diff options
author | Yann Ylavic <ylavic@apache.org> | 2021-10-29 15:04:55 +0000 |
---|---|---|
committer | Yann Ylavic <ylavic@apache.org> | 2021-10-29 15:04:55 +0000 |
commit | 524ff11d3c7cc5c35ce6c6568cbeb3c9309dc482 (patch) | |
tree | 25b4113ed609457e9e3cf87d6ae719853868d253 /atomic | |
parent | abc1e1784d67dd93ac7f80fb215d98b2676f9f7f (diff) | |
download | apr-524ff11d3c7cc5c35ce6c6568cbeb3c9309dc482.tar.gz |
apr_atomic: Use __atomic builtins when available.
Unlike Intel's atomic builtins (__sync_*), the more recent __atomic builtins
provide atomic load and store for weakly ordered architectures like ARM32 or
powerpc[64], so use them when available (gcc 4.6.3+).
git-svn-id: https://svn.apache.org/repos/asf/apr/apr/trunk@1894618 13f79535-47bb-0310-9956-ffa450edef68
Diffstat (limited to 'atomic')
-rw-r--r-- | atomic/unix/builtins.c | 56 | ||||
-rw-r--r-- | atomic/unix/builtins64.c | 38 | ||||
-rw-r--r-- | atomic/unix/mutex.c | 2 | ||||
-rw-r--r-- | atomic/unix/ppc.c | 63 |
4 files changed, 131 insertions, 28 deletions
diff --git a/atomic/unix/builtins.c b/atomic/unix/builtins.c index ebf8833d1..d0f1b454c 100644 --- a/atomic/unix/builtins.c +++ b/atomic/unix/builtins.c @@ -25,57 +25,97 @@ APR_DECLARE(apr_status_t) apr_atomic_init(apr_pool_t *p) APR_DECLARE(apr_uint32_t) apr_atomic_read32(volatile apr_uint32_t *mem) { +#if HAVE__ATOMIC_BUILTINS + return __atomic_load_n(mem, __ATOMIC_SEQ_CST); +#else return *mem; +#endif } APR_DECLARE(void) apr_atomic_set32(volatile apr_uint32_t *mem, apr_uint32_t val) { +#if HAVE__ATOMIC_BUILTINS + __atomic_store_n(mem, val, __ATOMIC_SEQ_CST); +#else *mem = val; +#endif } APR_DECLARE(apr_uint32_t) apr_atomic_add32(volatile apr_uint32_t *mem, apr_uint32_t val) { +#if HAVE__ATOMIC_BUILTINS + return __atomic_fetch_add(mem, val, __ATOMIC_SEQ_CST); +#else return __sync_fetch_and_add(mem, val); +#endif } APR_DECLARE(void) apr_atomic_sub32(volatile apr_uint32_t *mem, apr_uint32_t val) { +#if HAVE__ATOMIC_BUILTINS + __atomic_fetch_sub(mem, val, __ATOMIC_SEQ_CST); +#else __sync_fetch_and_sub(mem, val); +#endif } APR_DECLARE(apr_uint32_t) apr_atomic_inc32(volatile apr_uint32_t *mem) { +#if HAVE__ATOMIC_BUILTINS + return __atomic_fetch_add(mem, 1, __ATOMIC_SEQ_CST); +#else return __sync_fetch_and_add(mem, 1); +#endif } APR_DECLARE(int) apr_atomic_dec32(volatile apr_uint32_t *mem) { +#if HAVE__ATOMIC_BUILTINS + return __atomic_sub_fetch(mem, 1, __ATOMIC_SEQ_CST); +#else return __sync_sub_and_fetch(mem, 1); +#endif } -APR_DECLARE(apr_uint32_t) apr_atomic_cas32(volatile apr_uint32_t *mem, apr_uint32_t with, +APR_DECLARE(apr_uint32_t) apr_atomic_cas32(volatile apr_uint32_t *mem, apr_uint32_t val, apr_uint32_t cmp) { - return __sync_val_compare_and_swap(mem, cmp, with); +#if HAVE__ATOMIC_BUILTINS + __atomic_compare_exchange_n(mem, &cmp, val, 0, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST); + return cmp; +#else + return __sync_val_compare_and_swap(mem, cmp, val); +#endif } APR_DECLARE(apr_uint32_t) apr_atomic_xchg32(volatile apr_uint32_t *mem, apr_uint32_t val) { +#if HAVE__ATOMIC_BUILTINS + return __atomic_exchange_n(mem, val, __ATOMIC_SEQ_CST); +#else __sync_synchronize(); - return __sync_lock_test_and_set(mem, val); +#endif } -APR_DECLARE(void*) apr_atomic_casptr(void *volatile *mem, void *with, const void *cmp) +APR_DECLARE(void*) apr_atomic_casptr(void *volatile *mem, void *ptr, const void *cmp) { - return (void*) __sync_val_compare_and_swap(mem, cmp, with); +#if HAVE__ATOMIC_BUILTINS + __atomic_compare_exchange_n(mem, (void **)&cmp, ptr, 0, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST); + return (void *)cmp; +#else + return (void *)__sync_val_compare_and_swap(mem, (void *)cmp, ptr); +#endif } -APR_DECLARE(void*) apr_atomic_xchgptr(void *volatile *mem, void *with) +APR_DECLARE(void*) apr_atomic_xchgptr(void *volatile *mem, void *ptr) { +#if HAVE__ATOMIC_BUILTINS + return __atomic_exchange_n(mem, ptr, __ATOMIC_SEQ_CST); +#else __sync_synchronize(); - - return (void*) __sync_lock_test_and_set(mem, with); + return __sync_lock_test_and_set(mem, ptr); +#endif } #endif /* USE_ATOMICS_BUILTINS */ diff --git a/atomic/unix/builtins64.c b/atomic/unix/builtins64.c index 4a4b685c7..0ac950c15 100644 --- a/atomic/unix/builtins64.c +++ b/atomic/unix/builtins64.c @@ -20,45 +20,77 @@ APR_DECLARE(apr_uint64_t) apr_atomic_read64(volatile apr_uint64_t *mem) { +#if HAVE__ATOMIC_BUILTINS + return __atomic_load_n(mem, __ATOMIC_SEQ_CST); +#else return *mem; +#endif } APR_DECLARE(void) apr_atomic_set64(volatile apr_uint64_t *mem, apr_uint64_t val) { +#if HAVE__ATOMIC_BUILTINS + __atomic_store_n(mem, val, __ATOMIC_SEQ_CST); +#else *mem = val; +#endif } APR_DECLARE(apr_uint64_t) apr_atomic_add64(volatile apr_uint64_t *mem, apr_uint64_t val) { +#if HAVE__ATOMIC_BUILTINS + return __atomic_fetch_add(mem, val, __ATOMIC_SEQ_CST); +#else return __sync_fetch_and_add(mem, val); +#endif } APR_DECLARE(void) apr_atomic_sub64(volatile apr_uint64_t *mem, apr_uint64_t val) { +#if HAVE__ATOMIC_BUILTINS + __atomic_fetch_sub(mem, val, __ATOMIC_SEQ_CST); +#else __sync_fetch_and_sub(mem, val); +#endif } APR_DECLARE(apr_uint64_t) apr_atomic_inc64(volatile apr_uint64_t *mem) { +#if HAVE__ATOMIC_BUILTINS + return __atomic_fetch_add(mem, 1, __ATOMIC_SEQ_CST); +#else return __sync_fetch_and_add(mem, 1); +#endif } APR_DECLARE(int) apr_atomic_dec64(volatile apr_uint64_t *mem) { +#if HAVE__ATOMIC_BUILTINS + return __atomic_sub_fetch(mem, 1, __ATOMIC_SEQ_CST); +#else return __sync_sub_and_fetch(mem, 1); +#endif } -APR_DECLARE(apr_uint64_t) apr_atomic_cas64(volatile apr_uint64_t *mem, apr_uint64_t with, +APR_DECLARE(apr_uint64_t) apr_atomic_cas64(volatile apr_uint64_t *mem, apr_uint64_t val, apr_uint64_t cmp) { - return __sync_val_compare_and_swap(mem, cmp, with); +#if HAVE__ATOMIC_BUILTINS + __atomic_compare_exchange_n(mem, &cmp, val, 0, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST); + return cmp; +#else + return __sync_val_compare_and_swap(mem, cmp, val); +#endif } APR_DECLARE(apr_uint64_t) apr_atomic_xchg64(volatile apr_uint64_t *mem, apr_uint64_t val) { +#if HAVE__ATOMIC_BUILTINS + return __atomic_exchange_n(mem, val, __ATOMIC_SEQ_CST); +#else __sync_synchronize(); - return __sync_lock_test_and_set(mem, val); +#endif } #endif /* USE_ATOMICS_BUILTINS */ diff --git a/atomic/unix/mutex.c b/atomic/unix/mutex.c index 78ad75336..b9d1cedc9 100644 --- a/atomic/unix/mutex.c +++ b/atomic/unix/mutex.c @@ -89,7 +89,7 @@ static APR_INLINE apr_thread_mutex_t *mutex_hash(volatile apr_uint32_t *mem) APR_DECLARE(apr_status_t) apr_atomic_init(apr_pool_t *p) { - return apr__atomic_generic64_init(p); + return APR_SUCCESS; } #endif /* APR_HAS_THREADS */ diff --git a/atomic/unix/ppc.c b/atomic/unix/ppc.c index 55bbdd50c..ffba3c27a 100644 --- a/atomic/unix/ppc.c +++ b/atomic/unix/ppc.c @@ -35,24 +35,39 @@ APR_DECLARE(apr_status_t) apr_atomic_init(apr_pool_t *p) APR_DECLARE(apr_uint32_t) apr_atomic_read32(volatile apr_uint32_t *mem) { - return *mem; + apr_uint32_t val; + asm volatile (" sync\n" /* full barrier */ + " lwz %0,%1\n" /* load */ + " cmpw 7,%0,%0\n" /* compare (always equal) */ + " bne- 7,$+4\n" /* goto next in any case */ + " isync" /* acquire barrier (bc+isync) */ + : "=r"(val) + : "m"(*mem) + : "cc", "memory"); + return val; } APR_DECLARE(void) apr_atomic_set32(volatile apr_uint32_t *mem, apr_uint32_t val) { - *mem = val; + asm volatile (" sync\n" /* full barrier */ + " stw %1,%0" /* store */ + : "=m"(*mem) + : "r"(val) + : "memory"); } APR_DECLARE(apr_uint32_t) apr_atomic_add32(volatile apr_uint32_t *mem, apr_uint32_t val) { apr_uint32_t prev, temp; - asm volatile ("1:\n" /* lost reservation */ + asm volatile (" sync\n" /* full barrier */ + "1:\n" /* lost reservation */ " lwarx %0,0,%3\n" /* load and reserve */ " add %1,%0,%4\n" /* add val and prev */ PPC405_ERR77_SYNC /* ppc405 Erratum 77 */ - " stwcx. %1,0,%3\n" /* store new value */ + " stwcx. %1,0,%3\n" /* store if still reserved */ " bne- 1b\n" /* loop if lost */ + " isync\n" /* acquire barrier (bc+isync) */ : "=&r" (prev), "=&r" (temp), "=m" (*mem) : "b" (mem), "r" (val) : "cc", "memory"); @@ -64,12 +79,14 @@ APR_DECLARE(void) apr_atomic_sub32(volatile apr_uint32_t *mem, apr_uint32_t val) { apr_uint32_t temp; - asm volatile ("1:\n" /* lost reservation */ + asm volatile (" sync\n" /* full barrier */ + "1:\n" /* lost reservation */ " lwarx %0,0,%2\n" /* load and reserve */ " subf %0,%3,%0\n" /* subtract val */ PPC405_ERR77_SYNC /* ppc405 Erratum 77 */ " stwcx. %0,0,%2\n" /* store new value */ " bne- 1b\n" /* loop if lost */ + " isync\n" /* acquire barrier (bc+isync) */ : "=&r" (temp), "=m" (*mem) : "b" (mem), "r" (val) : "cc", "memory"); @@ -79,13 +96,15 @@ APR_DECLARE(apr_uint32_t) apr_atomic_inc32(volatile apr_uint32_t *mem) { apr_uint32_t prev; - asm volatile ("1:\n" /* lost reservation */ + asm volatile (" sync\n" /* full barrier */ + "1:\n" /* lost reservation */ " lwarx %0,0,%2\n" /* load and reserve */ " addi %0,%0,1\n" /* add immediate */ PPC405_ERR77_SYNC /* ppc405 Erratum 77 */ " stwcx. %0,0,%2\n" /* store new value */ " bne- 1b\n" /* loop if lost */ " subi %0,%0,1\n" /* return old value */ + " isync\n" /* acquire barrier (bc+isync) */ : "=&b" (prev), "=m" (*mem) : "b" (mem), "m" (*mem) : "cc", "memory"); @@ -97,12 +116,14 @@ APR_DECLARE(int) apr_atomic_dec32(volatile apr_uint32_t *mem) { apr_uint32_t prev; - asm volatile ("1:\n" /* lost reservation */ + asm volatile (" sync\n" /* full barrier */ + "1:\n" /* lost reservation */ " lwarx %0,0,%2\n" /* load and reserve */ " subi %0,%0,1\n" /* subtract immediate */ PPC405_ERR77_SYNC /* ppc405 Erratum 77 */ " stwcx. %0,0,%2\n" /* store new value */ " bne- 1b\n" /* loop if lost */ + " isync\n" /* acquire barrier (bc+isync) */ : "=&b" (prev), "=m" (*mem) : "b" (mem), "m" (*mem) : "cc", "memory"); @@ -115,7 +136,8 @@ APR_DECLARE(apr_uint32_t) apr_atomic_cas32(volatile apr_uint32_t *mem, apr_uint3 { apr_uint32_t prev; - asm volatile ("1:\n" /* lost reservation */ + asm volatile (" sync\n" /* full barrier */ + "1:\n" /* lost reservation */ " lwarx %0,0,%1\n" /* load and reserve */ " cmpw %0,%3\n" /* compare operands */ " bne- exit_%=\n" /* skip if not equal */ @@ -123,6 +145,7 @@ APR_DECLARE(apr_uint32_t) apr_atomic_cas32(volatile apr_uint32_t *mem, apr_uint3 " stwcx. %2,0,%1\n" /* store new value */ " bne- 1b\n" /* loop if lost */ "exit_%=:\n" /* not equal */ + " isync\n" /* acquire barrier (bc+isync) */ : "=&r" (prev) : "b" (mem), "r" (with), "r" (cmp) : "cc", "memory"); @@ -134,11 +157,13 @@ APR_DECLARE(apr_uint32_t) apr_atomic_xchg32(volatile apr_uint32_t *mem, apr_uint { apr_uint32_t prev; - asm volatile ("1:\n" /* lost reservation */ + asm volatile (" sync\n" /* full barrier */ + "1:\n" /* lost reservation */ " lwarx %0,0,%1\n" /* load and reserve */ PPC405_ERR77_SYNC /* ppc405 Erratum 77 */ " stwcx. %2,0,%1\n" /* store new value */ - " bne- 1b" /* loop if lost */ + " bne- 1b\n" /* loop if lost */ + " isync\n" /* acquire barrier (bc+isync) */ : "=&r" (prev) : "b" (mem), "r" (val) : "cc", "memory"); @@ -150,7 +175,8 @@ APR_DECLARE(void*) apr_atomic_casptr(void *volatile *mem, void *with, const void { void *prev; #if APR_SIZEOF_VOIDP == 4 - asm volatile ("1:\n" /* lost reservation */ + asm volatile (" sync\n" /* full barrier */ + "1:\n" /* lost reservation */ " lwarx %0,0,%1\n" /* load and reserve */ " cmpw %0,%3\n" /* compare operands */ " bne- 2f\n" /* skip if not equal */ @@ -158,11 +184,13 @@ APR_DECLARE(void*) apr_atomic_casptr(void *volatile *mem, void *with, const void " stwcx. %2,0,%1\n" /* store new value */ " bne- 1b\n" /* loop if lost */ "2:\n" /* not equal */ + " isync\n" /* acquire barrier (bc+isync) */ : "=&r" (prev) : "b" (mem), "r" (with), "r" (cmp) : "cc", "memory"); #elif APR_SIZEOF_VOIDP == 8 - asm volatile ("1:\n" /* lost reservation */ + asm volatile (" sync\n" /* full barrier */ + "1:\n" /* lost reservation */ " ldarx %0,0,%1\n" /* load and reserve */ " cmpd %0,%3\n" /* compare operands */ " bne- 2f\n" /* skip if not equal */ @@ -170,6 +198,7 @@ APR_DECLARE(void*) apr_atomic_casptr(void *volatile *mem, void *with, const void " stdcx. %2,0,%1\n" /* store new value */ " bne- 1b\n" /* loop if lost */ "2:\n" /* not equal */ + " isync\n" /* acquire barrier (bc+isync) */ : "=&r" (prev) : "b" (mem), "r" (with), "r" (cmp) : "cc", "memory"); @@ -183,22 +212,24 @@ APR_DECLARE(void*) apr_atomic_xchgptr(void *volatile *mem, void *with) { void *prev; #if APR_SIZEOF_VOIDP == 4 - asm volatile ("1:\n" /* lost reservation */ + asm volatile (" sync\n" /* full barrier */ + "1:\n" /* lost reservation */ " lwarx %0,0,%1\n" /* load and reserve */ PPC405_ERR77_SYNC /* ppc405 Erratum 77 */ " stwcx. %2,0,%1\n" /* store new value */ " bne- 1b\n" /* loop if lost */ - " isync\n" /* memory barrier */ + " isync\n" /* acquire barrier (bc+isync) */ : "=&r" (prev) : "b" (mem), "r" (with) : "cc", "memory"); #elif APR_SIZEOF_VOIDP == 8 - asm volatile ("1:\n" /* lost reservation */ + asm volatile (" sync\n" /* full barrier */ + "1:\n" /* lost reservation */ " ldarx %0,0,%1\n" /* load and reserve */ PPC405_ERR77_SYNC /* ppc405 Erratum 77 */ " stdcx. %2,0,%1\n" /* store new value */ " bne- 1b\n" /* loop if lost */ - " isync\n" /* memory barrier */ + " isync\n" /* acquire barrier (bc+isync) */ : "=&r" (prev) : "b" (mem), "r" (with) : "cc", "memory"); |