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

Web: Remove/rationalize a set of *_EXTENSIONS, using GLSLANG_WEB.

Browse source 
Focus was on the front end (not SPIR-V), minus the grammar.
Reduces #ifdef count by around 320 and makes the web build 270K smaller,
which is about 90% the  target size.

The grammar and scanner will be another step, as will the SPIR-V backend.
This makes heavy use of methods #ifdef'd to return false as a global way
of turning off code, relying on C++ DCE to do the rest.
This commit is contained in:
John Kessenich 2019-08-01 03:28:08 -06:00
parent e66dace97e
commit 7015bd658e
32 changed files with 2661 additions and 2712 deletions
Download patch file Download diff file Expand all files Collapse all files

81
glslang/MachineIndependent/linkValidate.cpp
View file

@ -78,11 +78,15 @@ void TIntermediate::warn(TInfoSink& infoSink, const char* message)
//
void TIntermediate::merge(TInfoSink& infoSink, TIntermediate& unit)
{
#ifndef GLSLANG_WEB
mergeCallGraphs(infoSink, unit);
mergeModes(infoSink, unit);
mergeTrees(infoSink, unit);
#endif
}
#ifndef GLSLANG_WEB
void TIntermediate::mergeCallGraphs(TInfoSink& infoSink, TIntermediate& unit)
{
if (unit.getNumEntryPoints() > 0) {
@ -142,18 +146,13 @@ void TIntermediate::mergeModes(TInfoSink& infoSink, TIntermediate& unit)
if (vertices == TQualifier::layoutNotSet)
vertices = unit.vertices;
else if (vertices != unit.vertices) {
if (language == EShLangGeometry
#ifdef NV_EXTENSIONS
|| language == EShLangMeshNV
#endif
)
if (language == EShLangGeometry || language == EShLangMeshNV)
error(infoSink, "Contradictory layout max_vertices values");
else if (language == EShLangTessControl)
error(infoSink, "Contradictory layout vertices values");
else
assert(0);
}
#ifdef NV_EXTENSIONS
if (primitives == TQualifier::layoutNotSet)
primitives = unit.primitives;
else if (primitives != unit.primitives) {
@ -162,7 +161,6 @@ void TIntermediate::mergeModes(TInfoSink& infoSink, TIntermediate& unit)
else
assert(0);
}
#endif
if (inputPrimitive == ElgNone)
inputPrimitive = unit.inputPrimitive;
@ -224,21 +222,16 @@ 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
}
MERGE_TRUE(multiStream);
#ifdef NV_EXTENSIONS
MERGE_TRUE(layoutOverrideCoverage);
MERGE_TRUE(geoPassthroughEXT);
#endif
for (unsigned int i = 0; i < unit.shiftBinding.size(); ++i) {
if (unit.shiftBinding[i] > 0)
@ -287,13 +280,8 @@ void TIntermediate::mergeTrees(TInfoSink& infoSink, TIntermediate& unit)
}
// Getting this far means we have two existing trees to merge...
#ifdef NV_EXTENSIONS
numShaderRecordNVBlocks += unit.numShaderRecordNVBlocks;
#endif
#ifdef NV_EXTENSIONS
numTaskNVBlocks += unit.numTaskNVBlocks;
#endif
// Get the top-level globals of each unit
TIntermSequence& globals = treeRoot->getAsAggregate()->getSequence();
@ -493,6 +481,7 @@ void TIntermediate::mergeImplicitArraySizes(TType& type, const TType& unitType)
for (int i = 0; i < (int)type.getStruct()->size(); ++i)
mergeImplicitArraySizes(*(*type.getStruct())[i].type, *(*unitType.getStruct())[i].type);
}
#endif // not GLSLANG_WEB
//
// Compare two global objects from two compilation units and see if they match
@ -547,7 +536,7 @@ void TIntermediate::mergeErrorCheck(TInfoSink& infoSink, const TIntermSymbol& sy
symbol.getQualifier().flat != unitSymbol.getQualifier().flat ||
symbol.getQualifier().sample != unitSymbol.getQualifier().sample ||
symbol.getQualifier().patch != unitSymbol.getQualifier().patch ||
symbol.getQualifier().nopersp != unitSymbol.getQualifier().nopersp) {
symbol.getQualifier().isNonPerspective() != unitSymbol.getQualifier().isNonPerspective()) {
error(infoSink, "Interpolation and auxiliary storage qualifiers must match:");
writeTypeComparison = true;
}
@ -615,7 +604,7 @@ void TIntermediate::finalCheck(TInfoSink& infoSink, bool keepUncalled)
warn(infoSink, "Entry point not found");
}
if (numPushConstants > 1)
if (getNumPushConstants() > 1)
error(infoSink, "Only one push_constant block is allowed per stage");
// recursion and missing body checking
@ -629,6 +618,7 @@ void TIntermediate::finalCheck(TInfoSink& infoSink, bool keepUncalled)
if (invocations == TQualifier::layoutNotSet)
invocations = 1;
#ifndef GLSLANG_WEB
if (inIoAccessed("gl_ClipDistance") && inIoAccessed("gl_ClipVertex"))
error(infoSink, "Can only use one of gl_ClipDistance or gl_ClipVertex (gl_ClipDistance is preferred)");
if (inIoAccessed("gl_CullDistance") && inIoAccessed("gl_ClipVertex"))
@ -642,12 +632,10 @@ 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
@ -668,16 +656,11 @@ 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)) {
@ -686,7 +669,6 @@ void TIntermediate::finalCheck(TInfoSink& infoSink, bool keepUncalled)
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)) {
@ -730,8 +712,6 @@ void TIntermediate::finalCheck(TInfoSink& infoSink, bool keepUncalled)
break;
case EShLangCompute:
break;
#ifdef NV_EXTENSIONS
case EShLangRayGenNV:
case EShLangIntersectNV:
case EShLangAnyHitNV:
@ -764,8 +744,6 @@ void TIntermediate::finalCheck(TInfoSink& infoSink, bool keepUncalled)
if (numTaskNVBlocks > 1)
error(infoSink, "Only one taskNV interface block is allowed per shader");
break;
#endif
default:
error(infoSink, "Unknown Stage.");
break;
@ -787,6 +765,7 @@ void TIntermediate::finalCheck(TInfoSink& infoSink, bool keepUncalled)
} finalLinkTraverser;
treeRoot->traverse(&finalLinkTraverser);
#endif
}
//
@ -1187,14 +1166,10 @@ int TIntermediate::computeTypeLocationSize(const TType& type, EShLanguage stage)
// TODO: perf: this can be flattened by using getCumulativeArraySize(), and a deref that discards all arrayness
// TODO: are there valid cases of having an unsized array with a location? If so, running this code too early.
TType elementType(type, 0);
if (type.isSizedArray()
#ifdef NV_EXTENSIONS
&& !type.getQualifier().isPerView()
#endif
)
if (type.isSizedArray() && !type.getQualifier().isPerView())
return type.getOuterArraySize() * computeTypeLocationSize(elementType, stage);
else {
#ifdef NV_EXTENSIONS
#ifndef GLSLANG_WEB
// unset perViewNV attributes for arrayed per-view outputs: "perviewNV vec4 v[MAX_VIEWS][3];"
elementType.getQualifier().perViewNV = false;
#endif
@ -1273,6 +1248,7 @@ int TIntermediate::computeTypeUniformLocationSize(const TType& type)
return 1;
}
#ifndef GLSLANG_WEB
// Accumulate xfb buffer ranges and check for collisions as the accumulation is done.
//
// Returns < 0 if no collision, >= 0 if collision and the value returned is a colliding value.
@ -1285,11 +1261,7 @@ 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);
@ -1309,15 +1281,10 @@ 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 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.
@ -1330,44 +1297,32 @@ 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;
}
@ -1375,14 +1330,12 @@ 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;
}
@ -1402,7 +1355,6 @@ 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;
@ -1412,11 +1364,8 @@ unsigned int TIntermediate::computeTypeXfbSize(const TType& type, bool& contains
contains32BitType = true;
return 4 * numComponents;
}
#else
} else
return 4 * numComponents;
#endif
}
#endif // not GLSLANG_WEB
const int baseAlignmentVec4Std140 = 16;
@ -1741,7 +1690,7 @@ int TIntermediate::getBlockSize(const TType& blockType)
int TIntermediate::computeBufferReferenceTypeSize(const TType& type)
{
assert(type.getBasicType() == EbtReference);
assert(type.isReference());
int size = getBlockSize(*type.getReferentType());
int align = type.getBufferReferenceAlignment();