GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-09-13 01:43:13 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix ODR issues with Intel's AVX512 TRSM kernels.

Browse Source
This commit is contained in:
Antonio Sánchez 2023-02-27 07:54:52 +00:00
parent 826627f653
commit 62d5cfe835
2 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
Eigen/src/Core/arch/AVX512/TrsmKernel.h
View File

@ -110,7 +110,7 @@ int64_t avx512_trsm_cutoff(int64_t L2Size, int64_t N, double L2Cap) {
* Used by gemmKernel for the case A/B row-major and C col-major.
*/
template <typename Scalar, typename vec, int64_t unrollM, int64_t unrollN, bool remM, bool remN>
static EIGEN_ALWAYS_INLINE void transStoreC(PacketBlock<vec, EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS> &zmm,
EIGEN_ALWAYS_INLINE void transStoreC(PacketBlock<vec, EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS> &zmm,
Scalar *C_arr, int64_t LDC, int64_t remM_ = 0, int64_t remN_ = 0) {
EIGEN_UNUSED_VARIABLE(remN_);
EIGEN_UNUSED_VARIABLE(remM_);
@ -721,7 +721,7 @@ void gemmKernel(Scalar *A_arr, Scalar *B_arr, Scalar *C_arr, int64_t M, int64_t
* The B matrix (RHS) is assumed to be row-major
*/
template <typename Scalar, typename vec, int64_t unrollM, bool isARowMajor, bool isFWDSolve, bool isUnitDiag>
static EIGEN_ALWAYS_INLINE void triSolveKernel(Scalar *A_arr, Scalar *B_arr, int64_t K, int64_t LDA, int64_t LDB) {
EIGEN_ALWAYS_INLINE void triSolveKernel(Scalar *A_arr, Scalar *B_arr, int64_t K, int64_t LDA, int64_t LDB) {
static_assert(unrollM <= EIGEN_AVX_MAX_NUM_ROW, "unrollM should be equal to EIGEN_AVX_MAX_NUM_ROW");
using urolls = unrolls::trsm<Scalar>;
constexpr int64_t U3 = urolls::PacketSize * 3;
@ -802,7 +802,7 @@ void triSolveKernelLxK(Scalar *A_arr, Scalar *B_arr, int64_t M, int64_t K, int64
*
*/
template <typename Scalar, bool toTemp = true, bool remM = false>
static EIGEN_ALWAYS_INLINE void copyBToRowMajor(Scalar *B_arr, int64_t LDB, int64_t K, Scalar *B_temp, int64_t LDB_,
EIGEN_ALWAYS_INLINE void copyBToRowMajor(Scalar *B_arr, int64_t LDB, int64_t K, Scalar *B_temp, int64_t LDB_,
int64_t remM_ = 0) {
EIGEN_UNUSED_VARIABLE(remM_);
using urolls = unrolls::transB<Scalar>;

2
Eigen/src/Core/arch/AVX512/TrsmUnrolls.inc
View File

@ -11,7 +11,7 @@
#define EIGEN_CORE_ARCH_AVX512_TRSM_UNROLLS_H
template <bool isARowMajor = true>
static EIGEN_ALWAYS_INLINE int64_t idA(int64_t i, int64_t j, int64_t LDA) {
EIGEN_ALWAYS_INLINE int64_t idA(int64_t i, int64_t j, int64_t LDA) {
EIGEN_IF_CONSTEXPR(isARowMajor) return i * LDA + j;
else return i + j * LDA;
}