This patch adds a generic implementation of HAVE_RM_CTX using standard
fenv calls. As a result math functions using SET_RESTORE_ROUND* macros
do not suffer from a large slowdown on targets which do not implement
optimized libc_fe*_ctx inline functions. Most of the libc_fe* inline
functions are now unused and could be removed in the future (there are
a few math functions left which use a mixture of standard fenv calls
and libc_fe* inline functions - they could be updated to use
SET_RESTORE_ROUND or improved to avoid expensive fenv manipulations
across just a few FP instructions).
libc_feholdsetround*_noex_ctx is added to enable better optimization of
SET_RESTORE_ROUND_NOEX* implementations.
Performance measurements on ARM and x86 of sin() show significant gains
over the current default, fairly close to a highly optimized fenv_private:
ARM x86
no fenv_private : 100% 100%
generic HAVE_RM_CTX : 250% 350%
fenv_private (CTX) : 250% 450%
2014-06-23 Will Newton <will.newton@linaro.org>
Wilco <wdijkstr@arm.com>
* sysdeps/generic/math_private.h: Add generic HAVE_RM_CTX
implementation. Include get-rounding-mode.h.
[!HAVE_RM_CTX]: Define HAVE_RM_CTX to zero.
[!libc_feholdsetround_noex_ctx]: Define
libc_feholdsetround_noex_ctx.
[!libc_feholdsetround_noexf_ctx]: Define
libc_feholdsetround_noexf_ctx.
[!libc_feholdsetround_noexl_ctx]: Define
libc_feholdsetround_noexl_ctx.
(libc_feholdsetround_ctx): New function.
(libc_feresetround_ctx): New function.
(libc_feholdsetround_noex_ctx): New function.
(libc_feresetround_noex_ctx): New function.
ChangeLog:
2014-03-17 Will Newton <will.newton@linaro.org>
* sysdeps/generic/math_private.h: Check whether
HAVE_RM_CTX is defined with #ifdef rather
than #if.
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.
[BZ #6794]
Following Joseph comments about bug 6794, here is a proposed fix. It turned out
to be a large fix mainly because I had to move some file along to follow libm
files/names conventions.
Basically I have added wrappers (w_ilogb.c, w_ilogbf.c, w_ilogbl.c) that now calls
the symbol '__ieee754_ilogb'. The wrappers checks for '__ieee754_ilogb' output and
set the errno and raise exceptions as expected.
The '__ieee754_ilogb' is implemented in sysdeps. I have moved the 's_ilogb[f|l]' files
to e_ilogb[f|l] and renamed the '__ilogb[f|l]' to '__ieee754_ilogb[f|l]'.
I also found out a bug in i386 and x86-64 assembly coded ilogb implementation where
it raises a FE_DIVBYZERO when argument is '0.0'. I corrected this issue as well.
Finally I added the errno and FE_INVALID tests for 0.0, NaN and +-InF argument. Tested
on i386, x86-64, ppc32 and ppc64.
