From 9268dc9d887a3d54cd1f008dcb628aaa5b5bd90a Mon Sep 17 00:00:00 2001 From: Tim Dettmers Date: Mon, 25 Jul 2022 19:30:37 -0700 Subject: Some progress on build script; added multi-cuda install script. --- csrc/kernels.cuh | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'csrc/kernels.cuh') diff --git a/csrc/kernels.cuh b/csrc/kernels.cuh index cbfbeba..4e65e96 100644 --- a/csrc/kernels.cuh +++ b/csrc/kernels.cuh @@ -107,7 +107,7 @@ template __global__ void kPercentileCl __global__ void kHistogramScatterAdd2D(float* histogram, int *index1, int *index2, float *src, const int maxidx1, const int n); -template __global__ void kspmm_coo_very_sparse_naive(int *max_count, int *max_idx, int *offset_rowidx, int *rowidx, int *colidx, half *values, T *B, half *out, float *dequant_stats, int nnz, int rowsA, int rowsB, int colsB); +template __global__ void kspmm_coo_very_sparse_naive(int *max_count, int *max_idx, int *offset_rowidx, int *rowidx, int *colidx, half *values, T *B, half *out, float * __restrict__ const dequant_stats, int nnz, int rowsA, int rowsB, int colsB); template __global__ void kdequant_mm_int32_fp16( int *__restrict__ const A, float *__restrict__ const rowStats, float *__restrict__ const colStats, -- cgit v1.2.3