diff options
Diffstat (limited to 'tensorflow/core/kernels/stack_ops.cc')
| -rw-r--r-- | tensorflow/core/kernels/stack_ops.cc | 68 |
1 files changed, 2 insertions, 66 deletions
diff --git a/tensorflow/core/kernels/stack_ops.cc b/tensorflow/core/kernels/stack_ops.cc index 2db3e5ef77..b4698a8053 100644 --- a/tensorflow/core/kernels/stack_ops.cc +++ b/tensorflow/core/kernels/stack_ops.cc @@ -40,9 +40,6 @@ namespace tensorflow { typedef Eigen::ThreadPoolDevice CPUDevice; typedef Eigen::GpuDevice GPUDevice; -#ifdef TENSORFLOW_USE_SYCL -typedef Eigen::SyclDevice SYCLDevice; -#endif // TENSORFLOW_USE_SYCL class Stack : public ResourceBase { public: @@ -185,10 +182,6 @@ class StackOp : public OpKernel { REGISTER_KERNEL_BUILDER(Name("Stack").Device(DEVICE_CPU), StackOp); REGISTER_KERNEL_BUILDER(Name("Stack").Device(DEVICE_GPU).HostMemory("handle"), StackOp); -#ifdef TENSORFLOW_USE_SYCL -REGISTER_KERNEL_BUILDER(Name("Stack").Device(DEVICE_SYCL).HostMemory("handle"), - StackOp); -#endif // TENSORFLOW_USE_SYCL template <typename Device> class StackPushOp : public AsyncOpKernel { @@ -220,11 +213,7 @@ class StackPushOp : public AsyncOpKernel { static constexpr int kCopyThreshold = 2048; static constexpr double kOccupancy = 0.7; if (swap_memory_ && !alloc_attrs.on_host() && - ( std::is_same<Device, GPUDevice>::value -#ifdef TENSORFLOW_USE_SYCL - || std::is_same<Device, SYCLDevice>::value -#endif // TENSORFLOW_USE_SYCL - ) && + std::is_same<Device, GPUDevice>::value && tensor.TotalBytes() > kCopyThreshold && stack->IsUsefulToSwap(tensor)) { DeviceContext* device_ctxt = ctx->op_device_context(); auto device = static_cast<tensorflow::Device*>(ctx->device()); @@ -300,31 +289,6 @@ REGISTER_GPU_HOST_KERNEL(bool); #undef REGISTER_GPU_HOST_KERNEL -#ifdef TENSORFLOW_USE_SYCL -#define REGISTER_SYCL_KERNEL(type) \ - REGISTER_KERNEL_BUILDER(Name("StackPush") \ - .Device(DEVICE_SYCL) \ - .HostMemory("handle") \ - .TypeConstraint<type>("T"), \ - StackPushOp<SYCLDevice>); - -TF_CALL_GPU_NUMBER_TYPES(REGISTER_SYCL_KERNEL); - -#define REGISTER_SYCL_HOST_KERNEL(type) \ - REGISTER_KERNEL_BUILDER(Name("StackPush") \ - .Device(DEVICE_SYCL) \ - .HostMemory("handle") \ - .HostMemory("elem") \ - .HostMemory("output") \ - .TypeConstraint<type>("T"), \ - StackPushOp<SYCLDevice>) - -REGISTER_SYCL_HOST_KERNEL(int32); -REGISTER_SYCL_HOST_KERNEL(bool); -#undef REGISTER_SYCL_KERNEL -#undef REGISTER_SYCL_HOST_KERNEL -#endif // TENSORFLOW_USE_SYCL - class StackPopOp : public AsyncOpKernel { public: explicit StackPopOp(OpKernelConstruction* context) : AsyncOpKernel(context) {} @@ -395,31 +359,6 @@ REGISTER_GPU_HOST_KERNEL(bool); #undef REGISTER_GPU_HOST_KERNEL -#ifdef TENSORFLOW_USE_SYCL -#define REGISTER_SYCL_KERNEL(type) \ - REGISTER_KERNEL_BUILDER(Name("StackPop") \ - .Device(DEVICE_SYCL) \ - .HostMemory("handle") \ - .TypeConstraint<type>("elem_type"), \ - StackPopOp) - -TF_CALL_GPU_NUMBER_TYPES(REGISTER_SYCL_KERNEL); - -#define REGISTER_SYCL_HOST_KERNEL(type) \ - REGISTER_KERNEL_BUILDER(Name("StackPop") \ - .Device(DEVICE_SYCL) \ - .HostMemory("handle") \ - .HostMemory("elem") \ - .TypeConstraint<type>("elem_type"), \ - StackPopOp) - -REGISTER_SYCL_HOST_KERNEL(int32); -REGISTER_SYCL_HOST_KERNEL(bool); - -#undef REGISTER_SYCL_KERNEL -#undef REGISTER_SYCL_HOST_KERNEL -#endif // TENSORFLOW_USE_SYCL - class StackCloseOp : public OpKernel { public: explicit StackCloseOp(OpKernelConstruction* context) : OpKernel(context) {} @@ -437,8 +376,5 @@ class StackCloseOp : public OpKernel { REGISTER_KERNEL_BUILDER(Name("StackClose").Device(DEVICE_CPU), StackCloseOp); REGISTER_KERNEL_BUILDER( Name("StackClose").Device(DEVICE_GPU).HostMemory("handle"), StackCloseOp); -#ifdef TENSORFLOW_USE_SYCL -REGISTER_KERNEL_BUILDER( - Name("StackClose").Device(DEVICE_SYCL).HostMemory("handle"), StackCloseOp); -#endif // TENSORFLOW_USE_SYCL + } // namespace tensorflow |