Fix vectorization logic unit test for AVX512
diff --git a/test/vectorization_logic.cpp b/test/vectorization_logic.cpp
index 2d0a8ce..8efee92 100644
--- a/test/vectorization_logic.cpp
+++ b/test/vectorization_logic.cpp
@@ -37,6 +37,7 @@
template<typename Dst, typename Src>
bool test_assign(const Dst&, const Src&, int traversal, int unrolling)
{
+ EIGEN_PREDICATE_SAME_MATRIX_SIZE(Dst,Src);
typedef internal::copy_using_evaluator_traits<internal::evaluator<Dst>,internal::evaluator<Src>, internal::assign_op<typename Dst::Scalar,typename Src::Scalar> > traits;
bool res = traits::Traversal==traversal;
if(unrolling==InnerUnrolling+CompleteUnrolling)
@@ -61,6 +62,7 @@
template<typename Dst, typename Src>
bool test_assign(int traversal, int unrolling)
{
+ EIGEN_PREDICATE_SAME_MATRIX_SIZE(Dst,Src);
typedef internal::copy_using_evaluator_traits<internal::evaluator<Dst>,internal::evaluator<Src>, internal::assign_op<typename Dst::Scalar,typename Src::Scalar> > traits;
bool res = traits::Traversal==traversal && traits::Unrolling==unrolling;
if(!res)
@@ -117,26 +119,26 @@
typedef Matrix<Scalar,Dynamic,1> VectorX;
typedef Matrix<Scalar,Dynamic,Dynamic> MatrixXX;
typedef Matrix<Scalar,PacketSize,PacketSize> Matrix11;
- typedef Matrix<Scalar,2*PacketSize,2*PacketSize> Matrix22;
+ typedef Matrix<Scalar,(Matrix11::Flags&RowMajorBit)?8:2*PacketSize,(Matrix11::Flags&RowMajorBit)?2*PacketSize:8> Matrix22;
typedef Matrix<Scalar,(Matrix11::Flags&RowMajorBit)?16:4*PacketSize,(Matrix11::Flags&RowMajorBit)?4*PacketSize:16> Matrix44;
typedef Matrix<Scalar,(Matrix11::Flags&RowMajorBit)?16:4*PacketSize,(Matrix11::Flags&RowMajorBit)?4*PacketSize:16,DontAlign|EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION> Matrix44u;
typedef Matrix<Scalar,4*PacketSize,4*PacketSize,ColMajor> Matrix44c;
typedef Matrix<Scalar,4*PacketSize,4*PacketSize,RowMajor> Matrix44r;
typedef Matrix<Scalar,
- (PacketSize==8 ? 4 : PacketSize==4 ? 2 : PacketSize==2 ? 1 : /*PacketSize==1 ?*/ 1),
- (PacketSize==8 ? 2 : PacketSize==4 ? 2 : PacketSize==2 ? 2 : /*PacketSize==1 ?*/ 1)
+ (PacketSize==16 ? 8 : PacketSize==8 ? 4 : PacketSize==4 ? 2 : PacketSize==2 ? 1 : /*PacketSize==1 ?*/ 1),
+ (PacketSize==16 ? 2 : PacketSize==8 ? 2 : PacketSize==4 ? 2 : PacketSize==2 ? 2 : /*PacketSize==1 ?*/ 1)
> Matrix1;
typedef Matrix<Scalar,
- (PacketSize==8 ? 4 : PacketSize==4 ? 2 : PacketSize==2 ? 1 : /*PacketSize==1 ?*/ 1),
- (PacketSize==8 ? 2 : PacketSize==4 ? 2 : PacketSize==2 ? 2 : /*PacketSize==1 ?*/ 1),
+ (PacketSize==16 ? 8 : PacketSize==8 ? 4 : PacketSize==4 ? 2 : PacketSize==2 ? 1 : /*PacketSize==1 ?*/ 1),
+ (PacketSize==16 ? 2 : PacketSize==8 ? 2 : PacketSize==4 ? 2 : PacketSize==2 ? 2 : /*PacketSize==1 ?*/ 1),
DontAlign|((Matrix1::Flags&RowMajorBit)?RowMajor:ColMajor)> Matrix1u;
// this type is made such that it can only be vectorized when viewed as a linear 1D vector
typedef Matrix<Scalar,
- (PacketSize==8 ? 4 : PacketSize==4 ? 6 : PacketSize==2 ? ((Matrix11::Flags&RowMajorBit)?2:3) : /*PacketSize==1 ?*/ 1),
- (PacketSize==8 ? 6 : PacketSize==4 ? 2 : PacketSize==2 ? ((Matrix11::Flags&RowMajorBit)?3:2) : /*PacketSize==1 ?*/ 3)
+ (PacketSize==16 ? 4 : PacketSize==8 ? 4 : PacketSize==4 ? 6 : PacketSize==2 ? ((Matrix11::Flags&RowMajorBit)?2:3) : /*PacketSize==1 ?*/ 1),
+ (PacketSize==16 ? 12 : PacketSize==8 ? 6 : PacketSize==4 ? 2 : PacketSize==2 ? ((Matrix11::Flags&RowMajorBit)?3:2) : /*PacketSize==1 ?*/ 3)
> Matrix3;
#if !EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT
@@ -202,7 +204,7 @@
VERIFY(test_assign(Matrix11(), Matrix11()+Matrix11(),InnerVectorizedTraversal,CompleteUnrolling));
- VERIFY(test_assign(Matrix11(),Matrix<Scalar,17,17>().template block<PacketSize,PacketSize>(2,3)+Matrix<Scalar,17,17>().template block<PacketSize,PacketSize>(8,4),
+ VERIFY(test_assign(Matrix11(),Matrix<Scalar,21,21>().template block<PacketSize,PacketSize>(2,3)+Matrix<Scalar,21,21>().template block<PacketSize,PacketSize>(3,2),
(EIGEN_UNALIGNED_VECTORIZE) ? InnerVectorizedTraversal : DefaultTraversal, CompleteUnrolling|InnerUnrolling));
VERIFY(test_assign(Vector1(),Matrix11()*Vector1(),
@@ -294,19 +296,19 @@
// typedef Matrix<Scalar,4*PacketSize,4*PacketSize,RowMajor> Matrix44r;
typedef Matrix<Scalar,
- (PacketSize==8 ? 4 : PacketSize==4 ? 2 : PacketSize==2 ? 1 : /*PacketSize==1 ?*/ 1),
- (PacketSize==8 ? 2 : PacketSize==4 ? 2 : PacketSize==2 ? 2 : /*PacketSize==1 ?*/ 1)
+ (PacketSize==16 ? 8 : PacketSize==8 ? 4 : PacketSize==4 ? 2 : PacketSize==2 ? 1 : /*PacketSize==1 ?*/ 1),
+ (PacketSize==16 ? 2 : PacketSize==8 ? 2 : PacketSize==4 ? 2 : PacketSize==2 ? 2 : /*PacketSize==1 ?*/ 1)
> Matrix1;
typedef Matrix<Scalar,
- (PacketSize==8 ? 4 : PacketSize==4 ? 2 : PacketSize==2 ? 1 : /*PacketSize==1 ?*/ 1),
- (PacketSize==8 ? 2 : PacketSize==4 ? 2 : PacketSize==2 ? 2 : /*PacketSize==1 ?*/ 1),
+ (PacketSize==16 ? 8 : PacketSize==8 ? 4 : PacketSize==4 ? 2 : PacketSize==2 ? 1 : /*PacketSize==1 ?*/ 1),
+ (PacketSize==16 ? 2 : PacketSize==8 ? 2 : PacketSize==4 ? 2 : PacketSize==2 ? 2 : /*PacketSize==1 ?*/ 1),
DontAlign|((Matrix1::Flags&RowMajorBit)?RowMajor:ColMajor)> Matrix1u;
// this type is made such that it can only be vectorized when viewed as a linear 1D vector
typedef Matrix<Scalar,
- (PacketSize==8 ? 4 : PacketSize==4 ? 6 : PacketSize==2 ? ((Matrix11::Flags&RowMajorBit)?2:3) : /*PacketSize==1 ?*/ 1),
- (PacketSize==8 ? 6 : PacketSize==4 ? 2 : PacketSize==2 ? ((Matrix11::Flags&RowMajorBit)?3:2) : /*PacketSize==1 ?*/ 3)
+ (PacketSize==16 ? 4 : PacketSize==8 ? 4 : PacketSize==4 ? 6 : PacketSize==2 ? ((Matrix11::Flags&RowMajorBit)?2:3) : /*PacketSize==1 ?*/ 1),
+ (PacketSize==16 ? 12 : PacketSize==8 ? 6 : PacketSize==4 ? 2 : PacketSize==2 ? ((Matrix11::Flags&RowMajorBit)?3:2) : /*PacketSize==1 ?*/ 3)
> Matrix3;
#if !EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT
@@ -359,7 +361,8 @@
NoUnrolling));
VERIFY(test_assign(Matrix11(),Matrix<Scalar,17,17>().template block<PacketSize,PacketSize>(2,3)+Matrix<Scalar,17,17>().template block<PacketSize,PacketSize>(8,4),
- EIGEN_UNALIGNED_VECTORIZE ? InnerVectorizedTraversal : DefaultTraversal,PacketSize>4?InnerUnrolling:CompleteUnrolling));
+ EIGEN_UNALIGNED_VECTORIZE ? InnerVectorizedTraversal : DefaultTraversal,InnerUnrolling+CompleteUnrolling));
+
VERIFY(test_assign(Vector1(),Matrix11()*Vector1(),
InnerVectorizedTraversal,CompleteUnrolling));
@@ -391,10 +394,10 @@
VERIFY((test_assign<
Map<Matrix<Scalar,EIGEN_PLAIN_ENUM_MAX(2,PacketSize),EIGEN_PLAIN_ENUM_MAX(2,PacketSize)>, AlignedMax, InnerStride<3*PacketSize> >,
Matrix<Scalar,EIGEN_PLAIN_ENUM_MAX(2,PacketSize),EIGEN_PLAIN_ENUM_MAX(2,PacketSize)>
- >(DefaultTraversal,CompleteUnrolling)));
+ >(DefaultTraversal,PacketSize>4?InnerUnrolling:CompleteUnrolling)));
VERIFY((test_assign(Matrix57(), Matrix<Scalar,5*PacketSize,3>()*Matrix<Scalar,3,7>(),
- InnerVectorizedTraversal, InnerUnrolling|CompleteUnrolling)));
+ InnerVectorizedTraversal, InnerUnrolling+CompleteUnrolling)));
#endif
}
};