diff options
Diffstat (limited to 'csrc/kernels.cu')
| -rw-r--r-- | csrc/kernels.cu | 7 |
1 files changed, 7 insertions, 0 deletions
diff --git a/csrc/kernels.cu b/csrc/kernels.cu index 1c3e723..78170d0 100644 --- a/csrc/kernels.cu +++ b/csrc/kernels.cu @@ -2592,10 +2592,17 @@ __global__ void kspmm_coo_very_sparse_naive(int *max_count, int *max_idx, int *o } } +template <int FORMAT> __global__ void kExtractOutliers(char *A, int *idx, char *out, int rowsA, int colsA, int tiledRowsA, int tiledColsA) +{ +} + //============================================================== // TEMPLATE DEFINITIONS //============================================================== +template __global__ void kExtractOutliers<COL_TURING>(char *A, int *idx, char *out, int rowsA, int colsA, int tiledRowsA, int tiledColsA); +template __global__ void kExtractOutliers<COL_AMPERE>(char *A, int *idx, char *out, int rowsA, int colsA, int tiledRowsA, int tiledColsA); + template __global__ void kspmm_coo_very_sparse_naive<half, 8, 16>(int *max_count, int *max_idx, int *offset_rowidx, int *rowidx, int *colidx, half *values, half *B, half *out, float *dequant_stats, int nnz, int rowsA, int rowsB, int colsB); template __global__ void kspmm_coo_very_sparse_naive<half, 16, 16>(int *max_count, int *max_idx, int *offset_rowidx, int *rowidx, int *colidx, half *values, half *B, half *out, float *dequant_stats, int nnz, int rowsA, int rowsB, int colsB); template __global__ void kspmm_coo_very_sparse_naive<half, 32, 16>(int *max_count, int *max_idx, int *offset_rowidx, int *rowidx, int *colidx, half *values, half *B, half *out, float *dequant_stats, int nnz, int rowsA, int rowsB, int colsB); |