zdoom/gzdoom
0
0
Fork 0
mirror of https://github.com/ZDoom/gzdoom.git synced 2024-11-27 22:33:17 +00:00
Code Issues Projects Releases Packages Wiki Activity

Remove use of the llvm.round intrinsic as it apparently doesn't work in some environments

Browse source
This commit is contained in:
Magnus Norddahl 2016-11-13 08:57:29 +01:00
parent 87e9ba0f05
commit af7a7ab7d7
4 changed files with 18 additions and 7 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

19
src/r_compiler/fixedfunction/drawtrianglecodegen.cpp
View file

@ -39,18 +39,25 @@ void DrawTriangleCodegen::Generate(TriDrawVariant variant, bool truecolor, SSAVa
LoopBlockY(variant, truecolor); LoopBlockY(variant, truecolor);
} }
SSAInt DrawTriangleCodegen::FloatTo28_4(SSAFloat v)
{
// SSAInt(SSAFloat::round(16.0f * v), false);
SSAInt a = SSAInt(v * 32.0f, false);
return (a + (a.ashr(31) | SSAInt(1))).ashr(1);
}
void DrawTriangleCodegen::Setup(TriDrawVariant variant, bool truecolor) void DrawTriangleCodegen::Setup(TriDrawVariant variant, bool truecolor)
{ {
int pixelsize = truecolor ? 4 : 1; int pixelsize = truecolor ? 4 : 1;
// 28.4 fixed-point coordinates // 28.4 fixed-point coordinates
Y1 = SSAInt(SSAFloat::round(16.0f * v1.y), false); Y1 = FloatTo28_4(v1.y);
Y2 = SSAInt(SSAFloat::round(16.0f * v2.y), false); Y2 = FloatTo28_4(v2.y);
Y3 = SSAInt(SSAFloat::round(16.0f * v3.y), false); Y3 = FloatTo28_4(v3.y);
X1 = SSAInt(SSAFloat::round(16.0f * v1.x), false); X1 = FloatTo28_4(v1.x);
X2 = SSAInt(SSAFloat::round(16.0f * v2.x), false); X2 = FloatTo28_4(v2.x);
X3 = SSAInt(SSAFloat::round(16.0f * v3.x), false); X3 = FloatTo28_4(v3.x);
// Deltas // Deltas
DX12 = X1 - X2; DX12 = X1 - X2;

1
src/r_compiler/fixedfunction/drawtrianglecodegen.h
View file

@ -40,6 +40,7 @@ private:
SSATriVertex LoadTriVertex(SSAValue v); SSATriVertex LoadTriVertex(SSAValue v);
void LoadUniforms(SSAValue uniforms); void LoadUniforms(SSAValue uniforms);
void Setup(TriDrawVariant variant, bool truecolor); void Setup(TriDrawVariant variant, bool truecolor);
SSAInt FloatTo28_4(SSAFloat v);
void LoopBlockY(TriDrawVariant variant, bool truecolor); void LoopBlockY(TriDrawVariant variant, bool truecolor);
void LoopBlockX(TriDrawVariant variant, bool truecolor); void LoopBlockX(TriDrawVariant variant, bool truecolor);
void LoopFullBlock(TriDrawVariant variant, bool truecolor); void LoopFullBlock(TriDrawVariant variant, bool truecolor);

3
src/r_compiler/ssa/ssa_float.cpp
View file

@ -112,12 +112,15 @@ SSAFloat SSAFloat::fma(SSAFloat a, SSAFloat b, SSAFloat c)
return SSAFloat::from_llvm(SSAScope::builder().CreateCall(SSAScope::intrinsic(llvm::Intrinsic::fma, params), args, SSAScope::hint())); return SSAFloat::from_llvm(SSAScope::builder().CreateCall(SSAScope::intrinsic(llvm::Intrinsic::fma, params), args, SSAScope::hint()));
} }
/* This intrinsic isn't always available..
SSAFloat SSAFloat::round(SSAFloat val) SSAFloat SSAFloat::round(SSAFloat val)
{ {
std::vector<llvm::Type *> params; std::vector<llvm::Type *> params;
params.push_back(SSAFloat::llvm_type()); params.push_back(SSAFloat::llvm_type());
return SSAFloat::from_llvm(SSAScope::builder().CreateCall(SSAScope::intrinsic(llvm::Intrinsic::round, params), val.v, SSAScope::hint())); return SSAFloat::from_llvm(SSAScope::builder().CreateCall(SSAScope::intrinsic(llvm::Intrinsic::round, params), val.v, SSAScope::hint()));
} }
*/
SSAFloat SSAFloat::floor(SSAFloat val) SSAFloat SSAFloat::floor(SSAFloat val)
{ {

2
src/r_compiler/ssa/ssa_float.h
View file

@ -43,7 +43,7 @@ public:
static SSAFloat exp(SSAFloat val); static SSAFloat exp(SSAFloat val);
static SSAFloat log(SSAFloat val); static SSAFloat log(SSAFloat val);
static SSAFloat fma(SSAFloat a, SSAFloat b, SSAFloat c); static SSAFloat fma(SSAFloat a, SSAFloat b, SSAFloat c);
static SSAFloat round(SSAFloat val); //static SSAFloat round(SSAFloat val);
static SSAFloat floor(SSAFloat val); static SSAFloat floor(SSAFloat val);
static SSAFloat MIN(SSAFloat a, SSAFloat b); static SSAFloat MIN(SSAFloat a, SSAFloat b);
static SSAFloat MAX(SSAFloat a, SSAFloat b); static SSAFloat MAX(SSAFloat a, SSAFloat b);