Polish codes.

PaddlePaddle · Sep 7, 2023 · d1c06cd · d1c06cd
1 parent 71289f7
commit d1c06cd
Showing 1 changed file with 22 additions and 25 deletions.
diff --git a/paddle/phi/kernels/funcs/broadcast_function.h b/paddle/phi/kernels/funcs/broadcast_function.h
@@ -36,7 +36,7 @@ struct BroadcastTypeClassifier {
   int64_t numel{0};
   int vec_size{4};
   int broadcast_num{0};                   // Not used for XPU
-  bool all_elementwise{false};            // Not used for XPU
+  bool all_elementwise{true};             // Not used for XPU
   phi::Array<bool, Arity> use_broadcast;  // Not used for XPU
   phi::Array<kps::details::BroadcastConfig, Arity> configs;
   phi::Array<const _ptr_ char *__restrict__, Arity> ins_data;
@@ -46,21 +46,11 @@ struct BroadcastTypeClassifier {
   BroadcastTypeClassifier(const std::vector<const DenseTensor *> &ins,
                           std::vector<DenseTensor *> *outs,
                           int axis) {
-#ifdef PADDLE_WITH_XPU_KP
-    PADDLE_ENFORCE_EQ(
-        ins.size(),
-        2,
-        phi::errors::InvalidArgument(
-            "XPU only support inputs is 2, but received %d", ins.size()));
-#endif
-
     numel = (*outs)[0]->numel();
 
-#ifndef PADDLE_WITH_XPU_KP
-    broadcast_num = 0;
-    all_elementwise = true;
     for (size_t i = 0; i < ins.size(); ++i) {
       ins_data[i] = (const _ptr_ char *)(ins[i]->data());
+#ifndef PADDLE_WITH_XPU_KP
       bool is_same_dim = ins[i]->numel() == numel;
       if (is_same_dim) {
         use_broadcast[i] = false;
@@ -69,8 +59,8 @@ struct BroadcastTypeClassifier {
         broadcast_num++;
       }
       all_elementwise &= is_same_dim;
-    }
 #endif
+    }
 
     for (int i = 0; i < NumOuts; ++i) {
       outs_data[i] = (*outs)[i]->data<OutT>();
@@ -90,10 +80,6 @@ struct BroadcastTypeClassifier {
   void InitBroadcastConfigs(const std::vector<const DenseTensor *> &ins,
                             std::vector<DenseTensor *> *outs,
                             int axis) {
-    if (all_elementwise) {
-      return;
-    }
-
     const auto dims_simplifier =
         BroadcastDimsSimplifier(ins, (*outs)[0]->dims(), axis);
     if (VLOG_IS_ON(6)) {
@@ -111,14 +97,16 @@ struct BroadcastTypeClassifier {
                                                dims_simplifier.in_dims[0],
                                                dims_simplifier.rank);
 #else
-    for (int i = 0; i < Arity; ++i) {
-      // if data shape is[m, n], then you should set data_dim = {n, m}
-      // eg: out's shape [3, 45, 1]. then out_dims = {1, 45, 3}
-      // if (ins[i]->numel() != (*outs)[0]->numel()) {
-      if (ins[i]->numel()) {
-        configs[i] = kps::details::BroadcastConfig(dims_simplifier.out_dims,
-                                                   dims_simplifier.in_dims[i],
-                                                   dims_simplifier.rank);
+    if (!all_elementwise) {
+      for (int i = 0; i < Arity; ++i) {
+        // if data shape is[m, n], then you should set data_dim = {n, m}
+        // eg: out's shape [3, 45, 1]. then out_dims = {1, 45, 3}
+        // if (ins[i]->numel() != (*outs)[0]->numel()) {
+        if (ins[i]->numel()) {
+          configs[i] = kps::details::BroadcastConfig(dims_simplifier.out_dims,
+                                                     dims_simplifier.in_dims[i],
+                                                     dims_simplifier.rank);
+        }
       }
     }
 #endif
@@ -963,6 +951,15 @@ void BroadcastKernel(const KPDevice &ctx,
   // maximum rank of all inputs.
   using Traits = phi::funcs::FunctionTraits<Functor>;
   const int kArity = Traits::arity;
+
+#ifdef PADDLE_WITH_XPU_KP
+  PADDLE_ENFORCE_EQ(
+      ins.size(),
+      2,
+      phi::errors::InvalidArgument(
+          "XPU only support inputs is 2, but received %d", ins.size()));
+#endif
+
   PADDLE_ENFORCE_EQ(
       ins.size(),
       kArity,