Sanjay Patel

Right - I don't think we'd ever turn plain FP instructions into the strict intrinsics in LLVM. And if we have rounding=dynamic, then most optimization/analysis is disabled. https://reviews.llvm.org/D118928 is a...

Looks like it will reduce the same as #56711, but we can leave it open for now.

This was helped by 9d218b6, but not fixed completely. It's now virtually the same pattern as #56653, so it would be fixed by https://reviews.llvm.org/D131356.

Fixed with the previously mentioned commit and: 15e3d869119289991072679f82622072818468e6

It would be better to open new issues for different examples. At first glance, the new example here has more in common with #57152 than with the original example in...

Also note: we do not have a general-purpose boolean logic solver, so you can create more complicated (and less realistic) code sequences that will not reduce, but it is unlikely...

I don't think this is a hang. It's just taking forever to create the SDAG because those global arrays are so big. If you replace with smaller constants, it should...

computeKnownBits takes the common known bits from incoming values, but it doesn't use dominating conditions AFAICT. This might be a job for CVP?

We have three x86 codegen folds so far based on this: https://reviews.llvm.org/rGf060aa1cf3f4 https://reviews.llvm.org/rG1ce05ad619a5 https://reviews.llvm.org/rG4c41caa28710 And the example here is based on a sequence in a Bullet benchmark where GCC was...

Not sure if this example suggests a different approach, but we added it with https://reviews.llvm.org/D106684 : ``` %class.btAxis = type { %struct.btBroadphaseProxy.base, [3 x i16], [3 x i16], %struct.btBroadphaseProxy* }...