diff options
Diffstat (limited to 'win/tclWinTime.c')
| -rw-r--r-- | win/tclWinTime.c | 21 |
1 files changed, 16 insertions, 5 deletions
diff --git a/win/tclWinTime.c b/win/tclWinTime.c index d90e8ba..d0aa9ec 100644 --- a/win/tclWinTime.c +++ b/win/tclWinTime.c @@ -9,7 +9,7 @@ * See the file "license.terms" for information on usage and redistribution * of this file, and for a DISCLAIMER OF ALL WARRANTIES. * - * RCS: @(#) $Id: tclWinTime.c,v 1.9 2001/11/21 02:36:21 hobbs Exp $ + * RCS: @(#) $Id: tclWinTime.c,v 1.10 2002/04/01 20:44:36 kennykb Exp $ */ #include "tclWinInt.h" @@ -254,17 +254,28 @@ Tcl_GetTime(timePtr) * - unpredictable changes in performance counter frequency * on "gearshift" processors such as Transmeta and * SpeedStep. + * * There seems to be no way to test whether the performance * counter is reliable, but a useful heuristic is that * if its frequency is 1.193182 MHz or 3.579545 MHz, it's * derived from a colorburst crystal and is therefore - * the RTC rather than the TSC. If it's anything else, we - * presume that the performance counter is unreliable. + * the RTC rather than the TSC. + * + * A sloppier but serviceable heuristic is that the RTC crystal + * is normally less than 15 MHz while the TSC crystal is + * virtually assured to be greater than 100 MHz. Since Win98SE + * appears to fiddle with the definition of the perf counter + * frequency (perhaps in an attempt to calibrate the clock?) + * we use the latter rule rather than an exact match. */ if ( timeInfo.perfCounterAvailable - && timeInfo.curCounterFreq.QuadPart != (LONGLONG) 1193182 - && timeInfo.curCounterFreq.QuadPart != (LONGLONG) 3579545 ) { + /* The following lines would do an exact match on + * crystal frequency: + * && timeInfo.curCounterFreq.QuadPart != (LONGLONG) 1193182 + * && timeInfo.curCounterFreq.QuadPart != (LONGLONG) 3579545 + */ + && timeInfo.curCounterFreq.QuadPart > (LONGLONG) 15000000 ) { timeInfo.perfCounterAvailable = FALSE; } |