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

restore rm code to fix hang issue

Browse source
This commit is contained in:
jianyuzh 2024-01-18 20:13:58 +08:00 • committed by Meng, Hengyu
parent d5f7d364f6
commit 35a0daaaa1
1 changed files with 22 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

22
ggml.c
View file

@ -16570,6 +16570,28 @@ static int ggml_get_n_tasks(struct ggml_tensor * node, int n_threads) {
//n_tasks = MIN(n_threads, MAX(1, nr0/128));
//printf("nr0 = %8d, nr1 = %8d, nr0*nr1 = %8d, n_tasks%d\n", nr0, nr1, nr0*nr1, n_tasks);
#if defined(GGML_USE_CUBLAS)
if (ggml_cuda_can_mul_mat(node->src[0], node->src[1], node)) {
n_tasks = 1; // TODO: this actually is doing nothing
// the threads are still spinning
}
#elif defined(GGML_USE_CLBLAST)
if (ggml_cl_can_mul_mat(node->src[0], node->src[1], node)) {
n_tasks = 1; // TODO: this actually is doing nothing
// the threads are still spinning
}
#elif defined(GGML_USE_SYCL)
if (ggml_sycl_can_mul_mat(node->src[0], node->src[1], node)) {
n_tasks = 1;
}
#endif
#if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS)
if (ggml_compute_forward_mul_mat_use_blas(node->src[0], node->src[1], node)) {
n_tasks = 1; // TODO: this actually is doing nothing
// the threads are still spinning
}
#endif
} break;
case GGML_OP_MUL_MAT_ID:
{