This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH 2/2] PowerPC: libc single-thread lock optimization
- From: Will Newton <will dot newton at linaro dot org>
- To: Adhemerval Zanella <azanella at linux dot vnet dot ibm dot com>
- Cc: "GNU C. Library" <libc-alpha at sourceware dot org>
- Date: Thu, 1 May 2014 10:11:14 +0100
- Subject: Re: [PATCH 2/2] PowerPC: libc single-thread lock optimization
- Authentication-results: sourceware.org; auth=none
- References: <5361013D dot 9020005 at linux dot vnet dot ibm dot com>
On 30 April 2014 14:57, Adhemerval Zanella <azanella@linux.vnet.ibm.com> wrote:
> This patch adds a single-thread optimization for locks used within
> libc.so. For each lock operations it checks it the process has already
> spawned one thread and if not use non-atomic operations. Other libraries
> (libpthread.so for instance) are unaffected by this change.
>
> This is a respin on my first patch to add such optimization, but now the
> code is focused only on lowlevellock.h, the atomic.h is untouched.
>
> Tested on powerpc32 and powerpc64.
>
> --
>
> * nptl/sysdeps/unix/sysv/linux/powerpc/lowlevellock.h
> (__lll_is_single_thread): New macro to check if process has spawned
> any threads.
> (__lll_robust_trylock): Add optimization to avoid atomic operations in
> single thread case.
> (lll_lock): Likewise.
> (lll_robust_lock): Likewise.
> (lll_cond_lock): Likewise.
> (lll_robust_cond_lock): Likewise.
> (lll_timedlock): Likewise.
> (lll_robust_timedlock): Likewise.
> (lll_unlock): Likewise.
> (lll_robust_unlock): Likewise.
>
> ---
>
> diff --git a/nptl/sysdeps/unix/sysv/linux/powerpc/lowlevellock.h b/nptl/sysdeps/unix/sysv/linux/powerpc/lowlevellock.h
> index ab92c3f..38529a4 100644
> --- a/nptl/sysdeps/unix/sysv/linux/powerpc/lowlevellock.h
> +++ b/nptl/sysdeps/unix/sysv/linux/powerpc/lowlevellock.h
> @@ -76,6 +76,16 @@
> # endif
> #endif
>
> +/* For internal libc.so lock calls in single-thread process we use normal
> + load/stores. */
> +#if !defined NOT_IN_libc || defined UP
Where does this UP define come from? It seems to date back to the
early days of NPTL but I don't know when it gets set.
> +# define __lll_is_single_thread \
> + __glibc_likely (THREAD_GETMEM (THREAD_SELF, header.multiple_threads) == 0)
> +#else
> +# define __lll_is_single_thread (0)
> +#endif
> +
> +
> #define lll_futex_wait(futexp, val, private) \
> lll_futex_timed_wait (futexp, val, NULL, private)
>
> @@ -205,7 +215,9 @@
> /* Set *futex to ID if it is 0, atomically. Returns the old value */
> #define __lll_robust_trylock(futex, id) \
> ({ int __val; \
> - __asm __volatile ("1: lwarx %0,0,%2" MUTEX_HINT_ACQ "\n" \
> + if (!__lll_is_single_thread) \
> + __asm __volatile ( \
> + "1: lwarx %0,0,%2" MUTEX_HINT_ACQ "\n" \
> " cmpwi 0,%0,0\n" \
> " bne 2f\n" \
> " stwcx. %3,0,%2\n" \
> @@ -214,6 +226,12 @@
> : "=&r" (__val), "=m" (*futex) \
> : "r" (futex), "r" (id), "m" (*futex) \
> : "cr0", "memory"); \
> + else \
> + { \
> + __val = *futex; \
> + if (__val == 0) \
> + *futex = id; \
> + } \
> __val; \
> })
>
> @@ -237,8 +255,16 @@ extern int __lll_robust_lock_wait (int *futex, int private) attribute_hidden;
> #define lll_lock(lock, private) \
> (void) ({ \
> int *__futex = &(lock); \
> - if (__builtin_expect (atomic_compare_and_exchange_val_acq (__futex, 1, 0),\
> - 0) != 0) \
> + int __tmp; \
> + if (!__lll_is_single_thread) \
> + __tmp = atomic_compare_and_exchange_val_acq (__futex, 1, 0); \
> + else \
> + { \
> + __tmp = *__futex; \
> + if (__tmp == 0) \
> + *__futex = 1; \
> + } \
> + if (__builtin_expect (__tmp, 0) != 0) \
Should this be __glibc_unlikely? e.g.:
__glibc_unlikely(__tmp != 0)
> { \
> if (__builtin_constant_p (private) && (private) == LLL_PRIVATE) \
> __lll_lock_wait_private (__futex); \
> @@ -251,8 +277,16 @@ extern int __lll_robust_lock_wait (int *futex, int private) attribute_hidden;
> ({ \
> int *__futex = &(lock); \
> int __val = 0; \
> - if (__builtin_expect (atomic_compare_and_exchange_bool_acq (__futex, id, \
> - 0), 0)) \
> + int __tmp; \
> + if (!__lll_is_single_thread) \
> + __tmp = atomic_compare_and_exchange_bool_acq (__futex, id, 0); \
> + else \
> + { \
> + __tmp = (*__futex == 0); \
> + if (__tmp) \
> + *__futex = id; \
> + } \
> + if (__builtin_expect (__tmp, 0)) \
Likewise (and more places below).
> __val = __lll_robust_lock_wait (__futex, private); \
> __val; \
> })
> @@ -260,8 +294,16 @@ extern int __lll_robust_lock_wait (int *futex, int private) attribute_hidden;
> #define lll_cond_lock(lock, private) \
> (void) ({ \
> int *__futex = &(lock); \
> - if (__builtin_expect (atomic_compare_and_exchange_val_acq (__futex, 2, 0),\
> - 0) != 0) \
> + int __tmp; \
> + if (!__lll_is_single_thread) \
> + __tmp = atomic_compare_and_exchange_val_acq (__futex, 2, 0); \
> + else \
> + { \
> + __tmp = *__futex; \
> + if (__tmp == 0) \
> + *__futex = 2; \
> + } \
> + if (__builtin_expect (__tmp, 0) != 0) \
> __lll_lock_wait (__futex, private); \
> })
>
> @@ -269,9 +311,17 @@ extern int __lll_robust_lock_wait (int *futex, int private) attribute_hidden;
> ({ \
> int *__futex = &(lock); \
> int __val = 0; \
> + int __tmp; \
> int __id = id | FUTEX_WAITERS; \
> - if (__builtin_expect (atomic_compare_and_exchange_bool_acq (__futex, __id,\
> - 0), 0)) \
> + if (!__lll_is_single_thread) \
> + __tmp = atomic_compare_and_exchange_bool_acq (__futex, __id, 0); \
> + else \
> + { \
> + __tmp = (*__futex == 0); \
> + if (__tmp) \
> + *__futex = id; \
> + } \
> + if (__builtin_expect (__tmp, 0)) \
> __val = __lll_robust_lock_wait (__futex, private); \
> __val; \
> })
> @@ -286,8 +336,16 @@ extern int __lll_robust_timedlock_wait
> ({ \
> int *__futex = &(lock); \
> int __val = 0; \
> - if (__builtin_expect (atomic_compare_and_exchange_val_acq (__futex, 1, 0),\
> - 0) != 0) \
> + int __tmp; \
> + if (!__lll_is_single_thread) \
> + __tmp = atomic_compare_and_exchange_val_acq (__futex, 1, 0); \
> + else \
> + { \
> + __tmp = *__futex; \
> + if (__tmp == 0) \
> + *__futex = 1; \
> + } \
> + if (__builtin_expect (__tmp, 0) != 0) \
> __val = __lll_timedlock_wait (__futex, abstime, private); \
> __val; \
> })
> @@ -296,8 +354,16 @@ extern int __lll_robust_timedlock_wait
> ({ \
> int *__futex = &(lock); \
> int __val = 0; \
> - if (__builtin_expect (atomic_compare_and_exchange_bool_acq (__futex, id, \
> - 0), 0)) \
> + int __tmp; \
> + if (!__lll_is_single_thread) \
> + __tmp = atomic_compare_and_exchange_bool_acq (__futex, id, 0); \
> + else \
> + { \
> + __tmp = (*__futex == 0); \
> + if (__tmp) \
> + *__futex = id; \
> + } \
> + if (__builtin_expect (__tmp, 0)) \
> __val = __lll_robust_timedlock_wait (__futex, abstime, private); \
> __val; \
> })
> @@ -305,7 +371,14 @@ extern int __lll_robust_timedlock_wait
> #define lll_unlock(lock, private) \
> ((void) ({ \
> int *__futex = &(lock); \
> - int __val = atomic_exchange_rel (__futex, 0); \
> + int __val; \
> + if (!__lll_is_single_thread) \
> + __val = atomic_exchange_rel (__futex, 0); \
> + else \
> + { \
> + __val = *__futex; \
> + *__futex = 0; \
> + } \
> if (__glibc_unlikely (__val > 1)) \
> lll_futex_wake (__futex, 1, private); \
> }))
> @@ -313,7 +386,14 @@ extern int __lll_robust_timedlock_wait
> #define lll_robust_unlock(lock, private) \
> ((void) ({ \
> int *__futex = &(lock); \
> - int __val = atomic_exchange_rel (__futex, 0); \
> + int __val; \
> + if (!__lll_is_single_thread) \
> + __val = atomic_exchange_rel (__futex, 0); \
> + else \
> + { \
> + __val = *__futex; \
> + *__futex = 0; \
> + } \
> if (__glibc_unlikely (__val & FUTEX_WAITERS)) \
> lll_futex_wake (__futex, 1, private); \
> }))
> --
> 1.8.4.2
>
--
Will Newton
Toolchain Working Group, Linaro