From:	hasdi AT umich DOT edu (Hasdi Rodzmann Hashim)
Newsgroups:	comp.os.msdos.djgpp
Subject:	Re: Using inline MMX assembly
Date:	18 Mar 1998 01:48:07 GMT
Lines:	31
Message-ID:	<6en94n$lgr@srvr1.engin.umich.edu>
References:	<Pine DOT SUN DOT 3 DOT 91 DOT 980317185245 DOT 19045L-100000 AT is>
NNTP-Posting-Host:	joust.rs.itd.umich.edu
To:	djgpp AT delorie DOT com
DJ-Gateway:	from newsgroup comp.os.msdos.djgpp

Eli Zaretskii (eliz AT is DOT elta DOT co DOT il) wrote:
: On Tue, 17 Mar 1998, Salvador Eduardo Tropea (SET) wrote:

: > Dirk Stevens <djsteven AT us DOT oracle DOT com> wrote:
: > 
: > > Can this be done with DJGPP ??
: > 
: > Yes, but there are a bug in one of the opcodes (PAND or something like this).
: > A patch for it anb a GCC  adapted to generate MMX code can be found in the 
: > following link:

: How about patching the DJGPP-compiled binaries on SimTel.NET?

If you are talking about binutils, I'll say go ahead, it's only a one line
patch. On the other hand, this bug was reported to cygnus (check dejanews)
and they plan to release the fix on the next release (I don't know if we
have to pay them for that).

For GCC+MMX patch, I'll say wait first. This patch is not fit for public
consumption (just for general feedback). If I get some encouragement, I'll
speed up the process, persuade the gcc maintainers to patch up 2.7.2.x
tree. I plan to add support VIS, MAX (PA-RISC 2.0) and MAX (AXP) and plain
ol 486.  Let me tell you right now that you can perform 8 parallel
addition to at most 6 cycles on *ANY* 64-bit machine. With MMX, this drops
to 1 cycle. 

Let me know if you have problems using the patch. Later

Hasdi
GCC-MMX extensions for 2.7.2.x:
http://www-personal.umich.edu/~hasdi/mmx.html

- Raw text -