refactor(mge/grad_manager): refactor gradmanager, add allreduce callback

GitOrigin-RevId: 086e2871e8

imperative/python/megengine/autodiff/grad_manager.py

@@ -0,0 +1,52 @@
+from contextlib import contextmanager
+
+from ..core.autodiff.grad import Grad
+from ..tensor import tensor
+
+
+class GradManager:
+    def __init__(self):
+        self._call_back_pair = []
+        self._recording = False
+        self._grad = None
+
+    def register(self, params, callback=None):
+        self._call_back_pair.append([params, callback])
+
+    def backward(self, ys, dys=None):
+        if not self._recording:
+            raise RuntimeError(
+                "no computation history. "
+                "did you forget record() or "
+                "call a method that clears the history?"
+            )
+        assert self._grad is not None
+        if not isinstance(ys, (tuple, list)):
+            ys = [ys]
+        if dys is None:
+            dys = [tensor(1).broadcast(y.shape) for y in ys]
+        if not isinstance(dys, (tuple, list)):
+            dys = [dys]
+        try:
+            self._grad(ys, dys)
+        finally:
+            self._grad = None
+
+    def record(self):
+        @contextmanager
+        def recorder():
+            grad = Grad()
+            if self._recording:
+                raise RuntimeError("already recording!")
+            try:
+                self._recording = True
+                self._grad = grad
+                for params, callbacks in self._call_back_pair:
+                    grad.wrt(*params, callback=callbacks)
+                with grad:
+                    yield
+            finally:
+                self._recording = False
+                self._grad = None
+
+        return recorder()

imperative/python/megengine/core/autodiff/grad.py

@@ -260,9 +260,13 @@ class Grad:
                     cache[v] = g
                 if last_written_to[v] == (seqno, i):
                     if v.callback:
-                        v.callback(
+                        grad = v.callback(
                             v.owner(), Wrapper(cache[v]) if Wrapper else cache[v]
                         )
+                    if getattr(v.owner(), "grad", None) is None:
+                        v.owner().grad = grad
+                    else:
+                        v.owner().grad += grad
                     if v.opnode is None:
                         # won't read by backward, mark consumed
                         cache[v] = None

imperative/python/megengine/distributed/__init__.py

@@ -19,7 +19,7 @@ from .group import (
     is_distributed,
     new_group,
 )
-from .helper import synchronized
+from .helper import bcast_params_, make_allreduce_cb, synchronized
 from .launcher import launcher
 from .server import Client, Server
 from .util import get_free_ports

imperative/python/megengine/distributed/helper.py

@@ -12,7 +12,8 @@ from typing import Callable
 
 from megengine.device import get_device_count
 
-from .group import group_barrier, is_distributed
+from .functional import all_reduce_sum, broadcast
+from .group import WORLD, group_barrier, is_distributed
 
 
 def synchronized(func: Callable):
@@ -42,3 +43,23 @@ def get_device_count_by_fork(device_type: str):
     p.start()
     p.join()
     return q.get()
+
+
+def bcast_params_(params, group):
+    for p in params:
+        p._reset(broadcast(p, group))
+
+
+class AllreduceCallback:
+    def __init__(self, reduce_method, group=WORLD):
+        self._reduce_method = reduce_method
+        self._group = group
+
+    def __call__(self, param, grad):
+        ret = all_reduce_sum(grad, self._group)
+        if self._reduce_method == "MEAN":
+            ret = ret / self._group.size
+        return ret
+
+
+make_allreduce_cb = AllreduceCallback