Skip to content

S
swiftshader
Commit 0f448078 by Alexis Hetu Committed by Alexis Hétu
Options

Fixed some unary operators

There were a few issues in unary operators: - Many were not compiling because the promote function had not been adjusted to take the new builtin functions into account - abs and sign had not been implemented for int - For the integer abs version, used pabsd. Removed the extra argument, which seemed unnecessary (abs should have 1 input, 1 output, AFAIK). Change-Id: If02c5040438e8c45c99fc7b3c55107448c85cf58 Reviewed-on: https://swiftshader-review.googlesource.com/4970Tested-by: 's avatarAlexis Hétu <sugoi@google.com> Reviewed-by: 's avatarNicolas Capens <capn@google.com>
parent dc2bc1b0
Showing with 85 additions and 5 deletions
src/OpenGL/compiler/Intermediate.cpp
...@@ -752,6 +752,20 @@ bool TIntermUnary::promote(TInfoSink&) ...@@ -752,6 +752,20 @@ bool TIntermUnary::promote(TInfoSink&)
case EOpAny: case EOpAny:
case EOpAll: case EOpAll:
case EOpVectorLogicalNot: case EOpVectorLogicalNot:
case EOpAbs:
case EOpSign:
case EOpIsNan:
case EOpIsInf:
case EOpFloatBitsToInt:
case EOpFloatBitsToUint:
case EOpIntBitsToFloat:
case EOpUintBitsToFloat:
case EOpPackSnorm2x16:
case EOpPackUnorm2x16:
case EOpPackHalf2x16:
case EOpUnpackSnorm2x16:
case EOpUnpackUnorm2x16:
case EOpUnpackHalf2x16:
return true; return true;
default: default:
......
src/OpenGL/compiler/OutputASM.cpp
...@@ -368,6 +368,24 @@ namespace glsl ...@@ -368,6 +368,24 @@ namespace glsl
default: default:
return op; return op;
} }
case sw::Shader::OPCODE_ABS:
switch(baseType)
{
case EbtInt:
return sw::Shader::OPCODE_IABS;
case EbtFloat:
default:
return op;
}
case sw::Shader::OPCODE_SGN:
switch(baseType)
{
case EbtInt:
return sw::Shader::OPCODE_ISGN;
case EbtFloat:
default:
return op;
}
case sw::Shader::OPCODE_ADD: case sw::Shader::OPCODE_ADD:
switch(baseType) switch(baseType)
{ {
...@@ -984,8 +1002,8 @@ namespace glsl ...@@ -984,8 +1002,8 @@ namespace glsl
case EOpLog2: if(visit == PostVisit) emit(sw::Shader::OPCODE_LOG2, result, arg); break; case EOpLog2: if(visit == PostVisit) emit(sw::Shader::OPCODE_LOG2, result, arg); break;
case EOpSqrt: if(visit == PostVisit) emit(sw::Shader::OPCODE_SQRT, result, arg); break; case EOpSqrt: if(visit == PostVisit) emit(sw::Shader::OPCODE_SQRT, result, arg); break;
case EOpInverseSqrt: if(visit == PostVisit) emit(sw::Shader::OPCODE_RSQ, result, arg); break; case EOpInverseSqrt: if(visit == PostVisit) emit(sw::Shader::OPCODE_RSQ, result, arg); break;
case EOpAbs: if(visit == PostVisit) emit(sw::Shader::OPCODE_ABS, result, arg); break; case EOpAbs: if(visit == PostVisit) emit(getOpcode(sw::Shader::OPCODE_ABS, result), result, arg); break;
case EOpSign: if(visit == PostVisit) emit(sw::Shader::OPCODE_SGN, result, arg); break; case EOpSign: if(visit == PostVisit) emit(getOpcode(sw::Shader::OPCODE_SGN, result), result, arg); break;
case EOpFloor: if(visit == PostVisit) emit(sw::Shader::OPCODE_FLOOR, result, arg); break; case EOpFloor: if(visit == PostVisit) emit(sw::Shader::OPCODE_FLOOR, result, arg); break;
case EOpTrunc: if(visit == PostVisit) emit(sw::Shader::OPCODE_TRUNC, result, arg); break; case EOpTrunc: if(visit == PostVisit) emit(sw::Shader::OPCODE_TRUNC, result, arg); break;
case EOpRound: if(visit == PostVisit) emit(sw::Shader::OPCODE_ROUND, result, arg); break; case EOpRound: if(visit == PostVisit) emit(sw::Shader::OPCODE_ROUND, result, arg); break;
......
src/Reactor/Nucleus.cpp
...@@ -3579,6 +3579,19 @@ namespace sw ...@@ -3579,6 +3579,19 @@ namespace sw
return x86::pmaddwd(x, y); // FIXME: Fallback required return x86::pmaddwd(x, y); // FIXME: Fallback required
} }
RValue<Int4> Abs(RValue<Int4> x)
{
if(CPUID::supportsSSSE3())
{
return x86::pabsd(x);
}
else
{
Int4 mask = (x >> 31);
return (mask ^ x) - mask;
}
}
RValue<Short8> MulHigh(RValue<Short8> x, RValue<Short8> y) RValue<Short8> MulHigh(RValue<Short8> x, RValue<Short8> y)
{ {
return x86::pmulhw(x, y); // FIXME: Fallback required return x86::pmulhw(x, y); // FIXME: Fallback required
...@@ -7157,12 +7170,12 @@ namespace sw ...@@ -7157,12 +7170,12 @@ namespace sw
return cmpss(x, y, 7); return cmpss(x, y, 7);
} }
RValue<Int4> pabsd(RValue<Int4> x, RValue<Int4> y) RValue<Int4> pabsd(RValue<Int4> x)
{ {
Module *module = Nucleus::getModule(); Module *module = Nucleus::getModule();
llvm::Function *pabsd = Intrinsic::getDeclaration(module, Intrinsic::x86_ssse3_pabs_d_128); llvm::Function *pabsd = Intrinsic::getDeclaration(module, Intrinsic::x86_ssse3_pabs_d_128);
return RValue<Int4>(Nucleus::createCall(pabsd, x.value, y.value)); return RValue<Int4>(Nucleus::createCall(pabsd, x.value));
} }
RValue<Short4> paddsw(RValue<Short4> x, RValue<Short4> y) RValue<Short4> paddsw(RValue<Short4> x, RValue<Short4> y)
......
src/Reactor/Nucleus.hpp
...@@ -1115,6 +1115,7 @@ namespace sw ...@@ -1115,6 +1115,7 @@ namespace sw
RValue<Short8> MulHigh(RValue<Short8> x, RValue<Short8> y); RValue<Short8> MulHigh(RValue<Short8> x, RValue<Short8> y);
RValue<Int4> MulAdd(RValue<Short8> x, RValue<Short8> y); RValue<Int4> MulAdd(RValue<Short8> x, RValue<Short8> y);
RValue<Int4> Abs(RValue<Int4> x);
class UShort8 : public Variable<UShort8> class UShort8 : public Variable<UShort8>
{ {
......
src/Reactor/x86.hpp
...@@ -58,7 +58,7 @@ namespace sw ...@@ -58,7 +58,7 @@ namespace sw
RValue<Float> cmpnless(RValue<Float> x, RValue<Float> y); RValue<Float> cmpnless(RValue<Float> x, RValue<Float> y);
RValue<Float> cmpordss(RValue<Float> x, RValue<Float> y); RValue<Float> cmpordss(RValue<Float> x, RValue<Float> y);
RValue<Int4> pabsd(RValue<Int4> x, RValue<Int4> y); RValue<Int4> pabsd(RValue<Int4> x);
RValue<Short4> paddsw(RValue<Short4> x, RValue<Short4> y); RValue<Short4> paddsw(RValue<Short4> x, RValue<Short4> y);
RValue<Short4> psubsw(RValue<Short4> x, RValue<Short4> y); RValue<Short4> psubsw(RValue<Short4> x, RValue<Short4> y);
......
src/Shader/PixelProgram.cpp
...@@ -231,6 +231,7 @@ namespace sw ...@@ -231,6 +231,7 @@ namespace sw
case Shader::OPCODE_POWX: powx(d, s0, s1, pp); break; case Shader::OPCODE_POWX: powx(d, s0, s1, pp); break;
case Shader::OPCODE_POW: pow(d, s0, s1, pp); break; case Shader::OPCODE_POW: pow(d, s0, s1, pp); break;
case Shader::OPCODE_SGN: sgn(d, s0); break; case Shader::OPCODE_SGN: sgn(d, s0); break;
case Shader::OPCODE_ISGN: isgn(d, s0); break;
case Shader::OPCODE_CRS: crs(d, s0, s1); break; case Shader::OPCODE_CRS: crs(d, s0, s1); break;
case Shader::OPCODE_FORWARD1: forward1(d, s0, s1, s2); break; case Shader::OPCODE_FORWARD1: forward1(d, s0, s1, s2); break;
case Shader::OPCODE_FORWARD2: forward2(d, s0, s1, s2); break; case Shader::OPCODE_FORWARD2: forward2(d, s0, s1, s2); break;
...@@ -248,6 +249,7 @@ namespace sw ...@@ -248,6 +249,7 @@ namespace sw
case Shader::OPCODE_NRM3: nrm3(d, s0, pp); break; case Shader::OPCODE_NRM3: nrm3(d, s0, pp); break;
case Shader::OPCODE_NRM4: nrm4(d, s0, pp); break; case Shader::OPCODE_NRM4: nrm4(d, s0, pp); break;
case Shader::OPCODE_ABS: abs(d, s0); break; case Shader::OPCODE_ABS: abs(d, s0); break;
case Shader::OPCODE_IABS: iabs(d, s0); break;
case Shader::OPCODE_SINCOS: sincos(d, s0, pp); break; case Shader::OPCODE_SINCOS: sincos(d, s0, pp); break;
case Shader::OPCODE_COS: cos(d, s0, pp); break; case Shader::OPCODE_COS: cos(d, s0, pp); break;
case Shader::OPCODE_SIN: sin(d, s0, pp); break; case Shader::OPCODE_SIN: sin(d, s0, pp); break;
......
src/Shader/Shader.cpp
...@@ -821,7 +821,9 @@ namespace sw ...@@ -821,7 +821,9 @@ namespace sw
case OPCODE_POWX: return "powx"; case OPCODE_POWX: return "powx";
case OPCODE_CRS: return "crs"; case OPCODE_CRS: return "crs";
case OPCODE_SGN: return "sgn"; case OPCODE_SGN: return "sgn";
case OPCODE_ISGN: return "isgn";
case OPCODE_ABS: return "abs"; case OPCODE_ABS: return "abs";
case OPCODE_IABS: return "iabs";
case OPCODE_NRM2: return "nrm2"; case OPCODE_NRM2: return "nrm2";
case OPCODE_NRM3: return "nrm3"; case OPCODE_NRM3: return "nrm3";
case OPCODE_NRM4: return "nrm4"; case OPCODE_NRM4: return "nrm4";
......
src/Shader/Shader.hpp
...@@ -244,6 +244,8 @@ namespace sw ...@@ -244,6 +244,8 @@ namespace sw
// Integer opcodes // Integer opcodes
OPCODE_INEG, OPCODE_INEG,
OPCODE_IABS,
OPCODE_ISGN,
OPCODE_IADD, OPCODE_IADD,
OPCODE_ISUB, OPCODE_ISUB,
OPCODE_IMUL, OPCODE_IMUL,
......
src/Shader/ShaderCore.cpp
...@@ -1357,6 +1357,14 @@ namespace sw ...@@ -1357,6 +1357,14 @@ namespace sw
sgn(dst.w, src.w); sgn(dst.w, src.w);
} }
void ShaderCore::isgn(Vector4f &dst, const Vector4f &src)
{
isgn(dst.x, src.x);
isgn(dst.y, src.y);
isgn(dst.z, src.z);
isgn(dst.w, src.w);
}
void ShaderCore::abs(Vector4f &dst, const Vector4f &src) void ShaderCore::abs(Vector4f &dst, const Vector4f &src)
{ {
dst.x = Abs(src.x); dst.x = Abs(src.x);
...@@ -1365,6 +1373,14 @@ namespace sw ...@@ -1365,6 +1373,14 @@ namespace sw
dst.w = Abs(src.w); dst.w = Abs(src.w);
} }
void ShaderCore::iabs(Vector4f &dst, const Vector4f &src)
{
dst.x = As<Float4>(Abs(As<Int4>(src.x)));
dst.y = As<Float4>(Abs(As<Int4>(src.y)));
dst.z = As<Float4>(Abs(As<Int4>(src.z)));
dst.w = As<Float4>(Abs(As<Int4>(src.w)));
}
void ShaderCore::nrm2(Vector4f &dst, const Vector4f &src, bool pp) void ShaderCore::nrm2(Vector4f &dst, const Vector4f &src, bool pp)
{ {
Float4 dot = dot2(src, src); Float4 dot = dot2(src, src);
...@@ -1595,6 +1611,13 @@ namespace sw ...@@ -1595,6 +1611,13 @@ namespace sw
dst = As<Float4>(neg | pos); dst = As<Float4>(neg | pos);
} }
void ShaderCore::isgn(Float4 &dst, const Float4 &src)
{
Int4 neg = CmpLT(As<Int4>(src), Int4(0)) & Int4(-1);
Int4 pos = CmpNLE(As<Int4>(src), Int4(0)) & Int4(1);
dst = As<Float4>(neg | pos);
}
void ShaderCore::cmp0(Float4 &dst, const Float4 &src0, const Float4 &src1, const Float4 &src2) void ShaderCore::cmp0(Float4 &dst, const Float4 &src0, const Float4 &src1, const Float4 &src2)
{ {
Int4 pos = CmpLE(Float4(0.0f), src0); Int4 pos = CmpLE(Float4(0.0f), src0);
......
src/Shader/ShaderCore.hpp
...@@ -335,7 +335,9 @@ namespace sw ...@@ -335,7 +335,9 @@ namespace sw
void refract3(Vector4f &dst, const Vector4f &src0, const Vector4f &src1, const Float4 &src2); void refract3(Vector4f &dst, const Vector4f &src0, const Vector4f &src1, const Float4 &src2);
void refract4(Vector4f &dst, const Vector4f &src0, const Vector4f &src1, const Float4 &src2); void refract4(Vector4f &dst, const Vector4f &src0, const Vector4f &src1, const Float4 &src2);
void sgn(Vector4f &dst, const Vector4f &src); void sgn(Vector4f &dst, const Vector4f &src);
void isgn(Vector4f &dst, const Vector4f &src);
void abs(Vector4f &dst, const Vector4f &src); void abs(Vector4f &dst, const Vector4f &src);
void iabs(Vector4f &dst, const Vector4f &src);
void nrm2(Vector4f &dst, const Vector4f &src, bool pp = false); void nrm2(Vector4f &dst, const Vector4f &src, bool pp = false);
void nrm3(Vector4f &dst, const Vector4f &src, bool pp = false); void nrm3(Vector4f &dst, const Vector4f &src, bool pp = false);
void nrm4(Vector4f &dst, const Vector4f &src, bool pp = false); void nrm4(Vector4f &dst, const Vector4f &src, bool pp = false);
...@@ -373,6 +375,7 @@ namespace sw ...@@ -373,6 +375,7 @@ namespace sw
private: private:
void sgn(Float4 &dst, const Float4 &src); void sgn(Float4 &dst, const Float4 &src);
void isgn(Float4 &dst, const Float4 &src);
void cmp0(Float4 &dst, const Float4 &src0, const Float4 &src1, const Float4 &src2); void cmp0(Float4 &dst, const Float4 &src0, const Float4 &src1, const Float4 &src2);
void cmp0i(Float4 &dst, const Float4 &src0, const Float4 &src1, const Float4 &src2); void cmp0i(Float4 &dst, const Float4 &src0, const Float4 &src1, const Float4 &src2);
void select(Float4 &dst, RValue<Int4> src0, const Float4 &src1, const Float4 &src2); void select(Float4 &dst, RValue<Int4> src0, const Float4 &src1, const Float4 &src2);
......
src/Shader/VertexProgram.cpp
...@@ -157,6 +157,7 @@ namespace sw ...@@ -157,6 +157,7 @@ namespace sw
case Shader::OPCODE_DEFB: break; case Shader::OPCODE_DEFB: break;
case Shader::OPCODE_NOP: break; case Shader::OPCODE_NOP: break;
case Shader::OPCODE_ABS: abs(d, s0); break; case Shader::OPCODE_ABS: abs(d, s0); break;
case Shader::OPCODE_IABS: iabs(d, s0); break;
case Shader::OPCODE_ADD: add(d, s0, s1); break; case Shader::OPCODE_ADD: add(d, s0, s1); break;
case Shader::OPCODE_IADD: iadd(d, s0, s1); break; case Shader::OPCODE_IADD: iadd(d, s0, s1); break;
case Shader::OPCODE_CRS: crs(d, s0, s1); break; case Shader::OPCODE_CRS: crs(d, s0, s1); break;
...@@ -256,6 +257,7 @@ namespace sw ...@@ -256,6 +257,7 @@ namespace sw
case Shader::OPCODE_DIST4: dist4(d.x, s0, s1, pp); break; case Shader::OPCODE_DIST4: dist4(d.x, s0, s1, pp); break;
case Shader::OPCODE_SGE: step(d, s1, s0); break; case Shader::OPCODE_SGE: step(d, s1, s0); break;
case Shader::OPCODE_SGN: sgn(d, s0); break; case Shader::OPCODE_SGN: sgn(d, s0); break;
case Shader::OPCODE_ISGN: isgn(d, s0); break;
case Shader::OPCODE_SINCOS: sincos(d, s0, pp); break; case Shader::OPCODE_SINCOS: sincos(d, s0, pp); break;
case Shader::OPCODE_COS: cos(d, s0, pp); break; case Shader::OPCODE_COS: cos(d, s0, pp); break;
case Shader::OPCODE_SIN: sin(d, s0, pp); break; case Shader::OPCODE_SIN: sin(d, s0, pp); break;
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment