From 4ee7560eceebfa60f7a91240cd39c98931025474 Mon Sep 17 00:00:00 2001
From: hebotao <hebotao5@huawei.com>
Date: Mon, 26 Apr 2021 16:13:45 +0800
Subject: [PATCH] =?UTF-8?q?=E5=91=8A=E8=AD=A6=E6=B8=85=E7=90=86?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 .../cpu/elu_grad_cpu_kernel.cc                | 38 ++++---------------
 .../kernel_compiler/cpu/elu_grad_cpu_kernel.h |  2 -
 .../kernel_compiler/cpu/select_cpu_kernel.cc  |  2 +-
 .../kernel_compiler/cpu/select_cpu_kernel.h   |  2 +-
 4 files changed, 9 insertions(+), 35 deletions(-)

diff --git a/mindspore/ccsrc/backend/kernel_compiler/cpu/elu_grad_cpu_kernel.cc b/mindspore/ccsrc/backend/kernel_compiler/cpu/elu_grad_cpu_kernel.cc
index 70476811d2..5bd1680548 100644
--- a/mindspore/ccsrc/backend/kernel_compiler/cpu/elu_grad_cpu_kernel.cc
+++ b/mindspore/ccsrc/backend/kernel_compiler/cpu/elu_grad_cpu_kernel.cc
@@ -21,14 +21,6 @@
 
 namespace mindspore {
 namespace kernel {
-template <typename T>
-void EluGradCPUKernel::EluGrad(const T *input0, const T *input1, T *out, size_t start, size_t end) {
-  const T alpha = static_cast<T>(1);
-  for (size_t i = start; i < end; i++) {
-    out[i] = (input1[i] < static_cast<T>(0)) ? input0[i] * (input1[i] + alpha) : input0[i];
-  }
-}
-
 void EluGradCPUKernel::InitKernel(const CNodePtr &kernel_node) {
   MS_EXCEPTION_IF_NULL(kernel_node);
   dtype_ = AnfAlgo::GetPrevNodeOutputInferDataType(kernel_node, 0);
@@ -57,29 +49,13 @@ void EluGradCPUKernel::LaunchKernel(const std::vector<AddressPtr> &inputs, const
   T *output = reinterpret_cast<T *>(outputs[0]->addr);
 
   size_t lens = outputs[0]->size > 0 ? static_cast<size_t>(outputs[0]->size / sizeof(T)) : 1;
-  auto max_thread_num = std::thread::hardware_concurrency();
-  size_t thread_num = lens < 128 * max_thread_num ? std::ceil(lens / 128.0) : max_thread_num;
-  MS_LOG(INFO) << "Lens=" << lens << "; use thread_num=" << thread_num << "; max_thread_num: " << max_thread_num;
-  std::vector<std::thread> threads;
-  if (thread_num < 1) {
-    MS_LOG(ERROR) << "Invalid value: thread_num " << thread_num;
-    return;
-  }
-  threads.reserve(thread_num);
-  size_t start = 0;
-  size_t once_compute_size = (lens + thread_num - 1) / thread_num;
-  if (once_compute_size < 1) {
-    MS_LOG(ERROR) << "Invalid value: once_compute_size " << once_compute_size;
-    return;
-  }
-  while (start < lens) {
-    size_t end = (start + once_compute_size) > lens ? lens : (start + once_compute_size);
-    threads.emplace_back(std::thread(&EluGradCPUKernel::EluGrad<T>, this, input0, input1, output, start, end));
-    start += once_compute_size;
-  }
-  for (size_t i = 0; i < threads.size(); ++i) {
-    threads[i].join();
-  }
+  auto task = [input0, input1, output](const size_t start, const size_t end) {
+    const T alpha = T(1);
+    for (size_t i = start; i < end; i++) {
+      output[i] = (input1[i] < static_cast<T>(0)) ? input0[i] * (input1[i] + alpha) : input0[i];
+    }
+  };
+  CPUKernelUtils::ParallelFor(task, lens);
 }
 }  // namespace kernel
 }  // namespace mindspore
diff --git a/mindspore/ccsrc/backend/kernel_compiler/cpu/elu_grad_cpu_kernel.h b/mindspore/ccsrc/backend/kernel_compiler/cpu/elu_grad_cpu_kernel.h
index 07629f33ea..b9f61bec00 100644
--- a/mindspore/ccsrc/backend/kernel_compiler/cpu/elu_grad_cpu_kernel.h
+++ b/mindspore/ccsrc/backend/kernel_compiler/cpu/elu_grad_cpu_kernel.h
@@ -35,8 +35,6 @@ class EluGradCPUKernel : public CPUKernel {
   void LaunchKernel(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &outputs);
 
  private:
-  template <typename T>
-  void EluGrad(const T *input1, const T *input2, T *out, size_t start, size_t end);
   TypeId dtype_{kTypeUnknown};
 };
 
diff --git a/mindspore/ccsrc/backend/kernel_compiler/cpu/select_cpu_kernel.cc b/mindspore/ccsrc/backend/kernel_compiler/cpu/select_cpu_kernel.cc
index 470fb42955..da7224b657 100644
--- a/mindspore/ccsrc/backend/kernel_compiler/cpu/select_cpu_kernel.cc
+++ b/mindspore/ccsrc/backend/kernel_compiler/cpu/select_cpu_kernel.cc
@@ -37,7 +37,7 @@ void SelectCPUKernel<T>::InitKernel(const CNodePtr &kernel_node) {
 }
 
 template <typename T>
-bool SelectCPUKernel<T>::Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &workspace,
+bool SelectCPUKernel<T>::Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &,
                                 const std::vector<AddressPtr> &outputs) {
   auto input_cond = reinterpret_cast<bool *>(inputs[0]->addr);
   auto input_x = reinterpret_cast<T *>(inputs[1]->addr);
diff --git a/mindspore/ccsrc/backend/kernel_compiler/cpu/select_cpu_kernel.h b/mindspore/ccsrc/backend/kernel_compiler/cpu/select_cpu_kernel.h
index 92eb09a2d8..b17c286111 100644
--- a/mindspore/ccsrc/backend/kernel_compiler/cpu/select_cpu_kernel.h
+++ b/mindspore/ccsrc/backend/kernel_compiler/cpu/select_cpu_kernel.h
@@ -28,7 +28,7 @@ class SelectCPUKernel : public CPUKernel {
   SelectCPUKernel() = default;
   ~SelectCPUKernel() override = default;
 
-  bool Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &workspace,
+  bool Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &,
               const std::vector<AddressPtr> &outputs) override;
 
   void InitKernel(const CNodePtr &kernel_node) override;