N-archive/citra
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

gl_shader_decompiler: Improve performance of accurate_mul on Android.

Browse source
This commit is contained in:
bunnei 2019-08-24 15:51:04 -04:00 committed by SachinVin
parent 8a68c58070
commit 9fbda04891
1 changed files with 18 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

18
src/video_core/renderer_opengl/gl_shader_decompiler.cpp
View file

@ -514,11 +514,21 @@ private:
}
case OpCode::Id::RCP: {
if (!sanitize_mul) {
// When accurate multiplication is OFF, NaN are not really handled. This is a
// workaround to cheaply avoid NaN. Fixes graphical issues in Ocarina of Time.
shader.AddLine("if ({}.x != 0.0)", src1);
}
SetDest(swizzle, dest_reg, fmt::format("(1.0 / {}.x)", src1), 4, 1);
break;
}
case OpCode::Id::RSQ: {
if (!sanitize_mul) {
// When accurate multiplication is OFF, NaN are not really handled. This is a
// workaround to cheaply avoid NaN. Fixes graphical issues in Ocarina of Time.
shader.AddLine("if ({}.x > 0.0)", src1);
}
SetDest(swizzle, dest_reg, fmt::format("inversesqrt({}.x)", src1), 4, 1);
break;
}
@ -807,6 +817,13 @@ private:
void Generate() {
if (sanitize_mul) {
#ifdef ANDROID
// Use a cheaper sanitize_mul on Android, as mobile GPUs struggle here
// This seems to be sufficient at least for Ocarina of Time and Attack on Titan accurate
// multiplication bugs
shader.AddLine(
"#define sanitize_mul(lhs, rhs) mix(lhs * rhs, vec4(0.0), isnan(lhs * rhs))");
#else
shader.AddLine("vec4 sanitize_mul(vec4 lhs, vec4 rhs) {{");
++shader.scope;
shader.AddLine("vec4 product = lhs * rhs;");
@ -814,6 +831,7 @@ private:
"isnan(lhs)), isnan(product));");
--shader.scope;
shader.AddLine("}}\n");
#endif
}
// Add declarations for registers