【PaddlePaddle Hackathon 3 No.33】为 Paddle 优化 erfinv op 在 GPU 上的计算性能 (#45057)

* erfinv

* fix some tiny issues

paddle/phi/kernels/gpu/erfinv_kernel.cu

@@ -13,9 +13,25 @@
 // limitations under the License.
 
 #include "paddle/phi/kernels/erfinv_kernel.h"
-
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/impl/erfinv_kernel_impl.h"
+#include "paddle/phi/kernels/funcs/elementwise_base.h"
+
+namespace phi {
+
+template <typename T>
+struct ErfinvFunctor {
+  HOSTDEVICE inline T operator()(const T x) const { return erfinv(x); }
+};
+
+template <typename T, typename Context>
+void ErfinvKernel(const Context& ctx, const DenseTensor& x, DenseTensor* out) {
+  ctx.template Alloc<T>(out);
+  std::vector<const DenseTensor*> ins = {&x};
+  std::vector<DenseTensor*> outs = {out};
+  phi::funcs::ElementwiseKernel<T>(ctx, ins, &outs, ErfinvFunctor<T>());
+}
+
+}  // namespace phi
 
 PD_REGISTER_KERNEL(erfinv, GPU, ALL_LAYOUT, phi::ErfinvKernel, float, double) {}