From 316a0495335638c296fdacae9dadc618a9c35b3b Mon Sep 17 00:00:00 2001 From: pidack Date: Thu, 29 Aug 2024 10:36:33 +0800 Subject: [PATCH] add restrict for dst --- ggml/src/ggml-cuda/ssm_conv.cu | 2 +- ggml/src/ggml-cuda/ssm_scan.cu | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/ggml/src/ggml-cuda/ssm_conv.cu b/ggml/src/ggml-cuda/ssm_conv.cu index ec25376c55c9f..abb0177f09fac 100644 --- a/ggml/src/ggml-cuda/ssm_conv.cu +++ b/ggml/src/ggml-cuda/ssm_conv.cu @@ -5,7 +5,7 @@ static __global__ void ssm_conv_f32( const float * __restrict__ src0, const float * __restrict__ src1, const int src0_nb0, const int src0_nb1, const int src0_nb2, const int src1_nb1, - float * dst, + float * __restrict__ dst, const int dst_nb0, const int dst_nb1, const int dst_nb2, const int nc, const int ncs, const int nr, const int n_t, const int n_s) { diff --git a/ggml/src/ggml-cuda/ssm_scan.cu b/ggml/src/ggml-cuda/ssm_scan.cu index cf08f6e0f9f19..cc8dac9e6e159 100644 --- a/ggml/src/ggml-cuda/ssm_scan.cu +++ b/ggml/src/ggml-cuda/ssm_scan.cu @@ -10,7 +10,7 @@ static __global__ void ssm_scan_f32( const int src3_nb1, const int src4_nb1, const int src4_nb2, const int src5_nb1, const int src5_nb2, - float * dst, + float * __restrict__ dst, const int nc, const int nr, const int n_t, const int n_s) { // const int row = blockIdx.x*blockDim.y + threadIdx.y;