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

Support different data type between input and output #32823

Merged
merged 2 commits into from
May 10, 2021
Merged

Support different data type between input and output #32823

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
04811a4
Support different data type between input and output
ZzSean May 10, 2021
2424597
support abs
ZzSean May 10, 2021
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
97 changes: 66 additions & 31 deletions paddle/fluid/operators/abs_op.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,44 +13,79 @@
// limitations under the License.

#include "paddle/fluid/operators/abs_op.h"
#include "paddle/fluid/operators/elementwise/elementwise_op_impl.cu.h"
#include "paddle/fluid/platform/complex128.h"
#include "paddle/fluid/platform/complex64.h"
#include "paddle/fluid/platform/float16.h"

namespace paddle {
namespace operators {

template <typename T, typename Enable = void>
struct CudaAbsFunctor;

template <typename T>
struct CudaAbsFunctor<T, math::Complex<T, math::Real<T>>> {
__device__ __forceinline__ math::Real<T> operator()(const T* args) const {
return abs(args[0]);
}
};

template <typename T>
struct CudaAbsFunctor<T, math::NoComplex<T, math::Real<T>>> {
__device__ __forceinline__ T operator()(const T* args) const {
return std::abs(args[0]);
}
};

template <typename T>
class AbsKernel<platform::CUDADeviceContext, T>
: public framework::OpKernel<T> {
public:
void Compute(const framework::ExecutionContext& context) const override {
const Tensor* x = context.Input<Tensor>("X");
Tensor* out = context.Output<Tensor>("Out");
out->mutable_data<math::Real<T>>(context.GetPlace());

auto& dev_ctx =
context.template device_context<platform::CUDADeviceContext>();
std::vector<const framework::Tensor*> ins = {x};
std::vector<framework::Tensor*> outs = {out};
auto functor = CudaAbsFunctor<T>();
LaunchElementwiseCudaKernel<ElementwiseType::kUnary, T, math::Real<T>>(
dev_ctx, ins, &outs, functor);
}
};

} // namespace operators
} // namespace paddle

namespace ops = paddle::operators;
namespace plat = paddle::platform;

REGISTER_OP_CUDA_KERNEL(
abs, ops::AbsKernel<paddle::platform::CUDADeviceContext, float>,
ops::AbsKernel<paddle::platform::CUDADeviceContext, double>,
ops::AbsKernel<paddle::platform::CUDADeviceContext, int>,
ops::AbsKernel<paddle::platform::CUDADeviceContext, int64_t>,
ops::AbsKernel<paddle::platform::CUDADeviceContext,
paddle::platform::float16>,
ops::AbsKernel<paddle::platform::CUDADeviceContext,
paddle::platform::complex64>,
ops::AbsKernel<paddle::platform::CUDADeviceContext,
paddle::platform::complex128>);
abs, ops::AbsKernel<plat::CUDADeviceContext, float>,
ops::AbsKernel<plat::CUDADeviceContext, double>,
ops::AbsKernel<plat::CUDADeviceContext, int>,
ops::AbsKernel<plat::CUDADeviceContext, int64_t>,
ops::AbsKernel<plat::CUDADeviceContext, plat::float16>,
ops::AbsKernel<plat::CUDADeviceContext, plat::complex64>,
ops::AbsKernel<plat::CUDADeviceContext, plat::complex128>);

REGISTER_OP_CUDA_KERNEL(
abs_grad, ops::AbsGradKernel<paddle::platform::CUDADeviceContext, float>,
ops::AbsGradKernel<paddle::platform::CUDADeviceContext, double>,
ops::AbsGradKernel<paddle::platform::CUDADeviceContext, int>,
ops::AbsGradKernel<paddle::platform::CUDADeviceContext, int64_t>,
ops::AbsGradKernel<paddle::platform::CUDADeviceContext,
paddle::platform::float16>,
ops::AbsGradKernel<paddle::platform::CUDADeviceContext,
paddle::platform::complex64>,
ops::AbsGradKernel<paddle::platform::CUDADeviceContext,
paddle::platform::complex128>);
abs_grad, ops::AbsGradKernel<plat::CUDADeviceContext, float>,
ops::AbsGradKernel<plat::CUDADeviceContext, double>,
ops::AbsGradKernel<plat::CUDADeviceContext, int>,
ops::AbsGradKernel<plat::CUDADeviceContext, int64_t>,
ops::AbsGradKernel<plat::CUDADeviceContext, plat::float16>,
ops::AbsGradKernel<plat::CUDADeviceContext, plat::complex64>,
ops::AbsGradKernel<plat::CUDADeviceContext, plat::complex128>);

REGISTER_OP_CUDA_KERNEL(
abs_grad_grad,
ops::AbsDoubleGradKernel<paddle::platform::CUDADeviceContext, float>,
ops::AbsDoubleGradKernel<paddle::platform::CUDADeviceContext, double>,
ops::AbsDoubleGradKernel<paddle::platform::CUDADeviceContext, int>,
ops::AbsDoubleGradKernel<paddle::platform::CUDADeviceContext, int64_t>,
ops::AbsDoubleGradKernel<paddle::platform::CUDADeviceContext,
paddle::platform::float16>,
ops::AbsDoubleGradKernel<paddle::platform::CUDADeviceContext,
paddle::platform::complex64>,
ops::AbsDoubleGradKernel<paddle::platform::CUDADeviceContext,
paddle::platform::complex128>);
abs_grad_grad, ops::AbsDoubleGradKernel<plat::CUDADeviceContext, float>,
ops::AbsDoubleGradKernel<plat::CUDADeviceContext, double>,
ops::AbsDoubleGradKernel<plat::CUDADeviceContext, int>,
ops::AbsDoubleGradKernel<plat::CUDADeviceContext, int64_t>,
ops::AbsDoubleGradKernel<plat::CUDADeviceContext, plat::float16>,
ops::AbsDoubleGradKernel<plat::CUDADeviceContext, plat::complex64>,
ops::AbsDoubleGradKernel<plat::CUDADeviceContext, plat::complex128>);
16 changes: 8 additions & 8 deletions paddle/fluid/operators/activation_op.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1315,8 +1315,8 @@ class ActivationCudaKernel
for (auto& attr : attrs) {
*attr.second = ctx.Attr<float>(attr.first);
}
LaunchElementwiseCudaKernel<ElementwiseType::kUnary, T>(dev_ctx, ins, &outs,
functor);
LaunchElementwiseCudaKernel<ElementwiseType::kUnary, T, T>(dev_ctx, ins,
&outs, functor);
}
};

Expand Down Expand Up @@ -1345,17 +1345,17 @@ class ActivationGradCudaKernel
if (static_cast<int>(Functor::FwdDeps()) == static_cast<int>(kDepOut)) {
// Only need forward output Out
ins.push_back(out);
LaunchElementwiseCudaKernel<ElementwiseType::kBinary, T>(dev_ctx, ins,
&outs, functor);
LaunchElementwiseCudaKernel<ElementwiseType::kBinary, T, T>(
dev_ctx, ins, &outs, functor);
} else if (static_cast<int>(Functor::FwdDeps()) ==
static_cast<int>(kDepX)) {
// Only need forward input X
ins.push_back(x);
LaunchElementwiseCudaKernel<ElementwiseType::kBinary, T>(dev_ctx, ins,
&outs, functor);
LaunchElementwiseCudaKernel<ElementwiseType::kBinary, T, T>(
dev_ctx, ins, &outs, functor);
} else {
LaunchElementwiseCudaKernel<ElementwiseType::kUnary, T>(dev_ctx, ins,
&outs, functor);
LaunchElementwiseCudaKernel<ElementwiseType::kUnary, T, T>(
dev_ctx, ins, &outs, functor);
}
}
};
Expand Down
2 changes: 1 addition & 1 deletion paddle/fluid/operators/elementwise/elementwise_add_op.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -45,7 +45,7 @@ struct SameDimsElemwiseAdd<platform::CUDADeviceContext, T> {
framework::Tensor* z) {
std::vector<const framework::Tensor*> ins = {x, y};
std::vector<framework::Tensor*> outs = {z};
LaunchElementwiseCudaKernel<ElementwiseType::kBinary, T>(
LaunchElementwiseCudaKernel<ElementwiseType::kBinary, T, T>(
ctx.template device_context<platform::CUDADeviceContext>(), ins, &outs,
CudaAddFunctor<T>());
}
Expand Down
100 changes: 54 additions & 46 deletions paddle/fluid/operators/elementwise/elementwise_op_impl.cu.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,77 +49,81 @@ int GetVectorizedSizeImpl(const T *pointer) {
return 1;
}

template <typename T>
template <typename InT, typename OutT>
int GetVectorizedSize(const std::vector<const framework::Tensor *> &ins,
const std::vector<framework::Tensor *> &outs) {
int vec_size = 4;
for (auto iter = ins.begin(); iter != ins.end(); ++iter) {
vec_size =
std::min<int>(vec_size, GetVectorizedSizeImpl((*iter)->data<T>()));
std::min<int>(vec_size, GetVectorizedSizeImpl((*iter)->data<InT>()));
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

这个循环的写法其实可以简写成如下：

for (auto in : ins) {
  vec_size =
        std::min<int>(vec_size, GetVectorizedSizeImpl(in->data<InT>()));
}

}
for (auto iter = outs.begin(); iter != outs.end(); ++iter) {
vec_size =
std::min<int>(vec_size, GetVectorizedSizeImpl((*iter)->data<T>()));
std::min<int>(vec_size, GetVectorizedSizeImpl((*iter)->data<OutT>()));
}
return vec_size;
}

template <ElementwiseType ET, int VecSize, typename T>
template <ElementwiseType ET, int VecSize, typename InT, typename OutT>
struct ElementwiseDataWrapper {
T *out;
const T *in0;
const T *in1;
__device__ ElementwiseDataWrapper(T *out, const T *in0,
const T *in1 = nullptr)
OutT *out;
const InT *in0;
const InT *in1;
__device__ ElementwiseDataWrapper(OutT *out, const InT *in0,
const InT *in1 = nullptr)
: out(out), in0(in0), in1(in1) {}

using VecType = CudaAlignedVector<T, VecSize>;
using InVecType = CudaAlignedVector<InT, VecSize>;
using OutVecType = CudaAlignedVector<OutT, VecSize>;

inline __device__ void load_vector(VecType args[], int idx) {
const VecType *x_vec = reinterpret_cast<const VecType *>(in0);
inline __device__ void load_vector(InVecType args[], int idx) {
const InVecType *x_vec = reinterpret_cast<const InVecType *>(in0);
args[0] = x_vec[idx];
if (ET == ElementwiseType::kBinary) {
const VecType *y_vec = reinterpret_cast<const VecType *>(in1);
const InVecType *y_vec = reinterpret_cast<const InVecType *>(in1);
args[1] = y_vec[idx];
}
}

inline __device__ void load_scalar(T args[], int idx) {
inline __device__ void load_scalar(InT args[], int idx) {
args[0] = in0[idx];
if (ET == ElementwiseType::kBinary) {
args[1] = in1[idx];
}
}

inline __device__ void store_vector(VecType res, int idx) {
VecType *out_vec = reinterpret_cast<VecType *>(out);
inline __device__ void store_vector(OutVecType res, int idx) {
OutVecType *out_vec = reinterpret_cast<OutVecType *>(out);
out_vec[idx] = res;
}

inline __device__ void store_scalar(T res, int idx) { out[idx] = res; }
inline __device__ void store_scalar(OutT res, int idx) { out[idx] = res; }
};

template <ElementwiseType ET, int VecSize, typename T, typename Functor>
template <ElementwiseType ET, int VecSize, typename InT, typename OutT,
typename Functor>
__device__ void VectorizedKernelImpl(
ElementwiseDataWrapper<ET, VecSize, T> data, Functor func, int tid) {
using VecType = CudaAlignedVector<T, VecSize>;
VecType ins_vec[ET];
VecType out_vec;
T *ins_ptr[ET];
T *out_ptr;
ElementwiseDataWrapper<ET, VecSize, InT, OutT> data, Functor func,
int tid) {
using InVecType = CudaAlignedVector<InT, VecSize>;
using OutVecType = CudaAlignedVector<OutT, VecSize>;
InVecType ins_vec[ET];
OutVecType out_vec;
InT *ins_ptr[ET];
OutT *out_ptr;
#pragma unroll
for (int i = 0; i < ET; ++i) {
ins_ptr[i] = reinterpret_cast<T *>(&(ins_vec[i]));
ins_ptr[i] = reinterpret_cast<InT *>(&(ins_vec[i]));
}
out_ptr = reinterpret_cast<T *>(&out_vec);
out_ptr = reinterpret_cast<OutT *>(&out_vec);

// load
data.load_vector(ins_vec, tid);

// compute
#pragma unroll
for (int i = 0; i < VecSize; ++i) {
T ins[ET];
InT ins[ET];
#pragma unroll
for (int j = 0; j < ET; ++j) {
ins[j] = ins_ptr[j][i];
Expand All @@ -131,11 +135,13 @@ __device__ void VectorizedKernelImpl(
data.store_vector(out_vec, tid);
}

template <ElementwiseType ET, int VecSize, typename T, typename Functor>
__device__ void ScalarKernelImpl(ElementwiseDataWrapper<ET, VecSize, T> data,
Functor func, int start, int remain) {
T ins[ET];
T out;
template <ElementwiseType ET, int VecSize, typename InT, typename OutT,
typename Functor>
__device__ void ScalarKernelImpl(
ElementwiseDataWrapper<ET, VecSize, InT, OutT> data, Functor func,
int start, int remain) {
InT ins[ET];
OutT out;

for (int i = 0; i < remain; ++i) {
int idx = start + i;
Expand All @@ -148,45 +154,47 @@ __device__ void ScalarKernelImpl(ElementwiseDataWrapper<ET, VecSize, T> data,
}
}

template <ElementwiseType ET, int VecSize, typename T, typename Functor>
__global__ void VectorizedKernel(const T *__restrict__ in0,
const T *__restrict__ in1, T *out, int size,
Functor func) {
template <ElementwiseType ET, int VecSize, typename InT, typename OutT,
typename Functor>
__global__ void VectorizedKernel(const InT *__restrict__ in0,
const InT *__restrict__ in1, OutT *out,
int size, Functor func) {
int tid = blockIdx.x * blockDim.x + threadIdx.x;
int remain = size - VecSize * tid;
remain = remain > 0 ? remain : 0;
auto data = ElementwiseDataWrapper<ET, VecSize, T>(out, in0, in1);
auto data = ElementwiseDataWrapper<ET, VecSize, InT, OutT>(out, in0, in1);
if (remain >= VecSize) {
VectorizedKernelImpl(data, func, tid);
} else {
ScalarKernelImpl(data, func, tid * VecSize, remain);
}
}

template <ElementwiseType ET, typename T, typename Functor>
__global__ void ScalarKernel(const T *__restrict__ in0,
const T *__restrict__ in1, T *out, int size,
template <ElementwiseType ET, typename InT, typename OutT, typename Functor>
__global__ void ScalarKernel(const InT *__restrict__ in0,
const InT *__restrict__ in1, OutT *out, int size,
Functor func) {
auto data = ElementwiseDataWrapper<ET, 1, T>(out, in0, in1);
auto data = ElementwiseDataWrapper<ET, 1, InT, OutT>(out, in0, in1);
int tid = blockIdx.x * blockDim.x + threadIdx.x;
int remain = tid < size ? 1 : 0;
ScalarKernelImpl(data, func, tid, remain);
}

template <ElementwiseType ET, typename T, typename Functor>
template <ElementwiseType ET, typename InT, typename OutT, typename Functor>
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

可以设置默认值OutT = InT吗？

void LaunchElementwiseCudaKernel(
const platform::CUDADeviceContext &ctx,
const std::vector<const framework::Tensor *> &ins,
std::vector<framework::Tensor *> *outs, Functor func) {
// calculate the max vec_size for all ins and outs
auto size = ins[0]->numel();
int vec_size = GetVectorizedSize<T>(ins, *outs);
int vec_size = GetVectorizedSize<InT, OutT>(ins, *outs);
int block_size = ELEMENTWISE_BLOCK_SIZE;
int grid_size =
((size + vec_size - 1) / vec_size + block_size - 1) / block_size;
const T *in0 = ins[0]->data<T>();
const T *in1 = (ET == ElementwiseType::kBinary) ? ins[1]->data<T>() : nullptr;
T *out = (*outs)[0]->data<T>();
const InT *in0 = ins[0]->data<InT>();
const InT *in1 =
(ET == ElementwiseType::kBinary) ? ins[1]->data<InT>() : nullptr;
OutT *out = (*outs)[0]->data<OutT>();
// cuda kernel
auto stream = ctx.stream();
switch (vec_size) {
Expand Down