vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Fix ROCM build by relaxing constness

Browse source
This commit is contained in:
KerfuffleV2 2023-11-01 19:24:11 -06:00
parent 0eb332a10f
commit 9ea1c9bf99
1 changed files with 3 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

6
ggml-cuda.cu
View file

@ -7369,9 +7369,9 @@ static void ggml_cuda_mul_mat_mat_batched_cublas(const ggml_tensor * src0, const
CUBLAS_CHECK(
cublasGemmBatchedEx(g_cublas_handles[id], CUBLAS_OP_T, CUBLAS_OP_N,
ne01, ne11, ne10,
&alpha_f16, (const void * const *) (ptrs_as + 0*ne23), CUDA_R_16F, nb01/sizeof(half),
(const void * const *) (ptrs_as + 1*ne23), CUDA_R_16F, nb11/sizeof(float),
&beta_f16, ( void ** ) (ptrs_as + 2*ne23), CUDA_R_16F, ne01,
&alpha_f16, (const void **) (ptrs_as + 0*ne23), CUDA_R_16F, nb01/sizeof(half),
(const void **) (ptrs_as + 1*ne23), CUDA_R_16F, nb11/sizeof(float),
&beta_f16, ( void **) (ptrs_as + 2*ne23), CUDA_R_16F, ne01,
ne23,
CUBLAS_COMPUTE_16F,
CUBLAS_GEMM_DEFAULT_TENSOR_OP));