Hi Niels, On Mon, 2025-03-17 at 09:21 +0100, Niels Möller wrote: > John Paul Adrian Glaubitz <[email protected]> writes: > > > nettle fails to build from source on ppc64 due to uncondtional use of the > > lxvb16x > > VSX instruction [1]: > > > > Illegal instruction > > FAIL: gcm > > > > This seems to have been fixed upstream [1], but the fix has not been > > released yet. > > I take it this could motivate a bugfix release, but on the debian side I > guess you may want to backport fixes.
I'm fine with either way. There is nothing that blocks a new upstream release on Debian. > > > [2] > > > https://git.lysator.liu.se/nettle/nettle/-/commit/2dffb58ae9e11a1626aa12a1ca9432338458fb44 > > I think you also need "Avoid using stxv/lxv instructions in powerpc64/p8 > files.", > https://git.lysator.liu.se/nettle/nettle/-/commit/d891462ba3f6e39e3696b657dcb69e9869ad9762 > > BTW: You also ask in a different email that nettle shouldn't assume > power8 (simplest way would be to disable all powerpc assembly for > earlier processors), would you be able to propose or test configure > changes for that? (https://git.lysator.liu.se/nettle/nettle/-/issues/7 > suggested testing for __VSX__, but I suspect that's slightly different from > your use case?). Testing for __VSX__ is already a very good idea. While at it, I would also test for the presence of AltiVec as not all PowerPC targets automatically enable it (especially embedded targets): glaubitz@perotto:~$ echo | gcc -maltivec -E -dM - |grep ALTIVEC #define __ALTIVEC__ 1 #define __APPLE_ALTIVEC__ 1 glaubitz@perotto:~$ echo | gcc -mvsx -E -dM - |grep VSX #define __VSX__ 1 glaubitz@perotto:~$ I wasn't aware that lxvb16x was a VSX instruction when I wrote my first mail. Thanks, Adrian -- .''`. John Paul Adrian Glaubitz : :' : Debian Developer `. `' Physicist `- GPG: 62FF 8A75 84E0 2956 9546 0006 7426 3B37 F5B5 F913
