[GCU] Support inference for GCU (#14142)

Author: EnflameGCU

tools/infer/utility.py

@@ -41,6 +41,12 @@ def init_args():
     parser.add_argument("--use_xpu", type=str2bool, default=False)
     parser.add_argument("--use_npu", type=str2bool, default=False)
     parser.add_argument("--use_mlu", type=str2bool, default=False)
+    parser.add_argument(
+        "--use_gcu",
+        type=str2bool,
+        default=False,
+        help="Use Enflame GCU(General Compute Unit)",
+    )
     parser.add_argument("--ir_optim", type=str2bool, default=True)
     parser.add_argument("--use_tensorrt", type=str2bool, default=False)
     parser.add_argument("--min_subgraph_size", type=int, default=15)
@@ -298,6 +304,34 @@ def create_predictor(args, mode, logger):
             config.enable_custom_device("mlu")
         elif args.use_xpu:
             config.enable_xpu(10 * 1024 * 1024)
+        elif args.use_gcu:  # for Enflame GCU(General Compute Unit)
+            assert paddle.device.is_compiled_with_custom_device("gcu"), (
+                "Args use_gcu cannot be set as True while your paddle "
+                "is not compiled with gcu! \nPlease try: \n"
+                "\t1. Install paddle-custom-gcu to run model on GCU. \n"
+                "\t2. Set use_gcu as False in args to run model on CPU."
+            )
+            import paddle_custom_device.gcu.passes as gcu_passes
+
+            gcu_passes.setUp()
+            if args.precision == "fp16":
+                config.enable_custom_device(
+                    "gcu", 0, paddle.inference.PrecisionType.Half
+                )
+                gcu_passes.set_exp_enable_mixed_precision_ops(config)
+            else:
+                config.enable_custom_device("gcu")
+
+            if paddle.framework.use_pir_api():
+                config.enable_new_ir(True)
+                config.enable_new_executor(True)
+                kPirGcuPasses = gcu_passes.inference_passes(
+                    use_pir=True, name="PaddleOCR"
+                )
+                config.enable_custom_passes(kPirGcuPasses, True)
+            else:
+                pass_builder = config.pass_builder()
+                gcu_passes.append_passes_for_legacy_ir(pass_builder, "PaddleOCR")
         else:
             config.disable_gpu()
             if args.enable_mkldnn:
@@ -314,7 +348,8 @@ def create_predictor(args, mode, logger):
         # enable memory optim
         config.enable_memory_optim()
         config.disable_glog_info()
-        config.delete_pass("conv_transpose_eltwiseadd_bn_fuse_pass")
+        if not args.use_gcu:  # for Enflame GCU(General Compute Unit)
+            config.delete_pass("conv_transpose_eltwiseadd_bn_fuse_pass")
         config.delete_pass("matmul_transpose_reshape_fuse_pass")
         if mode == "rec" and args.rec_algorithm == "SRN":
             config.delete_pass("gpu_cpu_map_matmul_v2_to_matmul_pass")

tools/program.py

@@ -115,7 +115,7 @@ def merge_config(config, opts):
     return config
 
 
-def check_device(use_gpu, use_xpu=False, use_npu=False, use_mlu=False):
+def check_device(use_gpu, use_xpu=False, use_npu=False, use_mlu=False, use_gcu=False):
     """
     Log error and exit when set use_gpu=true in paddlepaddle
     cpu version.
@@ -154,6 +154,9 @@ def check_device(use_gpu, use_xpu=False, use_npu=False, use_mlu=False):
         if use_mlu and not paddle.device.is_compiled_with_mlu():
             print(err.format("use_mlu", "mlu", "mlu", "use_mlu"))
             sys.exit(1)
+        if use_gcu and not paddle.device.is_compiled_with_custom_device("gcu"):
+            print(err.format("use_gcu", "gcu", "gcu", "use_gcu"))
+            sys.exit(1)
     except Exception as e:
         pass
 
@@ -799,6 +802,7 @@ def preprocess(is_train=False):
     use_xpu = config["Global"].get("use_xpu", False)
     use_npu = config["Global"].get("use_npu", False)
     use_mlu = config["Global"].get("use_mlu", False)
+    use_gcu = config["Global"].get("use_gcu", False)
 
     alg = config["Architecture"]["algorithm"]
     assert alg in [
@@ -853,9 +857,11 @@ def preprocess(is_train=False):
         device = "npu:{0}".format(os.getenv("FLAGS_selected_npus", 0))
     elif use_mlu:
         device = "mlu:{0}".format(os.getenv("FLAGS_selected_mlus", 0))
+    elif use_gcu:  # Use Enflame GCU(General Compute Unit)
+        device = "gcu:{0}".format(os.getenv("FLAGS_selected_gcus", 0))
     else:
         device = "gpu:{}".format(dist.ParallelEnv().dev_id) if use_gpu else "cpu"
-    check_device(use_gpu, use_xpu, use_npu, use_mlu)
+    check_device(use_gpu, use_xpu, use_npu, use_mlu, use_gcu)
 
     device = paddle.set_device(device)