Skip to content

P
Paddle

PaddlePaddle / Paddle
大约 1 年 前同步成功

通知 2297
Star 20931
Fork 5422
P
Paddle

体验新版 GitCode，发现更多精彩内容 >>

提交 b3115fb0 编写于 作者: L liaogang
浏览文件
操作

Add SetDeviceId in memcpy

上级 0897d18a
隐藏空白更改
内联 并排
Showing with 9 addition and 4 deletion
paddle/memory/memory.cc
浏览文件 @ b3115fb0
...@@ -86,18 +86,22 @@ size_t Used<platform::GPUPlace>(platform::GPUPlace place) { ...@@ -86,18 +86,22 @@ size_t Used<platform::GPUPlace>(platform::GPUPlace place) {
} }
template <> template <>
void Copy<platform::CPUPlace, platform::GPUPlace>(platform::CPUPlace, void* dst, void Copy<platform::CPUPlace, platform::GPUPlace>(platform::CPUPlace dst_place,
platform::GPUPlace, void* dst,
platform::GPUPlace src_place,
const void* src, size_t num, const void* src, size_t num,
cudaStream_t stream) { cudaStream_t stream) {
platform::SetDeviceId(src_place.device);
platform::GpuMemcpyAsync(dst, src, num, cudaMemcpyDeviceToHost, stream); platform::GpuMemcpyAsync(dst, src, num, cudaMemcpyDeviceToHost, stream);
} }
template <> template <>
void Copy<platform::GPUPlace, platform::CPUPlace>(platform::GPUPlace, void* dst, void Copy<platform::GPUPlace, platform::CPUPlace>(platform::GPUPlace dst_place,
platform::CPUPlace, void* dst,
platform::CPUPlace src_place,
const void* src, size_t num, const void* src, size_t num,
cudaStream_t stream) { cudaStream_t stream) {
platform::SetDeviceId(dst_place.device);
platform::GpuMemcpyAsync(dst, src, num, cudaMemcpyHostToDevice, stream); platform::GpuMemcpyAsync(dst, src, num, cudaMemcpyHostToDevice, stream);
} }
...@@ -108,6 +112,7 @@ void Copy<platform::GPUPlace, platform::GPUPlace>(platform::GPUPlace dst_place, ...@@ -108,6 +112,7 @@ void Copy<platform::GPUPlace, platform::GPUPlace>(platform::GPUPlace dst_place,
const void* src, size_t num, const void* src, size_t num,
cudaStream_t stream) { cudaStream_t stream) {
if (dst_place == src_place) { if (dst_place == src_place) {
platform::SetDeviceId(src_place.device);
platform::GpuMemcpyAsync(dst, src, num, cudaMemcpyDeviceToDevice, stream); platform::GpuMemcpyAsync(dst, src, num, cudaMemcpyDeviceToDevice, stream);
} else { } else {
platform::GpuMemcpyPeer(dst, dst_place.device, src, src_place.device, num, platform::GpuMemcpyPeer(dst, dst_place.device, src, src_place.device, num,
......
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或