提交 b4ff2e43 编写于 作者: L liaogang

Remove GPUPlaceGuard

上级 7829bdbd
add_subdirectory(detail) add_subdirectory(detail)
cc_library(memory SRCS memory.cc) cc_library(memory SRCS memory.cc)
cc_library(memcpy SRCS memcpy.cc DEPS device_context) cc_library(memcpy SRCS memcpy.cc)
cc_library(paddle_memory cc_library(paddle_memory
DEPS DEPS
......
...@@ -35,7 +35,7 @@ void Copy<platform::CPUPlace, platform::GPUPlace>(platform::CPUPlace dst_place, ...@@ -35,7 +35,7 @@ void Copy<platform::CPUPlace, platform::GPUPlace>(platform::CPUPlace dst_place,
platform::GPUPlace src_place, platform::GPUPlace src_place,
const void* src, size_t num, const void* src, size_t num,
cudaStream_t stream) { cudaStream_t stream) {
platform::GPUPlaceGuard g(src_place.device); platform::SetDeviceId(src_place.device);
platform::GpuMemcpyAsync(dst, src, num, cudaMemcpyDeviceToHost, stream); platform::GpuMemcpyAsync(dst, src, num, cudaMemcpyDeviceToHost, stream);
} }
...@@ -45,7 +45,7 @@ void Copy<platform::GPUPlace, platform::CPUPlace>(platform::GPUPlace dst_place, ...@@ -45,7 +45,7 @@ void Copy<platform::GPUPlace, platform::CPUPlace>(platform::GPUPlace dst_place,
platform::CPUPlace src_place, platform::CPUPlace src_place,
const void* src, size_t num, const void* src, size_t num,
cudaStream_t stream) { cudaStream_t stream) {
platform::GPUPlaceGuard g(dst_place.device); platform::SetDeviceId(dst_place.device);
platform::GpuMemcpyAsync(dst, src, num, cudaMemcpyHostToDevice, stream); platform::GpuMemcpyAsync(dst, src, num, cudaMemcpyHostToDevice, stream);
} }
...@@ -56,7 +56,7 @@ void Copy<platform::GPUPlace, platform::GPUPlace>(platform::GPUPlace dst_place, ...@@ -56,7 +56,7 @@ 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::GPUPlaceGuard g(src_place.device); 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% .
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