[Qemu-devel] TCG: 128 bit ops?

[Qemu-devel] TCG: 128 bit ops?

[Andreas Färber]

Hello,

Am I correct in seeing that TCG offers no i128 ops? Would this be  
useful to add to TCG, internally reusing the i64/i32 code? ppc needs  
128-bit temporaries for Altivec instructions. It uses a union  
ppc_avr_t of (among others) two uint64_t.

Or should I just use two i64 temporaries with a local macro/helper?
It's not used much in ppc but I thought it might be useful to share  
with other architectures.

Andreas

Re: [Qemu-devel] TCG: 128 bit ops?

[Aurelien Jarno]

On Thu, Sep 04, 2008 at 01:10:39PM +0200, Andreas Färber wrote:
> Hello,
>
> Am I correct in seeing that TCG offers no i128 ops? Would this be useful 
> to add to TCG, internally reusing the i64/i32 code? ppc needs 128-bit 
> temporaries for Altivec instructions. It uses a union ppc_avr_t of (among 
> others) two uint64_t.
>
> Or should I just use two i64 temporaries with a local macro/helper?
> It's not used much in ppc but I thought it might be useful to share with 
> other architectures.
>

Currently TCG doesn't support i128 ops, but I also agree it would be
nice to have it. I think it can benefits for example to sparc64 which 
has 128-bit fp support.

-- 
  .''`.  Aurelien Jarno	            | GPG: 1024D/F1BCDB73
 : :' :  Debian developer           | Electrical Engineer
 `. `'   aurel32@debian.org         | aurelien@aurel32.net
   `-    people.debian.org/~aurel32 | www.aurel32.net

Re: [Qemu-devel] TCG: 128 bit ops?

[Blue Swirl]

On 9/4/08, Aurelien Jarno <aurelien@aurel32.net> wrote:
> On Thu, Sep 04, 2008 at 01:10:39PM +0200, Andreas Färber wrote:
>  > Hello,
>  >
>  > Am I correct in seeing that TCG offers no i128 ops? Would this be useful
>  > to add to TCG, internally reusing the i64/i32 code? ppc needs 128-bit
>  > temporaries for Altivec instructions. It uses a union ppc_avr_t of (among
>  > others) two uint64_t.
>  >
>  > Or should I just use two i64 temporaries with a local macro/helper?
>  > It's not used much in ppc but I thought it might be useful to share with
>  > other architectures.
>  >
>
>
> Currently TCG doesn't support i128 ops, but I also agree it would be
>  nice to have it. I think it can benefits for example to sparc64 which
>  has 128-bit fp support.

I agree that i128 would be useful, though float32/64/80/128 would be
even more so.

Sparc (32 and 64) specifications have included 128 bit float ops since
day one, but so far they haven't been implemented in real CPUs.

Re: [Qemu-devel] TCG: 128 bit ops?

[Paul Brook]

On Thursday 04 September 2008, Andreas Färber wrote:
> Hello,
>
> Am I correct in seeing that TCG offers no i128 ops? Would this be
> useful to add to TCG, internally reusing the i64/i32 code? ppc needs
> 128-bit temporaries for Altivec instructions. It uses a union
> ppc_avr_t of (among others) two uint64_t.
>
> Or should I just use two i64 temporaries with a local macro/helper?
> It's not used much in ppc but I thought it might be useful to share
> with other architectures.

Does it really have 128-bit ops, or just 128-bit vectors with smaller 
elements?

I expect that implementing native 128-bit support would end up fairly tricky. 
It's bad enough having to handle 64-bit ops 32-bit x86.

Paul