[Tinycc-devel] RE :Re: how hard to add SSE2 opcodes ?

[Christian JULLIEN]

I think Anaël wants to add new macros to generate SSE2 instructions as it is done by gcc and VC++ intrinsics.
Using those macros is as you were using ymm / xmm register directly in assembler.


> ----- message d'origine -----
> De : "Basile Starynkevitch" <address@hidden>
> date jeu. 10/09/2015 12:46 (GMT +02:00)
> À : "address@hidden" <address@hidden>, "address@hidden" <address@hidden>
> Objet : Re: [Tinycc-devel] how hard to add SSE2 opcodes ?
>
> On Thu, Sep 10, 2015 at 12:32:00PM +0200, Anaël Seghezzi wrote:
> > Hi all,
> > 
> > what would it take to add SSE2 opcodes to tcc assembler, at least for x64 ?
> > 
> > I'm working on a small interactive sandbox to play with C in realtime to create small demos/games with graphics and sounds,
> > a sort of mix of "ShaderToy" and "Pico8". I'm using TCC and it's all very effective, it could just gain some speed with a bit of SSE2 instructions.
> > 
>
>
> You won't get much speed, because TCC is known to very quickly
> generate poor machine code, e.g. without using the register
> efficiently. BTW, having some efficient register allocator would speed
> up the generated code significantly (independently of SSE2, you could
> hope a 3x to 10x speedup), but then you'll need a real-sized compiler.
>
> If execution speed of the generated code is a major concern, TCC is
> not the right tool for you. You could consider using some JIT library.
> See http://stackoverflow.com/a/12322672/841108 which gives more
> references.
>
> Regards.
> -- 
> Basile Starynkevitch           http://starynkevitch.net/Basile/
> France
>
> _______________________________________________
> Tinycc-devel mailing list
> address@hidden
> https://lists.nongnu.org/mailman/listinfo/tinycc-devel