Ivan Maidanski [Fri, 27 Jan 2017 16:35:20 +0000 (19:35 +0300)]
Use generalized load/store if AO_PREFER_GENERALIZED (gcc/generic)
Note: AO_PREFER_GENERALIZED macro is for testing generalize.h mainly.
* src/atomic_ops/sysdeps/gcc/generic-small.template (AO_XSIZE_load,
AO_XSIZE_load_acquire, AO_XSIZE_store, AO_XSIZE_store_release): Do not
define if AO_GCC_HAVE_XSIZE_SYNC_CAS and AO_PREFER_GENERALIZED.
* src/atomic_ops/sysdeps/gcc/generic-small.h: Regenerate.
* src/atomic_ops/sysdeps/gcc/generic.h [AO_HAVE_DOUBLE_PTR_STORAGE]
(AO_GCC_HAVE_double_SYNC_CAS): Move the definition upper to be before
AO_double_load.
* src/atomic_ops/sysdeps/gcc/generic.h [AO_HAVE_DOUBLE_PTR_STORAGE]
(AO_double_load, AO_double_load_acquire, AO_double_store,
AO_double_store_release): Do not define if AO_GCC_HAVE_double_SYNC_CAS
and AO_PREFER_GENERALIZED.
Ivan Maidanski [Thu, 26 Jan 2017 20:07:57 +0000 (23:07 +0300)]
Enable limited testing in Makefile.msft without Cygwin
New target "check-noautogen" is introduced in Makefile.msft to build and
run only test_malloc and test_stack (which do not need test code
auto-generation by sed).
* src/Makefile.msft: Refine comment about tests execution; add comment
about "check-noautogen" target.
* src/Makefile.msft (check): Add dependency on check-noautogen;
move test_malloc and test_stack to check-noautogen.
* src/Makefile.msft (check-noautogen): New target (executes test_malloc
and test_stack only).
Ivan Maidanski [Thu, 26 Jan 2017 07:38:21 +0000 (10:38 +0300)]
Implement AO_CLEAR using C11 atomic intrinsic (GCC)
* src/atomic_ops/sysdeps/gcc/generic.h [!AO_PREFER_GENERALIZED]
(AO_CLEAR): Implement using GCC built-in __atomic_clear (thus
overriding the default one based on AO_char_store_release).
Hans Boehm [Thu, 26 Jan 2017 07:16:54 +0000 (10:16 +0300)]
Implement compiler_barrier using C11 __atomic_signal_fence (GCC)
Use __atomic_signal_fence(__ATOMIC_SEQ_CST) as AO_compiler_barrier.
That was the intended use of __atomic_signal_fence. The only
standards-relevant context in which you can tell apart a compiler and
real fence is in the case if signal handlers.
* src/atomic_ops/sysdeps/gcc/generic.h (AO_compiler_barrier): Redefine
using GCC atomic intrinsic (__atomic_signal_fence).
Ivan Maidanski [Tue, 24 Jan 2017 19:46:57 +0000 (22:46 +0300)]
Fix test_atomic failure caused unaligned AO_double_t access on x86
The failure is caused by violation of an assertion that checks
AO_double_t variable is 8-byte aligned on x86.
* doc/README.txt (AO_double_t): Add note about required alignment.
* src/atomic_ops/sysdeps/standard_ao_double_t.h (AO_double_t): Add
comment about alignment.
* src/atomic_ops_stack.h (AO_stack_t): Likewise.
* tests/test_atomic_include.template (test_atomicXX): Define old_w, w
local variables as static (as otherwise, e.g., they could have 4-byte
alignment on x86); add comment.
The failure is actually caused by the fact that default alignment of
stack on x86 is 4 bytes.
So, another solution should be proposed to fix the issue for other
compilers (and, potentially, other targets). E.g. do not place
AO_double_t volatile variables on stack.
Ivan Maidanski [Fri, 20 Jan 2017 08:58:13 +0000 (11:58 +0300)]
Fix test_atomic failure caused unaligned AO_double_t access on x86 (VC++)
Passing an unaligned AO_double_t pointer to AO double-wide primitives
results in an undefined behavior of the latter when running on x86 (or
violation of the corresponding assertion on the alignment).
MS VC++/x86 aligns AO_double_t values on a 4-byte boundary by default.
For the proper alignment, __declspec(align(8)) should be applied to
variables defined in the client code (which uses double-wide AO
primitives). Unfortunately, the attribute cannot be added to
AO_double_t definition itself because the compiler does not allow the
attribute for function arguments.
This patch introduces AO_DOUBLE_ALIGN attribute for use by clients of
the double-wide AO primitives (and, thus, AO_stack clients).
Matters only Visual Studio compiler for X86.
The inner clients (atomic_ops_malloc, test_atomic, test_stack) are
updated to use this attribute.
* doc/README_win32.txt [x86] (AO_DOUBLE_ALIGN): Document.
* src/atomic_ops/sysdeps/generic_pthread.h (AO_DOUBLE_ALIGN): Define
(as empty).
* src/atomic_ops/sysdeps/standard_ao_double_t.h [!_WIN64 && _WIN32
&& !__GNUC__ && _MSC_VER] (AO_DOUBLE_ALIGN): Define as declspec
align(8); document it.
* src/atomic_ops/sysdeps/standard_ao_double_t.h [!AO_DOUBLE_ALIGN]
(AO_DOUBLE_ALIGN): Define as empty (otherwise).
* src/atomic_ops_stack.h [AO_USE_ALMOST_LOCK_FREE && !AO_DOUBLE_ALIGN]
(AO_DOUBLE_ALIGN): Likewise.
* src/atomic_ops_malloc.c (AO_free_list): Use AO_DOUBLE_ALIGN attribute.
* tests/test_stack.c (the_list): Likewise.
* src/atomic_ops_stack.h [!AO_USE_ALMOST_LOCK_FREE] (AO_stack_t):
Document AO_DOUBLE_ALIGN usage (by clients).
* tests/test_atomic_include.template (test_atomicXX): Use
AO_DOUBLE_ALIGN attribute for old_w and w double-wide local variables
(to avoid alignment assertion violation or AO primitives undefined
behavior on x86 if the test code is compiled by VC++).
Ivan Maidanski [Fri, 20 Jan 2017 00:03:54 +0000 (03:03 +0300)]
Add assertion that double-wide CAS target is aligned (msftc/x86[_64])
* src/atomic_ops/sysdeps/msftc/x86.h [AO_ASSUME_VISTA]
(AO_double_compare_and_swap_full): Add assert() that addr pointer is
aligned according to the size of AO_double_t.
* src/atomic_ops/sysdeps/msftc/x86_64.h [AO_CMPXCHG16B_AVAILABLE
&& _MSC_VER>=1500] (AO_compare_double_and_swap_double_full): Likewise.
Ivan Maidanski [Thu, 19 Jan 2017 07:03:59 +0000 (10:03 +0300)]
Fix turn off optimization in case of configure --enable-gcov
(fix commit 0d0a16d)
* configure.ac [enable_gcov] (CFLAGS): Change sed regular expression
to avoid [] (as they are not properly handled during autogen), and to
also filter out -Os and -Ofast (-O0 is ignored).
Ivan Maidanski [Wed, 7 Dec 2016 09:35:47 +0000 (12:35 +0300)]
Eliminate 'printf format specifies type void*' GCC pedantic warnings
This commit also eliminates 'cast from pointer to integer' compiler
warnings and '32-bit value shift followed by expansion to 64-bit'
static analysis tool warning.
* src/atomic_ops_malloc.c [AO_TRACE_MALLOC] (AO_malloc, AO_free):
Use %p (instead of %x) format specifier to print pthread_self() value
(and cast pthread_self() to void*).
* src/atomic_ops_malloc.c [AO_TRACE_MALLOC] (AO_malloc): Cast result+1
to void* (to match %p printf format specifier).
* src/atomic_ops_malloc.c [AO_TRACE_MALLOC] (AO_free): Use "UL" suffix
for 1 and remove the cast to unsigned long (instead of performing left
shift of an integer value and extending the result to unsigned long).
* src/atomic_ops/sysdeps/gcc/generic.h [AO_HAVE_DOUBLE_PTR_STORAGE
&& !AO_HAVE_double_store] (AO_double_store): Do not define if
AO_SKIPATOMIC_double_store.
* src/atomic_ops/sysdeps/gcc/generic.h [AO_HAVE_DOUBLE_PTR_STORAGE
&& !AO_HAVE_double_store_release] (AO_double_store_release): Do not
define if AO_SKIPATOMIC_double_load.
* src/atomic_ops/sysdeps/gcc/x86.h [AO_GCC_ATOMIC_TEST_AND_SET
&& __clang__ && !__x86_64__ && !__APPLE_CC__ && !__CYGWIN__
&& !AO_PREFER_BUILTIN_ATOMICS] (AO_SKIPATOMIC_double_load,
AO_SKIPATOMIC_double_load_acquire, AO_SKIPATOMIC_double_store,
AO_SKIPATOMIC_double_store_release): Define internal macro (undefined
at the end of file).
* src/atomic_ops/sysdeps/gcc/x86.h [AO_GCC_ATOMIC_TEST_AND_SET
&& (__clang__ && !__x86_64__ && !__APPLE_CC__ && !__CYGWIN__
&& !AO_PREFER_BUILTIN_ATOMICS) || (__APPLE_CC__ && __x86_64__))]
(AO_SKIPATOMIC_double_compare_and_swap_ANY): Move definition here
(remove at other places in this file); undefine at the end of the file.
* src/atomic_ops/sysdeps/gcc/x86.h [AO_GCC_ATOMIC_TEST_AND_SET]: Check
AO_SKIPATOMIC_double_compare_and_swap_ANY instead of checking multiple
macros to detect whether a workaround for the CAS (and others)
double-wide operations is needed.
Ivan Maidanski [Tue, 22 Nov 2016 01:18:08 +0000 (04:18 +0300)]
Report gcc/clang pedantic warnings (configure)
* configure.ac [GCC] (WPEDANTIC): New variable (set to
"-Wpedantic -Wno-long-long" if supported by the compiler).
* configure.ac [GCC] (CFLAGS): Add $WPEDANTIC.
Tobias Leich [Thu, 10 Nov 2016 08:02:56 +0000 (09:02 +0100)]
Implement fetch-CAS for s390[x] (gcc)
* src/atomic_ops/sysdeps/gcc/s390.h (AO_compare_and_swap_full): Do not
define if AO_GENERALIZE_ASM_BOOL_CAS.
* src/atomic_ops/sysdeps/gcc/s390.h (AO_fetch_compare_and_swap_full):
Implement; remove TODO item.
Ivan Maidanski [Tue, 1 Nov 2016 08:25:39 +0000 (11:25 +0300)]
Fix reporting about missing and/or/xor_dd_acquire_read (test_atomic)
AO_XSIZE_and/or/xor_dd_acquire_read are meaningless and not implemented,
thus their absence is not reported in the tests.
* tests/test_atomic_include.template [!AO_HAVE_orXX] (test_atomicXX):
Do not call MISSING(AO_or) if all AO_HAVE_or* excluding
AO_HAVE_or_dd_acquire_read are not defined.
* tests/test_atomic_include.template [!AO_HAVE_xorXX] (test_atomicXX):
Do not call MISSING(AO_xor) if all AO_HAVE_xor* excluding
AO_HAVE_xor_dd_acquire_read are not defined.
* tests/test_atomic_include.template [!AO_HAVE_andXX] (test_atomicXX):
Do not call MISSING(AO_and) if all AO_HAVE_and* excluding
AO_HAVE_and_dd_acquire_read are not defined.
This matters only if AO_double_compare_and_swap[_acquire_read] actually
defined in generalize.h too.
* src/atomic_ops/generalize.h
[!AO_HAVE_double_compare_and_swap_dd_acquire_read]
(AO_double_compare_and_swap_dd_acquire_read): Define (same as definition
as auto-generated in generalize-small.h).
Ivan Maidanski [Wed, 26 Oct 2016 08:57:13 +0000 (11:57 +0300)]
Workaround 'value of MAP_ANON unknown' cppcheck info message
(fix commit 4c303bc)
* src/atomic_ops_malloc.c [USE_MMAP_ANON && CPPCHECK] (OPT_MAP_ANON):
Define to immediate non-zero value (taken from Linux libc) instead of
MAP_ANON (because the latter is used in ifdef).
Ivan Maidanski [Tue, 18 Oct 2016 07:44:10 +0000 (10:44 +0300)]
Workaround 'value of macro is unknown' cppcheck information messages
* src/atomic_ops_malloc.c [USE_MMAP_ANON] (OPT_MAP_ANON): Define to 0
(do not use MAP_ANONYMOUS value) if CPPCHECK.
* src/atomic_ops_malloc.c [SIZE_MAX] (AO_SIZE_MAX): Do not use SIZE_MAX
value if CPPCHECK.
Ivan Maidanski [Mon, 17 Oct 2016 05:22:53 +0000 (08:22 +0300)]
Fix 'variable new value is never used' cppcheck style warning
Assertion for v value after fetch_and_sub1(&v) is added.
* tests/test_atomic_include.template (test_atomicXX): Call TA_assert
for s, b, zz values after they are decremented (use volatile pointer to
avoid "condition is always true" compiler warning).
Ivan Maidanski [Mon, 17 Oct 2016 05:08:18 +0000 (08:08 +0300)]
Workaround 'overflow in pointer subtraction' cppcheck warning
* src/atomic_ops_malloc.c (AO_free): Do not perform subtraction from
"p" argument if its value is null; change type of "base" local variable
from char* to AO_t*.
Ivan Maidanski [Tue, 11 Oct 2016 14:52:54 +0000 (17:52 +0300)]
Eliminate 'ISO C90 does not support long long' compiler pedantic warning
Long (32-bit) type is sufficient to represent time delta (duration)
in test_stack.
* tests/test_stack.c (get_msecs): Change return type from long long
to unsigned long (use lowest 32 bits of time value).
* tests/test_stack.c (main): Change type of start_time local variable
to unsigned long; remove redundant type cast.
Ivan Maidanski [Tue, 11 Oct 2016 07:01:35 +0000 (10:01 +0300)]
Eliminate 'scope of variable can be reduced' cppcheck warning in AO_stack
* src/atomic_ops_stack.c [AO_HAVE_compare_and_swap_double]
(AO_stack_push_release): Move local variable declaration to the inner
scope where the variable is actually used.
Jean Girardet [Mon, 10 Oct 2016 08:04:24 +0000 (11:04 +0300)]
Support AIX/ppc (gcc)
* src/atomic_ops/sysdeps/gcc/powerpc.h (AO_PPC_L, AO_PPC_BR_A): New
internal macros (undefined at the end of file); definition depends on
_AIX macro presence.
* src/atomic_ops/sysdeps/gcc/powerpc.h (AO_load_acquire,
AO_test_and_set, AO_compare_and_swap, AO_fetch_compare_and_swap,
AO_fetch_and_add): Use AO_PPC_L for label definitions; use AO_PPC_BR_A
to specify jump target.
Ivan Maidanski [Fri, 7 Oct 2016 16:27:09 +0000 (19:27 +0300)]
Code refactoring of gcc/powerpc.h to avoid code duplication
* src/atomic_ops/sysdeps/gcc/powerpc.h (AO_PPC_LD, AO_PPC_LxARX,
AO_PPC_CMPx, AO_PPC_STxCXd, AO_PPC_LOAD_CLOBBER): New internal macros
(undefined at the end of the file).
* src/atomic_ops/sysdeps/gcc/powerpc.h (AO_T_IS_INT): Move definition
upper (to have single check of __ppc64__).
* src/atomic_ops/sysdeps/gcc/powerpc.h (AO_load_acquire): Use AO_PPC_LD,
AO_PPC_LOAD_CLOBBER (and eliminate code duplication).
* src/atomic_ops/sysdeps/gcc/powerpc.h (AO_test_and_set,
AO_compare_and_swap, AO_fetch_compare_and_swap, AO_fetch_and_add): Use
AO_PPC_LxARX, AO_PPC_CMPx, AO_PPC_STxCXd (and eliminate code
duplication).
* tests/test_atomic_include.template (test_atomicXX): Treat "x"
variable as volatile one to suppress cppcheck warning that x==13 is
always true.
* tests/test_atomic_include.template (test_atomicXX): Make
TA_assert(x==117) unconditional to suppress cppcheck warning that the
value assigned by x=117 is never used.
Ivan Maidanski [Tue, 23 Aug 2016 21:50:49 +0000 (00:50 +0300)]
Eliminate 'scope of variable can be reduced' cppcheck warnings
* src/atomic_ops_malloc.c (msb): Declare "v" variable exactly in the
scope where the variable is used in.
* tests/run_parallel.h [USE_PTHREADS || USE_WINTHREADS] (run_parallel):
Declare "code" variable exactly in the scope where the variable is
used in.
Ivan Maidanski [Mon, 22 Aug 2016 17:12:19 +0000 (20:12 +0300)]
Fix size value wrap around in AO_malloc_large
AO_malloc(SIZE_MAX) should return NULL now.
* src/atomic_ops_malloc.c (SIZET_SAT_ADD): New macro.
* src/atomic_ops_malloc.c (AO_malloc_large): Use SIZET_SAT_ADD to
avoid integer overflow when computing the memory size to map (i.e.,
malloc should handle arguments close to SIZE_MAX correctly).
Ivan Maidanski [Sat, 20 Aug 2016 14:46:48 +0000 (17:46 +0300)]
Use C11 double-wide atomic intrinsics for Clang/x86 on Cygwin and OS X
Others (e.g., NDK clang/i686) still might need -latomic linker option.
* src/atomic_ops/sysdeps/gcc/x86.h
[AO_GCC_ATOMIC_TEST_AND_SET && __clang__]: Do not fallback to
non-intrinsic implementation if __APPLE_CC__ or __CYGWIN__.
Ivan Maidanski [Wed, 17 Aug 2016 16:51:30 +0000 (19:51 +0300)]
Fix (delete) comment for AO_and_full (x86)
* src/atomic_ops/sysdeps/gcc/x86.h [!AO_PREFER_GENERALIZED]
(AO_and_full): Remove incorrect comment (the code should work even for
i386).
* src/atomic_ops/sysdeps/sunc/x86.h [!AO_PREFER_GENERALIZED]
(AO_and_full): Likewise.
Hans Boehm [Mon, 15 Aug 2016 08:32:33 +0000 (11:32 +0300)]
Fix store-load ordering in AO_stack_pop_explicit_aux_acquire (PowerPC)
Issue #15.
The core issue is that AO_stack_pop_explicit_aux_acquire really needs
to ensure that the store to the blacklist via
AO_compare_and_swap_acquire becomes visible before the load to check
the list head. This effectively needs store-load ordering.
Currently the only ordering here is imposed by the _acquire on the
compare_and_swap. On PowerPC that turns into an lwsync, which is too
weak to enforce store to load ordering.
This patch should fix the issue. But this is suboptimal on x86, and
we may want to make the fence conditional on "not x86", where the CAS
already includes sufficient ordering. (With C++11 atomics, this would
also be tricky and probably involve making a bunch of accesses seq_cst.)
* src/atomic_ops_stack.c [AO_USE_ALMOST_LOCK_FREE]
(AO_stack_pop_explicit_aux_acquire): Call AO_compare_and_swap instead
of AO_compare_and_swap_acquire; call AO_nop_full just before
(first != AO_load(list)).
projects / libatomic_ops / log