diff options
Diffstat (limited to 'csrc/kernels.cuh')
| -rw-r--r-- | csrc/kernels.cuh | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/csrc/kernels.cuh b/csrc/kernels.cuh index 2447494..bdf61b2 100644 --- a/csrc/kernels.cuh +++ b/csrc/kernels.cuh @@ -111,7 +111,7 @@ template <typename T, int SPMM_ITEMS, int BITS> __global__ void kspmm_coo_very_s template <int ITEMS_PER_THREAD, int SUBTILE_ROWS, int THREADS>__global__ void kdequant_mm_int32_fp16( int *__restrict__ const A, float *__restrict__ const rowStats, float *__restrict__ const colStats, - half *out, float* newRowStats, float* newcolStats, const int numRows, const int numCols, const int tileCols, const int n); + half *out, float* newRowStats, float* newcolStats, half * __restrict__ const bias, const int numRows, const int numCols, const int tileCols, const int n); template<typename T, int THREADS, int ITEMS_PER_THREAD, int TILE_ROWS, int TILE_COLS, int SPARSE_DECOMP> __global__ void kgetColRowStats(T * __restrict__ A, float *rowStats, float *colStats, int * nnz_count_row, float nnz_threshold, int rows, int cols, int tiledRows, int tiledCols); template <int THREADS, int ITEMS_PER_THREAD, int TILE_ROWS, int TILE_COLS, int SPARSE_DECOMP> __global__ void kDoubleRowColQuant(half *__restrict__ const A, float *__restrict__ const rowStats, float * __restrict__ const colStats, char *out_col_normed, char *out_row_normed, int *rowidx, int *colidx, half *val, int * __restrict__ nnz_block_ptr, float threshold, int rows, int cols, int tiledCols); |