GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-14 04:35:57 +08:00
Code Issues Packages Projects Releases Wiki Activity

Enable runtime stack alignment in gemm_blocking_space.

Browse Source
This commit is contained in:
Gael Guennebaud 2015-08-06 14:00:26 +02:00
parent d4f5efc51a
commit becd89df29
2 changed files with 11 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
Eigen/src/Core/products/GeneralBlockPanelKernel.h
View File

@ -2066,7 +2066,7 @@ EIGEN_DONT_INLINE void gemm_pack_rhs<Scalar, Index, DataMapper, nr, ColMajor, Co
const LinearMapper dm3 = rhs.getLinearMapper(0, j2 + 3); const LinearMapper dm3 = rhs.getLinearMapper(0, j2 + 3);
Index k=0; Index k=0;
if((PacketSize%4)==0) // TODO enbale vectorized transposition for PacketSize==2 ?? if((PacketSize%4)==0) // TODO enable vectorized transposition for PacketSize==2 ??
{ {
for(; k<peeled_k; k+=PacketSize) { for(; k<peeled_k; k+=PacketSize) {
PacketBlock<Packet,(PacketSize%4)==0?4:PacketSize> kernel; PacketBlock<Packet,(PacketSize%4)==0?4:PacketSize> kernel;

10
Eigen/src/Core/products/GeneralMatrixMatrix.h
View File

@ -293,8 +293,13 @@ class gemm_blocking_space<StorageOrder,_LhsScalar,_RhsScalar,MaxRows, MaxCols, M
SizeB = ActualCols * MaxDepth SizeB = ActualCols * MaxDepth
}; };
#if EIGEN_MAX_STATIC_ALIGN_BYTES >= EIGEN_DEFAULT_ALIGN_BYTES
EIGEN_ALIGN_MAX LhsScalar m_staticA[SizeA]; EIGEN_ALIGN_MAX LhsScalar m_staticA[SizeA];
EIGEN_ALIGN_MAX RhsScalar m_staticB[SizeB]; EIGEN_ALIGN_MAX RhsScalar m_staticB[SizeB];
#else
EIGEN_ALIGN_MAX char m_staticA[SizeA * sizeof(LhsScalar) + EIGEN_DEFAULT_ALIGN_BYTES-1];
EIGEN_ALIGN_MAX char m_staticB[SizeB * sizeof(RhsScalar) + EIGEN_DEFAULT_ALIGN_BYTES-1];
#endif
public: public:
@ -303,8 +308,13 @@ class gemm_blocking_space<StorageOrder,_LhsScalar,_RhsScalar,MaxRows, MaxCols, M
this->m_mc = ActualRows; this->m_mc = ActualRows;
this->m_nc = ActualCols; this->m_nc = ActualCols;
this->m_kc = MaxDepth; this->m_kc = MaxDepth;
#if EIGEN_MAX_STATIC_ALIGN_BYTES >= EIGEN_DEFAULT_ALIGN_BYTES
this->m_blockA = m_staticA; this->m_blockA = m_staticA;
this->m_blockB = m_staticB; this->m_blockB = m_staticB;
#else
this->m_blockA = reinterpret_cast<LhsScalar*>((std::size_t(m_staticA) + (EIGEN_DEFAULT_ALIGN_BYTES-1)) & ~std::size_t(EIGEN_DEFAULT_ALIGN_BYTES-1));
this->m_blockB = reinterpret_cast<RhsScalar*>((std::size_t(m_staticB) + (EIGEN_DEFAULT_ALIGN_BYTES-1)) & ~std::size_t(EIGEN_DEFAULT_ALIGN_BYTES-1));
#endif
} }
void initParallel(Index, Index, Index, Index) void initParallel(Index, Index, Index, Index)