chore: import upstream snapshot with attribution

paddle/phi/kernels/gpu/full_kernel.cu

@@ -0,0 +1,185 @@
+/* Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License. */
+
+#include "paddle/phi/kernels/full_kernel.h"
+
+#include "paddle/phi/backends/gpu/gpu_context.h"
+#include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/kernels/funcs/elementwise_base.h"
+#include "paddle/phi/kernels/impl/full_with_tensor_kernel_impl.h"
+
+namespace phi {
+
+template <typename InT, typename OutT = InT>
+struct FullFunctor {
+  OutT value;
+
+  template <typename VType>
+  explicit inline FullFunctor(VType val) {
+    value = static_cast<OutT>(val);
+  }
+
+  __device__ __forceinline__ OutT operator()() const {
+    return static_cast<OutT>(value);
+  }
+};
+
+template <typename T, typename Context>
+void FullKernel(const Context& dev_ctx,
+                const IntArray& shape,
+                const Scalar& val,
+                DataType dtype,
+                DenseTensor* out) {
+  out->Resize(shape.GetData());
+  int64_t numel = out->numel();
+  dev_ctx.template Alloc<T>(out);
+
+  if (numel > 0) {
+    // in transformer model the numel of output will be zero.
+    std::vector<const DenseTensor*> inputs = {};
+    std::vector<DenseTensor*> outputs = {out};
+    // This function has no input, so the inputs.size() == 0. Use kUnary, but
+    // the data will not be loaded in the kernel because the number of
+    // parameters in the operator is 0
+    funcs::ElementwiseKernel<T>(
+        dev_ctx, inputs, &outputs, FullFunctor<T>(val.to<T>()));
+  }
+}
+
+template <typename T, typename Context>
+void FullLikeKernel(const Context& dev_ctx,
+                    const DenseTensor& x,
+                    const Scalar& val,
+                    DataType dtype,
+                    DenseTensor* out) {
+  std::vector<const DenseTensor*> inputs = {};
+  std::vector<DenseTensor*> outputs = {out};
+  dev_ctx.template Alloc<T>(out);
+  // This function has no input, so the inputs.size() == 0. Use kUnary, but the
+  // data will not be loaded in the kernel because the number of parameters in
+  // the operator is 0
+  int64_t numel = out->numel();
+
+  if (!std::is_same<T, complex64>::value &&
+      !std::is_same<T, complex128>::value && !std::is_same<T, int64_t>::value) {
+    auto value = val.to<double>();
+    using CommonType = typename std::common_type<
+        float,
+        typename std::conditional<std::is_same<T, float16>::value ||
+                                      std::is_same<T, bfloat16>::value,
+                                  float,
+                                  T>::type>::type;
+    auto common_type_value = static_cast<CommonType>(value);
+
+    // Check whether the filled value is valid
+    bool is_out_range = true;
+    if (std::isinf(value) || std::isnan(value)) {
+      is_out_range = false;
+    }
+
+    if ((common_type_value >=
+         static_cast<CommonType>(std::numeric_limits<T>::lowest())) &&
+        (common_type_value <=
+         static_cast<CommonType>(std::numeric_limits<T>::max()))) {
+      is_out_range = false;
+    }
+
+    PADDLE_ENFORCE_EQ(
+        is_out_range,
+        false,
+        common::errors::InvalidArgument(
+            "The filled value is out of range for target type, "
+            "current kernel type is %s, the range should between %f "
+            "and %f, but now value is %f.",
+            typeid(T).name(),
+            static_cast<CommonType>(std::numeric_limits<T>::lowest()),
+            static_cast<CommonType>(std::numeric_limits<T>::max()),
+            static_cast<float>(value)));
+
+    if (numel > 0) {
+      funcs::ElementwiseKernel<T>(
+          dev_ctx, inputs, &outputs, FullFunctor<T>(value));
+    }
+  } else {
+    if (numel > 0) {
+      funcs::ElementwiseKernel<T>(
+          dev_ctx, inputs, &outputs, FullFunctor<T>(val.to<T>()));
+    }
+  }
+}
+#ifdef _WIN32
+INSTANTIATE_FULL_KERNEL(float, GPUContext)
+INSTANTIATE_FULL_KERNEL(double, GPUContext)
+INSTANTIATE_FULL_KERNEL(int, GPUContext)
+INSTANTIATE_FULL_KERNEL(int64_t, GPUContext)
+#endif
+}  // namespace phi
+
+PD_REGISTER_KERNEL(full,
+                   GPU,
+                   ALL_LAYOUT,
+                   phi::FullKernel,
+                   float,
+                   double,
+                   int8_t,
+                   uint8_t,
+                   int16_t,
+                   int,
+                   int64_t,
+                   bool,
+                   phi::float8_e4m3fn,
+                   phi::float8_e5m2,
+                   phi::float16,
+                   phi::bfloat16,
+                   phi::complex64,
+                   phi::complex128) {}
+
+PD_REGISTER_KERNEL(full_like,
+                   GPU,
+                   ALL_LAYOUT,
+                   phi::FullLikeKernel,
+                   bool,
+                   float,
+                   double,
+                   int,
+                   int8_t,
+                   int64_t,
+                   int16_t,
+                   uint8_t,
+                   phi::float8_e4m3fn,
+                   phi::float16,
+                   phi::bfloat16,
+                   phi::complex64,
+                   phi::complex128) {
+  kernel->InputAt(0).SetBackend(phi::Backend::ALL_BACKEND);
+}
+
+PD_REGISTER_KERNEL(full_with_tensor,
+                   GPU,
+                   ALL_LAYOUT,
+                   phi::FullWithTensorKernel,
+                   float,
+                   double,
+                   int8_t,
+                   uint8_t,
+                   int16_t,
+                   int,
+                   int64_t,
+                   bool,
+                   phi::float16,
+                   phi::bfloat16,
+                   phi::complex64,
+                   phi::complex128) {
+  kernel->InputAt(0).SetBackend(phi::Backend::CPU);
+}