[TableGen] Fix computeRegUnitLaneMasks for ad hoc aliasing #79462
Conversation
With ad hoc aliasing (using the Aliases field in register definitions) we can have subregisters with disjoint lane masks that nevertheless share a regunit. In this case we need to take the union of their lane masks, not the intersection. This was inadvertently broken by https://reviews.llvm.org/D157864
|
@kaz7 can you write a test for this? |
|
Thank you! I've noticed this PR just now. I'll try to create test cases. I can create test cases for VE like checking live-in problems described in #78942 . However, such tests work only on VE environment. So, I think it's better to have different tests something like to check the output of TableGen directly. I'll try to create such one. |
|
@jayfoad , I made a test in main...kaz7:llvm-project:computeregunitlanemasks-test. I tried to push it to your fork, but it won't work. Can you cherry-pick my patch? This test work fine with a tablegen after this PR. And, it causes errors if someone try to calculate lanemask using intersection. |
Sorry about that. Recreated as #79835. |
With ad hoc aliasing (using the Aliases field in register definitions)
we can have subregisters with disjoint lane masks that nevertheless
share a regunit. In this case we need to take the union of their
lane masks, not the intersection.
This was inadvertently broken by https://reviews.llvm.org/D157864
Fixes #78942