]> git.djapps.eu Git - pkg/ggml/sources/whisper.cpp/commit
CANN: fix CPU memory leak in CANN backend (llama/16549)
authorChenguang Li <redacted>
Mon, 13 Oct 2025 09:01:24 +0000 (17:01 +0800)
committerGeorgi Gerganov <redacted>
Wed, 15 Oct 2025 06:29:17 +0000 (09:29 +0300)
commit417ecdddc5a1919094965d85bdf4caa604d26288
treef3afcf9f0f001e264cc31c17cae661a4d5107dde
parentbfd88b8b6ee31ce7733e42f6107f0731a83cadc7
CANN: fix CPU memory leak in CANN backend (llama/16549)

This commit fixes a CPU-side memory leak issue in the CANN backend,
which occurred when intermediate aclTensorList objects were not properly
released after operator execution. The leak happened during repeated
invocations of CANN ops (e.g., FlashAttention), leading to increasing
host memory usage over time.

Proper resource cleanup (aclDestroyTensorList and related release logic)
has been added to ensure that all temporary tensors are correctly freed.
ggml/src/ggml-cann/aclnn_ops.cpp