[bitc-dev] Bitc and Simd

[Ben Kloosterman]

Yes it is possible without intrinsics which would make BitC code less HW
dependent. 

 

Using the reg as a  GP registers is easy ( and outline as phase 2 ) in my
earlier mails, A optimizing compiler should even be able to place as an int
in a XMM registers and use it as a GP.

 

In my proposed phases Phase 2 would allow

 

Int I = 12; 

Int j = 3; 

xmm r1 =  (xmm)  I + j;  // note that if I + j are greater than max int it
will wrap.  For GP purposes xmm is 32 bits. 

 

Phase 3 would allow  this 

 

xmm r1 = {1,2,3,4};

Xmm r2  = {4,3,2,1};

 

Let r3 = r1+r2; 

 

However implicitly converting to  SIMD is hard , how do you  communicate
intent ? 

 

Ben

 

 

From: Tyler Tricker [mailto:tntricker at gmail.com] 
Sent: Sunday, August 15, 2010 11:05 AM
To: bklooste at gmail.com; Discussions about the BitC language
Subject: Re: [bitc-dev] Bitc and Simd

 

Is there anyway to provide SIMD without resorting to using just intrinsic
functions(so applications that want to support them don't need to access
them)?

 

To me regardless of the architecture, operations over arrays and unions
should be able to support implicit SIMD functionality.

 

No virus found in this incoming message.
Checked by AVG - www.avg.com
Version: 9.0.851 / Virus Database: 271.1.1/3069 - Release Date: 08/15/10
02:34:00

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.coyotos.org/pipermail/bitc-dev/attachments/20100815/03b1c1bb/attachment.html