[Qemu-devel] GCC 4 patches

[Qemu-devel] GCC 4 patches

[Gwenole Beauchesne]

Hi,

The following patches makes it possible to build QEMU with GCC 4 on x86.

However, there is still a problem with x86_64-softmmu. It may be a genuine
GCC bug though I don't see what's wrong with do_interrupt_real(). IOW, if
I build this function with GCC 3.3, x86_64-softmmu works fine, otherwise
QEMU hangs early. That's with GCC 4.0.3. With GCC 4.0.1, the problem looks
elsewhere, namely cpu-exec.c needs to be compiled with GCC 3 to work...

Anyhow, I am listing the current patches so that a GCC 4 compiled 
i386-softmmu binary works. Sorry, I don't have the time to look further 
the x86_64-softmmu problems.

Patches are available at:
<http://cvs.mandriva.com/cgi-bin/cvsweb.cgi/contrib-SPECS/qemu/>

* qemu-0.7.0-gcc4.patch
  Original patch from Paul Brook + an addition to correctly decode imul.

* qemu-0.7.2-dyngen-check-stack-clobbers.patch
  Check stack is not clobbered in synthetic opcodes using GOTO_LABEL_PARAM
  [ this might exhibit a gcc4 performance regression though, insofar as 
  gcc4 feels the need to generate stack variables ]

* qemu-0.7.2-gcc4-hacks.patch
  Various workarounds to GCC ICEs.

* qemu-0.7.2-gcc4-opts.patch
  Don't generate stack variables for x86_64-softmmu case in functions
  using GOTO_LABEL_PARAM.

IMHO, it's not worth looking hard at compiling QEMU with GCC 4 on x86. The 
new hand written backend is a better alternative.

Bye,
Gwenolé.

Re: [Qemu-devel] GCC 4 patches

[Fabrice Bellard]

Gwenole Beauchesne wrote:
> [...]
> IMHO, it's not worth looking hard at compiling QEMU with GCC 4 on x86. The 
> new hand written backend is a better alternative.

This is my opinion too !

Fabrice.