[Mesa-dev] Optimizations in mesa shader compiler

[Segovia, Benjamin]

Yep, I start to read it and it is clearly what I was looking for. But I am still wondering if this is so useful to make it here since some back ends (like Intel GPU fragment programs and maybe g80s (???) ) use struct of array layouts which do not need swizzle.  

BTW, the simpler solution for these machines should be to transform the Mesa GPU code into a scalar code. So:
** mul a b c 
becomes
mul a.x b.xxxx c.xxxx
mul a.y b.yyyy c.yyyy
mul a.z b.zzzz c.zzzz
mul a.w b.wwww c.wwww
and then, we re-analyze the program as a scalar one.

Anyway, it may be still nice to have better programs higher in the sw stack.

Thanks for your help. I am studying the r300 code

Ben
 
________________________________________
From: Tom Stellard [tstellar at gmail.com]
Sent: Wednesday, July 14, 2010 10:33 PM
To: Segovia, Benjamin
Cc: mesa-dev at lists.freedesktop.org
Subject: Re: [Mesa-dev] Optimizations in mesa shader compiler

On Wed, Jul 14, 2010 at 09:16:15PM -0700, Segovia, Benjamin wrote:
> Thanks for that. It may be clearly useful.
>
> FYW, I output big shaders from nexuiz (> 100 inst) and clearly swizzling kills opportunities of optimizations. A serie of instructions may be totally useless but swizzles make the flow analysis more complicated.
> For big nexuiz fragment shaders, almost all the moves may be removed. This may lead to remove almost 40% of the instructions in these cases.
>
> I am not sure of the final speed up at the end. Intel GPUs use SoA computations for the fragment programs. So, it should be easy to completely remove the swizzle and to consider everything scalar. In that case, flow analysis becomes easier.
> However, despite that, the Intel Gen output still includes uncessary moves in particular from GPR to message registers. Not sure it is still related.
>
> Hmmm. Cool. We may be able anyway to make it faster :)
>
> Ben
>

Hi,

As Alex suggested, take a look at the r300 compiler.  If you are
looking to eliminate  MOV instructions, the peephole() function in
src/mesa/drivers/dri/r300/compiler/radeon_optimize.c might interest you.
It does copy propagation and is a good example of dataflow analysis
with swizzles.

-Tom