[Mesa-dev] [PATCH 6/6] gallivm: Add a new interface for doing TGSI->LLVM conversions

[Jose Fonseca]

----- Original Message -----
> On Fri, Jan 27, 2012 at 09:17:54AM -0800, Jose Fonseca wrote:
> > 
> > 
> > ----- Original Message -----
> > > On Mon, Jan 16, 2012 at 2:27 PM, Jose Fonseca
> > > <jfonseca at vmware.com>
> > > wrote:
> > > > Tom,
> > > >
> > > > Looks good in principle!
> > > >
> > > > But I need to test this myself before I can be comfortable w/
> > > > merging it into master.
> > > >
> > > 
> > > Hi Jose,
> > > 
> > > Have you had a chance to test this out yet?
> > 
> > Hi Tom,
> > 
> > Sorry for dropping the ball.  I just tried your branch now, and I'm
> > hitting this segfault:
> > 
> > #0  0x00007ffff46ab3f9 in lp_build_emit_fetch
> > (bld_base=0x7fffffffb010, inst=0xb36760, src_op=0,
> > chan_index=4294967295)
> >     at src/gallium/auxiliary/gallivm/lp_bld_tgsi.c:312
> > 312	   assert(reg->Register.Index <=
> > bld_base->info->file_max[reg->Register.File]);
> > (gdb) p bld_base->info
> > $4 = (const struct tgsi_shader_info *) 0x0
> > (gdb)
> > 
> > Apart of fixing this, I need to properly understand how aos is
> > being changed.
> > 
> > Jose
> >
> 
> Hi Jose,
> 
> Are you hitting that segfault with lp_build_tgsi_aos()?  If so, the
> attached patch should fix it. 

Yep. That fixed it. Thanks.

> Please note though, that the aos code
> has only been compile tested, since it isn't used anywhere in Mesa.

I was aware of that, but I still needed to ensure that it was not broken, or could be easily fixed later. But it seems to be producing exact the same LLVM IR as before so I'm happy. 

So this change is good by me.

Jose