It's not an AVX instruction, as Wireshark runs on a Core2Duo pre Sandy Bridge when AVX arrived.
I agree that it looks too heavy for a P2, and last time I ran it under *nix it would have been the gtk version, I was surprised at all the dependencies it pulled in. This issue is only likely to get worse, is it wise to do a CPU check, and refuse to install some packages? PK On 16/03/2016, Vadim Zhukov <persg...@gmail.com> wrote: > 2016-03-15 21:33 GMT+03:00 Stuart Henderson <s...@spacehopper.org>: >> On 2016-03-15, Peter Kay <syllops...@syllopsium.co.uk> wrote: >>> It's a MOVSD SSE instruction. Tshark is ok. I can cope with that or >>> tcpdump >>> if need be, but here's the output : >> >> I think this variant of MOVSD might be AVX? >> >>> Starting program: /usr/local/bin/wireshark >>> warning: Lowest section in /usr/local/lib/libicudata.so.9.0 is .hash at >>> 00000154 >>> >>> Program received signal SIGILL, Illegal instruction. >>> 0x06d685fb in _GLOBAL__sub_I_qguiapplication.cpp () from >>> /usr/local/lib/qt5/./libQt5Gui.so.1.1 >> >> Looks like it's in Qt5 then. Wireshark still has the "legacy" gtk GUI >> (it's in a subpackage), you could try that instead for now. >> >> Looks like Qt autodetects at build time, we probably want to configure >> on i386 with no-avx, no-avx2, no-sse4.1, no-sse4.2, maybe no-ssse3. >> (SSE2 is probably reasonable to expect for Qt5 apps, it's present on >> Netburst, Pentium-M, Atom, C7 etc. which seems a sane cut-off point >> for heavy GUI apps). > > Yeah. I've patched qmake files, but missed those bits. Modern amd64 > makes you blind sometimes. :( > > -- > WBR, > Vadim Zhukov
