X-Git-Url: https://git.libre-soc.org/?p=mesa.git;a=blobdiff_plain;f=src%2Futil%2Fu_atomic.h;h=db56835e9d4f9ba6ed219770849ed1375e624b83;hp=cf7fff3f35b32609f4124b3e966623777b4fa8d9;hb=HEAD;hpb=fbc3e030e658c2c6d44bac084c3166d1fb737c7e diff --git a/src/util/u_atomic.h b/src/util/u_atomic.h index cf7fff3f35b..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. * @@ -34,13 +37,39 @@ #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) #define p_atomic_inc(v) (void) __sync_add_and_fetch((v), 1) #define p_atomic_dec(v) (void) __sync_sub_and_fetch((v), 1) +#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)) @@ -60,8 +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) ((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 @@ -84,36 +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; -} - -#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. @@ -144,6 +145,16 @@ char _InterlockedCompareExchange8(char volatile *Destination8, char Exchange8, c sizeof *(_v) == sizeof(__int64) ? InterlockedDecrement64 ((__int64 *)(_v)) : \ (assert(!"should not get here"), 0)) +#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)) : \ + sizeof *(_v) == sizeof(__int64) ? InterlockedExchangeAdd64((__int64 *)(_v), (_i)) : \ + (assert(!"should not get here"), 0)) + #define p_atomic_cmpxchg(_v, _old, _new) (\ sizeof *(_v) == sizeof(char) ? _InterlockedCompareExchange8 ((char *) (_v), (char) (_new), (char) (_old)) : \ sizeof *(_v) == sizeof(short) ? _InterlockedCompareExchange16((short *) (_v), (short) (_new), (short) (_old)) : \ @@ -177,28 +188,42 @@ char _InterlockedCompareExchange8(char volatile *Destination8, char Exchange8, c 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_cmpxchg(v, old, _new) ((typeof(*v)) \ +#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_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)) : \ @@ -211,6 +236,33 @@ char _InterlockedCompareExchange8(char volatile *Destination8, char Exchange8, c #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 */