Created attachment 413678 [details] emerge --info On my hardened ~amd64 machine with gcc-5.2.0 I see the following failure for cinelerra. ibtool: compile: x86_64-pc-linux-gnu-gcc -DHAVE_CONFIG_H -I. -I.. -I../libmpeg3 -D__STDC_CONSTANT_MACROS -D_LARGEFILE_SOURCE -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 -DHAVE_MMX -DUSE_MMX -DX86_CPU -I/usr/include/mjpegtools -I/usr/include/mjpegtools/mpeg2enc -I/usr/include/mjpegtools/mplex -DENCORE_INCLUDE=\"encore50/encore.h\" -I/usr/include/libavcodec -I/usr/include/libswscale -DHAVE_SWSCALER -O2 -pipe -march=native -ggdb -floop-interchange -floop-strip-mine -floop-block -finline-functions -Wformat -Werror=format-security -c libdv.c -fPIC -DPIC -o .libs/libdv.o In file included from libdv.c:13:0: mmx.h: In function ‘mm_support’: mmx.h:118:2: error: ‘asm’ operand has impossible constraints __asm__ __volatile__ ( ^
Created attachment 413680 [details] build log
it'd be worth to report a minimal test case upstream. clobber list is: "eax", "ebx", "ecx", "edx" which seems plenty enough to be able to solve it; but since, basically, inline asm is defined by what gcc supports, they might reply some of these registers are reserved, which means it'd have to be patched here, but before that I'd like to have a clear statement on what's wrong.
Hmmm: /* * Drop mm_support() and mmx_ok() as these do not build with clang and * are unused by Cinelerra CV. Inspired by change found in * <URL: http://hg.libsdl.org/SDL/file/32f0f603a0c8/src/video/mmx.h > */ So, it might be even simpler by ifdefing that stuff out (but maybe still worth reporting to gcc upstream so that they can have a look at it)
You may be hit by -fstack-check it need one registers to work and it is default on hardened. check if it works if you pass -fstack-check=no
(In reply to Magnus Granberg from comment #4) > You may be hit by -fstack-check it need one registers to work > and it is default on hardened. > check if it works if you pass -fstack-check=no no, that's not it, I have it without hardened too, and clobbering only "eax" makes it fail the same way. anyway, this asm seems very wrong, and I'm not even sure "eax" is valid on amd64.
commit cc1aa656e022d809b2e9d2ef64c2c4360c75a9d3 Author: Alexis Ballier <aballier@gentoo.org> Date: Sat Oct 10 11:04:52 2015 +0200 media-video/cinelerra: Fix build with gcc 5.2 by disabling old useless code. Bug #562208. this code was definitely buggy and useless, so I went the simple way :)
