!38651 [MS][LITE] add op coders for micro train

Merge pull request !38651 from jianghui58/micro_train_dev

cmake/package_micro.cmake

@@ -10,6 +10,10 @@ function(__install_micro_wrapper)
             COMPONENT ${RUNTIME_COMPONENT_NAME} FILES_MATCHING PATTERN "*.h")
     install(DIRECTORY ${NNACL_DIR}/fp32 DESTINATION ${CODEGEN_ROOT_DIR}/include/nnacl
             COMPONENT ${RUNTIME_COMPONENT_NAME} FILES_MATCHING PATTERN "*.h")
+    install(FILES ${NNACL_DIR}/fp32_grad/activation_grad.h DESTINATION
+            ${CODEGEN_ROOT_DIR}/include/nnacl/fp32_grad COMPONENT ${RUNTIME_COMPONENT_NAME})
+    install(FILES ${NNACL_DIR}/fp32_grad/softmax_cross_entropy_with_logits.h DESTINATION
+            ${CODEGEN_ROOT_DIR}/include/nnacl/fp32_grad COMPONENT ${RUNTIME_COMPONENT_NAME})
     install(DIRECTORY ${NNACL_DIR}/kernel DESTINATION ${CODEGEN_ROOT_DIR}/include/nnacl
             COMPONENT ${RUNTIME_COMPONENT_NAME} FILES_MATCHING PATTERN "*.h")
     install(DIRECTORY ${NNACL_DIR}/infer DESTINATION ${CODEGEN_ROOT_DIR}/include/nnacl

mindspore/ccsrc/plugin/device/cpu/kernel/nnacl/fp32/adam_fp32.c

@@ -209,3 +209,31 @@ size_t FusedCastAdamFp16Fp32(int16_t *var16, const float *gradient32, float *m,
                   global_norm_reciprocal, end);
   return c1;
 }
+
+int DoAdam(float *m, float *v, const float *gradient, float *weight, float beta1, float beta2, float *beta1_power,
+           float *beta2_power, float eps, float learning_rate, bool nesterov, int start, int end) {
+  if ((1.f - beta1_power[0]) <= 0.0f) {
+    return NNACL_PARAM_INVALID;
+  }
+  if ((1.f - beta2_power[0]) < 0.0f) {
+    return NNACL_ERRCODE_SQRT_NEGATIVE;
+  }
+
+  float update_lr = learning_rate * sqrtf(1.f - beta2_power[0]) / (1.f - beta1_power[0]);
+  const float one_minus_beta1 = 1.f - beta1;
+  const float one_minus_beta2 = 1.f - beta2;
+  if (nesterov) {  // Nadam
+    for (int i = start; i < end; ++i) {
+      m[i] += (gradient[i] - m[i]) * one_minus_beta1;
+      v[i] += (gradient[i] * gradient[i] - v[i]) * one_minus_beta2;
+      weight[i] -= update_lr * (m[i] * beta1 + one_minus_beta1 * gradient[i]) / (sqrtf(v[i]) + eps);
+    }
+  } else {
+    for (int i = start; i < end; ++i) {
+      m[i] += (gradient[i] - m[i]) * one_minus_beta1;
+      v[i] += (gradient[i] * gradient[i] - v[i]) * one_minus_beta2;
+      weight[i] -= update_lr * m[i] / (sqrtf(v[i]) + eps);
+    }
+  }
+  return NNACL_OK;
+}

mindspore/ccsrc/plugin/device/cpu/kernel/nnacl/fp32/adam_fp32.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2021 Huawei Technologies Co., Ltd
+ * Copyright 2021-2022 Huawei Technologies Co., Ltd
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -40,6 +40,9 @@ size_t FusedCastAdamFp16Fp16(int16_t *var16, const int16_t *gradient16, float *m
 size_t FusedCastAdamFp16Fp32(int16_t *var16, const float *gradient32, float *m, float *v, float lr, float beta1,
                              float beta2, float epsilon, float decay, float global_norm_reciprocal, size_t start,
                              size_t end);
+
+int DoAdam(float *m, float *v, const float *gradient, float *weight, float beta1, float beta2, float *beta1_power,
+           float *beta2_power, float eps, float learning_rate, bool nesterov, int start, int end);
 #ifdef __cplusplus
 }
 #endif

mindspore/ccsrc/plugin/device/cpu/kernel/nnacl/fp32_grad/softmax_cross_entropy_with_logits.c

@@ -0,0 +1,43 @@
+/**
+ * Copyright 2022 Huawei Technologies Co., Ltd
+ *
+ * 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.
+ */
+
+#include "nnacl/fp32_grad/softmax_cross_entropy_with_logits.h"
+#include <math.h>
+
+void ForwardPostExecute(const float *labels, const float *logits, float *grads, float *output2,
+                        size_t number_of_classes, int batch_size) {
+  float eps = 1e-6;
+  if (grads != NULL) {
+    for (size_t i = 0; i < (size_t)(batch_size); ++i) {
+      float loss = 0.f;
+      for (size_t j = 0; j < number_of_classes; ++j) {
+        float logit = -logf(logits[i * number_of_classes + j] <= 0.0 ? eps : logits[i * number_of_classes + j]);
+        grads[i * number_of_classes + j] = (logits[i * number_of_classes + j] - labels[i * number_of_classes + j]);
+        loss += labels[i * number_of_classes + j] * logit;
+      }
+      output2[i] = loss;
+    }
+  } else {
+    for (size_t i = 0; i < (size_t)(batch_size); ++i) {
+      float loss = 0.f;
+      for (size_t j = 0; j < number_of_classes; ++j) {
+        float logit = -logf(logits[i * number_of_classes + j] <= 0.0 ? eps : logits[i * number_of_classes + j]);
+        loss += labels[i * number_of_classes + j] * logit;
+      }
+      output2[i] = loss;
+    }
+  }
+}

mindspore/ccsrc/plugin/device/cpu/kernel/nnacl/fp32_grad/softmax_cross_entropy_with_logits.h

@@ -0,0 +1,33 @@
+/**
+ * Copyright 2022 Huawei Technologies Co., Ltd
+ *
+ * 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.
+ */
+
+#ifndef MINDSPORE_CCSRC_PLUGIN_DEVICE_CPU_KERNEL_NNACL_FP32_GRAD_SOFTMAX_CROSS_ENTROPY_WITH_LOGITS_H_
+#define MINDSPORE_CCSRC_PLUGIN_DEVICE_CPU_KERNEL_NNACL_FP32_GRAD_SOFTMAX_CROSS_ENTROPY_WITH_LOGITS_H_
+
+#include <stddef.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void ForwardPostExecute(const float *labels, const float *logits, float *grads, float *output2,
+                        size_t number_of_classes, int batch_size);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // MINDSPORE_CCSRC_PLUGIN_DEVICE_CPU_KERNEL_NNACL_FP32_GRAD_SOFTMAX_CROSS_ENTROPY_WITH_LOGITS_H_

mindspore/lite/examples/train_lenet_cpp/src/net_runner.cc

@@ -1,5 +1,5 @@
 /**
- * Copyright 2020 Huawei Technologies Co., Ltd
+ * Copyright 2020-2022 Huawei Technologies Co., Ltd
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -15,11 +15,11 @@
  */
 
 #include "src/net_runner.h"
-#include <math.h>
 #include <getopt.h>
 #include <stdio.h>
 #include <malloc.h>
 #include <cstring>
+#include <cmath>
 #include <chrono>
 #include <iostream>
 #include <fstream>
@@ -63,7 +63,7 @@ constexpr static int kElem2Print = 10;
 class Rescaler : public mindspore::TrainCallBack {
  public:
   explicit Rescaler(float scale) : scale_(scale) {
-    if (scale_ == 0) {
+    if (std::fabs(scale) <= std::numeric_limits<float>::epsilon()) {
       scale_ = 1.0;
     }
   }

mindspore/lite/src/runtime/kernel/cpu/fp32_grad/adam.cc

@@ -20,6 +20,7 @@
 #include "schema/model_generated.h"
 #include "src/runtime/kernel_registry.h"
 #include "include/errorcode.h"
+#include "nnacl/fp32/adam_fp32.h"
 #include "plugin/device/cpu/kernel/nnacl/op_base.h"
 
 using mindspore::kernel::KERNEL_ARCH;
@@ -41,43 +42,13 @@ constexpr static int kGradientIdx = 9;
 
 int AdamCPUKernel::ReSize() { return RET_OK; }
 
-static int DoAdam(float *m, float *v, const float *gradient, float *weight, float beta1, float beta2, float beta1_power,
-                  float beta2_power, float eps, float learning_rate, bool nesterov, int start, int end) {
-  if ((1.f - beta1_power) <= 0.0f) {
-    MS_LOG(ERROR) << "divisor cannot be 0 or below";
-    return RET_ERROR;
-  }
-  if ((1.f - beta2_power) < 0.0f) {
-    MS_LOG(ERROR) << "sqrt cannot be negative";
-    return RET_ERROR;
-  }
-
-  auto update_lr = learning_rate * std::sqrt(1.f - beta2_power) / (1.f - beta1_power);
-  const float one_minus_beta1 = 1.f - beta1;
-  const float one_minus_beta2 = 1.f - beta2;
-  if (nesterov) {  // Nadam
-    for (int i = start; i < end; ++i) {
-      m[i] += (gradient[i] - m[i]) * one_minus_beta1;
-      v[i] += (gradient[i] * gradient[i] - v[i]) * one_minus_beta2;
-      weight[i] -= update_lr * (m[i] * beta1 + one_minus_beta1 * gradient[i]) / (std::sqrt(v[i]) + eps);
-    }
-  } else {
-    for (int i = start; i < end; ++i) {
-      m[i] += (gradient[i] - m[i]) * one_minus_beta1;
-      v[i] += (gradient[i] * gradient[i] - v[i]) * one_minus_beta2;
-      weight[i] -= update_lr * m[i] / (std::sqrt(v[i]) + eps);
-    }
-  }
-  return RET_OK;
-}
-
 int AdamCPUKernel::DoExecute(int task_id) {
   CHECK_LESS_RETURN(in_tensors_.size(), DIMENSION_10D);
   auto weight = reinterpret_cast<float *>(in_tensors_.at(kWeightIdx)->MutableData());
   auto m = reinterpret_cast<float *>(in_tensors_.at(kMomentVector1stIdx)->MutableData());
   auto v = reinterpret_cast<float *>(in_tensors_.at(kMomentVector2stIdx)->MutableData());
-  auto beta1_power = reinterpret_cast<float *>(in_tensors_.at(kBeta1PowerIdx)->MutableData())[0];
+  auto beta1_power = reinterpret_cast<float *>(in_tensors_.at(kBeta1PowerIdx)->MutableData());
-  auto beta2_power = reinterpret_cast<float *>(in_tensors_.at(kBeta2PowerIdx)->MutableData())[0];
+  auto beta2_power = reinterpret_cast<float *>(in_tensors_.at(kBeta2PowerIdx)->MutableData());
   auto learning_rate = lr_;
   auto beta1 = reinterpret_cast<float *>(in_tensors_.at(kBeta1Idx)->MutableData())[0];
   auto beta2 = reinterpret_cast<float *>(in_tensors_.at(kBeta2Idx)->MutableData())[0];
@@ -146,8 +117,8 @@ int AdamCPUKernel::OptimizerStep() {
   auto weight = reinterpret_cast<float *>(in_tensors_.at(kWeightIdx)->MutableData());
   auto m = reinterpret_cast<float *>(in_tensors_.at(kMomentVector1stIdx)->MutableData());
   auto v = reinterpret_cast<float *>(in_tensors_.at(kMomentVector2stIdx)->MutableData());
-  auto beta1_power = reinterpret_cast<float *>(in_tensors_.at(kBeta1PowerIdx)->MutableData())[0];
+  auto beta1_power = reinterpret_cast<float *>(in_tensors_.at(kBeta1PowerIdx)->MutableData());
-  auto beta2_power = reinterpret_cast<float *>(in_tensors_.at(kBeta2PowerIdx)->MutableData())[0];
+  auto beta2_power = reinterpret_cast<float *>(in_tensors_.at(kBeta2PowerIdx)->MutableData());
   auto learning_rate = lr_;
   auto beta1 = reinterpret_cast<float *>(in_tensors_.at(kBeta1Idx)->MutableData())[0];
   auto beta2 = reinterpret_cast<float *>(in_tensors_.at(kBeta2Idx)->MutableData())[0];

mindspore/lite/src/runtime/kernel/cpu/fp32_grad/softmax_cross_entropy_with_logits.cc

@@ -18,6 +18,7 @@
 #include "src/runtime/kernel_registry.h"
 #include "nnacl/softmax_parameter.h"
 #include "nnacl/fp32/softmax_fp32.h"
+#include "nnacl/fp32_grad/softmax_cross_entropy_with_logits.h"
 #include "src/runtime/kernel/cpu/fp32_grad/softmax_cross_entropy_with_logits.h"
 #include "include/errorcode.h"
 
@@ -29,34 +30,6 @@ using mindspore::schema::PrimitiveType_SoftmaxCrossEntropyWithLogits;
 namespace mindspore::kernel {
 int SoftmaxCrossEntropyWithLogitsCPUKernel::Prepare() { return ReSize(); }
 
-void SoftmaxCrossEntropyWithLogitsCPUKernel::ForwardPostExecute(const float *labels, const float *logits, float *grads,
-                                                                float *output2) const {
-  float eps = 1e-6;
-  if (grads != nullptr) {
-    for (size_t i = 0; i < static_cast<size_t>(param_->batch_size_); ++i) {
-      float loss = 0.f;
-      for (size_t j = 0; j < param_->number_of_classes_; ++j) {
-        float logit =
-          -logf(logits[i * param_->number_of_classes_ + j] <= 0.0 ? eps : logits[i * param_->number_of_classes_ + j]);
-        grads[i * param_->number_of_classes_ + j] =
-          (logits[i * param_->number_of_classes_ + j] - labels[i * param_->number_of_classes_ + j]);
-        loss += labels[i * param_->number_of_classes_ + j] * logit;
-      }
-      output2[i] = loss;
-    }
-  } else {
-    for (size_t i = 0; i < static_cast<size_t>(param_->batch_size_); ++i) {
-      float loss = 0.f;
-      for (size_t j = 0; j < param_->number_of_classes_; ++j) {
-        float logit =
-          -logf(logits[i * param_->number_of_classes_ + j] <= 0.0 ? eps : logits[i * param_->number_of_classes_ + j]);
-        loss += labels[i * param_->number_of_classes_ + j] * logit;
-      }
-      output2[i] = loss;
-    }
-  }
-}
-
 int SoftmaxCrossEntropyWithLogitsCPUKernel::DoExecute(int task_id) {
   auto ins = reinterpret_cast<float *>(in_tensors_.at(0)->data());
   CHECK_NULL_RETURN(ins);
@@ -75,7 +48,7 @@ int SoftmaxCrossEntropyWithLogitsCPUKernel::DoExecute(int task_id) {
   std::fill(losses_, losses_ + data_size, 0);
   std::fill(sum_data_, sum_data_ + sm_params_.input_shape_[0], 0);
   Softmax(ins, losses_, sum_data_, &sm_params_);
-  ForwardPostExecute(labels, losses_, grads, out);
+  ForwardPostExecute(labels, losses_, grads, out, param_->number_of_classes_, param_->batch_size_);
   return RET_OK;
 }
 

mindspore/lite/src/runtime/kernel/cpu/fp32_grad/softmax_cross_entropy_with_logits.h

@@ -35,8 +35,6 @@ class SoftmaxCrossEntropyWithLogitsCPUKernel : public LossKernel {
   }
   ~SoftmaxCrossEntropyWithLogitsCPUKernel() override = default;
 
-  void ForwardPostExecute(const float *labels, const float *logits, float *output1, float *output2) const;
-
   int Prepare() override;
   int ReSize() override;
   int Run() override;
@@ -46,7 +44,6 @@ class SoftmaxCrossEntropyWithLogitsCPUKernel : public LossKernel {
   SoftmaxCrossEntropyParameter *param_;
   SoftmaxParameter sm_params_;
 };
-
 }  // namespace mindspore::kernel
 
 #endif  // MINDSPORE_LITE_SRC_RUNTIME_KERNEL_CPU_FP32_GRAD_SOFTMAX_CROSS_ENTROPY_WITH_LOGITS_H_

mindspore/lite/tools/converter/micro/cmake/file_list.cmake

@@ -100,6 +100,12 @@ set(CODER_OPCODERS_SRC
         ${MICRO_DIR}/coder/opcoders/nnacl/fp32/exp_fp32_coder.cc
         ${MICRO_DIR}/coder/opcoders/nnacl/fp32/deconv2d_fp32_coder.cc
         ${MICRO_DIR}/coder/opcoders/nnacl/fp32/prelu_fp32_coder.cc
+        #### nnacl fp32_grad coder
+        ${MICRO_DIR}/coder/opcoders/nnacl/fp32_grad/activation_grad_coder.cc
+        ${MICRO_DIR}/coder/opcoders/nnacl/fp32_grad/adam_coder.cc
+        ${MICRO_DIR}/coder/opcoders/nnacl/fp32_grad/assign_coder.cc
+        ${MICRO_DIR}/coder/opcoders/nnacl/fp32_grad/biasadd_grad_coder.cc
+        ${MICRO_DIR}/coder/opcoders/nnacl/fp32_grad/softmax_cross_entropy_with_logits_coder.cc
         #### nnacl int8 coder
         ${MICRO_DIR}/coder/opcoders/nnacl/int8/activation_int8_coder.cc
         ${MICRO_DIR}/coder/opcoders/nnacl/int8/affine_int8_coder.cc

mindspore/lite/tools/converter/micro/coder/coder.cc

@@ -120,18 +120,18 @@ int Coder::MicroSourceCodeGeneration(const schema::MetaGraphT &graph, const std:
   return RET_OK;
 }
 
-int Coder::Init(const std::string code_mode, const std::string target, bool support_parallel, bool debug_mode) const {
+int Coder::Init(const std::string &code_mode, const std::string &target, bool support_parallel, bool debug_mode) const {
   static const std::map<std::string, Target> kTargetMap = {
     {"x86", kX86}, {"Cortex-M", kCortex_M}, {"ARM32", kARM32}, {"ARM64", kARM64}, {"All", kAllTargets}};
   static const std::map<std::string, CodeMode> kCodeModeMap = {{"Inference", Inference}, {"Train", Train}};
   Configurator *config = Configurator::GetInstance();
 
   auto target_item = kTargetMap.find(target);
-  MS_CHECK_TRUE_RET_BOOL(target_item != kTargetMap.end(), "unsupported target: " + target);
+  MS_CHECK_TRUE_MSG(target_item != kTargetMap.end(), RET_ERROR, "unsupported target: " + target);
   config->set_target(target_item->second);
 
   auto code_item = kCodeModeMap.find(code_mode);
-  MS_CHECK_TRUE_RET_BOOL(code_item != kCodeModeMap.end(), "unsupported code mode: " + code_mode);
+  MS_CHECK_TRUE_MSG(code_item != kCodeModeMap.end(), RET_ERROR, "unsupported code mode: " + code_mode);
   config->set_code_mode(code_item->second);
 
   if (support_parallel && config->target() == kCortex_M) {

mindspore/lite/tools/converter/micro/coder/coder.h

@@ -35,7 +35,7 @@ class Coder final {
                                        bool support_parallel, bool debug_mode);
 
  private:
-  int Init(const std::string code_mode, const std::string target, bool support_parallel, bool debug_mode_) const;
+  int Init(const std::string &code_mode, const std::string &target, bool support_parallel, bool debug_mode_) const;
   int Run(const void *model_buff, size_t size);
   bool InitPath(const std::string &output_path);
   std::shared_ptr<CoderSession> session_{nullptr};

mindspore/lite/tools/converter/micro/coder/generator/component/common_component.cc

@@ -229,7 +229,7 @@ void CodeInputImplement(std::ofstream &ofs, const std::unique_ptr<CoderContext>
   // input tensors
   std::vector<Tensor *> inputs = ctx->graph_inputs();
   for (size_t i = 0; i < inputs.size(); ++i) {
-    ofs << "static const unsigned char *" << ctx->input_name() + std::to_string(i) << " = 0;\n";
+    ofs << "const unsigned char *" << ctx->input_name() + std::to_string(i) << " = 0;\n";
   }
   size_t size = inputs.size();
   ofs << "int "

mindspore/lite/tools/converter/micro/coder/generator/component/const_blocks/debug_utils.cc

@@ -1,5 +1,5 @@
 /**
- * Copyright 2021 Huawei Technologies Co., Ltd
+ * Copyright 2021-2022 Huawei Technologies Co., Ltd
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -41,6 +41,7 @@ const char debug_utils_h[] = R"RAW(
 #include <sys/time.h>
 #include <time.h>
 #include <stdint.h>
+#include "nnacl/op_base.h"
 
 #define MICRO_INFO(content, args...) \
   { printf("[INFO] %s|%d: " #content "\r\n", __func__, __LINE__, ##args); }
@@ -62,24 +63,9 @@ enum DataType {
   DataType_MAX = DataType_DT_UNDEFINED
 };
 
-enum Format {
-  Format_NCHW = 0,
-  Format_NHWC = 1,
-  Format_HWKC = 2,
-  Format_HWCK = 3,
-  Format_KCHW = 4,
-  Format_CKHW = 5,
-  Format_KHWC = 6,
-  Format_CHWK = 7,
-  Format_NC4HW4 = 100,
-  Format_NUM_OF_FORMAT = 101,
-  Format_MIN = Format_NCHW,
-  Format_MAX = Format_NUM_OF_FORMAT
-};
-
 typedef struct {
   enum DataType type;
-  enum Format format;
+  enum FormatC format;
   int ndim;
   int *dim;
   void *data;
@@ -140,7 +126,7 @@ static const char *const TypeNames[] = {"DT_FLOAT", "DT_FLOAT16", "DT_INT8",   "
                                         "",         "",           "DT_UINT32", "DT_INT64", "DT_UINT16",    "",
                                         "",         "",           "",          "",         "DT_UNDEFINED", ""};
 
-const char *EnumNameFormat(enum Format e) {
+const char *EnumNameFormat(enum FormatC e) {
   switch (e) {
     case Format_NCHW:
       return "NCHW";
@@ -160,8 +146,6 @@ const char *EnumNameFormat(enum Format e) {
       return "CHWK";
     case Format_NC4HW4:
       return "NC4HW4";
-    case Format_NUM_OF_FORMAT:
-      return "NUM_OF_FORMAT";
     default:
       return "";
   }

mindspore/lite/tools/converter/micro/coder/generator/generator.cc

@@ -254,7 +254,10 @@ int Generator::CodeWeightFile() {
   cofs << g_hwLicense;
   cofs << "#include \"" << net_weight_hfile_ << "\"\n\n";
   cofs << "int  " << gThreadNum << " = 1; \n";
-
+  std::vector<Tensor *> inputs = ctx_->graph_inputs();
+  for (size_t i = 0; i < inputs.size(); ++i) {
+    cofs << "extern const unsigned char *" << ctx_->input_name() + std::to_string(i) << ";\n";
+  }
   if (config_->target() != kCortex_M) {
     cofs << "unsigned char * " << ctx_->buffer_name() << " = 0; \n";
     cofs << "unsigned char * " << ctx_->weight_name() << " = 0; \n";

mindspore/lite/tools/converter/micro/coder/opcoders/nnacl/fp32/activation_fp32_coder.cc

@@ -14,9 +14,7 @@
  * limitations under the License.
  */
 #include "coder/opcoders/nnacl/fp32/activation_fp32_coder.h"
-#include <string>
 #include "nnacl/fp32/activation_fp32.h"
-#include "nnacl/op_base.h"
 #include "coder/opcoders/serializers/nnacl_serializer/nnacl_fp32_serializer.h"
 #include "coder/opcoders/file_collector.h"
 #include "coder/opcoders/parallel.h"

mindspore/lite/tools/converter/micro/coder/opcoders/nnacl/fp32/matmul_fp32_base_coder.cc

@@ -204,10 +204,7 @@ int MatMulFP32BaseCoder::DoCode(CoderContext *const context) {
   // do const value packing to init
   if (!params_->a_const_) {
     code.CodeFunction("InitMatrixA", input_tensor_, a_pack_ptr_, "&mat_mul_parameter", vec_matmul_);
-    if (!params_->b_const_) {
+    if (params_->b_const_) {
-      init_code.CodeMallocExpression(b_pack_ptr_, b_pack_ptr_size_);
-      init_code.CodeFunction("memset", b_pack_ptr_, 0, b_pack_ptr_size_);
-    } else {
       init_code.CodeBufferOffsetExpression(b_pack_ptr_, context->weight_name(), context->weight_offset_name(),
                                            context->weight_size_name(), b_pack_ptr_size_);
       w_buf_size += b_pack_ptr_size_;
@@ -223,10 +220,7 @@ int MatMulFP32BaseCoder::DoCode(CoderContext *const context) {
     init_code.CodeFunction("InitMatrixB", b_src_str, b_pack_ptr_, "&mat_mul_parameter", vec_matmul_);
   }
   if (!params_->b_const_) {
-    if (!params_->a_const_) {
+    if (params_->a_const_) {
-      init_code.CodeMallocExpression(a_pack_str, a_pack_ptr_size_);
-      init_code.CodeFunction("memset", a_pack_ptr_, 0, a_pack_ptr_size_);
-    } else {
       init_code.CodeBufferOffsetExpression(a_pack_ptr_, context->weight_name(), context->weight_offset_name(),
                                            context->weight_size_name(), a_pack_ptr_size_);
       w_buf_size += a_pack_ptr_size_;

mindspore/lite/tools/converter/micro/coder/opcoders/nnacl/fp32_grad/activation_grad_coder.cc

@@ -0,0 +1,59 @@
+/**
+ * Copyright 2022 Huawei Technologies Co., Ltd
+ *
+ * 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.
+ */
+
+#include "coder/opcoders/nnacl/fp32_grad/activation_grad_coder.h"
+#include "nnacl/fp32_grad/activation_grad.h"
+#include "coder/opcoders/serializers/nnacl_serializer/nnacl_fp32_serializer.h"
+#include "coder/opcoders/file_collector.h"
+
+using mindspore::schema::PrimitiveType_ActivationGrad;
+
+namespace mindspore::lite::micro::nnacl {
+int ActivationGradCoder::DoCode(CoderContext *const context) {
+  MS_CHECK_TRUE(input_tensors_.size() == DIMENSION_2D, "inputs size is not equal to two");
+  Tensor *input0 = input_tensors_.at(0);
+  Tensor *input1 = input_tensors_.at(1);
+  // attribute
+  auto *activation_parameter = reinterpret_cast<ActivationParameter *>(parameter_);
+  int count = input_tensor_->ElementsNum();
+  Collect(context,
+          {
+            "nnacl/fp32_grad/activation_grad.h",
+          },
+          {
+            "activation_grad.c",
+          });
+  NNaclFp32Serializer code;
+
+  switch (activation_parameter->type_) {
+    case schema::ActivationType_RELU:
+      code.CodeFunction("ReluGrad", input0, input1, count, output_tensor_);
+      break;
+    case schema::ActivationType_ELU:
+      code.CodeFunction("EluGrad", input0, input1, count, output_tensor_, activation_parameter->alpha_);
+      break;
+    default:
+      MS_LOG(ERROR) << "Activation type error";
+      return RET_ERROR;
+  }
+  MS_LOG(DEBUG) << "ActivationGradCode has been called";
+  context->AppendCode(code.str());
+  return lite::RET_OK;
+}
+
+REG_OPERATOR_CODER(kAllTargets, kNumberTypeFloat32, PrimitiveType_ActivationGrad,
+                   CPUOpCoderCreator<ActivationGradCoder>)
+}  // namespace mindspore::lite::micro::nnacl

mindspore/lite/tools/converter/micro/coder/opcoders/nnacl/fp32_grad/activation_grad_coder.h

@@ -0,0 +1,38 @@
+/**
+ * Copyright 2022 Huawei Technologies Co., Ltd
+ *
+ * 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.
+ */
+
+#ifndef MINDSPORE_LITE_TOOLS_CONVERTER_MICRO_CODER_OPCODERS_NNACL_FP32_GRAD_ACTIVATION_GRAD_CODER_H_
+#define MINDSPORE_LITE_TOOLS_CONVERTER_MICRO_CODER_OPCODERS_NNACL_FP32_GRAD_ACTIVATION_GRAD_CODER_H_
+
+#include <vector>
+#include "coder/opcoders/op_coder.h"
+
+namespace mindspore::lite::micro::nnacl {
+class ActivationGradCoder final : public OperatorCoder {
+ public:
+  ActivationGradCoder(const std::vector<Tensor *> &in_tensors, const std::vector<Tensor *> &out_tensors,
+                      const LiteGraph::Node *node, size_t node_index, Target target)
+      : OperatorCoder(in_tensors, out_tensors, node, node_index, target) {}
+
+  ~ActivationGradCoder() override = default;
+
+  int Prepare(CoderContext *const context) override { return RET_OK; }
+
+  int DoCode(CoderContext *const context) override;
+};
+}  // namespace mindspore::lite::micro::nnacl
+
+#endif  // MINDSPORE_LITE_TOOLS_CONVERTER_MICRO_CODER_OPCODERS_NNACL_FP32_GRAD_ACTIVATION_GRAD_CODER_H_

mindspore/lite/tools/converter/micro/coder/opcoders/nnacl/fp32_grad/adam_coder.cc

@@ -0,0 +1,69 @@
+/**
+ * Copyright 2022 Huawei Technologies Co., Ltd
+ *
+ * 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.
+ */
+
+#include "coder/opcoders/nnacl/fp32_grad/adam_coder.h"
+#include "nnacl/fp32_grad/optimizer.h"
+#include "coder/opcoders/serializers/nnacl_serializer/nnacl_fp32_serializer.h"
+#include "coder/opcoders/file_collector.h"
+
+using mindspore::schema::PrimitiveType_Adam;
+
+namespace mindspore::lite::micro::nnacl {
+namespace {
+constexpr int kWeightIdx = 0;
+constexpr int kMomentVector1stIdx = 1;
+constexpr int kMomentVector2stIdx = 2;
+constexpr int kBeta1PowerIdx = 3;
+constexpr int kBeta2PowerIdx = 4;
+constexpr int kLearningRateIdx = 5;
+constexpr int kBeta1Idx = 6;
+constexpr int kBeta2Idx = 7;
+constexpr int kEpsilonIdx = 8;
+constexpr int kGradientIdx = 9;
+}  // namespace
+int AdamCoder::DoCode(CoderContext *const context) {
+  MS_CHECK_TRUE(input_tensors_.size() >= DIMENSION_10D, "inputs size is less than 10");
+  auto weight = input_tensors_.at(kWeightIdx);
+  auto m = input_tensors_.at(kMomentVector1stIdx);
+  auto v = input_tensors_.at(kMomentVector2stIdx);
+  auto beta1_power = input_tensors_.at(kBeta1PowerIdx);
+  auto beta2_power = input_tensors_.at(kBeta2PowerIdx);
+  auto learning_rate = reinterpret_cast<float *>(
+    input_tensors_.at(kLearningRateIdx)->MutableData())[0];  // use model origin lr, unsupported to config
+  auto beta1 = reinterpret_cast<float *>(input_tensors_.at(kBeta1Idx)->MutableData())[0];
+  auto beta2 = reinterpret_cast<float *>(input_tensors_.at(kBeta2Idx)->MutableData())[0];
+  auto eps = reinterpret_cast<float *>(input_tensors_.at(kEpsilonIdx)->MutableData())[0];
+  auto gradient = input_tensors_.at(kGradientIdx);
+  int length = input_tensors_.at(kWeightIdx)->ElementsNum();
+
+  // attribute
+  auto *adam_param = reinterpret_cast<AdamParameter *>(parameter_);
+  Collect(context,
+          {
+            "nnacl/fp32/adam_fp32.h",
+          },
+          {
+            "adam_fp32.c",
+          });
+  NNaclFp32Serializer code;
+  code.CodeFunction("DoAdam", m, v, gradient, weight, beta1, beta2, beta1_power, beta2_power, eps, learning_rate,
+                    adam_param->use_nesterov_, 0, length);
+  context->AppendCode(code.str());
+  return lite::RET_OK;
+}
+
+REG_OPERATOR_CODER(kAllTargets, kNumberTypeFloat32, PrimitiveType_Adam, CPUOpCoderCreator<AdamCoder>)
+}  // namespace mindspore::lite::micro::nnacl

mindspore/lite/tools/converter/micro/coder/opcoders/nnacl/fp32_grad/adam_coder.h

@@ -0,0 +1,38 @@
+/**
+ * Copyright 2022 Huawei Technologies Co., Ltd
+ *
+ * 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.
+ */
+
+#ifndef MINDSPORE_LITE_TOOLS_CONVERTER_MICRO_CODER_OPCODERS_NNACL_FP32_GRAD_ADAM_CODER_H_
+#define MINDSPORE_LITE_TOOLS_CONVERTER_MICRO_CODER_OPCODERS_NNACL_FP32_GRAD_ADAM_CODER_H_
+
+#include <vector>
+#include "coder/opcoders/op_coder.h"
+
+namespace mindspore::lite::micro::nnacl {
+class AdamCoder final : public OperatorCoder {
+ public:
+  AdamCoder(const std::vector<Tensor *> &in_tensors, const std::vector<Tensor *> &out_tensors,
+            const LiteGraph::Node *node, size_t node_index, Target target)
+      : OperatorCoder(in_tensors, out_tensors, node, node_index, target) {}
+
+  ~AdamCoder() override = default;
+
+  int Prepare(CoderContext *const context) override { return RET_OK; }
+
+  int DoCode(CoderContext *const context) override;
+};
+}  // namespace mindspore::lite::micro::nnacl
+
+#endif  // MINDSPORE_LITE_TOOLS_CONVERTER_MICRO_CODER_OPCODERS_NNACL_FP32_GRAD_ADAM_CODER_H_

mindspore/lite/tools/converter/micro/coder/opcoders/nnacl/fp32_grad/assign_coder.cc

@@ -0,0 +1,48 @@
+/**
+ * Copyright 2022 Huawei Technologies Co., Ltd
+ *
+ * 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.
+ */
+
+#include "coder/opcoders/nnacl/fp32_grad/assign_coder.h"
+#include <string>
+#include "schema/inner/ops_generated.h"
+#include "coder/opcoders/serializers/nnacl_serializer/nnacl_fp32_serializer.h"
+
+namespace mindspore::lite::micro::nnacl {
+using mindspore::schema::PrimitiveType_Assign;
+
+int AssignCoder::Prepare(CoderContext *const context) { return RET_OK; }
+
+int AssignCoder::DoCode(CoderContext *const context) {
+  MS_CHECK_TRUE(input_tensors_.size() == DIMENSION_2D, "inputs size is not equal to two");
+  Tensor *input0 = input_tensors_.at(0);
+  Tensor *input1 = input_tensors_.at(1);
+  if (input0->Size() != input1->Size()) {
+    MS_LOG(ERROR) << "input0 size: " << input0->Size() << ", input1 size: " << input1->Size();
+    return RET_ERROR;
+  }
+
+  NNaclFp32Serializer code;
+  // Get Tensor Pointer
+  std::string input0_str = allocator_->GetRuntimeAddr(input0);
+  std::string input1_str = allocator_->GetRuntimeAddr(input1);
+  size_t data_size = input0->Size();
+  // assign, just assign input1'data to input0
+  code.CodeFunction("memcpy", input0_str, input1_str, data_size);
+  context->AppendCode(code.str());
+  return RET_OK;
+}
+
+REG_OPERATOR_CODER(kAllTargets, kNumberTypeFloat32, PrimitiveType_Assign, CPUOpCoderCreator<AssignCoder>)
+}  // namespace mindspore::lite::micro::nnacl

mindspore/lite/tools/converter/micro/coder/opcoders/nnacl/fp32_grad/assign_coder.h

@@ -0,0 +1,37 @@
+/**
+ * Copyright 2022 Huawei Technologies Co., Ltd
+ *
+ * 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.
+ */
+
+#ifndef MINDSPORE_LITE_TOOLS_CONVERTER_MICRO_CODER_OPCODERS_NNACL_FP32_GRAD_ASSIGN_CODER_H_
+#define MINDSPORE_LITE_TOOLS_CONVERTER_MICRO_CODER_OPCODERS_NNACL_FP32_GRAD_ASSIGN_CODER_H_
+
+#include <vector>
+#include "coder/opcoders/op_coder.h"
+
+namespace mindspore::lite::micro::nnacl {
+class AssignCoder final : public OperatorCoder {
+ public:
+  AssignCoder(const std::vector<Tensor *> &in_tensors, const std::vector<Tensor *> &out_tensors,
+              const LiteGraph::Node *node, size_t node_index, Target target)
+      : OperatorCoder(in_tensors, out_tensors, node, node_index, target) {}
+  ~AssignCoder() override = default;
+
+  int Prepare(CoderContext *const context) override;
+
+  int DoCode(CoderContext *const context) override;
+};
+}  // namespace mindspore::lite::micro::nnacl
+
+#endif  // MINDSPORE_LITE_TOOLS_CONVERTER_MICRO_CODER_OPCODERS_NNACL_FP32_GRAD_ASSIGN_CODER_H_

mindspore/lite/tools/converter/micro/coder/opcoders/nnacl/fp32_grad/biasadd_grad_coder.cc

@@ -0,0 +1,68 @@
+/**
+ * Copyright 2022 Huawei Technologies Co., Ltd
+ *
+ * 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.
+ */
+
+#include "coder/opcoders/nnacl/fp32_grad/biasadd_grad_coder.h"
+#include <string>
+#include "schema/inner/ops_generated.h"
+#include "coder/opcoders/serializers/nnacl_serializer/nnacl_fp32_serializer.h"
+
+namespace mindspore::lite::micro::nnacl {
+using mindspore::schema::PrimitiveType_BiasAddGrad;
+
+int BiasAddGradCoder::Prepare(CoderContext *const context) {
+  auto dims = input_tensor_->shape();
+  auto *bias_param = reinterpret_cast<ArithmeticParameter *>(parameter_);
+  bias_param->ndim_ = dims.size();
+  for (unsigned int i = 0; i < bias_param->ndim_; i++) {
+    bias_param->in_shape0_[i] = dims[i];
+    bias_param->out_shape_[i] = 1;  // 1 dimension for N,H,W,
+  }
+  bias_param->out_shape_[bias_param->ndim_ - 1] = dims[bias_param->ndim_ - 1];
+  for (auto i = bias_param->ndim_; i < DIMENSION_4D; i++) {
+    bias_param->in_shape0_[i] = 0;
+    bias_param->out_shape_[i] = 0;
+  }
+  return RET_OK;
+}
+
+int BiasAddGradCoder::DoCode(CoderContext *const context) {
+  auto *bias_param = reinterpret_cast<ArithmeticParameter *>(parameter_);
+  size_t nhw_size = 1;
+  size_t channels = bias_param->in_shape0_[bias_param->ndim_ - 1];  // C in NHWC
+  for (size_t i = 0; i < bias_param->ndim_ - 1; i++) {
+    nhw_size *= static_cast<size_t>(bias_param->in_shape0_[i]);
+  }
+
+  size_t total_size = channels * nhw_size;
+
+  NNaclFp32Serializer code;
+  // Get Tensor Pointer
+  std::string input_str = allocator_->GetRuntimeAddr(input_tensor_);
+  std::string output_str = allocator_->GetRuntimeAddr(output_tensor_);
+
+  code << "\t\tfor (size_t c = 0; c < " << channels << "; ++c) {\n";
+  code << "\t\t\t(" << output_str << ")[c] = 0;\n";
+  code << "\t\t\tfor (size_t offset = 0; offset < " << total_size << "; offset += " << channels << ") {\n";
+  code << "\t\t\t\t(" << output_str << ")[c] += (" << input_str << ")[offset + c];\n";
+  code << "\t\t\t}\n";
+  code << "\t\t}\n";
+
+  context->AppendCode(code.str());
+  return RET_OK;
+}
+
+REG_OPERATOR_CODER(kAllTargets, kNumberTypeFloat32, PrimitiveType_BiasAddGrad, CPUOpCoderCreator<BiasAddGradCoder>)
+}  // namespace mindspore::lite::micro::nnacl

mindspore/lite/tools/converter/micro/coder/opcoders/nnacl/fp32_grad/biasadd_grad_coder.h

@@ -0,0 +1,37 @@
+/**
+ * Copyright 2022 Huawei Technologies Co., Ltd
+ *
+ * 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.
+ */
+
+#ifndef MINDSPORE_LITE_TOOLS_CONVERTER_MICRO_CODER_OPCODERS_NNACL_FP32_GRAD_BIASADD_GRAD_CODER_H_
+#define MINDSPORE_LITE_TOOLS_CONVERTER_MICRO_CODER_OPCODERS_NNACL_FP32_GRAD_BIASADD_GRAD_CODER_H_
+
+#include <vector>
+#include "coder/opcoders/op_coder.h"
+
+namespace mindspore::lite::micro::nnacl {
+class BiasAddGradCoder final : public OperatorCoder {
+ public:
+  BiasAddGradCoder(const std::vector<Tensor *> &in_tensors, const std::vector<Tensor *> &out_tensors,
+                   const LiteGraph::Node *node, size_t node_index, Target target)
+      : OperatorCoder(in_tensors, out_tensors, node, node_index, target) {}
+  ~BiasAddGradCoder() override = default;
+
+  int Prepare(CoderContext *const context) override;
+
+  int DoCode(CoderContext *const context) override;
+};
+}  // namespace mindspore::lite::micro::nnacl
+
+#endif  // MINDSPORE_LITE_TOOLS_CONVERTER_MICRO_CODER_OPCODERS_NNACL_FP32_GRAD_BIASADD_GRAD_CODER_H_

mindspore/lite/tools/converter/micro/coder/opcoders/nnacl/fp32_grad/softmax_cross_entropy_with_logits_coder.cc

@@ -0,0 +1,80 @@
+/**
+ * Copyright 2022 Huawei Technologies Co., Ltd
+ *
+ * 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.
+ */
+
+#include "coder/opcoders/nnacl/fp32_grad/softmax_cross_entropy_with_logits_coder.h"
+#include <string>
+#include "nnacl/fp32_grad/softmax_crossentropy_parameter.h"
+#include "coder/opcoders/file_collector.h"
+#include "schema/inner/ops_generated.h"
+#include "coder/opcoders/serializers/nnacl_serializer/nnacl_fp32_serializer.h"
+
+namespace mindspore::lite::micro::nnacl {
+using mindspore::schema::PrimitiveType_SoftmaxCrossEntropyWithLogits;
+
+int SoftmaxCrossEntropyWithLogitsCoder::Prepare(CoderContext *const context) {
+  MS_CHECK_TRUE(input_tensor_ != nullptr, "input_tensor is nullptr.");
+  size_t data_size = input_tensor_->ElementsNum();
+  auto dims = input_tensor_->shape();
+  auto *softmax_cross_entropy_param = reinterpret_cast<SoftmaxCrossEntropyParameter *>(parameter_);
+  softmax_cross_entropy_param->n_dim_ = DIMENSION_2D;
+  CHECK_LESS_RETURN(dims.size(), DIMENSION_2D);
+  softmax_cross_entropy_param->number_of_classes_ = dims.at(1);
+  softmax_cross_entropy_param->batch_size_ = dims.at(0);
+
+  losses_ = reinterpret_cast<float *>(allocator_->Malloc(kNumberTypeFloat32, data_size * sizeof(float), kWorkspace));
+  sum_data_ = reinterpret_cast<float *>(allocator_->Malloc(kNumberTypeFloat32, dims[0] * sizeof(float), kWorkspace));
+  softmax_params_.n_dim_ = DIMENSION_2D;
+  softmax_params_.element_size_ = data_size;
+  softmax_params_.axis_ = 1;
+  for (size_t i = 0; i < dims.size(); i++) {
+    softmax_params_.input_shape_[i] = dims.at(i);
+  }
+  return RET_OK;
+}
+
+int SoftmaxCrossEntropyWithLogitsCoder::DoCode(CoderContext *const context) {
+  MS_CHECK_TRUE(input_tensors_.size() == DIMENSION_2D, "inputs size is not equal to two");
+  Collect(context,
+          {
+            "nnacl/fp32/softmax_fp32.h",
+            "nnacl/fp32_grad/softmax_cross_entropy_with_logits.h",
+          },
+          {
+            "softmax_fp32.c",
+            "softmax_cross_entropy_with_logits.c",
+          });
+  NNaclFp32Serializer code, init_code;
+  code.CodeStruct("softmax_params", softmax_params_);
+  // Get Tensor Pointer
+  std::string in_str = allocator_->GetRuntimeAddr(input_tensor_);
+  std::string labels_str = allocator_->GetRuntimeAddr(input_tensors_.at(1));
+  std::string out_str = allocator_->GetRuntimeAddr(output_tensor_);
+  std::string grad_str = "NULL";
+  if (output_tensors_.size() > 1) {
+    grad_str = allocator_->GetRuntimeAddr(output_tensors_.at(1));
+  }
+  auto *softmax_cross_entropy_param = reinterpret_cast<SoftmaxCrossEntropyParameter *>(parameter_);
+  code.CodeFunction("Softmax", in_str, losses_, sum_data_, "&softmax_params");
+  code.CodeFunction("ForwardPostExecute", labels_str, losses_, grad_str, out_str,
+                    softmax_cross_entropy_param->number_of_classes_, softmax_cross_entropy_param->batch_size_);
+
+  context->AppendCode(code.str());
+  return RET_OK;
+}
+
+REG_OPERATOR_CODER(kAllTargets, kNumberTypeFloat32, PrimitiveType_SoftmaxCrossEntropyWithLogits,
+                   CPUOpCoderCreator<SoftmaxCrossEntropyWithLogitsCoder>)
+}  // namespace mindspore::lite::micro::nnacl

mindspore/lite/tools/converter/micro/coder/opcoders/nnacl/fp32_grad/softmax_cross_entropy_with_logits_coder.h

@@ -0,0 +1,43 @@
+/**
+ * Copyright 2022 Huawei Technologies Co., Ltd
+ *
+ * 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.
+ */
+
+#ifndef MINDSPORE_LITE_TOOLS_CONVERTER_MICRO_CODER_OPCODERS_NNACL_FP32_GRAD_SOFTMAX_CROSS_ENTROPY_WITH_LOGITS_CODER_H_
+#define MINDSPORE_LITE_TOOLS_CONVERTER_MICRO_CODER_OPCODERS_NNACL_FP32_GRAD_SOFTMAX_CROSS_ENTROPY_WITH_LOGITS_CODER_H_
+
+#include <vector>
+#include "nnacl/softmax_parameter.h"
+#include "coder/opcoders/op_coder.h"
+
+namespace mindspore::lite::micro::nnacl {
+class SoftmaxCrossEntropyWithLogitsCoder final : public OperatorCoder {
+ public:
+  SoftmaxCrossEntropyWithLogitsCoder(const std::vector<Tensor *> &in_tensors, const std::vector<Tensor *> &out_tensors,
+                                     const LiteGraph::Node *node, size_t node_index, Target target)
+      : OperatorCoder(in_tensors, out_tensors, node, node_index, target) {}
+  ~SoftmaxCrossEntropyWithLogitsCoder() override = default;
+
+  int Prepare(CoderContext *const context) override;
+
+  int DoCode(CoderContext *const context) override;
+
+ private:
+  SoftmaxParameter softmax_params_;
+  float *losses_{nullptr};
+  float *sum_data_{nullptr};
+};
+}  // namespace mindspore::lite::micro::nnacl
+
+#endif  // MICRO_CODER_OPCODERS_NNACL_FP32_GRAD_SOFTMAX_CROSS_ENTROPY_WITH_LOGITS_CODER_H_

mindspore/lite/tools/converter/micro/coder/session.cc

@@ -42,7 +42,8 @@
 namespace mindspore::lite::micro {
 CoderSession::CoderSession() { allocator_ = MemoryAllocator::GetInstance(); }
 
-void CoderSession::EndCode() {
+int CoderSession::EndCode() {
+  int ret = RET_OK;
   context_->set_tensor_map(allocator_->tensors_map());
   context_->set_saved_weights(allocator_->saved_weights());
   size_t de_quant_max_workspace_size = nnacl::Dequant::GetInstance()->de_quant_max_workspace();
@@ -59,8 +60,10 @@ void CoderSession::EndCode() {
     context_->set_code_blocks(blocks);
   }
   if (config->code_mode() == Train) {
-    Train::TransformGraphForTrain(context_.get(), op_coders_, schema_version_);
+    ret = Train::TransformGraphForTrain(context_.get(), op_coders_, schema_version_);
+    MS_CHECK_RET_CODE(ret, "transform graph for train failed.");
   }
+  return ret;
 }
 
 int CoderSession::Run() {
@@ -85,7 +88,8 @@ int CoderSession::Run() {
     MS_CHECK_RET_CODE(ret, "do coder " << op_coder->name() << " failed");
   }
 
-  this->EndCode();
+  ret = this->EndCode();
+  MS_CHECK_RET_CODE(ret, "End code failed.");
   MS_LOG(INFO) << "run opcoders success";
   return RET_OK;
 }
@@ -220,6 +224,8 @@ int CoderSession::CreateOpCoders() {
   CodeMode code_mode = config->code_mode();
   bool support_parallel = config->support_parallel();
   uint32_t nodes_size = model->graph_.all_nodes_.size();
+  std::vector<lite::Tensor *> all_tensors = coder_graph_->all_tensors();
+  MS_CHECK_TRUE_MSG(!all_tensors.empty(), RET_ERROR, "coder_graph has no any tensors");
   OpCoderBuilder builder;
   for (uint32_t i = 0; i < nodes_size; ++i) {
     const auto *node = model->graph_.all_nodes_.at(i);
@@ -227,19 +233,9 @@ int CoderSession::CreateOpCoders() {
       MS_LOG(ERROR) << "node is nullptr";
       return RET_ERROR;
     }
-    std::vector<lite::Tensor *> all_tensors = coder_graph_->all_tensors();
-    if (all_tensors.empty()) {
-      MS_LOG(ERROR) << "coder_graph has no any tensors";
-      return RET_ERROR;
-    }
     // set op_coder's inputs && outputs info
-    std::vector<uint32_t> input_indices;
+    std::vector<uint32_t> input_indices(node->input_indices_);
-    std::vector<uint32_t> node_input_indices = node->input_indices_;
+    std::vector<uint32_t> output_indices(node->output_indices_);
-    input_indices.insert(input_indices.end(), node_input_indices.begin(), node_input_indices.end());
-    std::vector<uint32_t> output_indices;
-    std::vector<uint32_t> node_output_indices = node->output_indices_;
-    output_indices.insert(output_indices.end(), node_output_indices.begin(), node_output_indices.end());
-
     std::vector<lite::Tensor *> inputs;
     std::vector<lite::Tensor *> outputs;
     for (auto in_index : input_indices) {

mindspore/lite/tools/converter/micro/coder/session.h

@@ -50,7 +50,7 @@ class CoderSession {
   int CreateOpCoders();
   int InitCodeGraph();
   int CompileGraph();
-  void EndCode();
+  int EndCode();
 
   std::unique_ptr<CoderGraph> coder_graph_{nullptr};
   std::unique_ptr<CoderContext> context_{nullptr};

mindspore/lite/tools/converter/micro/coder/train.cc

@@ -60,12 +60,13 @@ int Train::TransformGraphForTrain(CoderContext *context, const std::vector<std::
     MS_LOG(INFO) << "input context invalid";
     return RET_ERROR;
   }
-  const std::array<int, 6> loss_types = {schema::PrimitiveType_SparseSoftmaxCrossEntropyWithLogits,
+  const std::array<int, 7> loss_types = {schema::PrimitiveType_SparseSoftmaxCrossEntropyWithLogits,
                                          schema::PrimitiveType_BinaryCrossEntropy,
                                          schema::PrimitiveType_SmoothL1Loss,
                                          schema::PrimitiveType_SmoothL1LossGrad,
                                          schema::PrimitiveType_SigmoidCrossEntropyWithLogits,
-                                         schema::PrimitiveType_SigmoidCrossEntropyWithLogitsGrad};
+                                         schema::PrimitiveType_SigmoidCrossEntropyWithLogitsGrad,
+                                         schema::PrimitiveType_SoftmaxCrossEntropyWithLogits};
   OperatorCoder *loss_op = nullptr;
   for (const auto &opcoder : op_coders) {
     const LiteGraph::Node *node = opcoder->node();

mindspore/lite/tools/converter/micro/coder/utils/coder_utils.cc

@@ -1,5 +1,5 @@
 /**
- * Copyright 2021 Huawei Technologies Co., Ltd
+ * Copyright 2021-2022 Huawei Technologies Co., Ltd
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -112,6 +112,11 @@ std::vector<std::string> AddDumpDataInfo(const std::vector<std::string> &blocks,
     auto &opcoder = opcoders.at(i);
     std::string code = blocks.at(i);
     std::string name = opcoder->name();
+    auto pos = name.find_first_of('/');
+    while (pos != std::string::npos) {
+      name.replace(pos, 1, ".");
+      pos = name.find_first_of('/');
+    }
     code += "    {\n";
     code += "      FILE *output_file = fopen(\"./" + name + ".ir\", \"w\");\n";
     code += "      fprintf(output_file, \"Node:" + name + "\\n\");\n";