Exverge/suyu
3
0
Fork 0
forked from suyu/suyu
Code Pull requests Activity

gl_arb_decompiler: Implement FSwizzleAdd

Browse source
This commit is contained in:
ReinUsesLisp 2020-06-09 03:03:41 -03:00
parent a63a0daa5e
commit 87011a97f9
1 changed files with 27 additions and 4 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

31
src/video_core/renderer_opengl/gl_arb_decompiler.cpp
View file

@ -786,6 +786,8 @@ ARBDecompiler::ARBDecompiler(const Device& device, const ShaderIR& ir, const Reg
ShaderType stage, std::string_view identifier) ShaderType stage, std::string_view identifier)
: device{device}, ir{ir}, registry{registry}, stage{stage} { : device{device}, ir{ir}, registry{registry}, stage{stage} {
AddLine("TEMP RC;"); AddLine("TEMP RC;");
AddLine("TEMP FSWZA[4];");
AddLine("TEMP FSWZB[4];");
if (ir.IsDecompiled()) { if (ir.IsDecompiled()) {
DecompileAST(); DecompileAST();
} else { } else {
@ -991,6 +993,15 @@ void ARBDecompiler::DeclareInternalFlags() {
} }
void ARBDecompiler::InitializeVariables() { void ARBDecompiler::InitializeVariables() {
AddLine("MOV.F32 FSWZA[0], -1;");
AddLine("MOV.F32 FSWZA[1], 1;");
AddLine("MOV.F32 FSWZA[2], -1;");
AddLine("MOV.F32 FSWZA[3], 0;");
AddLine("MOV.F32 FSWZB[0], -1;");
AddLine("MOV.F32 FSWZB[1], -1;");
AddLine("MOV.F32 FSWZB[2], 1;");
AddLine("MOV.F32 FSWZB[3], -1;");
if (stage == ShaderType::Vertex || stage == ShaderType::Geometry) { if (stage == ShaderType::Vertex || stage == ShaderType::Geometry) {
AddLine("MOV.F result.position, {{0, 0, 0, 1}};"); AddLine("MOV.F result.position, {{0, 0, 0, 1}};");
} }
@ -1570,10 +1581,22 @@ std::string ARBDecompiler::FSqrt(Operation operation) {
} }
std::string ARBDecompiler::FSwizzleAdd(Operation operation) { std::string ARBDecompiler::FSwizzleAdd(Operation operation) {
LOG_WARNING(Render_OpenGL, "(STUBBED)"); const std::string temporary = AllocVectorTemporary();
const std::string temporary = AllocTemporary(); if (!device.HasWarpIntrinsics()) {
AddLine("ADD.F {}, {}, {};", temporary, Visit(operation[0]), Visit(operation[1])); LOG_ERROR(Render_OpenGL,
return temporary; "NV_shader_thread_shuffle is missing. Kepler or better is required.");
AddLine("ADD.F {}.x, {}, {};", temporary, Visit(operation[0]), Visit(operation[1]));
return fmt::format("{}.x", temporary);
}
const std::string lut = AllocVectorTemporary();
AddLine("AND.U {}.z, {}.threadid, 3;", temporary, StageInputName(stage));
AddLine("SHL.U {}.z, {}.z, 1;", temporary, temporary);
AddLine("SHR.U {}.z, {}, {}.z;", temporary, Visit(operation[2]), temporary);
AddLine("AND.U {}.z, {}.z, 3;", temporary, temporary);
AddLine("MUL.F32 {}.x, {}, FSWZA[{}.z];", temporary, Visit(operation[0]), temporary);
AddLine("MUL.F32 {}.y, {}, FSWZB[{}.z];", temporary, Visit(operation[1]), temporary);
AddLine("ADD.F32 {}.x, {}.x, {}.y;", temporary, temporary, temporary);
return fmt::format("{}.x", temporary);
} }
std::string ARBDecompiler::HAdd2(Operation operation) { std::string ARBDecompiler::HAdd2(Operation operation) {