| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 1244.1 | This is being worked on as CRTL problem number 1656 | TLE::D_SMITH | Duane Smith -- DEC C RTL | Thu Jan 09 1997 13:45 | 1 |
| 1244.2 | only sprintf() problem? | TAVENG::BORIS | Boris Gubenko, ISE | Thu Jan 09 1997 14:38 | 75 |
| 1244.3 | 2.5? | CAIRN::HARRIS | Kevin Harris, dtn 381-2039 | Thu Jan 09 1997 15:52 | 26 |
| 1244.4 | Workarounds?... | CSC32::R_GROSSMAN | Don't forget to dance. | Thu Jan 09 1997 16:42 | 19 |
| 1244.5 | | DECCXL::WIBECAN | That's the way it is, in Engineering! | Thu Jan 09 1997 16:51 | 9 |
| 1244.6 | | CSC32::D_DERAMO | Dan D'Eramo, Customer Support Center | Thu Jan 09 1997 18:53 | 23 |
| 1244.7 | thanks for the correction, Kevin | TLE::CRTL | | Fri Jan 10 1997 03:14 | 32 |
| 1244.8 | I wonder why I still see 2 1/2 slower response? | CSC32::R_GROSSMAN | Don't forget to dance. | Tue Jan 14 1997 13:14 | 27 |
| 1244.9 | | CXXC::MJHANS | Matthew Hanselman, DEC C | Tue Jan 14 1997 15:39 | 8 |
| 1244.10 | | CXXC::MJHANS | Matthew Hanselman, DEC C | Wed Jan 15 1997 13:23 | 6 |
| 1244.11 | Went picking low hanging fruit | TLE::D_SMITH | Duane Smith -- DEC C RTL | Wed Jan 29 1997 11:44 | 12 |
| I've looked at the sprintf performance problem reported in .0. While
the dec c sprintf does support more options than the vax c sprintf, I
did not want to dismiss this real user problem so quickly.
While the actual 2.5 times slower number was not reproduced locally, I
have been able to significantly reduce the number of CPU seconds taken
by this benchmark. On an Alpha machine, the number of CPU seconds has
been reduced from 29 to 13. That's pretty danged significant if you
ask me. Given that this is all common code between VAX and Alpha, I
expect most of the win to carry over to VAX.
This work will appear in the OpenVMS release after V7.1.
|
| 1244.12 | Almost caught VAX C | TLE::D_SMITH | Duane Smith -- DEC C RTL | Thu Feb 27 1997 16:03 | 17 |
| Russ Grossman was kind enough to remeasure the performance numbers for
me. There are three sets of numbers shown in the following chart
comparing the VAXC, DECC V7.1, and new DECC RAVEN numbers. The
customer was looking for a solid commitment to improving the printf
performance:
VAX C V7.1 RAVEN
10000 sprintf loop 5.77 14.39 6.45
10000 fprintf loop 4.34 9.83 4.85
Now we can argue that DEC C is "a little bit" slower due to an
increased number of format specifiers... 8-)
This performance work will be included in the next round of ECO kits
for OpenVMS V7.1 and in the next point release after V7.1.
|