[XPU] fix XPU linking error on Clang (#71852)

Author: lj970926

paddle/phi/kernels/custom/c_embedding_grad_kernel.cc

@@ -31,7 +31,7 @@ void CEmbeddingGradKernel(const Context& dev_ctx,
  int64_t start_index,
  DenseTensor* w_grad) {
  w_grad->Resize(w.dims());
- dev_ctx.template Alloc(w_grad, w.dtype());
+ dev_ctx.Alloc(w_grad, w.dtype());
  const auto& index_type = ids.dtype();
  if (index_type == phi::DataType::INT32 ||
  index_type == phi::DataType::INT64) {

paddle/phi/kernels/xpu/c_concat_kernel.cc

@@ -58,7 +58,7 @@ void CConcatKernel(const Context& dev_ctx,
  phi::DDim temp_out_dims = x->dims();
  temp_out_dims[0] *= nranks;
  temp_out.Resize(temp_out_dims);
- dev_ctx.template Alloc(&temp_out, x->dtype());
+ dev_ctx.Alloc(&temp_out, x->dtype());
 
  XPUStream stream = nullptr;
  phi::distributed::BKCLCommContext* comm_ctx = nullptr;
@@ -86,7 +86,7 @@ void CConcatKernel(const Context& dev_ctx,
 
  phi::funcs::ConcatFunctor<phi::XPUContext, T> functor;
  out->Resize(out_dims);
- dev_ctx.template Alloc(out, x->dtype());
+ dev_ctx.Alloc(out, x->dtype());
  functor(dev_ctx, inputs, axis, out);
 #else
  PADDLE_THROW(common::errors::PreconditionNotMet(

paddle/phi/kernels/xpu/c_embedding_kernel_grad.cc

@@ -29,7 +29,7 @@ void CEmbeddingGradKernel(const Context& dev_ctx,
  int64_t start_index,
  DenseTensor* w_grad) {
  w_grad->Resize(w.dims());
- dev_ctx.template Alloc(w_grad, w.dtype());
+ dev_ctx.Alloc(w_grad, w.dtype());
  T* table_grad_data = static_cast<T*>(w_grad->data());
  using XPUType = typename XPUTypeTrait<T>::Type;
 

paddle/phi/kernels/xpu/c_softmax_with_cross_entropy_kernel.cc

@@ -161,8 +161,8 @@ struct CSoftmaxWithCrossEntropyFunctor<phi::XPUContext, T> {
  stream = dev_ctx.stream();
 
  // allocate memory on device.
- dev_ctx.template Alloc(softmax, logits->dtype());
- dev_ctx.template Alloc(loss, logits->dtype());
+ dev_ctx.Alloc(softmax, logits->dtype());
+ dev_ctx.Alloc(loss, logits->dtype());
 
  const auto& logits_dims = logits->dims();
 

paddle/phi/kernels/xpu/c_split_kernel.cc

@@ -59,7 +59,7 @@ void CSplitKernel(const Context& dev_ctx,
 
  dims[dims_size - 1] /= nranks;
  out->Resize(dims);
- dev_ctx.template Alloc(out, x.dtype());
+ dev_ctx.Alloc(out, x.dtype());
 
  std::vector<XPUType*> output_list(nranks, nullptr);
  output_list.at(rank) = reinterpret_cast<XPUType*>(out->data<T>());

paddle/phi/kernels/xpu/embedding_grad_kernel.cc

@@ -93,8 +93,7 @@ void EmbeddingSparseGradKernel(const Context& ctx,
  std::vector<int64_t> ids;
  DenseTensor ids_cpu;
  ids_cpu.Resize(input.dims());
- ctx.template HostAlloc(
- &ids_cpu, input.dtype(), input.numel() * sizeof(int64_t));
+ ctx.HostAlloc(&ids_cpu, input.dtype(), input.numel() * sizeof(int64_t));
  if (input.dtype() == phi::DataType::INT64) {
  phi::Copy(ctx, input, CPUPlace(), false, &ids_cpu);
 

paddle/phi/kernels/xpu/reduce_sum_grad_kernel.cc

@@ -33,7 +33,7 @@ void ReduceSumGradKernel(const Context& dev_ctx,
  using XPUType = typename XPUTypeTrait<T>::Type;
  reduce_all = recompute_reduce_all(x, dims_arr, reduce_all);
  auto dims = dims_arr.GetData();
- dev_ctx.template Alloc(x_grad, x.dtype());
+ dev_ctx.Alloc(x_grad, x.dtype());
  auto* out_data = reinterpret_cast<const XPUType*>(out_grad.data());
  auto* x_grad_data = reinterpret_cast<XPUType*>(x_grad->data());
  const auto& input_dim_size = x.dims().size();