DAG version of subset of D39417
Diff Detail
Diff Detail
Event Timeline
Comment Actions
Test is hard for this short of testing debug info, which is fragile. This prevents some regressions in a future AMDGPU commit
Comment Actions
If it does break a subsequent commit, could you commit that first, and then apply this to fix the regressions introduced by the other commit?
Comment Actions
This would be the 3rd time I've had to touch all of the tests impacted by the, and each time it's been a pain. That patch adds folding an addressing mode that doesn't support overflows, so directly checks NUW. Once that's in it should be more straightforward to write the tests with these situations directly feeding into the address expression.