From b90c80bdbf17e298533b16d71296327d46de1390 Mon Sep 17 00:00:00 2001 From: JohannesGaessler Date: Sat, 8 Jul 2023 22:53:43 +0200 Subject: [PATCH] Add __restrict__ to dequantize_mul_mat kernels --- ggml-cuda.cu | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/ggml-cuda.cu b/ggml-cuda.cu index 8ba0830d0..7227646ee 100644 --- a/ggml-cuda.cu +++ b/ggml-cuda.cu @@ -1485,7 +1485,8 @@ static __global__ void dequantize_mul_mat_vec(const void * vx, const dfloat * y, template static __global__ void dequantize_mul_mat( - const void * vx, const float * y, float * dst, const int ncols_x, const int nrows_x, const int ncols_y, const int nrows_dst) { + const void * __restrict__ vx, const float * __restrict__ y, float * __restrict__ dst, + const int ncols_x, const int nrows_x, const int ncols_y, const int nrows_dst) { const int nrows_y = ncols_x; const int ncols_dst = ncols_y;