On Fri, Jun 14, 2013 at 8:30 AM, Nathan Kurz <[email protected]> wrote:
> I guess it would allow you to get a correct backtrace if there was a
> segfault or breakpoint in the single 'mov' instruction before the 'jmp'.

It's gone after the commit I just made, which hacks in `-fomit-frame-pointer`.

Dump of assembler code for function cfish_thunk112:
0x00000001000715f0 <cfish_thunk112+0>: mov    0x8(%rdi),%rax
0x00000001000715f4 <cfish_thunk112+4>: jmpq   *0x70(%rax)
0x00000001000715f7 <cfish_thunk112+7>: nopw   0x0(%rax,%rax,1)
End of assembler dump.

Curiously we were already at optimization level `-O2`.  Whatever.  If this
proof-of-concept works out we can write per-architecture assembler per Nick's
suggestion.  It's not a lot of code.  Load-time symbol resolution is a harder
problem.

Marvin Humphrey

Reply via email to