From 082b74fa241ba71f27b2e3fdf756824a0467f7f4 Mon Sep 17 00:00:00 2001 From: Yuri Kunde Schlesner Date: Mon, 24 Aug 2015 01:46:58 -0300 Subject: [PATCH] Shaders: Explicitly conform to PICA semantics in MAX/MIN --- src/video_core/shader/shader_interpreter.cpp | 10 ++++++++-- src/video_core/shader/shader_jit_x64.cpp | 2 ++ 2 files changed, 10 insertions(+), 2 deletions(-) diff --git a/src/video_core/shader/shader_interpreter.cpp b/src/video_core/shader/shader_interpreter.cpp index ae5a30441..69e4efa68 100644 --- a/src/video_core/shader/shader_interpreter.cpp +++ b/src/video_core/shader/shader_interpreter.cpp @@ -177,7 +177,10 @@ void RunInterpreter(UnitState& state) { if (!swizzle.DestComponentEnabled(i)) continue; - dest[i] = std::max(src1[i], src2[i]); + // NOTE: Exact form required to match NaN semantics to hardware: + // max(0, NaN) -> NaN + // max(NaN, 0) -> 0 + dest[i] = (src1[i] > src2[i]) ? src1[i] : src2[i]; } Record(state.debug, iteration, dest); break; @@ -190,7 +193,10 @@ void RunInterpreter(UnitState& state) { if (!swizzle.DestComponentEnabled(i)) continue; - dest[i] = std::min(src1[i], src2[i]); + // NOTE: Exact form required to match NaN semantics to hardware: + // min(0, NaN) -> NaN + // min(NaN, 0) -> 0 + dest[i] = (src1[i] < src2[i]) ? src1[i] : src2[i]; } Record(state.debug, iteration, dest); break; diff --git a/src/video_core/shader/shader_jit_x64.cpp b/src/video_core/shader/shader_jit_x64.cpp index e56bff2b8..456c8567d 100644 --- a/src/video_core/shader/shader_jit_x64.cpp +++ b/src/video_core/shader/shader_jit_x64.cpp @@ -467,6 +467,7 @@ void JitCompiler::Compile_FLR(Instruction instr) { void JitCompiler::Compile_MAX(Instruction instr) { Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1); Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2); + // SSE semantics match PICA200 ones: In case of NaN, SRC2 is returned. MAXPS(SRC1, R(SRC2)); Compile_DestEnable(instr, SRC1); } @@ -474,6 +475,7 @@ void JitCompiler::Compile_MAX(Instruction instr) { void JitCompiler::Compile_MIN(Instruction instr) { Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1); Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2); + // SSE semantics match PICA200 ones: In case of NaN, SRC2 is returned. MINPS(SRC1, R(SRC2)); Compile_DestEnable(instr, SRC1); }