From ac1e188f3bf09eac9a3ab5e0ab7431a2c3db58d8 Mon Sep 17 00:00:00 2001
From: John Kessenich <cepheus@frii.com>
Date: Fri, 31 Jan 2014 02:40:19 +0000
Subject: [PATCH] GL_ARB_enhanced_layouts, part 6: Numerical side of uniform
offset and align semantics. Included - moving offset calculations for
std140/std430 from reflection to linkValidate.cpp - applying the
offset/align rules on top of std140/std430 - removing caching the
structure's number of components (and correcting that this is components, not
size)
git-svn-id: https://cvs.khronos.org/svn/repos/ogl/trunk/ecosystem/public/sdk/tools/glslang@25174 e7fa87d3-cd2b-0410-9028-fcbf551c1848
---
Test/440.frag | 51 +++++-
Test/baseResults/300layout.vert.out | 14 +-
Test/baseResults/440.frag.out | 18 +-
Test/baseResults/specExamples.vert.out | 2 +-
glslang/Include/Types.h | 30 ++--
glslang/Include/revision.h | 4 +-
glslang/MachineIndependent/Constant.cpp | 70 ++++----
glslang/MachineIndependent/Intermediate.cpp | 2 +-
glslang/MachineIndependent/ParseHelper.cpp | 58 ++++++-
glslang/MachineIndependent/SymbolTable.cpp | 15 --
glslang/MachineIndependent/intermOut.cpp | 2 +-
glslang/MachineIndependent/linkValidate.cpp | 159 +++++++++++++++++-
.../MachineIndependent/localintermediate.h | 2 +
glslang/MachineIndependent/parseConst.cpp | 12 +-
glslang/MachineIndependent/reflection.cpp | 147 +++-------------
15 files changed, 367 insertions(+), 219 deletions(-)
diff --git a/Test/440.frag b/Test/440.frag
index 22d46f56b..a32706c04 100644
--- a/Test/440.frag
+++ b/Test/440.frag
@@ -79,13 +79,58 @@ uniform ubl11 {
layout(std140) uniform block {
vec4 a; // a takes offsets 0-15
- layout(offset = 20) vec3 b; // b takes offsets 32-43
+ layout(offset = 32) vec3 b; // b takes offsets 32-43
layout(offset = 40) vec2 c; // ERROR, lies within previous member
+ layout(align = 6) double g; // ERROR, 6 is not a power of 2
+ layout(offset=68) double h; // ERROR, offset not aligned
+} specExampleErrors;
+
+layout(std140) uniform block2 {
+ vec4 a; // a takes offsets 0-15
+ layout(offset = 32) vec3 b; // b takes offsets 32-43
layout(offset = 48) vec2 d; // d takes offsets 48-55
layout(align = 16) float e; // e takes offsets 64-67
layout(align = 2) double f; // f takes offsets 72-79
- layout(align = 6) double g; // ERROR, 6 is not a power of 2
layout(offset = 80) float h; // h takes offsets 80-83
layout(align = 64) dvec3 i; // i takes offsets 128-151
- layout(offset = 153, align = 8) float j; // j takes offsets 160-163
+ layout(offset = 164, align = 8) float j; // j takes offsets 168-171
} specExample;
+
+layout(std430) uniform block430 {
+ vec4 a; // a takes offsets 0-15
+ layout(offset = 32) vec3 b; // b takes offsets 32-43
+ layout(offset = 40) vec2 c; // ERROR, lies within previous member
+ layout(align = 6) double g; // ERROR, 6 is not a power of 2
+ layout(offset=68) double h; // ERROR, offset not aligned
+} specExampleErrors430;
+
+layout(std430) uniform block2430 {
+ vec4 a; // a takes offsets 0-15
+ layout(offset = 32) vec3 b; // b takes offsets 32-43
+ layout(offset = 48) vec2 d; // d takes offsets 48-55
+ layout(align = 16) float e; // e takes offsets 64-67
+ layout(align = 2) double f; // f takes offsets 72-79
+ layout(offset = 80) float h; // h takes offsets 80-83
+ layout(align = 64) dvec3 i; // i takes offsets 128-151
+ layout(offset = 164, align = 8) float j; // j takes offsets 168-171
+} specExample430;
+
+layout(std430, align = 128) uniform block24300 {
+ vec4 a;
+ vec3 b;
+ vec2 d;
+ float e;
+ double f;
+ float h;
+ dvec3 i;
+} specExample4300;
+
+layout(std430, align = 128) uniform block24301 {
+ vec4 a;
+ vec3 b;
+ vec2 d;
+ layout(offset=388) float e;
+ layout(align=8) double f;
+ float h;
+ dvec3 i;
+} specExample4301;
diff --git a/Test/baseResults/300layout.vert.out b/Test/baseResults/300layout.vert.out
index 0ce4b3c79..4d2c3a5b9 100644
--- a/Test/baseResults/300layout.vert.out
+++ b/Test/baseResults/300layout.vert.out
@@ -32,12 +32,12 @@ ERROR: node is still EOpNull!
0:46 add (highp 4X4 matrix of float)
0:46 add (highp 4X4 matrix of float)
0:46 add (highp 4X4 matrix of float)
-0:46 M1: direct index for structure (layout(row_major std140 ) uniform highp 4X4 matrix of float)
-0:46 'tblock' (layout(row_major std140 ) uniform block{layout(row_major std140 ) uniform highp 4X4 matrix of float M1, layout(column_major std140 ) uniform highp 4X4 matrix of float M2, layout(row_major std140 ) uniform highp 3X3 matrix of float N1, layout(row_major std140 ) centroid uniform highp float badf, layout(row_major std140 ) uniform highp float badg, layout(row_major std140 ) uniform highp float bad1, layout(row_major shared ) uniform highp float bad2, layout(row_major packed ) uniform highp float bad3})
+0:46 M1: direct index for structure (layout(row_major std140 offset=0 ) uniform highp 4X4 matrix of float)
+0:46 'tblock' (layout(row_major std140 ) uniform block{layout(row_major std140 offset=0 ) uniform highp 4X4 matrix of float M1, layout(column_major std140 offset=64 ) uniform highp 4X4 matrix of float M2, layout(row_major std140 offset=128 ) uniform highp 3X3 matrix of float N1, layout(row_major std140 offset=176 ) centroid uniform highp float badf, layout(row_major std140 offset=180 ) uniform highp float badg, layout(row_major std140 offset=184 ) uniform highp float bad1, layout(row_major shared offset=188 ) uniform highp float bad2, layout(row_major packed offset=192 ) uniform highp float bad3})
0:46 Constant:
0:46 0 (const int)
-0:46 M2: direct index for structure (layout(column_major std140 ) uniform highp 4X4 matrix of float)
-0:46 'tblock' (layout(row_major std140 ) uniform block{layout(row_major std140 ) uniform highp 4X4 matrix of float M1, layout(column_major std140 ) uniform highp 4X4 matrix of float M2, layout(row_major std140 ) uniform highp 3X3 matrix of float N1, layout(row_major std140 ) centroid uniform highp float badf, layout(row_major std140 ) uniform highp float badg, layout(row_major std140 ) uniform highp float bad1, layout(row_major shared ) uniform highp float bad2, layout(row_major packed ) uniform highp float bad3})
+0:46 M2: direct index for structure (layout(column_major std140 offset=64 ) uniform highp 4X4 matrix of float)
+0:46 'tblock' (layout(row_major std140 ) uniform block{layout(row_major std140 offset=0 ) uniform highp 4X4 matrix of float M1, layout(column_major std140 offset=64 ) uniform highp 4X4 matrix of float M2, layout(row_major std140 offset=128 ) uniform highp 3X3 matrix of float N1, layout(row_major std140 offset=176 ) centroid uniform highp float badf, layout(row_major std140 offset=180 ) uniform highp float badg, layout(row_major std140 offset=184 ) uniform highp float bad1, layout(row_major shared offset=188 ) uniform highp float bad2, layout(row_major packed offset=192 ) uniform highp float bad3})
0:46 Constant:
0:46 1 (const int)
0:46 M4: direct index for structure (layout(row_major shared ) uniform highp 4X4 matrix of float)
@@ -56,8 +56,8 @@ ERROR: node is still EOpNull!
0:47 'color' (smooth out highp 3-component vector of float)
0:47 vector-times-matrix (highp 3-component vector of float)
0:47 'c' (layout(location=7 ) in highp 3-component vector of float)
-0:47 N1: direct index for structure (layout(row_major std140 ) uniform highp 3X3 matrix of float)
-0:47 'tblock' (layout(row_major std140 ) uniform block{layout(row_major std140 ) uniform highp 4X4 matrix of float M1, layout(column_major std140 ) uniform highp 4X4 matrix of float M2, layout(row_major std140 ) uniform highp 3X3 matrix of float N1, layout(row_major std140 ) centroid uniform highp float badf, layout(row_major std140 ) uniform highp float badg, layout(row_major std140 ) uniform highp float bad1, layout(row_major shared ) uniform highp float bad2, layout(row_major packed ) uniform highp float bad3})
+0:47 N1: direct index for structure (layout(row_major std140 offset=128 ) uniform highp 3X3 matrix of float)
+0:47 'tblock' (layout(row_major std140 ) uniform block{layout(row_major std140 offset=0 ) uniform highp 4X4 matrix of float M1, layout(column_major std140 offset=64 ) uniform highp 4X4 matrix of float M2, layout(row_major std140 offset=128 ) uniform highp 3X3 matrix of float N1, layout(row_major std140 offset=176 ) centroid uniform highp float badf, layout(row_major std140 offset=180 ) uniform highp float badg, layout(row_major std140 offset=184 ) uniform highp float bad1, layout(row_major shared offset=188 ) uniform highp float bad2, layout(row_major packed offset=192 ) uniform highp float bad3})
0:47 Constant:
0:47 2 (const int)
0:? Linker Objects
@@ -68,7 +68,7 @@ ERROR: node is still EOpNull!
0:? 'pos' (smooth out highp 4-component vector of float)
0:? 'color' (smooth out highp 3-component vector of float)
0:? 'badm4' (layout(column_major shared ) uniform highp 4X4 matrix of float)
-0:? 'tblock' (layout(row_major std140 ) uniform block{layout(row_major std140 ) uniform highp 4X4 matrix of float M1, layout(column_major std140 ) uniform highp 4X4 matrix of float M2, layout(row_major std140 ) uniform highp 3X3 matrix of float N1, layout(row_major std140 ) centroid uniform highp float badf, layout(row_major std140 ) uniform highp float badg, layout(row_major std140 ) uniform highp float bad1, layout(row_major shared ) uniform highp float bad2, layout(row_major packed ) uniform highp float bad3})
+0:? 'tblock' (layout(row_major std140 ) uniform block{layout(row_major std140 offset=0 ) uniform highp 4X4 matrix of float M1, layout(column_major std140 offset=64 ) uniform highp 4X4 matrix of float M2, layout(row_major std140 offset=128 ) uniform highp 3X3 matrix of float N1, layout(row_major std140 offset=176 ) centroid uniform highp float badf, layout(row_major std140 offset=180 ) uniform highp float badg, layout(row_major std140 offset=184 ) uniform highp float bad1, layout(row_major shared offset=188 ) uniform highp float bad2, layout(row_major packed offset=192 ) uniform highp float bad3})
0:? '__anon__0' (layout(row_major shared ) uniform block{layout(row_major shared ) uniform bool b, layout(row_major shared ) uniform highp 4X4 matrix of float t2m})
0:? '__anon__2' (out block{out highp float f})
0:? 'badoutA' (layout(location=10 ) smooth out highp 4-component vector of float)
diff --git a/Test/baseResults/440.frag.out b/Test/baseResults/440.frag.out
index b81870cb1..c020f070d 100644
--- a/Test/baseResults/440.frag.out
+++ b/Test/baseResults/440.frag.out
@@ -39,8 +39,13 @@ ERROR: 0:58: 'align' : can only be used with std140 or std430 layout packing
ERROR: 0:63: 'align' : can only be used with std140 or std430 layout packing
ERROR: 0:62: 'layout' : offset/align can only be used on a uniform or buffer
ERROR: 0:63: 'layout' : offset/align can only be used on a uniform or buffer
-ERROR: 0:87: 'align' : must be a power of 2
-ERROR: 40 compilation errors. No code generated.
+ERROR: 0:84: 'align' : must be a power of 2
+ERROR: 0:83: 'offset' : cannot lie in previous members
+ERROR: 0:85: 'offset' : must be a multiple of the member's alignment
+ERROR: 0:103: 'align' : must be a power of 2
+ERROR: 0:102: 'offset' : cannot lie in previous members
+ERROR: 0:104: 'offset' : must be a multiple of the member's alignment
+ERROR: 45 compilation errors. No code generated.
ERROR: node is still EOpNull!
@@ -56,7 +61,7 @@ ERROR: node is still EOpNull!
0:? 'inst1' (layout(column_major shared offset=12 ) uniform block{layout(column_major shared ) uniform int a})
0:? 'inst2' (layout(offset=12 ) in block{in int a})
0:? 'inst3' (layout(offset=12 ) out block{out int a})
-0:? 'inst4' (layout(column_major std140 align=16 ) uniform block{layout(column_major std140 align=16 ) uniform int a})
+0:? 'inst4' (layout(column_major std140 align=16 ) uniform block{layout(column_major std140 offset=0 align=16 ) uniform int a})
0:? 'inst8' (layout(column_major shared align=16 ) uniform block{layout(column_major shared ) uniform int a})
0:? 'inst5' (layout(align=16 ) in block{in int a})
0:? 'inst6' (layout(align=16 ) out block{out int a})
@@ -71,7 +76,12 @@ ERROR: node is still EOpNull!
0:? 'inst10' (in block{layout(offset=12 ) in float f, layout(align=4 ) in float g})
0:? 'inst9' (layout(column_major std430 align=32 ) uniform block{layout(column_major std430 align=32 ) uniform float e, layout(column_major std430 offset=12 align=4 ) uniform float f, layout(column_major std430 offset=20 align=32 ) uniform float g, layout(column_major std430 align=32 ) uniform float h})
0:? 'inst11' (layout(column_major std430 ) uniform block{layout(column_major std430 offset=12 align=4 ) uniform float f, layout(column_major std430 ) uniform float g})
-0:? 'specExample' (layout(column_major std140 ) uniform block{layout(column_major std140 ) uniform 4-component vector of float a, layout(column_major std140 offset=20 ) uniform 3-component vector of float b, layout(column_major std140 offset=40 ) uniform 2-component vector of float c, layout(column_major std140 offset=48 ) uniform 2-component vector of float d, layout(column_major std140 align=16 ) uniform float e, layout(column_major std140 align=2 ) uniform double f, layout(column_major std140 ) uniform double g, layout(column_major std140 offset=80 ) uniform float h, layout(column_major std140 align=64 ) uniform 3-component vector of double i, layout(column_major std140 offset=153 align=8 ) uniform float j})
+0:? 'specExampleErrors' (layout(column_major std140 ) uniform block{layout(column_major std140 offset=0 ) uniform 4-component vector of float a, layout(column_major std140 offset=32 ) uniform 3-component vector of float b, layout(column_major std140 offset=48 ) uniform 2-component vector of float c, layout(column_major std140 offset=56 ) uniform double g, layout(column_major std140 offset=72 ) uniform double h})
+0:? 'specExample' (layout(column_major std140 ) uniform block{layout(column_major std140 offset=0 ) uniform 4-component vector of float a, layout(column_major std140 offset=32 ) uniform 3-component vector of float b, layout(column_major std140 offset=48 ) uniform 2-component vector of float d, layout(column_major std140 offset=64 align=16 ) uniform float e, layout(column_major std140 offset=72 align=2 ) uniform double f, layout(column_major std140 offset=80 ) uniform float h, layout(column_major std140 offset=128 align=64 ) uniform 3-component vector of double i, layout(column_major std140 offset=168 align=8 ) uniform float j})
+0:? 'specExampleErrors430' (layout(column_major std430 ) uniform block{layout(column_major std430 offset=0 ) uniform 4-component vector of float a, layout(column_major std430 offset=32 ) uniform 3-component vector of float b, layout(column_major std430 offset=48 ) uniform 2-component vector of float c, layout(column_major std430 offset=56 ) uniform double g, layout(column_major std430 offset=72 ) uniform double h})
+0:? 'specExample430' (layout(column_major std430 ) uniform block{layout(column_major std430 offset=0 ) uniform 4-component vector of float a, layout(column_major std430 offset=32 ) uniform 3-component vector of float b, layout(column_major std430 offset=48 ) uniform 2-component vector of float d, layout(column_major std430 offset=64 align=16 ) uniform float e, layout(column_major std430 offset=72 align=2 ) uniform double f, layout(column_major std430 offset=80 ) uniform float h, layout(column_major std430 offset=128 align=64 ) uniform 3-component vector of double i, layout(column_major std430 offset=168 align=8 ) uniform float j})
+0:? 'specExample4300' (layout(column_major std430 align=128 ) uniform block{layout(column_major std430 offset=0 align=128 ) uniform 4-component vector of float a, layout(column_major std430 offset=128 align=128 ) uniform 3-component vector of float b, layout(column_major std430 offset=256 align=128 ) uniform 2-component vector of float d, layout(column_major std430 offset=384 align=128 ) uniform float e, layout(column_major std430 offset=512 align=128 ) uniform double f, layout(column_major std430 offset=640 align=128 ) uniform float h, layout(column_major std430 offset=768 align=128 ) uniform 3-component vector of double i})
+0:? 'specExample4301' (layout(column_major std430 align=128 ) uniform block{layout(column_major std430 offset=0 align=128 ) uniform 4-component vector of float a, layout(column_major std430 offset=128 align=128 ) uniform 3-component vector of float b, layout(column_major std430 offset=256 align=128 ) uniform 2-component vector of float d, layout(column_major std430 offset=512 align=128 ) uniform float e, layout(column_major std430 offset=520 align=8 ) uniform double f, layout(column_major std430 offset=640 align=128 ) uniform float h, layout(column_major std430 offset=768 align=128 ) uniform 3-component vector of double i})
Linked fragment stage:
diff --git a/Test/baseResults/specExamples.vert.out b/Test/baseResults/specExamples.vert.out
index b69695a9c..0a7c408d9 100644
--- a/Test/baseResults/specExamples.vert.out
+++ b/Test/baseResults/specExamples.vert.out
@@ -289,7 +289,7 @@ ERROR: node is still EOpNull!
0:? 'var5' (smooth out 4-component vector of float)
0:? '__anon__2' (out block{out 4-component vector of float var6})
0:? 'var7' (smooth out 4-component vector of float)
-0:? '__anon__3' (layout(row_major std140 ) uniform block{layout(row_major std140 ) uniform 4X4 matrix of float M1, layout(column_major std140 ) uniform 4X4 matrix of float M2, layout(row_major std140 ) uniform 3X3 matrix of float N1})
+0:? '__anon__3' (layout(row_major std140 ) uniform block{layout(row_major std140 offset=0 ) uniform 4X4 matrix of float M1, layout(column_major std140 offset=64 ) uniform 4X4 matrix of float M2, layout(row_major std140 offset=128 ) uniform 3X3 matrix of float N1})
0:? '__anon__4' (layout(column_major shared ) uniform block{layout(column_major shared ) uniform 4X4 matrix of float M13, layout(row_major shared ) uniform 4X4 matrix of float m14, layout(column_major shared ) uniform 3X3 matrix of float N12})
0:? 's17' (layout(binding=3 ) uniform sampler2D)
0:? 'a2' (layout(binding=2 offset=4 ) uniform int)
diff --git a/glslang/Include/Types.h b/glslang/Include/Types.h
index a7435d00c..3a712cbbc 100644
--- a/glslang/Include/Types.h
+++ b/glslang/Include/Types.h
@@ -42,7 +42,7 @@
namespace glslang {
-const int GlslangMaxTypeLength = 200;
+const int GlslangMaxTypeLength = 200; // TODO: need to print block/struct one member per line, so this can stay bounded
//
// Details within a sampler type
@@ -643,7 +643,7 @@ public:
// for "empty" type (no args) or simple scalar/vector/matrix
explicit TType(TBasicType t = EbtVoid, TStorageQualifier q = EvqTemporary, int vs = 1, int mc = 0, int mr = 0) :
basicType(t), vectorSize(vs), matrixCols(mc), matrixRows(mr), arraySizes(0),
- structure(0), structureSize(0), fieldName(0), typeName(0)
+ structure(0), fieldName(0), typeName(0)
{
sampler.clear();
qualifier.clear();
@@ -652,7 +652,7 @@ public:
// for explicit precision qualifier
TType(TBasicType t, TStorageQualifier q, TPrecisionQualifier p, int vs = 1, int mc = 0, int mr = 0) :
basicType(t), vectorSize(vs), matrixCols(mc), matrixRows(mr), arraySizes(0),
- structure(0), structureSize(0), fieldName(0), typeName(0)
+ structure(0), fieldName(0), typeName(0)
{
sampler.clear();
qualifier.clear();
@@ -663,7 +663,7 @@ public:
// for turning a TPublicType into a TType
explicit TType(const TPublicType& p) :
basicType(p.basicType), vectorSize(p.vectorSize), matrixCols(p.matrixCols), matrixRows(p.matrixRows), arraySizes(p.arraySizes),
- structure(0), structureSize(0), fieldName(0), typeName(0)
+ structure(0), fieldName(0), typeName(0)
{
if (basicType == EbtSampler)
sampler = p.sampler;
@@ -723,7 +723,6 @@ public:
matrixRows = copyOf.matrixRows;
arraySizes = copyOf.arraySizes;
structure = copyOf.structure;
- structureSize = copyOf.structureSize;
fieldName = copyOf.fieldName;
typeName = copyOf.typeName;
}
@@ -1015,24 +1014,25 @@ public:
TTypeList* getStruct() { return structure; }
TTypeList* getStruct() const { return structure; }
- int getObjectSize() const
+ int computeNumComponents() const
{
- int totalSize;
+ int components = 0;
- if (getBasicType() == EbtStruct || getBasicType() == EbtBlock)
- totalSize = getStructSize();
- else if (matrixCols)
- totalSize = matrixCols * matrixRows;
+ if (getBasicType() == EbtStruct || getBasicType() == EbtBlock) {
+ for (TTypeList::iterator tl = getStruct()->begin(); tl != getStruct()->end(); tl++)
+ components += ((*tl).type)->computeNumComponents();
+ } else if (matrixCols)
+ components = matrixCols * matrixRows;
else
- totalSize = vectorSize;
+ components = vectorSize;
if (isArray()) {
// this function can only be used in paths that don't allow unsized arrays
assert(getArraySize() > 0);
- totalSize *= getArraySize();
+ components *= getArraySize();
}
- return totalSize;
+ return components;
}
// append this type's mangled name to the passed in 'name'
@@ -1117,7 +1117,6 @@ protected:
TType& operator=(const TType& type);
void buildMangledName(TString&);
- int getStructSize() const;
TBasicType basicType : 8;
int vectorSize : 4;
@@ -1129,7 +1128,6 @@ protected:
TArraySizes* arraySizes;
TTypeList* structure; // 0 unless this is a struct
- mutable int structureSize; // a cache, updated on first access
TString *fieldName; // for structure field names
TString *typeName; // for structure type name
};
diff --git a/glslang/Include/revision.h b/glslang/Include/revision.h
index 1a2b8fe55..a8b56f3c3 100644
--- a/glslang/Include/revision.h
+++ b/glslang/Include/revision.h
@@ -9,5 +9,5 @@
// source have to figure out how to create revision.h just to get a build
// going. However, if it is not updated, it can be a version behind.
-#define GLSLANG_REVISION "25043"
-#define GLSLANG_DATE "2014/01/27 13:02:12"
+#define GLSLANG_REVISION "25092"
+#define GLSLANG_DATE "2014/01/28 14:13:59"
diff --git a/glslang/MachineIndependent/Constant.cpp b/glslang/MachineIndependent/Constant.cpp
index 801cf1d10..4648b559e 100644
--- a/glslang/MachineIndependent/Constant.cpp
+++ b/glslang/MachineIndependent/Constant.cpp
@@ -101,50 +101,52 @@ TIntermTyped* TIntermConstantUnion::fold(TOperator op, const TIntermTyped* const
TConstUnionArray rightUnionArray = node->getConstArray();
// Figure out the size of the result
- int objectSize;
+ int newComps;
+ int constComps;
switch(op) {
case EOpMatrixTimesMatrix:
- objectSize = getMatrixRows() * node->getMatrixCols();
+ newComps = getMatrixRows() * node->getMatrixCols();
break;
case EOpMatrixTimesVector:
- objectSize = getMatrixRows();
+ newComps = getMatrixRows();
break;
case EOpVectorTimesMatrix:
- objectSize = node->getMatrixCols();
+ newComps = node->getMatrixCols();
break;
default:
- objectSize = getType().getObjectSize();
- if (constantNode->getType().getObjectSize() == 1 && getType().getObjectSize() > 1) {
+ newComps = getType().computeNumComponents();
+ constComps = constantNode->getType().computeNumComponents();
+ if (constComps == 1 && newComps > 1) {
// for a case like vec4 f = vec4(2,3,4,5) + 1.2;
- TConstUnionArray smearedArray(objectSize, node->getConstArray()[0]);
+ TConstUnionArray smearedArray(newComps, node->getConstArray()[0]);
rightUnionArray = smearedArray;
- } else if (constantNode->getType().getObjectSize() > 1 && getType().getObjectSize() == 1) {
+ } else if (constComps > 1 && newComps == 1) {
// for a case like vec4 f = 1.2 + vec4(2,3,4,5);
- objectSize = constantNode->getType().getObjectSize();
+ newComps = constComps;
rightUnionArray = node->getConstArray();
- TConstUnionArray smearedArray(objectSize, getConstArray()[0]);
+ TConstUnionArray smearedArray(newComps, getConstArray()[0]);
unionArray = smearedArray;
returnType.shallowCopy(node->getType());
}
break;
}
- TConstUnionArray newConstArray(objectSize);
+ TConstUnionArray newConstArray(newComps);
switch(op) {
case EOpAdd:
- for (int i = 0; i < objectSize; i++)
+ for (int i = 0; i < newComps; i++)
newConstArray[i] = unionArray[i] + rightUnionArray[i];
break;
case EOpSub:
- for (int i = 0; i < objectSize; i++)
+ for (int i = 0; i < newComps; i++)
newConstArray[i] = unionArray[i] - rightUnionArray[i];
break;
case EOpMul:
case EOpVectorTimesScalar:
case EOpMatrixTimesScalar:
- for (int i = 0; i < objectSize; i++)
+ for (int i = 0; i < newComps; i++)
newConstArray[i] = unionArray[i] * rightUnionArray[i];
break;
case EOpMatrixTimesMatrix:
@@ -159,7 +161,7 @@ TIntermTyped* TIntermConstantUnion::fold(TOperator op, const TIntermTyped* const
returnType.shallowCopy(TType(getType().getBasicType(), EvqConst, 0, getMatrixRows(), node->getMatrixCols()));
break;
case EOpDiv:
- for (int i = 0; i < objectSize; i++) {
+ for (int i = 0; i < newComps; i++) {
switch (getType().getBasicType()) {
case EbtDouble:
case EbtFloat:
@@ -211,7 +213,7 @@ TIntermTyped* TIntermConstantUnion::fold(TOperator op, const TIntermTyped* const
break;
case EOpMod:
- for (int i = 0; i < objectSize; i++) {
+ for (int i = 0; i < newComps; i++) {
if (rightUnionArray[i] == 0)
newConstArray[i] = unionArray[i];
else
@@ -220,40 +222,40 @@ TIntermTyped* TIntermConstantUnion::fold(TOperator op, const TIntermTyped* const
break;
case EOpRightShift:
- for (int i = 0; i < objectSize; i++)
+ for (int i = 0; i < newComps; i++)
newConstArray[i] = unionArray[i] >> rightUnionArray[i];
break;
case EOpLeftShift:
- for (int i = 0; i < objectSize; i++)
+ for (int i = 0; i < newComps; i++)
newConstArray[i] = unionArray[i] << rightUnionArray[i];
break;
case EOpAnd:
- for (int i = 0; i < objectSize; i++)
+ for (int i = 0; i < newComps; i++)
newConstArray[i] = unionArray[i] & rightUnionArray[i];
break;
case EOpInclusiveOr:
- for (int i = 0; i < objectSize; i++)
+ for (int i = 0; i < newComps; i++)
newConstArray[i] = unionArray[i] | rightUnionArray[i];
break;
case EOpExclusiveOr:
- for (int i = 0; i < objectSize; i++)
+ for (int i = 0; i < newComps; i++)
newConstArray[i] = unionArray[i] ^ rightUnionArray[i];
break;
case EOpLogicalAnd: // this code is written for possible future use, will not get executed currently
- for (int i = 0; i < objectSize; i++)
+ for (int i = 0; i < newComps; i++)
newConstArray[i] = unionArray[i] && rightUnionArray[i];
break;
case EOpLogicalOr: // this code is written for possible future use, will not get executed currently
- for (int i = 0; i < objectSize; i++)
+ for (int i = 0; i < newComps; i++)
newConstArray[i] = unionArray[i] || rightUnionArray[i];
break;
case EOpLogicalXor:
- for (int i = 0; i < objectSize; i++) {
+ for (int i = 0; i < newComps; i++) {
switch (getType().getBasicType()) {
case EbtBool: newConstArray[i].setBConst((unionArray[i] == rightUnionArray[i]) ? false : true); break;
default: assert(false && "Default missing");
@@ -309,6 +311,7 @@ TIntermTyped* TIntermConstantUnion::fold(TOperator op, const TType& returnType)
int resultSize;
bool componentWise = true;
+ int objectSize = getType().computeNumComponents();
switch (op) {
case EOpDeterminant:
case EOpAny:
@@ -339,18 +342,17 @@ TIntermTyped* TIntermConstantUnion::fold(TOperator op, const TType& returnType)
case EOpNormalize:
componentWise = false;
- resultSize = getType().getObjectSize();
+ resultSize = objectSize;
break;
default:
- resultSize = getType().getObjectSize();
+ resultSize = objectSize;
break;
}
// Set up for processing
TConstUnionArray newConstArray(resultSize);
const TConstUnionArray& unionArray = getConstArray();
- int objectSize = getType().getObjectSize();
// Process non-component-wise operations
switch (op) {
@@ -593,13 +595,13 @@ TIntermTyped* TIntermediate::fold(TIntermAggregate* aggrNode)
case EOpVectorEqual:
case EOpVectorNotEqual:
componentwise = true;
- objectSize = children[0]->getAsConstantUnion()->getType().getObjectSize();
+ objectSize = children[0]->getAsConstantUnion()->getType().computeNumComponents();
break;
case EOpCross:
case EOpReflect:
case EOpRefract:
case EOpFaceForward:
- objectSize = children[0]->getAsConstantUnion()->getType().getObjectSize();
+ objectSize = children[0]->getAsConstantUnion()->getType().computeNumComponents();
break;
case EOpDistance:
case EOpDot:
@@ -726,7 +728,7 @@ TIntermTyped* TIntermediate::fold(TIntermAggregate* aggrNode)
} else {
// Non-componentwise...
- int numComps = children[0]->getAsConstantUnion()->getType().getObjectSize();
+ int numComps = children[0]->getAsConstantUnion()->getType().computeNumComponents();
double dot;
switch (aggrNode->getOp()) {
@@ -788,7 +790,7 @@ TIntermTyped* TIntermediate::fold(TIntermAggregate* aggrNode)
case EOpOuterProduct:
{
int numRows = numComps;
- int numCols = children[1]->getAsConstantUnion()->getType().getObjectSize();
+ int numCols = children[1]->getAsConstantUnion()->getType().computeNumComponents();
for (int row = 0; row < numRows; ++row)
for (int col = 0; col < numCols; ++col)
newConstArray[col * numRows + row] = childConstUnions[0][row] * childConstUnions[1][col];
@@ -828,7 +830,7 @@ TIntermTyped* TIntermediate::foldConstructor(TIntermAggregate* aggrNode)
{
bool error = false;
- TConstUnionArray unionArray(aggrNode->getType().getObjectSize());
+ TConstUnionArray unionArray(aggrNode->getType().computeNumComponents());
if (aggrNode->getSequence().size() == 1)
error = parseConstTree(aggrNode, unionArray, aggrNode->getOp(), aggrNode->getType(), true);
else
@@ -850,13 +852,13 @@ TIntermTyped* TIntermediate::foldDereference(TIntermTyped* node, int index, TSou
TType dereferencedType(node->getType(), index);
dereferencedType.getQualifier().storage = EvqConst;
TIntermTyped* result = 0;
- int size = dereferencedType.getObjectSize();
+ int size = dereferencedType.computeNumComponents();
int start;
if (node->isStruct()) {
start = 0;
for (int i = 0; i < index; ++i)
- start += (*node->getType().getStruct())[i].type->getObjectSize();
+ start += (*node->getType().getStruct())[i].type->computeNumComponents();
} else
start = size * index;
diff --git a/glslang/MachineIndependent/Intermediate.cpp b/glslang/MachineIndependent/Intermediate.cpp
index 952d103e7..80c98f1be 100644
--- a/glslang/MachineIndependent/Intermediate.cpp
+++ b/glslang/MachineIndependent/Intermediate.cpp
@@ -1415,7 +1415,7 @@ void TIntermTyped::propagatePrecision(TPrecisionQualifier newPrecision)
TIntermTyped* TIntermediate::promoteConstantUnion(TBasicType promoteTo, TIntermConstantUnion* node)
{
const TConstUnionArray& rightUnionArray = node->getConstArray();
- int size = node->getType().getObjectSize();
+ int size = node->getType().computeNumComponents();
TConstUnionArray leftUnionArray(size);
diff --git a/glslang/MachineIndependent/ParseHelper.cpp b/glslang/MachineIndependent/ParseHelper.cpp
index 34699c543..91ef2dd2a 100644
--- a/glslang/MachineIndependent/ParseHelper.cpp
+++ b/glslang/MachineIndependent/ParseHelper.cpp
@@ -1602,13 +1602,13 @@ bool TParseContext::constructorError(TSourceLoc loc, TIntermNode* node, TFunctio
bool matrixInMatrix = false;
bool arrayArg = false;
for (int i = 0; i < function.getParamCount(); ++i) {
- size += function[i].type->getObjectSize();
+ size += function[i].type->computeNumComponents();
if (constructingMatrix && function[i].type->isMatrix())
matrixInMatrix = true;
if (full)
overFull = true;
- if (op != EOpConstructStruct && ! type.isArray() && size >= type.getObjectSize())
+ if (op != EOpConstructStruct && ! type.isArray() && size >= type.computeNumComponents())
full = true;
if (function[i].type->getQualifier().storage != EvqConst)
constType = false;
@@ -1649,8 +1649,8 @@ bool TParseContext::constructorError(TSourceLoc loc, TIntermNode* node, TFunctio
return true;
}
- if ((op != EOpConstructStruct && size != 1 && size < type.getObjectSize()) ||
- (op == EOpConstructStruct && size < type.getObjectSize())) {
+ if ((op != EOpConstructStruct && size != 1 && size < type.computeNumComponents()) ||
+ (op == EOpConstructStruct && size < type.computeNumComponents())) {
error(loc, "not enough data provided for construction", "constructor", "");
return true;
}
@@ -4097,10 +4097,58 @@ void TParseContext::fixBlockXfbOffsets(TSourceLoc loc, TQualifier& qualifier, TT
qualifier.layoutXfbOffset = TQualifier::layoutXfbOffsetEnd;
}
+// Calculate and save the offset of each block member, using the recursively
+// defined block offset rules and the user-provided offset and align.
+//
+// Also, compute and save the total size of the block. For the block's size, arrayness
+// is not taken into account, as each element is backed by a separate buffer.
+//
void TParseContext::fixBlockUniformOffsets(TSourceLoc loc, TQualifier& qualifier, TTypeList& typeList)
{
- if (qualifier.storage != EvqUniform || qualifier.storage != EvqBuffer)
+ if (qualifier.storage != EvqUniform && qualifier.storage != EvqBuffer)
return;
+ if (qualifier.layoutPacking != ElpStd140 && qualifier.layoutPacking != ElpStd430)
+ return;
+
+ int offset = 0;
+ int memberSize;
+ for (unsigned int member = 0; member < typeList.size(); ++member) {
+ TQualifier& memberQualifier = typeList[member].type->getQualifier();
+ TSourceLoc memberLoc = typeList[member].loc;
+
+ // "When align is applied to an array, it effects only the start of the array, not the array's internal stride."
+
+ int memberAlignment = intermediate.getBaseAlignment(*typeList[member].type, memberSize, qualifier.layoutPacking == ElpStd140);
+ if (memberQualifier.hasOffset()) {
+ // "The specified offset must be a multiple
+ // of the base alignment of the type of the block member it qualifies, or a compile-time error results."
+ if (! IsMultipleOfPow2(memberQualifier.layoutOffset, memberAlignment))
+ error(memberLoc, "must be a multiple of the member's alignment", "offset", "");
+
+ // "It is a compile-time error to specify an offset that is smaller than the offset of the previous
+ // member in the block or that lies within the previous member of the block"
+ if (memberQualifier.layoutOffset < offset)
+ error(memberLoc, "cannot lie in previous members", "offset", "");
+
+ // "The offset qualifier forces the qualified member to start at or after the specified
+ // integral-constant expression, which will be its byte offset from the beginning of the buffer.
+ // "The actual offset of a member is computed as
+ // follows: If offset was declared, start with that offset, otherwise start with the next available offset."
+ offset = std::max(offset, memberQualifier.layoutOffset);
+ }
+
+ // "The actual alignment of a member will be the greater of the specified align alignment and the standard
+ // (e.g., std140) base alignment for the member's type."
+ if (memberQualifier.hasAlign())
+ memberAlignment = std::max(memberAlignment, memberQualifier.layoutAlign);
+
+ // "If the resulting offset is not a multiple of the actual alignment,
+ // increase it to the first offset that is a multiple of
+ // the actual alignment."
+ RoundToPow2(offset, memberAlignment);
+ typeList[member].type->getQualifier().layoutOffset = offset;
+ offset += memberSize;
+ }
}
// For an identifier that is already declared, add more qualification to it.
diff --git a/glslang/MachineIndependent/SymbolTable.cpp b/glslang/MachineIndependent/SymbolTable.cpp
index 1384c4660..9a2f3b25b 100644
--- a/glslang/MachineIndependent/SymbolTable.cpp
+++ b/glslang/MachineIndependent/SymbolTable.cpp
@@ -118,20 +118,6 @@ void TType::buildMangledName(TString& mangledName)
}
}
-int TType::getStructSize() const
-{
- if (! isStruct()) {
- assert(false && "Not a struct");
- return 0;
- }
-
- if (structureSize == 0)
- for (TTypeList::iterator tl = getStruct()->begin(); tl != getStruct()->end(); tl++)
- structureSize += ((*tl).type)->getObjectSize();
-
- return structureSize;
-}
-
//
// Dump functions.
//
@@ -256,7 +242,6 @@ TVariable::TVariable(const TVariable& copyOf) : TSymbol(copyOf)
if (! copyOf.unionArray.empty()) {
assert(! copyOf.type.isStruct());
- assert(copyOf.type.getObjectSize() == 1);
TConstUnionArray newArray(1);
newArray[0] = copyOf.unionArray[0];
unionArray = newArray;
diff --git a/glslang/MachineIndependent/intermOut.cpp b/glslang/MachineIndependent/intermOut.cpp
index f51035340..9f89cadba 100644
--- a/glslang/MachineIndependent/intermOut.cpp
+++ b/glslang/MachineIndependent/intermOut.cpp
@@ -411,7 +411,7 @@ bool TOutputTraverser::visitSelection(TVisit /* visit */, TIntermSelection* node
void OutputConstantUnion(TInfoSink& out, const TIntermTyped* node, const TConstUnionArray& constUnion, int depth)
{
- int size = node->getType().getObjectSize();
+ int size = node->getType().computeNumComponents();
for (int i = 0; i < size; i++) {
OutputTreeText(out, node, depth);
diff --git a/glslang/MachineIndependent/linkValidate.cpp b/glslang/MachineIndependent/linkValidate.cpp
index 491b637aa..ec32c9009 100644
--- a/glslang/MachineIndependent/linkValidate.cpp
+++ b/glslang/MachineIndependent/linkValidate.cpp
@@ -60,6 +60,10 @@ void TIntermediate::error(TInfoSink& infoSink, const char* message)
++numErrors;
}
+// TODO: 4.4 offset/align: "Two blocks linked together in the same program with the same block
+// name must have the exact same set of members qualified with offset and their integral-constant
+// expression values must be the same, or a link-time error results."
+
//
// Merge the information from 'unit' into 'this'
//
@@ -266,7 +270,9 @@ void TIntermediate::mergeErrorCheck(TInfoSink& infoSink, const TIntermSymbol& sy
}
// Layouts...
- // TODO: 4.4 enhanced layouts: generalize to include offset/align
+ // TODO: 4.4 enhanced layouts: Generalize to include offset/align: currrent spec
+ // requires separate user-supplied offset from actual computed offset, but
+ // current implementation only has one offset.
if (symbol.getQualifier().layoutMatrix != unitSymbol.getQualifier().layoutMatrix ||
symbol.getQualifier().layoutPacking != unitSymbol.getQualifier().layoutPacking ||
symbol.getQualifier().layoutLocation != unitSymbol.getQualifier().layoutLocation ||
@@ -321,7 +327,7 @@ void TIntermediate::finalCheck(TInfoSink& infoSink)
// "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
- // in different compilation units. While xfb_stridecan be declared multiple times for the same buffer, it is a
+ // in different compilation units. While xfb_stride can be declared multiple times for the same buffer, it is a
// compile-time or link-time error to have different values specified for the stride for the same buffer."
if (xfbBuffers[b].stride != TQualifier::layoutXfbStrideEnd && xfbBuffers[b].implicitStride > xfbBuffers[b].stride) {
error(infoSink, "xfb_stride is too small to hold all buffer entries:");
@@ -740,4 +746,153 @@ unsigned int TIntermediate::computeTypeXfbSize(const TType& type, bool& contains
return 4 * numComponents;
}
+const int baseAlignmentVec4Std140 = 16;
+
+// Return the size and alignment of a scalar.
+// The size is returned in the 'size' parameter
+// Return value is the alignment of the type.
+int TIntermediate::getBaseAlignmentScalar(const TType& type, int& size) const
+{
+ switch (type.getBasicType()) {
+ case EbtDouble: size = 8; return 8;
+ default: size = 4; return 4;
+ }
+}
+
+// Implement base-alignment and size rules from section 7.6.2.2 Standard Uniform Block Layout
+// Operates recursively.
+//
+// If std140 is true, it does the rounding up to vec4 size required by std140,
+// otherwise it does not, yielding std430 rules.
+//
+// The size is returned in the 'size' parameter
+// Return value is the alignment of the type.
+int TIntermediate::getBaseAlignment(const TType& type, int& size, bool std140) const
+{
+ int alignment;
+
+ // When using the std140 storage layout, structures will be laid out in buffer
+ // storage with its members stored in monotonically increasing order based on their
+ // location in the declaration. A structure and each structure member have a base
+ // offset and a base alignment, from which an aligned offset is computed by rounding
+ // the base offset up to a multiple of the base alignment. The base offset of the first
+ // member of a structure is taken from the aligned offset of the structure itself. The
+ // base offset of all other structure members is derived by taking the offset of the
+ // last basic machine unit consumed by the previous member and adding one. Each
+ // structure member is stored in memory at its aligned offset. The members of a top-
+ // level uniform block are laid out in buffer storage by treating the uniform block as
+ // a structure with a base offset of zero.
+ //
+ // 1. If the member is a scalar consuming N basic machine units, the base alignment is N.
+ //
+ // 2. If the member is a two- or four-component vector with components consuming N basic
+ // machine units, the base alignment is 2N or 4N, respectively.
+ //
+ // 3. If the member is a three-component vector with components consuming N
+ // basic machine units, the base alignment is 4N.
+ //
+ // 4. If the member is an array of scalars or vectors, the base alignment and array
+ // stride are set to match the base alignment of a single array element, according
+ // to rules (1), (2), and (3), and rounded up to the base alignment of a vec4. The
+ // array may have padding at the end; the base offset of the member following
+ // the array is rounded up to the next multiple of the base alignment.
+ //
+ // 5. If the member is a column-major matrix with C columns and R rows, the
+ // matrix is stored identically to an array of C column vectors with R
+ // components each, according to rule (4).
+ //
+ // 6. If the member is an array of S column-major matrices with C columns and
+ // R rows, the matrix is stored identically to a row of S C column vectors
+ // with R components each, according to rule (4).
+ //
+ // 7. If the member is a row-major matrix with C columns and R rows, the matrix
+ // is stored identically to an array of R row vectors with C components each,
+ // according to rule (4).
+ //
+ // 8. If the member is an array of S row-major matrices with C columns and R
+ // rows, the matrix is stored identically to a row of S R row vectors with C
+ // components each, according to rule (4).
+ //
+ // 9. If the member is a structure, the base alignment of the structure is N , where
+ // N is the largest base alignment value of any of its members, and rounded
+ // up to the base alignment of a vec4. The individual members of this substructure
+ // are then assigned offsets by applying this set of rules recursively,
+ // where the base offset of the first member of the sub-structure is equal to the
+ // aligned offset of the structure. The structure may have padding at the end;
+ // the base offset of the member following the sub-structure is rounded up to
+ // the next multiple of the base alignment of the structure.
+ //
+ // 10. If the member is an array of S structures, the S elements of the array are laid
+ // out in order, according to rule (9).
+
+ // rules 4, 6, and 8
+ if (type.isArray()) {
+ TType derefType(type, 0);
+ alignment = getBaseAlignment(derefType, size, std140);
+ if (std140)
+ alignment = std::max(baseAlignmentVec4Std140, alignment);
+ RoundToPow2(size, alignment);
+ size *= type.getArraySize();
+ return alignment;
+ }
+
+ // rule 9
+ if (type.getBasicType() == EbtStruct) {
+ const TTypeList& memberList = *type.getStruct();
+
+ size = 0;
+ int maxAlignment = std140 ? baseAlignmentVec4Std140 : 0;
+ for (size_t m = 0; m < memberList.size(); ++m) {
+ int memberSize;
+ int memberAlignment = getBaseAlignment(*memberList[m].type, memberSize, std140);
+ maxAlignment = std::max(maxAlignment, memberAlignment);
+ RoundToPow2(size, memberAlignment);
+ size += memberSize;
+ }
+
+ return maxAlignment;
+ }
+
+ // rule 1
+ if (type.isScalar())
+ return getBaseAlignmentScalar(type, size);
+
+ // rules 2 and 3
+ if (type.isVector()) {
+ int scalarAlign = getBaseAlignmentScalar(type, size);
+ switch (type.getVectorSize()) {
+ case 2:
+ size *= 2;
+ return 2 * scalarAlign;
+ default:
+ size *= type.getVectorSize();
+ return 4 * scalarAlign;
+ }
+ }
+
+ // rules 5 and 7
+ if (type.isMatrix()) {
+ TType derefType(type, 0);
+
+ // rule 5: deref to row, not to column, meaning the size of vector is num columns instead of num rows
+ if (type.getQualifier().layoutMatrix == ElmRowMajor)
+ derefType.setElementType(derefType.getBasicType(), type.getMatrixCols(), 0, 0, 0);
+
+ alignment = getBaseAlignment(derefType, size, std140);
+ if (std140)
+ alignment = std::max(baseAlignmentVec4Std140, alignment);
+ RoundToPow2(size, alignment);
+ if (type.getQualifier().layoutMatrix == ElmRowMajor)
+ size *= type.getMatrixRows();
+ else
+ size *= type.getMatrixCols();
+
+ return alignment;
+ }
+
+ assert(0); // all cases should be covered above
+ size = baseAlignmentVec4Std140;
+ return baseAlignmentVec4Std140;
+}
+
} // end namespace glslang
diff --git a/glslang/MachineIndependent/localintermediate.h b/glslang/MachineIndependent/localintermediate.h
index c3d7588a8..9bb17099e 100644
--- a/glslang/MachineIndependent/localintermediate.h
+++ b/glslang/MachineIndependent/localintermediate.h
@@ -239,6 +239,7 @@ public:
}
int addXfbBufferOffset(const TType&);
unsigned int computeTypeXfbSize(const TType&, bool& containsDouble) const;
+ int getBaseAlignment(const TType&, int& size, bool std140) const;
protected:
void error(TInfoSink& infoSink, const char*);
@@ -249,6 +250,7 @@ protected:
void inOutLocationCheck(TInfoSink&);
TIntermSequence& findLinkerObjects() const;
bool userOutputUsed() const;
+ int getBaseAlignmentScalar(const TType&, int& size) const;
protected:
const EShLanguage language;
diff --git a/glslang/MachineIndependent/parseConst.cpp b/glslang/MachineIndependent/parseConst.cpp
index ac643b680..978dc8a41 100644
--- a/glslang/MachineIndependent/parseConst.cpp
+++ b/glslang/MachineIndependent/parseConst.cpp
@@ -82,7 +82,7 @@ bool TConstTraverser::visitAggregate(TVisit /* visit */, TIntermAggregate* node)
if (flag) {
singleConstantParam = true;
constructorType = node->getOp();
- size = node->getType().getObjectSize();
+ size = node->getType().computeNumComponents();
if (node->getType().isMatrix()) {
isMatrix = true;
@@ -115,13 +115,13 @@ bool TConstTraverser::visitAggregate(TVisit /* visit */, TIntermAggregate* node)
void TConstTraverser::visitConstantUnion(TIntermConstantUnion* node)
{
TConstUnionArray leftUnionArray(unionArray);
- int instanceSize = type.getObjectSize();
+ int instanceSize = type.computeNumComponents();
if (index >= instanceSize)
return;
if (! singleConstantParam) {
- int rightUnionSize = node->getType().getObjectSize();
+ int rightUnionSize = node->getType().computeNumComponents();
const TConstUnionArray& rightUnionArray = node->getConstArray();
for (int i = 0; i < rightUnionSize; i++) {
@@ -136,6 +136,7 @@ void TConstTraverser::visitConstantUnion(TIntermConstantUnion* node)
const TConstUnionArray& rightUnionArray = node->getConstArray();
if (! isMatrix) {
int count = 0;
+ int nodeComps = node->getType().computeNumComponents();
for (int i = index; i < endIndex; i++) {
if (i >= instanceSize)
return;
@@ -144,7 +145,7 @@ void TConstTraverser::visitConstantUnion(TIntermConstantUnion* node)
(index)++;
- if (node->getType().getObjectSize() > 1)
+ if (nodeComps > 1)
count++;
}
} else {
@@ -169,6 +170,7 @@ void TConstTraverser::visitConstantUnion(TIntermConstantUnion* node)
// matrix from vector
int count = 0;
const int startIndex = index;
+ int nodeComps = node->getType().computeNumComponents();
for (int i = startIndex; i < endIndex; i++) {
if (i >= instanceSize)
return;
@@ -179,7 +181,7 @@ void TConstTraverser::visitConstantUnion(TIntermConstantUnion* node)
index++;
- if (node->getType().getObjectSize() > 1)
+ if (nodeComps > 1)
count++;
}
}
diff --git a/glslang/MachineIndependent/reflection.cpp b/glslang/MachineIndependent/reflection.cpp
index 164ced57b..709fc45ec 100644
--- a/glslang/MachineIndependent/reflection.cpp
+++ b/glslang/MachineIndependent/reflection.cpp
@@ -108,139 +108,42 @@ public:
}
}
- static const int baseAlignmentVec4Std140;
-
- // align a value: if 'value' is not aligned to 'alignment', move it up to a multiple of alignment
- void align(int& value, int alignment)
- {
- int error = value % alignment;
- if (error)
- value += alignment - error;
- }
-
- // return the size and alignment of a scalar
- int getBaseAlignmentScalar(const TType& type, int& size)
- {
- switch (type.getBasicType()) {
- case EbtDouble: size = 8; return 8;
- default: size = 4; return 4;
- }
- }
-
- // Implement base-alignment and size rules from section 7.6.2.2 Standard Uniform Block Layout
- // Operates recursively.
- // If std140 is true, it does the rounding up to vec4 size required by std140,
- // otherwise it does not, yielding std430 rules.
- //
- // Returns the size of the type.
- int getBaseAlignment(const TType& type, int& size, bool std140)
+ // Lookup or calculate the offset of a block member, using the recursively
+ // defined block offset rules.
+ int getOffset(const TType& type, int index)
{
- int alignment;
-
- // rules 4, 6, and 8
- if (type.isArray()) {
- TType derefType(type, 0);
- alignment = getBaseAlignment(derefType, size, std140);
- if (std140)
- alignment = std::max(baseAlignmentVec4Std140, alignment);
- align(size, alignment);
- size *= type.getArraySize();
- return alignment;
- }
-
- // rule 9
- if (type.getBasicType() == EbtStruct) {
- const TTypeList& memberList = *type.getStruct();
-
- size = 0;
- int maxAlignment = std140 ? baseAlignmentVec4Std140 : 0;
- for (size_t m = 0; m < memberList.size(); ++m) {
- int memberSize;
- int memberAlignment = getBaseAlignment(*memberList[m].type, memberSize, std140);
- maxAlignment = std::max(maxAlignment, memberAlignment);
- align(size, memberAlignment);
- size += memberSize;
- }
-
- return maxAlignment;
- }
-
- // rule 1
- if (type.isScalar())
- return getBaseAlignmentScalar(type, size);
-
- // rules 2 and 3
- if (type.isVector()) {
- int scalarAlign = getBaseAlignmentScalar(type, size);
- switch (type.getVectorSize()) {
- case 2:
- size *= 2;
- return 2 * scalarAlign;
- default:
- size *= type.getVectorSize();
- return 4 * scalarAlign;
- }
- }
-
- // rules 5 and 7
- if (type.isMatrix()) {
- TType derefType(type, 0);
-
- // rule 5: deref to row, not to column, meaning the size of vector is num columns instead of num rows
- if (type.getQualifier().layoutMatrix == ElmRowMajor)
- derefType.setElementType(derefType.getBasicType(), type.getMatrixCols(), 0, 0, 0);
-
- alignment = getBaseAlignment(derefType, size, std140);
- if (std140)
- alignment = std::max(baseAlignmentVec4Std140, alignment);
- align(size, alignment);
- if (type.getQualifier().layoutMatrix == ElmRowMajor)
- size *= type.getMatrixRows();
- else
- size *= type.getMatrixCols();
-
- return alignment;
- }
+ const TTypeList& memberList = *type.getStruct();
- assert(0); // all cases should be covered above
- size = baseAlignmentVec4Std140;
- return baseAlignmentVec4Std140;
- }
-
- // Calculate the offset of a block member, using the recursively defined
- // block offset rules.
- int getBlockMemberOffset(const TType& blockType, int index)
- {
- // TODO: reflection performance: cache intermediate results instead of recomputing them
+ // Don't calculate offset if one is present, it could be user supplied
+ // and different than what would be calculated. That is, this is faster,
+ // but not just an optimization.
+ if (memberList[index].type->getQualifier().hasOffset())
+ return memberList[index].type->getQualifier().layoutOffset;
- int offset = 0;
- const TTypeList& memberList = *blockType.getStruct();
int memberSize;
- for (int m = 0; m < index; ++m) {
- int memberAlignment = getBaseAlignment(*memberList[m].type, memberSize, blockType.getQualifier().layoutPacking == ElpStd140);
- align(offset, memberAlignment);
- offset += memberSize;
+ int offset = 0;
+ for (int m = 0; m <= index; ++m) {
+ int memberAlignment = intermediate.getBaseAlignment(*memberList[m].type, memberSize, type.getQualifier().layoutPacking == ElpStd140);
+ RoundToPow2(offset, memberAlignment);
+ if (m < index)
+ offset += memberSize;
}
- int memberAlignment = getBaseAlignment(*memberList[index].type, memberSize, blockType.getQualifier().layoutPacking == ElpStd140);
- align(offset, memberAlignment);
return offset;
}
// Calculate the block data size.
- // Arrayness is not taken into account, each element is backed by a separate buffer.
+ // Block arrayness is not taken into account, each element is backed by a separate buffer.
int getBlockSize(const TType& blockType)
{
- int size = 0;
- const TTypeList& memberList = *blockType.getStruct();
- int memberSize;
- for (size_t m = 0; m < memberList.size(); ++m) {
- int memberAlignment = getBaseAlignment(*memberList[m].type, memberSize, blockType.getQualifier().layoutPacking == ElpStd140);
- align(size, memberAlignment);
- size += memberSize;
- }
+ const TTypeList& memberList = *blockType.getStruct();
+ int lastIndex = memberList.size() - 1;
+ int lastOffset = getOffset(blockType, lastIndex);
- return size;
+ int lastMemberSize;
+ intermediate.getBaseAlignment(*memberList[lastIndex].type, lastMemberSize, blockType.getQualifier().layoutPacking == ElpStd140);
+
+ return lastOffset + lastMemberSize;
}
// Traverse the provided deref chain, including the base, and
@@ -283,7 +186,7 @@ public:
case EOpIndexDirectStruct:
index = visitNode->getRight()->getAsConstantUnion()->getConstArray()[0].getIConst();
if (offset >= 0)
- offset += getBlockMemberOffset(visitNode->getLeft()->getType(), index);
+ offset += getOffset(visitNode->getLeft()->getType(), index);
if (name.size() > 0)
name.append(".");
name.append((*visitNode->getLeft()->getType().getStruct())[index].type->getFieldName());
@@ -715,8 +618,6 @@ public:
std::set<const TIntermNode*> processedDerefs;
};
-const int TLiveTraverser::baseAlignmentVec4Std140 = 16;
-
//
// Implement the traversal functions of interest.
//
--
GitLab