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), value);
} }
void EmitStorageAtomicAnd32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding, void EmitStorageAtomicAnd32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) { const IR::Value& offset, std::string_view value) {
throw NotImplementedException("GLSL Instrucion"); LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to 32x2");
ctx.AddU32x2("{}=uvec2(atomicAnd({}_ssbo{}[{}>>2],{}.x),atomicAnd({}_ssbo{}[({}>>2)+1],{}.y));",
inst, ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), value,
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), value);
} }
void EmitStorageAtomicOr32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding, void EmitStorageAtomicOr32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) { const IR::Value& offset, std::string_view value) {
throw NotImplementedException("GLSL Instrucion"); LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to 32x2");
ctx.AddU32x2("{}=uvec2(atomicOr({}_ssbo{}[{}>>2],{}.x),atomicOr({}_ssbo{}[({}>>2)+1],{}.y));",
inst, ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), value,
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), value);
} }
void EmitStorageAtomicXor32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding, void EmitStorageAtomicXor32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) { const IR::Value& offset, std::string_view value) {
throw NotImplementedException("GLSL Instrucion"); LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to 32x2");
ctx.AddU32x2("{}=uvec2(atomicXor({}_ssbo{}[{}>>2],{}.x),atomicXor({}_ssbo{}[({}>>2)+1],{}.y));",
inst, ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), value,
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), value);
} }
void EmitStorageAtomicExchange32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding, void EmitStorageAtomicExchange32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) { const IR::Value& offset, std::string_view value) {
throw NotImplementedException("GLSL Instrucion"); LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to 32x2");
ctx.AddU32x2("{}=uvec2(atomicExchange({}_ssbo{}[{}>>2],{}.x),atomicExchange({}_ssbo{}[({}>>2)+"
"1],{}.y));",
inst, ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), value,
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), value);
} }
void EmitStorageAtomicAddF32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding, void EmitStorageAtomicAddF32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,