xlf -O3 -qstrict -C = bad code 
Author Message
 xlf -O3 -qstrict -C = bad code

Hi there,

I've just discovered that my code produces massively NaN's when
I used the following command on RS/6000 590

        xlf -O3 -qstrict -C -qarch=pwr2 -qtune=pwr2

Is such combination forbidden? In 10,000 lines of code the only
improperly generated instruction resides inside doubly nested
DO-loop and all arrays are inside commons, i.e. there is no
local or undeclared variables.

Anyone has similar experience?

Tomek
--
Saturday, 17:46 MET, 11-Mar-95
_______________________________________________________________________________


  Max-Planck-Institut fuer Astrophysik   phone: +49-89-32993200 (office)
  Karl-Schwarzschild-Strasse 1                         3206852  (home)
  Postfach 1523                          FAX:          32993235
  85740 Garching b. Muenchen
  Germany                                 http://www.*-*-*.com/ ~tomek/
_______________________________________________________________________________



Thu, 28 Aug 1997 00:47:12 GMT  
 
 [ 1 post ] 

 Relevant Pages 

1. GNAT cs.nyu.edu good and bad

2. assume cs:code,ds:code

3. Good code/bad code & looping

4. Good code/bad code & looping

5. RS6000 Stuff XLF<->C<->XLF

6. Porting DEC, SUN code to IBM RS6K (xlf)

7. to CS: or not to CS: in F-PC assembler

8. Initializing DX to Code Segment(CS) Memory Address...?!?

9. Initializing DX to Code Segment(CS) Memory Address…?!?

10. Problems with printing and video in O3 for DOS

11. Data_Error - GNAT -O3 problem

12. Examples of short programs showing --O3 wrong results

 

 
Powered by phpBB® Forum Software