cuda: fixes races in mmid and mmf (#28475)

2026-09-06 19:45:01 +08:00
parent 9e0e220594
commit 73a43d1f69
4 changed files with 27 additions and 3 deletions
@@ -121,6 +121,12 @@
 #    define GGML_CUDA_USE_PDL
 #endif  // !defined(GGML_USE_HIP) && !defined(GGML_USE_MUSA) && (CUDART_VERSION >= 12030 || (!(defined(_MSC_VER) && !defined(__clang__)) && CUDART_VERSION >= 11080))

+static __device__ __forceinline__ void ggml_cuda_syncwarp() {
+#ifndef GGML_USE_HIP
+    __syncwarp();
+#endif // GGML_USE_HIP
+}
+
 static __device__ __forceinline__ void ggml_cuda_pdl_sync() {
 #if defined(GGML_CUDA_USE_PDL) && defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= GGML_CUDA_CC_HOPPER
    cudaGridDependencySynchronize();