Doesn't try the all the tricks computeKnownBits does.
Details
Details
Diff Detail
Diff Detail
Event Timeline
Comment Actions
My nofpclass inference test seems to be unaffected, I'm assuming there's some attributor magic happening with the phi input values?
Comment Actions
LG, though I'd prefer if we could change the signed bit stuff such that we don't need "is first", hence use 4 states and define |= properly.
Comment Actions
It looks like this crashes clang during compilation of rc_quadratic.c from lencod from llvm-test-suite (https://llvm-compile-time-tracker.com/show_error.php?commit=cac9e427eb1ff3dabda8ac08968b998c3bc5ab47). Could you please revert?