Complex dispatch should be disabled on min/max functions

[ngimel]

As of #36377, min/max functions were disabled for complex inputs (via dtype checks), however, min/max kernels are still compiled and dispatched for complex, see e.g.

pytorch/aten/src/ATen/native/cpu/ReduceAllOpsKernel.cpp

Line 77 in e44b2b7

AT_DISPATCH_ALL_TYPES_AND_COMPLEX(input.scalar_type(), "min_all", [&] {

This dispatch should be disabled, and we should rely on errors produced by dispatch macro to not run those ops on complex, instead of doing redundant dtype checks.

cc @ezyang @anjali411 @dylanbespalko @mruberry
cc @t-vi, thanks for reporting!

[rishabhvarshney14]

Can I work on it @anjali411?

[anjali411]

@rishabhvarshney14 yes absolutely! please feel free to add me as a reviewer for your PR

[ngimel]

Reopening. The linked PR only disables dispatch on cpu reductions, it does not handle pointwise operations, and it does not handle cuda operations.
I also asked for redundant type checks to be removed (because dispatch would produce necessary error messages), and they are still in place.

[imaginary-person]

Reopening. The linked PR only disables dispatch on cpu reductions, it does not handle pointwise operations, and it does not handle cuda operations.
I also asked for redundant type checks to be removed (because dispatch would produce necessary error messages), and they are still in place.

Hello @ngimel, Sorry about that! l had realized that I hadn't disabled the type-checks introduced in #36377, after Richard suggested writing a test, and I recently submitted another PR (#50465). For CPU min/max ops in TensorCompareKernel.cpp, I had skipped it by mistake due to some confusion between the names of max_stub & maximum_stub. So, I changed them in this PR.

As for the CUDA operations, their kernels do not seem to be compiled & dispatched for complex types anyway, so I think that no further changes are required. Please confirm if my rationale is correct. Thanks!

Basically, the dispatch macros currently being used don't have cases (of switch statements) for complex types, so they'd default to error for complex types.
For example,

1. the reduce CUDA ops use AT_DISPATCH_ALL_TYPES_AND2 in ReduceMinMaxKernel.cu, and that macro doesn't allow complex types.

2. In MinMaxElementwiseKernel.cu, the CUDA pointwise ops use AT_DISPATCH_FLOATING_TYPES_AND2 for non-integral & non-boolean types, and this marco doesn't have a case for complex types either.

[imaginary-person]

Duplicate comment from #50465 for the convenience of any potential readers in the future. Please skip if you read the PR #50465.

There are a few cases in which the methods corresponding to min_stub() or max_stub() are not called, so dispatch macros don't get invoked, resulting in no exceptions being raised. Hence, dtype checks are necessary at 3 places:

1. pytorch/aten/src/ATen/native/TensorCompare.cpp

   Line 342 in 52dcc72

   if (_dimreduce_return_trivial_no_ident(max, self, dim, keepdim, "max")) {

2. pytorch/aten/src/ATen/native/TensorCompare.cpp

   Line 422 in 52dcc72

   if (_dimreduce_return_trivial_no_ident(min, self, dim, keepdim, "min")) {

3. pytorch/aten/src/ATen/native/TensorCompare.cpp

   Line 389 in 52dcc72

   if (_dimreduce_return_trivial_no_ident(min, self, dim, keepdim, "min") &&

The first dtype check requirement can be verified from the following example Python code:

import unittest 
import torch

class MyTestCase(unittest.TestCase): 
  
   def test_1(self):
      t = torch.tensor((1 + 1j), device='cpu', dtype=torch.complex128) 
      with self.assertRaises(RuntimeError): 
         torch.max(t, dim=0)
  
if __name__ == '__main__':  
    unittest.main()