From 3d1741b79403fe5424ebae6b5d55f33f8bae2362 Mon Sep 17 00:00:00 2001
From: liuyuhui <liuyuhui@baidu.com>
Date: Mon, 28 Dec 2020 13:20:02 +0800
Subject: [PATCH] [Kunlun] bug fix of PR2: Support MultiDevicePass and BKCL in
 parallel executor  (#29926)

---
 .../multi_devices_graph_pass/multi_devices_graph_pass.h   | 3 +++
 paddle/fluid/framework/parallel_executor.cc               | 8 +++++---
 paddle/fluid/platform/device_context.h                    | 2 +-
 3 files changed, 9 insertions(+), 4 deletions(-)

diff --git a/paddle/fluid/framework/ir/multi_devices_graph_pass/multi_devices_graph_pass.h b/paddle/fluid/framework/ir/multi_devices_graph_pass/multi_devices_graph_pass.h
index 42d22bfe6d4..97d3a40874b 100644
--- a/paddle/fluid/framework/ir/multi_devices_graph_pass/multi_devices_graph_pass.h
+++ b/paddle/fluid/framework/ir/multi_devices_graph_pass/multi_devices_graph_pass.h
@@ -39,10 +39,13 @@ class Graph;
 
 namespace paddle {
 namespace platform {
+#if defined(PADDLE_WITH_NCCL)
 class NCCLContextMap;
 class NCCLCommunicator;
+#elif defined(PADDLE_WITH_XPU_BKCL)
 class BKCLContextMap;
 class BKCLCommunicator;
+#endif
 }
 
 namespace framework {
diff --git a/paddle/fluid/framework/parallel_executor.cc b/paddle/fluid/framework/parallel_executor.cc
index 8f38a56e98f..947a3c9455f 100644
--- a/paddle/fluid/framework/parallel_executor.cc
+++ b/paddle/fluid/framework/parallel_executor.cc
@@ -968,9 +968,6 @@ void ParallelExecutor::BCastParamsToDevices(
       continue;
     }
     auto &dims = main_tensor.dims();
-
-    VLOG(1) << "bcast var=" << var;
-
     if (paddle::platform::is_gpu_place(main_tensor.place())) {
 #if defined(PADDLE_WITH_NCCL)
       std::vector<void *> buffers;
@@ -1013,6 +1010,11 @@ void ParallelExecutor::BCastParamsToDevices(
       std::vector<void *> buffers;
       buffers.reserve(member_->places_.size());
       size_t numel = main_tensor.numel();
+      // TODO(liuyuhui): BKCL only support parameters using float type,
+      // other parameters need to be strongly converted to float before
+      // broadcasting,
+      // but broadcast is equivalent to no type of operation, does not affect
+      // correctness.
       BKCLDataType data_type = BKCL_FLOAT;
       // BKCLDataType data_type = platform::ToBKCLDataType(main_tensor.type());
       for (size_t i = 0; i < member_->places_.size(); ++i) {
diff --git a/paddle/fluid/platform/device_context.h b/paddle/fluid/platform/device_context.h
index 8e5363fafa3..4e79e645aaa 100644
--- a/paddle/fluid/platform/device_context.h
+++ b/paddle/fluid/platform/device_context.h
@@ -123,7 +123,7 @@ class XPUDeviceContext : public DeviceContext {
   void Wait() const override;
 
 #ifdef PADDLE_WITH_XPU_BKCL
-  /*! \brief  Return nccl context. */
+  /*! \brief  Return bkcl context. */
   BKCLContext_t bkcl_context() const { return bkcl_context_; }
 
   /*! \brief  Set bkcl context. */
-- 
GitLab