arcadiaphy commented on a change in pull request #14963: Add matrix inversion operator in linalg URL: https://github.com/apache/incubator-mxnet/pull/14963#discussion_r284547249
########## File path: src/operator/linalg_impl.h ########## @@ -1233,4 +1234,252 @@ LINALG_GPU_SYEVD_WORKSPACE_QUERY(DnDsyevd, double) #endif // __CUDACC__ +//////////////////////////////// GETRF //////////////////////////////////////////// + +// CPU/GPU-versions of LAPACK function "getrf" + +// The input of this function should be col-major for performance. +// Tensor work holds space for ipiv in getrf +#define LINALG_CPU_GETRF(fname, DType) \ +template<> inline \ +void linalg_getrf<cpu, DType>(const Tensor<cpu, 2, DType>& A, \ + const Tensor<cpu, 1, DType>& work, \ + Stream<cpu> *s) { \ + int *ipiv = reinterpret_cast<int *>(work.dptr_); \ + int ret(MXNET_LAPACK_##fname(MXNET_LAPACK_COL_MAJOR, A.size(1), A.size(0), \ + A.dptr_, A.stride_, ipiv)); \ + CHECK_EQ(ret, 0) << #fname << " failed in lapack on cpu."; \ +} + +LINALG_CPU_GETRF(sgetrf, float) +LINALG_CPU_GETRF(dgetrf, double) + +#ifdef __CUDACC__ + +struct set_matrix : public mxnet::op::mxnet_op::tunable { + template<typename DType> + MSHADOW_XINLINE static void Map(int i, DType **p, DType *m, int step) { + p[i] = m + i * step; Review comment: Right, I'll fix it. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services