[PHI Decoupling] remove "paddle/fluid/platform/float16.h" and...

[PHI Decoupling] remove "paddle/fluid/platform/float16.h" and "paddle/fluid/platform/for_range.h" in phi. (#47817)

* rm "paddle/fluid/platform/float16.h" in phi

* rm "paddle/fluid/platform/for_range.h" in phi

paddle/phi/kernels/cpu/fill_grad_kernel.cc

@@ -25,6 +25,6 @@ PD_REGISTER_KERNEL(fill_grad,
                    double,
                    int64_t,
                    int,
-                   paddle::platform::float16,
+                   phi::dtype::float16,
                    paddle::platform::bfloat16,
                    bool) {}

paddle/phi/kernels/cpu/fill_kernel.cc

@@ -25,6 +25,6 @@ PD_REGISTER_KERNEL(fill,
                    double,
                    int64_t,
                    int,
-                   paddle::platform::float16,
+                   phi::dtype::float16,
                    paddle::platform::bfloat16,
                    bool) {}

paddle/phi/kernels/funcs/math_function.cc

@@ -27,9 +27,9 @@ limitations under the License. */
 #include <vector>
 
 #include "paddle/fluid/platform/bfloat16.h"
-#include "paddle/fluid/platform/float16.h"
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/common/data_type.h"
+#include "paddle/phi/common/float16.h"
 #include "paddle/phi/kernels/funcs/eigen/common.h"
 #include "paddle/phi/kernels/funcs/math_function_impl.h"
 #include "unsupported/Eigen/CXX11/Tensor"

paddle/phi/kernels/funcs/math_function.cu

@@ -17,9 +17,9 @@ limitations under the License. */
 #include "paddle/fluid/memory/malloc.h"
 #include "paddle/fluid/memory/memcpy.h"
 #include "paddle/fluid/platform/bfloat16.h"
-#include "paddle/fluid/platform/float16.h"
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/common/data_type.h"
+#include "paddle/phi/common/float16.h"
 #include "paddle/phi/kernels/funcs/blas/blas.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 #include "paddle/phi/kernels/funcs/math_function_impl.h"

paddle/phi/kernels/gpu/bmm_grad_kernel.cu

@@ -24,4 +24,4 @@ PD_REGISTER_KERNEL(bmm_grad,
                    phi::BmmGradKernel,
                    float,
                    double,
-                   paddle::platform::float16) {}
+                   phi::dtype::float16) {}

paddle/phi/kernels/gpu/bmm_kernel.cu

@@ -18,10 +18,5 @@
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/kernels/impl/bmm_kernel_impl.h"
 
-PD_REGISTER_KERNEL(bmm,
-                   GPU,
-                   ALL_LAYOUT,
-                   phi::BmmKernel,
-                   float,
-                   double,
-                   paddle::platform::float16) {}
+PD_REGISTER_KERNEL(
+    bmm, GPU, ALL_LAYOUT, phi::BmmKernel, float, double, phi::dtype::float16) {}

paddle/phi/kernels/gpu/cholesky_kernel.cu

@@ -23,10 +23,10 @@ limitations under the License. */
 #include <vector>
 
 #include "paddle/fluid/memory/memory.h"
-#include "paddle/fluid/platform/for_range.h"
 #include "paddle/phi/backends/dynload/cusolver.h"
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/kernels/funcs/for_range.h"
 
 namespace phi {
 
@@ -130,7 +130,7 @@ void CholeskyKernel(const Context& dev_ctx,
       upper ? CUBLAS_FILL_MODE_LOWER : CUBLAS_FILL_MODE_UPPER;
   // portf is inplace, thus copy the triangular part of the input matrices to
   // the output and set the other triangular part to 0 firstly
-  paddle::platform::ForRange<GPUContext> for_range(dev_ctx, tensor_size);
+  phi::funcs::ForRange<GPUContext> for_range(dev_ctx, tensor_size);
   if (upper) {
     MatrixBandPartFunctor<T> matrix_band_part_functor(m,
                                                       m,

paddle/phi/kernels/gpu/fill_grad_kernel.cu

@@ -26,6 +26,6 @@ PD_REGISTER_KERNEL(fill_grad,
                    double,
                    int64_t,
                    int,
-                   paddle::platform::float16,
+                   phi::dtype::float16,
                    paddle::platform::bfloat16,
                    bool) {}

paddle/phi/kernels/gpu/fill_kernel.cu

@@ -26,6 +26,6 @@ PD_REGISTER_KERNEL(fill,
                    double,
                    int64_t,
                    int,
-                   paddle::platform::float16,
+                   phi::dtype::float16,
                    paddle::platform::bfloat16,
                    bool) {}

paddle/phi/kernels/gpu/overlap_add_grad_kernel.cu

@@ -160,6 +160,6 @@ PD_REGISTER_KERNEL(overlap_add_grad,
                    int64_t,
                    float,
                    double,
-                   paddle::platform::float16,
+                   phi::dtype::float16,
                    paddle::platform::complex<float>,
                    paddle::platform::complex<double>) {}

paddle/phi/kernels/gpu/overlap_add_kernel.cu

@@ -146,6 +146,6 @@ PD_REGISTER_KERNEL(overlap_add,
                    int64_t,
                    float,
                    double,
-                   paddle::platform::float16,
+                   phi::dtype::float16,
                    paddle::platform::complex<float>,
                    paddle::platform::complex<double>) {}

paddle/phi/kernels/gpu/set_value_kernel.cu

@@ -28,7 +28,7 @@ PD_REGISTER_KERNEL(set_value,
                    int,
                    int64_t,
                    bool,
-                   paddle::platform::float16,
+                   phi::dtype::float16,
                    phi::dtype::complex<float>,
                    phi::dtype::complex<double>) {}
 PD_REGISTER_KERNEL(set_value_with_tensor,
@@ -40,6 +40,6 @@ PD_REGISTER_KERNEL(set_value_with_tensor,
                    int,
                    int64_t,
                    bool,
-                   paddle::platform::float16,
+                   phi::dtype::float16,
                    phi::dtype::complex<float>,
                    phi::dtype::complex<double>) {}

paddle/phi/kernels/gpudnn/conv_grad_kernel.cu

@@ -24,7 +24,6 @@
 #endif
 
 #include "paddle/fluid/platform/cudnn_workspace_helper.h"
-#include "paddle/fluid/platform/float16.h"
 #include "paddle/fluid/platform/profiler.h"
 #include "paddle/phi/common/bfloat16.h"
 #include "paddle/phi/common/float16.h"

paddle/phi/kernels/impl/atan2_grad_kernel_impl.h

@@ -14,9 +14,9 @@
 
 #pragma once
 
-#include "paddle/fluid/platform/for_range.h"
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/kernels/atan2_grad_kernel.h"
+#include "paddle/phi/kernels/funcs/for_range.h"
 
 namespace phi {
 

paddle/phi/kernels/impl/atan2_kernel_impl.h

@@ -14,9 +14,9 @@
 
 #pragma once
 
-#include "paddle/fluid/platform/for_range.h"
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/kernels/atan2_kernel.h"
+#include "paddle/phi/kernels/funcs/for_range.h"
 
 namespace phi {
 template <typename T>

paddle/phi/kernels/impl/cholesky_grad_kernel_impl.h

@@ -14,9 +14,9 @@ limitations under the License. */
 
 #pragma once
 
-#include "paddle/fluid/platform/for_range.h"
 #include "paddle/phi/kernels/cholesky_grad_kernel.h"
 #include "paddle/phi/kernels/funcs/blas/blas.h"
+#include "paddle/phi/kernels/funcs/for_range.h"
 
 namespace phi {
 
@@ -280,7 +280,7 @@ void CholeskyGradKernel(const Context& dev_ctx,
   blas.MatMul(l, trans_desc, l_grad, no_trans_desc, T(1), &middle, T(0));
 
   /*! phi.tril_().diagonal(0, -2, -1).mul_(0.5) */
-  paddle::platform::ForRange<Context> for_range(dev_ctx, tensor_size);
+  phi::funcs::ForRange<Context> for_range(dev_ctx, tensor_size);
   MatrixBandPartScaleEndFunctor<T> matrix_band_part_scale_end_functor(
       m,
       m,

paddle/phi/kernels/impl/conv_cudnn_impl.h

@@ -24,9 +24,9 @@
 #endif
 
 #include "paddle/fluid/platform/cudnn_workspace_helper.h"
-#include "paddle/fluid/platform/float16.h"
 #include "paddle/fluid/platform/profiler.h"
 #include "paddle/phi/backends/dynload/cudnn.h"
+#include "paddle/phi/common/float16.h"
 #include "paddle/phi/kernels/cpu/conv_util.h"
 #include "paddle/phi/kernels/funcs/batch_norm_utils.h"
 #include "paddle/phi/kernels/funcs/padding.h"

paddle/phi/kernels/impl/eye_kernel_impl.h

@@ -14,8 +14,8 @@
 
 #pragma once
 
-#include "paddle/fluid/platform/for_range.h"
 #include "paddle/phi/common/scalar.h"
+#include "paddle/phi/kernels/funcs/for_range.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 
 namespace phi {
@@ -48,7 +48,7 @@ void EyeKernel(const Context& ctx,
   phi::funcs::SetConstant<Context, T> set_zero;
   set_zero(ctx, out, static_cast<T>(0));
   int64_t num_eyes = (std::min)(rows, columns);
-  paddle::platform::ForRange<Context> for_range(ctx, num_eyes);
+  phi::funcs::ForRange<Context> for_range(ctx, num_eyes);
   EyeFunctor<T> functor(columns, out_data);
   for_range(functor);
 }

paddle/phi/kernels/impl/selu_grad_kernel_impl.h

@@ -28,7 +28,7 @@ void SeluGradKernel(const Context& dev_ctx,
   SeluGradFunctor<T> functor(
       out.data<T>(), dout.data<T>(), alpha, scale, dx_ptr);
   size_t limit = static_cast<size_t>(out.numel());
-  paddle::platform::ForRange<Context> for_range(dev_ctx, limit);
+  phi::funcs::ForRange<Context> for_range(dev_ctx, limit);
   for_range(functor);
 }
 }  // namespace phi

paddle/phi/kernels/impl/selu_kernel_impl.h

@@ -16,8 +16,8 @@
 #include <string>
 
 #include "paddle/fluid/operators/math.h"
-#include "paddle/fluid/platform/for_range.h"
 #include "paddle/phi/core/dense_tensor.h"
+#include "paddle/phi/kernels/funcs/for_range.h"
 
 namespace phi {
 
@@ -86,7 +86,7 @@ void SeluKernel(const Context& dev_ctx,
   auto out_ptr = dev_ctx.template Alloc<T>(out);
   SeluFunctor<T> functor(x.data<T>(), alpha, scale, out_ptr);
   size_t limit = static_cast<size_t>(x.numel());
-  paddle::platform::ForRange<Context> for_range(dev_ctx, limit);
+  phi::funcs::ForRange<Context> for_range(dev_ctx, limit);
   for_range(functor);
 }
 }  // namespace phi

paddle/phi/kernels/xpu/bmm_grad_kernel.cc

@@ -27,7 +27,7 @@ void MatMul(const Context& dev_ctx,
             DenseTensor* out) {
   dev_ctx.template Alloc<T>(out);
   xpu::Context* xpu_ctx = dev_ctx.x_context();
-  if (std::is_same<paddle::platform::float16, T>::value) {
+  if (std::is_same<phi::dtype::float16, T>::value) {
     MatMulXPUFunction<T, int16_t>(a, b, out, trans_a, trans_b, xpu_ctx);
   } else {
     if (std::getenv("XPU_PADDLE_FC_INT32") != nullptr) {
@@ -99,9 +99,6 @@ void BmmGradKernel(const Context& dev_ctx,
 
 }  // namespace phi
 
-PD_REGISTER_KERNEL(bmm_grad,
-                   XPU,
-                   ALL_LAYOUT,
-                   phi::BmmGradKernel,
-                   float,
-                   paddle::platform::float16) {}
+PD_REGISTER_KERNEL(
+    bmm_grad, XPU, ALL_LAYOUT, phi::BmmGradKernel, float, phi::dtype::float16) {
+}

paddle/phi/kernels/xpu/bmm_kernel.cc

@@ -62,7 +62,7 @@ void BmmKernel(const Context& dev_ctx,
           y_dims[1]));
 
   xpu::Context* xpu_ctx = dev_ctx.x_context();
-  if (std::is_same<paddle::platform::float16, T>::value) {
+  if (std::is_same<phi::dtype::float16, T>::value) {
     MatMulXPUFunction<T, int16_t>(x, y, out, trans_x, trans_y, xpu_ctx);
   } else {
     if (std::getenv("XPU_PADDLE_FC_INT32") != nullptr) {
@@ -77,4 +77,4 @@ void BmmKernel(const Context& dev_ctx,
 }  // namespace phi
 
 PD_REGISTER_KERNEL(
-    bmm, XPU, ALL_LAYOUT, phi::BmmKernel, float, paddle::platform::float16) {}
+    bmm, XPU, ALL_LAYOUT, phi::BmmKernel, float, phi::dtype::float16) {}

paddle/phi/kernels/xpu/xpu_api_wrapper.h

@@ -34,7 +34,7 @@ enum XPUFCCalcType {
 
 template <typename T>
 XPUFCCalcType FCCalcType() {
-  if (std::is_same<paddle::platform::float16, T>::value ||
+  if (std::is_same<phi::dtype::float16, T>::value ||
       std::is_same<float16, T>::value) {
     return XPUFCCalcType::FC_INT16;
   } else if (std::getenv("XPU_PADDLE_FC_INT32") != nullptr) {