Skip to content

[release/3.4] fix: use ThrustAllocator in argsort 1D path to avoid implicit cudaStr… #78733

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
risemeup1111 wants to merge 1 commit into
base: release/3.4
Choose a base branch
from
+13 −4
Open

[release/3.4] fix: use ThrustAllocator in argsort 1D path to avoid implicit cudaStr… #78733

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 13 additions & 4 deletions paddle/phi/kernels/gpu/argsort_kernel.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,6 +21,7 @@
#include "paddle/phi/backends/gpu/gpu_context.h"
#include "paddle/phi/backends/gpu/gpu_info.h"
#include "paddle/phi/backends/gpu/gpu_launch_config.h"
#include "paddle/phi/common/memory_utils.h"
#include "paddle/phi/core/kernel_registry.h"
#include "paddle/phi/kernels/funcs/blas/blas.h"
#include "paddle/phi/kernels/funcs/cub.h"
Expand Down Expand Up @@ -283,9 +284,13 @@ void PerSort(const GPUContext& dev_ctx,
bool stable,
bool descending) {
#ifdef PADDLE_WITH_CUDA
const auto& exec_policy = thrust::cuda::par.on(dev_ctx.stream());
phi::memory_utils::ThrustAllocator<cudaStream_t> allocator(dev_ctx.GetPlace(),
dev_ctx.stream());
const auto& exec_policy = thrust::cuda::par(allocator).on(dev_ctx.stream());
#else
const auto& exec_policy = thrust::hip::par.on(dev_ctx.stream());
phi::memory_utils::ThrustAllocator<hipStream_t> allocator(dev_ctx.GetPlace(),
dev_ctx.stream());
const auto& exec_policy = thrust::hip::par(allocator).on(dev_ctx.stream());
#endif
if (stable) {
if (descending) {
Expand Down Expand Up @@ -349,9 +354,13 @@ void ArgsortKernel(const Context& dev_ctx,
T* out_data = dev_ctx.template Alloc<T>(output);
int64_t* ids_data = dev_ctx.template Alloc<int64_t>(indices);
#ifdef PADDLE_WITH_CUDA
const auto& exec_policy = thrust::cuda::par.on(dev_ctx.stream());
phi::memory_utils::ThrustAllocator<cudaStream_t> allocator(
dev_ctx.GetPlace(), dev_ctx.stream());
const auto& exec_policy = thrust::cuda::par(allocator).on(dev_ctx.stream());
#else
const auto& exec_policy = thrust::hip::par.on(dev_ctx.stream());
phi::memory_utils::ThrustAllocator<hipStream_t> allocator(
dev_ctx.GetPlace(), dev_ctx.stream());
const auto& exec_policy = thrust::hip::par(allocator).on(dev_ctx.stream());
#endif
auto cu_stream = dev_ctx.stream();
thrust::sequence(exec_policy, ids_data, ids_data + size);
Expand Down
Loading