support_tensor_fusion_xpu (#71785)

AndSonder · web-flow · commit 1e1a138b11de · 2025-04-03T11:01:13.000+08:00
diff --git a/paddle/fluid/framework/new_executor/instruction/instruction_util.cc b/paddle/fluid/framework/new_executor/instruction/instruction_util.cc
@@ -107,8 +107,7 @@ phi::DeviceContext* ParseDeviceContext(pir::Operation* op,
 
   // only gpu need update. xpu not need, because xpu memcpy op kernel is
   // synchronous.
-  if (phi::is_gpu_place(place) || phi::is_custom_place(place) ||
-      phi::is_xpu_place(place)) {
+  if (phi::is_gpu_place(place) || phi::is_custom_place(place)) {
     VLOG(6) << "Parse DeviceContext for " << op_name
             << ", execution stream = " << execution_stream;
     if (execution_stream != kDefaultStream) {
diff --git a/paddle/fluid/framework/new_executor/interpreter/stream_analyzer.cc b/paddle/fluid/framework/new_executor/interpreter/stream_analyzer.cc
@@ -202,8 +202,7 @@ DeviceContext* StreamAnalyzer::ParseDeviceContext(
 
   // only gpu need update. xpu not need, because xpu memcpy op kernel is
   // synchronous.
-  if (phi::is_gpu_place(place_) || phi::is_custom_place(place_) ||
-      phi::is_xpu_place(place_)) {
+  if (phi::is_gpu_place(place_) || phi::is_custom_place(place_)) {
     VLOG(6) << "Parse DeviceContext for " << op_type
             << ", execution stream = " << execution_stream;
     if (execution_stream != kDefaultStream) {
diff --git a/paddle/phi/backends/xpu/xpu3_op_list.cc b/paddle/phi/backends/xpu/xpu3_op_list.cc
@@ -303,7 +303,9 @@ XPUOpMap& get_kl3_ops() {
                      phi::DataType::INT64,
                      phi::DataType::INT32})},
       {"coalesce_tensor",
-       XPUKernelSet({phi::DataType::FLOAT32, phi::DataType::FLOAT16})},
+       XPUKernelSet({phi::DataType::FLOAT32,
+                     phi::DataType::FLOAT16,
+                     phi::DataType::BFLOAT16})},
       {"concat_grad",
        XPUKernelSet({phi::DataType::FLOAT32,
                      phi::DataType::FLOAT16,
@@ -982,6 +984,10 @@ XPUOpMap& get_kl3_ops() {
       {"nearest_interp_v2_grad", XPUKernelSet({phi::DataType::FLOAT32})},
       {"nll_loss", XPUKernelSet({phi::DataType::FLOAT32})},
       {"nll_loss_grad", XPUKernelSet({phi::DataType::FLOAT32})},
+      {"nop",
+       XPUKernelSet({phi::DataType::FLOAT32,
+                     phi::DataType::FLOAT16,
+                     phi::DataType::BFLOAT16})},
       {"norm",
        XPUKernelSet({
            phi::DataType::FLOAT32,
@@ -1665,6 +1671,26 @@ XPUOpMap& get_kl3_ops() {
                      phi::DataType::FLOAT32})},
       {"variable_length_memory_efficient_attention",
        XPUKernelSet({phi::DataType::FLOAT32, phi::DataType::FLOAT16})},
+      {"view_shape",
+       XPUKernelSet({phi::DataType::BOOL,
+                     phi::DataType::INT32,
+                     phi::DataType::INT8,
+                     phi::DataType::UINT8,
+                     phi::DataType::INT64,
+                     phi::DataType::FLOAT32,
+                     phi::DataType::FLOAT64,
+                     phi::DataType::FLOAT16,
+                     phi::DataType::BFLOAT16})},
+      {"view_slice",
+       XPUKernelSet({phi::DataType::BOOL,
+                     phi::DataType::INT32,
+                     phi::DataType::INT8,
+                     phi::DataType::UINT8,
+                     phi::DataType::INT64,
+                     phi::DataType::FLOAT32,
+                     phi::DataType::FLOAT64,
+                     phi::DataType::FLOAT16,
+                     phi::DataType::BFLOAT16})},
       {"warpctc_grad", XPUKernelSet({phi::DataType::FLOAT32})},
       {"warpctc", XPUKernelSet({phi::DataType::FLOAT32})},
       {"where_index",
diff --git a/paddle/phi/kernels/coalesce_tensor_kernel.cc b/paddle/phi/kernels/coalesce_tensor_kernel.cc
@@ -316,6 +316,7 @@ PD_REGISTER_KERNEL(coalesce_tensor,
                    ALL_LAYOUT,
                    phi::CoalesceTensorKernel,
                    phi::dtype::float16,
+                   phi::dtype::bfloat16,
                    int,
                    float,
                    double) {
diff --git a/paddle/phi/kernels/xpu/nop_kernel.cc b/paddle/phi/kernels/xpu/nop_kernel.cc
@@ -0,0 +1,24 @@
+// Copyright (c) 2025 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
+//
+//     https://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/nop_kernel.h"
+#include "paddle/phi/core/kernel_registry.h"
+
+PD_REGISTER_KERNEL(nop,
+                   XPU,
+                   ALL_LAYOUT,
+                   phi::NopKernel,
+                   float,
+                   phi::dtype::bfloat16,
+                   phi::dtype::float16) {}
