change CMakelist for FPGA track

CMakeLists.txt

@@ -2,9 +2,9 @@ cmake_minimum_required(VERSION 3.0)
 project(paddle-mobile)
 
 # select the platform to build
-option(CPU "armv7 with neon support" ON)
+option(CPU "armv7 with neon support" OFF)
 option(MALI_GPU "mali gpu support" OFF)
-option(FPGA "fpga support" OFF)
+option(FPGA "fpga support" ON)
 
 option(USE_OPENMP "openmp support" OFF)
 option(DEBUGING "enable debug mode" ON)
@@ -29,7 +29,10 @@ if(DEBUGING)
     message(STATUS "debugging mode")
     add_definitions(-DPADDLE_MOBILE_DEBUG)
 else()
+    if(FPGA)
+    else()
         add_definitions(-fvisibility=hidden -fvisibility-inlines-hidden)
+    endif()
 endif()
 
 if(USE_EXCEPTION)
@@ -93,8 +96,7 @@ else()
 endif()
 
 if(FPGA)
-    set(DEBUGING ON)
-    add_definitions(-DPADDLE_MOBILE_DEBUG)
+    message("FPGA mode enabled")
     add_definitions(-DPADDLE_MOBILE_FPGA)
 else()
     file(GLOB_RECURSE _tmp_list src/operators/kernel/fpga/*.cpp src/operators/kernel/fpga/*.cc)
@@ -177,6 +179,10 @@ if(DEBUGING)
     else()
         add_subdirectory(test)
     endif()
+elseif(FPGA)
+    add_subdirectory(test)
 endif()
 
 
+
+

src/common/variant.h

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include <cstdlib>
-
+#include <cstring>
 #include "common/enforce.h"
 #include "common/log.h"
 

src/fpga/api.cpp

@@ -22,7 +22,7 @@ limitations under the License. */
 #include "fpga/filter.h"
 #include "fpga/image.h"
 #define FPGA_TEST_MODE
-//#define PADDLE_MOBILE_OS_LINUX
+// #define PADDLE_MOBILE_OS_LINUX
 
 namespace paddle_mobile {
 namespace fpga {
@@ -149,7 +149,7 @@ int ComputeBasicConv(const struct ConvArgs &args) {
   return do_ioctl(IOCTL_CONFIG_CONV, &args);
 }
 
-int ComputeFpgaConv(const struct WrapperConvArgs &args) {
+int ComputeFpgaConv(const struct SplitConvArgs &args) {
 #ifdef FPGA_TEST_MODE
   DLOG << "=============ComputeFPGAConv===========";
   DLOG << "   filter_num:" << args.filter_num
@@ -194,8 +194,8 @@ int ComputeFpgaEWAdd(const struct EWAddArgs &args) {
 #ifdef FPGA_TEST_MODE
   DLOG << "=============ComputeFpgaEWAdd===========";
   DLOG << "   relu_enabled:" << args.relu_enabled
-       << "   const0:" << fp16_2_fp32(short(args.const0))
-       << "   const1:" << fp16_2_fp32(short(args.const1));
+       << "   const0:" << fp16_2_fp32(int16_t(args.const0))
+       << "   const1:" << fp16_2_fp32(int16_t(args.const1));
   DLOG << "   image0_address:" << args.image0.address
        << "   image0_scale_address:" << args.image0.scale_address
        << "   image0_channels:" << args.image0.channels
@@ -383,10 +383,10 @@ void format_concat_output(framework::Tensor *out, int height, int width,
   out->reset_data_ptr(data_ptr);
 }
 
-void fill_conv_arg(struct WrapperConvArgs *arg, framework::Tensor *input,
+void fill_split_arg(struct SplitConvArgs *arg, framework::Tensor *input,
                     framework::Tensor *out, framework::Tensor *filter,
-                   bool relu_enabled, int group_num, int stride_h, int stride_w,
-                   int padding_h, int padding_w, float *bs_ptr) {
+                    bool relu_enabled, int group_num, int stride_h,
+                    int stride_w, int padding_h, int padding_w, float *bs_ptr) {
   auto input_ptr = input->data<float>();
   auto filter_ptr = filter->data<float>();
   auto out_ptr = out->data<float>();

src/fpga/api.h

@@ -89,7 +89,7 @@ struct ConcatArgs {
   uint32_t width;
 };
 
-struct WrapperConvArgs {
+struct SplitConvArgs {
   uint32_t split_num;
   uint32_t group_num;
   uint32_t filter_num;
@@ -98,6 +98,14 @@ struct WrapperConvArgs {
   struct ConcatArgs concat_arg;
 };
 
+struct GroupConvArgs {
+  uint32_t group_num;
+  uint32_t filter_num;
+  struct ImageOutputArgs output;
+  struct SplitConvArgs* conv_args;
+  struct ConcatArgs concat_arg;
+};
+
 struct PoolingArgs {
   int16_t mode;  // mode: 0:max, 1:avg
   half kernel_reciprocal;
@@ -159,30 +167,6 @@ struct MemoryCacheArgs {
 #define IOCTL_FPGA_REG_READ _IOW(IOCTL_FPGA_MAGIC, 28, struct FpgaRegReadArgs)
 #define IOCTL_FPGA_REG_WRITE _IOW(IOCTL_FPGA_MAGIC, 29, struct FpgaRegWriteArgs)
 
-enum FPGA_ERR_TYPE {
-  ERR_IOCTL_CMD = -1,
-  ERR_TIMEOUT = -2,
-  ERR_COMPLETION_TIMEOUT = -3,
-  ERR_INVALID_FPGA_ADDR = -4,
-  ERR_NOMEM = -5,
-  ERR_NO_RESERVE_MEM = -6,
-  ERR_COPY_FROM_USER = -7,
-  ERR_COPY_TO_USER = -8,
-  ERR_DEL_TIMER = -9,
-  ERR_ENABLE_MSI = -10,
-  ERR_REGISTER_IRQ = -11,
-  ERR_PCIE_REGISTER = -12,
-  ERR_PCIE_PROBE = -13,
-  ERR_REGISTER_BLOCK = -14,
-  ERR_ALLOC_GENDISK = -15,
-  ERR_INIT_QUEUE = -16,
-  ERR_WAIT = -17,
-  ERR_ECC_ERROR = -31,
-  ERR_FPGA_FAIL_STOP = -64,
-  ERR_FPGA_DEBUG_STOP = -113,
-  DEV_TMP_UNAVAILABLE = -128
-};
-
 //============================== API =============================
 
 int open_device();
@@ -195,7 +179,7 @@ int fpga_flush(void* address, size_t size);
 int fpga_invalidate(void* address, size_t size);
 
 int PerformBypass(const struct BypassArgs& args);
-int ComputeFpgaConv(const struct WrapperConvArgs& args);
+int ComputeFpgaConv(const struct SplitConvArgs& args);
 int ComputeFpgaPool(const struct PoolingArgs& args);
 int ComputeFpgaEWAdd(const struct EWAddArgs& args);
 int ComputeFPGAConcat(const struct ConcatArgs& args);
@@ -220,10 +204,10 @@ void format_bias_scale_array(float** bias_scale_array,
 void format_concat_output(framework::Tensor* out, int height, int width,
                           int image_num, uint32_t* channel_num);
 
-void fill_conv_arg(struct WrapperConvArgs* arg, framework::Tensor* input,
+void fill_split_arg(struct SplitConvArgs* arg, framework::Tensor* input,
                     framework::Tensor* out, framework::Tensor* filter,
-                   bool relu_enabled, int group_num, int stride_h, int stride_w,
-                   int padding_h, int padding_w, float* bs_ptr);
+                    bool relu_enabled, int group_num, int stride_h,
+                    int stride_w, int padding_h, int padding_w, float* bs_ptr);
 
 half fp32_2_fp16(float fp32_num);
 float fp16_2_fp32(half fp16_num);

src/fpga/filter.cpp

@@ -21,7 +21,10 @@ namespace paddle_mobile {
 namespace fpga {
 namespace filter {
 
-int calc_division_capacity(int chw) { return 2048 / ((chw + 15) / 16) * 32; }
+int calc_division_capacity(int chw) {
+  int n = 2048 / ((chw + 15) / 16) * 32;
+  return n < 2048 ? n : 2048;
+}
 
 int calc_split_num(int num, int division_capacity) {
   return (num + division_capacity - 1) / division_capacity;

src/operators/kernel/fpga/conv_add_bn_kernel.cpp

@@ -66,10 +66,11 @@ bool ConvAddBNKernel<FPGA, float>::Init(FusionConvAddBNParam<FPGA> *param) {
   fpga::format_bias_scale_array(&bs_ptr, element_num_per_div, channel);
   fpga::format_fp16_ofm(out);
 
-  fpga::WrapperConvArgs conv_arg = {0};
-  fpga::fill_conv_arg(&conv_arg, input, out, filter, relu_enabled,
-                      param->Groups(), param->Strides()[0], param->Strides()[1],
-                      param->Paddings()[0], param->Paddings()[1], bs_ptr);
+  fpga::SplitConvArgs conv_arg = {0};
+  fpga::fill_split_arg(&conv_arg, input, out, filter, relu_enabled,
+                       param->Groups(), param->Strides()[0],
+                       param->Strides()[1], param->Paddings()[0],
+                       param->Paddings()[1], bs_ptr);
   param->SetFpgaArgs(conv_arg);
 
   return true;

src/operators/kernel/fpga/conv_add_bn_relu_kernel.cpp

@@ -65,10 +65,11 @@ bool ConvAddBNReluKernel<FPGA, float>::Init(
 
   fpga::format_fp16_ofm(out);
 
-  fpga::WrapperConvArgs conv_arg = {0};
-  fpga::fill_conv_arg(&conv_arg, input, out, filter, relu_enabled,
-                      param->Groups(), param->Strides()[0], param->Strides()[1],
-                      param->Paddings()[0], param->Paddings()[1], bs_ptr);
+  fpga::SplitConvArgs conv_arg = {0};
+  fpga::fill_split_arg(&conv_arg, input, out, filter, relu_enabled,
+                       param->Groups(), param->Strides()[0],
+                       param->Strides()[1], param->Paddings()[0],
+                       param->Paddings()[1], bs_ptr);
   param->SetFpgaArgs(conv_arg);
   return true;
 }

src/operators/kernel/fpga/conv_add_relu_kernel.cpp

@@ -47,10 +47,11 @@ bool ConvAddReluKernel<FPGA, float>::Init(FusionConvAddReluParam<FPGA> *param) {
 
   fpga::format_fp16_ofm(out);
 
-  fpga::WrapperConvArgs conv_arg = {0};
-  fpga::fill_conv_arg(&conv_arg, input, out, filter, relu_enabled,
-                      param->Groups(), param->Strides()[0], param->Strides()[1],
-                      param->Paddings()[0], param->Paddings()[1], bs_ptr);
+  fpga::SplitConvArgs conv_arg = {0};
+  fpga::fill_split_arg(&conv_arg, input, out, filter, relu_enabled,
+                       param->Groups(), param->Strides()[0],
+                       param->Strides()[1], param->Paddings()[0],
+                       param->Paddings()[1], bs_ptr);
   param->SetFpgaArgs(conv_arg);
   return true;
 }

src/operators/kernel/fpga/conv_bn_kernel.cpp

@@ -59,10 +59,11 @@ bool ConvBNKernel<FPGA, float>::Init(FusionConvBNParam<FPGA> *param) {
 
   fpga::format_fp16_ofm(out);
 
-  fpga::WrapperConvArgs conv_arg = {0};
-  fpga::fill_conv_arg(&conv_arg, input, out, filter, relu_enabled,
-                      param->Groups(), param->Strides()[0], param->Strides()[1],
-                      param->Paddings()[0], param->Paddings()[1], bs_ptr);
+  fpga::SplitConvArgs conv_arg = {0};
+  fpga::fill_split_arg(&conv_arg, input, out, filter, relu_enabled,
+                       param->Groups(), param->Strides()[0],
+                       param->Strides()[1], param->Paddings()[0],
+                       param->Paddings()[1], bs_ptr);
   param->SetFpgaArgs(conv_arg);
   return true;
 }

src/operators/kernel/fpga/conv_bn_relu_kernel.cpp

@@ -59,10 +59,11 @@ bool ConvBNReluKernel<FPGA, float>::Init(FusionConvBNReluParam<FPGA> *param) {
 
   fpga::format_fp16_ofm(out);
 
-  fpga::WrapperConvArgs conv_arg = {0};
-  fpga::fill_conv_arg(&conv_arg, input, out, filter, relu_enabled,
-                      param->Groups(), param->Strides()[0], param->Strides()[1],
-                      param->Paddings()[0], param->Paddings()[1], bs_ptr);
+  fpga::SplitConvArgs conv_arg = {0};
+  fpga::fill_split_arg(&conv_arg, input, out, filter, relu_enabled,
+                       param->Groups(), param->Strides()[0],
+                       param->Strides()[1], param->Paddings()[0],
+                       param->Paddings()[1], bs_ptr);
   param->SetFpgaArgs(conv_arg);
   return true;
 }

src/operators/kernel/fpga/fc_relu_kernel.cpp

@@ -53,9 +53,9 @@ bool FusionFcReluKernel<FPGA, float>::Init(FusionFcReluParam<FPGA> *param) {
   fpga::format_bias_scale_array(&bs_ptr, element_num_per_div, channel);
   fpga::format_fp16_ofm(out);
 
-  fpga::WrapperConvArgs conv_arg = {0};
-  fpga::fill_conv_arg(&conv_arg, input_x, out, filter, relu_enabled, 1, 1, 1, 0,
-                      0, bs_ptr);
+  fpga::SplitConvArgs conv_arg = {0};
+  fpga::fill_split_arg(&conv_arg, input_x, out, filter, relu_enabled, 1, 1, 1,
+                       0, 0, bs_ptr);
   param->SetFpgaArgs(conv_arg);
   return true;
 }

src/operators/kernel/fpga/fusion_fc_kernel.cpp

@@ -54,9 +54,9 @@ bool FusionFcKernel<FPGA, float>::Init(FusionFcParam<FPGA> *param) {
   fpga::format_bias_scale_array(&bs_ptr, element_num_per_div, channel);
   fpga::format_fp16_ofm(out);
 
-  fpga::WrapperConvArgs conv_arg = {0};
-  fpga::fill_conv_arg(&conv_arg, input_x, out, filter, relu_enabled, 1, 1, 1, 0,
-                      0, bs_ptr);
+  fpga::SplitConvArgs conv_arg = {0};
+  fpga::fill_split_arg(&conv_arg, input_x, out, filter, relu_enabled, 1, 1, 1,
+                       0, 0, bs_ptr);
   param->SetFpgaArgs(conv_arg);
   return true;
 }

src/operators/kernel/fpga/mul_kernel.cpp

-/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
-
-    http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License. */
-
-#ifdef MUL_OP
-
-#include "operators/kernel/mul_kernel.h"
-
-namespace paddle_mobile {
-namespace operators {
-
-template <>
-bool MulKernel<FPGA, float>::Init(MulParam<FPGA> *param) {
-  bool relu_enabled = false;
-  auto input_x = const_cast<LoDTensor *>(param->InputX());
-  auto filter = const_cast<LoDTensor *>(param->InputY());
-  auto out = param->Out();
-
-  PADDLE_MOBILE_ENFORCE(input_x->dims()[1] == filter->dims()[0],
-                        "Image channel should be equal to weight number");
-  int channel = (uint32_t)out->dims()[1];
-  auto bs_ptr =
-      (float *)fpga::fpga_malloc(2 * channel * sizeof(float));  // NOLINT
-  for (int i = 0; i < channel; i++) {
-    bs_ptr[i + channel] = 1;
-    bs_ptr[i] = 0;
-  }
-  int num = (uint32_t)filter->dims()[1];
-  int chw = (uint32_t)filter->dims()[0];
-  PADDLE_MOBILE_ENFORCE(
-      chw == input_x->numel(),
-      "Filter element num should be equal to IFM element num");
-  int height = (uint32_t)input_x->dims()[2];
-  int width = (uint32_t)input_x->dims()[3];
-  int filter_channel = chw / height / width;
-
-  out->Resize(framework::make_ddim({1, channel, 1, 1}));
-  filter->Resize(framework::make_ddim({num, filter_channel, height, width}));
-  float max_value = fpga::filter_find_max(filter);
-  fpga::format_fc_filter(filter, max_value);
-
-  int element_num_per_div = fpga::get_filter_num_per_div(filter, 1);
-  fpga::format_bias_scale_array(&bs_ptr, element_num_per_div, channel);
-  fpga::format_fp16_ofm(out);
-
-  fpga::WrapperConvArgs conv_arg = {0};
-  fpga::fill_conv_arg(&conv_arg, input_x, out, filter, relu_enabled, 1, 1, 1, 0,
-                      0, bs_ptr);
-  param->SetFpgaArgs(conv_arg);
-  return true;
-}
-
-template <>
-void MulKernel<FPGA, float>::Compute(const MulParam<FPGA> &param) const {
-  fpga::ComputeFpgaConv(param.FpgaArgs());
-}
-
-}  // namespace operators
-}  // namespace paddle_mobile
-
-#endif

src/operators/math/gemm_int8.cpp

@@ -652,7 +652,7 @@ void Gemm::WriteBasic(int32_t mc, int32_t nc, int32_t *c, int32_t *C,
   int32_t *C0, *c0;
   c_ptr = c;
   C_ptr = C;
-  if (nc1 > 0) {
+  /*if (nc1 > 0) {
     asm volatile(
         "subs       %[mc], %[mc], #1        \n\t"
         "blt        end_mc_%=               \n\t"
@@ -684,7 +684,7 @@ void Gemm::WriteBasic(int32_t mc, int32_t nc, int32_t *c, int32_t *C,
         : [C_ptr] "r"(C_ptr), [c_ptr] "r"(c_ptr), [mc] "r"(m), [nc1] "r"(nc1),
           [step] "r"(step), [step1] "r"(step1)
         : "memory", "r5", "r6", "q0", "q1", "q2", "q3");
-  }
+  }*/
 
   if (_nc1 != 0) {
     for (int32_t i = 0; i < mc; i++) {

src/operators/op_param.h

@@ -546,11 +546,11 @@ class MulParam : OpParam {
 #ifdef PADDLE_MOBILE_FPGA
 
  private:
-  fpga::WrapperConvArgs fpga_conv_args;
+  fpga::SplitConvArgs fpga_conv_args;
 
  public:
-  const fpga::WrapperConvArgs &FpgaArgs() const { return fpga_conv_args; }
-  void SetFpgaArgs(const fpga::WrapperConvArgs &args) { fpga_conv_args = args; }
+  const fpga::SplitConvArgs &FpgaArgs() const { return fpga_conv_args; }
+  void SetFpgaArgs(const fpga::SplitConvArgs &args) { fpga_conv_args = args; }
 #endif
 };
 #endif
@@ -1401,11 +1401,11 @@ class FusionFcParam : public OpParam {
 #ifdef PADDLE_MOBILE_FPGA
 
  private:
-  fpga::WrapperConvArgs fpga_conv_args;
+  fpga::SplitConvArgs fpga_conv_args;
 
  public:
-  const fpga::WrapperConvArgs &FpgaArgs() const { return fpga_conv_args; }
-  void SetFpgaArgs(const fpga::WrapperConvArgs &args) { fpga_conv_args = args; }
+  const fpga::SplitConvArgs &FpgaArgs() const { return fpga_conv_args; }
+  void SetFpgaArgs(const fpga::SplitConvArgs &args) { fpga_conv_args = args; }
 #endif
 };
 
@@ -1441,11 +1441,11 @@ class FusionConvAddParam : public ConvParam<Dtype> {
 #ifdef PADDLE_MOBILE_FPGA
 
  private:
-  fpga::WrapperConvArgs fpga_conv_args;
+  fpga::SplitConvArgs fpga_conv_args;
 
  public:
-  const fpga::WrapperConvArgs &FpgaArgs() const { return fpga_conv_args; }
-  void SetFpgaArgs(const fpga::WrapperConvArgs &args) { fpga_conv_args = args; }
+  const fpga::SplitConvArgs &FpgaArgs() const { return fpga_conv_args; }
+  void SetFpgaArgs(const fpga::SplitConvArgs &args) { fpga_conv_args = args; }
 #endif
 };
 
@@ -1496,11 +1496,11 @@ class FusionConvAddPReluParam : public ConvParam<Dtype> {
 #ifdef PADDLE_MOBILE_FPGA
 
  private:
-  fpga::WrapperConvArgs fpga_conv_args;
+  fpga::SplitConvArgs fpga_conv_args;
 
  public:
-  const fpga::WrapperConvArgs &FpgaArgs() const { return fpga_conv_args; }
-  void SetFpgaArgs(const fpga::WrapperConvArgs &args) { fpga_conv_args = args; }
+  const fpga::SplitConvArgs &FpgaArgs() const { return fpga_conv_args; }
+  void SetFpgaArgs(const fpga::SplitConvArgs &args) { fpga_conv_args = args; }
 #endif
 };
 #endif
@@ -1554,11 +1554,11 @@ class FusionConvAddAddPReluParam : public ConvParam<Dtype> {
 #ifdef PADDLE_MOBILE_FPGA
 
  private:
-  fpga::WrapperConvArgs fpga_conv_args;
+  fpga::SplitConvArgs fpga_conv_args;
 
  public:
-  const fpga::WrapperConvArgs &FpgaArgs() const { return fpga_conv_args; }
-  void SetFpgaArgs(const fpga::WrapperConvArgs &args) { fpga_conv_args = args; }
+  const fpga::SplitConvArgs &FpgaArgs() const { return fpga_conv_args; }
+  void SetFpgaArgs(const fpga::SplitConvArgs &args) { fpga_conv_args = args; }
 #endif
 };
 #endif
@@ -1629,11 +1629,11 @@ class FusionConvAddBNReluParam : public ConvParam<Dtype> {
 #ifdef PADDLE_MOBILE_FPGA
 
  private:
-  fpga::WrapperConvArgs fpga_conv_args;
+  fpga::SplitConvArgs fpga_conv_args;
 
  public:
-  const fpga::WrapperConvArgs &FpgaArgs() const { return fpga_conv_args; }
-  void SetFpgaArgs(const fpga::WrapperConvArgs &args) { fpga_conv_args = args; }
+  const fpga::SplitConvArgs &FpgaArgs() const { return fpga_conv_args; }
+  void SetFpgaArgs(const fpga::SplitConvArgs &args) { fpga_conv_args = args; }
 #endif
 };
 #endif
@@ -1715,11 +1715,11 @@ class FusionConvBNAddReluParam : public ConvParam<Dtype> {
 #ifdef PADDLE_MOBILE_FPGA
 
  private:
-  fpga::WrapperConvArgs fpga_conv_args;
+  fpga::SplitConvArgs fpga_conv_args;
 
  public:
-  const fpga::WrapperConvArgs &FpgaArgs() const { return fpga_conv_args; }
-  void SetFpgaArgs(const fpga::WrapperConvArgs &args) { fpga_conv_args = args; }
+  const fpga::SplitConvArgs &FpgaArgs() const { return fpga_conv_args; }
+  void SetFpgaArgs(const fpga::SplitConvArgs &args) { fpga_conv_args = args; }
 #endif
 };
 #endif
@@ -1782,11 +1782,11 @@ class FusionConvBNParam : public ConvParam<Dtype> {
 #ifdef PADDLE_MOBILE_FPGA
 
  private:
-  fpga::WrapperConvArgs fpga_conv_args;
+  fpga::SplitConvArgs fpga_conv_args;
 
  public:
-  const fpga::WrapperConvArgs &FpgaArgs() const { return fpga_conv_args; }
-  void SetFpgaArgs(const fpga::WrapperConvArgs &args) { fpga_conv_args = args; }
+  const fpga::SplitConvArgs &FpgaArgs() const { return fpga_conv_args; }
+  void SetFpgaArgs(const fpga::SplitConvArgs &args) { fpga_conv_args = args; }
 #endif
 };
 #endif
@@ -1857,11 +1857,11 @@ class FusionConvAddBNParam : public ConvParam<Dtype> {
 #ifdef PADDLE_MOBILE_FPGA
 
  private:
-  fpga::WrapperConvArgs fpga_conv_args;
+  fpga::SplitConvArgs fpga_conv_args;
 
  public:
-  const fpga::WrapperConvArgs &FpgaArgs() const { return fpga_conv_args; }
-  void SetFpgaArgs(const fpga::WrapperConvArgs &args) { fpga_conv_args = args; }
+  const fpga::SplitConvArgs &FpgaArgs() const { return fpga_conv_args; }
+  void SetFpgaArgs(const fpga::SplitConvArgs &args) { fpga_conv_args = args; }
 #endif
 };
 #endif
@@ -1983,11 +1983,11 @@ class FusionConvBNReluParam : public ConvParam<Dtype> {
 #ifdef PADDLE_MOBILE_FPGA
 
  private:
-  fpga::WrapperConvArgs fpga_conv_args;
+  fpga::SplitConvArgs fpga_conv_args;
 
  public:
-  const fpga::WrapperConvArgs &FpgaArgs() const { return fpga_conv_args; }
-  void SetFpgaArgs(const fpga::WrapperConvArgs &args) { fpga_conv_args = args; }
+  const fpga::SplitConvArgs &FpgaArgs() const { return fpga_conv_args; }
+  void SetFpgaArgs(const fpga::SplitConvArgs &args) { fpga_conv_args = args; }
 #endif
 };
 #endif

src/operators/sum_op.cpp

@@ -65,7 +65,6 @@ REGISTER_OPERATOR_CPU(sum, ops::SumOp);
 REGISTER_OPERATOR_MALI_GPU(sum, ops::ConcatOp);
 #endif
 #ifdef PADDLE_MOBILE_FPGA
-REGISTER_OPERATOR_FPGA(sum, ops::ConcatOp);
 #endif
 
 #endif

test/CMakeLists.txt

@@ -61,38 +61,11 @@ endif ()
 
 list(FIND NET "FPGAnets" CON)
 if (CON GREATER -1)
-    ADD_EXECUTABLE(test-resnet net/test_resnet.cpp test_helper.h test_include.h executor_for_test.h)
-    target_link_libraries(test-resnet paddle-mobile)
-
     ADD_EXECUTABLE(test-resnet50 fpga/test_resnet50.cpp test_helper.h test_include.h executor_for_test.h)
     target_link_libraries(test-resnet50 paddle-mobile)
 
-    ADD_EXECUTABLE(test-fpga-EW fpga/test_fpga_EW.cpp test_helper.h test_include.h executor_for_test.h)
-    target_link_libraries(test-fpga-EW paddle-mobile)
-
-    ADD_EXECUTABLE(test-fpga-conv fpga/test_fpga_conv.cpp test_helper.h test_include.h executor_for_test.h)
-    target_link_libraries(test-fpga-conv paddle-mobile)
-
-    ADD_EXECUTABLE(test-fpga-pooling fpga/test_fpga_pooling.cpp test_helper.h test_include.h executor_for_test.h)
-    target_link_libraries(test-fpga-pooling paddle-mobile)
-
-    ADD_EXECUTABLE(test-fpga-bypass fpga/test_fpga_bypass.cpp test_helper.h test_include.h executor_for_test.h)
-    target_link_libraries(test-fpga-bypass paddle-mobile)
-
-    ADD_EXECUTABLE(test-fpga-softmax fpga/test_fpga_softmax.cpp test_helper.h test_include.h executor_for_test.h)
-    target_link_libraries(test-fpga-softmax paddle-mobile)
-
-    ADD_EXECUTABLE(test-fpga-concat fpga/test_fpga_concat.cpp test_helper.h test_include.h executor_for_test.h)
-    target_link_libraries(test-fpga-concat paddle-mobile)
-
-    ADD_EXECUTABLE(test-tensor-quant fpga/test_tensor_quant.cpp test_helper.h test_include.h executor_for_test.h)
-    target_link_libraries(test-tensor-quant paddle-mobile)
-
-    ADD_EXECUTABLE(test-fpga-concat-op fpga/test_concat_op.cpp test_helper.h test_include.h)
-    target_link_libraries(test-fpga-concat-op paddle-mobile)
-
-    ADD_EXECUTABLE(test-format-data fpga/test_format_data.cpp test_helper.h test_include.h)
-    target_link_libraries(test-format-data paddle-mobile)
+#    ADD_EXECUTABLE(test-resnet net/test_resnet.cpp test_helper.h test_include.h executor_for_test.h)
+#    target_link_libraries(test-resnet paddle-mobile)
     set(FOUND_MATCH ON)
 
 endif ()

test/fpga/test_resnet50.cpp

@@ -30,7 +30,11 @@ int main() {
                              input_tensor.data<float>() + input_tensor.numel());
 
     paddle_mobile.FeedData(input_tensor);
+    for (int i = 0; i < 1000; i++) {
       paddle_mobile.Predict_To(-1);
+      if (i % 100 == 0) std::cout << i << std::endl;
+    }
+
     //    paddle_mobile.Predict_From(73);
     //    paddle_mobile.Predict_From_To(72, 73);
 

tools/op.cmake

@@ -118,12 +118,9 @@ if (CON GREATER -1)
   set(POOL_OP ON)
   set(CONCAT_OP ON)
   set(SOFTMAX_OP ON)
-  set(DROPOUT_OP ON)
   set(FUSION_CONVBNRELU_OP ON)
   set(FUSION_CONVBN_OP ON)
   set(FUSION_CONVADD_OP ON)
-  set(MUL_OP ON)
-
   set(FOUND_MATCH ON)
 endif()