fix some paras error and rename device name.

dicp/dicp/vendor/TopsGraph/codegen/enflame.py

@@ -256,33 +256,21 @@ def gen_run_func_code(self):
         func_body = IndentedBuffer()
         func_body.writeline('std::vector<void *> input_ptrs;')
         for i in range(0, len(self.input_args)):
-            func_body.writeline(
-                f'input_ptrs.emplace_back(static_cast<void *>(input_ptr{str(i)}));')
+            func_body.writeline(f'input_ptrs.emplace_back(inputs_ptr[{str(i)}]);')
 
         func_body.writeline("")
         func_body.writeline("std::vector<void *> output_ptrs;")
+        output_ptr_count = 0
         for i in range(0, len(self.output_args)):
             if not isinstance(self.output_args[i], type(None)):
-                func_body.writeline(
-                    f'output_ptrs.emplace_back(output_ptr{str(i)});')
-
+                func_body.writeline(f'output_ptrs.emplace_back(outputs_ptr[{output_ptr_count}]);')
+                output_ptr_count += 1
         func_body.writeline("")
-        func_body.writeline(
-            f'run(exe_ptr, dipu_stream, input_ptrs, output_ptrs, {self.device_id}, {"true" if dipu_flag else "false"});')
-
-        input_paras = []
-        for i in range(0, len(self.input_args)):
-            input_paras.append(f"float* input_ptr{str(i)}")
 
-        output_paras = []
-        for i in range(0, len(self.output_args)):
-            if not isinstance(self.output_args[i], type(None)):
-                output_paras.append(f"float* output_ptr{str(i)}")
-        paras = input_paras + output_paras
+        func_body.writeline(f'run(exe_ptr, dipu_stream, input_ptrs, output_ptrs, {self.device_id}, {"true" if dipu_flag else "false"});')
 
         run_func_code = IndentedBuffer()
-        run_func_code.writeline(
-            f'extern "C" void run(void *dipu_stream, {", ".join(paras)}) {"{"}')
+        run_func_code.writeline(f'extern "C" void run(void *dipu_stream, void **inputs_ptr, void **outputs_ptr) {"{"}')
         with run_func_code.indent():
             run_func_code.splice(func_body)
         run_func_code.splice('}')
@@ -395,7 +383,7 @@ def check_res(a, b, graph_name):
 
     def gen_tensor(self, prefix, tensor):
         if dipu_flag:
-            res = f"{prefix}({tuple(tensor.shape)}, {tensor.stride()}, device='xpu:{self.device_id}', dtype={tensor.dtype})"
+            res = f"{prefix}({tuple(tensor.shape)}, {tensor.stride()}, device='dipu:{self.device_id}', dtype={tensor.dtype})"
         else:
             res = f"{prefix}({tuple(tensor.shape)}, {tensor.stride()}, device='{tensor.device.type}', dtype={tensor.dtype})"
         # makes a copy of the tensor for view ops
@@ -913,7 +901,7 @@ def Concatenate(op_var, out_shape, out_dtype, tensors, dim):
         return f"builder::Op {op_var} = builder::Concatenate({'{' + ', '.join(tensors) + '}'}, {dim});"
 
     @staticmethod
-    def Softmax(op_var, out_shape, out_dtype, x, y, z):
+    def Softmax(op_var, out_shape, out_dtype, x, y):
         return f"builder::Op {op_var} = builder::Softmax({x}, {y}, true);"
 
     @staticmethod

dicp/dicp/vendor/TopsGraph/conversion.py

@@ -210,14 +210,10 @@ def HardswishBackward(self, *args, **kwargs):
     def Clone(self, *args, **kwargs):
         return self.get_proxy(tops_op.Clone, args, kwargs)
 
-    @register_conversion(aten.copy.default)
+    @register_conversion([aten.copy.default, aten.copy_.default])
     def Copy(self, *args, **kwargs):
         return self.get_proxy(tops_op.Copy, args, kwargs)
 
-    @register_conversion(aten.copy_.default)
-    def Copy_(self, *args, **kwargs):
-        return self.get_proxy(tops_op.Copy, args, kwargs)
-
     @register_conversion(aten.lift_fresh_copy.default)
     def LiftFreshCopy(self, *args, **kwargs):
         return self.get_proxy(tops_op.LiftFreshCopy, args, kwargs)
@@ -247,13 +243,9 @@ def Less(self, *args, **kwargs):
     def LessEqual(self, *args, **kwargs):
         return self.get_proxy(tops_op.LessEqual, args, kwargs)
 
-    @register_conversion(aten.eq.Tensor)
-    def Equal_Tensor(self, *args, **kwargs):
-        return self.get_proxy(tops_op.Equal_Tensor, args, kwargs)
-
-    @register_conversion(aten.eq.Scalar)
-    def Equal_Scalar(self, *args, **kwargs):
-        return self.get_proxy(tops_op.Equal_Scalar, args, kwargs)
+    @register_conversion([aten.eq.Tensor, aten.eq.Scalar])
+    def Equal(self, *args, **kwargs):
+        return self.get_proxy(tops_op.Equal, args, kwargs)
 
     @register_conversion(aten.ne.Scalar)
     def NotEqual(self, a, b):
@@ -332,7 +324,7 @@ def BatchNormBackward(*args, **kwargs):
     def Softmax(self, a, dim, half_to_float):
         out_shape = fx_traceback.get_current_meta()["val"].shape
         dim = dim + len(out_shape) if dim < 0 else dim
-        return self.get_proxy(tops_op.Softmax, (a, dim, half_to_float))
+        return self.get_proxy(tops_op.Softmax, (a, dim))
 
     @register_conversion(aten.mm)
     def Gemm(self, *args, **kwargs):

dicp/dicp/vendor/TopsGraph/tops_op.py

@@ -78,22 +78,6 @@ def __init__(self, *args, **kwargs):
     def __call__(self, *args, **kwargs):
         new_args = args[1:]
         return super().__call__(*new_args, **kwargs)
-
-
-class Equal_Tensor(Operator):
-    def __init__(self, *args, **kwargs):
-        super().__init__("Equal")
-        self.args = args
-        self.kwargs = kwargs
-        self.torch_op = aten.eq.Tensor
-
-
-class Equal_Scalar(Operator):
-    def __init__(self, *args, **kwargs):
-        super().__init__("Equal")
-        self.args = args
-        self.kwargs = kwargs
-        self.torch_op = aten.eq.Scalar
 
 
 class Mul(Operator):
@@ -257,15 +241,7 @@ def __init__(self, *args, **kwargs):
         super().__init__("Copy")
         self.args = args
         self.kwargs = kwargs
-        self.torch_op = torch.ops.aten.copy.default
-
-
-class Copy_(Operator):
-    def __init__(self, *args, **kwargs):
-        super().__init__("Copy")
-        self.args = args
-        self.kwargs = kwargs
-        self.torch_op = torch.ops.aten.copy_.default
+        self.torch_op = torch.ops.aten.copy
 
 
 class LiftFreshCopy(Operator):
@@ -515,6 +491,14 @@ def __init__(self, *args, **kwargs):
         self.torch_op = aten.new_empty_strided.default
 
 
+class Equal(Operator):
+    def __init__(self, *args, **kwargs):
+        super().__init__("Equal")
+        self.args = args
+        self.kwargs = kwargs
+        self.torch_op = aten.eq
+
+
 class Expand(Operator):
     def __init__(self, *args, **kwargs):
         super().__init__("Expand")