add restrict for dst

ggerganov · Aug 29, 2024 · 316a049 · 316a049
1 parent 5999d6d
commit 316a049
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 2 deletions.
diff --git 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
@@ -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;