ElaMay 15
You might not like it, but this is what peak compiler performance looks like.
Show threadmirabilosMay 15
@ela this stumped me until I saw clang decides to return bool in al, not eax, for some reason.
Show threadwaldiMay 16
@mirabilos @ela does "and al,1" clear the upper bits?
Show threadmirabilos
@waldi @ela only seven of them
May 16 at 6:16amWeb
Show threadElaMay 16
@mirabilos @waldi To be unfathomably pedantic, that would be all of them for al.
Show threadmirabilosMay 16
@ela @waldi yes, but it using only al ipv eax for the return value strikes me as interesting ABI choice.
Show threadwaldiMay 16
@mirabilos @ela Ah, because _Bool is defined as 8bit value where only bit 0 matters, bits 1 to 7 are set to 0 and 8 to 63 are undefined.
Show threadmirabilosMay 16
@waldi @ela except for actual storage (where one bit would be better in structs for example), blowing it up to the full 32 bits would reduce the number of instructions executed overall in typical programs (and incidentally even the code size)
Show threadElaMay 16
@mirabilos @waldi Well, it's what the spec says.
Show threadmirabilosMay 16
@ela @waldi I admit only vague familiarity with the psABI for amd64 (normally preferring i386), and specifically not that they chose a (surprising, to me) representation for booleans