https://bugs.kde.org/show_bug.cgi?id=484426
--- Comment #3 from Paul Floyd <pjfl...@wanadoo.fr> --- I think that I understand the problem. arm64 has two overlapping ways of doing floating point rounding. The mnemonics are FRINT (Float Round to Int) and then one more letter for the kind of rounding. The FPCR (Floating Point Control Register) has 2 bits for the rounding mode (00 nearest 01 towards +inf 10 towards -inf 11 towards zero) I - use the current rounding mode P - like rounding mode 01 M - like rounding mode 10 Z - like rounding mode 11 but then there is N - to nearest, ties to even A - to nearest, ties away from zero X - exact, like I but causes inexect FPE or sets IneXact Computing bit in FPSR Out problem is that we're trying to use FRINTI for everything in the generated code. That's not too bad for FRINTX (we don't support FPEs). But it's wrong for N and A. -- You are receiving this mail because: You are watching all bug changes.
