mirrors
/
dxvk
mirror of https://github.com/doitsujin/dxvk
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

[dxbc] Implement sample operations with sparse feedback

This commit is contained in:
Philip Rebohle 2022-08-22 03:54:48 +02:00
parent d5b68b3642
commit 0faba649da
1 changed files with 73 additions and 29 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

102
src/dxbc/dxbc_compiler.cpp
View File

@ -3536,14 +3536,17 @@ namespace dxvk {
// Load reference value for depth-compare operations
const bool isDepthCompare = ins.op == DxbcOpcode::SampleC
|| ins.op == DxbcOpcode::SampleClz;
|| ins.op == DxbcOpcode::SampleClz
|| ins.op == DxbcOpcode::SampleCClampS
|| ins.op == DxbcOpcode::SampleClzS;
const DxbcRegisterValue referenceValue = isDepthCompare
? emitRegisterLoad(ins.src[3], DxbcRegMask(true, false, false, false))
: DxbcRegisterValue();
// Load explicit gradients for sample operations that require them
const bool hasExplicitGradients = ins.op == DxbcOpcode::SampleD;
const bool hasExplicitGradients = ins.op == DxbcOpcode::SampleD
|| ins.op == DxbcOpcode::SampleDClampS;
const DxbcRegisterValue explicitGradientX = hasExplicitGradients
? emitRegisterLoad(ins.src[3], DxbcRegMask::firstN(imageLayerDim))
@ -3555,16 +3558,29 @@ namespace dxvk {
// LOD for certain sample operations
const bool hasLod = ins.op == DxbcOpcode::SampleL
|| ins.op == DxbcOpcode::SampleB;
|| ins.op == DxbcOpcode::SampleLS
|| ins.op == DxbcOpcode::SampleB
|| ins.op == DxbcOpcode::SampleBClampS;
const DxbcRegisterValue lod = hasLod
? emitRegisterLoad(ins.src[3], DxbcRegMask(true, false, false, false))
: DxbcRegisterValue();
// Min LOD for certain sparse operations
const bool hasMinLod = ins.op == DxbcOpcode::SampleClampS
|| ins.op == DxbcOpcode::SampleBClampS
|| ins.op == DxbcOpcode::SampleDClampS
|| ins.op == DxbcOpcode::SampleCClampS;
const DxbcRegisterValue minLod = hasMinLod && ins.src[ins.srcCount - 1].type != DxbcOperandType::Null
? emitRegisterLoad(ins.src[ins.srcCount - 1], DxbcRegMask(true, false, false, false))
: DxbcRegisterValue();
// Accumulate additional image operands. These are
// not part of the actual operand token in SPIR-V.
SpirvImageOperands imageOperands;
imageOperands.sparse = ins.dstCount == 2;
if (ins.sampleControls.u != 0 || ins.sampleControls.v != 0 || ins.sampleControls.w != 0) {
const std::array<uint32_t, 3> offsetIds = {
imageLayerDim >= 1 ? m_module.consti32(ins.sampleControls.u) : 0,
@ -3577,70 +3593,89 @@ namespace dxvk {
getVectorTypeId({ DxbcScalarType::Sint32, imageLayerDim }),
imageLayerDim, offsetIds.data());
}
if (hasMinLod) {
m_module.enableCapability(spv::CapabilityMinLod);
imageOperands.flags |= spv::ImageOperandsMinLodMask;
imageOperands.sMinLod = minLod.id;
}
// Combine the texture and the sampler into a sampled image
uint32_t sampledImageId = emitLoadSampledImage(texture, sampler, isDepthCompare);
// Sampling an image always returns a four-component
// vector, whereas depth-compare ops return a scalar.
DxbcRegisterValue result;
result.type.ctype = texture.sampledType;
result.type.ccount = isDepthCompare ? 1 : 4;
DxbcVectorType texelType;
texelType.ctype = texture.sampledType;
texelType.ccount = isDepthCompare ? 1 : 4;
uint32_t texelTypeId = getVectorTypeId(texelType);
uint32_t resultTypeId = texelTypeId;
uint32_t resultId = 0;
if (imageOperands.sparse)
resultTypeId = getSparseResultTypeId(texelTypeId);
switch (ins.op) {
// Simple image sample operation
case DxbcOpcode::Sample: {
result.id = m_module.opImageSampleImplicitLod(
getVectorTypeId(result.type),
sampledImageId, coord.id,
case DxbcOpcode::Sample:
case DxbcOpcode::SampleClampS: {
resultId = m_module.opImageSampleImplicitLod(
resultTypeId, sampledImageId, coord.id,
imageOperands);
} break;
// Depth-compare operation
case DxbcOpcode::SampleC: {
result.id = m_module.opImageSampleDrefImplicitLod(
getVectorTypeId(result.type), sampledImageId, coord.id,
case DxbcOpcode::SampleC:
case DxbcOpcode::SampleCClampS: {
resultId = m_module.opImageSampleDrefImplicitLod(
resultTypeId, sampledImageId, coord.id,
referenceValue.id, imageOperands);
} break;
// Depth-compare operation on mip level zero
case DxbcOpcode::SampleClz: {
case DxbcOpcode::SampleClz:
case DxbcOpcode::SampleClzS: {
imageOperands.flags |= spv::ImageOperandsLodMask;
imageOperands.sLod = m_module.constf32(0.0f);
result.id = m_module.opImageSampleDrefExplicitLod(
getVectorTypeId(result.type), sampledImageId, coord.id,
resultId = m_module.opImageSampleDrefExplicitLod(
resultTypeId, sampledImageId, coord.id,
referenceValue.id, imageOperands);
} break;
// Sample operation with explicit gradients
case DxbcOpcode::SampleD: {
case DxbcOpcode::SampleD:
case DxbcOpcode::SampleDClampS: {
imageOperands.flags |= spv::ImageOperandsGradMask;
imageOperands.sGradX = explicitGradientX.id;
imageOperands.sGradY = explicitGradientY.id;
result.id = m_module.opImageSampleExplicitLod(
getVectorTypeId(result.type), sampledImageId, coord.id,
resultId = m_module.opImageSampleExplicitLod(
resultTypeId, sampledImageId, coord.id,
imageOperands);
} break;
// Sample operation with explicit LOD
case DxbcOpcode::SampleL: {
case DxbcOpcode::SampleL:
case DxbcOpcode::SampleLS: {
imageOperands.flags |= spv::ImageOperandsLodMask;
imageOperands.sLod = lod.id;
result.id = m_module.opImageSampleExplicitLod(
getVectorTypeId(result.type), sampledImageId, coord.id,
resultId = m_module.opImageSampleExplicitLod(
resultTypeId, sampledImageId, coord.id,
imageOperands);
} break;
// Sample operation with LOD bias
case DxbcOpcode::SampleB: {
case DxbcOpcode::SampleB:
case DxbcOpcode::SampleBClampS: {
imageOperands.flags |= spv::ImageOperandsBiasMask;
imageOperands.sLodBias = lod.id;
result.id = m_module.opImageSampleImplicitLod(
getVectorTypeId(result.type), sampledImageId, coord.id,
resultId = m_module.opImageSampleImplicitLod(
resultTypeId, sampledImageId, coord.id,
imageOperands);
} break;
@ -3651,6 +3686,12 @@ namespace dxvk {
return;
}
DxbcRegisterValue result;
result.type = texelType;
result.id = imageOperands.sparse
? emitExtractSparseTexel(texelTypeId, resultId)
: resultId;
// Swizzle components using the texture swizzle
// and the destination operand's write mask
if (result.type.ccount != 1) {
@ -3659,6 +3700,9 @@ namespace dxvk {
}
emitRegisterStore(ins.dst[0], result);
if (imageOperands.sparse)
emitStoreSparseFeedback(ins.dst[1], resultId);
}