streams on Challange

From: Robert van Liere (Robert.van.Liere@cwi.nl)
Date: Tue Jul 19 1994 - 08:07:51 CDT


John,

[For what its worth...]

Here are some results of your streams program on a multi-processor
Challange @ 150 Mhz running IRIX 5.2.

Included are three cases:
   - sequential code
   - parallel code -- compiled with -pfa, the POWER FORTRAN accelerator.
        -- on a 2 CPU Challange
        -- on a 6 CPU Challange

The -pfa does all sorts of things like loop unrolling, some data
dependency analysis, etc. (you probably know more abt this than i).
I have the impression that (some) improvement can be made by
finding better compiler options.

Cheers,
-- Robert van Liere

Included for each of three cases is:
        - compiler options
        - uname -a
        - hardware inventory | grep 150
        - a.out output
1. --------------------------------------------------------------------
; f77 -mips2 -O3 -non_shared stream_d.f
; uname -a
IRIX artemis 5.2 02282015 IP19 mips
; a.out
--------------------------------------
 Double precision appears to have 16 digits of accuracy
 Assuming 8 bytes per DOUBLEPRECISION word
--------------------------------------
 Timing calibration ; time = 159.0000024065375 hundredths of a second
 Increase the size of the arrays if this is <30
  and your clock precision is =<1/100 second
 ---------------------------------------------------
Function Rate (MB/s) RMS time Min time Max time
Assignment: 62.7451 0.5231 0.5100 0.5400
Scaling : 58.1819 0.5520 0.5500 0.5600
Summing : 63.1580 0.7700 0.7600 0.7800
SAXPYing : 70.5882 0.6920 0.6800 0.7000

2. --------------------------------------------------------------------
; f77 -pfa -mips2 -O3 -non_shared stream_d.f
; hinv | grep 150
2 150 MHZ IP19 Processors
; uname -a
IRIX artemis 5.2 02282015 IP19 mips
; a.out
--------------------------------------
 Double precision appears to have 16 digits of accuracy
 Assuming 8 bytes per DOUBLEPRECISION word
--------------------------------------
 Timing calibration ; time = 178.0000081285834 hundredths of a second
 Increase the size of the arrays if this is <30
  and your clock precision is =<1/100 second
 ---------------------------------------------------
Function Rate (MB/s) RMS time Min time Max time
Assignment: 133.3337 0.2461 0.2400 0.2600
Scaling : 139.1304 0.2441 0.2300 0.2500
Summing : 133.3338 0.3721 0.3600 0.3800
SAXPYing : 141.1764 0.3551 0.3400 0.3600

3. --------------------------------------------------------------------
; f77 -pfa -mips2 -O3 -non_shared stream_d.f
; hinv | grep 150
6 150 MHZ IP19 Processors
; uname -a
IRIX zeus 5.2 02282015 IP19 mips
; a.out
--------------------------------------
 Double precision appears to have 16 digits of accuracy
 Assuming 8 bytes per DOUBLEPRECISION word
--------------------------------------
 Timing calibration ; time = 80.99999930709600 hundredths of a second
 Increase the size of the arrays if this is <30
  and your clock precision is =<1/100 second
 ---------------------------------------------------
Function Rate (MB/s) RMS time Min time Max time
Assignment: 355.5568 0.1018 0.0900 0.1300
Scaling : 355.5559 0.1015 0.0900 0.1200
Summing : 369.2308 0.1444 0.1300 0.1600
SAXPYing : 369.2304 0.1467 0.1300 0.1700

--------------------------------------------------
Robert van Liere
robertl@cwi.nl
<http://www.cwi.nl/cwi/people/Robert.van.Liere.html>



This archive was generated by hypermail 2b29 : Tue Apr 18 2000 - 05:23:04 CDT