[SYCL] Change sub_group_mask copy-assignment operator to default

[steffenlarsen]

As part of version two of sycl_ext_oneapi_sub_group_mask the copy-asssignment operator is defined as default but were deleted as part of #10564. This changes them to default ahead of version 2 of the extension to avoid regressions.

[masterleinad]

Ah, I had #10885 already opened.

As part of version two of sycl_ext_oneapi_sub_group_mask the copy-asssignment operator is defined as default but were deleted as part of #10564. This changes them to default ahead of version 2 of the extension to avoid regressions.

Either would work since version 1 didn't have any user-provided constructor or assignments operators but they were implicitly generated (and not deleted). The problem really is that the current implementation neither conforms to version 1 nor 2, probably because the existence of the private

sub_group_mask(BitsType rhs, size_t bn)

avoids all other constructors being implicitly defined.

[steffenlarsen]

Ah, I had #10885 already opened.

As part of version two of sycl_ext_oneapi_sub_group_mask the copy-asssignment operator is defined as default but were deleted as part of #10564. This changes them to default ahead of version 2 of the extension to avoid regressions.

Either would work since version 1 didn't have any user-provided constructor or assignments operators but they were implicitly generated (and not deleted). The problem really is that the current implementation neither conforms to version 1 nor 2, probably because the existence of the private

sub_group_mask(BitsType rhs, size_t bn)

avoids all other constructors being implicitly defined.

I was made aware of your patch right after I opened this! Personally, I would prefer default as long as it is possible for the class. I have opened an issue regarding the problems caused by the private ctor. For now I think it is best for this PR to just revert the regression.

[maarquitos14]

LGTM

[steffenlarsen]

CI failure caused by device issue similar to #10460.