[Fixbug] Fix a bug in trace_from when the inputs are directly use…

…d as outputs (#76)

* fix a bug in trace_from when the inputs are directly used as outputs

* .

python/hidet/graph/ir/flow_graph.py

@@ -331,20 +331,24 @@ def load(model_file: str) -> FlowGraph:
         return ret
 
     def update_nodes(self):
-        inputs, self.nodes, self.usage_count = self._analyze(self.outputs)
+        free_vars, self.nodes, self.usage_count = self._analyze(self.outputs)
         if self.inputs:
-            if len(inputs) != len(self.inputs):
-                raise ValueError('Found {} symbol inputs, but {} given'.format(len(inputs), len(self.inputs)))
-            if any(all(a is not v for v in self.inputs) for a in inputs):
-                raise ValueError('There is a symbol tensor not given in inputs')
+            non_bound_free_vars: Set[Tensor] = set(free_vars) - set(self.inputs)
+            if len(non_bound_free_vars) > 0:
+                msg = ['There is free variable(s) not given in inputs:']
+                for v in non_bound_free_vars:
+                    msg.append('  {}'.format(v.signature()))
+                raise ValueError('\n'.join(msg))
         else:
-            if len(inputs) > 1:
+            if len(free_vars) > 1:
                 raise ValueError(
-                    f'The traced graph has found {len(inputs)} symbol inputs. When there are multiple symbol inputs, \n'
-                    'It is mandatory to specify the "inputs" argument explicitly when calling hidet.trace_from(...):\n'
-                    '    hidet.trace_from(..., inputs=[tensor1, tensor2, ...])\n'
+                    f'The traced graph has found {len(free_vars)} free varaibles. '
+                    f'When there are multiple free '
+                    f'variables, it is mandatory to specify the "inputs" argument explicitly when calling '
+                    f'hidet.trace_from(...):\n'
+                    '    hidet.trace_from(..., free_vars=[tensor1, tensor2, ...])\n'
                 )
-            self.inputs = inputs
+            self.inputs = free_vars
         return self
 
     def cuda_graph(self):
@@ -409,7 +413,27 @@ def latency(
 
     @staticmethod
     def _analyze(outputs: List[Tensor]) -> Tuple[List[Tensor], List[Operator], Dict[Tensor, int]]:
-        inputs = []
+        """
+        Analyze the implicit flow graph by backwards traversing the graph from given outputs.
+
+        Parameters
+        ----------
+        outputs: List[Tensor]
+            The outputs of the flow graph to traversing from.
+
+        Returns
+        -------
+        free_vars, nodes, usage_count: Tuple[List[Tensor], List[Operator], Dict[Tensor, int]]
+            The free variables, nodes and usage count of the flow graph.
+
+            The free variables are the free symbolic tensors that are not produced by any operators and do not contain
+            the non-None storage attribute.
+
+            The nodes are the operators that are used to produce the outputs, in topological order.
+
+            The usage count contains the number of times each tensor is used.
+        """
+        free_vars = []
         nodes: List[Operator] = []
         # find out all nodes
         all_nodes: Set[Operator] = set()
@@ -449,9 +473,9 @@ def find_all_nodes(u: Operator):
             nodes.append(op)
             for it in op.inputs:
                 if it.op is None:
-                    if it.storage is None and all(it is not v for v in inputs):
+                    if it.storage is None and all(it is not v for v in free_vars):
                         # input
-                        inputs.append(it)
+                        free_vars.append(it)
                 else:
                     out_degree[it.op] -= 1
                     if out_degree[it.op] == 0:
@@ -467,7 +491,7 @@ def find_all_nodes(u: Operator):
         for graph_output in outputs:
             usage_count[graph_output] += 1
 
-        return inputs, nodes, usage_count
+        return free_vars, nodes, usage_count
 
 
 def trace_from(tensor: Union[Tensor, List[Tensor]], inputs: Optional[Union[Tensor, List[Tensor]]] = None) -> FlowGraph:

python/hidet/ir/dtypes/floats.py

@@ -105,7 +105,7 @@ def finfo(self) -> FloatInfo:
     np.finfo(np.float16).min,
     np.finfo(np.float16).max,
     np.finfo(np.float16).eps,
-    np.finfo(np.float16).smallest_normal,
+    np.finfo(np.float16).tiny,
 )
 float32 = FloatType(
     'float32',
@@ -114,7 +114,7 @@ def finfo(self) -> FloatInfo:
     np.finfo(np.float32).min,
     np.finfo(np.float32).max,
     np.finfo(np.float32).eps,
-    np.finfo(np.float32).smallest_normal,
+    np.finfo(np.float32).tiny,
 )
 float64 = FloatType(
     'float64',
@@ -123,7 +123,7 @@ def finfo(self) -> FloatInfo:
     np.finfo(np.float64).min,
     np.finfo(np.float64).max,
     np.finfo(np.float64).eps,
-    np.finfo(np.float64).smallest_normal,
+    np.finfo(np.float64).tiny,
 )
 bfloat16 = FloatType('bfloat16', 'bf16', 2, -3.4e38, 3.4e38, None, None)  # TODO: find correct values
 tfloat32 = FloatType('tfloat32', 'tf32', 4, -3.4e38, 3.4e38, None, None)