diff options
| author | Kawrakow <48489457+ikawrakow@users.noreply.github.com> | 2024-07-27 07:55:01 +0200 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2024-07-27 07:55:01 +0200 |
| commit | 154e0d75fccf1784fe9ff6fd76a630b66563da3d (patch) | |
| tree | 81ce6dbb5b1900c1aa78a879f0593c694cab9d27 /ggml/src/ggml-cuda/cpy.cuh | |
| parent | 0684c3e9c70d49323b4fc517128cbe222cab7f96 (diff) | |
Merge mainline llama.cpp (#3)
* Merging mainline - WIP
* Merging mainline - WIP
AVX2 and CUDA appear to work.
CUDA performance seems slightly (~1-2%) lower as it is so often
the case with llama.cpp/ggml after some "improvements" have been made.
* Merging mainline - fix Metal
* Remove check
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
Diffstat (limited to 'ggml/src/ggml-cuda/cpy.cuh')
| -rw-r--r-- | ggml/src/ggml-cuda/cpy.cuh | 9 |
1 files changed, 9 insertions, 0 deletions
diff --git a/ggml/src/ggml-cuda/cpy.cuh b/ggml/src/ggml-cuda/cpy.cuh new file mode 100644 index 00000000..79616742 --- /dev/null +++ b/ggml/src/ggml-cuda/cpy.cuh @@ -0,0 +1,9 @@ +#include "common.cuh" + +#define CUDA_CPY_BLOCK_SIZE 32 + +void ggml_cuda_cpy(ggml_backend_cuda_context & ctx, const ggml_tensor * src0, ggml_tensor * src1); + +void ggml_cuda_dup(ggml_backend_cuda_context & ctx, ggml_tensor * dst); + +void* ggml_cuda_cpy_fn(const ggml_tensor * src0, ggml_tensor * src1); |