diff options
author | Benoit Steiner <benoit.steiner.goog@gmail.com> | 2014-02-18 18:06:44 -0800 |
---|---|---|
committer | Benoit Steiner <benoit.steiner.goog@gmail.com> | 2014-02-18 18:06:44 -0800 |
commit | 7ed9441ea472031bb9357b5bec80151cae7ed2cb (patch) | |
tree | ecd2a8a0973068987bdb5ec631a79e8b82641e56 /Eigen/src/Core/products/GeneralMatrixVector.h | |
parent | 64a85800bd3573a7da7a396fde9707dce87a58d9 (diff) |
Reverted the definition of the EIGEN_ALIGN to its former meaning (i.e. a boolean)
Created a new EIGEN_ALIGN_BYTES define to encode how the data should be aligned
Fixed a few remaining alignment issues exposed when the Eigen code is compiled with avx enabled.
Created a new EIGEN_ALIGN_DEFAULT define, which is set to the minimum alignment value required for the chosen instruction set. Use this value instead of EIGEN_ALIGN32 to preserve the existing alignment on SSE/Altivec/Neon.
Diffstat (limited to 'Eigen/src/Core/products/GeneralMatrixVector.h')
-rw-r--r-- | Eigen/src/Core/products/GeneralMatrixVector.h | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/Eigen/src/Core/products/GeneralMatrixVector.h b/Eigen/src/Core/products/GeneralMatrixVector.h index 7f35cc53a..7bf350c42 100644 --- a/Eigen/src/Core/products/GeneralMatrixVector.h +++ b/Eigen/src/Core/products/GeneralMatrixVector.h @@ -449,7 +449,7 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,RowMajor,Co Index rowBound = ((rows-skipRows)/rowsAtOnce)*rowsAtOnce + skipRows; for (Index i=skipRows; i<rowBound; i+=rowsAtOnce) { - EIGEN_ALIGN32 ResScalar tmp0 = ResScalar(0); + EIGEN_ALIGN_DEFAULT ResScalar tmp0 = ResScalar(0); ResScalar tmp1 = ResScalar(0), tmp2 = ResScalar(0), tmp3 = ResScalar(0); // this helps the compiler generating good binary code @@ -558,7 +558,7 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,RowMajor,Co { for (Index i=start; i<end; ++i) { - EIGEN_ALIGN32 ResScalar tmp0 = ResScalar(0); + EIGEN_ALIGN_DEFAULT ResScalar tmp0 = ResScalar(0); ResPacket ptmp0 = pset1<ResPacket>(tmp0); const LhsScalar* lhs0 = lhs + i*lhsStride; // process first unaligned result's coeffs |