diff options
Diffstat (limited to 'csrc/pythonInterface.c')
| -rw-r--r-- | csrc/pythonInterface.c | 6 |
1 files changed, 6 insertions, 0 deletions
diff --git a/csrc/pythonInterface.c b/csrc/pythonInterface.c index a6a4b13..7356c11 100644 --- a/csrc/pythonInterface.c +++ b/csrc/pythonInterface.c @@ -105,6 +105,9 @@ void transform_row2turingT(char * A, char *out, int rows, int cols){ transformRo void transform_row2ampere(char * A, char *out, int rows, int cols){ transformRowToFormat<COL_AMPERE, 0>(A, out, rows, cols); } void transform_row2ampereT(char * A, char *out, int rows, int cols){ transformRowToFormat<COL_AMPERE, 1>(A, out, rows, cols); } +void extractOutliers_turing(char * A, int *idx, char *out, int idx_size, int rows, int cols){ extractOutliers<COL_TURING>(A, idx, out, idx_size, rows, cols); } +void extractOutliers_ampere(char * A, int *idx, char *out, int idx_size, int rows, int cols){ extractOutliers<COL_AMPERE>(A, idx, out, idx_size, rows, cols); } + int igemmlt_turing_32(cublasLtHandle_t ltHandle, int m, int n, int k, const int8_t *A, const int8_t *B, void *C, float *row_scale, int lda, int ldb, int ldc) { return igemmlt<COL_TURING, 32, 0>(ltHandle, m, n, k, A, B, C, row_scale, lda, ldb, ldc); } @@ -280,6 +283,9 @@ extern "C" void cspmm_coo_very_sparse_naive_int8(int *max_count, int *max_idx, int *offset_rowidx, int *rowidx, int *colidx, half *values, signed char *B, half *out, float *dequant_stats, int nnz_rows, int nnz, int rowsA, int rowsB, int colsB) { spmm_coo_very_sparse_naive_int8(max_count, max_idx, offset_rowidx, rowidx, colidx, values, B, out, dequant_stats, nnz_rows, nnz, rowsA, rowsB, colsB); } + void cextractOutliers_turing(char * A, int *idx, char *out, int idx_size, int rows, int cols){ extractOutliers_turing(A, idx, out, idx_size, rows, cols); } + void cextractOutliers_ampere(char * A, int *idx, char *out, int idx_size, int rows, int cols){ extractOutliers_ampere(A, idx, out, idx_size, rows, cols); } + #endif void cquantize_blockwise_cpu_fp32(float *code, float *A, float *absmax, unsigned char *out, const int n){ quantize_cpu(code, A, absmax, out, n); } void cdequantize_blockwise_cpu_fp32(float *code, unsigned char *A, float *absmax, float *out, const int n){ dequantize_cpu(code, A, absmax, out, n); } |