Control: tags -1 + upstream
Hello Claude Heiland-Allen, I tried just to collect some more information for the maintainer. The issue could be reproduced in a qemu VM with '-cpu host' on a Ryzen 7 1700. The resulting binary crashes on Windows at the same instruction, so I guess Wine can be ruled out. It seems that the vmovapd instruction expects a 32-byte aligned value in register $rcx. Unfortunately it is just 16-byte aligned and therefore the SIGSEGV. This could be confirmed by setting a breakpoint to that instruction and set $rcx to the next aligned value below, then the vmovapd could be executed. I think following upstream reports and messages are releated, mostly 54412. There are also some workarounds mentioned. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=49001 https://gcc.gnu.org/bugzilla/show_bug.cgi?id=54412 https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61730 https://sourceforge.net/p/mingw-w64/mailman/message/36287627/ Kind regards, Bernhard $ wine winedbg --gdb a.exe 4 ... Wine-gdb> cont Continuing. Program received signal SIGSEGV, Segmentation fault. 0x000000000040166d in ?? () Wine-gdb> bt #0 0x000000000040166d in ?? () #1 0x00000000004031a8 in ?? () #2 0x0000000000000000 in ?? () Wine-gdb> print/x $rcx $1 = 0x33fa10 Wine-gdb> print/x $rcx % 32 $2 = 0x10 Wine-gdb> print/x $rcx % 16 $3 = 0x0 $ x86_64-w64-mingw32-objdump --disassemble a.exe ... 0000000000401665 <_Z9broadcastIDv4_ddET_T0_>: 401665: c4 e2 7d 19 c9 vbroadcastsd %xmm1,%ymm1 40166a: 48 89 c8 mov %rcx,%rax 40166d: c5 fd 29 09 vmovapd %ymm1,(%rcx) 401671: c5 f8 77 vzeroupper 401674: c3 retq ... 0000000000402e90 <main>: ... 40319b: 48 8d 8c 24 e0 00 00 lea 0xe0(%rsp),%rcx 4031a2: 00 4031a3: e8 bd e4 ff ff callq 401665 <_Z9broadcastIDv4_ddET_T0_> 4031a8: c5 fd 28 94 24 e0 00 vmovapd 0xe0(%rsp),%ymm2 ... That function _Z9broadcastIDv4_ddET_T0_ should unmangled look like: double __vector(4) broadcast<double __vector(4), double>(double) That should be line 53: 51 template <> 52 __attribute__ ((noinline)) 53 double4 broadcast<double4 ,double>(double x) { double4 r = { x, x, x, x }; return r; }
