Revert "fix(mge): replace _full_sync by sync"

This reverts commit e36ef45464bf8292d51aae4f92d49dd0915364af

GitOrigin-RevId: 2d913c8ac99423858be509d33110a172558df2df

imperative/python/megengine/__init__.py

@@ -76,7 +76,7 @@ if sys.platform == "win32":
     kernel32.SetErrorMode(old_error_mode)
 
 from .core._imperative_rt.core2 import close as _close
-from .core._imperative_rt.core2 import full_sync as sync
+from .core._imperative_rt.core2 import full_sync as _full_sync
 from .core._imperative_rt.core2 import sync as _sync
 from .core._imperative_rt.utils import _set_fork_exec_path_for_timed_func
 from .config import *

imperative/python/megengine/device.py

@@ -165,7 +165,7 @@ def get_cuda_compute_capability(device: int, device_type=DeviceType.CUDA) -> int
 def get_allocated_memory(device: Optional[str] = None):
     r"""Returns the current memory occupied by tensors on the computing device in bytes.
 
-    Due to the asynchronous execution of MegEngine, please call megengine.sync
+    Due to the asynchronous execution of MegEngine, please call megengine._full_sync
     before calling this function in order to get accurate value.
     """
     if device is None:
@@ -176,7 +176,7 @@ def get_allocated_memory(device: Optional[str] = None):
 def get_reserved_memory(device: Optional[str] = None):
     r"""Returns the current memory managed by the caching allocator on the computing device in bytes.
 
-    Due to the asynchronous execution of MegEngine, please call megengine.sync
+    Due to the asynchronous execution of MegEngine, please call megengine._full_sync
     before calling this function in order to get accurate value.
     """
     if device is None:
@@ -187,7 +187,7 @@ def get_reserved_memory(device: Optional[str] = None):
 def get_max_reserved_memory(device: Optional[str] = None):
     r"""Returns the maximum memory managed by the caching allocator on the computing device in bytes.
 
-    Due to the asynchronous execution of MegEngine, please call megengine.sync
+    Due to the asynchronous execution of MegEngine, please call megengine._full_sync
     before calling this function in order to get accurate value.
     """
     if device is None:
@@ -198,7 +198,7 @@ def get_max_reserved_memory(device: Optional[str] = None):
 def get_max_allocated_memory(device: Optional[str] = None):
     r"""Returns the maximum memory occupied by tensors on the computing device in bytes.
 
-    Due to the asynchronous execution of MegEngine, please call megengine.sync
+    Due to the asynchronous execution of MegEngine, please call megengine._full_sync
     before calling this function in order to get accurate value.
     """
     if device is None:
@@ -209,7 +209,7 @@ def get_max_allocated_memory(device: Optional[str] = None):
 def reset_max_memory_stats(device: Optional[str] = None):
     r"""Resets the maximum stats on the computing device.
 
-    Due to the asynchronous execution of MegEngine, please call megengine.sync
+    Due to the asynchronous execution of MegEngine, please call megengine._full_sync
     before calling this function in order to properly reset memory stats.
     """
     if device is None:
@@ -255,7 +255,7 @@ def coalesce_free_memory():
     small pieces may not be returned.
     
     because of the async processing of megengine, the effect of this func may not be reflected 
-    immediately. if you want to see the effect immediately, you can call megengine.sync after
+    immediately. if you want to see the effect immediately, you can call megengine._full_sync after
     this func was called
 
     .. note::

imperative/python/megengine/tools/benchmark_op.py

@@ -400,14 +400,14 @@ test_cases = [
 
 def perf_func(func, inps, reps, unpack_inps, is_mge):
     if is_mge:
-        mge.sync()
+        mge._full_sync()
         tik = time.time()
         for _ in range(reps):
             if unpack_inps:
                 out = func(*inps)
             else:
                 out = func(inps)
-        mge.sync()
+        mge._full_sync()
     else:
         torch.cuda.synchronize()
         with torch.no_grad():