#3904: Bug in shipped gcc makes profiling lose resolution
-----------------------------+----------------------------------------------
Reporter: augustss | Owner:
Type: bug | Status: new
Priority: normal | Milestone:
Component: Build System | Version: 6.12.1
Keywords: | Difficulty:
Os: Windows | Testcase:
Architecture: x86 | Failure: None/Unknown
-----------------------------+----------------------------------------------
Comment(by simonmar):
Replying to [comment:3 augustss]:
> Switching back to using (unsigned) int seems like the simplest fix then.
The patch that introduced the 64-bit values was:
{{{
Mon Sep 11 23:32:10 BST 2006 Ian Lynagh <[email protected]>
* Fix ~2000 second profiling time wrapping bug
{
hunk ./rts/ProfHeap.c 389
- fprintf(hp_file, "%s %d.%02d\n",
+ fprintf(hp_file, "%s %" FMT_Word64 ".%02" FMT_Word64 "\n",
hunk ./rts/ProfHeap.c 391
- (int)integralPart, (int)(fractionalPart * 100));
+ (StgWord64)integralPart, (StgWord64)(fractionalPart *
100));
hunk ./rts/posix/GetTime.c 47
- return (t.ru_utime.tv_sec * TICKS_PER_SECOND + [_$_]
+ return ((Ticks)t.ru_utime.tv_sec * TICKS_PER_SECOND + [_$_]
hunk ./rts/posix/GetTime.c 55
- return (tv.tv_sec * TICKS_PER_SECOND +
+ return ((Ticks)tv.tv_sec * TICKS_PER_SECOND +
hunk ./rts/posix/GetTime.c 123
- return (ts.tv_sec * TICKS_PER_SECOND + [_$_]
+ return ((Ticks)ts.tv_sec * TICKS_PER_SECOND + [_$_]
}
}}}
I can see that the changes to `GetTime` were necessary to avoid
overflowing 32-bit integers (TICKS_PER_SECOND==1000000), but I don't see
why we need to use `StgWord64` in `printSample`, since the values there
are seconds, not microseconds. Ian, any idea why this was done?
--
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/3904#comment:4>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler
_______________________________________________
Glasgow-haskell-bugs mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs
