From 5c9aaf7d5c5bde858f567857e309e12ceeab0a1f Mon Sep 17 00:00:00 2001
From: JamesLim-sy <santamonic@sina.com>
Date: Thu, 7 Jul 2022 02:18:58 +0000
Subject: [PATCH 1/4] first commit

---
 paddle/phi/kernels/autotune/auto_tune_base.h | 9 ++-------
 1 file changed, 2 insertions(+), 7 deletions(-)

diff --git a/paddle/phi/kernels/autotune/auto_tune_base.h b/paddle/phi/kernels/autotune/auto_tune_base.h
index 91685c2ed547c..3fd3b090c77b5 100644
--- a/paddle/phi/kernels/autotune/auto_tune_base.h
+++ b/paddle/phi/kernels/autotune/auto_tune_base.h
@@ -153,20 +153,15 @@ template <typename T, typename KernelType>
 class TransposeAutoTuner : public AutoTuneBase<T, KernelType> {
  public:
   static AutoTuneBase<T, KernelType>* Instance(KernelType kernel) {
+    static std::once_flag transpose_init_flag_;
     static std::unique_ptr<AutoTuneBase<T, KernelType>> instance_;
-    std::call_once(init_flag_, [&] {
+    std::call_once(transpose_init_flag_, [&] {
       instance_.reset(new AutoTuneBase<T, KernelType>(kernel));
     });
     return instance_.get();
   }
-
- private:
-  static std::once_flag init_flag_;
 };
 
-template <typename T, typename KernelType>
-std::once_flag TransposeAutoTuner<T, KernelType>::init_flag_;
-
 template <typename T, typename RetureType, typename... Args>
 static AutoTuneBase<T, KernelCallback<T, RetureType, Args...>>*
 MakeTransposeTuner(RetureType (*func)(Args...)) {

From 6d2d61587ede9ea97d5eacd167968b7152b80503 Mon Sep 17 00:00:00 2001
From: JamesLim-sy <santamonic@sina.com>
Date: Tue, 19 Jul 2022 02:18:00 +0000
Subject: [PATCH 2/4] clarify the quotes

---
 paddle/phi/kernels/autotune/auto_tune_base.h | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/paddle/phi/kernels/autotune/auto_tune_base.h b/paddle/phi/kernels/autotune/auto_tune_base.h
index 3fd3b090c77b5..4831ea3e55485 100644
--- a/paddle/phi/kernels/autotune/auto_tune_base.h
+++ b/paddle/phi/kernels/autotune/auto_tune_base.h
@@ -120,8 +120,7 @@ class AutoTuneBase {
 
   template <typename Context, typename... Args>
   float RunAndMeasureKernel(const Context& ctx, const int idx, Args&&... args) {
-    // Regard 1st run as warmup. Judge the result by the time cost of rest run
-    // cycles.
+    // Regard 1st run as warmup, judge the result by the time cost of rest cycles.
     constexpr int repeats = 3;
     phi::GpuTimer timer;
     float time_cost = 0;

From 66db14cd69e65834e04b5a960bb99cfbd92d55a7 Mon Sep 17 00:00:00 2001
From: JamesLim-sy <santamonic@sina.com>
Date: Tue, 19 Jul 2022 02:52:21 +0000
Subject: [PATCH 3/4] change code style format

---
 paddle/phi/kernels/autotune/auto_tune_base.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/paddle/phi/kernels/autotune/auto_tune_base.h b/paddle/phi/kernels/autotune/auto_tune_base.h
index 4831ea3e55485..054006d84934f 100644
--- a/paddle/phi/kernels/autotune/auto_tune_base.h
+++ b/paddle/phi/kernels/autotune/auto_tune_base.h
@@ -120,7 +120,8 @@ class AutoTuneBase {
 
   template <typename Context, typename... Args>
   float RunAndMeasureKernel(const Context& ctx, const int idx, Args&&... args) {
-    // Regard 1st run as warmup, judge the result by the time cost of rest cycles.
+    // Regard 1st run as warmup and judge the compare result by the time cost
+    // of rest cycles.
     constexpr int repeats = 3;
     phi::GpuTimer timer;
     float time_cost = 0;

From e8419fe61f382e7f5b2d3762444a53155a9f9f60 Mon Sep 17 00:00:00 2001
From: JamesLim-sy <santamonic@sina.com>
Date: Thu, 22 Sep 2022 19:14:42 +0800
Subject: [PATCH 4/4] support bfloat16

---
 paddle/phi/kernels/kps/compare_kernel.cu | 18 ++++++++++++------
 1 file changed, 12 insertions(+), 6 deletions(-)

diff --git a/paddle/phi/kernels/kps/compare_kernel.cu b/paddle/phi/kernels/kps/compare_kernel.cu
index b981d802255a2..b882fcc2a6c96 100644
--- a/paddle/phi/kernels/kps/compare_kernel.cu
+++ b/paddle/phi/kernels/kps/compare_kernel.cu
@@ -114,7 +114,8 @@ PD_REGISTER_KERNEL(less_than,
                    int64_t,
                    float,
                    double,
-                   phi::dtype::float16) {}
+                   phi::dtype::float16,
+                   phi::dtype::bfloat16) {}
 PD_REGISTER_KERNEL(less_equal,
                    KPS,
                    ALL_LAYOUT,
@@ -125,7 +126,8 @@ PD_REGISTER_KERNEL(less_equal,
                    int64_t,
                    float,
                    double,
-                   phi::dtype::float16) {}
+                   phi::dtype::float16,
+                   phi::dtype::bfloat16) {}
 PD_REGISTER_KERNEL(greater_than,
                    KPS,
                    ALL_LAYOUT,
@@ -136,7 +138,8 @@ PD_REGISTER_KERNEL(greater_than,
                    int64_t,
                    float,
                    double,
-                   phi::dtype::float16) {}
+                   phi::dtype::float16,
+                   phi::dtype::bfloat16) {}
 PD_REGISTER_KERNEL(greater_equal,
                    KPS,
                    ALL_LAYOUT,
@@ -147,7 +150,8 @@ PD_REGISTER_KERNEL(greater_equal,
                    int64_t,
                    float,
                    double,
-                   phi::dtype::float16) {}
+                   phi::dtype::float16,
+                   phi::dtype::bfloat16) {}
 PD_REGISTER_KERNEL(equal,
                    KPS,
                    ALL_LAYOUT,
@@ -158,7 +162,8 @@ PD_REGISTER_KERNEL(equal,
                    int64_t,
                    float,
                    double,
-                   phi::dtype::float16) {}
+                   phi::dtype::float16,
+                   phi::dtype::bfloat16) {}
 PD_REGISTER_KERNEL(not_equal,
                    KPS,
                    ALL_LAYOUT,
@@ -169,7 +174,8 @@ PD_REGISTER_KERNEL(not_equal,
                    int64_t,
                    float,
                    double,
-                   phi::dtype::float16) {}
+                   phi::dtype::float16,
+                   phi::dtype::bfloat16) {}
 
 PD_REGISTER_KERNEL(equal_all,
                    KPS,