Skip to content

make range functions respect current stream #21619

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

Closed
ngimel wants to merge 2 commits into from
Closed

make range functions respect current stream #21619

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
9600392
make range functions respect current stream
Jun 10, 2019
75afec0
try fixing ROCm build
Jun 11, 2019
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
18 changes: 14 additions & 4 deletions aten/src/ATen/native/cuda/RangeFactories.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,11 +2,13 @@
#include <ATen/NativeFunctions.h>
#include <ATen/AccumulateType.h>
#include <ATen/cuda/Exceptions.h>
#include <ATen/cuda/CUDAContext.h>
#include <cmath>
#include <limits>

#include <thrust/device_ptr.h>
#include <thrust/sequence.h>
#include <thrust/execution_policy.h>

namespace at {
namespace native {
Expand Down Expand Up @@ -56,7 +58,9 @@ Tensor& linspace_cuda_out(Tensor& result, Scalar start, Scalar end, int64_t step
scalar_t step = (scalar_end - scalar_start) / static_cast<scalar_t>(steps - 1);
LinspaceOp<scalar_t> linspace_method(scalar_start, step);
thrust::device_ptr<scalar_t> data_(r.data<scalar_t>());
thrust::tabulate(data_, data_ + steps, linspace_method);
cudaStream_t stream = at::cuda::getCurrentCUDAStream();
auto policy = thrust::cuda::par.on(stream);
thrust::tabulate(policy, data_, data_ + steps, linspace_method);
});
}

Expand Down Expand Up @@ -87,7 +91,9 @@ Tensor& logspace_cuda_out(Tensor& result, Scalar start, Scalar end, int64_t step
scalar_t step = (scalar_end - scalar_start) / static_cast<scalar_t>(steps - 1);
LogspaceOp<scalar_t> logspace_method(scalar_start, step, scalar_base);
thrust::device_ptr<scalar_t> data_(r.data<scalar_t>());
thrust::tabulate(data_, data_ + steps, logspace_method);
cudaStream_t stream = at::cuda::getCurrentCUDAStream();
auto policy = thrust::cuda::par.on(stream);
thrust::tabulate(policy, data_, data_ + steps, logspace_method);
});
}

Expand Down Expand Up @@ -117,8 +123,10 @@ Tensor& range_cuda_out(Tensor& result, Scalar start, Scalar end, Scalar step) {
}
Tensor r = result.is_contiguous() ? result : result.contiguous();
LinspaceOp<scalar_t, accscalar_t> linspace_method(xstart, xstep);
cudaStream_t stream = at::cuda::getCurrentCUDAStream();
auto policy = thrust::cuda::par.on(stream);
thrust::device_ptr<scalar_t> data_ptr(r.data<scalar_t>());
thrust::tabulate(data_ptr, data_ptr + size, linspace_method);
thrust::tabulate(policy, data_ptr, data_ptr + size, linspace_method);

if (!result.is_contiguous()) {
result.copy_(r);
Expand Down Expand Up @@ -168,8 +176,10 @@ Tensor& arange_cuda_out(Tensor& result, Scalar start, Scalar end, Scalar step) {
}
Tensor r = result.is_contiguous() ? result : result.contiguous();
LinspaceOp<scalar_t, accscalar_t> linspace_method(xstart, xstep);
cudaStream_t stream = at::cuda::getCurrentCUDAStream();
auto policy = thrust::cuda::par.on(stream);
thrust::device_ptr<scalar_t> data_ptr(r.data<scalar_t>());
thrust::tabulate(data_ptr, data_ptr + size, linspace_method);
thrust::tabulate(policy, data_ptr, data_ptr + size, linspace_method);

if (!result.is_contiguous()) {
result.copy_(r);
Expand Down