Thread: Re: Alpha spinlock
Tom Lane wrote:
> For a while I though it might be because we are using an alpha TAS inThe current code uses semaphores, which has the advantage that it works well even on multi-processor machines, but the disadvantage that it is not the fastest way possible. Writing a spinlock on Alpha for SMP machines is very difficult, as you need to deal with memory barriers. A real mess. But then one of the people at Compaq pointed out to us that there is a ready-made routine on Alpha. We implemented it with the two patches below. I ran tests with lots of parallel back-ends and got around a 10% speed increase. I include the two patches. Perhaps some of the other people running Tru64 can have a look at these as well.
> the spinlock rather than the old semaphore. I replaced our spinlock
> with the standard one and it made no difference. We have been running
> with our spinlock implementation for nearly 2 months on a production
> database now without a hitch, so I think it is ok. Did I ever submit
> any patches for the Alpha spinlock?Not that I recall. We did get some advice from some Alpha gurus at DEC
who seemed to think the existing TAS code is OK. What was it that you
felt needed to be improved?
Cheers,
Adriaan
Adriaan Joubert <a.joubert@albourne.com> writes:
> Tom Lane wrote:
>> Not that I recall. We did get some advice from some Alpha gurus at DEC
>> who seemed to think the existing TAS code is OK. What was it that you
>> felt needed to be improved?
> The current code uses semaphores,
Oh, I'm sorry, I was thinking about the Linux-Alpha code, for which
there has been an inline-assembly version of tas() for a long time;
and it was that code that we asked the DEC people about. I had
forgotten that the OSF port uses a different semaphore method.
Is there a reason we couldn't use the Linux-Alpha code on OSF too?
I'd just as soon minimize the difference between the two ports ...
regards, tom lane
Tom, what do you want to do with this patch? > > > Tom Lane wrote: > > > > For a while I though it might be because we are using an alpha TAS in > > > the spinlock rather than the old semaphore. I replaced our spinlock > > > with the standard one and it made no difference. We have been running > > > with our spinlock implementation for nearly 2 months on a production > > > database now without a hitch, so I think it is ok. Did I ever submit > > > any patches for the Alpha spinlock? > > > > Not that I recall. We did get some advice from some Alpha gurus at DEC > > who seemed to think the existing TAS code is OK. What was it that you > > felt needed to be improved? > > The current code uses semaphores, which has the advantage that it works > well even on multi-processor machines, but the disadvantage that it is not > the fastest way possible. Writing a spinlock on Alpha for SMP machines is > very difficult, as you need to deal with memory barriers. A real mess. But > then one of the people at Compaq pointed out to us that there is a > ready-made routine on Alpha. We implemented it with the two patches below. > I ran tests with lots of parallel back-ends and got around a 10% speed > increase. I include the two patches. Perhaps some of the other people > running Tru64 can have a look at these as well. > > Cheers, > > Adriaan > *** src/include/storage/s_lock.h Sat Sep 30 12:13:42 2000 > --- src/include/storage/s_lock.h- Sat Sep 30 12:13:42 2000 > *************** > *** 252,266 **** > * Note that slock_t on the Alpha AXP is msemaphore instead of char > * (see storage/ipc.h). > */ > - #include <alpha/builtins.h> > - #if 0 > #define TAS(lock) (msem_lock((lock), MSEM_IF_NOWAIT) < 0) > #define S_UNLOCK(lock) msem_unlock((lock), 0) > #define S_INIT_LOCK(lock) msem_init((lock), MSEM_UNLOCKED) > #define S_LOCK_FREE(lock) (!(lock)->msem_state) > - #else > - #define TAS(lock) (__INTERLOCKED_TESTBITSS_QUAD((lock),0)) > - #endif > > #else /* i.e. not __osf__ */ > > --- 252,261 ---- > *** src/include/port/alpha.h Sat Sep 30 12:13:21 2000 > --- src/include/port/alpha.h- Sat Sep 30 12:13:21 2000 > *************** > *** 1,10 **** > #define USE_POSIX_TIME > #define DISABLE_XOPEN_NLS > #define HAS_TEST_AND_SET > ! /*#include <sys/mman.h>*/ /* for msemaphore */ > ! /*typedef msemaphore slock_t;*/ > ! #include <alpha/builtins.h> > ! typedef volatile long slock_t; > > /* some platforms define __alpha, but not __alpha__ */ > #if defined(__alpha) && !defined(__alpha__) > --- 1,8 ---- > #define USE_POSIX_TIME > #define DISABLE_XOPEN_NLS > #define HAS_TEST_AND_SET > ! #include <sys/mman.h> /* for msemaphore */ > ! typedef msemaphore slock_t; > > /* some platforms define __alpha, but not __alpha__ */ > #if defined(__alpha) && !defined(__alpha__) -- Bruce Momjian | http://candle.pha.pa.us pgman@candle.pha.pa.us | (610) 853-3000 + If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill, Pennsylvania 19026
Bruce Momjian <pgman@candle.pha.pa.us> writes:
> Tom, what do you want to do with this patch?
Apply away ...
regards, tom lane
Applied, but include/port/alpha.h is now include/port/osf.h. > > > Tom Lane wrote: > > > > For a while I though it might be because we are using an alpha TAS in > > > the spinlock rather than the old semaphore. I replaced our spinlock > > > with the standard one and it made no difference. We have been running > > > with our spinlock implementation for nearly 2 months on a production > > > database now without a hitch, so I think it is ok. Did I ever submit > > > any patches for the Alpha spinlock? > > > > Not that I recall. We did get some advice from some Alpha gurus at DEC > > who seemed to think the existing TAS code is OK. What was it that you > > felt needed to be improved? > > The current code uses semaphores, which has the advantage that it works > well even on multi-processor machines, but the disadvantage that it is not > the fastest way possible. Writing a spinlock on Alpha for SMP machines is > very difficult, as you need to deal with memory barriers. A real mess. But > then one of the people at Compaq pointed out to us that there is a > ready-made routine on Alpha. We implemented it with the two patches below. > I ran tests with lots of parallel back-ends and got around a 10% speed > increase. I include the two patches. Perhaps some of the other people > running Tru64 can have a look at these as well. > > Cheers, > > Adriaan > *** src/include/storage/s_lock.h Sat Sep 30 12:13:42 2000 > --- src/include/storage/s_lock.h- Sat Sep 30 12:13:42 2000 > *************** > *** 252,266 **** > * Note that slock_t on the Alpha AXP is msemaphore instead of char > * (see storage/ipc.h). > */ > - #include <alpha/builtins.h> > - #if 0 > #define TAS(lock) (msem_lock((lock), MSEM_IF_NOWAIT) < 0) > #define S_UNLOCK(lock) msem_unlock((lock), 0) > #define S_INIT_LOCK(lock) msem_init((lock), MSEM_UNLOCKED) > #define S_LOCK_FREE(lock) (!(lock)->msem_state) > - #else > - #define TAS(lock) (__INTERLOCKED_TESTBITSS_QUAD((lock),0)) > - #endif > > #else /* i.e. not __osf__ */ > > --- 252,261 ---- > *** src/include/port/alpha.h Sat Sep 30 12:13:21 2000 > --- src/include/port/alpha.h- Sat Sep 30 12:13:21 2000 > *************** > *** 1,10 **** > #define USE_POSIX_TIME > #define DISABLE_XOPEN_NLS > #define HAS_TEST_AND_SET > ! /*#include <sys/mman.h>*/ /* for msemaphore */ > ! /*typedef msemaphore slock_t;*/ > ! #include <alpha/builtins.h> > ! typedef volatile long slock_t; > > /* some platforms define __alpha, but not __alpha__ */ > #if defined(__alpha) && !defined(__alpha__) > --- 1,8 ---- > #define USE_POSIX_TIME > #define DISABLE_XOPEN_NLS > #define HAS_TEST_AND_SET > ! #include <sys/mman.h> /* for msemaphore */ > ! typedef msemaphore slock_t; > > /* some platforms define __alpha, but not __alpha__ */ > #if defined(__alpha) && !defined(__alpha__) -- Bruce Momjian | http://candle.pha.pa.us pgman@candle.pha.pa.us | (610) 853-3000 + If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill, Pennsylvania 19026