Sebastian Pop [Wed, 10 Jul 2019 18:31:27 +0000 (18:31 +0000)]
vectorize string-escape with NEON
On this benchmark:
function simple_string_escape() {
$a = "test'asd'asd'' asd\'\"asdfasdf";
for($i=0; $i<512; $i++) {
$a .= chr($i%256);
}
for ($i = 0; $i < 100000; $i++) {
if ($a === stripslashes(addslashes($a)))
$a .= chr($i%256);
else {
echo "error at i=".$i."\n";
return;
}
}
}
the execution time goes from 21.619s to 8.139s (165% speedup) on an A1 Graviton instance.
When removing the characters that need escaping, i.e., this benchmark:
function simple_string() {
$a = "testasdasd asdasdfasdf";
for ($i = 0; $i < 10000; $i++) {
if ($a === stripslashes(addslashes($a)))
$a .= "test dedeasdf";
else {
echo "error at i=".$i."\n";
return;
}
}
}
the execution time goes from 2.932s down to 0.516s (468% speedup) on an A1 Graviton instance.
New opcache directives have been added recently which are returned
if using `ini_get_all('zend opcache')` but are not listed in the
directives if using `opcache_get_configuration()`. This fix adds
those missing directives as well as if `opcache.mmap_base` is used
instead of `opcache.lockfile_path`. Also adds a test to ensure the
directives match with both methods of fetching.
Peter Kokot [Sun, 14 Jul 2019 12:45:27 +0000 (14:45 +0200)]
Remove duplicate PTHREADS template define
The undef PTHREADS converts to define if thread safety is configured.
This step is already done by pthreads m4 macros from TSRM so this now
removes duplicated PTHREADS defines from php_config.h.
Peter Kokot [Sun, 14 Jul 2019 11:50:05 +0000 (13:50 +0200)]
Remove timelib outdated defines
- TIMELIB_OMIT_STDINT is not used anymore since
https://github.com/derickr/timelib/commit/a171f99cf02baf39549ab7209e37565b8b8f6529
- HAVE_GETTIMEOFDAY was defined multiple times in Windows headers
- ext/date/lib/timelib_config.h.win32 does not seem to be used
This test does not work on Windows, if the child process has OPcache
enabled via php.ini, since TEMP is not set in the environment, so
OPcache can't find the base address file, and terminates the childs.
To avoid this (and potentially some other issues regarding missing
environment variables, such as PATH), we force the child process to
ignore any ini files.
The fix for bug #78241 assumed that `time_t` would always be 64bit, but
actually is 32bit for x86. We therefore enforce 64bit arithmetic to
avoid wrapping.
Peter Kokot [Fri, 12 Jul 2019 00:57:52 +0000 (02:57 +0200)]
Remove tsrm_config_common.h
- Checks and defines are not relevant for files that include it anymore
- Some code is not used anymore
- Defines are a bit duplicated in zend_portability.h and TSRM.h file
- MAXPATHLEN defs moved to zend_virtual_cwd.h
This will make descriptor 2 on the child end a dup'd descriptor 1.
This is mainly useful in conjunction with shell-less mode, because
we don't have an easy way to do "2>&1" there.
Additionally we support:
[1 => ['pipe', 'w'], 2 => ['null']]
Which would be the same as a >/dev/null or >nul redirect, depending
on platform.
export() methods were implemented in a roundabout way, where they
would call Reflection::export(), which would then call __toString().
Cut out the middleman by directly calling __toString().
Since syslog's ident and facility parameters have been added to
config[1], vsyslog() segfaults on Windows, if openlog() has not been
called before. We bring back the removed lines to fix this.
The test currently assumes that we'll first read the data of
stdout and stderr and then see eof on stdout and stderr. However
we could also read stdout, see eof on stdout, read stderr and see
eof on stderr, depending on timing.
Avoid output ordering issues by collecting events into a per-pipe
array, so interleaving is not visible.
Sebastian Pop [Wed, 3 Jul 2019 20:10:38 +0000 (20:10 +0000)]
neon vectorization for base64
A similar algorithm is used to vectorize on x86_64, with a good description in
https://arxiv.org/abs/1704.00605 . On AArch64 the implementation differs in that
instead of using multiplies to shift bits around, it uses the vld3+vst4 and
vld4+vst3 combinations to load and store interleaved data. This patch is based
on the NEON implementation of Wojciech Mula:
https://github.com/WojciechMula/base64simd/blob/master/encode/encode.neon.cpp
https://github.com/WojciechMula/base64simd/blob/master/encode/lookup.neon.cpp
and
https://github.com/WojciechMula/base64simd/blob/master/encode/encode.neon.cpp
https://github.com/WojciechMula/base64simd/blob/master/encode/encode.neon.cpp
adapted to php/ext/standard/base64.c and vectorized with factor 16 instead of 8.
On a Graviton A1 instance and on the synthetic benchmarks in
https://github.com/lemire/fastbase64 I see 175% speedup on base64 encoding and
60% speedup on base64 decode compared to the scalar implementation.
The patch passes `make test` regression testing on aarch64-linux.
Peter Kokot [Thu, 11 Jul 2019 00:00:51 +0000 (02:00 +0200)]
Refactor undefining PACKAGE_* symbols
Instead of patching configuration headers template generated by
the given tools - autoheader, this moves patching these symbols to
the configure step before creating and invoking the config.status
and before the configuration header file is generated from the
patched template.
projects / php / log