replica
/
llama.cpp
mirror of https://github.com/ggml-org/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

CANN: Replace aclrtMemsetSync with aclnnInplaceZero operator (#14002) 

Co-authored-by: luyuhong <luyuhong@kylinos.cn>
This commit is contained in:
luyhcsu 2025-07-04 11:50:07 +08:00 committed by GitHub
parent 28657a8229
commit 499a8f5a78
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 3 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

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

@ -67,6 +67,7 @@
#include <aclnnop/aclnn_pow.h>
#include <aclnnop/aclnn_grouped_matmul_v3.h>
#include <aclnnop/aclnn_fused_infer_attention_score_v2.h>
#include <aclnnop/aclnn_zero.h>
#include <float.h>
#include <cmath>
@ -804,10 +805,11 @@ static aclTensor* aclnn_zero(ggml_backend_cann_context& ctx, void* buffer,
nb[i] = nb[i - 1] * ne[i - 1];
}
ggml_cann_async_memset(ctx, buffer, n_bytes, 0);
aclTensor* zero =
ggml_cann_create_tensor(buffer, type, type_size, ne, nb, dims);
GGML_CANN_CALL_ACLNN_OP(ctx, InplaceZero, zero);
return zero;
GGML_UNUSED(n_bytes);
}
/**