Date: Thu, 17 Jun 2004 18:24:12 -0700 From: Sean McNeil <sean@mcneil.com> To: freebsd-amd64@freebsd.org, freebsd-gnome@freebsd.org Subject: Re: bad compiler generated code? Message-ID: <1087521851.32817.27.camel@server.mcneil.com> In-Reply-To: <20040616015703.GA17852@dragon.nuxi.com> References: <1087346887.66377.1.camel@server.mcneil.com> <20040616015703.GA17852@dragon.nuxi.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, 2004-06-15 at 18:57, David O'Brien wrote:
> On Tue, Jun 15, 2004 at 05:48:08PM -0700, Sean McNeil wrote:
> > I'm recompiling firefox right now and I get a whole bunch of the
> > following:
> >
> > {standard input}:766: Warning: indirect jmp without `*'
> > {standard input}:816: Warning: indirect jmp without `*'
> ...
> > this happens on quite a number of C++ source files. Looks like the
> > compiler is generating some bad assembly.
>
> This is not a useful bug report. Can you provide the preprocesed C file?
> Or at least provide command line(s) invocation that produces this?
I took a look into this and it appears that the bad code generation of
c++ is related to my issues with firefox. Here is what I saw:
In xpcom/base/nsExceptionService.cpp it is generating thunks as
_ZThn8_N18nsExceptionService6AddRefEv:
addq $-8, %rdi
jmp _ZN18nsExceptionService6AddRefEv(%rip)
.size _ZThn8_N18nsExceptionService6AddRefEv, .-_ZThn8_N18nsExceptionService6AddRefEv
when they should be
_ZThn8_N18nsExceptionService6AddRefEv:
addq $-8, %rdi
jmp _ZN18nsExceptionService6AddRefEv
.size _ZThn8_N18nsExceptionService6AddRefEv, .-_ZThn8_N18nsExceptionService6AddRefEv
This has been fixed in the cvs version of gcc and involves the following patch:
*** contrib/gcc/config/i386/i386.c.orig Thu Jun 17 18:20:11 2004
--- contrib/gcc/config/i386/i386.c Thu Jun 17 18:10:55 2004
*************** x86_output_mi_thunk (file, thunk, delta,
*** 14607,14621 ****
output_asm_insn ("mov{l}\t{%0, %1|%1, %0}", xops);
}
! xops[0] = DECL_RTL (function);
if (TARGET_64BIT)
{
if (!flag_pic || (*targetm.binds_local_p) (function))
output_asm_insn ("jmp\t%P0", xops);
else
{
! tmp = XEXP (xops[0], 0);
! tmp = gen_rtx_UNSPEC (Pmode, gen_rtvec (1, tmp), UNSPEC_GOTPCREL);
tmp = gen_rtx_CONST (Pmode, tmp);
tmp = gen_rtx_MEM (QImode, tmp);
xops[0] = tmp;
--- 14607,14620 ----
output_asm_insn ("mov{l}\t{%0, %1|%1, %0}", xops);
}
! xops[0] = XEXP (DECL_RTL (function), 0);
if (TARGET_64BIT)
{
if (!flag_pic || (*targetm.binds_local_p) (function))
output_asm_insn ("jmp\t%P0", xops);
else
{
! tmp = gen_rtx_UNSPEC (Pmode, gen_rtvec (1, xops[0]), UNSPEC_GOTPCREL);
tmp = gen_rtx_CONST (Pmode, tmp);
tmp = gen_rtx_MEM (QImode, tmp);
xops[0] = tmp;
I am rebuilding firefox with this patch and will report back if firefox works.
Cheers,
Sean
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?1087521851.32817.27.camel>