archive/rpcs3
1
0
Fork 0
mirror of https://github.com/RPCS3/rpcs3.git synced 2025-07-16 11:48:36 +12:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

rsx: Implement strict clamp16 operation needed for NVIDIA cards

Browse source
This commit is contained in:
kd-11 2019-04-12 16:25:12 +03:00 • committed by kd-11
parent 2ade3c594c
commit ee319f7c13
1 changed files with 20 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

23
rpcs3/Emu/RSX/Common/FragmentProgramDecompiler.cpp
View file

@ -249,7 +249,7 @@ std::string FragmentProgramDecompiler::ClampValue(const std::string& code, u32 p
case 0: case 0:
break; break;
case 1: case 1:
return "clamp(" + code + ", -65504., 65504.)"; return "clamp16(" + code + ")";
case 2: case 2:
return "clamp(" + code + ", -2., 2.)"; return "clamp(" + code + ", -2., 2.)";
case 3: case 3:
@ -545,10 +545,27 @@ std::string FragmentProgramDecompiler::BuildCode()
insertOutputs(OS); insertOutputs(OS);
OS << "\n"; OS << "\n";
//Insert global function definitions // Insert global function definitions
insertGlobalFunctions(OS); insertGlobalFunctions(OS);
//Declare register gather/merge if needed // Accurate float to half clamping (preserves IEEE-754 NaN)
OS <<
"vec4 clamp16(vec4 x)\n"
"{\n"
" bvec4 sel = isnan(x);\n"
" vec4 clamped = clamp(x, -65504., +65504.);\n"
" if (!any(sel))\n"
" {\n"
" return clamped;\n"
" }\n\n"
" return _select(clamped, x, sel);\n"
"}\n\n"
"vec3 clamp16(vec3 x){ return clamp16(x.xyzz).xyz; }\n"
"vec2 clamp16(vec2 x){ return clamp16(x.xyxy).xy; }\n"
"float clamp16(float x){ return isnan(x)? x : clamp(x, -65504., +65504.); }\n";
// Declare register gather/merge if needed
if (properties.has_gather_op) if (properties.has_gather_op)
{ {
std::string float2 = getFloatTypeName(2); std::string float2 = getFloatTypeName(2);