CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-14 11:07:10 +00:00
Code Issues Packages Projects Releases Wiki Activity

CANN: Improve device ID handling and aclnnArange checks (#16752)

Browse Source 
* cann: improve device ID handling and aclnnArange checks

- Stop relying on CANN's internal device ID retrieval; use a global variable instead.
- Enforce stricter dimension validation in aclnnArange for better compatibility across CANN versions.

* cann: use thread local var
This commit is contained in:
Chenguang Li
2025-10-28 10:54:53 +08:00
committed by GitHub
parent 463bbf20bf
commit 3479efd112
2 changed files with 18 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
ggml/src/ggml-cann/aclnn_ops.cpp
View File

@@ -2234,7 +2234,7 @@ static void aclnn_cache_init(ggml_backend_cann_context & ctx,
ACL_MEM_MALLOC_HUGE_FIRST));
acl_theta_scale_tensor = ggml_cann_create_tensor(ctx.rope_cache.theta_scale_cache, ACL_FLOAT, sizeof(float),
theta_scale_ne, theta_scale_nb, GGML_MAX_DIMS);
theta_scale_ne, theta_scale_nb, 1);
float start = 0;
float step = 1;
@@ -2251,7 +2251,7 @@ static void aclnn_cache_init(ggml_backend_cann_context & ctx,
yarn_ramp_allocator.alloc(theta_scale_length * sizeof(float));
void * yarn_ramp_buffer = yarn_ramp_allocator.get();
acl_yarn_ramp_tensor = ggml_cann_create_tensor(yarn_ramp_buffer, ACL_FLOAT, sizeof(float), theta_scale_ne,
theta_scale_nb, GGML_MAX_DIMS);
theta_scale_nb, 1);
float zero_value = 0, one_value = 1;
float denom_safe_value = MAX(0.001f, corr_dims[1] - corr_dims[0]);
aclScalar * low = aclCreateScalar(&corr_dims[0], aclDataType::ACL_FLOAT);