[NPU] add npu support for waveflow, test=develop

Author: qili93

.gitignore

@@ -4,4 +4,5 @@ output
 paddlerec.egg-info/
 *~
 *.pyc
-*.DS_Store
+*.DS_Store
+kernel_meta/

models/rank/wide_deep/net.py

@@ -35,10 +35,13 @@ def __init__(self, sparse_feature_number, sparse_feature_dim,
                 initializer=paddle.nn.initializer.TruncatedNormal(
                     mean=0.0, std=1.0 / math.sqrt(self.dense_feature_dim))))
 
+        use_sparse = True
+        if paddle.is_compiled_with_npu():
+            use_sparse = False
         self.embedding = paddle.nn.Embedding(
             self.sparse_feature_number,
             self.sparse_feature_dim,
-            sparse=True,
+            sparse=use_sparse,
             weight_attr=paddle.ParamAttr(
                 name="SparseFeatFactors",
                 initializer=paddle.nn.initializer.Uniform()))

tools/infer.py

@@ -66,6 +66,7 @@ def main(args):
     # tools.vars
     use_gpu = config.get("runner.use_gpu", True)
     use_xpu = config.get("runner.use_xpu", False)
+    use_npu = config.get("runner.use_npu", False)
     use_visual = config.get("runner.use_visual", False)
     test_data_dir = config.get("runner.test_data_dir", None)
     print_interval = config.get("runner.print_interval", None)
@@ -76,14 +77,18 @@ def main(args):
 
     logger.info("**************common.configs**********")
     logger.info(
-        "use_gpu: {}, use_xpu: {}, use_visual: {}, infer_batch_size: {}, test_data_dir: {}, start_epoch: {}, end_epoch: {}, print_interval: {}, model_load_path: {}".
-        format(use_gpu, use_xpu, use_visual, infer_batch_size, test_data_dir,
-               start_epoch, end_epoch, print_interval, model_load_path))
+        "use_gpu: {}, use_xpu: {}, use_npu: {}, use_visual: {}, infer_batch_size: {}, test_data_dir: {}, start_epoch: {}, end_epoch: {}, print_interval: {}, model_load_path: {}".
+        format(use_gpu, use_xpu, use_npu, use_visual, infer_batch_size,
+               test_data_dir, start_epoch, end_epoch, print_interval,
+               model_load_path))
     logger.info("**************common.configs**********")
 
     if use_xpu:
         xpu_device = 'xpu:{0}'.format(os.getenv('FLAGS_selected_xpus', 0))
         place = paddle.set_device(xpu_device)
+    elif use_npu:
+        npu_device = 'npu:{0}'.format(os.getenv('FLAGS_selected_npus', 0))
+        place = paddle.set_device(npu_device)
     else:
         place = paddle.set_device('gpu' if use_gpu else 'cpu')
 

tools/trainer.py

@@ -65,6 +65,7 @@ def main(args):
 
     # tools.vars
     use_gpu = config.get("runner.use_gpu", True)
+    use_npu = config.get("runner.use_npu", False)
     use_xpu = config.get("runner.use_xpu", False)
     use_visual = config.get("runner.use_visual", False)
     train_data_dir = config.get("runner.train_data_dir", None)
@@ -77,14 +78,17 @@ def main(args):
 
     logger.info("**************common.configs**********")
     logger.info(
-        "use_gpu: {}, use_xpu: {}, use_visual: {}, train_batch_size: {}, train_data_dir: {}, epochs: {}, print_interval: {}, model_save_path: {}".
-        format(use_gpu, use_xpu, use_visual, train_batch_size, train_data_dir,
-               epochs, print_interval, model_save_path))
+        "use_gpu: {}, use_xpu: {}, use_npu: {}, use_visual: {}, train_batch_size: {}, train_data_dir: {}, epochs: {}, print_interval: {}, model_save_path: {}".
+        format(use_gpu, use_xpu, use_npu, use_visual, train_batch_size,
+               train_data_dir, epochs, print_interval, model_save_path))
     logger.info("**************common.configs**********")
 
     if use_xpu:
         xpu_device = 'xpu:{0}'.format(os.getenv('FLAGS_selected_xpus', 0))
         place = paddle.set_device(xpu_device)
+    elif use_npu:
+        npu_device = 'npu:{0}'.format(os.getenv('FLAGS_selected_npus', 0))
+        place = paddle.set_device(npu_device)
     else:
         place = paddle.set_device('gpu' if use_gpu else 'cpu')