From 8700e9e6d1c5317cb015d29b09e454ca0748b378 Mon Sep 17 00:00:00 2001 From: John Kessenich Date: Fri, 30 Aug 2013 00:45:57 +0000 Subject: [PATCH] Add more constant folding cases for min, max, step, smoothstep, mix, clamp, atan, and pow. git-svn-id: https://cvs.khronos.org/svn/repos/ogl/trunk/ecosystem/public/sdk/tools/glslang@22903 e7fa87d3-cd2b-0410-9028-fcbf551c1848 --- Test/baseResults/constFold.frag.out | 199 +++++++++++++++--------- Test/constFold.frag | 21 ++- glslang/MachineIndependent/Constant.cpp | 147 ++++++++++++----- 3 files changed, 256 insertions(+), 111 deletions(-) diff --git a/Test/baseResults/constFold.frag.out b/Test/baseResults/constFold.frag.out index e8f180ec..343077f6 100644 --- a/Test/baseResults/constFold.frag.out +++ b/Test/baseResults/constFold.frag.out @@ -1,81 +1,134 @@ 0:? Sequence -0:26 Function Definition: main( (void) -0:26 Function Parameters: -0:28 Sequence -0:28 Sequence -0:28 move second child to first child (4-component vector of float) -0:28 'dx' (4-component vector of float) -0:28 dPdx (4-component vector of float) -0:28 'inv' (smooth in 4-component vector of float) -0:35 move second child to first child (4-component vector of float) -0:35 'FragColor' (out 4-component vector of float) -0:35 2.000000 -0:35 6.000000 -0:35 3.000000 -0:35 171.887339 -0:40 move second child to first child (4-component vector of float) -0:40 'FragColor' (out 4-component vector of float) -0:40 3.000000 -0:40 2.000000 -0:40 0.001593 -0:40 -0.999999 -0:41 move second child to first child (2-component vector of float) -0:41 'out2' (out 2-component vector of float) -0:41 5.600000 -0:41 5.800000 +0:28 Function Definition: main( (void) +0:28 Function Parameters: +0:30 Sequence +0:30 Sequence +0:30 move second child to first child (4-component vector of float) +0:30 'dx' (4-component vector of float) +0:30 dPdx (4-component vector of float) +0:30 'inv' (smooth in 4-component vector of float) +0:37 move second child to first child (4-component vector of float) +0:37 'FragColor' (out 4-component vector of float) +0:37 2.000000 +0:37 6.000000 +0:37 3.000000 +0:37 171.887339 0:42 move second child to first child (4-component vector of float) -0:42 'out3' (out 4-component vector of float) -0:42 20.085537 -0:42 2.302585 -0:42 16.000000 -0:42 8.000000 -0:43 move second child to first child (4-component vector of float) -0:43 'out4' (out 4-component vector of float) -0:43 10.000000 -0:43 0.100000 -0:43 4.700000 -0:43 10.900000 -0:44 move second child to first child (4-component vector of int) -0:44 'out5' (out 4-component vector of int) -0:44 8 (const int) -0:44 17 (const int) -0:44 -1 (const int) -0:44 1 (const int) -0:45 move second child to first child (3-component vector of float) -0:45 'out6' (out 3-component vector of float) -0:45 -1.000000 -0:45 1.000000 -0:45 0.000000 -0:46 move second child to first child (4-component vector of float) -0:46 'out7' (out 4-component vector of float) -0:46 4.000000 -0:46 -4.000000 -0:46 5.000000 -0:46 -5.000000 -0:47 move second child to first child (4-component vector of float) -0:47 'out8' (out 4-component vector of float) -0:47 4.000000 -0:47 5.000000 -0:47 4.000000 -0:47 -6.000000 +0:42 'FragColor' (out 4-component vector of float) +0:42 3.000000 +0:42 2.000000 +0:42 0.001593 +0:42 -0.999999 +0:43 move second child to first child (2-component vector of float) +0:43 'out2' (out 2-component vector of float) +0:43 5.600000 +0:43 5.800000 +0:44 move second child to first child (4-component vector of float) +0:44 'out3' (out 4-component vector of float) +0:44 20.085537 +0:44 2.302585 +0:44 16.000000 +0:44 8.000000 +0:45 move second child to first child (4-component vector of float) +0:45 'out4' (out 4-component vector of float) +0:45 10.000000 +0:45 0.100000 +0:45 4.700000 +0:45 10.900000 +0:46 move second child to first child (4-component vector of int) +0:46 'out5' (out 4-component vector of int) +0:46 8 (const int) +0:46 17 (const int) +0:46 -1 (const int) +0:46 1 (const int) +0:47 move second child to first child (3-component vector of float) +0:47 'out6' (out 3-component vector of float) +0:47 -1.000000 +0:47 1.000000 +0:47 0.000000 0:48 move second child to first child (4-component vector of float) -0:48 'out9' (out 4-component vector of float) -0:48 8.000000 +0:48 'out7' (out 4-component vector of float) +0:48 4.000000 0:48 -4.000000 -0:48 0.345000 -0:48 0.400000 +0:48 5.000000 +0:48 -5.000000 0:49 move second child to first child (4-component vector of float) -0:49 'out10' (out 4-component vector of float) -0:49 1.000000 -0:49 1.000000 -0:49 0.000000 -0:49 0.000000 +0:49 'out8' (out 4-component vector of float) +0:49 4.000000 +0:49 5.000000 +0:49 4.000000 +0:49 -6.000000 0:50 move second child to first child (4-component vector of float) -0:50 'out11' (out 4-component vector of float) -0:50 0.000000 -0:50 0.000000 -0:50 1.000000 -0:50 0.000000 +0:50 'out9' (out 4-component vector of float) +0:50 8.000000 +0:50 -4.000000 +0:50 0.345000 +0:50 0.400000 +0:51 move second child to first child (4-component vector of float) +0:51 'out10' (out 4-component vector of float) +0:51 1.000000 +0:51 1.000000 +0:51 0.000000 +0:51 0.000000 +0:52 move second child to first child (4-component vector of float) +0:52 'out11' (out 4-component vector of float) +0:52 0.000000 +0:52 0.000000 +0:52 1.000000 +0:52 0.000000 +0:53 move second child to first child (4-component vector of float) +0:53 'out11' (out 4-component vector of float) +0:53 1.029639 +0:53 0.799690 +0:53 0.674741 +0:53 1.570696 +0:54 move second child to first child (4-component vector of float) +0:54 'out11' (out 4-component vector of float) +0:54 0.000000 +0:54 0.523599 +0:54 1.570796 +0:54 1.047198 +0:58 move second child to first child (4-component vector of float) +0:58 'out11' (out 4-component vector of float) +0:58 1.373401 +0:58 0.000000 +0:58 0.896055 +0:58 -0.380506 +0:62 move second child to first child (2-component vector of int) +0:62 'out12' (out 2-component vector of int) +0:62 15 (const int) +0:62 16 (const int) +0:63 move second child to first child (2-component vector of int) +0:63 'out12' (out 2-component vector of int) +0:63 17 (const int) +0:63 17 (const int) +0:64 move second child to first child (2-component vector of float) +0:64 'out2' (out 2-component vector of float) +0:64 871.421253 +0:64 4913.000000 +0:65 move second child to first child (3-component vector of uint) +0:65 'out13' (out 3-component vector of uint) +0:65 10 (const uint) +0:65 20 (const uint) +0:65 30 (const uint) +0:66 move second child to first child (2-component vector of float) +0:66 'out2' (out 2-component vector of float) +0:66 3.000000 +0:66 6.000000 +0:67 move second child to first child (2-component vector of float) +0:67 'out2' (out 2-component vector of float) +0:67 3.500000 +0:67 4.500000 +0:68 move second child to first child (2-component vector of float) +0:68 'out2' (out 2-component vector of float) +0:68 0.000000 +0:68 1.000000 +0:69 move second child to first child (4-component vector of float) +0:69 'out11' (out 4-component vector of float) +0:69 0.000000 +0:69 0.028000 +0:69 0.500000 +0:69 1.000000 0:? Linker Objects 0:? 'inv' (smooth in 4-component vector of float) 0:? 'FragColor' (out 4-component vector of float) @@ -89,4 +142,6 @@ 0:? 'out9' (out 4-component vector of float) 0:? 'out10' (out 4-component vector of float) 0:? 'out11' (out 4-component vector of float) +0:? 'out12' (out 2-component vector of int) +0:? 'out13' (out 3-component vector of uint) diff --git a/Test/constFold.frag b/Test/constFold.frag index 61e5f3b1..0769390c 100644 --- a/Test/constFold.frag +++ b/Test/constFold.frag @@ -21,7 +21,9 @@ out vec4 out7; out vec4 out8; out vec4 out9; out vec4 out10; -out vec4 out11; +out vec4 out11; +out ivec2 out12; +out uvec3 out13; void main() { @@ -48,4 +50,21 @@ void main() out9 = vec4(roundEven(7.5), roundEven(-4.5), fract(2.345), fract(-2.6)); // 8, -4, .345, 0.4 out10 = vec4(isinf(4.0/0.0), isinf(-3.0/0.0), isinf(0.0/0.0), isinf(-93048593405938405938405.0)); // true, true, false, false -> 1.0, 1.0, 0.0, 0.0 out11 = vec4(isnan(4.0/0.0), isnan(-3.0/0.0), isnan(0.0/0.0), isnan(-93048593405938405938405.0)); // false, false, true, false -> 0.0, 1.0, 0.0, 0.0 + out11 = vec4(tan(0.8), atan(1.029), atan(8.0, 10.0), atan(10000.0)); // 1.029, 0.8, 0.6747, 1.57 + out11 = vec4(asin(0.0), asin(0.5), acos(0.0), acos(0.5)); // 0.0, .523599, 1.57, 1.047 + + const vec4 v1 = vec4(1.0, 0.0, 0.5, -0.2); + const vec4 v2 = vec4(0.2, 0.3, 0.4, 0.5); + out11 = atan(v1, v2); // 1.373401, 0.0, 0.896055, -0.380506 + + const ivec2 v3 = ivec2(15.0, 17.0); + const ivec2 v4 = ivec2(17.0, 15.0); + out12 = min(v3, 16); // 15, 16 + out12 = max(v3, v4); // 17, 17 + out2 = pow(vec2(v3), vec2(2.5, 3.0)); // 871.4, 4913 + out13 = clamp(uvec3(1, 20, 50), 10u, 30u); // 10, 20, 30 + out2 = mix(vec2(3.0, 4.0), vec2(5.0, 6.0), bvec2(false, true)); // 3.0, 6.0 + out2 = mix(vec2(3.0, 4.0), vec2(5.0, 6.0), 0.25); // 3.5, 4.5 + out2 = step(0.5, vec2(0.2, 0.6)); // 0.0, 1.0 + out11 = smoothstep(50.0, 60.0, vec4(40.0, 51.0, 55.0, 70.0)); // 0.0, 0.028, 0.5, 1.0 } diff --git a/glslang/MachineIndependent/Constant.cpp b/glslang/MachineIndependent/Constant.cpp index dcccf164..174aa55a 100644 --- a/glslang/MachineIndependent/Constant.cpp +++ b/glslang/MachineIndependent/Constant.cpp @@ -373,7 +373,7 @@ TIntermTyped* TIntermConstantUnion::fold(TOperator op, TIntermTyped* constantNod break; default: - infoSink.info.message(EPrefixInternalError, "Invalid operator for constant folding", getLoc()); + infoSink.info.message(EPrefixInternalError, "Invalid binary operator for constant folding", getLoc()); return 0; } @@ -633,21 +633,26 @@ TIntermTyped* TIntermediate::fold(TIntermAggregate* aggrNode) // First, see if this is an operation to constant fold, kick out if not, // see what size the result is if so. + + bool componentwise = false; // will also say componentwise if a scalar argument gets repeated to make per-component results int objectSize; switch (aggrNode->getOp()) { + case EOpAtan: + case EOpPow: case EOpMin: case EOpMax: + case EOpMix: + case EOpClamp: + componentwise = true; + objectSize = children[0]->getAsConstantUnion()->getType().getObjectSize(); + break; + case EOpCross: case EOpReflect: case EOpRefract: case EOpFaceForward: - case EOpAtan: - case EOpPow: - case EOpClamp: - case EOpMix: - case EOpDistance: - case EOpCross: objectSize = children[0]->getAsConstantUnion()->getType().getObjectSize(); break; + case EOpDistance: case EOpDot: objectSize = 1; break; @@ -656,10 +661,12 @@ TIntermTyped* TIntermediate::fold(TIntermAggregate* aggrNode) children[1]->getAsTyped()->getType().getVectorSize(); break; case EOpStep: + componentwise = true; objectSize = std::max(children[0]->getAsTyped()->getType().getVectorSize(), children[1]->getAsTyped()->getType().getVectorSize()); break; case EOpSmoothStep: + componentwise = true; objectSize = std::max(children[0]->getAsTyped()->getType().getVectorSize(), children[2]->getAsTyped()->getType().getVectorSize()); break; @@ -669,44 +676,108 @@ TIntermTyped* TIntermediate::fold(TIntermAggregate* aggrNode) TConstUnion* newConstArray = new TConstUnion[objectSize]; TVector childConstUnions; - for (unsigned int i = 0; i < children.size(); ++i) - childConstUnions.push_back(children[i]->getAsConstantUnion()->getUnionArrayPointer()); + for (unsigned int arg = 0; arg < children.size(); ++arg) + childConstUnions.push_back(children[arg]->getAsConstantUnion()->getUnionArrayPointer()); // Second, do the actual folding - // TODO: Functionality: constant folding: separate component-wise from non-component-wise - switch (aggrNode->getOp()) { - case EOpMin: - case EOpMax: - for (int i = 0; i < objectSize; i++) { - if (aggrNode->getOp() == EOpMax) - newConstArray[i].setDConst(std::max(childConstUnions[0]->getDConst(), childConstUnions[1]->getDConst())); - else - newConstArray[i].setDConst(std::min(childConstUnions[0]->getDConst(), childConstUnions[1]->getDConst())); + bool isFloatingPoint = children[0]->getAsTyped()->getBasicType() == EbtFloat || + children[0]->getAsTyped()->getBasicType() == EbtDouble; + bool isSigned = children[0]->getAsTyped()->getBasicType() == EbtInt; + if (componentwise) { + for (int comp = 0; comp < objectSize; comp++) { + + // some arguments are scalars instead of matching vectors; simulate a smear + int arg0comp = std::min(comp, children[0]->getAsTyped()->getType().getVectorSize() - 1); + int arg1comp; + if (children.size() > 1) + arg1comp = std::min(comp, children[1]->getAsTyped()->getType().getVectorSize() - 1); + int arg2comp; + if (children.size() > 2) + arg2comp = std::min(comp, children[2]->getAsTyped()->getType().getVectorSize() - 1); + + switch (aggrNode->getOp()) { + case EOpAtan: + newConstArray[comp].setDConst(atan2(childConstUnions[0][arg0comp].getDConst(), childConstUnions[1][arg1comp].getDConst())); + break; + case EOpPow: + newConstArray[comp].setDConst(pow(childConstUnions[0][arg0comp].getDConst(), childConstUnions[1][arg1comp].getDConst())); + break; + case EOpMin: + if (isFloatingPoint) + newConstArray[comp].setDConst(std::min(childConstUnions[0][arg0comp].getDConst(), childConstUnions[1][arg1comp].getDConst())); + else if (isSigned) + newConstArray[comp].setIConst(std::min(childConstUnions[0][arg0comp].getIConst(), childConstUnions[1][arg1comp].getIConst())); + else + newConstArray[comp].setUConst(std::min(childConstUnions[0][arg0comp].getUConst(), childConstUnions[1][arg1comp].getUConst())); + break; + case EOpMax: + if (isFloatingPoint) + newConstArray[comp].setDConst(std::max(childConstUnions[0][arg0comp].getDConst(), childConstUnions[1][arg1comp].getDConst())); + else if (isSigned) + newConstArray[comp].setIConst(std::max(childConstUnions[0][arg0comp].getIConst(), childConstUnions[1][arg1comp].getIConst())); + else + newConstArray[comp].setUConst(std::max(childConstUnions[0][arg0comp].getUConst(), childConstUnions[1][arg1comp].getUConst())); + break; + case EOpClamp: + if (isFloatingPoint) + newConstArray[comp].setDConst(std::min(std::max(childConstUnions[0][arg0comp].getDConst(), childConstUnions[1][arg1comp].getDConst()), + childConstUnions[2][arg2comp].getDConst())); + else if (isSigned) + newConstArray[comp].setIConst(std::min(std::max(childConstUnions[0][arg0comp].getIConst(), childConstUnions[1][arg1comp].getIConst()), + childConstUnions[2][arg2comp].getIConst())); + else + newConstArray[comp].setUConst(std::min(std::max(childConstUnions[0][arg0comp].getUConst(), childConstUnions[1][arg1comp].getUConst()), + childConstUnions[2][arg2comp].getUConst())); + break; + case EOpMix: + if (children[2]->getAsTyped()->getBasicType() == EbtBool) + newConstArray[comp].setDConst(childConstUnions[2][arg2comp].getBConst() ? childConstUnions[1][arg1comp].getDConst() : + childConstUnions[0][arg0comp].getDConst()); + else + newConstArray[comp].setDConst(childConstUnions[0][arg0comp].getDConst() * (1.0 - childConstUnions[2][arg2comp].getDConst()) + + childConstUnions[1][arg1comp].getDConst() * childConstUnions[2][arg2comp].getDConst()); + break; + case EOpStep: + newConstArray[comp].setDConst(childConstUnions[1][arg1comp].getDConst() < childConstUnions[0][arg0comp].getDConst() ? 0.0 : 1.0); + break; + case EOpSmoothStep: + { + double t = (childConstUnions[2][arg2comp].getDConst() - childConstUnions[0][arg0comp].getDConst()) / + (childConstUnions[1][arg1comp].getDConst() - childConstUnions[0][arg0comp].getDConst()); + if (t < 0.0) + t = 0.0; + if (t > 1.0) + t = 1.0; + newConstArray[comp].setDConst(t * t * (3.0 - 2.0 * t)); + break; + } + default: + infoSink.info.message(EPrefixInternalError, "componentwise constant folding operation not implemented", aggrNode->getLoc()); + return aggrNode; + } } - break; + } else { + // Non-componentwise... - // TODO: Functionality: constant folding: the rest of the ops have to be fleshed out + switch (aggrNode->getOp()) { - case EOpAtan: - case EOpPow: - case EOpModf: - case EOpClamp: - case EOpMix: - case EOpStep: - case EOpSmoothStep: - case EOpDistance: - case EOpDot: - case EOpCross: - case EOpFaceForward: - case EOpReflect: - case EOpRefract: - case EOpOuterProduct: - infoSink.info.message(EPrefixInternalError, "constant folding operation not implemented", aggrNode->getLoc()); - return aggrNode; + // TODO: Functionality: constant folding: the rest of the ops have to be fleshed out - default: - return aggrNode; + case EOpModf: + case EOpDistance: + case EOpDot: + case EOpCross: + case EOpFaceForward: + case EOpReflect: + case EOpRefract: + case EOpOuterProduct: + infoSink.info.message(EPrefixInternalError, "constant folding operation not implemented", aggrNode->getLoc()); + return aggrNode; + + default: + return aggrNode; + } } TIntermConstantUnion *newNode = new TIntermConstantUnion(newConstArray, aggrNode->getType()); -- 2.34.1