[bitc-dev] Bitc and Simd

[Ben Kloosterman]

Were talking about the use of the XMM registers ( rather than SIMD but its
an extention).  

 

Now there are several approaches here 

 

1)      You can give the language first grade support for 128 and 256 GP
registers ( which are limited to 32 bits for + - / *  but 128/256 for all
other ops ) .

2)      You can try to get most common use cases and hand craft them in
assembly  in libs  eg memcpy 

3)      You can try to let the compiler optimize it by trying to work out if
SSE can be used  but at present this  just works for some stores and writes
and suffers from alignment issues. Hence 2.

 

Now the 2) case is common but not very portable ( hence my comments as the
L4 guys are very concerned about the IPC speed) and inferior to a language
solution .  Note it's true that L4 don't use much SSE in their OS , but
Linux does 

 

This probably should be a spate thread but it does relate.

 

Regards, 

 

Ben 

 

From: jonathan.s.shapiro at gmail.com [mailto:jonathan.s.shapiro at gmail.com] On
Behalf Of Jonathan S. Shapiro
Sent: Wednesday, August 18, 2010 6:37 AM
To: bklooste at gmail.com; Discussions about the BitC language
Subject: Re: [bitc-dev] Bitc and Simd

 

Tyler certainly doesn't speak for me on this, but the L4 guys aren't doing
much SIMD support in their OS, so I'm not sure what the point of this
exchange is, exactly.

On Mon, Aug 16, 2010 at 12:36 AM, Ben Kloosterman <bklooste at gmail.com>
wrote:

It seems wrong that all the guys working on the L4 series OS have got it to
the point where they removed  ( most of ?) the asm and made it more portable
while the libs are going the other way and we are seeing more assembly.

 

Ben

 

<snip> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.coyotos.org/pipermail/bitc-dev/attachments/20100818/60515e4f/attachment.html