optimize phi::CPUPlace in kernels/cpu/

Author: co63oc

paddle/phi/kernels/cpu/adam_kernel.cc

@@ -140,8 +140,7 @@ PADDLE_API void AdamDenseKernel(
   const T* grad_ptr = grad.data<T>();
 
   auto adam =
-      phi::jit::KernelFuncs<phi::jit::AdamTuple<T>, phi::CPUPlace>::Cache().At(
-          attr);
+      phi::jit::KernelFuncs<phi::jit::AdamTuple<T>, CPUPlace>::Cache().At(attr);
 
   static constexpr int64_t chunk_size = 512;
 

paddle/phi/kernels/cpu/adamw_kernel.cc

@@ -153,7 +153,7 @@ PADDLE_API void AdamwDenseKernel(
   const T* grad_ptr = grad.data<T>();
 
   auto adamw =
-      phi::jit::KernelFuncs<phi::jit::AdamWTuple<T>, phi::CPUPlace>::Cache().At(
+      phi::jit::KernelFuncs<phi::jit::AdamWTuple<T>, CPUPlace>::Cache().At(
           attr);
 
   static constexpr int64_t chunk_size = 512;

paddle/phi/kernels/cpu/crf_decoding_kernel.cc

@@ -41,9 +41,9 @@ void Decode(const Context& dev_ctx,
   DenseTensor track;
   track.Resize(emission_dims);
   int* track_value = dev_ctx.template Alloc<int>(&track);
-  auto ker = phi::jit::KernelFuncs<phi::jit::CRFDecodingTuple<T>,
-                                   phi::CPUPlace>::Cache()
-                 .At(tag_num);
+  auto ker =
+      phi::jit::KernelFuncs<phi::jit::CRFDecodingTuple<T>, CPUPlace>::Cache()
+          .At(tag_num);
   ker(static_cast<int>(seq_len), x, w, alpha_value, track_value, tag_num);
   T max_score = -std::numeric_limits<T>::max();
   int max_i = 0;

paddle/phi/kernels/cpu/eig.h

@@ -249,7 +249,7 @@ void MagmaEig(const Context& dev_ctx,
   // magma will modify original input, so copy to cpu at any case
   DenseTensor input_copy_cpu;
   input_copy_cpu.Resize(input.dims());
-  Copy(dev_ctx, input, phi::CPUPlace(), false, &input_copy_cpu);
+  Copy(dev_ctx, input, CPUPlace(), false, &input_copy_cpu);
 
   using RealT = typename phi::dtype::Real<T>;
   magma_vec_t jobvr = MagmaVec;
@@ -274,7 +274,7 @@ void MagmaEig(const Context& dev_ctx,
   phi::dtype::Real<T>* rwork_data = nullptr;
 
   rwork.Resize(common::make_ddim({lda * 2}));
-  auto cpu_place = phi::CPUPlace();
+  auto cpu_place = CPUPlace();
   phi::DeviceContextPool& pool = phi::DeviceContextPool::Instance();
   auto* cpu_ctx = static_cast<phi::CPUContext*>(pool.Get(cpu_place));
   rwork_data = (*cpu_ctx).template Alloc<phi::dtype::Real<T>>(&rwork);
@@ -345,7 +345,7 @@ void ApplyEigKernelMagma(const Context& dev_ctx,
 
   DenseTensor vectors_row_major_cpu;
   vectors_row_major_cpu.Resize(input.dims());
-  auto cpu_place = phi::CPUPlace();
+  auto cpu_place = CPUPlace();
   phi::DeviceContextPool& pool = phi::DeviceContextPool::Instance();
   auto* cpu_ctx = static_cast<phi::CPUContext*>(pool.Get(cpu_place));
   (*cpu_ctx).template Alloc<T>(&vectors_row_major_cpu);

paddle/phi/kernels/cpu/fusion_seqpool_concat_kernel.cc

@@ -55,8 +55,8 @@ void FusionSeqPoolConcatKernel(const Context& dev_ctx,
     attr.type = phi::jit::SeqPoolType::kSqrt;
   }
   auto seqpool =
-      phi::jit::KernelFuncs<phi::jit::SeqPoolTuple<T>, phi::CPUPlace>::Cache()
-          .At(attr);
+      phi::jit::KernelFuncs<phi::jit::SeqPoolTuple<T>, CPUPlace>::Cache().At(
+          attr);
   size_t n = ins.size();
   size_t dst_step_size = n * w;
   for (size_t i = 0; i < n; ++i) {

paddle/phi/kernels/cpu/index_select_impl.h

@@ -65,7 +65,7 @@ void IndexSelectInner(const Context& dev_ctx,
 
   DenseTensor index_cpu_copy;
   if (index.place().GetType() != AllocationType::CPU) {
-    Copy(dev_ctx, index, phi::CPUPlace(), true, &index_cpu_copy);
+    Copy(dev_ctx, index, CPUPlace(), true, &index_cpu_copy);
   }
   const IndexT* index_data = index.place().GetType() == AllocationType::CPU
                                  ? index.data<IndexT>()

paddle/phi/kernels/cpu/layer_norm_kernel.cc

@@ -133,8 +133,8 @@ void LayerNormKernel(const Context& dev_ctx,
   }
 
   auto ker =
-      phi::jit::KernelFuncs<phi::jit::LayerNormTuple<T>, phi::CPUPlace>::Cache()
-          .At(right);
+      phi::jit::KernelFuncs<phi::jit::LayerNormTuple<T>, CPUPlace>::Cache().At(
+          right);
   ker(x_tmp.data<T>(),
       out.data<T>(),
       mean_tmp.data<T>(),

paddle/phi/kernels/cpu/sgd_kernel.cc

@@ -35,8 +35,7 @@ void sgd_dense_param_dense_grad_impl(const DenseTensor& param,
   T* out_data = param_out->data<T>();
 
   auto sgd =
-      phi::jit::KernelFuncs<phi::jit::SgdTuple<T>, phi::CPUPlace>::Cache().At(
-          attr);
+      phi::jit::KernelFuncs<phi::jit::SgdTuple<T>, CPUPlace>::Cache().At(attr);
   sgd(lr, param_data, grad_data, &rows_idx, out_data, &attr);
 }
 
@@ -76,8 +75,7 @@ void sgd_dense_param_sparse_grad_impl(const DenseTensor& param,
   attr.selected_rows_size = static_cast<int>(grad_rows.size());
 
   auto sgd =
-      phi::jit::KernelFuncs<phi::jit::SgdTuple<T>, phi::CPUPlace>::Cache().At(
-          attr);
+      phi::jit::KernelFuncs<phi::jit::SgdTuple<T>, CPUPlace>::Cache().At(attr);
   sgd(lr, param_data, grad_data, rows_data, out_data, &attr);
 }