mirror of
https://gitlab.freedesktop.org/mesa/mesa.git
synced 2026-04-05 16:50:37 +02:00
8915f0c0de
This is motivated by the fact that p_atomic_read and p_atomic_set may somewhat surprisingly not do the right thing in the old version: while stores and loads are de facto atomic at least on x86, the compiler may apply re-ordering and speculation quite liberally. Basically, the old version uses the "relaxed" memory ordering. The new ordering always uses acquire/release ordering. This is the strongest possible memory ordering that doesn't require additional fence instructions on x86. (And the only stronger ordering is "sequentially consistent", which is usually more than you need anyway.) I would feel more comfortable if p_atomic_set/read in the old implementation were at least using volatile loads and stores, but I don't see a way to get there without typeof (which we cannot use here since the code is compiled with -std=c99). Eventually, we should really just move to something that is based on the atomics in C11 / C++11. Acked-by: Marek Olšák <marek.olsak@amd.com> Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
225 lines
9 KiB
C
225 lines
9 KiB
C
/**
|
|
* Many similar implementations exist. See for example libwsbm
|
|
* or the linux kernel include/atomic.h
|
|
*
|
|
* No copyright claimed on this file.
|
|
*
|
|
*/
|
|
|
|
#include "no_extern_c.h"
|
|
|
|
#ifndef U_ATOMIC_H
|
|
#define U_ATOMIC_H
|
|
|
|
#include <stdbool.h>
|
|
|
|
/* Favor OS-provided implementations.
|
|
*
|
|
* Where no OS-provided implementation is available, fall back to
|
|
* locally coded assembly, compiler intrinsic or ultimately a
|
|
* mutex-based implementation.
|
|
*/
|
|
#if defined(__sun)
|
|
#define PIPE_ATOMIC_OS_SOLARIS
|
|
#elif defined(_MSC_VER)
|
|
#define PIPE_ATOMIC_MSVC_INTRINSIC
|
|
#elif defined(__GNUC__)
|
|
#define PIPE_ATOMIC_GCC_INTRINSIC
|
|
#else
|
|
#error "Unsupported platform"
|
|
#endif
|
|
|
|
|
|
/* Implementation using GCC-provided synchronization intrinsics
|
|
*/
|
|
#if defined(PIPE_ATOMIC_GCC_INTRINSIC)
|
|
|
|
#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)
|
|
|
|
#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)
|
|
|
|
#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))
|
|
|
|
#endif
|
|
|
|
|
|
|
|
/* Unlocked version for single threaded environments, such as some
|
|
* windows kernel modules.
|
|
*/
|
|
#if defined(PIPE_ATOMIC_OS_UNLOCKED)
|
|
|
|
#define PIPE_ATOMIC "Unlocked"
|
|
|
|
#define p_atomic_set(_v, _i) (*(_v) = (_i))
|
|
#define p_atomic_read(_v) (*(_v))
|
|
#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_inc_return(_v) (++(*(_v)))
|
|
#define p_atomic_dec_return(_v) (--(*(_v)))
|
|
#define p_atomic_cmpxchg(_v, _old, _new) (*(_v) == (_old) ? (*(_v) = (_new), (_old)) : *(_v))
|
|
|
|
#endif
|
|
|
|
|
|
#if defined(PIPE_ATOMIC_MSVC_INTRINSIC)
|
|
|
|
#define PIPE_ATOMIC "MSVC Intrinsics"
|
|
|
|
/* We use the Windows header's Interlocked*64 functions instead of the
|
|
* _Interlocked*64 intrinsics wherever we can, as support for the latter varies
|
|
* with target CPU, whereas Windows headers take care of all portability
|
|
* issues: using intrinsics where available, falling back to library
|
|
* implementations where not.
|
|
*/
|
|
#ifndef WIN32_LEAN_AND_MEAN
|
|
#define WIN32_LEAN_AND_MEAN 1
|
|
#endif
|
|
#include <windows.h>
|
|
#include <intrin.h>
|
|
#include <assert.h>
|
|
|
|
/* 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.
|
|
* Therefore, we rely on implicit casting to LONGLONG for the functions that return
|
|
*/
|
|
|
|
#define p_atomic_set(_v, _i) (*(_v) = (_i))
|
|
#define p_atomic_read(_v) (*(_v))
|
|
|
|
#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_inc_return(_v) (\
|
|
sizeof *(_v) == sizeof(short) ? _InterlockedIncrement16((short *) (_v)) : \
|
|
sizeof *(_v) == sizeof(long) ? _InterlockedIncrement ((long *) (_v)) : \
|
|
sizeof *(_v) == sizeof(__int64) ? InterlockedIncrement64 ((__int64 *)(_v)) : \
|
|
(assert(!"should not get here"), 0))
|
|
|
|
#define p_atomic_dec(_v) \
|
|
((void) p_atomic_dec_return(_v))
|
|
|
|
#define p_atomic_dec_return(_v) (\
|
|
sizeof *(_v) == sizeof(short) ? _InterlockedDecrement16((short *) (_v)) : \
|
|
sizeof *(_v) == sizeof(long) ? _InterlockedDecrement ((long *) (_v)) : \
|
|
sizeof *(_v) == sizeof(__int64) ? InterlockedDecrement64 ((__int64 *)(_v)) : \
|
|
(assert(!"should not get here"), 0))
|
|
|
|
#define p_atomic_add(_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)) : \
|
|
sizeof *(_v) == sizeof(long) ? _InterlockedCompareExchange ((long *) (_v), (long) (_new), (long) (_old)) : \
|
|
sizeof *(_v) == sizeof(__int64) ? InterlockedCompareExchange64 ((__int64 *)(_v), (__int64)(_new), (__int64)(_old)) : \
|
|
(assert(!"should not get here"), 0))
|
|
|
|
#endif
|
|
|
|
#if defined(PIPE_ATOMIC_OS_SOLARIS)
|
|
|
|
#define PIPE_ATOMIC "Solaris OS atomic functions"
|
|
|
|
#include <atomic.h>
|
|
#include <assert.h>
|
|
|
|
#define p_atomic_set(_v, _i) (*(_v) = (_i))
|
|
#define p_atomic_read(_v) (*(_v))
|
|
|
|
#define p_atomic_dec_zero(v) (\
|
|
sizeof(*v) == sizeof(uint8_t) ? atomic_dec_8_nv ((uint8_t *)(v)) == 0 : \
|
|
sizeof(*v) == sizeof(uint16_t) ? atomic_dec_16_nv((uint16_t *)(v)) == 0 : \
|
|
sizeof(*v) == sizeof(uint32_t) ? atomic_dec_32_nv((uint32_t *)(v)) == 0 : \
|
|
sizeof(*v) == sizeof(uint64_t) ? atomic_dec_64_nv((uint64_t *)(v)) == 0 : \
|
|
(assert(!"should not get here"), 0))
|
|
|
|
#define p_atomic_inc(v) (void) (\
|
|
sizeof(*v) == sizeof(uint8_t) ? atomic_inc_8 ((uint8_t *)(v)) : \
|
|
sizeof(*v) == sizeof(uint16_t) ? atomic_inc_16((uint16_t *)(v)) : \
|
|
sizeof(*v) == sizeof(uint32_t) ? atomic_inc_32((uint32_t *)(v)) : \
|
|
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)) \
|
|
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) \
|
|
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)) \
|
|
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) \
|
|
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)) \
|
|
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)) : \
|
|
sizeof(*v) == sizeof(uint64_t) ? atomic_cas_64((uint64_t *)(v), (uint64_t)(old), (uint64_t)(_new)) : \
|
|
(assert(!"should not get here"), 0))
|
|
|
|
#endif
|
|
|
|
#ifndef PIPE_ATOMIC
|
|
#error "No pipe_atomic implementation selected"
|
|
#endif
|
|
|
|
|
|
|
|
#endif /* U_ATOMIC_H */
|