From 0358f919ca8eb2145116667e922b8e00923b68a3 Mon Sep 17 00:00:00 2001
From: sneaxiy <sneaxiy@126.com>
Date: Tue, 13 Sep 2022 07:55:03 +0000
Subject: [PATCH] remove useless codes

---
 paddle/fluid/platform/flags.cc                |  2 -
 paddle/phi/kernels/funcs/broadcast_function.h | 48 +------------------
 2 files changed, 1 insertion(+), 49 deletions(-)

diff --git a/paddle/fluid/platform/flags.cc b/paddle/fluid/platform/flags.cc
index 2f2eae6add537..28dddc1fbebdd 100644
--- a/paddle/fluid/platform/flags.cc
+++ b/paddle/fluid/platform/flags.cc
@@ -1023,5 +1023,3 @@ PADDLE_DEFINE_EXPORTED_bool(
 PADDLE_DEFINE_EXPORTED_string(jit_engine_type,
                               "PE",
                               "Choose default funciton type in JitLayer.");
-
-PADDLE_DEFINE_EXPORTED_bool(use_int32_kernel, false, "");
diff --git a/paddle/phi/kernels/funcs/broadcast_function.h b/paddle/phi/kernels/funcs/broadcast_function.h
index eb4a332887e9c..46ec31227346b 100644
--- a/paddle/phi/kernels/funcs/broadcast_function.h
+++ b/paddle/phi/kernels/funcs/broadcast_function.h
@@ -14,7 +14,6 @@ limitations under the License. */
 
 #pragma once
 
-#include "gflags/gflags.h"
 #include "paddle/phi/kernels/funcs/elementwise_base.h"
 
 #if defined(__NVCC__) || defined(__HIPCC__) || defined(__xpu__)
@@ -23,8 +22,6 @@ namespace kps = phi::kps;
 
 #endif
 
-DECLARE_bool(use_int32_kernel);
-
 namespace phi {
 namespace funcs {
 
@@ -472,14 +469,6 @@ void LaunchBroadcastKernel(
 }
 
 #ifndef PADDLE_WITH_XPU_KP
-#define CUDA_ASSERT(__cond) \
-  do {                      \
-    if (!(__cond)) {        \
-      printf(#__cond);      \
-      asm("trap;");         \
-    }                       \
-  } while (0)
-
 HOSTDEVICE static int64_t ConvertSrcIdxToDstIdx(
     int64_t src_idx,
     const phi::Array<int64_t, phi::DDim::kMaxRank + 1> &src_strides,
@@ -495,8 +484,6 @@ HOSTDEVICE static int64_t ConvertSrcIdxToDstIdx(
       dst_idx += local_idx * dst_strides[k + 1];
     }
   }
-  CUDA_ASSERT(src_idx >= 0 && src_idx < src_strides[0]);
-  CUDA_ASSERT(dst_idx >= 0 && dst_idx < dst_strides[0]);
   return dst_idx;
 }
 
@@ -518,15 +505,6 @@ HOSTDEVICE static void ReadVecDataWithInt64Index(
   } else {
     if (!need_broadcast) {
 #pragma unroll
-      for (int i = 0; i < VecSize; ++i) {
-        CUDA_ASSERT(idx + i >= 0 && idx + i < src_strides[0]);
-      }
-      for (int i = 0; i < rank; ++i) {
-        CUDA_ASSERT(src_strides[i] == dst_strides[i]);
-      }
-      CUDA_ASSERT(src_strides[rank] == 1);
-      CUDA_ASSERT(dst_strides[rank] == 1);
-
       phi::Load<T, VecSize>(in + idx, out);
     } else {
 #pragma unroll
@@ -772,28 +750,6 @@ struct LaunchBroadcastKernelWithInt64IndexHelper<InT,
     auto gpu_config =
         phi::backends::gpu::GetGpuLaunchConfig1D(ctx, numel, VecSize);
 
-    if (VLOG_IS_ON(10)) {
-      VLOG(10) << "-------------------------------------------------";
-      VLOG(10) << "Arity = " << Arity << " , VecSize = " << VecSize
-               << " , rank = " << rank << " , Functor = "
-               << phi::enforce::demangle(typeid(Functor).name())
-               << " , InT = " << phi::enforce::demangle(typeid(InT).name())
-               << " , OutT = " << phi::enforce::demangle(typeid(OutT).name());
-      for (int i = 0; i < Arity; ++i) {
-        VLOG(10) << "in " << i << " : dims = " << ins[i]->dims()
-                 << " broadcast_dims = "
-                 << phi::DDim(ins_expand_dims[i].Get(), rank)
-                 << " strides = " << phi::DDim(ins_strides[i].Get(), rank + 1)
-                 << " need_broadcast = " << need_broadcasts[i];
-      }
-      VLOG(10) << "out : dims = " << (*outs)[0]->dims()
-               << " strides = " << phi::DDim(out_strides.Get(), rank + 1);
-      VLOG(10) << "gpu_config.block_per_grid = " << gpu_config.block_per_grid.x
-               << " , gpu_config.thread_per_block = "
-               << gpu_config.thread_per_block.x;
-      VLOG(10) << "-------------------------------------------------";
-    }
-
     BroadcastKernelWithInt64Index<InT, OutT, Functor, VecSize, Arity>
         <<<gpu_config.block_per_grid,
            gpu_config.thread_per_block,
@@ -949,9 +905,7 @@ void BroadcastKernelForDifferentVecSize(
   bool use_int64_index_kernel =
       kEnabledInt64IndexKernel &&
       (*outs)[0]->numel() >= std::numeric_limits<int32_t>::max();
-  if (!FLAGS_use_int32_kernel) {
-    use_int64_index_kernel = kEnabledInt64IndexKernel;
-  }
+  use_int64_index_kernel = kEnabledInt64IndexKernel;
   if (use_int64_index_kernel) {
     int vec_size = GetVecsize<InT, OutT>(ins, outs);
     switch (vec_size) {