== [7.4.0] 2013-11-17 ==
* Add and/or/xor entries to list_atomic (tests).
* Add char/short/int/AO_double_t and dd_acquire cases to list_atomic (tests).
* Add compile-time assertion for size of 'standard' AO_double_t.
* Add double_store pthread-based implementation and tests.
* Add generalized CAS primitives of char/short/int size.
* Add generalized atomic and/or/xor operations for char/short/int types.
* Add generalized fetch_and_add_acquire/release (for ARMv6+).
* Add generic implementation of double_load primitives.
* Add information about AO_ASSUME_VISTA to README_win32.
* Add internal header containing only char/short/int/AO_t atomic loads.
* Add load/store primitives generalization based on CAS.
* Add lock-based implementation of char/short/int_fetch_compare_and_swap.
* Add makefile rule to test list_atomic.template syntax (tests).
* Add missing 'const' in aligned-atomic XSIZE_load implementation.
* Add missing double_compare_and_swap to generalization.
* Add missing generalization of no-barrier CAS in template.
* Add negative double-CAS test cases to test_atomic_include (tests).
* Add test_stack to Makefile.msft (tests).
* Adjust fprintf arguments type matching specifier in test_stack (tests).
* Adjust included filenames in atomic_ops_malloc and test_stack.
* Adjust quotes in echo command of Makefile.msft (Win32).
* Always use 'mfence' for nop_full if target CPU supports SSE2 (gcc/x86).
* Better document configure THREADDLLIBS variable.
* Cast away volatile on dereference in CAS-based generalization primitives.
* Change policy regarding version numbers ("micro" part instead of "alpha").
* Convert README to Markdown format.
* Define AO_NO_PTHREADS in configure if no pthreads (Win32 and VxWorks).
* Define AO_int_X operations for ARM and avr32.
* Define double-wide ordered loads/stores for x86.
* Define int_and/or/xor primitives in ao_t_is_int header.
* Define nop_full as compiler barrier for pre-ARMv6 single-core case.
* Do not duplicate BUILT_SOURCES entries in nobase_private_HEADERS (Makefile).
* Do not include standard_ao_double_t.h where double-CAS is unimplemented.
* Do not report absence of meaningless nop, load and store in test_atomic.
* Do not use deprecated AO_T and AO_TS_T (tests).
* Eliminate 'missing initializer' warning for AO_stack_t value initializer.
* Eliminate 64-bit compiler warnings in atomic_ops_malloc.
* Eliminate arithmetic shifts in double-CAS (gcc/arm, msftc/x86).
* Eliminate warning for fetch_and_add argument in test_atomic_include (tests).
* Enable Makefile.msft for Win64.
* Enable build using toolchain without pthreads.
* Enable double_compare_and_swap for non-cpp code (msftc/x86.h).
* Enable generalization of all variants of CAS via fetch_compare_and_swap.
* Enable test_stack for pthreads-w32 and Win32 with native threads.
* Fix generalized AO_char/short_compare_and_swap args (missing 'unsigned').
* Fix makefile sed rule for list_atomic (tests).
* Fix missing abort() usage in atomic_ops_malloc and tests on WinCE.
* Generalize compare_double_and_swap_double using double_compare_and_swap.
* Generalize double_load/store for x86_64 (GCC).
* Generate ao_t_is_int, 'loadstore' headers from templates.
* Generate generalized AO_t load/store/fetch_and_add primitives from template.
* Generate ordered_loads/stores_only headers from templates.
* Group all X_acquire_release_volatile.h and X_[aligned_]atomic_load_store.h.
* Implement and/or/xor, AO_double_load for ARM.
* Implement atomic store using direct write by default on ARMv6+.
* Implement char/short/int-wide primitives using GCC built-in atomic/sync.
* Implement char/short/int_fetch_and_add for msftc/x86[_64] (Win32).
* Implement char/short_fetch_and_add, char/short_load for ARMv6+ (GCC).
* Implement char/short_store primitives at aligned addresses for ARM.
* Implement compare_double_and_swap_double for SunCC/x86.
* Implement double_load/store based on guaranteed x86 access atomicity.
* Implement double_store for ARMv7 using LDREXD/STREXD.
* Implement load/store via simple LDR/STR for ARMv6+ (msftc).
* Implement nop_full/write using 'dmb' instruction if available (gcc/arm).
* Improve debug printing in test_stack (tests).
* Log messages to stdout instead of stderr (tests).
* Make AO_ASSUME_VISTA also enables Win98 code in msftc/x86.h (Win32).
* Minimize gcc/generic-arithm template by factoring out barriers.
* Move 'unsigned' keyword to XCTYPE in generalize-small template.
* Move default compiler options to CFLAGS in Makefile.msft (Win32).
* Move definitions of ordered loads/stores to inner separate headers.
* Move gcc-generic AO_t-wide primitives to generic-small/arithm headers.
* Move generalized arithmetical primitives to 'generalize-arithm' template.
* Optimize AO_spin manually to minimize compiler influence on its duration.
* Parameterize list_atomic template with XSIZE (tests).
* Perform only few list reversals in test_malloc if AO based on pthreads.
* Put autogen.sh to 'dist' package (Automake).
* Remote duplicate definition of test_and_set_acquire in generalize.h.
* Remove X_aligned_atomic_load_store headers and template.
* Remove duplicate AO_spin and AO_pause definition in atomic_ops_stack.
* Remove gcc/x86_64.h eliminating code duplication of gcc/x86.h.
* Remove nested AO_USE_PTHREAD_DEFS macro check in atomic_ops.h (gcc/arm).
* Remove redundant 'cc' clobber for LDREXD instruction (gcc/arm).
* Remove store_full from msftc/arm.h in favor of generalized primitive.
* Remove sunc/x86_64.h eliminating code duplication of sunc/x86.h.
* Remove unsafe emulation-based implementation of double CAS (SunCC/x86_64).
* Remove useless 'perror' call in run_parallel.h (tests).
* Reorder AO_double_t union elements for AO_DOUBLE_T_INITIALIZER portability.
* Replace atomic_load_store.template with atomic_load and atomic_store ones.
* Replace some FIXME items with TODO in atomic_ops.c and sysdeps headers.
* Specify fetch_and_add/sub1 result as unused in test_atomic (tests).
* Support AArch64 (64-bit ARM) target (GCC).
* Support ARMv8 target (gcc/arm).
* Test double_compare_and_swap in test_atomic (tests).
* Use AO_ prefix for internal functions in arm_v6.h, hppa.h.
* Use __atomic GCC built-in to implement generic double-wide CAS.
* Use built-in __sync CAS for double-CAS if AO_USE_SYNC_CAS_BUILTIN for x86.
* Workaround GCC 4.4.3 warning reported for 'val' of list_atomic.c (tests).
== [7.3alpha2] 2012-05-11 ==
* Add '-no-undefined' to LDFLAGS in src/Makefile.am.
* Add AO_and, AO_xor atomic operations.
* Add AO_fetch_compare_and_swap primitives.
* Add and fill in AUTHORS, TODO files.
* Add autogen.sh file.
* Adjust AO_..._H macros in public headers.
* Code refactoring of gcc/arm.h by introducing AO_ARM_HAVE_x macros.
* Define AO macros for libatomic_ops version identification.
* Do not define NDEBUG if '--enable-assertions' passed to configure.
* Eliminate compiler warnings in various functions and macros.
* Generalize AO_compare_and_swap primitives via AO_fetch_compare_and_swap.
* Generalize acquire/release/full CAS primitives for MIPS
* Implement fetch_and_add, test_and_set primitives for MIPS.
* Improve Makefile for MS VC++; pass '-W3' option to MS compiler.
* Include ao_t_is_int.h from atomic_ops.h after first generalization pass
* Merge all Makefile.am files in src tree.
* Minor code refactoring of atomic_ops.c, generic_pthread.h.
* Minor configure build improvements (e.g., ensure proper autoconf version).
* Place only major per-release changes description to ChangeLog (this file).
* Recognize AO_PREFER_GENERALIZED macro to favor generalization over assembly.
* Remove all auto-generated files except for generalize-small.h from the repo.
* Remove duplicate doc/COPYING and empty NEWS files.
* Replace atomic_ops_malloc static mmap-related empty functions with macros.
* Replace pointer relational comparisons with non-pointer ones.
* Require autoconf 2.61 instead of v2.64.
* Show extra compiler warnings (GCC only).
* Turn off AO primitives inlining if AO_NO_INLINE defined.
* Use __builtin_expect in CAS failure loop condition checks (GCC only).
== [7.2e] 2013-11-10 ==
* Fix (remove) invalid include of read_ordered.h for ARM.
* Fix AM_CONFIG_HEADER in configure for autoconf-2.69-1.
* Fix AO_pause sleep delay for particular argument values (Win32).
* Fix ARMv7 LDREXD/STREXD double-wide operand specification (GCC/Clang).
* Fix LDREXD/STREXD use for pre-Clang3.3/arm.
* Fix README regarding _acquire_read barrier.
* Fix XSIZE_load/store definition order in generalize-small template.
* Fix asm constraint of CAS memory operand for gcc/alpha, clang-3.1/mips.
* Fix asm constraints of primitives in sunc/x86.h.
* Fix cmpxchg16b-based compare_double_and_swap_double for SunCC/x86_64.
* Fix compare_double_and_swap_double and double_ptr_storage for gcc/x32.
* Fix compare_double_and_swap_double for clang3.0/x86 in PIC mode.
* Fix compare_double_and_swap_double_full definition condition in emul_cas.
* Fix generalize-small template adding missed CAS-based fetch_and_add.
* Fix generalized fetch_and_add function.
* Fix missing compiler barrier in nop_full for uniprocessor ARM.
* Fix ordered_except_wr header inclusion for s390.
* Fix return type of AO_int_X primitives defined in ao_t_is_int header.
* Fix return type of char/short/int_load_read() in read_ordered.h.
* Fix template-based headers regeneration order in src/Makefile.
* Fix typos in ao_t_is_int, atomic_ops.h, generalize.h, msftc/arm.h comments.
* Fix variable type to match printf format specifier in test_stack.
* Fix visibility and initial value of 'dummy' variable in atomic_ops_stack.
* Terminate tests with abort after error reported.
== [7.2d] 2012-08-09 ==
* Fix AO_compare_double_and_swap_double_full for gcc-4.2.1/x86 in PIC mode.
* Fix AO_compiler_barrier missing parentheses.
* Fix missing 'unsigned' for generalized AO_char/short_fetch_and_add result.
== [7.2] 2012-05-11 ==
* Add atomic_ops.pc.in and atomic_ops-uninstalled.pc.in to pkgconfig folder.
* Define and use AO_PTRDIFF_T in tests for casts between pointer and integer.
* Fix AO_compare_and_swap return type for s390 and PowerPC.
* Fix AO_compare_double_and_swap_double_full for gcc/x86 (PIC mode).
* Fix AO_stack_push_release to workaround bug in clang-1.1/x86 compiler.
* Fix AO_test_and_setXX in tests/list_atomic.template.
* Fix AO_test_and_set_full (gcc/x86[_64].h) to work-around a bug in LLVM v2.7.
* Fix AO_test_and_set_full on m68k.
* Fix __ARM_ARCH_5__ macro handling for Android NDK (ARMv7).
* Fix configure for Cygwin, mingw-w64/32.
* Fix configure to define __PIC__ macro explicitly if needed (GCC).
* Fix double_ptr_storage definition for GCC pre-v4 (x86_64).
* Fix for x32 by removing 'q' suffix in x86-64 instructions.
* Fix generalization for IA-64 (regarding AO_or, AO_..._read/write primitives)
* Fix generalized AO_<type>_fetch_and_add() return type.
* Fix test_atomic_include for the case of missing CAS primitive.
* Fix test_malloc - allocate less memory in case of missing mmap.
* Implement the basic atomic primitives for the hexagon CPU.
This changes the buildlink3.mk files to use an include guard for the
recursive include. The use of BUILDLINK_DEPTH, BUILDLINK_DEPENDS,
BUILDLINK_PACKAGES and BUILDLINK_ORDER is handled by a single new
variable BUILDLINK_TREE. Each buildlink3.mk file adds a pair of
enter/exit marker, which can be used to reconstruct the tree and
to determine first level includes. Avoiding := for large variables
(BUILDLINK_ORDER) speeds up parse time as += has linear complexity.
The include guard reduces system time by avoiding reading files over and
over again. For complex packages this reduces both %user and %sys time to
half of the former time.
The latest code is only found in the boehm-gc CVS repository (see
Makefile.version for information). I had to take this from there,
as there were bug fixes.
This library was imported primarily for the [hopefully]
upcoming audio/pulseaudio package.
TODO: Add support for NetBSD atomic ops?
This package provides semi-portable access to hardware provided
atomic memory operations. These might allow you to write code:
- That does more interesting things in signal handlers.
- Makes more effective use of multiprocessors by allowing you to write
clever lock-free code. Note that such code is very difficult to get
right, and will unavoidably be less portable than lock-based code. It
ia also not always faster than lock-based code. But it may occasionally
be a large performance win.
- To experiment with new and much better thread programming paradigms, etc.
As the latest code for this library is only available via CVS, this package
provides a snapshot.
