On Sunday, August 27, 2023 12:43:31 P.M. CDT Karine Crèvecœur wrote:
> (gdb) disassemble > … > 0x00007ffff6cc20e8 <+136>: movaps %xmm8,%xmm4 > 0x00007ffff6cc20ec <+140>: mov %edx,-0x4c(%rsp) > 0x00007ffff6cc20f0 <+144>: mov 0x38(%r9),%rdx > 0x00007ffff6cc20f4 <+148>: mulss %xmm5,%xmm4 > 0x00007ffff6cc20f8 <+152>: movss 0x3c(%r9),%xmm10 > 0x00007ffff6cc20fe <+158>: movq %r10,%xmm7 > => 0x00007ffff6cc2103 <+163>: extractps $0x3,%xmm12,%r11d > 0x00007ffff6cc210a <+170>: mov %rdx,%r15 > 0x00007ffff6cc210d <+173>: mov 0x28(%rax),%rdx > 0x00007ffff6cc2111 <+177>: movshdup %xmm7,%xmm7 > 0x00007ffff6cc2115 <+181>: extractps $0x2,%xmm12,%edi > 0x00007ffff6cc211c <+188>: mulss %xmm6,%xmm2 > 0x00007ffff6cc2120 <+192>: movq %rdx,%xmm15 > 0x00007ffff6cc2125 <+197>: mov %rdx,-0x40(%rsp) > … > > The instruction that leads to crash seems to be "extractps".According > to <https://www.felixcloutier.com/x86/extractps> it is an instruction > related to SSE4.1. Thank you! That is exactly what I needed to know. So it is clear that some folks need a build with SSE4 disabled. But SSE2 is supported on the machines of you and the original reporter so I'll try first with SSE2 but not SSE4. The remaining issue is: I need to figure a way to build two binary packages from the source. The other alternative is to disable SSE4 for everyone but I have no idea what performance impact that might have. Regards, -Steve
signature.asc
Description: This is a digitally signed message part.