From e760cd49bd7415f18acc29c1309e35d8756c0a15 Mon Sep 17 00:00:00 2001 From: bssrdf Date: Wed, 19 Nov 2025 07:47:11 -0500 Subject: [PATCH] fix CI --- ggml/src/ggml-cuda/conv2d-implicit.cu | 3 +++ 1 file changed, 3 insertions(+) diff --git a/ggml/src/ggml-cuda/conv2d-implicit.cu b/ggml/src/ggml-cuda/conv2d-implicit.cu index 917f3a6b1e..602bc37a0e 100644 --- a/ggml/src/ggml-cuda/conv2d-implicit.cu +++ b/ggml/src/ggml-cuda/conv2d-implicit.cu @@ -871,10 +871,13 @@ static __global__ void conv2d_implicit_kernel(const half * __restrict__ input, prepareIteratorA(thread_row, masks_a, element_offset_a, param); + +#ifdef CP_ASYNC_AVAILABLE unsigned int iter_src_idx = thread_row * param.weightKOffset; unsigned int iter_dst_idx = thread_row * TILE_COLS_VECTORIZED + thread_col; unsigned int krow_idx = thread_row + blockIdx.x * BN; const int ITER_SRC_STEPS = ROW_STEP * param.weightKOffset; +#endif // prefetch the first block tile of A,B into shared memory