Date: Sat, 23 Mar 2024 19:49:12 GMT From: Gleb Popov <arrowd@FreeBSD.org> To: ports-committers@FreeBSD.org, dev-commits-ports-all@FreeBSD.org, dev-commits-ports-main@FreeBSD.org Subject: git: 81243cb5ab62 - main - lang/ghc: Try to unbreak on AArch64 by pulling a patch from upstream. Message-ID: <202403231949.42NJnC1j095587@gitrepo.freebsd.org>
next in thread | raw e-mail | index | archive | help
The branch main has been updated by arrowd: URL: https://cgit.FreeBSD.org/ports/commit/?id=81243cb5ab6240ff32fb5c6d2c80bba36582dd17 commit 81243cb5ab6240ff32fb5c6d2c80bba36582dd17 Author: Gleb Popov <arrowd@FreeBSD.org> AuthorDate: 2024-03-23 19:47:02 +0000 Commit: Gleb Popov <arrowd@FreeBSD.org> CommitDate: 2024-03-23 19:47:57 +0000 lang/ghc: Try to unbreak on AArch64 by pulling a patch from upstream. --- lang/ghc/Makefile | 4 +- .../ghc/files/patch-rts_ARMOutlineAtomicsSymbols.h | 822 +++++++++++++++++++++ 2 files changed, 823 insertions(+), 3 deletions(-) diff --git a/lang/ghc/Makefile b/lang/ghc/Makefile index 3258a75a12f8..5e6fda6901af 100644 --- a/lang/ghc/Makefile +++ b/lang/ghc/Makefile @@ -1,6 +1,6 @@ PORTNAME= ghc PORTVERSION= ${GHC_VERSION} -PORTREVISION?= 0 +PORTREVISION?= 1 CATEGORIES= lang haskell MASTER_SITES= https://www.haskell.org/ghc/dist/${PORTVERSION}/:source \ LOCAL/arrowd/:boot @@ -15,8 +15,6 @@ LICENSE_FILE= ${WRKSRC}/LICENSE ONLY_FOR_ARCHS= aarch64 amd64 i386 -BROKEN_FreeBSD_14_aarch64= https://gitlab.haskell.org/ghc/ghc/-/issues/24580 - BUILD_DEPENDS= ${LOCALBASE}/lib/libgmp.so:math/gmp USES= autoreconf compiler:c11 gmake \ diff --git a/lang/ghc/files/patch-rts_ARMOutlineAtomicsSymbols.h b/lang/ghc/files/patch-rts_ARMOutlineAtomicsSymbols.h new file mode 100644 index 000000000000..2e956831a7f6 --- /dev/null +++ b/lang/ghc/files/patch-rts_ARMOutlineAtomicsSymbols.h @@ -0,0 +1,822 @@ +https://gitlab.haskell.org/ghc/ghc/-/merge_requests/12227 + +https://gitlab.haskell.org/ghc/ghc/-/issues/24580 + +--- rts/ARMOutlineAtomicsSymbols.h.orig 2024-01-08 13:24:59 UTC ++++ rts/ARMOutlineAtomicsSymbols.h +@@ -10,583 +10,583 @@ + #include <stdint.h> + #include <stdatomic.h> + +-uint8_t ghc___aarch64_cas1_relax(uint8_t old, uint8_t new, uint8_t* p); +-uint8_t ghc___aarch64_cas1_relax(uint8_t old, uint8_t new, uint8_t* p) { ++uint8_t ghc___aarch64_cas1_relax(uint8_t old, uint8_t new, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_cas1_relax(uint8_t old, uint8_t new, _Atomic uint8_t* p) { + atomic_compare_exchange_strong_explicit(p, &old, new, memory_order_relaxed, memory_order_relaxed); return old; + } + +-uint8_t ghc___aarch64_cas1_acq(uint8_t old, uint8_t new, uint8_t* p); +-uint8_t ghc___aarch64_cas1_acq(uint8_t old, uint8_t new, uint8_t* p) { ++uint8_t ghc___aarch64_cas1_acq(uint8_t old, uint8_t new, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_cas1_acq(uint8_t old, uint8_t new, _Atomic uint8_t* p) { + atomic_compare_exchange_strong_explicit(p, &old, new, memory_order_acquire, memory_order_acquire); return old; + } + +-uint8_t ghc___aarch64_cas1_acq_rel(uint8_t old, uint8_t new, uint8_t* p); +-uint8_t ghc___aarch64_cas1_acq_rel(uint8_t old, uint8_t new, uint8_t* p) { ++uint8_t ghc___aarch64_cas1_acq_rel(uint8_t old, uint8_t new, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_cas1_acq_rel(uint8_t old, uint8_t new, _Atomic uint8_t* p) { + atomic_compare_exchange_strong_explicit(p, &old, new, memory_order_acq_rel, memory_order_acquire); return old; + } + +-uint8_t ghc___aarch64_cas1_sync(uint8_t old, uint8_t new, uint8_t* p); +-uint8_t ghc___aarch64_cas1_sync(uint8_t old, uint8_t new, uint8_t* p) { ++uint8_t ghc___aarch64_cas1_sync(uint8_t old, uint8_t new, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_cas1_sync(uint8_t old, uint8_t new, _Atomic uint8_t* p) { + atomic_compare_exchange_strong_explicit(p, &old, new, memory_order_seq_cst, memory_order_seq_cst); return old; + } + +-uint16_t ghc___aarch64_cas2_relax(uint16_t old, uint16_t new, uint16_t* p); +-uint16_t ghc___aarch64_cas2_relax(uint16_t old, uint16_t new, uint16_t* p) { ++uint16_t ghc___aarch64_cas2_relax(uint16_t old, uint16_t new, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_cas2_relax(uint16_t old, uint16_t new, _Atomic uint16_t* p) { + atomic_compare_exchange_strong_explicit(p, &old, new, memory_order_relaxed, memory_order_relaxed); return old; + } + +-uint16_t ghc___aarch64_cas2_acq(uint16_t old, uint16_t new, uint16_t* p); +-uint16_t ghc___aarch64_cas2_acq(uint16_t old, uint16_t new, uint16_t* p) { ++uint16_t ghc___aarch64_cas2_acq(uint16_t old, uint16_t new, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_cas2_acq(uint16_t old, uint16_t new, _Atomic uint16_t* p) { + atomic_compare_exchange_strong_explicit(p, &old, new, memory_order_acquire, memory_order_acquire); return old; + } + +-uint16_t ghc___aarch64_cas2_acq_rel(uint16_t old, uint16_t new, uint16_t* p); +-uint16_t ghc___aarch64_cas2_acq_rel(uint16_t old, uint16_t new, uint16_t* p) { ++uint16_t ghc___aarch64_cas2_acq_rel(uint16_t old, uint16_t new, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_cas2_acq_rel(uint16_t old, uint16_t new, _Atomic uint16_t* p) { + atomic_compare_exchange_strong_explicit(p, &old, new, memory_order_acq_rel, memory_order_acquire); return old; + } + +-uint16_t ghc___aarch64_cas2_sync(uint16_t old, uint16_t new, uint16_t* p); +-uint16_t ghc___aarch64_cas2_sync(uint16_t old, uint16_t new, uint16_t* p) { ++uint16_t ghc___aarch64_cas2_sync(uint16_t old, uint16_t new, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_cas2_sync(uint16_t old, uint16_t new, _Atomic uint16_t* p) { + atomic_compare_exchange_strong_explicit(p, &old, new, memory_order_seq_cst, memory_order_seq_cst); return old; + } + +-uint32_t ghc___aarch64_cas4_relax(uint32_t old, uint32_t new, uint32_t* p); +-uint32_t ghc___aarch64_cas4_relax(uint32_t old, uint32_t new, uint32_t* p) { ++uint32_t ghc___aarch64_cas4_relax(uint32_t old, uint32_t new, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_cas4_relax(uint32_t old, uint32_t new, _Atomic uint32_t* p) { + atomic_compare_exchange_strong_explicit(p, &old, new, memory_order_relaxed, memory_order_relaxed); return old; + } + +-uint32_t ghc___aarch64_cas4_acq(uint32_t old, uint32_t new, uint32_t* p); +-uint32_t ghc___aarch64_cas4_acq(uint32_t old, uint32_t new, uint32_t* p) { ++uint32_t ghc___aarch64_cas4_acq(uint32_t old, uint32_t new, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_cas4_acq(uint32_t old, uint32_t new, _Atomic uint32_t* p) { + atomic_compare_exchange_strong_explicit(p, &old, new, memory_order_acquire, memory_order_acquire); return old; + } + +-uint32_t ghc___aarch64_cas4_acq_rel(uint32_t old, uint32_t new, uint32_t* p); +-uint32_t ghc___aarch64_cas4_acq_rel(uint32_t old, uint32_t new, uint32_t* p) { ++uint32_t ghc___aarch64_cas4_acq_rel(uint32_t old, uint32_t new, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_cas4_acq_rel(uint32_t old, uint32_t new, _Atomic uint32_t* p) { + atomic_compare_exchange_strong_explicit(p, &old, new, memory_order_acq_rel, memory_order_acquire); return old; + } + +-uint32_t ghc___aarch64_cas4_sync(uint32_t old, uint32_t new, uint32_t* p); +-uint32_t ghc___aarch64_cas4_sync(uint32_t old, uint32_t new, uint32_t* p) { ++uint32_t ghc___aarch64_cas4_sync(uint32_t old, uint32_t new, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_cas4_sync(uint32_t old, uint32_t new, _Atomic uint32_t* p) { + atomic_compare_exchange_strong_explicit(p, &old, new, memory_order_seq_cst, memory_order_seq_cst); return old; + } + +-uint64_t ghc___aarch64_cas8_relax(uint64_t old, uint64_t new, uint64_t* p); +-uint64_t ghc___aarch64_cas8_relax(uint64_t old, uint64_t new, uint64_t* p) { ++uint64_t ghc___aarch64_cas8_relax(uint64_t old, uint64_t new, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_cas8_relax(uint64_t old, uint64_t new, _Atomic uint64_t* p) { + atomic_compare_exchange_strong_explicit(p, &old, new, memory_order_relaxed, memory_order_relaxed); return old; + } + +-uint64_t ghc___aarch64_cas8_acq(uint64_t old, uint64_t new, uint64_t* p); +-uint64_t ghc___aarch64_cas8_acq(uint64_t old, uint64_t new, uint64_t* p) { ++uint64_t ghc___aarch64_cas8_acq(uint64_t old, uint64_t new, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_cas8_acq(uint64_t old, uint64_t new, _Atomic uint64_t* p) { + atomic_compare_exchange_strong_explicit(p, &old, new, memory_order_acquire, memory_order_acquire); return old; + } + +-uint64_t ghc___aarch64_cas8_acq_rel(uint64_t old, uint64_t new, uint64_t* p); +-uint64_t ghc___aarch64_cas8_acq_rel(uint64_t old, uint64_t new, uint64_t* p) { ++uint64_t ghc___aarch64_cas8_acq_rel(uint64_t old, uint64_t new, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_cas8_acq_rel(uint64_t old, uint64_t new, _Atomic uint64_t* p) { + atomic_compare_exchange_strong_explicit(p, &old, new, memory_order_acq_rel, memory_order_acquire); return old; + } + +-uint64_t ghc___aarch64_cas8_sync(uint64_t old, uint64_t new, uint64_t* p); +-uint64_t ghc___aarch64_cas8_sync(uint64_t old, uint64_t new, uint64_t* p) { ++uint64_t ghc___aarch64_cas8_sync(uint64_t old, uint64_t new, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_cas8_sync(uint64_t old, uint64_t new, _Atomic uint64_t* p) { + atomic_compare_exchange_strong_explicit(p, &old, new, memory_order_seq_cst, memory_order_seq_cst); return old; + } + +-uint8_t ghc___aarch64_swp1_relax(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_swp1_relax(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_swp1_relax(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_swp1_relax(uint8_t v, _Atomic uint8_t* p) { + return atomic_exchange_explicit(p, v, memory_order_relaxed); + } + +-uint8_t ghc___aarch64_swp1_acq(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_swp1_acq(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_swp1_acq(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_swp1_acq(uint8_t v, _Atomic uint8_t* p) { + return atomic_exchange_explicit(p, v, memory_order_acquire); + } + +-uint8_t ghc___aarch64_swp1_rel(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_swp1_rel(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_swp1_rel(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_swp1_rel(uint8_t v, _Atomic uint8_t* p) { + return atomic_exchange_explicit(p, v, memory_order_release); + } + +-uint8_t ghc___aarch64_swp1_acq_rel(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_swp1_acq_rel(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_swp1_acq_rel(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_swp1_acq_rel(uint8_t v, _Atomic uint8_t* p) { + return atomic_exchange_explicit(p, v, memory_order_acq_rel); + } + +-uint8_t ghc___aarch64_swp1_sync(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_swp1_sync(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_swp1_sync(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_swp1_sync(uint8_t v, _Atomic uint8_t* p) { + return atomic_exchange_explicit(p, v, memory_order_seq_cst); + } + +-uint16_t ghc___aarch64_swp2_relax(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_swp2_relax(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_swp2_relax(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_swp2_relax(uint16_t v, _Atomic uint16_t* p) { + return atomic_exchange_explicit(p, v, memory_order_relaxed); + } + +-uint16_t ghc___aarch64_swp2_acq(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_swp2_acq(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_swp2_acq(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_swp2_acq(uint16_t v, _Atomic uint16_t* p) { + return atomic_exchange_explicit(p, v, memory_order_acquire); + } + +-uint16_t ghc___aarch64_swp2_rel(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_swp2_rel(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_swp2_rel(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_swp2_rel(uint16_t v, _Atomic uint16_t* p) { + return atomic_exchange_explicit(p, v, memory_order_release); + } + +-uint16_t ghc___aarch64_swp2_acq_rel(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_swp2_acq_rel(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_swp2_acq_rel(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_swp2_acq_rel(uint16_t v, _Atomic uint16_t* p) { + return atomic_exchange_explicit(p, v, memory_order_acq_rel); + } + +-uint16_t ghc___aarch64_swp2_sync(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_swp2_sync(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_swp2_sync(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_swp2_sync(uint16_t v, _Atomic uint16_t* p) { + return atomic_exchange_explicit(p, v, memory_order_seq_cst); + } + +-uint32_t ghc___aarch64_swp4_relax(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_swp4_relax(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_swp4_relax(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_swp4_relax(uint32_t v, _Atomic uint32_t* p) { + return atomic_exchange_explicit(p, v, memory_order_relaxed); + } + +-uint32_t ghc___aarch64_swp4_acq(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_swp4_acq(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_swp4_acq(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_swp4_acq(uint32_t v, _Atomic uint32_t* p) { + return atomic_exchange_explicit(p, v, memory_order_acquire); + } + +-uint32_t ghc___aarch64_swp4_rel(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_swp4_rel(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_swp4_rel(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_swp4_rel(uint32_t v, _Atomic uint32_t* p) { + return atomic_exchange_explicit(p, v, memory_order_release); + } + +-uint32_t ghc___aarch64_swp4_acq_rel(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_swp4_acq_rel(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_swp4_acq_rel(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_swp4_acq_rel(uint32_t v, _Atomic uint32_t* p) { + return atomic_exchange_explicit(p, v, memory_order_acq_rel); + } + +-uint32_t ghc___aarch64_swp4_sync(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_swp4_sync(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_swp4_sync(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_swp4_sync(uint32_t v, _Atomic uint32_t* p) { + return atomic_exchange_explicit(p, v, memory_order_seq_cst); + } + +-uint64_t ghc___aarch64_swp8_relax(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_swp8_relax(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_swp8_relax(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_swp8_relax(uint64_t v, _Atomic uint64_t* p) { + return atomic_exchange_explicit(p, v, memory_order_relaxed); + } + +-uint64_t ghc___aarch64_swp8_acq(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_swp8_acq(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_swp8_acq(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_swp8_acq(uint64_t v, _Atomic uint64_t* p) { + return atomic_exchange_explicit(p, v, memory_order_acquire); + } + +-uint64_t ghc___aarch64_swp8_rel(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_swp8_rel(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_swp8_rel(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_swp8_rel(uint64_t v, _Atomic uint64_t* p) { + return atomic_exchange_explicit(p, v, memory_order_release); + } + +-uint64_t ghc___aarch64_swp8_acq_rel(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_swp8_acq_rel(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_swp8_acq_rel(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_swp8_acq_rel(uint64_t v, _Atomic uint64_t* p) { + return atomic_exchange_explicit(p, v, memory_order_acq_rel); + } + +-uint64_t ghc___aarch64_swp8_sync(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_swp8_sync(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_swp8_sync(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_swp8_sync(uint64_t v, _Atomic uint64_t* p) { + return atomic_exchange_explicit(p, v, memory_order_seq_cst); + } + +-uint8_t ghc___aarch64_ldadd1_relax(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_ldadd1_relax(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_ldadd1_relax(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_ldadd1_relax(uint8_t v, _Atomic uint8_t* p) { + return atomic_fetch_add_explicit(p, v, memory_order_relaxed); + } + +-uint8_t ghc___aarch64_ldadd1_acq(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_ldadd1_acq(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_ldadd1_acq(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_ldadd1_acq(uint8_t v, _Atomic uint8_t* p) { + return atomic_fetch_add_explicit(p, v, memory_order_acquire); + } + +-uint8_t ghc___aarch64_ldadd1_rel(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_ldadd1_rel(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_ldadd1_rel(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_ldadd1_rel(uint8_t v, _Atomic uint8_t* p) { + return atomic_fetch_add_explicit(p, v, memory_order_release); + } + +-uint8_t ghc___aarch64_ldadd1_acq_rel(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_ldadd1_acq_rel(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_ldadd1_acq_rel(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_ldadd1_acq_rel(uint8_t v, _Atomic uint8_t* p) { + return atomic_fetch_add_explicit(p, v, memory_order_acq_rel); + } + +-uint8_t ghc___aarch64_ldadd1_sync(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_ldadd1_sync(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_ldadd1_sync(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_ldadd1_sync(uint8_t v, _Atomic uint8_t* p) { + return atomic_fetch_add_explicit(p, v, memory_order_seq_cst); + } + +-uint16_t ghc___aarch64_ldadd2_relax(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_ldadd2_relax(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_ldadd2_relax(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_ldadd2_relax(uint16_t v, _Atomic uint16_t* p) { + return atomic_fetch_add_explicit(p, v, memory_order_relaxed); + } + +-uint16_t ghc___aarch64_ldadd2_acq(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_ldadd2_acq(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_ldadd2_acq(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_ldadd2_acq(uint16_t v, _Atomic uint16_t* p) { + return atomic_fetch_add_explicit(p, v, memory_order_acquire); + } + +-uint16_t ghc___aarch64_ldadd2_rel(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_ldadd2_rel(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_ldadd2_rel(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_ldadd2_rel(uint16_t v, _Atomic uint16_t* p) { + return atomic_fetch_add_explicit(p, v, memory_order_release); + } + +-uint16_t ghc___aarch64_ldadd2_acq_rel(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_ldadd2_acq_rel(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_ldadd2_acq_rel(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_ldadd2_acq_rel(uint16_t v, _Atomic uint16_t* p) { + return atomic_fetch_add_explicit(p, v, memory_order_acq_rel); + } + +-uint16_t ghc___aarch64_ldadd2_sync(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_ldadd2_sync(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_ldadd2_sync(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_ldadd2_sync(uint16_t v, _Atomic uint16_t* p) { + return atomic_fetch_add_explicit(p, v, memory_order_seq_cst); + } + +-uint32_t ghc___aarch64_ldadd4_relax(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_ldadd4_relax(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_ldadd4_relax(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_ldadd4_relax(uint32_t v, _Atomic uint32_t* p) { + return atomic_fetch_add_explicit(p, v, memory_order_relaxed); + } + +-uint32_t ghc___aarch64_ldadd4_acq(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_ldadd4_acq(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_ldadd4_acq(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_ldadd4_acq(uint32_t v, _Atomic uint32_t* p) { + return atomic_fetch_add_explicit(p, v, memory_order_acquire); + } + +-uint32_t ghc___aarch64_ldadd4_rel(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_ldadd4_rel(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_ldadd4_rel(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_ldadd4_rel(uint32_t v, _Atomic uint32_t* p) { + return atomic_fetch_add_explicit(p, v, memory_order_release); + } + +-uint32_t ghc___aarch64_ldadd4_acq_rel(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_ldadd4_acq_rel(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_ldadd4_acq_rel(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_ldadd4_acq_rel(uint32_t v, _Atomic uint32_t* p) { + return atomic_fetch_add_explicit(p, v, memory_order_acq_rel); + } + +-uint32_t ghc___aarch64_ldadd4_sync(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_ldadd4_sync(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_ldadd4_sync(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_ldadd4_sync(uint32_t v, _Atomic uint32_t* p) { + return atomic_fetch_add_explicit(p, v, memory_order_seq_cst); + } + +-uint64_t ghc___aarch64_ldadd8_relax(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_ldadd8_relax(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_ldadd8_relax(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_ldadd8_relax(uint64_t v, _Atomic uint64_t* p) { + return atomic_fetch_add_explicit(p, v, memory_order_relaxed); + } + +-uint64_t ghc___aarch64_ldadd8_acq(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_ldadd8_acq(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_ldadd8_acq(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_ldadd8_acq(uint64_t v, _Atomic uint64_t* p) { + return atomic_fetch_add_explicit(p, v, memory_order_acquire); + } + +-uint64_t ghc___aarch64_ldadd8_rel(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_ldadd8_rel(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_ldadd8_rel(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_ldadd8_rel(uint64_t v, _Atomic uint64_t* p) { + return atomic_fetch_add_explicit(p, v, memory_order_release); + } + +-uint64_t ghc___aarch64_ldadd8_acq_rel(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_ldadd8_acq_rel(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_ldadd8_acq_rel(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_ldadd8_acq_rel(uint64_t v, _Atomic uint64_t* p) { + return atomic_fetch_add_explicit(p, v, memory_order_acq_rel); + } + +-uint64_t ghc___aarch64_ldadd8_sync(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_ldadd8_sync(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_ldadd8_sync(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_ldadd8_sync(uint64_t v, _Atomic uint64_t* p) { + return atomic_fetch_add_explicit(p, v, memory_order_seq_cst); + } + +-uint8_t ghc___aarch64_ldclr1_relax(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_ldclr1_relax(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_ldclr1_relax(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_ldclr1_relax(uint8_t v, _Atomic uint8_t* p) { + return atomic_fetch_and_explicit(p, v, memory_order_relaxed); + } + +-uint8_t ghc___aarch64_ldclr1_acq(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_ldclr1_acq(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_ldclr1_acq(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_ldclr1_acq(uint8_t v, _Atomic uint8_t* p) { + return atomic_fetch_and_explicit(p, v, memory_order_acquire); + } + +-uint8_t ghc___aarch64_ldclr1_rel(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_ldclr1_rel(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_ldclr1_rel(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_ldclr1_rel(uint8_t v, _Atomic uint8_t* p) { + return atomic_fetch_and_explicit(p, v, memory_order_release); + } + +-uint8_t ghc___aarch64_ldclr1_acq_rel(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_ldclr1_acq_rel(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_ldclr1_acq_rel(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_ldclr1_acq_rel(uint8_t v, _Atomic uint8_t* p) { + return atomic_fetch_and_explicit(p, v, memory_order_acq_rel); + } + +-uint8_t ghc___aarch64_ldclr1_sync(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_ldclr1_sync(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_ldclr1_sync(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_ldclr1_sync(uint8_t v, _Atomic uint8_t* p) { + return atomic_fetch_and_explicit(p, v, memory_order_seq_cst); + } + +-uint16_t ghc___aarch64_ldclr2_relax(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_ldclr2_relax(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_ldclr2_relax(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_ldclr2_relax(uint16_t v, _Atomic uint16_t* p) { + return atomic_fetch_and_explicit(p, v, memory_order_relaxed); + } + +-uint16_t ghc___aarch64_ldclr2_acq(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_ldclr2_acq(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_ldclr2_acq(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_ldclr2_acq(uint16_t v, _Atomic uint16_t* p) { + return atomic_fetch_and_explicit(p, v, memory_order_acquire); + } + +-uint16_t ghc___aarch64_ldclr2_rel(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_ldclr2_rel(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_ldclr2_rel(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_ldclr2_rel(uint16_t v, _Atomic uint16_t* p) { + return atomic_fetch_and_explicit(p, v, memory_order_release); + } + +-uint16_t ghc___aarch64_ldclr2_acq_rel(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_ldclr2_acq_rel(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_ldclr2_acq_rel(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_ldclr2_acq_rel(uint16_t v, _Atomic uint16_t* p) { + return atomic_fetch_and_explicit(p, v, memory_order_acq_rel); + } + +-uint16_t ghc___aarch64_ldclr2_sync(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_ldclr2_sync(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_ldclr2_sync(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_ldclr2_sync(uint16_t v, _Atomic uint16_t* p) { + return atomic_fetch_and_explicit(p, v, memory_order_seq_cst); + } + +-uint32_t ghc___aarch64_ldclr4_relax(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_ldclr4_relax(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_ldclr4_relax(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_ldclr4_relax(uint32_t v, _Atomic uint32_t* p) { + return atomic_fetch_and_explicit(p, v, memory_order_relaxed); + } + +-uint32_t ghc___aarch64_ldclr4_acq(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_ldclr4_acq(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_ldclr4_acq(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_ldclr4_acq(uint32_t v, _Atomic uint32_t* p) { + return atomic_fetch_and_explicit(p, v, memory_order_acquire); + } + +-uint32_t ghc___aarch64_ldclr4_rel(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_ldclr4_rel(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_ldclr4_rel(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_ldclr4_rel(uint32_t v, _Atomic uint32_t* p) { + return atomic_fetch_and_explicit(p, v, memory_order_release); + } + +-uint32_t ghc___aarch64_ldclr4_acq_rel(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_ldclr4_acq_rel(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_ldclr4_acq_rel(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_ldclr4_acq_rel(uint32_t v, _Atomic uint32_t* p) { + return atomic_fetch_and_explicit(p, v, memory_order_acq_rel); + } + +-uint32_t ghc___aarch64_ldclr4_sync(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_ldclr4_sync(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_ldclr4_sync(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_ldclr4_sync(uint32_t v, _Atomic uint32_t* p) { + return atomic_fetch_and_explicit(p, v, memory_order_seq_cst); + } + +-uint64_t ghc___aarch64_ldclr8_relax(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_ldclr8_relax(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_ldclr8_relax(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_ldclr8_relax(uint64_t v, _Atomic uint64_t* p) { + return atomic_fetch_and_explicit(p, v, memory_order_relaxed); + } + +-uint64_t ghc___aarch64_ldclr8_acq(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_ldclr8_acq(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_ldclr8_acq(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_ldclr8_acq(uint64_t v, _Atomic uint64_t* p) { + return atomic_fetch_and_explicit(p, v, memory_order_acquire); + } + +-uint64_t ghc___aarch64_ldclr8_rel(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_ldclr8_rel(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_ldclr8_rel(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_ldclr8_rel(uint64_t v, _Atomic uint64_t* p) { + return atomic_fetch_and_explicit(p, v, memory_order_release); + } + +-uint64_t ghc___aarch64_ldclr8_acq_rel(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_ldclr8_acq_rel(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_ldclr8_acq_rel(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_ldclr8_acq_rel(uint64_t v, _Atomic uint64_t* p) { + return atomic_fetch_and_explicit(p, v, memory_order_acq_rel); + } + +-uint64_t ghc___aarch64_ldclr8_sync(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_ldclr8_sync(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_ldclr8_sync(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_ldclr8_sync(uint64_t v, _Atomic uint64_t* p) { + return atomic_fetch_and_explicit(p, v, memory_order_seq_cst); + } + +-uint8_t ghc___aarch64_ldeor1_relax(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_ldeor1_relax(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_ldeor1_relax(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_ldeor1_relax(uint8_t v, _Atomic uint8_t* p) { + return atomic_fetch_xor_explicit(p, v, memory_order_relaxed); + } + +-uint8_t ghc___aarch64_ldeor1_acq(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_ldeor1_acq(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_ldeor1_acq(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_ldeor1_acq(uint8_t v, _Atomic uint8_t* p) { + return atomic_fetch_xor_explicit(p, v, memory_order_acquire); + } + +-uint8_t ghc___aarch64_ldeor1_rel(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_ldeor1_rel(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_ldeor1_rel(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_ldeor1_rel(uint8_t v, _Atomic uint8_t* p) { + return atomic_fetch_xor_explicit(p, v, memory_order_release); + } + +-uint8_t ghc___aarch64_ldeor1_acq_rel(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_ldeor1_acq_rel(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_ldeor1_acq_rel(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_ldeor1_acq_rel(uint8_t v, _Atomic uint8_t* p) { + return atomic_fetch_xor_explicit(p, v, memory_order_acq_rel); + } + +-uint8_t ghc___aarch64_ldeor1_sync(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_ldeor1_sync(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_ldeor1_sync(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_ldeor1_sync(uint8_t v, _Atomic uint8_t* p) { + return atomic_fetch_xor_explicit(p, v, memory_order_seq_cst); + } + +-uint16_t ghc___aarch64_ldeor2_relax(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_ldeor2_relax(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_ldeor2_relax(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_ldeor2_relax(uint16_t v, _Atomic uint16_t* p) { + return atomic_fetch_xor_explicit(p, v, memory_order_relaxed); + } + +-uint16_t ghc___aarch64_ldeor2_acq(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_ldeor2_acq(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_ldeor2_acq(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_ldeor2_acq(uint16_t v, _Atomic uint16_t* p) { + return atomic_fetch_xor_explicit(p, v, memory_order_acquire); + } + +-uint16_t ghc___aarch64_ldeor2_rel(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_ldeor2_rel(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_ldeor2_rel(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_ldeor2_rel(uint16_t v, _Atomic uint16_t* p) { + return atomic_fetch_xor_explicit(p, v, memory_order_release); + } + +-uint16_t ghc___aarch64_ldeor2_acq_rel(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_ldeor2_acq_rel(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_ldeor2_acq_rel(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_ldeor2_acq_rel(uint16_t v, _Atomic uint16_t* p) { + return atomic_fetch_xor_explicit(p, v, memory_order_acq_rel); + } + +-uint16_t ghc___aarch64_ldeor2_sync(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_ldeor2_sync(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_ldeor2_sync(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_ldeor2_sync(uint16_t v, _Atomic uint16_t* p) { + return atomic_fetch_xor_explicit(p, v, memory_order_seq_cst); + } + +-uint32_t ghc___aarch64_ldeor4_relax(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_ldeor4_relax(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_ldeor4_relax(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_ldeor4_relax(uint32_t v, _Atomic uint32_t* p) { + return atomic_fetch_xor_explicit(p, v, memory_order_relaxed); + } + +-uint32_t ghc___aarch64_ldeor4_acq(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_ldeor4_acq(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_ldeor4_acq(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_ldeor4_acq(uint32_t v, _Atomic uint32_t* p) { + return atomic_fetch_xor_explicit(p, v, memory_order_acquire); + } + +-uint32_t ghc___aarch64_ldeor4_rel(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_ldeor4_rel(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_ldeor4_rel(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_ldeor4_rel(uint32_t v, _Atomic uint32_t* p) { + return atomic_fetch_xor_explicit(p, v, memory_order_release); + } + +-uint32_t ghc___aarch64_ldeor4_acq_rel(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_ldeor4_acq_rel(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_ldeor4_acq_rel(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_ldeor4_acq_rel(uint32_t v, _Atomic uint32_t* p) { + return atomic_fetch_xor_explicit(p, v, memory_order_acq_rel); + } + +-uint32_t ghc___aarch64_ldeor4_sync(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_ldeor4_sync(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_ldeor4_sync(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_ldeor4_sync(uint32_t v, _Atomic uint32_t* p) { + return atomic_fetch_xor_explicit(p, v, memory_order_seq_cst); + } + +-uint64_t ghc___aarch64_ldeor8_relax(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_ldeor8_relax(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_ldeor8_relax(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_ldeor8_relax(uint64_t v, _Atomic uint64_t* p) { + return atomic_fetch_xor_explicit(p, v, memory_order_relaxed); + } + +-uint64_t ghc___aarch64_ldeor8_acq(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_ldeor8_acq(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_ldeor8_acq(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_ldeor8_acq(uint64_t v, _Atomic uint64_t* p) { + return atomic_fetch_xor_explicit(p, v, memory_order_acquire); + } + +-uint64_t ghc___aarch64_ldeor8_rel(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_ldeor8_rel(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_ldeor8_rel(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_ldeor8_rel(uint64_t v, _Atomic uint64_t* p) { + return atomic_fetch_xor_explicit(p, v, memory_order_release); + } + +-uint64_t ghc___aarch64_ldeor8_acq_rel(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_ldeor8_acq_rel(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_ldeor8_acq_rel(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_ldeor8_acq_rel(uint64_t v, _Atomic uint64_t* p) { + return atomic_fetch_xor_explicit(p, v, memory_order_acq_rel); + } + +-uint64_t ghc___aarch64_ldeor8_sync(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_ldeor8_sync(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_ldeor8_sync(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_ldeor8_sync(uint64_t v, _Atomic uint64_t* p) { + return atomic_fetch_xor_explicit(p, v, memory_order_seq_cst); + } + +-uint8_t ghc___aarch64_ldset1_relax(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_ldset1_relax(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_ldset1_relax(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_ldset1_relax(uint8_t v, _Atomic uint8_t* p) { + return atomic_fetch_or_explicit(p, v, memory_order_relaxed); + } + +-uint8_t ghc___aarch64_ldset1_acq(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_ldset1_acq(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_ldset1_acq(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_ldset1_acq(uint8_t v, _Atomic uint8_t* p) { + return atomic_fetch_or_explicit(p, v, memory_order_acquire); + } + +-uint8_t ghc___aarch64_ldset1_rel(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_ldset1_rel(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_ldset1_rel(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_ldset1_rel(uint8_t v, _Atomic uint8_t* p) { + return atomic_fetch_or_explicit(p, v, memory_order_release); + } + +-uint8_t ghc___aarch64_ldset1_acq_rel(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_ldset1_acq_rel(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_ldset1_acq_rel(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_ldset1_acq_rel(uint8_t v, _Atomic uint8_t* p) { + return atomic_fetch_or_explicit(p, v, memory_order_acq_rel); + } + +-uint8_t ghc___aarch64_ldset1_sync(uint8_t v, uint8_t* p); +-uint8_t ghc___aarch64_ldset1_sync(uint8_t v, uint8_t* p) { ++uint8_t ghc___aarch64_ldset1_sync(uint8_t v, _Atomic uint8_t* p); ++uint8_t ghc___aarch64_ldset1_sync(uint8_t v, _Atomic uint8_t* p) { + return atomic_fetch_or_explicit(p, v, memory_order_seq_cst); + } + +-uint16_t ghc___aarch64_ldset2_relax(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_ldset2_relax(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_ldset2_relax(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_ldset2_relax(uint16_t v, _Atomic uint16_t* p) { + return atomic_fetch_or_explicit(p, v, memory_order_relaxed); + } + +-uint16_t ghc___aarch64_ldset2_acq(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_ldset2_acq(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_ldset2_acq(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_ldset2_acq(uint16_t v, _Atomic uint16_t* p) { + return atomic_fetch_or_explicit(p, v, memory_order_acquire); + } + +-uint16_t ghc___aarch64_ldset2_rel(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_ldset2_rel(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_ldset2_rel(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_ldset2_rel(uint16_t v, _Atomic uint16_t* p) { + return atomic_fetch_or_explicit(p, v, memory_order_release); + } + +-uint16_t ghc___aarch64_ldset2_acq_rel(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_ldset2_acq_rel(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_ldset2_acq_rel(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_ldset2_acq_rel(uint16_t v, _Atomic uint16_t* p) { + return atomic_fetch_or_explicit(p, v, memory_order_acq_rel); + } + +-uint16_t ghc___aarch64_ldset2_sync(uint16_t v, uint16_t* p); +-uint16_t ghc___aarch64_ldset2_sync(uint16_t v, uint16_t* p) { ++uint16_t ghc___aarch64_ldset2_sync(uint16_t v, _Atomic uint16_t* p); ++uint16_t ghc___aarch64_ldset2_sync(uint16_t v, _Atomic uint16_t* p) { + return atomic_fetch_or_explicit(p, v, memory_order_seq_cst); + } + +-uint32_t ghc___aarch64_ldset4_relax(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_ldset4_relax(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_ldset4_relax(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_ldset4_relax(uint32_t v, _Atomic uint32_t* p) { + return atomic_fetch_or_explicit(p, v, memory_order_relaxed); + } + +-uint32_t ghc___aarch64_ldset4_acq(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_ldset4_acq(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_ldset4_acq(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_ldset4_acq(uint32_t v, _Atomic uint32_t* p) { + return atomic_fetch_or_explicit(p, v, memory_order_acquire); + } + +-uint32_t ghc___aarch64_ldset4_rel(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_ldset4_rel(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_ldset4_rel(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_ldset4_rel(uint32_t v, _Atomic uint32_t* p) { + return atomic_fetch_or_explicit(p, v, memory_order_release); + } + +-uint32_t ghc___aarch64_ldset4_acq_rel(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_ldset4_acq_rel(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_ldset4_acq_rel(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_ldset4_acq_rel(uint32_t v, _Atomic uint32_t* p) { + return atomic_fetch_or_explicit(p, v, memory_order_acq_rel); + } + +-uint32_t ghc___aarch64_ldset4_sync(uint32_t v, uint32_t* p); +-uint32_t ghc___aarch64_ldset4_sync(uint32_t v, uint32_t* p) { ++uint32_t ghc___aarch64_ldset4_sync(uint32_t v, _Atomic uint32_t* p); ++uint32_t ghc___aarch64_ldset4_sync(uint32_t v, _Atomic uint32_t* p) { + return atomic_fetch_or_explicit(p, v, memory_order_seq_cst); + } + +-uint64_t ghc___aarch64_ldset8_relax(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_ldset8_relax(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_ldset8_relax(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_ldset8_relax(uint64_t v, _Atomic uint64_t* p) { + return atomic_fetch_or_explicit(p, v, memory_order_relaxed); + } + +-uint64_t ghc___aarch64_ldset8_acq(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_ldset8_acq(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_ldset8_acq(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_ldset8_acq(uint64_t v, _Atomic uint64_t* p) { + return atomic_fetch_or_explicit(p, v, memory_order_acquire); + } + +-uint64_t ghc___aarch64_ldset8_rel(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_ldset8_rel(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_ldset8_rel(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_ldset8_rel(uint64_t v, _Atomic uint64_t* p) { + return atomic_fetch_or_explicit(p, v, memory_order_release); + } + +-uint64_t ghc___aarch64_ldset8_acq_rel(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_ldset8_acq_rel(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_ldset8_acq_rel(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_ldset8_acq_rel(uint64_t v, _Atomic uint64_t* p) { + return atomic_fetch_or_explicit(p, v, memory_order_acq_rel); + } + +-uint64_t ghc___aarch64_ldset8_sync(uint64_t v, uint64_t* p); +-uint64_t ghc___aarch64_ldset8_sync(uint64_t v, uint64_t* p) { ++uint64_t ghc___aarch64_ldset8_sync(uint64_t v, _Atomic uint64_t* p); ++uint64_t ghc___aarch64_ldset8_sync(uint64_t v, _Atomic uint64_t* p) { + return atomic_fetch_or_explicit(p, v, memory_order_seq_cst); + } +
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?202403231949.42NJnC1j095587>