]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
CANN: fix CPU memory leak in CANN backend (#16549)
authorChenguang Li <redacted>
Mon, 13 Oct 2025 09:01:24 +0000 (17:01 +0800)
committerGitHub <redacted>
Mon, 13 Oct 2025 09:01:24 +0000 (17:01 +0800)
commit56fc38b9655fbe1869d8bd6cfb269418196cea69
tree19b89deb1f3488d02c8a07e85b2bc6a709bb046b
parent1fb9504eb744969a990bfe4cfcf1d3d7a479541c
CANN: fix CPU memory leak in CANN backend (#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