From: Kai T. <kti...@go...> - 2011-05-26 10:27:27
|
2011/5/26 Ozkan Sezer <se...@gm...>: > On Thu, May 26, 2011 at 11:42 AM, Kai Tietz <kti...@go...> wrote: >> 2011/5/25 Dongsheng Song <don...@gm...>: >>> After I review the source, the failure was caused by sem_init only >>> support PTHREAD_PROCESS_PRIVATE, >>> but benchtest.c required PTHREAD_PROCESS_SHARED. >>> >>> Here is my patch. >>> >>> On Wed, May 25, 2011 at 22:11, Dongsheng Song <don...@gm...> wrote: >>>> >>>> Hi, >>>> >>>> With the attached patch, I can building and testing WinPthread, but benchmark testing 5th failed on XP SP3: >>>> >>>> $ cd winpthreads/tests >>>> $ make clean GC >>>> ... >>>> ALL TESTS PASSED! Congratulations! >>>> >>>> $ make GC-bench >>>> Copying .././outlib/libpthreadGC2-32.dll.a >>>> Copying .././outlib/pthreadGC2-32.dll >>>> make -k TEST=GC CC=-gcc XXCFLAGS="-D__CLEANUP_C" XXLIBS="" all-bench >>>> make[1]: Entering directory `/o/vcs/svn/mingw-w64/experimental/winpthreads/tests >>>> ' >>>> Copying ../include/pthread.h >>>> Copying ../include/semaphore.h >>>> Running benchtest1 >>>> ./benchtest1 >>>> ============================================================================= >>>> >>>> Lock plus unlock on an unlocked mutex. >>>> 10000000 iterations >>>> >>>> Test Total(msec) average(usec) >>>> ----------------------------------------------------------------------------- >>>> Dummy call x 2 46 0.005 >>>> Dummy call -> Interlocked with cond x 2 344 0.034 >>>> InterlockedOp x 2 328 0.033 >>>> Simple Critical Section 360 0.036 >>>> Old PT Mutex using a Critical Section (WNT) 406 0.041 >>>> Old PT Mutex using a Win32 Mutex (W9x) 9969 0.997 >>>> ............................................................................. >>>> PTHREAD_MUTEX_DEFAULT (W9x,WNT) 12640 1.264 >>>> PTHREAD_MUTEX_NORMAL (W9x,WNT) 12610 1.261 >>>> PTHREAD_MUTEX_ERRORCHECK (W9x,WNT) 12656 1.266 >>>> PTHREAD_MUTEX_RECURSIVE (W9x,WNT) 12969 1.297 >>>> ============================================================================= >>>> Done >>>> Compiling benchtest2.exe >>>> -gcc -O3 -UNDEBUG -Wall -D__CLEANUP_C benchlib.o -o benchtest2.exe benchtest2.c >>>> -I./include -L../outlib -lpthreadGC2-32 -lsupc++ >>>> Running benchtest2 >>>> ./benchtest2 >>>> ============================================================================= >>>> >>>> Lock plus unlock on a locked mutex. >>>> 100000 iterations, four locks/unlocks per iteration. >>>> >>>> Test Total(msec) average(usec) >>>> ----------------------------------------------------------------------------- >>>> Simple Critical Section 375 0.938 >>>> Old PT Mutex using a Critical Section (WNT) 437 1.093 >>>> Old PT Mutex using a Win32 Mutex (W9x) 500 1.250 >>>> ............................................................................. >>>> PTHREAD_MUTEX_DEFAULT (W9x,WNT) 640 1.600 >>>> PTHREAD_MUTEX_NORMAL (W9x,WNT) 609 1.523 >>>> PTHREAD_MUTEX_ERRORCHECK (W9x,WNT) 656 1.640 >>>> PTHREAD_MUTEX_RECURSIVE (W9x,WNT) 594 1.485 >>>> ============================================================================= >>>> Done >>>> Compiling benchtest3.exe >>>> -gcc -O3 -UNDEBUG -Wall -D__CLEANUP_C benchlib.o -o benchtest3.exe benchtest3.c >>>> -I./include -L../outlib -lpthreadGC2-32 -lsupc++ >>>> Running benchtest3 >>>> ./benchtest3 >>>> ============================================================================= >>>> >>>> Trylock on a locked mutex. >>>> 10000000 iterations. >>>> >>>> Test Total(msec) average(usec) >>>> ----------------------------------------------------------------------------- >>>> Old PT Mutex using a Critical Section (WNT) 296 0.030 >>>> Old PT Mutex using a Win32 Mutex (W9x) 5375 0.537 >>>> ............................................................................. >>>> PTHREAD_MUTEX_DEFAULT (W9x,WNT) 1297 0.130 >>>> PTHREAD_MUTEX_NORMAL (W9x,WNT) 1297 0.130 >>>> PTHREAD_MUTEX_ERRORCHECK (W9x,WNT) 1297 0.130 >>>> PTHREAD_MUTEX_RECURSIVE (W9x,WNT) 1297 0.130 >>>> ============================================================================= >>>> Done >>>> Compiling benchtest4.exe >>>> -gcc -O3 -UNDEBUG -Wall -D__CLEANUP_C benchlib.o -o benchtest4.exe benchtest4.c >>>> -I./include -L../outlib -lpthreadGC2-32 -lsupc++ >>>> Running benchtest4 >>>> ./benchtest4 >>>> ============================================================================= >>>> Trylock plus unlock on an unlocked mutex. >>>> 10000000 iterations. >>>> >>>> Test Total(msec) average(usec) >>>> ----------------------------------------------------------------------------- >>>> Old PT Mutex using a Critical Section (WNT) 407 0.041 >>>> Old PT Mutex using a Win32 Mutex (W9x) 10031 1.003 >>>> ............................................................................. >>>> PTHREAD_MUTEX_DEFAULT (W9x,WNT) 12797 1.280 >>>> PTHREAD_MUTEX_NORMAL (W9x,WNT) 12828 1.283 >>>> PTHREAD_MUTEX_ERRORCHECK (W9x,WNT) 12797 1.280 >>>> PTHREAD_MUTEX_RECURSIVE (W9x,WNT) 12937 1.294 >>>> ============================================================================= >>>> Done >>>> Compiling benchtest5.exe >>>> -gcc -O3 -UNDEBUG -Wall -D__CLEANUP_C benchlib.o -o benchtest5.exe benchtest5.c >>>> -I./include -L../outlib -lpthreadGC2-32 -lsupc++ >>>> Running benchtest5 >>>> ./benchtest5 >>>> ============================================================================= >>>> >>>> Operations on a semaphore. >>>> 1000000 iterations >>>> >>>> Test Total(msec) average(usec) >>>> ----------------------------------------------------------------------------- >>>> W32 Post with no waiters 454 0.454 >>>> W32 Wait without blocking 515 0.515 >>>> Assertion failed: (sem_init(&sema, 0, 0) == 0), file benchtest5.c, line 134 >>>> make[1]: *** [benchtest5.bench] Error 1 >>>> make[1]: Target `all-bench' not remade because of errors. >>>> make[1]: Leaving directory `/o/vcs/svn/mingw-w64/experimental/winpthreads/tests' >>>> >>>> make: *** [GC-bench] Error 2 >>>> >> >> Thanks for the patches. I will apply them next days together with >> license-disclaimer change. >> >> Regards, >> Kai > > Off-topic: IIRC, Dongsheng's msvcr90 patch was OK'ed. Is it applied? > > -- > O.S. Hmm, AFAIR (but I might be wrong) you applied it to branch and trunk, isn't it? Kai |