AuroraMiddleware/glslang
1
0
Fork 0
mirror of https://github.com/KhronosGroup/glslang synced 2024-09-19 12:19:53 +00:00
Code Issues Packages Projects Releases Wiki Activity

HLSL: Include shape-changing conversions in overloaded signature selection.

Browse Source 
This also enables vecN -> vec1 shape conversions for all places doing shape
conversions.

For signature selection, makes shape changes worse than any other comparison
when deciding what conversions are better than others.
This commit is contained in:
John Kessenich 2016-08-25 15:57:56 -06:00
parent 90dd70f752
commit e3f2c8f98a
6 changed files with 1504 additions and 1026 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

97
Test/baseResults/hlsl.intrinsics.negative.frag.out
View File

@ -1,12 +1,6 @@
hlsl.intrinsics.negative.frag hlsl.intrinsics.negative.frag
ERROR: 0:8: 'cross' : no matching overloaded function found
ERROR: 0:9: 'D3DCOLORtoUBYTE4' : no matching overloaded function found
ERROR: 0:10: 'determinant' : no matching overloaded function found ERROR: 0:10: 'determinant' : no matching overloaded function found
ERROR: 0:12: 'f32tof16' : unimplemented intrinsic: handle natively ERROR: 0:12: 'f32tof16' : unimplemented intrinsic: handle natively
ERROR: 0:23: 'length' : no matching overloaded function found
ERROR: 0:25: 'normalize' : no matching overloaded function found
ERROR: 0:26: 'reflect' : no matching overloaded function found
ERROR: 0:27: 'refract' : no matching overloaded function found
ERROR: 0:28: 'refract' : no matching overloaded function found ERROR: 0:28: 'refract' : no matching overloaded function found
ERROR: 0:30: 'transpose' : no matching overloaded function found ERROR: 0:30: 'transpose' : no matching overloaded function found
ERROR: 0:39: 'GetRenderTargetSamplePosition' : no matching overloaded function found ERROR: 0:39: 'GetRenderTargetSamplePosition' : no matching overloaded function found
@ -23,7 +17,6 @@ ERROR: 0:67: 'determinant' : no matching overloaded function found
ERROR: 0:68: 'f32tof16' : unimplemented intrinsic: handle natively ERROR: 0:68: 'f32tof16' : unimplemented intrinsic: handle natively
ERROR: 0:73: 'transpose' : no matching overloaded function found ERROR: 0:73: 'transpose' : no matching overloaded function found
ERROR: 0:81: 'CheckAccessFullyMapped' : no matching overloaded function found ERROR: 0:81: 'CheckAccessFullyMapped' : no matching overloaded function found
ERROR: 0:83: 'cross' : no matching overloaded function found
ERROR: 0:84: 'determinant' : no matching overloaded function found ERROR: 0:84: 'determinant' : no matching overloaded function found
ERROR: 0:85: 'f32tof16' : unimplemented intrinsic: handle natively ERROR: 0:85: 'f32tof16' : unimplemented intrinsic: handle natively
ERROR: 0:90: 'transpose' : no matching overloaded function found ERROR: 0:90: 'transpose' : no matching overloaded function found
@ -66,7 +59,7 @@ ERROR: 0:133: 'normalize' : no matching overloaded function found
ERROR: 0:133: 'reflect' : no matching overloaded function found ERROR: 0:133: 'reflect' : no matching overloaded function found
ERROR: 0:133: 'refract' : no matching overloaded function found ERROR: 0:133: 'refract' : no matching overloaded function found
ERROR: 0:133: 'reversebits' : no matching overloaded function found ERROR: 0:133: 'reversebits' : no matching overloaded function found
ERROR: 67 compilation errors. No code generated. ERROR: 60 compilation errors. No code generated.
Shader version: 450 Shader version: 450
@ -91,10 +84,14 @@ ERROR: node is still EOpNull!
0:7 bitCount (global uint) 0:7 bitCount (global uint)
0:7 Convert float to uint (temp uint) 0:7 Convert float to uint (temp uint)
0:7 'inF0' (in float) 0:7 'inF0' (in float)
0:8 Constant: 0:8 cross-product (global 3-component vector of float)
0:8 0.000000 0:8 Construct vec3 (in 3-component vector of float)
0:9 Constant: 0:8 'inF0' (in float)
0:9 0.000000 0:8 Construct vec3 (in 3-component vector of float)
0:8 'inF1' (in float)
0:9 Function Call: D3DCOLORtoUBYTE4(vf4; (global 4-component vector of int)
0:9 Construct vec4 (in 4-component vector of float)
0:9 'inF0' (in float)
0:10 Constant: 0:10 Constant:
0:10 0.000000 0:10 0.000000
0:12 ERROR: Bad unary op 0:12 ERROR: Bad unary op
@ -107,8 +104,9 @@ ERROR: node is still EOpNull!
0:14 findLSB (global uint) 0:14 findLSB (global uint)
0:14 Convert float to uint (temp uint) 0:14 Convert float to uint (temp uint)
0:14 'inF0' (in float) 0:14 'inF0' (in float)
0:23 Constant: 0:23 length (global float)
0:23 0.000000 0:23 Construct vec2 (in 2-component vector of float)
0:23 'inF0' (in float)
0:24 Function Call: msad4(u1;vu2;vu4; (global 4-component vector of uint) 0:24 Function Call: msad4(u1;vu2;vu4; (global 4-component vector of uint)
0:24 Convert float to uint (temp uint) 0:24 Convert float to uint (temp uint)
0:24 'inF0' (in float) 0:24 'inF0' (in float)
@ -120,12 +118,20 @@ ERROR: node is still EOpNull!
0:24 0 (const uint) 0:24 0 (const uint)
0:24 0 (const uint) 0:24 0 (const uint)
0:24 0 (const uint) 0:24 0 (const uint)
0:25 Constant: 0:25 normalize (global 2-component vector of float)
0:25 0.000000 0:25 Construct vec2 (in 2-component vector of float)
0:26 Constant: 0:25 'inF0' (in float)
0:26 0.000000 0:26 reflect (global 2-component vector of float)
0:27 Constant: 0:26 Construct vec2 (in 2-component vector of float)
0:27 0.000000 0:26 'inF0' (in float)
0:26 Construct vec2 (in 2-component vector of float)
0:26 'inF1' (in float)
0:27 refract (global 2-component vector of float)
0:27 Construct vec2 (in 2-component vector of float)
0:27 'inF0' (in float)
0:27 Construct vec2 (in 2-component vector of float)
0:27 'inF1' (in float)
0:27 'inF2' (in float)
0:28 Constant: 0:28 Constant:
0:28 0.000000 0:28 0.000000
0:29 bitFieldReverse (global uint) 0:29 bitFieldReverse (global uint)
@ -239,8 +245,11 @@ ERROR: node is still EOpNull!
0:82 bitCount (global 4-component vector of uint) 0:82 bitCount (global 4-component vector of uint)
0:82 Convert float to uint (temp 4-component vector of uint) 0:82 Convert float to uint (temp 4-component vector of uint)
0:82 'inF0' (in 4-component vector of float) 0:82 'inF0' (in 4-component vector of float)
0:83 Constant: 0:83 cross-product (global 3-component vector of float)
0:83 0.000000 0:83 Construct vec3 (in 3-component vector of float)
0:83 'inF0' (in 4-component vector of float)
0:83 Construct vec3 (in 3-component vector of float)
0:83 'inF1' (in 4-component vector of float)
0:84 Constant: 0:84 Constant:
0:84 0.000000 0:84 0.000000
0:85 ERROR: Bad unary op 0:85 ERROR: Bad unary op
@ -423,10 +432,14 @@ ERROR: node is still EOpNull!
0:7 bitCount (global uint) 0:7 bitCount (global uint)
0:7 Convert float to uint (temp uint) 0:7 Convert float to uint (temp uint)
0:7 'inF0' (in float) 0:7 'inF0' (in float)
0:8 Constant: 0:8 cross-product (global 3-component vector of float)
0:8 0.000000 0:8 Construct vec3 (in 3-component vector of float)
0:9 Constant: 0:8 'inF0' (in float)
0:9 0.000000 0:8 Construct vec3 (in 3-component vector of float)
0:8 'inF1' (in float)
0:9 Function Call: D3DCOLORtoUBYTE4(vf4; (global 4-component vector of int)
0:9 Construct vec4 (in 4-component vector of float)
0:9 'inF0' (in float)
0:10 Constant: 0:10 Constant:
0:10 0.000000 0:10 0.000000
0:12 ERROR: Bad unary op 0:12 ERROR: Bad unary op
@ -439,8 +452,9 @@ ERROR: node is still EOpNull!
0:14 findLSB (global uint) 0:14 findLSB (global uint)
0:14 Convert float to uint (temp uint) 0:14 Convert float to uint (temp uint)
0:14 'inF0' (in float) 0:14 'inF0' (in float)
0:23 Constant: 0:23 length (global float)
0:23 0.000000 0:23 Construct vec2 (in 2-component vector of float)
0:23 'inF0' (in float)
0:24 Function Call: msad4(u1;vu2;vu4; (global 4-component vector of uint) 0:24 Function Call: msad4(u1;vu2;vu4; (global 4-component vector of uint)
0:24 Convert float to uint (temp uint) 0:24 Convert float to uint (temp uint)
0:24 'inF0' (in float) 0:24 'inF0' (in float)
@ -452,12 +466,20 @@ ERROR: node is still EOpNull!
0:24 0 (const uint) 0:24 0 (const uint)
0:24 0 (const uint) 0:24 0 (const uint)
0:24 0 (const uint) 0:24 0 (const uint)
0:25 Constant: 0:25 normalize (global 2-component vector of float)
0:25 0.000000 0:25 Construct vec2 (in 2-component vector of float)
0:26 Constant: 0:25 'inF0' (in float)
0:26 0.000000 0:26 reflect (global 2-component vector of float)
0:27 Constant: 0:26 Construct vec2 (in 2-component vector of float)
0:27 0.000000 0:26 'inF0' (in float)
0:26 Construct vec2 (in 2-component vector of float)
0:26 'inF1' (in float)
0:27 refract (global 2-component vector of float)
0:27 Construct vec2 (in 2-component vector of float)
0:27 'inF0' (in float)
0:27 Construct vec2 (in 2-component vector of float)
0:27 'inF1' (in float)
0:27 'inF2' (in float)
0:28 Constant: 0:28 Constant:
0:28 0.000000 0:28 0.000000
0:29 bitFieldReverse (global uint) 0:29 bitFieldReverse (global uint)
@ -571,8 +593,11 @@ ERROR: node is still EOpNull!
0:82 bitCount (global 4-component vector of uint) 0:82 bitCount (global 4-component vector of uint)
0:82 Convert float to uint (temp 4-component vector of uint) 0:82 Convert float to uint (temp 4-component vector of uint)
0:82 'inF0' (in 4-component vector of float) 0:82 'inF0' (in 4-component vector of float)
0:83 Constant: 0:83 cross-product (global 3-component vector of float)
0:83 0.000000 0:83 Construct vec3 (in 3-component vector of float)
0:83 'inF0' (in 4-component vector of float)
0:83 Construct vec3 (in 3-component vector of float)
0:83 'inF1' (in 4-component vector of float)
0:84 Constant: 0:84 Constant:
0:84 0.000000 0:84 0.000000
0:85 ERROR: Bad unary op 0:85 ERROR: Bad unary op

2119
Test/baseResults/hlsl.overload.frag.out
View File

File diff suppressed because it is too large Load Diff

227
Test/hlsl.overload.frag
View File

@ -1,117 +1,142 @@
// function selection under type conversion // function selection under type conversion
void foo1(double a, bool b) {} void foo1(double a, bool b) {}
void foo1(double a, uint b) {} void foo1(double a, uint b) {}
void foo1(double a, int b) {} void foo1(double a, int b) {}
void foo1(double a, float b) {} void foo1(double a, float b) {}
void foo1(double a, double b){} void foo1(double a, double b){}
// uint -> int // uint -> int
void foo2(int a, bool b) {} void foo2(int a, bool b) {}
void foo2(int a, uint b) {} void foo2(int a, uint b) {}
void foo2(int a, int b) {} void foo2(int a, int b) {}
void foo2(int a, float b) {} void foo2(int a, float b) {}
void foo2(int a, double b){} void foo2(int a, double b){}
// everything can promote // everything can promote
void foo3(bool b) {} void foo3(bool b) {}
void foo4(uint b) {} void foo4(uint b) {}
void foo5(int b) {} void foo5(int b) {}
void foo6(float b) {} void foo6(float b) {}
void foo7(double b){} void foo7(double b){}
// shorter forward chain better than longer or backward chain // shorter forward chain better than longer or backward chain
void foo8(float); void foo8(float) {}
void foo8(double); void foo8(double) {}
void foo9(int); void foo9(int) {}
void foo9(uint); void foo9(uint) {}
void foo10(bool); void foo10(bool) {}
void foo10(int); void foo10(int) {}
// shape change is worse
void foo11(float3) {}
void foo11(double) {}
void foo11(int3) {}
void foo11(uint) {}
void foo12(float1) {}
void foo12(double3) {}
void foo16(uint) {}
void foo16(uint2) {}
// shape change
void foo13(float3) {}
void foo14(int1) {}
void foo15(bool1) {}
float4 PixelShaderFunction(float4 input) : COLOR0 float4 PixelShaderFunction(float4 input) : COLOR0
{ {
bool b; bool b;
double d; double d;
uint u; uint u;
int i; int i;
float f; float f;
foo1(d, b);
foo1(d, d);
foo1(d, u);
foo1(d, i);
foo1(d, f);
foo1(f, b);
foo1(f, d);
foo1(f, u);
foo1(f, i);
foo1(f, f);
foo1(u, b);
foo1(u, d);
foo1(u, u);
foo1(u, i);
foo1(u, f);
foo1(i, b);
foo1(i, d);
foo1(i, u);
foo1(i, i);
foo1(i, f);
foo2(u, b);
foo2(u, d);
foo2(u, u);
foo2(u, i);
foo2(u, f);
foo2(i, b);
foo2(i, d);
foo2(i, u);
foo2(i, i);
foo2(i, f);
foo3(b);
foo3(d);
foo3(u);
foo3(i);
foo3(f);
foo4(b);
foo4(d);
foo4(u);
foo4(i);
foo4(f);
foo5(b);
foo5(d);
foo5(u);
foo5(i);
foo5(f);
foo6(b);
foo6(d);
foo6(u);
foo6(i);
foo6(f);
foo7(b);
foo7(d);
foo7(u);
foo7(i);
foo7(f);
foo8(b);
foo8(u);
foo8(i);
foo9(b);
foo9(f);
foo9(d);
foo10(u);
foo10(f);
foo10(d);
foo11(b);
foo11(f);
foo12(float3(f));
foo16(int2(i,i));
foo13(f);
foo14(int4(i));
foo15(b);
foo15(bool3(b));
foo1(d, b);
foo1(d, d);
foo1(d, u);
foo1(d, i);
foo1(d, f);
foo1(f, b);
foo1(f, d);
foo1(f, u);
foo1(f, i);
foo1(f, f);
foo1(u, b);
foo1(u, d);
foo1(u, u);
foo1(u, i);
foo1(u, f);
foo1(i, b);
foo1(i, d);
foo1(i, u);
foo1(i, i);
foo1(i, f);
foo2(u, b);
foo2(u, d);
foo2(u, u);
foo2(u, i);
foo2(u, f);
foo2(i, b);
foo2(i, d);
foo2(i, u);
foo2(i, i);
foo2(i, f);
foo3(b);
foo3(d);
foo3(u);
foo3(i);
foo3(f);
foo4(b);
foo4(d);
foo4(u);
foo4(i);
foo4(f);
foo5(b);
foo5(d);
foo5(u);
foo5(i);
foo5(f);
foo6(b);
foo6(d);
foo6(u);
foo6(i);
foo6(f);
foo7(b);
foo7(d);
foo7(u);
foo7(i);
foo7(f);
foo8(b);
foo8(u);
foo8(i);
foo9(b);
foo9(f);
foo9(d);
foo10(u);
foo10(f);
foo10(d);
return input; return input;
} }

2
glslang/Include/revision.h
View File

@ -2,5 +2,5 @@
// For the version, it uses the latest git tag followed by the number of commits. // For the version, it uses the latest git tag followed by the number of commits.
// For the date, it uses the current date (when then script is run). // For the date, it uses the current date (when then script is run).
#define GLSLANG_REVISION "Overload400-PrecQual.1434" #define GLSLANG_REVISION "Overload400-PrecQual.1438"
#define GLSLANG_DATE "25-Aug-2016" #define GLSLANG_DATE "25-Aug-2016"

10
glslang/MachineIndependent/Intermediate.cpp
View File

@ -702,6 +702,7 @@ TIntermTyped* TIntermediate::addShapeConversion(TOperator op, const TType& type,
case EOpGreaterThan: case EOpGreaterThan:
case EOpLessThanEqual: case EOpLessThanEqual:
case EOpGreaterThanEqual: case EOpGreaterThanEqual:
case EOpFunctionCall:
break; break;
default: default:
return node; return node;
@ -715,9 +716,11 @@ TIntermTyped* TIntermediate::addShapeConversion(TOperator op, const TType& type,
// The new node that handles the conversion // The new node that handles the conversion
TOperator constructorOp = mapTypeToConstructorOp(type); TOperator constructorOp = mapTypeToConstructorOp(type);
// scalar -> smeared -> vector // scalar -> smeared -> vector, or
if (type.isVector() && node->getType().isScalar()) // bigger vector -> smaller vector or scalar
return setAggregateOperator(node, constructorOp, type, node->getLoc()); if ((type.isVector() && node->getType().isScalar()) ||
(node->getVectorSize() > type.getVectorSize() && type.isVector()))
return setAggregateOperator(makeAggregate(node), constructorOp, type, node->getLoc());
return node; return node;
} }
@ -731,6 +734,7 @@ bool TIntermediate::canImplicitlyPromote(TBasicType from, TBasicType to, TOperat
if (profile == EEsProfile || version == 110) if (profile == EEsProfile || version == 110)
return false; return false;
// TODO: Move more policies into language-specific handlers.
// Some languages allow more general (or potentially, more specific) conversions under some conditions. // Some languages allow more general (or potentially, more specific) conversions under some conditions.
if (source == EShSourceHlsl) { if (source == EShSourceHlsl) {
const bool fromConvertable = (from == EbtFloat || from == EbtDouble || from == EbtInt || from == EbtUint || from == EbtBool); const bool fromConvertable = (from == EbtFloat || from == EbtDouble || from == EbtInt || from == EbtUint || from == EbtBool);

75
hlsl/hlslParseHelper.cpp
View File

@ -2001,6 +2001,7 @@ void HlslParseContext::addInputArgumentConversions(const TFunction& function, TI
// In-qualified arguments just need an extra node added above the argument to // In-qualified arguments just need an extra node added above the argument to
// convert to the correct type. // convert to the correct type.
arg = intermediate.addConversion(EOpFunctionCall, *function[i].type, arg); arg = intermediate.addConversion(EOpFunctionCall, *function[i].type, arg);
arg = intermediate.addShapeConversion(EOpFunctionCall, *function[i].type, arg);
if (arg) { if (arg) {
if (function.getParamCount() == 1) if (function.getParamCount() == 1)
arguments = arg; arguments = arg;
@ -3565,9 +3566,25 @@ const TFunction* HlslParseContext::findFunction(const TSourceLoc& loc, const TFu
auto convertible = [this](const TType& from, const TType& to) { auto convertible = [this](const TType& from, const TType& to) {
if (from == to) if (from == to)
return true; return true;
if (from.isArray() || to.isArray() || ! from.sameElementShape(to))
// no aggregate conversions
if (from.isArray() || to.isArray() ||
from.isStruct() || to.isStruct())
return false; return false;
return intermediate.canImplicitlyPromote(from.getBasicType(), to.getBasicType(), EOpFunctionCall);
// basic types have to be convertible
if (! intermediate.canImplicitlyPromote(from.getBasicType(), to.getBasicType(), EOpFunctionCall))
return false;
// shapes have to be convertible
if ((from.isScalar() && to.isScalar()) ||
(from.isScalar() && to.isVector()) ||
(from.isVector() && to.isVector() && from.getVectorSize() >= to.getVectorSize()))
return true;
// TODO: what are the matrix rules? they go here
return false;
}; };
// Is 'to2' a better conversion than 'to1'? // Is 'to2' a better conversion than 'to1'?
@ -3580,33 +3597,41 @@ const TFunction* HlslParseContext::findFunction(const TSourceLoc& loc, const TFu
if (from == to1) if (from == to1)
return false; return false;
// float -> double is better than any other float conversion // shape changes are always worse
if (from.getBasicType() == EbtFloat) { if (from.isScalar() || from.isVector()) {
if (to2.getBasicType() == EbtDouble && to1.getBasicType() != EbtDouble) if (from.getVectorSize() == to2.getVectorSize() &&
from.getVectorSize() != to1.getVectorSize())
return true; return true;
if (from.getVectorSize() == to1.getVectorSize() &&
from.getVectorSize() != to2.getVectorSize())
return false;
} }
// int -> uint is better than any other int conversion // Might or might not be changing shape, which means basic type might
if (from.getBasicType() == EbtInt) { // or might not match, so within that, the question is how big a
if (to2.getBasicType() == EbtUint && to1.getBasicType() != EbtUint) // basic-type conversion is being done.
return true; //
} // Use a hierarchy of domains, translated to order of magnitude
// in a linearized view:
// - floating-point vs. integer
// - 32 vs. 64 bit (or width in general)
// - bool vs. non bool
// - signed vs. not signed
auto linearize = [](const TBasicType& basicType) {
switch (basicType) {
case EbtBool: return 1;
case EbtInt: return 10;
case EbtUint: return 11;
case EbtInt64: return 20;
case EbtUint64: return 21;
case EbtFloat: return 100;
case EbtDouble: return 110;
default: return 0;
}
};
// TODO: these should be replaced by a more generic "shorter chain is better than longer chain" rule return std::abs(linearize(to2.getBasicType()) - linearize(from.getBasicType())) <
std::abs(linearize(to1.getBasicType()) - linearize(from.getBasicType()));
// -> float is better than -> double
if (to2.getBasicType() == EbtFloat && to1.getBasicType() == EbtDouble)
return true;
// -> int is better than -> bool
if ((to2.getBasicType() == EbtInt || to2.getBasicType() == EbtUint) && to1.getBasicType() == EbtBool)
return true;
// -> uint is better than -> int
if (to2.getBasicType() == EbtUint && to1.getBasicType() == EbtInt)
return true;
return false;
}; };
// for ambiguity reporting // for ambiguity reporting