Merge pull request #4381 from LDOUBLEV/test_v10

add precision to mkldnn

PTDN/test_train_inference_python.sh

@@ -141,10 +141,15 @@ function func_inference(){
                 fi
                 for threads in ${cpu_threads_list[*]}; do
                     for batch_size in ${batch_size_list[*]}; do
-                        precison="fp32"
-                        if [ ${use_mkldnn} = "False" ] && [ ${_flag_quant} = "True" ]; then
-                            precision="int8"
-                        fi
+                        for precision in ${precision_list[*]}; do
+                            if [ ${use_mkldnn} = "False" ] && [ ${precision} = "fp16" ]; then
+                                continue
+                            fi # skip when enable fp16 but disable mkldnn
+                            if [ ${_flag_quant} = "True" ] && [ ${precision} != "int8" ]; then
+                                continue
+                            fi # skip when quant model inference but precision is not int8
+                            set_precision=$(func_set_params "${precision_key}" "${precision}")
+                            
                             _save_log_path="${_log_path}/python_infer_cpu_usemkldnn_${use_mkldnn}_threads_${threads}_precision_${precision}_batchsize_${batch_size}.log"
                             set_infer_data=$(func_set_params "${image_dir_key}" "${_img_dir}")
                             set_benchmark=$(func_set_params "${benchmark_key}" "${benchmark_value}")
@@ -152,7 +157,7 @@ function func_inference(){
                             set_cpu_threads=$(func_set_params "${cpu_threads_key}" "${threads}")
                             set_model_dir=$(func_set_params "${infer_model_key}" "${_model_dir}")
                             set_infer_params1=$(func_set_params "${infer_key1}" "${infer_value1}")
-                        command="${_python} ${_script} ${use_gpu_key}=${use_gpu} ${use_mkldnn_key}=${use_mkldnn} ${set_cpu_threads} ${set_model_dir} ${set_batchsize} ${set_infer_data} ${set_benchmark} ${set_infer_params1} > ${_save_log_path} 2>&1 "
+                            command="${_python} ${_script} ${use_gpu_key}=${use_gpu} ${use_mkldnn_key}=${use_mkldnn} ${set_cpu_threads} ${set_model_dir} ${set_batchsize} ${set_infer_data} ${set_benchmark} ${set_precision} ${set_infer_params1} > ${_save_log_path} 2>&1 "
                             eval $command
                             last_status=${PIPESTATUS[0]}
                             eval "cat ${_save_log_path}"
@@ -160,6 +165,7 @@ function func_inference(){
                         done
                     done
                 done
+            done
         elif [ ${use_gpu} = "True" ] || [ ${use_gpu} = "gpu" ]; then
             for use_trt in ${use_trt_list[*]}; do
                 for precision in ${precision_list[*]}; do

tools/infer/utility.py

@@ -267,10 +267,11 @@ def create_predictor(args, mode, logger):
             # cache 10 different shapes for mkldnn to avoid memory leak
             config.set_mkldnn_cache_capacity(10)
             config.enable_mkldnn()
-
+            if args.precision == "fp16":
+                config.enable_mkldnn_bfloat16()
     # enable memory optim
     config.enable_memory_optim()
-    #config.disable_glog_info()
+    config.disable_glog_info()
 
     config.delete_pass("conv_transpose_eltwiseadd_bn_fuse_pass")
     if mode == 'table':