apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-07 17:41:52 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[cuda][win] add win cuda version check and cuda setting FAQ (#140)

Browse Source 
* [win][cuda] add cuda version check on windows

* [docs][win] add cuda setting FAQ

* [docs][win] add cuda setting FAQ
This commit is contained in:
DefTruth
2022-08-22 15:35:58 +08:00
committed by GitHub
parent e35d6dcfd4
commit 1ac1c35e95
2 changed files with 38 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

23
docs/FAQ.md Normal file
View File

@@ -0,0 +1,23 @@
# FastDeploy FAQ 文档
## 1. 在Windows 10 配置 CUDA v11.2 环境变量
FastDeploy Windows 10 x64 的 GPU 版本需要依赖 CUDA 11.2，在安装完 CUDA v11.2 之后，需要设置`CUDA_DIRECTORY`、`CUDA_HOME`、`CUDA_PATH`和`CUDA_ROOT`中**任意一个**环境变量，这样FastDeploy才能链接到相关的库。有两种方式设置环境变量，通过终端命令行设置以及在系统环境变量中设置。
- 方式一: 终端命令行设置。该方式只在当前终端有效。Windows菜单打开`x64 Native Tools Command Prompt for VS 2019`命令工具，假设你需要在该终端运行类似`python infer_ppyoloe.py`的命令。
```bat
% 选择以下任意一个环境变量设置即可 %
set CUDA_PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2
set CUDA_HOME=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2
set CUDA_ROOT=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2
set CUDA_DIRECTORY=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2
```
- 方式二: 系统环境变量设置。该方式会修改系统环境变量。设置步骤为：
- (1) 打开 "设置->系统->关于"
- (2) 找到 "高级系统设置"，点击打开
- (3) 点击右下角的 "环境变量设置"
- (4) 注意，在 "系统变量" 一栏右下角点击 "新建"，如果已有相关的环境变量，则只需确认路径是否正确
- (5) 设置`CUDA_DIRECTORY`、`CUDA_HOME`、`CUDA_PATH`和`CUDA_ROOT`中**任意一个**环境变量
- (6) 根据以下提示来设置环境变量，并点击确认
```text
变量名(N): CUDA_DIRECTORY、CUDA_HOME、CUDA_PATH和CUDA_ROOT中任意一个
变量值(V): 类似 C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2
```

17
fastdeploy/c_lib_wrap.py.in
View File

@@ -56,9 +56,10 @@ def add_cuda_shared_lib_dir_windows():
# paths. User should set it manually if the
# cuda toolkit is not locate in the default
# path we assume.
base_url = "https://github.com/PaddlePaddle/FastDeploy/blob/"
default_cuda_dir = get_default_cuda_directory()
default_cuda_version = os.path.basename(default_cuda_dir) # v11.2
cuda_shared_lib_dir = os.path.join(default_cuda_dir, "bin")
# TODO: add FAQ docs reference.
if not os.path.exists(cuda_shared_lib_dir):
# try to get cuda directory from user's local env
custom_cuda_dir = "NOTFOUNDED"
@@ -72,8 +73,20 @@ def add_cuda_shared_lib_dir_windows():
logging.warnings.warn(f"\n--- FastDeploy was built with gpu, \
\n--- but the default cuda directory does not exists. \
\n--- Please setup one of {custom_cuda_envs} manually, \
\n--- this path should look like: {default_cuda_dir}")
\n--- this path should look like: {default_cuda_dir}. \
\n--- Check FAQ: {base_url + 'develop/docs/FAQ.md'}")
return
# check cuda version
custom_cuda_version = os.path.basename(custom_cuda_dir) # v11.2
if default_cuda_version != custom_cuda_version:
logging.warnings.warn(
f"\n--- FastDeploy was built with CUDA version {default_cuda_version}, \
\n--- but found custom CUDA version {custom_cuda_version} at {custom_cuda_dir} \
\n--- Please setup one of {custom_cuda_envs} manually, \
\n--- this path should look like: {default_cuda_dir}. \
\n--- Check FAQ: {base_url + 'develop/docs/FAQ.md'}")
return
# path to cuda dlls
cuda_shared_lib_dir = os.path.join(custom_cuda_dir, "bin")
add_dll_search_dir(cuda_shared_lib_dir)
print(f"[FastDeploy][CUDA]: Found valid cuda directroy and added it: -> {cuda_shared_lib_dir}")