!4210 [MS][LITE] fix bug of arm cpu fp32 conv_depthwise: only support group equals output channel

Merge pull request !4210 from yangruoqi713/lite

mindspore/lite/src/ops/convolution_depthwise.cc

@@ -40,6 +40,7 @@ int DepthwiseConv2D::InferShape(std::vector<tensor::Tensor *> inputs_, std::vect
   auto in_shape = input->shape();
   int input_h = in_shape.at(1);
   int input_w = in_shape.at(2);
+  int input_channel = in_shape.at(3);
   int output_w = 0, output_h = 0;
 
   auto conv_prim = this->primitive->value_as_DepthwiseConv2D();
@@ -69,6 +70,10 @@ int DepthwiseConv2D::InferShape(std::vector<tensor::Tensor *> inputs_, std::vect
   std::vector<int> out_shape{input->shape()};
   out_shape.at(1) = output_h;
   out_shape.at(2) = output_w;
+  if (conv_prim->channelMultiplier() * input_channel != weight->shape()[0]) {
+    MS_LOG(ERROR) << "Conv depthwise only support group equals output channel.";
+    return RET_ERROR;
+  }
   out_shape.at(3) = weight->shape()[0] * weight->shape()[3];  // in_channel * out_channel
 
   output->set_shape(out_shape);

mindspore/lite/src/ops/deconvolution_depthwise.cc

@@ -40,6 +40,7 @@ int DeconvDepthwiseConv2D::InferShape(std::vector<tensor::Tensor *> inputs_, std
   auto in_shape = input->shape();
   int input_h = in_shape.at(1);
   int input_w = in_shape.at(2);
+  int input_channel = in_shape.at(3);
   int output_w = 0, output_h = 0;
 
   auto conv_prim = this->primitive->value_as_DeDepthwiseConv2D();
@@ -58,6 +59,10 @@ int DeconvDepthwiseConv2D::InferShape(std::vector<tensor::Tensor *> inputs_, std
   std::vector<int> out_shape{input->shape()};
   out_shape.at(1) = output_h;
   out_shape.at(2) = output_w;
+  if (conv_prim->channelMultiplier() * input_channel != weight->shape()[0]) {
+    MS_LOG(ERROR) << "Conv depthwise only support group equals output channel.";
+    return RET_ERROR;
+  }
   out_shape.at(3) = weight->shape()[0] * weight->shape()[3];  // in_channel * out_channel
 
   output->set_shape(out_shape);

mindspore/lite/src/runtime/kernel/arm/nnacl/fp32/conv_depthwise.cc

@@ -53,7 +53,7 @@ void InitSlidingParam(SlidingWindowParam *sliding, const ConvParameter *conv_par
   sliding->in_step_ = conv_param->input_h_ * conv_param->input_w_ * sliding->block_channel_;  // for batch loop
   sliding->in_h_step_ = conv_param->input_w_ * sliding->block_channel_;
   sliding->in_sh_step_ = conv_param->input_w_ * sliding->block_channel_ * conv_param->stride_h_;    // stride H
-  sliding->in_sw_step_ = sliding->block_channel_ * conv_param->stride_h_;                           // stride W
+  sliding->in_sw_step_ = sliding->block_channel_ * conv_param->stride_w_;                           // stride W
   sliding->in_kh_step_ = conv_param->input_w_ * sliding->block_channel_ * conv_param->dilation_h_;  // kernel H
   sliding->in_kw_step_ = sliding->block_channel_ * conv_param->dilation_w_;                         // kernel W
   sliding->kernel_step_ = conv_param->kernel_w_ * conv_param->kernel_h_ * block;

mindspore/lite/tools/converter/parser/caffe/caffe_convolution_parser.cc

@@ -20,7 +20,7 @@
 namespace mindspore {
 namespace lite {
 void CaffeConvolutionParser::ParseGroupConvolution(schema::CNodeT *op, schema::Conv2DT *attr) {
-  if (attr == nullptr || attr->group == 1 || attr->group != attr->channelOut) {
+  if (attr == nullptr || attr->group == 1) {
     return;
   }
   std::unique_ptr<schema::DepthwiseConv2DT> depthwiseConv2DParam(new schema::DepthwiseConv2DT());

mindspore/lite/tools/converter/parser/caffe/caffe_deconvolution_parser.cc

@@ -20,7 +20,7 @@
 namespace mindspore {
 namespace lite {
 void CaffeDeconvolutionParser::ParseGroupDeconvolution(schema::CNodeT *op, schema::DeConv2DT *attr) {
-  if (attr == nullptr || attr->group == 1 || attr->group != attr->channelIn) {
+  if (attr == nullptr || attr->group == 1) {
     return;
   }