[HLSL/Spir-V] fix for incorrect spir-v on int dot(int, int)
Decomposes OpDot into corresponding multiplies and additions.
diff --git a/SPIRV/GlslangToSpv.cpp b/SPIRV/GlslangToSpv.cpp
index 3569001..4a59ffe 100755
--- a/SPIRV/GlslangToSpv.cpp
+++ b/SPIRV/GlslangToSpv.cpp
@@ -6914,6 +6914,17 @@
// We might need the remaining arguments, e.g. in the EOpFrexp case.
std::vector<spv::Id> callArguments(operands.begin(), operands.begin() + consumedOperands);
id = builder.createBuiltinCall(typeId, extBuiltins >= 0 ? extBuiltins : stdBuiltins, libCall, callArguments);
+ } else if (opCode == spv::OpDot && !isFloat) {
+ // int dot(int, int)
+ // NOTE: never called for scalar/vector1, this is turned into simple mul before this can be reached
+ const int componentCount = builder.getNumComponents(operands[0]);
+ spv::Id mulOp = builder.createBinOp(spv::OpIMul, builder.getTypeId(operands[0]), operands[0], operands[1]);
+ builder.setPrecision(mulOp, precision);
+ id = builder.createCompositeExtract(mulOp, typeId, 0);
+ for (int i = 1; i < componentCount; ++i) {
+ builder.setPrecision(id, precision);
+ id = builder.createBinOp(spv::OpIAdd, typeId, id, builder.createCompositeExtract(operands[0], typeId, i));
+ }
} else {
switch (consumedOperands) {
case 0:
diff --git a/Test/baseResults/hlsl.int.dot.frag.out b/Test/baseResults/hlsl.int.dot.frag.out
new file mode 100644
index 0000000..afe44c8
--- /dev/null
+++ b/Test/baseResults/hlsl.int.dot.frag.out
@@ -0,0 +1,339 @@
+hlsl.int.dot.frag
+Shader version: 500
+gl_FragCoord origin is upper left
+0:? Sequence
+0:1 Function Definition: @main( ( temp 4-component vector of float)
+0:1 Function Parameters:
+0:? Sequence
+0:2 Sequence
+0:2 move second child to first child ( temp int)
+0:2 'i' ( temp int)
+0:2 Constant:
+0:2 1 (const int)
+0:3 Sequence
+0:3 move second child to first child ( temp 1-component vector of int)
+0:3 'i2' ( temp 1-component vector of int)
+0:3 Constant:
+0:3 2 (const int)
+0:4 Sequence
+0:4 move second child to first child ( temp 2-component vector of int)
+0:4 'i3' ( temp 2-component vector of int)
+0:4 Constant:
+0:4 3 (const int)
+0:4 3 (const int)
+0:5 Sequence
+0:5 move second child to first child ( temp 3-component vector of int)
+0:5 'i4' ( temp 3-component vector of int)
+0:5 Constant:
+0:5 4 (const int)
+0:5 4 (const int)
+0:5 4 (const int)
+0:6 Sequence
+0:6 move second child to first child ( temp 4-component vector of int)
+0:6 'i5' ( temp 4-component vector of int)
+0:6 Constant:
+0:6 5 (const int)
+0:6 5 (const int)
+0:6 5 (const int)
+0:6 5 (const int)
+0:8 move second child to first child ( temp int)
+0:8 'i' ( temp int)
+0:8 dot-product ( temp int)
+0:8 'i' ( temp int)
+0:8 'i' ( temp int)
+0:9 move second child to first child ( temp 1-component vector of int)
+0:9 'i2' ( temp 1-component vector of int)
+0:9 Construct int ( temp 1-component vector of int)
+0:9 dot-product ( temp int)
+0:9 Construct int ( in int)
+0:9 'i2' ( temp 1-component vector of int)
+0:9 Construct int ( in int)
+0:9 'i2' ( temp 1-component vector of int)
+0:10 move second child to first child ( temp 2-component vector of int)
+0:10 'i3' ( temp 2-component vector of int)
+0:10 Construct ivec2 ( temp 2-component vector of int)
+0:10 dot-product ( temp int)
+0:10 'i3' ( temp 2-component vector of int)
+0:10 'i3' ( temp 2-component vector of int)
+0:11 move second child to first child ( temp 3-component vector of int)
+0:11 'i4' ( temp 3-component vector of int)
+0:11 Construct ivec3 ( temp 3-component vector of int)
+0:11 dot-product ( temp int)
+0:11 'i4' ( temp 3-component vector of int)
+0:11 'i4' ( temp 3-component vector of int)
+0:12 move second child to first child ( temp 4-component vector of int)
+0:12 'i5' ( temp 4-component vector of int)
+0:12 Construct ivec4 ( temp 4-component vector of int)
+0:12 dot-product ( temp int)
+0:12 'i5' ( temp 4-component vector of int)
+0:12 'i5' ( temp 4-component vector of int)
+0:13 Branch: Return with expression
+0:13 Convert int to float ( temp 4-component vector of float)
+0:13 add ( temp 4-component vector of int)
+0:13 add ( temp 4-component vector of int)
+0:13 add ( temp 4-component vector of int)
+0:13 add ( temp 4-component vector of int)
+0:13 'i' ( temp int)
+0:13 Construct ivec4 ( temp 4-component vector of int)
+0:13 Construct int ( temp int)
+0:13 'i2' ( temp 1-component vector of int)
+0:13 vector swizzle ( temp 4-component vector of int)
+0:13 'i3' ( temp 2-component vector of int)
+0:13 Sequence
+0:13 Constant:
+0:13 0 (const int)
+0:13 Constant:
+0:13 1 (const int)
+0:13 Constant:
+0:13 0 (const int)
+0:13 Constant:
+0:13 1 (const int)
+0:13 vector swizzle ( temp 4-component vector of int)
+0:13 'i4' ( temp 3-component vector of int)
+0:13 Sequence
+0:13 Constant:
+0:13 0 (const int)
+0:13 Constant:
+0:13 1 (const int)
+0:13 Constant:
+0:13 2 (const int)
+0:13 Constant:
+0:13 0 (const int)
+0:13 'i5' ( temp 4-component vector of int)
+0:1 Function Definition: main( ( temp void)
+0:1 Function Parameters:
+0:? Sequence
+0:1 move second child to first child ( temp 4-component vector of float)
+0:? '@entryPointOutput' (layout( location=0) out 4-component vector of float)
+0:1 Function Call: @main( ( temp 4-component vector of float)
+0:? Linker Objects
+0:? '@entryPointOutput' (layout( location=0) out 4-component vector of float)
+
+
+Linked fragment stage:
+
+
+Shader version: 500
+gl_FragCoord origin is upper left
+0:? Sequence
+0:1 Function Definition: @main( ( temp 4-component vector of float)
+0:1 Function Parameters:
+0:? Sequence
+0:2 Sequence
+0:2 move second child to first child ( temp int)
+0:2 'i' ( temp int)
+0:2 Constant:
+0:2 1 (const int)
+0:3 Sequence
+0:3 move second child to first child ( temp 1-component vector of int)
+0:3 'i2' ( temp 1-component vector of int)
+0:3 Constant:
+0:3 2 (const int)
+0:4 Sequence
+0:4 move second child to first child ( temp 2-component vector of int)
+0:4 'i3' ( temp 2-component vector of int)
+0:4 Constant:
+0:4 3 (const int)
+0:4 3 (const int)
+0:5 Sequence
+0:5 move second child to first child ( temp 3-component vector of int)
+0:5 'i4' ( temp 3-component vector of int)
+0:5 Constant:
+0:5 4 (const int)
+0:5 4 (const int)
+0:5 4 (const int)
+0:6 Sequence
+0:6 move second child to first child ( temp 4-component vector of int)
+0:6 'i5' ( temp 4-component vector of int)
+0:6 Constant:
+0:6 5 (const int)
+0:6 5 (const int)
+0:6 5 (const int)
+0:6 5 (const int)
+0:8 move second child to first child ( temp int)
+0:8 'i' ( temp int)
+0:8 dot-product ( temp int)
+0:8 'i' ( temp int)
+0:8 'i' ( temp int)
+0:9 move second child to first child ( temp 1-component vector of int)
+0:9 'i2' ( temp 1-component vector of int)
+0:9 Construct int ( temp 1-component vector of int)
+0:9 dot-product ( temp int)
+0:9 Construct int ( in int)
+0:9 'i2' ( temp 1-component vector of int)
+0:9 Construct int ( in int)
+0:9 'i2' ( temp 1-component vector of int)
+0:10 move second child to first child ( temp 2-component vector of int)
+0:10 'i3' ( temp 2-component vector of int)
+0:10 Construct ivec2 ( temp 2-component vector of int)
+0:10 dot-product ( temp int)
+0:10 'i3' ( temp 2-component vector of int)
+0:10 'i3' ( temp 2-component vector of int)
+0:11 move second child to first child ( temp 3-component vector of int)
+0:11 'i4' ( temp 3-component vector of int)
+0:11 Construct ivec3 ( temp 3-component vector of int)
+0:11 dot-product ( temp int)
+0:11 'i4' ( temp 3-component vector of int)
+0:11 'i4' ( temp 3-component vector of int)
+0:12 move second child to first child ( temp 4-component vector of int)
+0:12 'i5' ( temp 4-component vector of int)
+0:12 Construct ivec4 ( temp 4-component vector of int)
+0:12 dot-product ( temp int)
+0:12 'i5' ( temp 4-component vector of int)
+0:12 'i5' ( temp 4-component vector of int)
+0:13 Branch: Return with expression
+0:13 Convert int to float ( temp 4-component vector of float)
+0:13 add ( temp 4-component vector of int)
+0:13 add ( temp 4-component vector of int)
+0:13 add ( temp 4-component vector of int)
+0:13 add ( temp 4-component vector of int)
+0:13 'i' ( temp int)
+0:13 Construct ivec4 ( temp 4-component vector of int)
+0:13 Construct int ( temp int)
+0:13 'i2' ( temp 1-component vector of int)
+0:13 vector swizzle ( temp 4-component vector of int)
+0:13 'i3' ( temp 2-component vector of int)
+0:13 Sequence
+0:13 Constant:
+0:13 0 (const int)
+0:13 Constant:
+0:13 1 (const int)
+0:13 Constant:
+0:13 0 (const int)
+0:13 Constant:
+0:13 1 (const int)
+0:13 vector swizzle ( temp 4-component vector of int)
+0:13 'i4' ( temp 3-component vector of int)
+0:13 Sequence
+0:13 Constant:
+0:13 0 (const int)
+0:13 Constant:
+0:13 1 (const int)
+0:13 Constant:
+0:13 2 (const int)
+0:13 Constant:
+0:13 0 (const int)
+0:13 'i5' ( temp 4-component vector of int)
+0:1 Function Definition: main( ( temp void)
+0:1 Function Parameters:
+0:? Sequence
+0:1 move second child to first child ( temp 4-component vector of float)
+0:? '@entryPointOutput' (layout( location=0) out 4-component vector of float)
+0:1 Function Call: @main( ( temp 4-component vector of float)
+0:? Linker Objects
+0:? '@entryPointOutput' (layout( location=0) out 4-component vector of float)
+
+// Module Version 10300
+// Generated by (magic number): 80007
+// Id's are bound by 84
+
+ Capability Shader
+ 1: ExtInstImport "GLSL.std.450"
+ MemoryModel Logical GLSL450
+ EntryPoint Fragment 4 "main" 82
+ ExecutionMode 4 OriginUpperLeft
+ Source HLSL 500
+ Name 4 "main"
+ Name 9 "@main("
+ Name 13 "i"
+ Name 15 "i2"
+ Name 19 "i3"
+ Name 24 "i4"
+ Name 29 "i5"
+ Name 82 "@entryPointOutput"
+ Decorate 82(@entryPointOutput) Location 0
+ 2: TypeVoid
+ 3: TypeFunction 2
+ 6: TypeFloat 32
+ 7: TypeVector 6(float) 4
+ 8: TypeFunction 7(fvec4)
+ 11: TypeInt 32 1
+ 12: TypePointer Function 11(int)
+ 14: 11(int) Constant 1
+ 16: 11(int) Constant 2
+ 17: TypeVector 11(int) 2
+ 18: TypePointer Function 17(ivec2)
+ 20: 11(int) Constant 3
+ 21: 17(ivec2) ConstantComposite 20 20
+ 22: TypeVector 11(int) 3
+ 23: TypePointer Function 22(ivec3)
+ 25: 11(int) Constant 4
+ 26: 22(ivec3) ConstantComposite 25 25 25
+ 27: TypeVector 11(int) 4
+ 28: TypePointer Function 27(ivec4)
+ 30: 11(int) Constant 5
+ 31: 27(ivec4) ConstantComposite 30 30 30 30
+ 81: TypePointer Output 7(fvec4)
+82(@entryPointOutput): 81(ptr) Variable Output
+ 4(main): 2 Function None 3
+ 5: Label
+ 83: 7(fvec4) FunctionCall 9(@main()
+ Store 82(@entryPointOutput) 83
+ Return
+ FunctionEnd
+ 9(@main(): 7(fvec4) Function None 8
+ 10: Label
+ 13(i): 12(ptr) Variable Function
+ 15(i2): 12(ptr) Variable Function
+ 19(i3): 18(ptr) Variable Function
+ 24(i4): 23(ptr) Variable Function
+ 29(i5): 28(ptr) Variable Function
+ Store 13(i) 14
+ Store 15(i2) 16
+ Store 19(i3) 21
+ Store 24(i4) 26
+ Store 29(i5) 31
+ 32: 11(int) Load 13(i)
+ 33: 11(int) Load 13(i)
+ 34: 11(int) IMul 32 33
+ Store 13(i) 34
+ 35: 11(int) Load 15(i2)
+ 36: 11(int) Load 15(i2)
+ 37: 11(int) IMul 35 36
+ Store 15(i2) 37
+ 38: 17(ivec2) Load 19(i3)
+ 39: 17(ivec2) Load 19(i3)
+ 40: 17(ivec2) IMul 38 39
+ 41: 11(int) CompositeExtract 40 0
+ 42: 11(int) CompositeExtract 38 1
+ 43: 11(int) IAdd 41 42
+ 44: 17(ivec2) CompositeConstruct 43 43
+ Store 19(i3) 44
+ 45: 22(ivec3) Load 24(i4)
+ 46: 22(ivec3) Load 24(i4)
+ 47: 22(ivec3) IMul 45 46
+ 48: 11(int) CompositeExtract 47 0
+ 49: 11(int) CompositeExtract 45 1
+ 50: 11(int) IAdd 48 49
+ 51: 11(int) CompositeExtract 45 2
+ 52: 11(int) IAdd 50 51
+ 53: 22(ivec3) CompositeConstruct 52 52 52
+ Store 24(i4) 53
+ 54: 27(ivec4) Load 29(i5)
+ 55: 27(ivec4) Load 29(i5)
+ 56: 27(ivec4) IMul 54 55
+ 57: 11(int) CompositeExtract 56 0
+ 58: 11(int) CompositeExtract 54 1
+ 59: 11(int) IAdd 57 58
+ 60: 11(int) CompositeExtract 54 2
+ 61: 11(int) IAdd 59 60
+ 62: 11(int) CompositeExtract 54 3
+ 63: 11(int) IAdd 61 62
+ 64: 27(ivec4) CompositeConstruct 63 63 63 63
+ Store 29(i5) 64
+ 65: 11(int) Load 13(i)
+ 66: 11(int) Load 15(i2)
+ 67: 27(ivec4) CompositeConstruct 66 66 66 66
+ 68: 27(ivec4) CompositeConstruct 65 65 65 65
+ 69: 27(ivec4) IAdd 68 67
+ 70: 17(ivec2) Load 19(i3)
+ 71: 27(ivec4) VectorShuffle 70 70 0 1 0 1
+ 72: 27(ivec4) IAdd 69 71
+ 73: 22(ivec3) Load 24(i4)
+ 74: 27(ivec4) VectorShuffle 73 73 0 1 2 0
+ 75: 27(ivec4) IAdd 72 74
+ 76: 27(ivec4) Load 29(i5)
+ 77: 27(ivec4) IAdd 75 76
+ 78: 7(fvec4) ConvertSToF 77
+ ReturnValue 78
+ FunctionEnd
diff --git a/Test/hlsl.int.dot.frag b/Test/hlsl.int.dot.frag
new file mode 100644
index 0000000..c293dc1
--- /dev/null
+++ b/Test/hlsl.int.dot.frag
@@ -0,0 +1,14 @@
+float4 main() : SV_Target {
+ int i = 1;
+ int1 i2 = 2;
+ int2 i3 = 3;
+ int3 i4 = 4;
+ int4 i5 = 5;
+
+ i = dot(i, i);
+ i2 = dot(i2, i2);
+ i3 = dot(i3, i3);
+ i4 = dot(i4, i4);
+ i5 = dot(i5, i5);
+ return i + i2.xxxx + i3.xyxy + i4.xyzx + i5;
+}
\ No newline at end of file
diff --git a/gtests/Hlsl.FromFile.cpp b/gtests/Hlsl.FromFile.cpp
index a490ba0..809e525 100644
--- a/gtests/Hlsl.FromFile.cpp
+++ b/gtests/Hlsl.FromFile.cpp
@@ -400,7 +400,8 @@
{"hlsl.wavequery.frag", "PixelShaderFunction"},
{"hlsl.wavereduction.comp", "CSMain"},
{"hlsl.wavevote.comp", "CSMain"},
- { "hlsl.type.type.conversion.valid.frag", "main" }
+ { "hlsl.type.type.conversion.valid.frag", "main" },
+ {"hlsl.int.dot.frag", "main"}
}),
FileNameAsCustomTestSuffix
);
