问题描述 Issue Description
- git下载飞浆2.6.0源码
- 编译GPU版本飞浆
cmake .. -GNinja -DWITH_GPU=ON -DWITH_UNITY_BUILD=ON -DCUDA_ARCH_BIN=35
- 执行编译
ninja
- 编译过程中多次出现错误
../paddle/phi/kernels/funcs/weight_only_gemv.cu(707): error: more than one conversion function from "half" to a built-in type applies:
function "__half::operator float() const"
function "__half::operator short() const"
function "__half::operator unsigned short() const"
function "__half::operator int() const"
function "__half::operator unsigned int() const"
function "__half::operator long long() const"
function "__half::operator unsigned long long() const"
function "__half::operator __nv_bool() const"
detected during:
instantiation of "void
- 编译的运行日志如下
log.txt
版本&环境信息 Version & Environment Information
Paddle version: N/A
Paddle With CUDA: N/A
OS: Windows 10
GCC version: N/A
Clang version: N/A
CMake version: version 3.17.0
Libc version: N/A
Python version: 3.8.19
CUDA version: 11.2.152
Build cuda_11.2.r11.2/compiler.29618528_0
cuDNN version: 8.2.1
Nvidia driver version: 474.82
Nvidia driver List:
GPU 0: Quadro K6000
2条答案
按热度按时间cpjpxq1n1#
可以参考下这个答案: NVIDIA/cutlass#4
laawzig22#
看文档好像是飞浆2.6版本开始不再支持GPU 算力在60以下了,我的GPU Quadro K6000 是 Kepler架构 sm_35,只能使用2.5版本的飞浆了