[Ascend] fix sub bug for ascend

dipu/scripts/autogen_diopi_wrapper/diopi_functions.yaml

@@ -76,12 +76,31 @@
   custom_code_at_the_beginning: |
     return dipu_add__tensor(self, other, -alpha);
 
+- schema: sub.Scalar(Tensor self, Scalar other, Scalar alpha=1) -> Tensor
+  custom_code_at_the_beginning: |
+    at::Tensor out = UnaryOpInferrer().infer_out(self);
+  interface: diopiSubScalar(ctx, out, self, other, alpha)
+
 - schema: "sub.Tensor(Tensor self, Tensor other, *, Scalar alpha=1) -> Tensor"
-  dummy_call_diopi: True
+  ins: [selfTmp]
   custom_code_at_the_beginning: |
     at::native::sub_check(self, other);
-    auto out = BinaryOpInferrer().infer_out(self, other);
-    return dipu_add_out(self, other, -alpha, out);
+
+    if (is_scalar_on_cpu(other)) {
+      at::native::alpha_check(self.scalar_type(), alpha);
+      return dipu_sub_scalar(self, other.item(), alpha);
+    }
+
+    at::Tensor selfTmp = self;
+    if (is_scalar_on_cpu(selfTmp)) {
+      selfTmp = nodispatch::empty({1}, self.options().device(other.device()));
+      dipu_fill__scalar(selfTmp, self.item());
+    }
+
+    at::native::alpha_check(selfTmp.scalar_type(), alpha);
+    at::Tensor out = BinaryOpInferrer().infer_out(selfTmp, other);
+
+  interface: diopiSub(ctx, out, selfTmp, other, alpha)
 
 - schema: "div.Scalar(Tensor self, Scalar other) -> Tensor"
   custom_code_at_the_beginning: |
@@ -1341,9 +1360,10 @@
 
 - schema: rsub.Tensor(Tensor self, Tensor other, *, Scalar alpha=1) -> Tensor
   custom_code_at_the_beginning: |
+    if (is_scalar_on_cpu(self)) {
+      return dipu_sub_scalar(other, self.item(), alpha);
+    }
     auto out = nodispatch::empty_like(self);
-    // NOLINTNEXTLINE(readability-suspicious-call-argument)
-    return dipu_sub_out(other, self, alpha, out);
   interface: diopiSub(ctx, out, other, self, alpha)
 
 - schema: "unique_dim(Tensor self, int dim, bool sorted=True, bool return_inverse=False, bool return_counts=False) -> (Tensor out, Tensor indices, Tensor counts)"

dipu/tests/python/unittests/test_rsub.py

@@ -23,8 +23,15 @@ def test_rsub(self):
         self._test_rsub(torch.ones(4, 5) * 1.1, torch.ones(4, 5) * 5, alpha=4)
 
     def test_rsub_scalar(self):
-        self._test_rsub_scalar(torch.ones(4, 5), 10)
-        self._test_rsub_scalar(torch.ones(4, 5), 10, 2.5)
+        # from torch:
+        # For integral input tensors, argument alpha must not be a floating point number
+        # Boolean alpha only supported for Boolean results
+        self._test_rsub_scalar(torch.ones(4, 5), 10, alpha=1)
+        self.assertRaisesRegex(
+            RuntimeError,
+            r"For integral input tensors, argument alpha must not be a floating point number\.",
+            lambda: self._test_rsub_scalar(torch.ones(4, 5), 10, 2.5),
+        )
 
 
 if __name__ == "__main__":