Use GCC atomic intrinsics for MIPS (GCC 4.9+ and clang 3.5+)

* src/atomic_ops/sysdeps/gcc/mips.h: Include generic.h and do not
include all_aligned_atomic_load_store.h if AO_PREFER_BUILTIN_ATOMICS or
GCC 4.9+ or clang 3.5+ unless AO_DISABLE_GCC_ATOMICS.
* src/atomic_ops/sysdeps/gcc/mips.h (AO_MIPS_SET_ISA, AO_MIPS_LL_1,
AO_MIPS_SC, AO_MIPS_LL, AO_nop_full, AO_fetch_and_add, AO_test_and_set,
AO_compare_and_swap, AO_fetch_compare_and_swap): Do not define if
generic.h included.

Implement CAS_acquire/release/full using __atomic_compare_exchange_n (gcc)

* src/atomic_ops/sysdeps/gcc/generic-small.h: Regenerate.
* src/atomic_ops/sysdeps/gcc/generic-small.template
(AO_XSIZE_fetch_compare_and_swap, AO_XSIZE_compare_and_swap): Use
__atomic_compare_exchange_n(RELAXED) instead of
__sync_val/bool_compare_and_swap.
* src/atomic_ops/sysdeps/gcc/generic.h (AO_double_compare_and_swap):
Likewise.
* src/atomic_ops/sysdeps/gcc/generic-small.template
[AO_GCC_HAVE_XSIZE_SYNC_CAS] (AO_XSIZE_fetch_compare_and_swap_acquire,
AO_XSIZE_fetch_compare_and_swap_release,
AO_XSIZE_fetch_compare_and_swap_full): New inline function; remove
TODO item.
* src/atomic_ops/sysdeps/gcc/generic-small.template
[AO_GCC_HAVE_XSIZE_SYNC_CAS && !AO_GENERALIZE_ASM_BOOL_CAS]
(AO_XSIZE_compare_and_swap_acquire, AO_XSIZE_compare_and_swap_release,
AO_XSIZE_compare_and_swap_full): Likewise.
* src/atomic_ops/sysdeps/gcc/generic.h [AO_GCC_HAVE_double_SYNC_CAS]
(AO_double_compare_and_swap_acquire,
AO_double_compare_and_swap_release, AO_double_compare_and_swap_full):
Likewise.
* src/atomic_ops/sysdeps/gcc/generic.h (AO_compare_and_swap_acquire):
Reformat code.

Avoid atomic_compare_exchange_n if no __GCC_HAVE_SYNC_COMPARE_AND_SWAP_n

* src/atomic_ops/sysdeps/gcc/aarch64.h (AO_GCC_FORCE_HAVE_CAS,
AO_GCC_HAVE_double_SYNC_CAS): Define macro before include generic.h
if __clang__ (workaround).
* src/atomic_ops/sysdeps/gcc/generic-small.h: Regenerate.
* src/atomic_ops/sysdeps/gcc/generic-small.template
(AO_XSIZE_fetch_compare_and_swap, AO_XSIZE_compare_and_swap): Do not
define unless AO_GCC_HAVE_XSIZE_SYNC_CAS.
* src/atomic_ops/sysdeps/gcc/generic.h (AO_GCC_HAVE_char_SYNC_CAS,
AO_GCC_HAVE_short_SYNC_CAS, AO_GCC_HAVE_int_SYNC_CAS,
AO_GCC_HAVE_SYNC_CAS): New internal macro (based on
__GCC_HAVE_SYNC_COMPARE_AND_SWAP_<n> or AO_GCC_FORCE_HAVE_CAS presence).
* src/atomic_ops/sysdeps/gcc/generic.h (AO_GCC_HAVE_double_SYNC_CAS):
New internal macro if AO_HAVE_DOUBLE_PTR_STORAGE.
* src/atomic_ops/sysdeps/gcc/generic.h (AO_double_compare_and_swap):
Check AO_GCC_HAVE_double_SYNC_CAS instead of AO_HAVE_DOUBLE_PTR_STORAGE.

Eliminate redundant lwsync 2nd call in CAS_full on fail (gcc/PowerPC)

* src/atomic_ops/sysdeps/gcc/powerpc.h (AO_compare_and_swap_full,
AO_fetch_compare_and_swap_full): Do not call AO_lwsync 2nd time if
CAS is not successful.

Define AO_TS_SET as __GCC_ATOMIC_TEST_AND_SET_TRUEVAL if applicable

* src/atomic_ops/sysdeps/gcc/aarch64.h (AO_GCC_ATOMIC_TEST_AND_SET):
Define (before include test_and_set_t_is_ao_t.h).
* src/atomic_ops/sysdeps/gcc/nios2.h (AO_GCC_ATOMIC_TEST_AND_SET):
Likewise.
* src/atomic_ops/sysdeps/test_and_set_t_is_ao_t.h: Refine comment.
* src/atomic_ops/sysdeps/test_and_set_t_is_char.h: Likewise.
* src/atomic_ops/sysdeps/test_and_set_t_is_ao_t.h (AO_TS_SET_TRUEVAL):
New internal macro (defined via __GCC_ATOMIC_TEST_AND_SET_TRUEVAL).
* src/atomic_ops/sysdeps/test_and_set_t_is_char.h (AO_TS_SET_TRUEVAL):
Likewise.
* src/atomic_ops/sysdeps/test_and_set_t_is_ao_t.h (AO_TS_val): Use
AO_TS_SET_TRUEVAL instead of 1.
* src/atomic_ops/sysdeps/test_and_set_t_is_char.h (AO_BYTE_TS_val):
Use AO_TS_SET_TRUEVAL instead of 0xff; reformat code.
* src/atomic_ops/sysdeps/test_and_set_t_is_char.h (AO_TS_t): Move down
(to be after AO_TS_SET).

New macro AO_PREFER_BUILTIN_ATOMICS to rely on C11 atomics fully (AArch64)

If AO_PREFER_BUILTIN_ATOMICS is defined then inline assembly
implementation is not used.

* src/atomic_ops/sysdeps/gcc/aarch64.h (AO_SKIPATOMIC_double_load,
AO_SKIPATOMIC_double_load_acquire): Define if AO_PREFER_BUILTIN_ATOMICS.
* src/atomic_ops/sysdeps/gcc/aarch64.h (AO_nop_write, AO_double_load,
AO_double_load_acquire, AO_double_store, AO_double_store_release,
AO_double_compare_and_swap, AO_double_compare_and_swap_acquire,
AO_double_compare_and_swap_release, AO_double_compare_and_swap_full):
Do not define if AO_PREFER_BUILTIN_ATOMICS.
* src/atomic_ops/sysdeps/gcc/aarch64.h (AO_nop_write): Add comment;
remove TODO.
* src/atomic_ops/sysdeps/gcc/generic.h: Remove TODO about including
this header for other targets.
* src/atomic_ops/sysdeps/gcc/generic.h (AO_double_load): Do not define
if AO_SKIPATOMIC_double_load.
* src/atomic_ops/sysdeps/gcc/generic.h (AO_double_load_acquire): Do not
define if AO_SKIPATOMIC_double_load_acquire.

Remove obsolete information from README about C++0x standard future

* doc/README.txt (Future directions): Remove.

Update ChangeLog file

Update ChangeLog file (v7.4 changes)

Update ChangeLog file
(v7.2 - v7.4 changes only)

Update AUTHORS file

Add initial nios2 architecture support

* src/Makefile.am (nobase_private_HEADERS): Add nios2.h.
* src/atomic_ops.h: Include nios2.h if __nios2__.
* src/atomic_ops/sysdeps/gcc/nios2.h: New file.

Update AUTHORS file

Initial ibmc/powerpc (xlc) support

* src/atomic_ops/sysdeps/ibmc/powerpc.h: Include
test_and_set_t_is_ao_t.h.
* src/atomic_ops/sysdeps/ibmc/powerpc.h (AO_test_and_set,
AO_fetch_compare_and_swap): Implement.

Eliminate 'signed-to-unsigned value extension' compiler warning in AO_malloc

* src/atomic_ops_malloc.c (msbs): Change type from int to unsigned
char.
* src/atomic_ops_malloc.c (msb): Change return type from int to
unsigned.
* src/atomic_ops_malloc.c (msb, AO_malloc): Change type of v, result,
log_sz local variables from int to unsigned.

Support NaCl/arm

* src/atomic_ops/sysdeps/gcc/arm.h (AO_MASK_PTR, AO_BR_ALIGN): New
macro (empty unless __native_client__).
* src/atomic_ops/sysdeps/gcc/arm.h (AO_THUMB_GO_ARM,
AO_THUMB_RESTORE_MODE, AO_store, AO_char_store, AO_short_store,
AO_test_and_set, AO_fetch_and_add, AO_fetch_and_add1,
AO_fetch_and_sub1, AO_and, AO_or, AO_xor, AO_char_fetch_and_add,
AO_short_fetch_and_add, AO_compare_and_swap,
AO_fetch_compare_and_swap): Insert AO_BR_ALIGN after label.
* src/atomic_ops/sysdeps/gcc/arm.h (AO_store, AO_char_store,
AO_short_store, AO_test_and_set, AO_fetch_and_add, AO_fetch_and_add1,
AO_fetch_and_sub1, AO_and, AO_or, AO_xor, AO_char_fetch_and_add,
AO_short_fetch_and_add, AO_compare_and_swap, AO_fetch_compare_and_swap,
AO_double_load, AO_double_store, AO_double_compare_and_swap,
AO_test_and_set_full): Insert AO_MASK_PTR before each pointer access.

Remove TODO file (items moved to Issues on Github)

Fix GCC 5.x compatibility for AArch64 double-wide primitives

Small change to make aarch64 support work with gcc 5.x. Otherwise it
demands gcc -latomic and still fails test_stack.

* src/atomic_ops/sysdeps/gcc/aarch64.h (AO_double_load,
AO_double_load_acquire, AO_double_store, AO_double_store_release,
AO_double_compare_and_swap, AO_double_compare_and_swap_acquire,
AO_double_compare_and_swap_release, AO_double_compare_and_swap_full):
Use assembly implementation even for GCC v5+ (not just GCC 4.x).

New configure option (--enable-werror) to treat warnings as compiler errors

* configure.ac (werror): Test --enable-werror option to add -Werror
to CFLAGS.

Update AUTHORS file

Support n32 ABI for mips64

Without the proposed change, the existing preprocessor checks for
mips64 support do not account for the n32 ABI (when building with
"gcc -mabi=n32", test_atomic and test_stack both fail, and
test_malloc appears to hang indefinitely while spinning at 100%).

* src/atomic_ops/sysdeps/gcc/mips.h (AO_MIPS_SET_ISA, AO_MIPS_LL_1,
AO_MIPS_SC): Define depending on _MIPS_SIM value.

Update ChangeLog file

Update ChangeLog file
(v7.2 - v7.4 changes only)

Update AUTHORS file

Update AUTHORS file

Minor fix of code alignment in mips AO_compare_and_swap

Remove inclusion of acquire_release_volatile.h on mips

I'm not entirely sure how this ended up here. It might have been the case
that MIPS processors used to do this, but the ISA manuals don't say anything
about volatile loads / stores having aquire / release semnatics, so just
remove it to be safe.

Use LLD and SCD instructions on mips64

Update AUTHORS file

Relax shareability domain for dmb st in AO_nop_write (ARM/AArch64)

* src/atomic_ops/sysdeps/gcc/aarch64.h (AO_nop_write): Replace "dmb st"
to "dmb ishst" (this way the barriers will occupy the inner shareable
domain).
* src/atomic_ops/sysdeps/gcc/arm.h (AO_nop_write): Likewise.

Fix missing output folder on making auto-generated test files (Automake)

* tests/Makefile.am (test_atomic_include.h, list_atomic.c,
list_atomic.i): Add mkdir for output folder.

Update AUTHORS file

Fix missing output folder on making auto-generated headers (Automake)

Create the directories before sed needs them (otherwise build fails
when sed tries to redirect output to a non-existent directory in case
of build root folder is not identical to source root folder).

* src/Makefile.am (atomic_ops/generalize-small.h,
atomic_ops/generalize-arithm.h, atomic_ops/sysdeps/ao_t_is_int.h,
atomic_ops/sysdeps/gcc/generic-arithm.h,
atomic_ops/sysdeps/loadstore/*.h): Add mkdir for output folder.

Eliminate 'variable set but not used' Cppcheck warnings in test_stack

* tests/test_stack.c (run_one_test): Define "j" local variable only
if VERBOSE.
* tests/test_stack.c (main): Do not define "sum" local variable (and
don not compute the sum) if NO_TIMES.

Fix missing casts to match printf format specifier in test_atomic

* tests/test_atomic.c (test_and_set_thr): Add missing casts to long
for locked_counter (which is unsigned long) to match printf format
specifier.

.gitignore: Ignore .dirstamp

.gitignore: Ignore tests/.libs (generated if configure --enable-shared)

Fix makefile preventing AO_pause undefined in libatomic_ops_gpl

* src/Makefile.am (libatomic_ops_gpl_la_LIBADD): New dependency.

Restore contribution info in ChangeLog for authors not listed in git log
(deleted in commit 36a9c66)

* ChangeLog (7.0, 1.0): Add information about contribution of persons
mentioned in AUTHORS but not recorded in git log.
* src/atomic_ops/sysdeps/gcc/cris.h: Move contribution information
from comment to ChangeLog.

Update ChangeLog file

Merge pull request #8 from badboy/patch-1

Fix a typo in doc/README.txt (remove redundant "an" article)

Removed unnecessary "an"

Update emails/links due to project site transition

* AUTHORS: Update email.
* README.md: Update links to BDWGC/libatomic_ops site.
* doc/README_stack.txt: Likewise.
* README.md: Update mailing list info.
* configure.ac: Likewise.

Bump atomic_ops version

* README.md: Bump version to 7.5.0.
* configure.ac: Likewise.
* src/atomic_ops/ao_version.h (AO_VERSION_MINOR): Likewise.

[7.4.0]

Bump version to 7.4.0; change policy regarding version numbers

* ChangeLog: Update.
* README.md: Bump version.
* configure.ac (AC_INIT): Likewise.
* src/atomic_ops/ao_version.h (AO_VERSION_MINOR): Likewise.
* src/atomic_ops/ao_version.h (AO_VERSION_ALPHA): Replace to
AO_VERSION_MICRO; add comment.

Put README.md to 'dist' package (Automake)

* Makefile.am (dist_pkgdata_DATA): Add README.md item.

Update ChangeLog file

Convert README to Markdown format

* README: Rename to README.md; adjust formatting to follow Markdown
syntax.

Remove broken link to INSTALL file from README

* README (Installation and Usage): Remove reference to INSTALL file
(not generated by autoreconf as AM_INIT_AUTOMAKE specifies "foreign"
strictness).
* .gitignore (INSTALL): Remove.

Put autogen.sh to 'dist' package (Automake)

* Makefile.am (EXTRA_DIST): Add "autogen.sh" item.

.gitignore: Ignore test-driver (generated by autoreconf)

.gitignore: Ignore test*.log, test*.trs
(Apply commit 5cce9c5 from 'release-7_2' branch.)

Conflicts:

    .gitignore

Update ChangeLog file

Do not use LDREXD/STREXD for Clang3.2/arm (and earlier)

* src/atomic_ops/sysdeps/gcc/arm.h (AO_ARM_HAVE_LDREXD): Do not define
for pre-Clang3.3 (since it does not allocate register pairs for
LDREXD/STREXD instructions properly); add comment.

Fix ARM char/short fetch_and_add and double-CAS operands width (GCC/Clang)

* src/atomic_ops/sysdeps/gcc/arm.h (AO_char_fetch_and_add,
AO_short_fetch_and_add): Use 32-bit int type for "tmp" and "result"
local variables instead of char/short type (resolve Clang3.3 warning
"size being stored is truncated, use a modifier to specify the size"
enabled by -Wasm-operand-widths compiler option); cast "incr"
argument to int to prevent Clang3.2 warning about value
truncation.
* src/atomic_ops/sysdeps/gcc/arm.h (AO_double_compare_and_swap): Swap
assembly code operands ("new_val.AO_whole" and "addr") to prevent
Clang3.3 warning about operand truncation.

Fix ARMv7 LDREXD/STREXD double-wide operand specification (GCC/Clang)

* src/atomic_ops/sysdeps/gcc/arm.h (AO_double_load, AO_double_store,
AO_double_compare_and_swap): Specify that LDREXD and STREXD use 2
adjacent registers (thus preventing Clang3.3 from register allocation
failures leading to "registers may not be the same" or
"even register required" GAS errors).

Tests: workaround GCC 4.4.3 warning reported for list_atomic.c 'val'

* tests/list_atomic.template (XSIZE_list_atomicXX): Use "static"
modifier for "val" local variable to initialize it (cannot use direct
assignment to zero because "val" type could be a SIMD vector type if
XCTYPE is AO_double_t) to workaround GCC v4.4.3
"'val' is used uninitialized in this function" warning reported for
AO_load.

Support ARMv8 target (gcc/arm)

* src/atomic_ops/sysdeps/gcc/arm.h (__ARM_ARCH_8A__): Detect new macro
(treated same as __ARM_ARCH_7A__).
* src/atomic_ops/sysdeps/gcc/arm.h (AO_ARM_HAVE_SWP): Do not define
for ARMv8 (since SWP{B} obsoleted); add comment.
* src/atomic_ops/sysdeps/gcc/arm.h (AO_compare_and_swap): Add TODO
item (for deprecated IT block containing wide Thumb instruction).

tests: Fix makefile sed rule for list_atomic

* tests/Makefile.am (test_atomic_include.h, list_atomic.c): Replace
all occurrences of 'XX' (instead of only the first one) on every line.

tests: Add char/short/int/AO_double_t and dd_acquire cases to list_atomic

* tests/Makefile.am (list_atomic.c): Iterate over XSIZE/XCTYPE (in
addition to XX); expand XX also as _dd_acquire_read.
* tests/list_atomic.template (XSIZE_list_atomicXX): Define 'val' local
variable only if used (needed for AO_double_t case).

tests: Add and/or/xor entries to list_atomic

* tests/list_atomic.template (list_atomicXX): Add AO_XSIZE_andXX,
AO_XSIZE_orXX, AO_XSIZE_xorXX invocations.

tests: Parameterize list_atomic template with XSIZE

* tests/Makefile.am (list_atomic.c): Substitute XSIZE/XCTYPE for AO_t.
* tests/list_atomic.template (list_atomicXX): Parameterize with XSIZE
and XCTYPE (except for AO_nop and AO_test_and_set).

list_atomic: Add makefile rule to test list_atomic.template syntax

* tests/Makefile.am (BUILT_SOURCES, CLEANFILES): Add "list_atomic.o".
* tests/Makefile.am (list_atomic.o): New rule (to verify list_atomic.c
is valid C code).
* tests/list_atomic.template (list_atomicXX): Use "volatile" for "val"
local variable.
* tests/list_atomic.template (list_atomicXX): Define "oldval", "newval",
"ts" and "incr" local variables only if used; cast string literals to
void (to prevent compiler warnings).
* tests/list_atomic.template (list_atomicXX): Use AO_t type for "incr"
local variable instead of C long type.
* tests/list_atomic.template (list_atomicXX): Define "oldval",
"newval", "incr" as statitc local variables, and remove explicit
initialization to zero (to void problems with initialization of these
variables for AO_double_t case if the latter is a struct).

tests: Do not use deprecated AO_T and AO_TS_T

* tests/list_atomic.template (list_atomicXX): Use AO_t instead of
deprecated AO_T.
* tests/list_atomic.template (list_atomicXX): Use AO_TS_t instead of
deprecated AO_TS_T.
* tests/test_atomic.c (lock): Likewise.

Fix typo in comment of generalize-small.template

* src/atomic_ops/generalize-arithm.template: Fix a typo in comment
("acquire").
* src/atomic_ops/generalize-arithm.h: Regenerate.

Add comment about store_full implementation for x86

* src/atomic_ops/sysdeps/gcc/x86.h: Add comment about AO_store_full
generalization.

Add non-generalized double-CAS-full implementation for AArch64

* src/atomic_ops/sysdeps/gcc/aarch64.h
(AO_double_compare_and_swap_full): New function.

Define AO_NO_PTHREADS in configure if no pthreads (Win32 and VxWorks)

* configure.ac (have_pthreads): Do not define to true if check for
pthread_self failed.
* configure.ac (AO_NO_PTHREADS): Define if not have_pthreads.

Better document configure THREADDLLIBS variable

* configure.ac (THREADDLLIBS, have_pthreads): Add comments.

Update AUTHORS

Enable build using toolchain without pthreads

* src/atomic_ops.c: Include pthread.h only if not AO_NO_PTHREADS (new
macro).
* src/atomic_ops.c (AO_pt_lock): Define only unless AO_NO_PTHREADS.
* tests/test_atomic.c (main): Skip test_atomic_pthreads if both
AO_NO_PTHREADS and AO_USE_PTHREAD_DEFS are defined (preventing test
link failure if AO_NO_PTHREADS passed via CFLAGS).

Add comment about double-wide load/store on x86_64 (GCC)

* src/atomic_ops/sysdeps/gcc/x86.h: Add comment about AO_double_load
and AO_double_store implementation in 64-bit mode; remove the
corresponding TODO item.

Fix generalized store_release_write for double-wide type on x64 (GCC)

* src/atomic_ops/generalize-small.template
(AO_XSIZE_store_release_write): Move store-write-based definition down
to follow AO_XSIZE_store_write one (since AO_double_store is currently
defined via CAS for gcc/x64).
* src/atomic_ops/generalize-small.h: Regenerate.

Update AUTHORS

Fix AM_CONFIG_HEADER in configure for autoconf-2.69-1

* configure.ac (AM_CONFIG_HEADER): Replace obsolete macro with with
AC_CONFIG_HEADERS one.

Generalize double_load/store for x86_64 (GCC)

* src/atomic_ops.h (AO_GENERALIZE_TWICE): Define in case of missing
double-wide load or store primitives while double-wide CAS is present
(required for x86-64, at least).

Fix fetch_compare_and_swap asm constraints in gcc/x86.h

* src/atomic_ops/sysdeps/gcc/x86.h (AO_fetch_compare_and_swap_full):
Use same asm constraints for old_val and new_val as in
AO_compare_and_swap_full.

Reformat asm instructions of sunc/x86.h

* src/atomic_ops/sysdeps/sunc/x86.h (AO_fetch_and_add_full,
AO_char_fetch_and_add_full, AO_short_fetch_and_add_full, AO_and_full,
AO_or_full, AO_xor_full, AO_compare_and_swap_full): Reformat code.

Fix asm constraints of primitives in sunc/x86.h

* src/atomic_ops/sysdeps/sunc/x86.h (AO_fetch_and_add_full,
AO_char_fetch_and_add_full, AO_short_fetch_and_add_full, AO_and_full,
AO_or_full, AO_xor_full, AO_test_and_set_full,
AO_compare_and_swap_full, AO_compare_double_and_swap_double_full,
AO_int_fetch_and_add_full): Use "+m" asm constraint for *addr instead
of "=m" (because the value pointed by addr is read and written by the
code).
* src/atomic_ops/sysdeps/sunc/x86.h (AO_fetch_compare_and_swap_full):
Use same asm constraints and parameters order as in
AO_compare_and_swap_full.

Merge branch 'add-aarch64-support'

Fix asm constraint of CAS memory operand for gcc/alpha

* src/atomic_ops/sysdeps/gcc/alpha.h (AO_compare_and_swap): Use "+m"
asm constraint for *addr instead of "=m" (because the value pointed by
addr is read and written by the code).

Fix asm constraint of fetch_and_add, test_and_set, fetch_CAS for MIPS (gcc)

* src/atomic_ops/sysdeps/gcc/mips.h (AO_fetch_and_add,
AO_test_and_set, AO_fetch_compare_and_swap): Use "+m" asm constraint
for *addr instead of "=m" (because the value pointed by addr is read
and written by the code).

Fix asm constraint of compare_and_swap for clang-3.1/mips

* src/atomic_ops/sysdeps/gcc/mips.h (AO_compare_and_swap): Use "+m"
asm constraint for *addr instead of "+R" (the letter is unsupported
by clang3.1 resulting in "invalid output constraint in asm" error).

Always use 'mfence' for nop_full if target CPU supports SSE2 (gcc/x86)

* src/atomic_ops/sysdeps/gcc/x86.h (AO_USE_PENTIUM4_INSTRS): Define
if __SSE2__ is defined; update comment (and remove TODO item).

Fix assembly comment prefix and identify memory clobbers for AArch64

* src/atomic_ops/sysdeps/gcc/aarch64.h: Revert comments beginning
marker to '//' (since '@' is not supported for A64 assembler).
* src/atomic_ops/sysdeps/gcc/aarch64.h (AO_double_load): Add comment
about the need for STXP instruction; remove TODO item.
* src/atomic_ops/sysdeps/gcc/aarch64.h (AO_double_store): Add comment
about absence of "cc" in clobber lists; remove TODO item.
* src/atomic_ops/sysdeps/gcc/aarch64.h (AO_double_load,
AO_double_load_acquire, AO_double_store, AO_double_store_release,
AO_double_compare_and_swap, AO_double_compare_and_swap_acquire,
AO_double_compare_and_swap_release): Use Q constraint (instead of "r")
for 'addr' argument to instruct the compiler about memory clobbers.

Use AO_EXPECT_FALSE for AArch64 double_load/store

* src/atomic_ops/sysdeps/gcc/aarch64.h (AO_HAVE_DOUBLE_PTR_STORAGE):
Do not check (since defined in standard_ao_double_t.h for AArch64).
* src/atomic_ops/sysdeps/gcc/aarch64.h (AO_double_load): Add TODO
items.
* src/atomic_ops/sysdeps/gcc/aarch64.h (AO_double_load,
AO_double_load_acquire, AO_double_store, AO_double_store_release):
Use AO_EXPECT_FALSE.

Workaround missing built-in double-wide primitives for AArch64 in gcc-4.7

* src/atomic_ops/sysdeps/gcc/aarch64.h (AO_double_load,
AO_double_load_acquire, AO_double_store, AO_double_store_release,
AO_double_compare_and_swap, AO_double_compare_and_swap_acquire,
AO_double_compare_and_swap_release): Define using inline assembly
(because GCC 4.7-2013.01 has no support of 16-byte atomic operations).
* src/atomic_ops/sysdeps/gcc/generic.h (AO_double_load,
AO_double_load_acquire, AO_double_store, AO_double_store_release,
AO_double_compare_and_swap): Do not define if the corresponding
AO_HAVE_double_X macro is defined.

gcc/x86.h: Add TODO items

Merge remote-tracking branch 'origin/master' into add-aarch64-support

Minimize gcc/generic-arithm template by factoring out barriers
(code refactoring)

* src/Makefile.am (atomic_ops/sysdeps/gcc/generic-arithm.h): Iterate
over XBAR/XGCCBAR (in addition to XSIZE/XCTYPE).
* src/atomic_ops/sysdeps/gcc/generic-arithm.template: Parametrize all
primitives with XBAR (replacing AO none/acquire/release/full barriers)
and with XGCCBAR (replacing GCC atomic RELAXED/ACQUIRE/RELEASE/SEQ_CST
barriers, respectively).
* src/atomic_ops/sysdeps/gcc/generic-arithm.h: Regenerate.

Implement char/short/int-wide primitives using GCC built-in atomic/sync

* src/Makefile.am (EXTRA_DIST): Add gcc/generic-arithm.template,
gcc/generic-small.template entries.
* src/Makefile.am (nobase_private_HEADERS): Move gcc/generic-arithm.h,
atomic_ops/sysdeps/gcc/generic-small.h to BUILT_SOURCES.
* src/Makefile.am (atomic_ops/sysdeps/gcc/generic-arithm.h,
atomic_ops/sysdeps/gcc/generic-small.h): New rules (to generate
AO_char/short/int_X primitives included from gcc/generic.h file.
* src/atomic_ops/sysdeps/gcc/generic-arithm.template: New template
file (based on code moved from generic-arithm.h).
* src/atomic_ops/sysdeps/gcc/generic-small.template: New template file
(based on code moved from generic-small.h).
* src/atomic_ops/sysdeps/gcc/generic-arithm.h: Regenerate.
* src/atomic_ops/sysdeps/gcc/generic-small.h: Likewise.

Move gcc-generic AO_t-wide primitives to generic-small/arithm headers
(code refactoring)

* src/atomic_ops/sysdeps/gcc/generic-small.h: New file.
* src/atomic_ops/sysdeps/gcc/generic-arithm.h: Likewise.
* src/atomic_ops/sysdeps/gcc/generic.h: Include generic-small.h file.
* src/atomic_ops/sysdeps/gcc/generic.h (AO_load, AO_load_acquire,
AO_store, AO_store_release, AO_fetch_compare_and_swap,
AO_compare_and_swap): Move to generic-small.h file.
* src/atomic_ops/sysdeps/gcc/generic.h: Include generic-arithm.h file
unless AO_PREFER_GENERALIZED.
* src/atomic_ops/sysdeps/gcc/generic.h (AO_fetch_and_add,
AO_fetch_and_add_acquire, AO_fetch_and_add_release,
AO_fetch_and_add_full, AO_and, AO_and_acquire, AO_and_release,
AO_and_full, AO_or, AO_or_acquire, AO_or_release, AO_or_full, AO_xor,
AO_xor_acquire, AO_xor_release, AO_xor_full): Move to generic-arithm.h
file.
* src/Makefile.am (nobase_private_HEADERS): Add gcc/generic-arithm.h,
gcc/generic-small.h entries.

Define nop_write as 'dmb st' for AArch64

* src/atomic_ops/sysdeps/gcc/aarch64.h (AO_nop_write): Define via asm
(dmb st) statement unless AO_UNIPROCESSOR (because it seems there is
no equivalent definition using compiler __atomic_thread_fence).
* src/atomic_ops/sysdeps/gcc/generic.h (AO_nop_write): Do not define
if already is.

Fix AO_double_t definition for AArch64

* src/atomic_ops/sysdeps/standard_ao_double_t.h (double_ptr_storage):
Define explicitly for AArch64 (use __int128 instead of long long).

Fix double_compare_and_swap compile-time error in gcc/generic.h

* src/atomic_ops/sysdeps/gcc/generic.h (AO_double_compare_and_swap):
Fix access to old_val value.

Remove improper comment for AO_XSIZE_load_read

* src/atomic_ops/generalize-small.template (AO_XSIZE_load_read):
Remove improper comment (because nop barrier type matches that of
load primitive).
* src/atomic_ops/generalize-small.template (AO_XSIZE_store_write):
Reorder AO_HAVE_nop_write and AO_HAVE_XSIZE_store checking (to match
that for other primitives in this file).
* src/atomic_ops/generalize-small.h: Regenerate.

Implement AO_and/or/xor efficiently for ARM

* src/atomic_ops/sysdeps/gcc/arm.h (AO_and, AO_or, AO_xor): Implement
directly using LDREX/STREX (enabled only if not AO_PREFER_GENERALIZED).

Add compile-time assertion for size of 'standard' AO_double_t

* src/atomic_ops/sysdeps/standard_ao_double_t.h
(AO_double_t_size_static_assert): Add compile-time assertion for
AO_double_t size.

Fix (remove) invalid include of read_ordered.h for ARM
(fix commit 2df9c1e partially)

* src/atomic_ops/sysdeps/gcc/arm.h: Do not include read_ordered.h
(because load_acquire should contain a DMB instruction in a multi-core
case, new load_acquire/read primitives implementation is generalized
using nop_full/read which either contain a DMB instruction, or is just
a compiler barrier for uniprocessor).
* src/atomic_ops/sysdeps/armcc/arm_v6.h: Likewise.
* src/atomic_ops/sysdeps/msftc/arm.h: Likewise.

Revert "Fix read_ordered.h inclusion for ARM"
(because it makes load_acquire/read incorrect for multi-core ARM)

This reverts commit fcc838be49c98486c26738764f895d34de0716be.