Hi all.
This patch fixes an oversight with ETIME/DTIME. While ETIME reports the
ressource usage since invokation of the application, DTIME is supposed to
give the ressource usage since the last invokation of DTIME. Up until now,
DTIME was an alias of ETIME.
The library implementation of DTIME uses a mutex to ensure data integregrity
if used with multiple threads, e.g. OpenMP. If no time-source is available,
DTIME returns -1 for each of its output values for each invokation.
Further, I took the opportunity to merge similar codes used dor CPU_TIME and
ETIME into a single function which now can be used by all three intrinsics,
CPU_TIME, DTIME and ETIME. This also implies that we now can provide a
working implementation of DTIME/ETIME for mingw.
FX, would you be so kind to test these functions there? :)
:ADDPATCH fortran:
gcc/fortran:
2007-12-21 Daniel Franke <franke.daniel@gmail.com>
* intrinsic.h (gfc_check_etime): Renamed to ...
(gfc_check_dtime_etime): ... this.
(gfc_check_etime_sub): Renamed to ...
(gfc_check_dtime_etime_sub): ... this.
(gfc_resolve_dtime_sub): New prototype.
* check.c (gfc_check_etime): Renamed to ...
(gfc_check_dtime_etime): ... this.
(gfc_check_etime_sub): Renamed to ...
(gfc_check_dtime_etime_sub): ... this.
* iresolve.c (gfc_resolve_dtime_sub): New implementation.
* intrinsic.c (add_functions): Removed alias from ETIME to DTIME,
added stand-alone intrinsic DTIME.
(add_subroutines): Adjusted check and resolve function names for
DTIME and ETIME.
* trans-intrinsic.c (gfc_conv_intrinsic_function): Added DTIME
to known functions in switch.
* intrinsic.texi (DTIME): Added paragraph about thread-safety,
fixed return value section.
(CPU_TIME): Clarified intent and added implementation notes.
libgfortran:
2007-12-21 Daniel Franke <franke.daniel@gmail.com>
* intrinsics/cpu_time.c: Moved code commonly usable for CPU_TIME,
DTIME and ETIME to ...
* intrinsics/time_1.h: ... here.
* intrinsics/dtime.c: New file.
* intrinsics/etime.c: Newly implemented using the common
time-aquisition function from time_1.h.
* gfortran.map (_gfortran_dtime, _gfortran_dtime_sub): New.
* Makefile.am: Added new file.
* Makefile.in: Regenerated.
* configure: Regenerated.
I'm not sure that I got all the various macros right, this might be a source
of easily-fixable regressions on some platforms ...
The patch does not contain the regenerated files as I currently do not have
the appropriate version of autoconf available.
Bootstrapped and regression tested on i686-pc-linux-gnu. Ok for trunk?
Regards
Daniel