GCC 4.8 enables -ftree-loop-distribute-patterns at -O3 by default and
this optimization may transform loops into memset/memmove calls. Without
proper handling this may generate unexpected PLT calls on GLIBC.
This patch fixes by create memset/memmove alias to internal GLIBC
__GI_memset/__GI_memmove symbols.
The most common use case of math functions is with default rounding
mode, i.e. rounding to nearest. Setting and restoring rounding mode
is an unnecessary overhead for this, so I've added support for a
context, which does the set/restore only if the FP status needs a
change. The code is written such that only x86 uses these. Other
architectures should be unaffected by it, but would definitely benefit
if the set/restore has as much overhead relative to the rest of the
code, as the x86 bits do.
Here's a summary of the performance improvement due to these
improvements; I've only mentioned functions that use the set/restore
and have benchmark inputs for x86_64:
Before:
cos(): ITERS:4.69335e+08: TOTAL:28884.6Mcy, MAX:4080.28cy, MIN:57.562cy, 16248.6 calls/Mcy
exp(): ITERS:4.47604e+08: TOTAL:28796.2Mcy, MAX:207.721cy, MIN:62.385cy, 15543.9 calls/Mcy
pow(): ITERS:1.63485e+08: TOTAL:28879.9Mcy, MAX:362.255cy, MIN:172.469cy, 5660.86 calls/Mcy
sin(): ITERS:3.89578e+08: TOTAL:28900Mcy, MAX:704.859cy, MIN:47.583cy, 13480.2 calls/Mcy
tan(): ITERS:7.0971e+07: TOTAL:28902.2Mcy, MAX:1357.79cy, MIN:388.58cy, 2455.55 calls/Mcy
After:
cos(): ITERS:6.0014e+08: TOTAL:28875.9Mcy, MAX:364.283cy, MIN:45.716cy, 20783.4 calls/Mcy
exp(): ITERS:5.48578e+08: TOTAL:28764.9Mcy, MAX:191.617cy, MIN:51.011cy, 19071.1 calls/Mcy
pow(): ITERS:1.70013e+08: TOTAL:28873.6Mcy, MAX:689.522cy, MIN:163.989cy, 5888.18 calls/Mcy
sin(): ITERS:4.64079e+08: TOTAL:28891.5Mcy, MAX:6959.3cy, MIN:36.189cy, 16062.8 calls/Mcy
tan(): ITERS:7.2354e+07: TOTAL:28898.9Mcy, MAX:1295.57cy, MIN:380.698cy, 2503.7 calls/Mcy
So the improvements are:
cos: 27.9089%
exp: 22.6919%
pow: 4.01564%
sin: 19.1585%
tan: 1.96086%
The downside of the change is that it will have an adverse performance
impact on non-default rounding modes, but I think the tradeoff is
justified.
__clock_gettime and other __clock_* functions could result in an extra
PLT reference within libc.so if it actually gets used. None of the
code currently uses them, which is why this probably went unnoticed.
Resolves: #15465
The program name may be unavailable if the user application tampers
with argc and argv[]. Some parts of the dynamic linker caters for
this while others don't, so this patch consolidates the check and
fallback into a single macro and updates all users.
Fixes BZ #15339.
NSS_STATUS_UNAVAIL may mean that a necessary input resource is not
available. This could occur in a number of cases including when the
network is down, system runs out of file descriptors, etc. The
correct differentiator in such a case is the h_errno, which gives the
nature of failure. In case of failures other than a simple 'not
found', we set h_errno as NETDB_INTERNAL and let errno be the
identifier for the exact error.
This implementation speed up memset in several ways. First is avoiding
expensive computed jump. Second is using fact that arguments of memset
are most of time aligned to 8 bytes.
Benchmark results on:
kam.mff.cuni.cz/~ondra/benchmark_string/memset_profile_result27_04_13.tar.bz2
We add new memcpy version that uses unaligned loads which are fast
on modern processors. This allows second improvement which is avoiding
computed jump which is relatively expensive operation.
Tests available here:
http://kam.mff.cuni.cz/~ondra/memcpy_profile_result27_04_13.tar.bz2
[BZ #15442] This adds support for the inverse interpretation of the
quiet bit of IEEE 754 floating-point NaN data that some processors
use. This includes in particular MIPS architecture processors; the
payload used for the canonical qNaN encoding is updated accordingly
so as not to interfere with the quiet bit.
The EXTRACT_WORDS64 and INSERT_WORDS64 macros use movd for a 64-bit
operation. Somehow gcc manages to turn this into movq, but LLVM won't.
2013-05-15 Peter Collingbourne <pcc@google.com>
* sysdeps/x86_64/fpu/math_private.h (MOVQ): New macro.
(EXTRACT_WORDS64) Use where appropriate.
(INSERT_WORDS64) Likewise.
While these instructions accept memory operands, only one operand
may be a memory operand. Giving two operands xm constraints gives
the compiler the option of using memory for both operands, which
would result in invalid assembly code. Using x for all operands is
more appropriate, as most x86_64 calling conventions will pass the
arguments in registers anyway.
2013-05-15 Peter Collingbourne <pcc@google.com>
* sysdeps/x86_64/fpu/multiarch/s_fma.c (__fma_fma4): Replace xm
constraints with x constraints.
* sysdeps/x86_64/fpu/multiarch/s_fmaf.c (__fmaf_fma4): Likewise.
