Tencent / Tencent/ncnn

GPU内存资源没有释放

Open
#4,367 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
C++
Stars
23.8k
Forks
4.5k
Avg merge
2d 20h
Merged PRs (30d)
37

Description

error log | 日志或报错信息 | ログ

context | 编译/运行环境 | バックグラウンド

版本:ncnn-20220721-android-vulkan
运行环境:Android 10,MTK

how to reproduce | 复现步骤 | 再現方法

1.加载网络,模型

ncnn::Option opt;
opt.lightmode = true;
opt.num_threads = 4;
opt.blob_allocator = &g_blob_pool_allocator;
opt.workspace_allocator = &g_workspace_pool_allocator;

// use vulkan compute
int gpu_count = ncnn::get_gpu_count();
if (gpu_count != 0)
    opt.use_vulkan_compute = true;
if(squeezenet==nullptr){
    squeezenet = new ncnn::Net();
}
squeezenet->opt = opt;

2.卸载网络,模型

g_blob_pool_allocator.clear();
g_workspace_pool_allocator.clear();
squeezenet_words.clear();
if(squeezenet!=nullptr){
    squeezenet->clear();
    delete squeezenet;
    squeezenet = nullptr;
}
ncnn::destroy_gpu_instance();

3.期望:

如果只是用CPU:
opt.use_vulkan_compute = false;
资源卸载没有问题;

但是如果使用了GPU:
opt.use_vulkan_compute = true;
内存会有2m没有释放。

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Reproduce the Android 10 MTK case using the shown ncnn::Option settings, comparing CPU mode with Vulkan compute enabled. Trace g_blob_pool_allocator.clear(), g_workspace_pool_allocator.clear(), squeezenet->clear(), deletion, and ncnn::destroy_gpu_instance(); done means the reported 2 MB is released after unloading the model.

Written by the indexing model from the issue text.

Assessment

Tech stack
android, cpp
Domain
machine-learning, mobile-dev
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.