egcs produces slow code on HPUX 10.20

egcs produces slow code on HPUX 10.20

[Jerry Quinn]

[-- Attachment #1: Type: text/plain, Size: 540 bytes --]

I've included a benchmark extracted out of our code that is
significantly slower for egcs 1.1a on an HPUX 10.20 pa-risc 2.0
machine.  In the file `results', note in particular the difference in
speed for the float test, since this is what we actually use.  The
constants are set up so as to avoid ending up in cache, because our
actual data is pretty large.

If you need .s files, I can generate them and send them on.

-- 
Jerry Quinn				Tel: (514) 761-8737
jquinn@nortel.ca			Fax: (514) 761-8505
Speech Recognition Research
speeders.tgz


[-- Attachment #2: speeders.tgz --]
[-- Type: application/x-gzip, Size: 3290 bytes --]

Re: egcs produces slow code on HPUX 10.20

[Jeffrey A Law]

  In message < c4b48993.3607a123@aol.com >you write:
  > In a message dated 9/21/98 10:14:32 PM Pacific Daylight Time,
  > Jerry.Quinn.jquinn@nt.com writes:
  > 
  > > code that is
  > >  significantly slower for egcs 1.1a on an HPUX 10.20 pa-risc 2.0
  > >  machine
  > You might be interested in the current main line snapshots.  I've seen
  > improvements up to 50% from egcs-1.1.
Note egcs doesn't optimize specifically for pa2.0 machines.  Certain opts
which improve code for pa1.1 (fmpyadd/fmpysub generation) are terrible for
performance on pa2.0 machines.

Jeff

Re: egcs produces slow code on HPUX 10.20

[N8TM]

In a message dated 9/21/98 10:14:32 PM Pacific Daylight Time,
Jerry.Quinn.jquinn@nt.com writes:

> code that is
>  significantly slower for egcs 1.1a on an HPUX 10.20 pa-risc 2.0
>  machine
You might be interested in the current main line snapshots.  I've seen
improvements up to 50% from egcs-1.1.