update sigmod_cross_entry_with_logits_*.cu for compiling optimize (#57631)

yinwei · web-flow · commit 1c5abfa9a004 · 2023-09-25T14:32:54.000+08:00
diff --git a/paddle/phi/kernels/gpu/sigmoid_cross_entropy_with_logits.h b/paddle/phi/kernels/gpu/sigmoid_cross_entropy_with_logits.h
@@ -51,15 +51,4 @@ struct NonzeroFunctor {
   }
 };
 
-template <typename T>
-struct DivFunctor {
-  const T norm_;
-  HOSTDEVICE inline DivFunctor(const T norm) : norm_(norm) {}
-
-  HOSTDEVICE inline T operator()(T loss) {
-    loss /= norm_;
-    return loss;
-  }
-};
-
 }  // namespace phi
diff --git a/paddle/phi/kernels/gpu/sigmoid_cross_entropy_with_logits_grad_kernel.cu b/paddle/phi/kernels/gpu/sigmoid_cross_entropy_with_logits_grad_kernel.cu
@@ -16,6 +16,7 @@
 
 #include "paddle/phi/common/memory_utils.h"
 #include "paddle/phi/kernels/gpu/sigmoid_cross_entropy_with_logits.h"
+#include "paddle/phi/kernels/scale_kernel.h"
 
 namespace phi {
 
@@ -144,10 +145,8 @@ void SigmoidCrossEntropyWithLogitsGradKernel(
     auto eps = static_cast<T>(1e-5);
     *norm_cpu_ptr = *norm_cpu_ptr > eps ? *norm_cpu_ptr : eps;
 
-    std::vector<const DenseTensor *> div_ins = {in_grad};
-    std::vector<DenseTensor *> div_outs = {in_grad};
-    auto div_functor = DivFunctor<T>(*norm_cpu_ptr);
-    phi::funcs::ElementwiseKernel<T>(dev_ctx, div_ins, &div_outs, div_functor);
+    phi::ScaleKernel<T>(
+        dev_ctx, *in_grad, (1.0 / *norm_cpu_ptr), 0.0f, false, in_grad);
 
     delete norm_tensor;
   }
diff --git a/paddle/phi/kernels/gpu/sigmoid_cross_entropy_with_logits_kernel.cu b/paddle/phi/kernels/gpu/sigmoid_cross_entropy_with_logits_kernel.cu
@@ -16,6 +16,7 @@
 
 #include "paddle/phi/common/memory_utils.h"
 #include "paddle/phi/kernels/gpu/sigmoid_cross_entropy_with_logits.h"
+#include "paddle/phi/kernels/scale_kernel.h"
 
 namespace phi {
 
@@ -146,10 +147,7 @@ void SigmoidCrossEntropyWithLogitsKernel(
     auto eps = static_cast<T>(1e-5);
     *norm_cpu_ptr = *norm_cpu_ptr > eps ? *norm_cpu_ptr : eps;
 
-    std::vector<const DenseTensor *> div_ins = {out};
-    std::vector<DenseTensor *> div_outs = {out};
-    auto div_functor = DivFunctor<T>(*norm_cpu_ptr);
-    phi::funcs::ElementwiseKernel<T>(dev_ctx, div_ins, &div_outs, div_functor);
+    phi::ScaleKernel<T>(dev_ctx, *out, 1.0 / (*norm_cpu_ptr), 0.0f, false, out);
 
     delete norm_tensor;
   }
