X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Futil%2Fu_atomic.h;h=db56835e9d4f9ba6ed219770849ed1375e624b83;hb=ed3f1e04c7d2b5c17885919db5fde310d4bdf822;hp=192cc8d1e0e283ef76f215fccc3225ea7268044a;hpb=1ba9f9e62c1e83a2a9a48435cce5128be3d3ea66;p=mesa.git diff --git a/src/util/u_atomic.h b/src/util/u_atomic.h index 192cc8d1e0e..db56835e9d4 100644 --- a/src/util/u_atomic.h +++ b/src/util/u_atomic.h @@ -6,10 +6,13 @@ * */ +#include "no_extern_c.h" + #ifndef U_ATOMIC_H #define U_ATOMIC_H #include +#include /* Favor OS-provided implementations. * @@ -34,6 +37,23 @@ #define PIPE_ATOMIC "GCC Sync Intrinsics" +#if defined(USE_GCC_ATOMIC_BUILTINS) + +/* The builtins with explicit memory model are available since GCC 4.7. */ +#define p_atomic_set(_v, _i) __atomic_store_n((_v), (_i), __ATOMIC_RELEASE) +#define p_atomic_read(_v) __atomic_load_n((_v), __ATOMIC_ACQUIRE) +#define p_atomic_dec_zero(v) (__atomic_sub_fetch((v), 1, __ATOMIC_ACQ_REL) == 0) +#define p_atomic_inc(v) (void) __atomic_add_fetch((v), 1, __ATOMIC_ACQ_REL) +#define p_atomic_dec(v) (void) __atomic_sub_fetch((v), 1, __ATOMIC_ACQ_REL) +#define p_atomic_add(v, i) (void) __atomic_add_fetch((v), (i), __ATOMIC_ACQ_REL) +#define p_atomic_inc_return(v) __atomic_add_fetch((v), 1, __ATOMIC_ACQ_REL) +#define p_atomic_dec_return(v) __atomic_sub_fetch((v), 1, __ATOMIC_ACQ_REL) +#define p_atomic_add_return(v, i) __atomic_add_fetch((v), (i), __ATOMIC_ACQ_REL) +#define p_atomic_xchg(v, i) __atomic_exchange_n((v), (i), __ATOMIC_ACQ_REL) +#define PIPE_NATIVE_ATOMIC_XCHG + +#else + #define p_atomic_set(_v, _i) (*(_v) = (_i)) #define p_atomic_read(_v) (*(_v)) #define p_atomic_dec_zero(v) (__sync_sub_and_fetch((v), 1) == 0) @@ -42,6 +62,14 @@ #define p_atomic_add(v, i) (void) __sync_add_and_fetch((v), (i)) #define p_atomic_inc_return(v) __sync_add_and_fetch((v), 1) #define p_atomic_dec_return(v) __sync_sub_and_fetch((v), 1) +#define p_atomic_add_return(v, i) __sync_add_and_fetch((v), (i)) + +#endif + +/* There is no __atomic_* compare and exchange that returns the current value. + * Also, GCC 5.4 seems unable to optimize a compound statement expression that + * uses an additional stack variable with __atomic_compare_exchange[_n]. + */ #define p_atomic_cmpxchg(v, old, _new) \ __sync_val_compare_and_swap((v), (old), (_new)) @@ -61,9 +89,10 @@ #define p_atomic_dec_zero(_v) (p_atomic_dec_return(_v) == 0) #define p_atomic_inc(_v) ((void) p_atomic_inc_return(_v)) #define p_atomic_dec(_v) ((void) p_atomic_dec_return(_v)) -#define p_atomic_add(_v, _i) (*(_v) = *(_v) + (_i)) +#define p_atomic_add(_v, _i) ((void) p_atomic_add_return((_v), (_i))) #define p_atomic_inc_return(_v) (++(*(_v))) #define p_atomic_dec_return(_v) (--(*(_v))) +#define p_atomic_add_return(_v, _i) (*(_v) = *(_v) + (_i)) #define p_atomic_cmpxchg(_v, _old, _new) (*(_v) == (_old) ? (*(_v) = (_new), (_old)) : *(_v)) #endif @@ -86,65 +115,6 @@ #include #include -#if _MSC_VER < 1600 - -/* Implement _InterlockedCompareExchange8 in terms of _InterlockedCompareExchange16 */ -static __inline char -_InterlockedCompareExchange8(char volatile *destination8, char exchange8, char comparand8) -{ - INT_PTR destinationAddr = (INT_PTR)destination8; - short volatile *destination16 = (short volatile *)(destinationAddr & ~1); - const short shift8 = (destinationAddr & 1) * 8; - const short mask8 = 0xff << shift8; - short initial16 = *destination16; - char initial8 = initial16 >> shift8; - while (initial8 == comparand8) { - /* initial *destination8 matches, so try exchange it while keeping the - * neighboring byte untouched */ - short exchange16 = (initial16 & ~mask8) | ((short)exchange8 << shift8); - short comparand16 = initial16; - short initial16 = _InterlockedCompareExchange16(destination16, exchange16, comparand16); - if (initial16 == comparand16) { - /* succeeded */ - return comparand8; - } - /* something changed, retry with the new initial value */ - initial8 = initial16 >> shift8; - } - return initial8; -} - -/* Implement _InterlockedExchangeAdd16 in terms of _InterlockedCompareExchange16 */ -static __inline short -_InterlockedExchangeAdd16(short volatile *addend, short value) -{ - short initial = *addend; - short comparand; - do { - short exchange = initial + value; - comparand = initial; - /* if *addend==comparand then *addend=exchange, return original *addend */ - initial = _InterlockedCompareExchange16(addend, exchange, comparand); - } while(initial != comparand); - return comparand; -} - -/* Implement _InterlockedExchangeAdd8 in terms of _InterlockedCompareExchange8 */ -static __inline char -_InterlockedExchangeAdd8(char volatile *addend, char value) -{ - char initial = *addend; - char comparand; - do { - char exchange = initial + value; - comparand = initial; - initial = _InterlockedCompareExchange8(addend, exchange, comparand); - } while(initial != comparand); - return comparand; -} - -#endif /* _MSC_VER < 1600 */ - /* MSVC supports decltype keyword, but it's only supported on C++ and doesn't * quite work here; and if a C++-only solution is worthwhile, then it would be * better to use templates / function overloading, instead of decltype magic. @@ -175,7 +145,10 @@ _InterlockedExchangeAdd8(char volatile *addend, char value) sizeof *(_v) == sizeof(__int64) ? InterlockedDecrement64 ((__int64 *)(_v)) : \ (assert(!"should not get here"), 0)) -#define p_atomic_add(_v, _i) (\ +#define p_atomic_add(_v, _i) \ + ((void) p_atomic_add_return((_v), (_i))) + +#define p_atomic_add_return(_v, _i) (\ sizeof *(_v) == sizeof(char) ? _InterlockedExchangeAdd8 ((char *) (_v), (_i)) : \ sizeof *(_v) == sizeof(short) ? _InterlockedExchangeAdd16((short *) (_v), (_i)) : \ sizeof *(_v) == sizeof(long) ? _InterlockedExchangeAdd ((long *) (_v), (_i)) : \ @@ -215,35 +188,42 @@ _InterlockedExchangeAdd8(char volatile *addend, char value) sizeof(*v) == sizeof(uint64_t) ? atomic_inc_64((uint64_t *)(v)) : \ (assert(!"should not get here"), 0)) -#define p_atomic_inc_return(v) ((typeof(*v)) \ +#define p_atomic_inc_return(v) (__typeof(*v))( \ sizeof(*v) == sizeof(uint8_t) ? atomic_inc_8_nv ((uint8_t *)(v)) : \ sizeof(*v) == sizeof(uint16_t) ? atomic_inc_16_nv((uint16_t *)(v)) : \ sizeof(*v) == sizeof(uint32_t) ? atomic_inc_32_nv((uint32_t *)(v)) : \ sizeof(*v) == sizeof(uint64_t) ? atomic_inc_64_nv((uint64_t *)(v)) : \ (assert(!"should not get here"), 0)) -#define p_atomic_dec(v) ((void) \ +#define p_atomic_dec(v) (void) ( \ sizeof(*v) == sizeof(uint8_t) ? atomic_dec_8 ((uint8_t *)(v)) : \ sizeof(*v) == sizeof(uint16_t) ? atomic_dec_16((uint16_t *)(v)) : \ sizeof(*v) == sizeof(uint32_t) ? atomic_dec_32((uint32_t *)(v)) : \ sizeof(*v) == sizeof(uint64_t) ? atomic_dec_64((uint64_t *)(v)) : \ (assert(!"should not get here"), 0)) -#define p_atomic_dec_return(v) ((typeof(*v)) \ +#define p_atomic_dec_return(v) (__typeof(*v))( \ sizeof(*v) == sizeof(uint8_t) ? atomic_dec_8_nv ((uint8_t *)(v)) : \ sizeof(*v) == sizeof(uint16_t) ? atomic_dec_16_nv((uint16_t *)(v)) : \ sizeof(*v) == sizeof(uint32_t) ? atomic_dec_32_nv((uint32_t *)(v)) : \ sizeof(*v) == sizeof(uint64_t) ? atomic_dec_64_nv((uint64_t *)(v)) : \ (assert(!"should not get here"), 0)) -#define p_atomic_add(v, i) ((void) \ +#define p_atomic_add(v, i) (void) ( \ sizeof(*v) == sizeof(uint8_t) ? atomic_add_8 ((uint8_t *)(v), (i)) : \ sizeof(*v) == sizeof(uint16_t) ? atomic_add_16((uint16_t *)(v), (i)) : \ sizeof(*v) == sizeof(uint32_t) ? atomic_add_32((uint32_t *)(v), (i)) : \ sizeof(*v) == sizeof(uint64_t) ? atomic_add_64((uint64_t *)(v), (i)) : \ (assert(!"should not get here"), 0)) -#define p_atomic_cmpxchg(v, old, _new) ((typeof(*v)) \ +#define p_atomic_add_return(v, i) (__typeof(*v)) ( \ + sizeof(*v) == sizeof(uint8_t) ? atomic_add_8_nv ((uint8_t *)(v), (i)) : \ + sizeof(*v) == sizeof(uint16_t) ? atomic_add_16_nv((uint16_t *)(v), (i)) : \ + sizeof(*v) == sizeof(uint32_t) ? atomic_add_32_nv((uint32_t *)(v), (i)) : \ + sizeof(*v) == sizeof(uint64_t) ? atomic_add_64_nv((uint64_t *)(v), (i)) : \ + (assert(!"should not get here"), 0)) + +#define p_atomic_cmpxchg(v, old, _new) (__typeof(*v))( \ sizeof(*v) == sizeof(uint8_t) ? atomic_cas_8 ((uint8_t *)(v), (uint8_t )(old), (uint8_t )(_new)) : \ sizeof(*v) == sizeof(uint16_t) ? atomic_cas_16((uint16_t *)(v), (uint16_t)(old), (uint16_t)(_new)) : \ sizeof(*v) == sizeof(uint32_t) ? atomic_cas_32((uint32_t *)(v), (uint32_t)(old), (uint32_t)(_new)) : \ @@ -256,6 +236,33 @@ _InterlockedExchangeAdd8(char volatile *addend, char value) #error "No pipe_atomic implementation selected" #endif +#ifndef PIPE_NATIVE_ATOMIC_XCHG +static inline uint32_t p_atomic_xchg_32(uint32_t *v, uint32_t i) +{ + uint32_t actual = p_atomic_read(v); + uint32_t expected; + do { + expected = actual; + actual = p_atomic_cmpxchg(v, expected, i); + } while (expected != actual); + return actual; +} + +static inline uint64_t p_atomic_xchg_64(uint64_t *v, uint64_t i) +{ + uint64_t actual = p_atomic_read(v); + uint64_t expected; + do { + expected = actual; + actual = p_atomic_cmpxchg(v, expected, i); + } while (expected != actual); + return actual; +} +#define p_atomic_xchg(v, i) (__typeof(*(v)))( \ + sizeof(*(v)) == sizeof(uint32_t) ? p_atomic_xchg_32((uint32_t *)(v), (uint32_t)(i)) : \ + sizeof(*(v)) == sizeof(uint64_t) ? p_atomic_xchg_64((uint64_t *)(v), (uint64_t)(i)) : \ + (assert(!"should not get here"), 0)) +#endif #endif /* U_ATOMIC_H */