tipc for npu

test_tipc/prepare_npu_test.sh

@@ -0,0 +1,25 @@
+#!/bin/bash
+
+BASEDIR=$(dirname "$0")
+
+function readlinkf() {
+    perl -MCwd -e 'print Cwd::abs_path shift' "$1";
+}
+
+REPO_ROOT_PATH=$(readlinkf ${BASEDIR}/../)
+
+config_files=$(find ${REPO_ROOT_PATH}/test_tipc/configs -name "train_infer_python.txt")
+for file in ${config_files}; do
+   echo $file
+   sed -i "s/runner.use_gpu/runner.use_npu/g" $file
+   sed -i '30s/$/ runner.use_gpu=False/' $file
+   sed -i '40s/$/ --use_gpu=false/' $file
+   sed -i "s/--use_gpu:/--use_npu:/g" $file
+done
+
+yaml_files=$(find ${REPO_ROOT_PATH}/models/ -name "*.yaml")
+for file in ${yaml_files}; do
+   echo $file
+   sed -i "s/use_gpu: True/use_gpu: False/g" $file
+done
+

test_tipc/test_train_inference_python.sh

@@ -399,7 +399,7 @@ else
                         status_check $? "${cmd}" "${status_log}"
                     done
                 else     # train with multi-machine
-                    cmd="${python} -m paddle.distributed.launch --ips=${ips} --gpus=${gpu} ${run_train} ${set_use_gpu} ${set_save_model} ${set_pretrain} ${set_epoch} ${set_autocast} ${set_batchsize} ${set_train_params1} ${set_amp_config}"
+                    cmd="${python} -m paddle.distributed.launch --ips=${ips} --devices=${gpu} ${run_train} ${set_use_gpu} ${set_save_model} ${set_pretrain} ${set_epoch} ${set_autocast} ${set_batchsize} ${set_train_params1} ${set_amp_config}"
                     eval "unset CUDA_VISIBLE_DEVICES"
                     eval $cmd
                     status_check $? "${cmd}" "${status_log}"

tools/paddle_infer.py

@@ -37,6 +37,7 @@ def parse_args():
     parser.add_argument("--params_file", type=str)
     parser.add_argument("--model_dir", type=str)
     parser.add_argument("--use_gpu", type=str)
+    parser.add_argument("--use_npu", type=str)
     parser.add_argument("--data_dir", type=str)
     parser.add_argument("--reader_file", type=str)
     parser.add_argument("--batchsize", type=int)
@@ -49,6 +50,7 @@ def parse_args():
     parser.add_argument("--precision", type=str)
     args = parser.parse_args()
     args.use_gpu = (True if args.use_gpu.lower() == "true" else False)
+    args.use_npu = (True if args.use_npu is not None and args.use_npu.lower() == "true" else False)
     args.enable_mkldnn = (True
                           if args.enable_mkldnn.lower() == "true" else False)
     args.enable_tensorRT = (True if args.enable_tensorRT.lower() == "true" else
@@ -86,6 +88,8 @@ def init_predictor(args):
                 max_batch_size=args.batchsize,
                 min_subgraph_size=9,
                 precision_mode=paddle.inference.PrecisionType.Float32)
+    elif args.use_npu:
+        config.enable_npu()
     else:
         config.disable_gpu()
         config.set_cpu_math_library_num_threads(args.cpu_threads)
@@ -114,7 +118,12 @@ def create_data_loader(args):
 
 def main(args):
     predictor, pred_config = init_predictor(args)
-    place = paddle.set_device('gpu' if args.use_gpu else 'cpu')
+    if args.use_gpu:
+        place = paddle.set_device('gpu')
+    elif args.use_npu:
+        place = paddle.set_device('npu')
+    else:
+        place = paddle.set_device('cpu')
     args.place = place
     input_names = predictor.get_input_names()
     output_names = predictor.get_output_names()
@@ -132,7 +141,7 @@ def main(args):
             inference_config=pred_config,
             pids=pid,
             process_name=None,
-            gpu_ids=0,
+            gpu_ids=0 if args.use_gpu else None,
             time_keys=[
                 'preprocess_time', 'inference_time', 'postprocess_time'
             ])