HLSL: allow destination swizzles when writing RWTexture/RWBuffer objects.

Reads and write syntax to UAV objects is turned into EOpImageLoad/Store
operations.  This translation did not support destination swizzles,
for example, "mybuffer[tc].zyx = 3;", so such statements would fail to
compile.  Now they work.

Parial updates are explicitly prohibited.

New test: hlsl.rw.swizzle.frag
This commit is contained in:
steve-lunarg 2016-12-28 10:03:58 -07:00
parent 807a0d9e2f
commit cd6829ba81
4 changed files with 381 additions and 4 deletions

View File

@ -0,0 +1,286 @@
hlsl.rw.swizzle.frag
Shader version: 450
gl_FragCoord origin is upper left
0:? Sequence
0:4 Function Definition: SomeValue( (temp 3-component vector of float)
0:4 Function Parameters:
0:? Sequence
0:4 Branch: Return with expression
0:? Constant:
0:? 1.000000
0:? 2.000000
0:? 3.000000
0:7 Function Definition: main( (temp 4-component vector of float)
0:7 Function Parameters:
0:? Sequence
0:8 Sequence
0:8 move second child to first child (temp 2-component vector of int)
0:8 'tc2' (temp 2-component vector of int)
0:8 Constant:
0:8 0 (const int)
0:8 0 (const int)
0:9 Sequence
0:9 move second child to first child (temp int)
0:9 'tc' (temp int)
0:9 Constant:
0:9 0 (const int)
0:12 Sequence
0:12 move second child to first child (temp 3-component vector of float)
0:12 vector swizzle (temp 3-component vector of float)
0:12 'storeTemp' (temp 3-component vector of float)
0:12 Sequence
0:12 Constant:
0:12 2 (const int)
0:12 Constant:
0:12 1 (const int)
0:12 Constant:
0:12 0 (const int)
0:? Constant:
0:? 1.000000
0:? 2.000000
0:? 3.000000
0:12 imageStore (temp void)
0:12 'rwtx' (layout(rgba32f ) uniform image2D)
0:12 'tc2' (temp 2-component vector of int)
0:12 'storeTemp' (temp 3-component vector of float)
0:12 'storeTemp' (temp 3-component vector of float)
0:13 Sequence
0:13 move second child to first child (temp 3-component vector of float)
0:13 vector swizzle (temp 3-component vector of float)
0:13 'storeTemp' (temp 3-component vector of float)
0:13 Sequence
0:13 Constant:
0:13 2 (const int)
0:13 Constant:
0:13 1 (const int)
0:13 Constant:
0:13 0 (const int)
0:13 Function Call: SomeValue( (temp 3-component vector of float)
0:13 imageStore (temp void)
0:13 'rwtx' (layout(rgba32f ) uniform image2D)
0:13 'tc2' (temp 2-component vector of int)
0:13 'storeTemp' (temp 3-component vector of float)
0:13 'storeTemp' (temp 3-component vector of float)
0:14 Sequence
0:14 move second child to first child (temp 3-component vector of float)
0:14 vector swizzle (temp 3-component vector of float)
0:14 'storeTemp' (temp 3-component vector of float)
0:14 Sequence
0:14 Constant:
0:14 2 (const int)
0:14 Constant:
0:14 1 (const int)
0:14 Constant:
0:14 0 (const int)
0:14 Constant:
0:14 2.000000
0:14 2.000000
0:14 2.000000
0:14 imageStore (temp void)
0:14 'rwtx' (layout(rgba32f ) uniform image2D)
0:14 'tc2' (temp 2-component vector of int)
0:14 'storeTemp' (temp 3-component vector of float)
0:14 'storeTemp' (temp 3-component vector of float)
0:27 Sequence
0:27 move second child to first child (temp 4-component vector of float)
0:? '@entryPointOutput' (layout(location=0 ) out 4-component vector of float)
0:27 Constant:
0:27 0.000000
0:27 0.000000
0:27 0.000000
0:27 0.000000
0:27 Branch: Return
0:? Linker Objects
0:? '@entryPointOutput' (layout(location=0 ) out 4-component vector of float)
0:? 'rwtx' (layout(rgba32f ) uniform image2D)
0:? 'buf' (layout(rgba32f ) uniform imageBuffer)
Linked fragment stage:
Shader version: 450
gl_FragCoord origin is upper left
0:? Sequence
0:4 Function Definition: SomeValue( (temp 3-component vector of float)
0:4 Function Parameters:
0:? Sequence
0:4 Branch: Return with expression
0:? Constant:
0:? 1.000000
0:? 2.000000
0:? 3.000000
0:7 Function Definition: main( (temp 4-component vector of float)
0:7 Function Parameters:
0:? Sequence
0:8 Sequence
0:8 move second child to first child (temp 2-component vector of int)
0:8 'tc2' (temp 2-component vector of int)
0:8 Constant:
0:8 0 (const int)
0:8 0 (const int)
0:9 Sequence
0:9 move second child to first child (temp int)
0:9 'tc' (temp int)
0:9 Constant:
0:9 0 (const int)
0:12 Sequence
0:12 move second child to first child (temp 3-component vector of float)
0:12 vector swizzle (temp 3-component vector of float)
0:12 'storeTemp' (temp 3-component vector of float)
0:12 Sequence
0:12 Constant:
0:12 2 (const int)
0:12 Constant:
0:12 1 (const int)
0:12 Constant:
0:12 0 (const int)
0:? Constant:
0:? 1.000000
0:? 2.000000
0:? 3.000000
0:12 imageStore (temp void)
0:12 'rwtx' (layout(rgba32f ) uniform image2D)
0:12 'tc2' (temp 2-component vector of int)
0:12 'storeTemp' (temp 3-component vector of float)
0:12 'storeTemp' (temp 3-component vector of float)
0:13 Sequence
0:13 move second child to first child (temp 3-component vector of float)
0:13 vector swizzle (temp 3-component vector of float)
0:13 'storeTemp' (temp 3-component vector of float)
0:13 Sequence
0:13 Constant:
0:13 2 (const int)
0:13 Constant:
0:13 1 (const int)
0:13 Constant:
0:13 0 (const int)
0:13 Function Call: SomeValue( (temp 3-component vector of float)
0:13 imageStore (temp void)
0:13 'rwtx' (layout(rgba32f ) uniform image2D)
0:13 'tc2' (temp 2-component vector of int)
0:13 'storeTemp' (temp 3-component vector of float)
0:13 'storeTemp' (temp 3-component vector of float)
0:14 Sequence
0:14 move second child to first child (temp 3-component vector of float)
0:14 vector swizzle (temp 3-component vector of float)
0:14 'storeTemp' (temp 3-component vector of float)
0:14 Sequence
0:14 Constant:
0:14 2 (const int)
0:14 Constant:
0:14 1 (const int)
0:14 Constant:
0:14 0 (const int)
0:14 Constant:
0:14 2.000000
0:14 2.000000
0:14 2.000000
0:14 imageStore (temp void)
0:14 'rwtx' (layout(rgba32f ) uniform image2D)
0:14 'tc2' (temp 2-component vector of int)
0:14 'storeTemp' (temp 3-component vector of float)
0:14 'storeTemp' (temp 3-component vector of float)
0:27 Sequence
0:27 move second child to first child (temp 4-component vector of float)
0:? '@entryPointOutput' (layout(location=0 ) out 4-component vector of float)
0:27 Constant:
0:27 0.000000
0:27 0.000000
0:27 0.000000
0:27 0.000000
0:27 Branch: Return
0:? Linker Objects
0:? '@entryPointOutput' (layout(location=0 ) out 4-component vector of float)
0:? 'rwtx' (layout(rgba32f ) uniform image2D)
0:? 'buf' (layout(rgba32f ) uniform imageBuffer)
// Module Version 10000
// Generated by (magic number): 80001
// Id's are bound by 58
Capability Shader
Capability SampledBuffer
1: ExtInstImport "GLSL.std.450"
MemoryModel Logical GLSL450
EntryPoint Fragment 4 "main" 51
ExecutionMode 4 OriginUpperLeft
Name 4 "main"
Name 9 "SomeValue("
Name 20 "tc2"
Name 24 "tc"
Name 26 "storeTemp"
Name 31 "rwtx"
Name 35 "storeTemp"
Name 42 "storeTemp"
Name 51 "@entryPointOutput"
Name 57 "buf"
Decorate 31(rwtx) DescriptorSet 0
Decorate 51(@entryPointOutput) Location 0
Decorate 57(buf) DescriptorSet 0
2: TypeVoid
3: TypeFunction 2
6: TypeFloat 32
7: TypeVector 6(float) 3
8: TypeFunction 7(fvec3)
11: 6(float) Constant 1065353216
12: 6(float) Constant 1073741824
13: 6(float) Constant 1077936128
14: 7(fvec3) ConstantComposite 11 12 13
17: TypeInt 32 1
18: TypeVector 17(int) 2
19: TypePointer Function 18(ivec2)
21: 17(int) Constant 0
22: 18(ivec2) ConstantComposite 21 21
23: TypePointer Function 17(int)
25: TypePointer Function 7(fvec3)
29: TypeImage 6(float) 2D nonsampled format:Rgba32f
30: TypePointer UniformConstant 29
31(rwtx): 30(ptr) Variable UniformConstant
43: 7(fvec3) ConstantComposite 12 12 12
49: TypeVector 6(float) 4
50: TypePointer Output 49(fvec4)
51(@entryPointOutput): 50(ptr) Variable Output
52: 6(float) Constant 0
53: 49(fvec4) ConstantComposite 52 52 52 52
55: TypeImage 6(float) Buffer nonsampled format:Rgba32f
56: TypePointer UniformConstant 55
57(buf): 56(ptr) Variable UniformConstant
4(main): 2 Function None 3
5: Label
20(tc2): 19(ptr) Variable Function
24(tc): 23(ptr) Variable Function
26(storeTemp): 25(ptr) Variable Function
35(storeTemp): 25(ptr) Variable Function
42(storeTemp): 25(ptr) Variable Function
Store 20(tc2) 22
Store 24(tc) 21
27: 7(fvec3) Load 26(storeTemp)
28: 7(fvec3) VectorShuffle 27 14 5 4 3
Store 26(storeTemp) 28
32: 29 Load 31(rwtx)
33: 18(ivec2) Load 20(tc2)
34: 7(fvec3) Load 26(storeTemp)
ImageWrite 32 33 34
36: 7(fvec3) FunctionCall 9(SomeValue()
37: 7(fvec3) Load 35(storeTemp)
38: 7(fvec3) VectorShuffle 37 36 5 4 3
Store 35(storeTemp) 38
39: 29 Load 31(rwtx)
40: 18(ivec2) Load 20(tc2)
41: 7(fvec3) Load 35(storeTemp)
ImageWrite 39 40 41
44: 7(fvec3) Load 42(storeTemp)
45: 7(fvec3) VectorShuffle 44 43 5 4 3
Store 42(storeTemp) 45
46: 29 Load 31(rwtx)
47: 18(ivec2) Load 20(tc2)
48: 7(fvec3) Load 42(storeTemp)
ImageWrite 46 47 48
Store 51(@entryPointOutput) 53
Return
FunctionEnd
9(SomeValue(): 7(fvec3) Function None 8
10: Label
ReturnValue 14
FunctionEnd

28
Test/hlsl.rw.swizzle.frag Normal file
View File

@ -0,0 +1,28 @@
RWTexture2D<float3> rwtx;
RWBuffer<float3> buf;
float3 SomeValue() { return float3(1,2,3); }
float4 main() : SV_Target0
{
int2 tc2 = { 0, 0 };
int tc = 0;
// Test swizzles and partial updates of L-values when writing to buffers and writable textures.
rwtx[tc2].zyx = float3(1,2,3); // full swizzle, simple RHS
rwtx[tc2].zyx = SomeValue(); // full swizzle, complex RHS
rwtx[tc2].zyx = 2; // full swizzle, modify op
// Partial updates not yet supported.
// Partial values, which will use swizzles.
// buf[tc].yz = 42; // partial swizzle, simple RHS
// buf[tc].yz = SomeValue().x; // partial swizzle, complex RHS
// buf[tc].yz += 43; // partial swizzle, modify op
// // Partial values, which will use index.
// buf[tc].y = 44; // single index, simple RHS
// buf[tc].y = SomeValue().x; // single index, complex RHS
// buf[tc].y += 45; // single index, modify op
return 0.0;
}

View File

@ -171,6 +171,7 @@ INSTANTIATE_TEST_CASE_P(
{"hlsl.rw.bracket.frag", "main"},
{"hlsl.rw.register.frag", "main"},
{"hlsl.rw.scalar.bracket.frag", "main"},
{"hlsl.rw.swizzle.frag", "main"},
{"hlsl.rw.vec2.bracket.frag", "main"},
{"hlsl.sample.array.dx10.frag", "main"},
{"hlsl.sample.basic.dx10.frag", "main"},

View File

@ -47,6 +47,7 @@
#include <algorithm>
#include <functional>
#include <cctype>
#include <array>
namespace glslang {
@ -144,6 +145,12 @@ bool HlslParseContext::shouldConvertLValue(const TIntermNode* node) const
return false;
const TIntermAggregate* lhsAsAggregate = node->getAsAggregate();
const TIntermBinary* lhsAsBinary = node->getAsBinaryNode();
// If it's a swizzled/indexed aggregate, look at the left node instead.
if (lhsAsBinary != nullptr &&
(lhsAsBinary->getOp() == EOpVectorSwizzle || lhsAsBinary->getOp() == EOpIndexDirect))
lhsAsAggregate = lhsAsBinary->getLeft()->getAsAggregate();
if (lhsAsAggregate != nullptr && lhsAsAggregate->getOp() == EOpImageLoad)
return true;
@ -282,6 +289,34 @@ TIntermTyped* HlslParseContext::handleLvalue(const TSourceLoc& loc, const char*
loc);
};
// Return true if swizzle or index writes all components of the given variable.
const auto writesAllComponents = [&](TIntermSymbol* var, TIntermBinary* swizzle) -> bool {
if (swizzle == nullptr) // not a swizzle or index
return true;
// Track which components are being set.
std::array<bool, 4> compIsSet;
compIsSet.fill(false);
const TIntermConstantUnion* asConst = swizzle->getRight()->getAsConstantUnion();
const TIntermAggregate* asAggregate = swizzle->getRight()->getAsAggregate();
// This could be either a direct index, or a swizzle.
if (asConst) {
compIsSet[asConst->getConstArray()[0].getIConst()] = true;
} else if (asAggregate) {
const TIntermSequence& seq = asAggregate->getSequence();
for (int comp=0; comp<int(seq.size()); ++comp)
compIsSet[seq[comp]->getAsConstantUnion()->getConstArray()[0].getIConst()] = true;
} else {
assert(0);
}
// Return true if all components are being set by the index or swizzle
return std::all_of(compIsSet.begin(), compIsSet.begin() + var->getType().getVectorSize(),
[](bool isSet) { return isSet; } );
};
// helper to create a temporary variable
const auto addTmpVar = [&](const char* name, const TType& derefType) -> TIntermSymbol* {
TVariable* tmpVar = makeInternalVariable(name, derefType);
@ -289,7 +324,24 @@ TIntermTyped* HlslParseContext::handleLvalue(const TSourceLoc& loc, const char*
return intermediate.addSymbol(*tmpVar, loc);
};
// Create swizzle matching input swizzle
const auto addSwizzle = [&](TIntermSymbol* var, TIntermBinary* swizzle) -> TIntermTyped* {
if (swizzle)
return intermediate.addBinaryNode(swizzle->getOp(), var, swizzle->getRight(), loc, swizzle->getType());
else
return var;
};
TIntermBinary* lhsAsBinary = lhs->getAsBinaryNode();
TIntermAggregate* lhsAsAggregate = lhs->getAsAggregate();
bool lhsIsSwizzle = false;
// If it's a swizzled L-value, remember the swizzle, and use the LHS.
if (lhsAsBinary != nullptr && (lhsAsBinary->getOp() == EOpVectorSwizzle || lhsAsBinary->getOp() == EOpIndexDirect)) {
lhsAsAggregate = lhsAsBinary->getLeft()->getAsAggregate();
lhsIsSwizzle = true;
}
TIntermTyped* object = lhsAsAggregate->getSequence()[0]->getAsTyped();
TIntermTyped* coord = lhsAsAggregate->getSequence()[1]->getAsTyped();
@ -336,16 +388,26 @@ TIntermTyped* HlslParseContext::handleLvalue(const TSourceLoc& loc, const char*
// OpSequence
// coordtmp = load's param1
// rhsTmp = OpImageLoad(object, coordTmp)
// rhsTmp op= rhs
// rhsTmp op = rhs
// OpImageStore(object, coordTmp, rhsTmp)
// rhsTmp
//
// If the lvalue is swizzled, we apply that when writing the temp variable, like so:
// ...
// rhsTmp.some_swizzle = ...
// For partial writes, an error is generated.
TIntermSymbol* rhsTmp = rhs->getAsSymbolNode();
TIntermTyped* coordTmp = coord;
if (rhsTmp == nullptr || isModifyOp) {
if (rhsTmp == nullptr || isModifyOp || lhsIsSwizzle) {
rhsTmp = addTmpVar("storeTemp", objDerefType);
// Partial updates not yet supported
if (!writesAllComponents(rhsTmp, lhsAsBinary)) {
error(loc, "unimplemented: partial image updates", "", "");
}
// Assign storeTemp = rhs
if (isModifyOp) {
// We have to make a temp var for the coordinate, to avoid evaluating it twice.
@ -355,7 +417,7 @@ TIntermTyped* HlslParseContext::handleLvalue(const TSourceLoc& loc, const char*
}
// rhsTmp op= rhs.
makeBinary(assignOp, intermediate.addSymbol(*rhsTmp), rhs);
makeBinary(assignOp, addSwizzle(intermediate.addSymbol(*rhsTmp), lhsAsBinary), rhs);
}
makeStore(object, coordTmp, rhsTmp); // add a store