[Fixbug] Fix a bug in IRModule.update_function

gallery/getting-started/quick-start.py

@@ -40,21 +40,15 @@
 )
 model = model.cuda().eval()
 
-# we should register the hidet backend for pytorch dynamo
-# only need to do this if you import hidet before torch. Otherwise, it is done automatically
-hidet.torch.register_dynamo_backends()
+# optimize the model with 'hidet' backend
+model_opt = torch.compile(model, backend='hidet')
 
-# currently, hidet only support inference
-with torch.no_grad():
-    # optimize the model with 'hidet' backend
-    model_opt = torch.compile(model, backend='hidet')
+# run the optimized model
+y1 = model_opt(x)
+y2 = model(x)
 
-    # run the optimized model
-    y1 = model_opt(x)
-    y2 = model(x)
-
-    # check the correctness
-    torch.testing.assert_close(actual=y1, expected=y2, rtol=1e-2, atol=1e-2)
+# check the correctness
+torch.testing.assert_close(actual=y1, expected=y2, rtol=1e-2, atol=1e-2)
 
 
 # benchmark the performance

gallery/tutorials/optimize-pytorch-model.py

@@ -68,8 +68,6 @@
 import torch.backends.cudnn
 import hidet
 
-hidet.torch.register_dynamo_backends()  # register hidet backend to torch dynamo
-
 x = torch.randn(1, 3, 224, 224).cuda()
 model = torch.hub.load(
     'pytorch/vision:v0.9.0', 'resnet18', pretrained=True, verbose=False

python/hidet/cli/bench/model.py

@@ -87,7 +87,6 @@ def bench_with_backend(self, backend: str, mode=None, warmup=3, number=10, repea
             raise RuntimeError('Torch Dynamo is not available, please install pytorch 2.0 or higher.')
         import torch._dynamo as dynamo
 
-        hidet.torch.register_dynamo_backends()
         torch.backends.cudnn.allow_tf32 = self.allow_tf32
         torch.backends.cuda.matmul.allow_tf32 = self.allow_tf32
 

python/hidet/ir/func.py

@@ -146,9 +146,15 @@ def lookup_var(self, name):
         return self.global_vars[name]
 
     def update_function(self, func: Function):
+        from hidet.ir.tools import rewrite
+
         self.functions[func.name] = func
         if func.name in self.global_vars:
-            self.global_vars[func.name].type = func.name, FuncType.from_func(func)
+            old_var = self.global_vars[func.name]
+            new_var = Var(func.name, FuncType.from_func(func))
+            self.global_vars[func.name] = new_var
+            for name, f in self.functions.items():
+                self.functions[name] = rewrite(f, {old_var: new_var})
 
     def add(self, name, func: Function):
         if name in self.functions:

python/hidet/ir/tools/util_functors.py

@@ -19,15 +19,7 @@
 class MapBasedRewriter(IRRewriter):
     def __init__(self, rmap):
         super().__init__()
-        self.rmap = rmap
-
-    def visit(self, node):
-        if node not in self.memo:
-            if node in self.rmap:
-                self.memo[node] = self.rmap[node]
-            else:
-                self.memo[node] = super().visit(node)
-        return self.memo[node]
+        self.memo.update(rmap)
 
 
 class IRCollector(IRVisitor):
@@ -102,7 +94,7 @@ def visit_Let(self, e: Let):
         return Let(v, self(e.value), self(e.body))
 
 
-def rewrite(node: Union[Expr, Stmt, tuple], rewrite_map: Mapping[Union[Stmt, Expr], Union[Stmt, Expr]]):
+def rewrite(node: Union[Function, Expr, Stmt, tuple], rewrite_map: Mapping[Union[Stmt, Expr], Union[Stmt, Expr]]):
     assert isinstance(rewrite_map, dict)
     rewriter = MapBasedRewriter(rewrite_map)
     return rewriter.rewrite(node)