!10754 fix npu fusion bug

From: @yeyunpeng2020
Reviewed-by: 
Signed-off-by:

mindspore/lite/src/ops/populate/deconv2d_populate.cc

@@ -46,6 +46,7 @@ OpParameter *PopulateDeconvParameter(const mindspore::lite::PrimitiveC *primitiv
   conv_param->pad_r_ = deconv_lite_primitive->PadRight();
   conv_param->dilation_h_ = conv_primitive->GetDilateH();
   conv_param->dilation_w_ = conv_primitive->GetDilateW();
+  conv_param->group_ = conv_primitive->GetGroup();
   auto act_type = conv_primitive->GetActivationType();
   switch (act_type) {
     case schema::ActivationType_RELU:

mindspore/lite/src/runtime/kernel/npu/deconvolution_npu.cc

@@ -24,13 +24,17 @@ using mindspore::schema::PrimitiveType_DeConv2D;
 namespace mindspore::kernel {
 int DeconvolutionNPUKernel::IsSupport(const std::vector<lite::Tensor *> &inputs,
                                       const std::vector<lite::Tensor *> &outputs, OpParameter *opParameter) {
+  if (conv_param_->group_ != 1) {
+    MS_LOG(WARNING) << "Only support group equals 1 for npu deconvolution op";
+    return RET_ERROR;
+  }
   return RET_OK;
 }
 
 int DeconvolutionNPUKernel::SetConvParam() {
   deconv_->set_attr_strides(ge::AttrValue::LIST_INT({conv_param_->stride_h_, conv_param_->stride_w_}));
   deconv_->set_attr_dilations(ge::AttrValue::LIST_INT({conv_param_->dilation_h_, conv_param_->dilation_w_}));
-  deconv_->set_attr_groups(1);
+  deconv_->set_attr_groups(conv_param_->group_);
 
   if (conv_param_->pad_mode_ == Pad_Same) {
     deconv_->set_attr_pad_mode(ge::AttrValue::STR{"SAME"});

mindspore/lite/test/models_npu.cfg

@@ -1,3 +1,4 @@
 mobilenet_v2_1.0_224.tflite 2.5
 squeezenet.tflite 2.5
 inception_v3.tflite 1
+porseg_tmp.onnx 1 2

mindspore/lite/test/models_with_several_inputs_or_without_outputs.cfg

@@ -30,3 +30,4 @@ ml_ei_facedetection.onnx
 quant_aware_bank_card_detection_inception.onnx
 quant_aware_bank_card_recognition_fcny.onnx
 quant_aware_identify_card_detect.onnx
+porseg_tmp.onnx;2

mindspore/lite/test/run_benchmark_nets.sh

@@ -1361,10 +1361,21 @@ function Run_arm64() {
     while read line; do
         model_name=`echo ${line}|awk -F ' ' '{print $1}'`
         accuracy_limit=`echo ${line}|awk -F ' ' '{print $2}'`
+        input_num=`echo ${line}|awk -F ' ' '{print $3}'`
+        data_path="/data/local/tmp/input_output/"
+        input_files=''
+        if [[ -z "$input_num" || $input_num == 1 ]]; then
+          input_files=${data_path}'input/'$model_name'.ms.bin'
+        elif [[ ! -z "$input_num" && $input_num -gt 1 ]]; then
+          for i in $(seq 1 $input_num)
+          do
+            input_files=$input_files${data_path}'input/'$model_name'.ms.bin_'$i','
+          done
+        fi
         echo "mindspore run npu: ${model_name}, accuracy limit:${accuracy_limit}" >> "${run_arm64_log_file}"
         echo 'cd  /data/local/tmp/benchmark_test' > adb_run_cmd.txt
-        echo 'export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/data/local/tmp/benchmark_test;./benchmark --device=NPU --modelFile='${model_name}'.ms --inDataFile=/data/local/tmp/input_output/input/'${model_name}'.ms.bin --benchmarkDataFile=/data/local/tmp/input_output/output/'${model_name}'.ms.out --accuracyThreshold='${accuracy_limit} >> "${run_arm64_log_file}"
-        echo 'export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/data/local/tmp/benchmark_test;./benchmark --device=NPU --modelFile='${model_name}'.ms --inDataFile=/data/local/tmp/input_output/input/'${model_name}'.ms.bin --benchmarkDataFile=/data/local/tmp/input_output/output/'${model_name}'.ms.out --accuracyThreshold='${accuracy_limit} >> adb_run_cmd.txt
+        echo 'export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/data/local/tmp/benchmark_test;./benchmark --device=NPU --modelFile='${model_name}'.ms --inDataFile='${input_files}' --benchmarkDataFile=/data/local/tmp/input_output/output/'${model_name}'.ms.out --accuracyThreshold='${accuracy_limit} >> "${run_arm64_log_file}"
+        echo 'export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/data/local/tmp/benchmark_test;./benchmark --device=NPU --modelFile='${model_name}'.ms --inDataFile='${input_files}' --benchmarkDataFile=/data/local/tmp/input_output/output/'${model_name}'.ms.out --accuracyThreshold='${accuracy_limit} >> adb_run_cmd.txt
         adb -s ${device_id} shell < adb_run_cmd.txt >> "${run_arm64_log_file}"
         if [ $? = 0 ]; then
             run_result='arm64_npu: '${model_name}' pass'; echo ${run_result} >> ${run_benchmark_result_file}
@@ -1385,7 +1396,8 @@ function Run_arm64() {
         input_files=''
         output_file=''
         data_path="/data/local/tmp/input_output/"
-        if [[ -z "$input_num" || $input_num == 1 ]] && [ -e ${data_path}'input/'${model_name}'.ms.bin' ]; then
+        model_x86_path="/home/workspace/mindspore_dataset/mslite/models/hiai/input_output/"
+        if [[ -z "$input_num" || $input_num == 1 ]] && [ -e ${model_x86_path}'input/'${model_name}'.ms.bin' ]; then
           input_files=${data_path}'input/'$model_name'.ms.bin'
         elif [[ ! -z "$input_num" && $input_num -gt 1 ]]; then
           for i in $(seq 1 $input_num)
@@ -1393,7 +1405,7 @@ function Run_arm64() {
             input_files=$input_files${data_path}'input/'$model_name'.ms.bin_'$i','
           done
         fi
-        if [ -e ${data_path}'output/'${model_name}'.ms.out' ]; then
+        if [ -e ${model_x86_path}'output/'${model_name}'.ms.out' ]; then
           output_file=${data_path}'output/'${model_name}'.ms.out'
         fi
         if [[ ${model_name##*.} == "caffemodel" ]]; then