mirror of
https://github.com/KhronosGroup/glslang
synced 2024-11-09 12:00:05 +00:00
HLSL: Include shape-changing conversions in overloaded signature selection.
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
parent
90dd70f752
commit
e3f2c8f98a
@ -1,12 +1,6 @@
|
||||
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: 0:12: '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: 0:30: 'transpose' : 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: 0:73: 'transpose' : 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: 0:85: 'f32tof16' : unimplemented intrinsic: handle natively
|
||||
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: 'refract' : 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
|
||||
@ -91,10 +84,14 @@ ERROR: node is still EOpNull!
|
||||
0:7 bitCount (global uint)
|
||||
0:7 Convert float to uint (temp uint)
|
||||
0:7 'inF0' (in float)
|
||||
0:8 Constant:
|
||||
0:8 0.000000
|
||||
0:9 Constant:
|
||||
0:9 0.000000
|
||||
0:8 cross-product (global 3-component vector of float)
|
||||
0:8 Construct vec3 (in 3-component vector of float)
|
||||
0:8 'inF0' (in float)
|
||||
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 0.000000
|
||||
0:12 ERROR: Bad unary op
|
||||
@ -107,8 +104,9 @@ ERROR: node is still EOpNull!
|
||||
0:14 findLSB (global uint)
|
||||
0:14 Convert float to uint (temp uint)
|
||||
0:14 'inF0' (in float)
|
||||
0:23 Constant:
|
||||
0:23 0.000000
|
||||
0:23 length (global float)
|
||||
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 Convert float to uint (temp uint)
|
||||
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)
|
||||
0:25 Constant:
|
||||
0:25 0.000000
|
||||
0:26 Constant:
|
||||
0:26 0.000000
|
||||
0:27 Constant:
|
||||
0:27 0.000000
|
||||
0:25 normalize (global 2-component vector of float)
|
||||
0:25 Construct vec2 (in 2-component vector of float)
|
||||
0:25 'inF0' (in float)
|
||||
0:26 reflect (global 2-component vector of float)
|
||||
0:26 Construct vec2 (in 2-component vector of float)
|
||||
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 0.000000
|
||||
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 Convert float to uint (temp 4-component vector of uint)
|
||||
0:82 'inF0' (in 4-component vector of float)
|
||||
0:83 Constant:
|
||||
0:83 0.000000
|
||||
0:83 cross-product (global 3-component vector of float)
|
||||
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 0.000000
|
||||
0:85 ERROR: Bad unary op
|
||||
@ -423,10 +432,14 @@ ERROR: node is still EOpNull!
|
||||
0:7 bitCount (global uint)
|
||||
0:7 Convert float to uint (temp uint)
|
||||
0:7 'inF0' (in float)
|
||||
0:8 Constant:
|
||||
0:8 0.000000
|
||||
0:9 Constant:
|
||||
0:9 0.000000
|
||||
0:8 cross-product (global 3-component vector of float)
|
||||
0:8 Construct vec3 (in 3-component vector of float)
|
||||
0:8 'inF0' (in float)
|
||||
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 0.000000
|
||||
0:12 ERROR: Bad unary op
|
||||
@ -439,8 +452,9 @@ ERROR: node is still EOpNull!
|
||||
0:14 findLSB (global uint)
|
||||
0:14 Convert float to uint (temp uint)
|
||||
0:14 'inF0' (in float)
|
||||
0:23 Constant:
|
||||
0:23 0.000000
|
||||
0:23 length (global float)
|
||||
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 Convert float to uint (temp uint)
|
||||
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)
|
||||
0:25 Constant:
|
||||
0:25 0.000000
|
||||
0:26 Constant:
|
||||
0:26 0.000000
|
||||
0:27 Constant:
|
||||
0:27 0.000000
|
||||
0:25 normalize (global 2-component vector of float)
|
||||
0:25 Construct vec2 (in 2-component vector of float)
|
||||
0:25 'inF0' (in float)
|
||||
0:26 reflect (global 2-component vector of float)
|
||||
0:26 Construct vec2 (in 2-component vector of float)
|
||||
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 0.000000
|
||||
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 Convert float to uint (temp 4-component vector of uint)
|
||||
0:82 'inF0' (in 4-component vector of float)
|
||||
0:83 Constant:
|
||||
0:83 0.000000
|
||||
0:83 cross-product (global 3-component vector of float)
|
||||
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 0.000000
|
||||
0:85 ERROR: Bad unary op
|
||||
|
||||
File diff suppressed because it is too large
Load Diff
@ -20,12 +20,27 @@ void foo6(float b) {}
|
||||
void foo7(double b){}
|
||||
|
||||
// shorter forward chain better than longer or backward chain
|
||||
void foo8(float);
|
||||
void foo8(double);
|
||||
void foo9(int);
|
||||
void foo9(uint);
|
||||
void foo10(bool);
|
||||
void foo10(int);
|
||||
void foo8(float) {}
|
||||
void foo8(double) {}
|
||||
void foo9(int) {}
|
||||
void foo9(uint) {}
|
||||
void foo10(bool) {}
|
||||
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
|
||||
{
|
||||
@ -113,5 +128,15 @@ float4 PixelShaderFunction(float4 input) : COLOR0
|
||||
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));
|
||||
|
||||
return input;
|
||||
}
|
||||
|
||||
@ -2,5 +2,5 @@
|
||||
// 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).
|
||||
|
||||
#define GLSLANG_REVISION "Overload400-PrecQual.1434"
|
||||
#define GLSLANG_REVISION "Overload400-PrecQual.1438"
|
||||
#define GLSLANG_DATE "25-Aug-2016"
|
||||
|
||||
@ -702,6 +702,7 @@ TIntermTyped* TIntermediate::addShapeConversion(TOperator op, const TType& type,
|
||||
case EOpGreaterThan:
|
||||
case EOpLessThanEqual:
|
||||
case EOpGreaterThanEqual:
|
||||
case EOpFunctionCall:
|
||||
break;
|
||||
default:
|
||||
return node;
|
||||
@ -715,9 +716,11 @@ TIntermTyped* TIntermediate::addShapeConversion(TOperator op, const TType& type,
|
||||
// The new node that handles the conversion
|
||||
TOperator constructorOp = mapTypeToConstructorOp(type);
|
||||
|
||||
// scalar -> smeared -> vector
|
||||
if (type.isVector() && node->getType().isScalar())
|
||||
return setAggregateOperator(node, constructorOp, type, node->getLoc());
|
||||
// scalar -> smeared -> vector, or
|
||||
// bigger vector -> smaller vector or scalar
|
||||
if ((type.isVector() && node->getType().isScalar()) ||
|
||||
(node->getVectorSize() > type.getVectorSize() && type.isVector()))
|
||||
return setAggregateOperator(makeAggregate(node), constructorOp, type, node->getLoc());
|
||||
|
||||
return node;
|
||||
}
|
||||
@ -731,6 +734,7 @@ bool TIntermediate::canImplicitlyPromote(TBasicType from, TBasicType to, TOperat
|
||||
if (profile == EEsProfile || version == 110)
|
||||
return false;
|
||||
|
||||
// TODO: Move more policies into language-specific handlers.
|
||||
// Some languages allow more general (or potentially, more specific) conversions under some conditions.
|
||||
if (source == EShSourceHlsl) {
|
||||
const bool fromConvertable = (from == EbtFloat || from == EbtDouble || from == EbtInt || from == EbtUint || from == EbtBool);
|
||||
|
||||
@ -2001,6 +2001,7 @@ void HlslParseContext::addInputArgumentConversions(const TFunction& function, TI
|
||||
// In-qualified arguments just need an extra node added above the argument to
|
||||
// convert to the correct type.
|
||||
arg = intermediate.addConversion(EOpFunctionCall, *function[i].type, arg);
|
||||
arg = intermediate.addShapeConversion(EOpFunctionCall, *function[i].type, arg);
|
||||
if (arg) {
|
||||
if (function.getParamCount() == 1)
|
||||
arguments = arg;
|
||||
@ -3565,9 +3566,25 @@ const TFunction* HlslParseContext::findFunction(const TSourceLoc& loc, const TFu
|
||||
auto convertible = [this](const TType& from, const TType& to) {
|
||||
if (from == to)
|
||||
return true;
|
||||
if (from.isArray() || to.isArray() || ! from.sameElementShape(to))
|
||||
|
||||
// no aggregate conversions
|
||||
if (from.isArray() || to.isArray() ||
|
||||
from.isStruct() || to.isStruct())
|
||||
return false;
|
||||
|
||||
// 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;
|
||||
return intermediate.canImplicitlyPromote(from.getBasicType(), to.getBasicType(), EOpFunctionCall);
|
||||
};
|
||||
|
||||
// Is 'to2' a better conversion than 'to1'?
|
||||
@ -3580,33 +3597,41 @@ const TFunction* HlslParseContext::findFunction(const TSourceLoc& loc, const TFu
|
||||
if (from == to1)
|
||||
return false;
|
||||
|
||||
// float -> double is better than any other float conversion
|
||||
if (from.getBasicType() == EbtFloat) {
|
||||
if (to2.getBasicType() == EbtDouble && to1.getBasicType() != EbtDouble)
|
||||
// shape changes are always worse
|
||||
if (from.isScalar() || from.isVector()) {
|
||||
if (from.getVectorSize() == to2.getVectorSize() &&
|
||||
from.getVectorSize() != to1.getVectorSize())
|
||||
return true;
|
||||
}
|
||||
|
||||
// int -> uint is better than any other int conversion
|
||||
if (from.getBasicType() == EbtInt) {
|
||||
if (to2.getBasicType() == EbtUint && to1.getBasicType() != EbtUint)
|
||||
return true;
|
||||
}
|
||||
|
||||
// TODO: these should be replaced by a more generic "shorter chain is better than longer chain" rule
|
||||
|
||||
// -> 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;
|
||||
|
||||
if (from.getVectorSize() == to1.getVectorSize() &&
|
||||
from.getVectorSize() != to2.getVectorSize())
|
||||
return false;
|
||||
}
|
||||
|
||||
// Might or might not be changing shape, which means basic type might
|
||||
// or might not match, so within that, the question is how big a
|
||||
// basic-type conversion is being done.
|
||||
//
|
||||
// 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;
|
||||
}
|
||||
};
|
||||
|
||||
return std::abs(linearize(to2.getBasicType()) - linearize(from.getBasicType())) <
|
||||
std::abs(linearize(to1.getBasicType()) - linearize(from.getBasicType()));
|
||||
};
|
||||
|
||||
// for ambiguity reporting
|
||||
|
||||
Loading…
Reference in New Issue
Block a user