LLM inference in C/C++
C++ ggml
cuda : warnings cleanup (#30420)
Adrien Gallouët30f7df1a831b
diff --git a/ggml/src/ggml-cuda/common.cuh b/ggml/src/ggml-cuda/common.cuhindex 928607c410..c6dc9dd622 100644--- a/ggml/src/ggml-cuda/common.cuh+++ b/ggml/src/ggml-cuda/common.cuh@@ -881,6 +881,7 @@ static __device__ __forceinline__ uint8_t ggml_cuda_fp32_to_ue4m3(float x) {const __nv_fp8_e4m3 xf(x);return xf.__x;#else+ GGML_UNUSED(x);NO_DEVICE_CODE; // Used only for NVFP4 Scales for Activations, only for Blackwell#endif // defined(BLACKWELL_MMA_AVAILABLE)}@@ -1448,10 +1449,10 @@ struct ggml_backend_cuda_context {std::string name;cudaEvent_t copy_event = nullptr;- cudaStream_t streams[GGML_CUDA_MAX_DEVICES][GGML_CUDA_MAX_STREAMS] = { { nullptr } };- cublasHandle_t cublas_handles[GGML_CUDA_MAX_DEVICES][GGML_CUDA_MAX_STREAMS] = {nullptr};- void * cublas_workspaces[GGML_CUDA_MAX_DEVICES][GGML_CUDA_MAX_STREAMS] = {nullptr};- size_t cublas_workspace_sizes[GGML_CUDA_MAX_DEVICES] = {0};+ cudaStream_t streams[GGML_CUDA_MAX_DEVICES][GGML_CUDA_MAX_STREAMS] = {};+ cublasHandle_t cublas_handles[GGML_CUDA_MAX_DEVICES][GGML_CUDA_MAX_STREAMS] = {};+ void * cublas_workspaces[GGML_CUDA_MAX_DEVICES][GGML_CUDA_MAX_STREAMS] = {};+ size_t cublas_workspace_sizes[GGML_CUDA_MAX_DEVICES] = {};int curr_stream_no = 0;