set device id of Place() to get GPUContext needed by LimitGridDim in ElemwiseGradBroadcast

[FlyingQianMM]

PR types

Bug fixes

PR changes

OPs

Describe

When use multiple gpu device to train a model, a error was raised as bellow:

NotImplementedError: 

--------------------------------------
C++ Traceback (most recent call last):
--------------------------------------
0   paddle::imperative::BasicEngine::Execute()
1   paddle::imperative::PreparedOp::Run(paddle::imperative::NameVariableWrapperMap const&, paddle::imperative::NameVariableWrapperMap const&, paddle::framework::AttributeMap const&, paddle::framework::AttributeMap const&)
2   phi::KernelImpl<void (*)(phi::GPUContext const&, phi::DenseTensor const&, phi::DenseTensor const&, phi::DenseTensor const&, paddle::optional<phi::DenseTensor const&>, paddle::optional<phi::DenseTensor const&>, int, phi::DenseTensor*, phi::DenseTensor*, phi::DenseTensor*), &(void phi::MultiplyDoubleGradKernel<float, phi::GPUContext>(phi::GPUContext const&, phi::DenseTensor const&, phi::DenseTensor const&, phi::DenseTensor const&, paddle::optional<phi::DenseTensor const&>, paddle::optional<phi::DenseTensor const&>, int, phi::DenseTensor*, phi::DenseTensor*, phi::DenseTensor*))>::Compute(phi::KernelContext*)
3   void phi::MultiplyDoubleGradKernel<float, phi::GPUContext>(phi::GPUContext const&, phi::DenseTensor const&, phi::DenseTensor const&, phi::DenseTensor const&, paddle::optional<phi::DenseTensor const&>, paddle::optional<phi::DenseTensor const&>, int, phi::DenseTensor*, phi::DenseTensor*, phi::DenseTensor*)
4   void phi::funcs::ElemwiseGradComputeWithBroadcast<float, phi::MulGradDX<float>, phi::MulGradDY<float>, float>(phi::GPUContext const&, phi::DDim const&, phi::DDim const&, phi::DenseTensor const&, phi::DenseTensor const&, phi::DenseTensor const&, phi::DenseTensor const&, int, phi::DenseTensor*, phi::DenseTensor*, phi::MulGradDX<float>, phi::MulGradDY<float>)
5   paddle::platform::DeviceContextPool::Get(phi::Place const&)
6   phi::enforce::EnforceNotMet::EnforceNotMet(phi::ErrorSummary const&, char const*, int)
7   phi::enforce::GetCurrentTraceBackString[abi:cxx11](bool)

----------------------
Error Message Summary:
----------------------
UnimplementedError: Place Place(gpu:0) is not supported. Please check that your paddle compiles with WITH_GPU, WITH_XPU, WITH_IPU, WITH_MLU or WITH_ASCEND_CL option or check that your train process set the correct device id if you use Executor. (at /ssd2/liuquanxiang/binary_search_pr/PaddleGAN_py37_102_docker_citest/paddle/paddle/fluid/platform/device_context.cc:139)

We find the the reason is that the current device id of a initialized Place() should be set to get a GPUContext needed by LimitGridDim in ElemwiseGradBroadcast, so we fix this bug.

[ZzSean]

LGTM