diff options
| author | FantasyGmm <16450052+FantasyGmm@users.noreply.github.com> | 2023-12-22 23:11:12 +0800 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2023-12-22 17:11:12 +0200 |
| commit | a55876955b1a83464171de8d578d3ab062a7b62d (patch) | |
| tree | 38b9d438a1fc56f3d2f53966673f42a2c70ae2f6 /ggml-cuda.cu | |
| parent | 6724ef16573ec7ecce620be56cbbff145856b2fb (diff) | |
cuda : fix jetson compile error (#4560)
* fix old jetson compile error
* Update Makefile
* update jetson detect and cuda version detect
* update cuda marco define
* update makefile and cuda,fix some issue
* Update README.md
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update Makefile
* Update README.md
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Diffstat (limited to 'ggml-cuda.cu')
| -rw-r--r-- | ggml-cuda.cu | 7 |
1 files changed, 7 insertions, 0 deletions
diff --git a/ggml-cuda.cu b/ggml-cuda.cu index da8fd1e0..b124774a 100644 --- a/ggml-cuda.cu +++ b/ggml-cuda.cu @@ -90,6 +90,13 @@ #include <cuda_runtime.h> #include <cublas_v2.h> #include <cuda_fp16.h> +// CUDA 10.2 does not have these macro definitions. +#ifndef CUBLAS_TF32_TENSOR_OP_MATH +#define CUBLAS_TF32_TENSOR_OP_MATH CUBLAS_TENSOR_OP_MATH +#define CUBLAS_COMPUTE_16F CUDA_R_16F +#define CUBLAS_COMPUTE_32F CUDA_R_32F +#define cublasComputeType_t cudaDataType_t +#endif #endif // defined(GGML_USE_HIPBLAS) #include "ggml-cuda.h" |