Skip to content
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

Jump to bottom

Feature/transform ctx #4177

Merged
merged 2 commits into from
Sep 19, 2017
Merged

Feature/transform ctx #4177

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
87e4e25
Change Transform API
reyoung Sep 19, 2017
eeaccc8
Merge branch 'develop' of github.com:baidu/Paddle into feature/transf…
reyoung Sep 19, 2017
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
2 changes: 1 addition & 1 deletion paddle/platform/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,4 +24,4 @@ cc_library(device_context SRCS device_context.cc DEPS memory buddy_allocator
nv_test(device_context_test SRCS device_context_test.cc DEPS device_context gpu_info)

nv_test(cudnn_helper_test SRCS cudnn_helper_test.cc DEPS dynload_cuda)
nv_test(transform_test SRCS transform_test.cu DEPS paddle_memory place)
nv_test(transform_test SRCS transform_test.cu DEPS paddle_memory place device_context)
31 changes: 19 additions & 12 deletions paddle/platform/transform.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,48 +14,55 @@

#pragma once

#include "paddle/platform/device_context.h"
#include "paddle/platform/enforce.h"
#include "paddle/platform/hostdevice.h"
#include "paddle/platform/place.h"

#include <algorithm>
#include <type_traits>
#ifdef __NVCC__
#include <thrust/execution_policy.h>
#include <thrust/transform.h>
#include "paddle/platform/details/device_ptr_cast.h"
#endif

namespace paddle {
namespace platform {
// Transform on host or device. It provides the same API in std library.
template <typename Place, typename InputIter, typename OutputIter,
typename UnaryOperation>
void Transform(Place place, InputIter first, InputIter last, OutputIter result,
UnaryOperation op) {
template <typename InputIter, typename OutputIter, typename UnaryOperation>
void Transform(const DeviceContext& context, InputIter first, InputIter last,
OutputIter result, UnaryOperation op) {
auto place = context.GetPlace();
if (is_cpu_place(place)) {
std::transform(first, last, result, op);
} else {
#ifdef __NVCC__
auto& ctx = reinterpret_cast<const CUDADeviceContext&>(context);
using namespace details;
thrust::transform(DevPtrCast(first), DevPtrCast(last), DevPtrCast(result),
op);
thrust::transform(thrust::cuda::par.on(ctx.stream()), DevPtrCast(first),
DevPtrCast(last), DevPtrCast(result), op);
#else
PADDLE_THROW("Do not invoke `Transform<GPUPlace>` in .cc file");
#endif
}
}

template <typename Place, typename InputIter1, typename InputIter2,
typename OutputIter, typename BinaryOperation>
void Transform(Place place, InputIter1 first1, InputIter1 last1,
InputIter2 first2, OutputIter result, BinaryOperation op) {
template <typename InputIter1, typename InputIter2, typename OutputIter,
typename BinaryOperation>
void Transform(const DeviceContext& context, InputIter1 first1,
InputIter1 last1, InputIter2 first2, OutputIter result,
BinaryOperation op) {
auto place = context.GetPlace();
if (is_cpu_place(place)) {
std::transform(first1, last1, first2, result, op);
} else {
#ifdef __NVCC__
auto& ctx = reinterpret_cast<const CUDADeviceContext&>(context);
using namespace details;
thrust::transform(DevPtrCast(first1), DevPtrCast(last1), DevPtrCast(first2),
DevPtrCast(result), op);
thrust::transform(thrust::cuda::par.on(ctx.stream()), DevPtrCast(first1),
DevPtrCast(last1), DevPtrCast(first2), DevPtrCast(result),
op);
#else
PADDLE_THROW("Do not invoke `Transform<GPUPlace>` in .cc file");
#endif
Expand Down
13 changes: 9 additions & 4 deletions paddle/platform/transform_test.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,8 +36,9 @@ class Multiply {

TEST(Transform, CPUUnary) {
using namespace paddle::platform;
CPUDeviceContext ctx;
float buf[4] = {0.1, 0.2, 0.3, 0.4};
Transform(CPUPlace(), buf, buf + 4, buf, Scale<float>(10));
Transform(ctx, buf, buf + 4, buf, Scale<float>(10));
for (int i = 0; i < 4; ++i) {
ASSERT_NEAR(buf[i], static_cast<float>(i + 1), 1e-5);
}
Expand All @@ -47,10 +48,12 @@ TEST(Transform, GPUUnary) {
using namespace paddle::platform;
using namespace paddle::memory;
GPUPlace gpu0(0);
CUDADeviceContext ctx(gpu0);
float cpu_buf[4] = {0.1, 0.2, 0.3, 0.4};
float* gpu_buf = static_cast<float*>(Alloc(gpu0, sizeof(float) * 4));
Copy(gpu0, gpu_buf, CPUPlace(), cpu_buf, sizeof(cpu_buf));
Transform(gpu0, gpu_buf, gpu_buf + 4, gpu_buf, Scale<float>(10));
Transform(ctx, gpu_buf, gpu_buf + 4, gpu_buf, Scale<float>(10));
ctx.Wait();
Copy(CPUPlace(), cpu_buf, gpu0, gpu_buf, sizeof(cpu_buf));
Free(gpu0, gpu_buf);
for (int i = 0; i < 4; ++i) {
Expand All @@ -62,7 +65,7 @@ TEST(Transform, CPUBinary) {
using namespace paddle::platform;
using namespace paddle::memory;
int buf[4] = {1, 2, 3, 4};
Transform(CPUPlace(), buf, buf + 4, buf, buf, Multiply<int>());
Transform(CPUDeviceContext(), buf, buf + 4, buf, buf, Multiply<int>());
for (int i = 0; i < 4; ++i) {
ASSERT_EQ((i + 1) * (i + 1), buf[i]);
}
Expand All @@ -73,9 +76,11 @@ TEST(Transform, GPUBinary) {
using namespace paddle::memory;
int buf[4] = {1, 2, 3, 4};
GPUPlace gpu0(0);
CUDADeviceContext ctx(gpu0);
int* gpu_buf = static_cast<int*>(Alloc(gpu0, sizeof(buf)));
Copy(gpu0, gpu_buf, CPUPlace(), buf, sizeof(buf));
Transform(gpu0, gpu_buf, gpu_buf + 4, gpu_buf, gpu_buf, Multiply<int>());
Transform(ctx, gpu_buf, gpu_buf + 4, gpu_buf, gpu_buf, Multiply<int>());
ctx.Wait();
Copy(CPUPlace(), buf, gpu0, gpu_buf, sizeof(buf));
Free(gpu0, gpu_buf);
for (int i = 0; i < 4; ++i) {
Expand Down