BeeEngine/glslang-zig
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Add 8-bit/16-bit transform feedback support for future use

Browse source
This commit is contained in:
Rex Xu 2019-01-21 16:50:17 +08:00
parent dc2d5673a1
commit eaf31ab98c
8 changed files with 288 additions and 9 deletions
Download patch file Download diff file Expand all files Collapse all files

80
glslang/MachineIndependent/linkValidate.cpp
View file

@ -224,6 +224,12 @@ void TIntermediate::mergeModes(TInfoSink& infoSink, TIntermediate& unit)
xfbBuffers[b].implicitStride = std::max(xfbBuffers[b].implicitStride, unit.xfbBuffers[b].implicitStride);
if (unit.xfbBuffers[b].contains64BitType)
xfbBuffers[b].contains64BitType = true;
#ifdef AMD_EXTENSIONS
if (unit.xfbBuffers[b].contains32BitType)
xfbBuffers[b].contains32BitType = true;
if (unit.xfbBuffers[b].contains16BitType)
xfbBuffers[b].contains16BitType = true;
#endif
// TODO: 4.4 link: enhanced layouts: compare ranges
}
@ -636,6 +642,12 @@ void TIntermediate::finalCheck(TInfoSink& infoSink, bool keepUncalled)
for (size_t b = 0; b < xfbBuffers.size(); ++b) {
if (xfbBuffers[b].contains64BitType)
RoundToPow2(xfbBuffers[b].implicitStride, 8);
#ifdef AMD_EXTENSIONS
else if (xfbBuffers[b].contains32BitType)
RoundToPow2(xfbBuffers[b].implicitStride, 4);
else if (xfbBuffers[b].contains16BitType)
RoundToPow2(xfbBuffers[b].implicitStride, 2);
#endif
// "It is a compile-time or link-time error to have
// any xfb_offset that overflows xfb_stride, whether stated on declarations before or after the xfb_stride, or
@ -656,12 +668,25 @@ void TIntermediate::finalCheck(TInfoSink& infoSink, bool keepUncalled)
error(infoSink, "xfb_stride must be multiple of 8 for buffer holding a double or 64-bit integer:");
infoSink.info.prefix(EPrefixError);
infoSink.info << " xfb_buffer " << (unsigned int)b << ", xfb_stride " << xfbBuffers[b].stride << "\n";
#ifdef AMD_EXTENSIONS
} else if (xfbBuffers[b].contains32BitType && ! IsMultipleOfPow2(xfbBuffers[b].stride, 4)) {
#else
} else if (! IsMultipleOfPow2(xfbBuffers[b].stride, 4)) {
#endif
error(infoSink, "xfb_stride must be multiple of 4:");
infoSink.info.prefix(EPrefixError);
infoSink.info << " xfb_buffer " << (unsigned int)b << ", xfb_stride " << xfbBuffers[b].stride << "\n";
}
#ifdef AMD_EXTENSIONS
// "If the buffer is capturing any
// outputs with half-precision or 16-bit integer components, the stride must be a multiple of 2"
else if (xfbBuffers[b].contains16BitType && ! IsMultipleOfPow2(xfbBuffers[b].stride, 2)) {
error(infoSink, "xfb_stride must be multiple of 2 for buffer holding a half float or 16-bit integer:");
infoSink.info.prefix(EPrefixError);
infoSink.info << " xfb_buffer " << (unsigned int)b << ", xfb_stride " << xfbBuffers[b].stride << "\n";
}
#endif
// "The resulting stride (implicit or explicit), when divided by 4, must be less than or equal to the
// implementation-dependent constant gl_MaxTransformFeedbackInterleavedComponents."
if (xfbBuffers[b].stride > (unsigned int)(4 * resources.maxTransformFeedbackInterleavedComponents)) {
@ -1260,7 +1285,11 @@ int TIntermediate::addXfbBufferOffset(const TType& type)
TXfbBuffer& buffer = xfbBuffers[qualifier.layoutXfbBuffer];
// compute the range
#ifdef AMD_EXTENSIONS
unsigned int size = computeTypeXfbSize(type, buffer.contains64BitType, buffer.contains32BitType, buffer.contains16BitType);
#else
unsigned int size = computeTypeXfbSize(type, buffer.contains64BitType);
#endif
buffer.implicitStride = std::max(buffer.implicitStride, qualifier.layoutXfbOffset + size);
TRange range(qualifier.layoutXfbOffset, qualifier.layoutXfbOffset + size - 1);
@ -1279,9 +1308,16 @@ int TIntermediate::addXfbBufferOffset(const TType& type)
// Recursively figure out how many bytes of xfb buffer are used by the given type.
// Return the size of type, in bytes.
// Sets contains64BitType to true if the type contains a double.
// Sets contains64BitType to true if the type contains a 64-bit data type.
#ifdef AMD_EXTENSIONS
// Sets contains32BitType to true if the type contains a 32-bit data type.
// Sets contains16BitType to true if the type contains a 16-bit data type.
// N.B. Caller must set contains64BitType, contains32BitType, and contains16BitType to false before calling.
unsigned int TIntermediate::computeTypeXfbSize(const TType& type, bool& contains64BitType, bool& contains32BitType, bool& contains16BitType) const
#else
// N.B. Caller must set contains64BitType to false before calling.
unsigned int TIntermediate::computeTypeXfbSize(const TType& type, bool& contains64BitType) const
#endif
{
// "...if applied to an aggregate containing a double or 64-bit integer, the offset must also be a multiple of 8,
// and the space taken in the buffer will be a multiple of 8.
@ -1294,22 +1330,44 @@ unsigned int TIntermediate::computeTypeXfbSize(const TType& type, bool& contains
// TODO: perf: this can be flattened by using getCumulativeArraySize(), and a deref that discards all arrayness
assert(type.isSizedArray());
TType elementType(type, 0);
#ifdef AMD_EXTENSIONS
return type.getOuterArraySize() * computeTypeXfbSize(elementType, contains64BitType, contains16BitType, contains16BitType);
#else
return type.getOuterArraySize() * computeTypeXfbSize(elementType, contains64BitType);
#endif
}
if (type.isStruct()) {
unsigned int size = 0;
bool structContains64BitType = false;
#ifdef AMD_EXTENSIONS
bool structContains32BitType = false;
bool structContains16BitType = false;
#endif
for (int member = 0; member < (int)type.getStruct()->size(); ++member) {
TType memberType(type, member);
// "... if applied to
// an aggregate containing a double or 64-bit integer, the offset must also be a multiple of 8,
// and the space taken in the buffer will be a multiple of 8."
bool memberContains64BitType = false;
#ifdef AMD_EXTENSIONS
bool memberContains32BitType = false;
bool memberContains16BitType = false;
int memberSize = computeTypeXfbSize(memberType, memberContains64BitType, memberContains32BitType, memberContains16BitType);
#else
int memberSize = computeTypeXfbSize(memberType, memberContains64BitType);
#endif
if (memberContains64BitType) {
structContains64BitType = true;
RoundToPow2(size, 8);
#ifdef AMD_EXTENSIONS
} else if (memberContains32BitType) {
structContains32BitType = true;
RoundToPow2(size, 4);
} else if (memberContains16BitType) {
structContains16BitType = true;
RoundToPow2(size, 2);
#endif
}
size += memberSize;
}
@ -1317,6 +1375,14 @@ unsigned int TIntermediate::computeTypeXfbSize(const TType& type, bool& contains
if (structContains64BitType) {
contains64BitType = true;
RoundToPow2(size, 8);
#ifdef AMD_EXTENSIONS
} else if (structContains32BitType) {
contains32BitType = true;
RoundToPow2(size, 4);
} else if (structContains16BitType) {
contains16BitType = true;
RoundToPow2(size, 2);
#endif
}
return size;
}
@ -1336,8 +1402,20 @@ unsigned int TIntermediate::computeTypeXfbSize(const TType& type, bool& contains
if (type.getBasicType() == EbtDouble || type.getBasicType() == EbtInt64 || type.getBasicType() == EbtUint64) {
contains64BitType = true;
return 8 * numComponents;
#ifdef AMD_EXTENSIONS
} else if (type.getBasicType() == EbtFloat16 || type.getBasicType() == EbtInt16 || type.getBasicType() == EbtUint16) {
contains16BitType = true;
return 2 * numComponents;
} else if (type.getBasicType() == EbtInt8 || type.getBasicType() == EbtUint8)
return numComponents;
else {
contains32BitType = true;
return 4 * numComponents;
}
#else
} else
return 4 * numComponents;
#endif
}
const int baseAlignmentVec4Std140 = 16;