Fix torch.histc not checking min > max on cuda for int8 tensors (#139372)

Fixes #139360

86e6513c86/aten/src/ATen/native/cuda/SummaryOps.cu (L323-L324)

Assign `min` and `max` to with low-precision input_t variable `minvalue` and `maxvalue` cause wrong comparing result in following check in here:

86e6513c86/aten/src/ATen/native/cuda/SummaryOps.cu (L353)

![image](https://github.com/user-attachments/assets/0d5c87f4-3dc6-48bb-bcc8-b1803e7cd487)

Change type of `minvalue` and `maxvalue` to fix it, similar like in line:

86e6513c86/aten/src/ATen/native/cuda/SummaryOps.cu (L280-L282)

**Test Result**
```bash
$ pytest test/test_reductions.py -vv
```
![image](https://github.com/user-attachments/assets/6b5d0d48-ebc2-4a8c-85f4-dbad147c086c)

```bash
$ lintrunner
```
![image](https://github.com/user-attachments/assets/f97c2d6d-78ea-4439-a1ba-907bc9defad7)

Pull Request resolved: https://github.com/pytorch/pytorch/pull/139372
Approved by: https://github.com/eqy

aten/src/ATen/native/cuda/SummaryOps.cu

@@ -320,8 +320,10 @@ Tensor _histc_cuda_template(
       std::nullopt /* layout */,
       DeviceType::CUDA,
       std::nullopt /* pin_memory */);
-  input_t minvalue = min;
-  input_t maxvalue = max;
+  using bounds_t = at::acc_type<input_t, /*is_cuda=*/true>;
+  bounds_t minvalue = min;
+  bounds_t maxvalue = max;
+
   if (min == max && self.numel() > 0) {
     minvalue = *self.min().cpu().const_data_ptr<input_t>();
     maxvalue = *self.max().cpu().const_data_ptr<input_t>();

test/test_reductions.py

@@ -3116,6 +3116,30 @@ class TestReductions(TestCase):
             actual)
         self.assertEqual(actual.dtype, dtype)
 
+    @dtypes(torch.uint8, torch.int8, torch.int, torch.long, torch.float, torch.double)
+    def test_histc_min_max_errors(self, device, dtype):
+        with self.assertRaisesRegex(RuntimeError, "max must be larger than min"):
+            torch.histc(torch.tensor([1., 2., 3.], dtype=dtype, device=device), bins=4, min=5, max=1)
+
+    @dtypes(torch.float, torch.double)
+    def test_histc_min_max_corner_cases(self, device, dtype):
+        actual = torch.histc(
+            torch.tensor([1., 2, 1], dtype=dtype, device=device),
+            bins=4, min=5, max=5)
+        self.assertEqual(
+            torch.tensor([2, 0, 0, 1], dtype=dtype, device=device),
+            actual)
+
+    @onlyCUDA
+    @dtypes(torch.uint8, torch.int8, torch.int, torch.long)
+    def test_histc_min_max_corner_cases_cuda(self, device, dtype):
+        actual = torch.histc(
+            torch.tensor([1., 2, 1], dtype=dtype, device=device),
+            bins=4, min=5, max=5)
+        self.assertEqual(
+            torch.tensor([2, 0, 0, 1], dtype=dtype, device=device),
+            actual)
+
     """
     Runs torch.histogram and numpy.histogram on the specified input parameters
     and asserts that their output is equal.

torch/testing/_internal/common_methods_invocations.py

@@ -19375,7 +19375,7 @@ op_db: List[OpInfo] = [
            )),
     OpInfo('histc',
            dtypes=floating_types_and(torch.bfloat16, torch.float16),
-           dtypesIfCUDA=floating_types_and(torch.int8, torch.int16, torch.int32, torch.int64),
+           dtypesIfCUDA=floating_types_and(torch.int8, torch.uint8, torch.int16, torch.int32, torch.int64),
            sample_inputs_func=sample_inputs_histc,
            supports_out=True,
            supports_autograd=False,