ComplexValueTy CodeGenFunction::EmitComplexDivSmiths(ComplexValueTy LHS, ComplexValueTy RHS) {
auto ElemTy = RHS.Re->getType();
// if(abs(d) <= abs(c)) then
auto FabsIntrinsic = GetIntrinsicFunction(llvm::Intrinsic::fabs, ElemTy);
auto Predicate = Builder.CreateFCmpOLE(Builder.CreateCall(FabsIntrinsic, RHS.Im),
Builder.CreateCall(FabsIntrinsic, RHS.Re));
auto ThenBlock = createBasicBlock("compdiv-then");
auto ElseBlock = createBasicBlock("compdiv-else");
auto MergeBlock = createBasicBlock("compdiv-done");
Builder.CreateCondBr(Predicate, ThenBlock, ElseBlock);
llvm::Value *R, *Den, *E, *F;
auto ResultRe = llvm::PHINode::Create(ElemTy, 2, "compdiv-re");
auto ResultIm = llvm::PHINode::Create(ElemTy, 2, "compdiv-im");
// r = d / c
// den = c + d * r
// e = (a + b * r) / den
// f = (b - a * r) / den
EmitBlock(ThenBlock);
R = Builder.CreateFDiv(RHS.Im, RHS.Re);
Den = Builder.CreateFAdd(RHS.Re, Builder.CreateFMul(RHS.Im, R));
E = Builder.CreateFDiv(Builder.CreateFAdd(LHS.Re,
Builder.CreateFMul(LHS.Im, R)), Den);
F = Builder.CreateFDiv(Builder.CreateFSub(LHS.Im,
Builder.CreateFMul(LHS.Re, R)), Den);
ResultRe->addIncoming(E, ThenBlock);
ResultIm->addIncoming(F, ThenBlock);
EmitBranch(MergeBlock);
// r = c / d
// den = c * r + d
// e = (a * r + b) / den
// f = (b * r - a) / den
EmitBlock(ElseBlock);
R = Builder.CreateFDiv(RHS.Re, RHS.Im);
Den = Builder.CreateFAdd(Builder.CreateFMul(RHS.Re, R), RHS.Im);
E = Builder.CreateFDiv(Builder.CreateFAdd(
Builder.CreateFMul(LHS.Re, R), LHS.Im), Den);
F = Builder.CreateFDiv(Builder.CreateFSub(
Builder.CreateFMul(LHS.Im, R), LHS.Re), Den);
ResultRe->addIncoming(E, ElseBlock);
ResultIm->addIncoming(F, ElseBlock);
EmitBranch(MergeBlock);
EmitBlock(MergeBlock);
Builder.Insert(ResultRe);
Builder.Insert(ResultIm);
return ComplexValueTy(ResultRe, ResultIm);
}
/*
========================
idDxtDecoder::DecompressNormalMapDXT1Renormalize
========================
*/
void idDxtDecoder::DecompressNormalMapDXT1Renormalize( const byte* inBuf, byte* outBuf, int width, int height )
{
byte block[64];
this->width = width;
this->height = height;
this->inData = inBuf;
for( int j = 0; j < height; j += 4 )
{
for( int i = 0; i < width; i += 4 )
{
DecodeColorValues( block, false, true );
for( int k = 0; k < 16; k++ )
{
float normal[3];
normal[0] = block[k * 4 + 0] / 255.0f * 2.0f - 1.0f;
normal[1] = block[k * 4 + 1] / 255.0f * 2.0f - 1.0f;
normal[2] = block[k * 4 + 2] / 255.0f * 2.0f - 1.0f;
float rsq = idMath::InvSqrt( normal[0] * normal[0] + normal[1] * normal[1] + normal[2] * normal[2] );
normal[0] *= rsq;
normal[1] *= rsq;
normal[2] *= rsq;
block[k * 4 + 0] = idMath::Ftob( ( normal[0] + 1.0f ) / 2.0f * 255.0f + 0.5f );
block[k * 4 + 1] = idMath::Ftob( ( normal[1] + 1.0f ) / 2.0f * 255.0f + 0.5f );
block[k * 4 + 2] = idMath::Ftob( ( normal[2] + 1.0f ) / 2.0f * 255.0f + 0.5f );
}
EmitBlock( outBuf, i, j, block );
}
}
}
/*
========================
idDxtDecoder::DecompressImageDXT1
========================
*/
void idDxtDecoder::DecompressImageDXT1( const byte* inBuf, byte* outBuf, int width, int height )
{
byte block[64];
this->width = width;
this->height = height;
this->inData = inBuf;
for( int j = 0; j < height; j += 4 )
{
for( int i = 0; i < width; i += 4 )
{
DecodeColorValues( block, false, true );
EmitBlock( outBuf, i, j, block );
}
}
}
/*
========================
idDxtDecoder::DecompressYCoCgCTX1DXT5A
========================
*/
void idDxtDecoder::DecompressYCoCgCTX1DXT5A( const byte* inBuf, byte* outBuf, int width, int height )
{
byte block[64];
this->width = width;
this->height = height;
this->inData = inBuf;
for( int j = 0; j < height; j += 4 )
{
for( int i = 0; i < width; i += 4 )
{
DecodeAlphaValues( block, 3 );
DecodeCTX1Values( block );
EmitBlock( outBuf, i, j, block );
}
}
}
/*
========================
idDxtDecoder::DecompressNormalMapDXT1
========================
*/
void idDxtDecoder::DecompressNormalMapDXT1( const byte* inBuf, byte* outBuf, int width, int height )
{
byte block[64];
this->width = width;
this->height = height;
this->inData = inBuf;
for( int j = 0; j < height; j += 4 )
{
for( int i = 0; i < width; i += 4 )
{
DecodeColorValues( block, false, true );
#if 1
float normals[16 * 4];
/*
for ( int k = 0; k < 16; k++ ) {
normals[k*4+0] = block[k*4+0] / 255.0f * 2.0f - 1.0f;
normals[k*4+1] = block[k*4+1] / 255.0f * 2.0f - 1.0f;
}
*/
UnRotateNormals( block, normals, block[0 * 4 + 2], 0 );
for( int k = 0; k < 16; k++ )
{
float x = normals[k * 4 + 0];
float y = normals[k * 4 + 1];
float z = 1.0f - x * x - y * y;
if( z < 0.0f ) z = 0.0f;
normals[k * 4 + 2] = sqrt( z );
}
for( int k = 0; k < 16; k++ )
{
block[k * 4 + 0] = idMath::Ftob( ( normals[k * 4 + 0] + 1.0f ) / 2.0f * 255.0f );
block[k * 4 + 1] = idMath::Ftob( ( normals[k * 4 + 1] + 1.0f ) / 2.0f * 255.0f );
block[k * 4 + 2] = idMath::Ftob( ( normals[k * 4 + 2] + 1.0f ) / 2.0f * 255.0f );
}
#else
DeriveNormalZValues( block );
#endif
EmitBlock( outBuf, i, j, block );
}
}
}
/*
========================
idDxtDecoder::DecompressNormalMapDXT5
========================
*/
void idDxtDecoder::DecompressNormalMapDXT5( const byte* inBuf, byte* outBuf, int width, int height )
{
byte block[64];
byte c0, c1;
this->width = width;
this->height = height;
this->inData = inBuf;
for( int j = 0; j < height; j += 4 )
{
for( int i = 0; i < width; i += 4 )
{
DecodeAlphaValues( block, 0 );
DecodeNormalYValues( block, 1, c0, c1 );
#if 1
float normals[16 * 4];
//BiasScaleNormals( block, normals, c0, c1 );
UnRotateNormals( block, normals, c0, c1 );
for( int k = 0; k < 16; k++ )
{
float x = normals[k * 4 + 0];
float y = normals[k * 4 + 1];
float z = 1.0f - x * x - y * y;
if( z < 0.0f ) z = 0.0f;
normals[k * 4 + 2] = sqrt( z );
}
for( int k = 0; k < 16; k++ )
{
block[k * 4 + 0] = idMath::Ftob( ( normals[k * 4 + 0] + 1.0f ) / 2.0f * 255.0f );
block[k * 4 + 1] = idMath::Ftob( ( normals[k * 4 + 1] + 1.0f ) / 2.0f * 255.0f );
block[k * 4 + 2] = idMath::Ftob( ( normals[k * 4 + 2] + 1.0f ) / 2.0f * 255.0f );
}
#else
BiasScaleNormalY( block, 1, c0, c1 );
DeriveNormalZValues( block );
#endif
EmitBlock( outBuf, i, j, block );
}
}
}