Mailinglist Archive: radeonhd (365 mails)

[Matthias Hopf <mhopf@xxxxxxx>]

On Mar 01, 09 18:42:37 +0100, Christian König wrote:
> the attached patch moves the r600 shaders from r600_exa.c into separate
> files. It also fixes a bunch of typos in the xv pixel shaders (DOT4
> instead of MOV instruction in the 4th ALU instruction), and replaces
> every C++ comment in r600_exa.c with C comments.

Thanks a lot, Christian :)

> Bye the way i'm using the r600 shader assembler and disassembler in from
> the ATI stream SDK
> (http://ati.amd.com/technology/streamcomputing/sdkdwnld.html)
> plus a self made awk script to decode/debug the shaders.

Interesting. Never have looked into it, but I probably should ;)

> > Ideally we'd use a shader compiler and we'd write the shaders in glsl
> > or some other IR, but we don't have one released yet.
> For a compiler we would first need an assembler/disassembler.
> So is anybody working on an assembler/disassembler already?
> Or can i start writing BNF for yacc/bison?
> Or should we stick with an simpler form?
> Somethink like an awk or perl script?

Given my experience of >10000 lines of awk scripts: don't. do. that.

But using perl is probably the best solution. It's pure text processing,
and the grammar should be sufficiently simple to make perl a feasible
option.

I eventually wanted to do that myself, but never had the time...

IMHO the most important thing an assembler has to do is to validate the
side conditions (bank swizzeling, barriers, last alu op, etc.).
I already had too many headaches with them.

CU

Matthias

-- 
Matthias Hopf <mhopf@xxxxxxx>      __        __   __
Maxfeldstr. 5 / 90409 Nuernberg   (_   | |  (_   |__          mat@xxxxxxxxx
Phone +49-911-74053-715           __)  |_|  __)  |__  R & D   www.mshopf.de
-- 
To unsubscribe, e-mail: radeonhd+unsubscribe@xxxxxxxxxxxx
For additional commands, e-mail: radeonhd+help@xxxxxxxxxxxx