PaddlePaddle
diff --git a/‎paddle/fluid/distributed/collective/process_group_bkcl.cc
Copy file name to clipboard
+2-3Lines changed: 2 additions & 3 deletions b/‎paddle/fluid/distributed/collective/process_group_bkcl.cc
Copy file name to clipboard
+2-3Lines changed: 2 additions & 3 deletions
diff --git a/‎paddle/fluid/framework/compiled_program.cc
Copy file name to clipboard
+2-2Lines changed: 2 additions & 2 deletions b/‎paddle/fluid/framework/compiled_program.cc
Copy file name to clipboard
+2-2Lines changed: 2 additions & 2 deletions
diff --git a/‎paddle/fluid/framework/data_type_transform.cc
Copy file name to clipboard
+2-2Lines changed: 2 additions & 2 deletions b/‎paddle/fluid/framework/data_type_transform.cc
Copy file name to clipboard
+2-2Lines changed: 2 additions & 2 deletions
diff --git a/‎paddle/fluid/framework/fleet/heter_ps/heter_comm_inl.h
Copy file name to clipboard
+8-8Lines changed: 8 additions & 8 deletions b/‎paddle/fluid/framework/fleet/heter_ps/heter_comm_inl.h
Copy file name to clipboard
+8-8Lines changed: 8 additions & 8 deletions
diff --git a/‎paddle/fluid/framework/fleet/ps_gpu_wrapper.kps
Copy file name to clipboard
+1-3Lines changed: 1 addition & 3 deletions b/‎paddle/fluid/framework/fleet/ps_gpu_wrapper.kps
Copy file name to clipboard
+1-3Lines changed: 1 addition & 3 deletions
diff --git a/‎paddle/fluid/framework/garbage_collector.cc
Copy file name to clipboard
+2-3Lines changed: 2 additions & 3 deletions b/‎paddle/fluid/framework/garbage_collector.cc
Copy file name to clipboard
+2-3Lines changed: 2 additions & 3 deletions
diff --git a/‎paddle/fluid/framework/heterxpu_trainer.cc
Copy file name to clipboard
+4-4Lines changed: 4 additions & 4 deletions b/‎paddle/fluid/framework/heterxpu_trainer.cc
Copy file name to clipboard
+4-4Lines changed: 4 additions & 4 deletions
diff --git a/‎paddle/fluid/framework/phi_utils.h
Copy file name to clipboard
+1-1Lines changed: 1 addition & 1 deletion b/‎paddle/fluid/framework/phi_utils.h
Copy file name to clipboard
+1-1Lines changed: 1 addition & 1 deletion
diff --git a/‎paddle/fluid/framework/tensor_util.cc
Copy file name to clipboard
+5-9Lines changed: 5 additions & 9 deletions b/‎paddle/fluid/framework/tensor_util.cc
Copy file name to clipboard
+5-9Lines changed: 5 additions & 9 deletions
diff --git a/‎paddle/fluid/framework/tensor_util.h
Copy file name to clipboard
+13-16Lines changed: 13 additions & 16 deletions b/‎paddle/fluid/framework/tensor_util.h
Copy file name to clipboard
+13-16Lines changed: 13 additions & 16 deletions
@@ -28,7 +28,6 @@
 
 namespace paddle {
 namespace distributed {
-using XPUDeviceContext = paddle::platform::XPUDeviceContext;
 
 ProcessGroupBKCL::BKCLTask::BKCLTask(const Place& place,
                                      int rank,
@@ -203,8 +202,8 @@ void ProcessGroupBKCL::CreateBKCLEnvCache(const Place& place,
   calc_event_ = std::make_shared<XPUEventManager>();
   auto* calc_ctx = static_cast<phi::XPUContext*>(
       phi::DeviceContextPool::Instance().Get(place));
-  // must use XPUDeviceContext here to make sure XPUContext::Init() is called
-  auto comm_ctx = std::make_unique<XPUDeviceContext>(place, true);
+  // must use phi::XPUContext here to make sure XPUContext::Init() is called
+  auto comm_ctx = std::make_unique<phi::XPUContext>(place, true);
   // comm_ctx does not require a pre-allocated GM buffer
   comm_ctx->x_context()->set_option("XPUAPI_DEFAULT_SIZE", "1");
   auto bkcl_comm_ctx = this->GetCommContext();
 
@@ -905,8 +905,8 @@ void CompiledProgram::PrepareNCCLCommunicator(Scope *global_scope) {
         global_scope, member_->places_);
     auto &pool = phi::DeviceContextPool::Instance();
     for (size_t dev_id = 0; dev_id < member_->places_.size(); ++dev_id) {
-      auto *dev_ctx = static_cast<platform::XPUDeviceContext *>(
-          pool.Get(member_->places_[dev_id]));
+      auto *dev_ctx =
+          static_cast<phi::XPUContext *>(pool.Get(member_->places_[dev_id]));
       auto &bkcl_ctx = bkcl_ctxs->at(member_->places_[dev_id]);
       dev_ctx->SetBkclContext(bkcl_ctx.comm());
     }
 
@@ -37,7 +37,7 @@ struct CastDataTypeFunctor {
 template <typename InType, typename OutType>
 static void XPUCastData(const phi::DenseTensor& in,
                         phi::DenseTensor* out,
-                        const platform::XPUDeviceContext* dev_ctx) {
+                        const phi::XPUContext* dev_ctx) {
   using XPUInTDType = typename XPUTypeTrait<InType>::Type;
   using XPUOutTDType = typename XPUTypeTrait<OutType>::Type;
   int r = xpu::cast<XPUInTDType, XPUOutTDType>(
@@ -55,7 +55,7 @@ static void XPUTransDataType(
     phi::DenseTensor* out,
     const paddle::framework::proto::VarType::Type& dst_type,
     const phi::DeviceContext* ctx) {
-  auto* context = static_cast<const platform::XPUDeviceContext*>(ctx);
+  auto* context = static_cast<const phi::XPUContext*>(ctx);
 
 #define XPUCastCallback(cpp_type, proto_type)          \
   do {                                                 \
 
@@ -1507,8 +1507,8 @@ void HeterComm<KeyType, ValType, GradType, GPUAccessor>::pull_merge_sparse(
   cudaMemsetAsync(d_right_ptr, -1, total_device * sizeof(int), stream);
 
 #elif defined(PADDLE_WITH_XPU_KP)
-  // get XPUDeviceContext according to xpu place
-  paddle::platform::XPUDeviceContext xpu_dev_ctx(place);
+  // get XPUContext according to xpu place
+  phi::XPUContext xpu_dev_ctx(place);
   auto xpu_context = xpu_dev_ctx.x_context();
 
   int r = xpu::constant<int>(xpu_context, d_left_ptr, total_device, -1);
@@ -1687,8 +1687,8 @@ void HeterComm<KeyType, ValType, GradType, GPUAccessor>::pull_normal_sparse(
   cudaMemsetAsync(d_right_ptr, -1, total_device * sizeof(int), stream);
 
 #elif defined(PADDLE_WITH_XPU_KP)
-  // get XPUDeviceContext according to xpu place
-  paddle::platform::XPUDeviceContext xpu_dev_ctx(place);
+  // get XPUContext according to xpu place
+  phi::XPUContext xpu_dev_ctx(place);
   auto xpu_context = xpu_dev_ctx.x_context();
 
   int r = xpu::constant<int>(xpu_context, d_left_ptr, total_device, -1);
@@ -1890,8 +1890,8 @@ void HeterComm<KeyType, ValType, GradType, GPUAccessor>::push_normal_sparse(
   cudaMemsetAsync(d_right_ptr, -1, total_device * sizeof(int), stream);
 
 #elif defined(PADDLE_WITH_XPU_KP)
-  // get XPUDeviceContext according to xpu place
-  paddle::platform::XPUDeviceContext xpu_dev_ctx(place);
+  // get XPUContext according to xpu place
+  phi::XPUContext xpu_dev_ctx(place);
   auto xpu_context = xpu_dev_ctx.x_context();
 
   int r = xpu::constant<int>(xpu_context, d_left_ptr, total_device, -1);
@@ -2065,8 +2065,8 @@ void HeterComm<KeyType, ValType, GradType, GPUAccessor>::push_sparse(
   cudaMemsetAsync(d_right_ptr, -1, total_device * sizeof(int), stream);
 
 #elif defined(PADDLE_WITH_XPU_KP)
-  // get XPUDeviceContext according to xpu place
-  paddle::platform::XPUDeviceContext xpu_dev_ctx(place);
+  // get XPUContext according to xpu place
+  phi::XPUContext xpu_dev_ctx(place);
   auto xpu_context = xpu_dev_ctx.x_context();
 
   int r = xpu::constant<int>(xpu_context, d_left_ptr, total_device, -1);
 
@@ -209,9 +209,7 @@ void PSGPUWrapper::CopyKeys(const phi::Place& place,
                             int total_len) {
   XPUStream stream = nullptr;
   auto dev_ctx = phi::DeviceContextPool::Instance().Get(place);
-  stream = static_cast<platform::XPUDeviceContext*>(dev_ctx)
-               ->x_context()
-               ->xpu_stream;
+  stream = static_cast<phi::XPUContext*>(dev_ctx)->x_context()->xpu_stream;
   unsigned long long* o_keys =
       reinterpret_cast<unsigned long long*>(origin_keys);
   unsigned long long* t_keys = (unsigned long long*)total_keys;
 
@@ -133,13 +133,12 @@ CustomDefaultStreamGarbageCollector::CustomDefaultStreamGarbageCollector(
     : GarbageCollector(place, max_memory_size) {}
 
 void CustomDefaultStreamGarbageCollector::Wait() const {
-  static_cast<platform::CustomDeviceContext *>(this->dev_ctx_)
-      ->WaitStreamCallback();
+  static_cast<phi::CustomContext *>(this->dev_ctx_)->WaitStreamCallback();
 }
 
 void CustomDefaultStreamGarbageCollector::ClearCallback(
     const std::function<void()> &callback) {
-  static_cast<platform::CustomDeviceContext *>(this->dev_ctx_)
+  static_cast<phi::CustomContext *>(this->dev_ctx_)
       ->AddStreamCallback(callback);
 }
 
 
@@ -364,8 +364,8 @@ int HeterXpuTrainer::EndPass(const HeterRequest* request,
         phi::backends::xpu::XPUDeviceGuard guard(dev_id);
         phi::DeviceContextPool& pool = phi::DeviceContextPool::Instance();
         phi::DeviceContext* dev_ctx = pool.Get(place);
-        const platform::XPUDeviceContext* xpu_ctx =
-            reinterpret_cast<const platform::XPUDeviceContext*>(dev_ctx);
+        const phi::XPUContext* xpu_ctx =
+            reinterpret_cast<const phi::XPUContext*>(dev_ctx);
         xpu::memset(
             xpu_ctx->x_context(),
             thread_tensor->data(),
@@ -400,8 +400,8 @@ int HeterXpuTrainer::EndPass(const HeterRequest* request,
       phi::backends::xpu::XPUDeviceGuard guard(dev_id);
       phi::DeviceContextPool& pool = phi::DeviceContextPool::Instance();
       phi::DeviceContext* dev_ctx = pool.Get(place);
-      const platform::XPUDeviceContext* xpu_ctx =
-          reinterpret_cast<const platform::XPUDeviceContext*>(dev_ctx);
+      const phi::XPUContext* xpu_ctx =
+          reinterpret_cast<const phi::XPUContext*>(dev_ctx);
       xpu::memset(
           xpu_ctx->x_context(),
           root_tensor->data(),
 
@@ -81,7 +81,7 @@ struct ConvertToPhiContext<phi::GPUContext> {
 
 #ifdef PADDLE_WITH_XPU
 template <>
-struct ConvertToPhiContext<platform::XPUDeviceContext> {
+struct ConvertToPhiContext<phi::XPUContext> {
   using TYPE = phi::XPUContext;
 };
 #endif
 
@@ -85,13 +85,11 @@ void TensorCopyImpl(const TENSOR& src,
 #ifdef PADDLE_WITH_CUSTOM_DEVICE
   else if (phi::is_custom_place(src_place) &&  // NOLINT
            phi::is_cpu_place(dst_place)) {
-    auto stream =
-        reinterpret_cast<const platform::CustomDeviceContext&>(ctx).stream();
+    auto stream = reinterpret_cast<const phi::CustomContext&>(ctx).stream();
     memory::Copy(dst_place, dst_ptr, src_place, src_ptr, size, stream);
   } else if (phi::is_cpu_place(src_place) &&  // NOLINT
              phi::is_custom_place(dst_place)) {
-    auto stream =
-        reinterpret_cast<const platform::CustomDeviceContext&>(ctx).stream();
+    auto stream = reinterpret_cast<const phi::CustomContext&>(ctx).stream();
     memory::Copy(dst_place, dst_ptr, src_place, src_ptr, size, stream);
   } else if (phi::is_custom_place(src_place) &&  // NOLINT
              phi::is_custom_place(dst_place)) {
@@ -100,8 +98,7 @@ void TensorCopyImpl(const TENSOR& src,
               << dst_place;
       return;
     }
-    auto stream =
-        reinterpret_cast<const platform::CustomDeviceContext&>(ctx).stream();
+    auto stream = reinterpret_cast<const phi::CustomContext&>(ctx).stream();
     memory::Copy(dst_place, dst_ptr, src_place, src_ptr, size, stream);
   }
 #endif
@@ -514,8 +511,7 @@ void TensorToStream(std::ostream& os,
 #ifdef PADDLE_WITH_XPU
       constexpr size_t kBufSize = 1024 * 1024 * 64;  // 64MB
       std::unique_ptr<char[]> buf(new char[kBufSize]);
-      auto& xpu_dev_ctx =
-          static_cast<const platform::XPUDeviceContext&>(dev_ctx);
+      auto& xpu_dev_ctx = static_cast<const phi::XPUContext&>(dev_ctx);
       phi::CPUPlace cpu;
       uintptr_t data = reinterpret_cast<uintptr_t>(data_ptr);
       while (size != 0) {
@@ -539,7 +535,7 @@ void TensorToStream(std::ostream& os,
       constexpr size_t kBufSize = 1024 * 1024 * 64;     // 64MB
       std::unique_ptr<char[]> buf(new char[kBufSize]);  // NOLINT
       auto& custom_device_context =
-          static_cast<const platform::CustomDeviceContext&>(dev_ctx);
+          static_cast<const phi::CustomContext&>(dev_ctx);
       phi::CPUPlace cpu;
       uintptr_t data = reinterpret_cast<uintptr_t>(data_ptr);
       while (size != 0) {
 
@@ -142,13 +142,12 @@ void TensorFromArray(const T* src,
 #endif
 #ifdef PADDLE_WITH_CUSTOM_DEVICE
   else if (phi::is_custom_place(dst_place)) {  // NOLINT
-    memory::Copy(
-        dst_place,
-        dst_ptr,
-        src_place,
-        src_ptr,
-        size,
-        reinterpret_cast<const platform::CustomDeviceContext&>(ctx).stream());
+    memory::Copy(dst_place,
+                 dst_ptr,
+                 src_place,
+                 src_ptr,
+                 size,
+                 reinterpret_cast<const phi::CustomContext&>(ctx).stream());
   }
 #endif
 #ifdef PADDLE_WITH_XPU
@@ -188,13 +187,12 @@ void TensorFromVector(const std::vector<T>& src,
 #endif
 #ifdef PADDLE_WITH_CUSTOM_DEVICE
   else if (phi::is_custom_place(dst_place)) {  // NOLINT
-    memory::Copy(
-        dst_place,
-        dst_ptr,
-        src_place,
-        src_ptr,
-        size,
-        reinterpret_cast<const platform::CustomDeviceContext&>(ctx).stream());
+    memory::Copy(dst_place,
+                 dst_ptr,
+                 src_place,
+                 src_ptr,
+                 size,
+                 reinterpret_cast<const phi::CustomContext&>(ctx).stream());
   }
 #endif
 #ifdef PADDLE_WITH_XPU
@@ -244,8 +242,7 @@ inline void TensorFromVector(const std::vector<bool>& src,
 #endif
 #ifdef PADDLE_WITH_CUSTOM_DEVICE
   else if (phi::is_custom_place(dst_place)) {  // NOLINT
-    auto stream =
-        reinterpret_cast<const platform::CustomDeviceContext&>(ctx).stream();
+    auto stream = reinterpret_cast<const phi::CustomContext&>(ctx).stream();
     memory::Copy(dst_place, dst_ptr, src_place, src_ptr, size, stream);
   }
 #endif
Original file line number	Diff line number	Diff line change
`@@ -133,13 +133,12 @@ CustomDefaultStreamGarbageCollector::CustomDefaultStreamGarbageCollector(`
`133`	`133`	`: GarbageCollector(place, max_memory_size) {}`
`134`	`134`
`135`	`135`	`void CustomDefaultStreamGarbageCollector::Wait() const {`
`136`		`- static_cast<platform::CustomDeviceContext *>(this->dev_ctx_)`
`137`		`- ->WaitStreamCallback();`
	`136`	`+ static_cast<phi::CustomContext *>(this->dev_ctx_)->WaitStreamCallback();`
`138`	`137`	`}`
`139`	`138`
`140`	`139`	`void CustomDefaultStreamGarbageCollector::ClearCallback(`
`141`	`140`	`const std::function<void()> &callback) {`
`142`		`- static_cast<platform::CustomDeviceContext *>(this->dev_ctx_)`
	`141`	`+ static_cast<phi::CustomContext *>(this->dev_ctx_)`
`143`	`142`	`->AddStreamCallback(callback);`
`144`	`143`	`}`
`145`	`144`