* Re: [RFC][PATCH] New expf, exp2f, logf, log2f and powf implementations
@ 2017-09-05 8:39 Wilco Dijkstra
2017-09-06 6:58 ` Corinna Vinschen
0 siblings, 1 reply; 10+ messages in thread
From: Wilco Dijkstra @ 2017-09-05 8:39 UTC (permalink / raw)
To: Joel Sherrill, newlib; +Cc: nd, Szabolcs Nagy
Hi,
> Any comment on the difference in code and data size? Just curious
> if we are considering it on all targets.
When statically linking with GLIBC, I noticed text size of binaries using
Szabolc's patch was around 1200 bytes smaller on AArch64 when using
expf, logf and powf. Newlib uses an almost identical fdlibm version so the
savings should be similar. Compile it on your favourite target, YMMV.
> What capabilities does it require of the target architecture? You
> mentioned VFP. Does it require vector operations. If something
> is assumed, if the minimum capabilities are not met, how well
> does this code perform?
Very little, if double precision operations have similar performance as
single precision, you will get a huge speedup. The source is standard C99
so it doesn't require any vector extensions. What is required is decent
support for uint64_t, so it works best on 32/64-bit ISAs.
For 8/16-bit ISAs or CPUs without hardware floating point you generally
have to use integer arithmetic using a fast 32x32->64 bit multiplier, but
that requires a different algorithm. If your CPU doesn't have a multiplier(!?!)
then CORDIC is your only option. However newlib isn't optimizing for these
targets today, so this patch isn't aiming to either.
Mixed-endian IEEE formats could be supported but the various asuint()
functions would need porting (same for issignaling). Non-IEEE formats would
be harder given the use of integer masking to manipulate floating point values.
Wilco
^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [RFC][PATCH] New expf, exp2f, logf, log2f and powf implementations
2017-09-05 8:39 [RFC][PATCH] New expf, exp2f, logf, log2f and powf implementations Wilco Dijkstra
@ 2017-09-06 6:58 ` Corinna Vinschen
2017-09-29 18:45 ` [PATCH v2] " Szabolcs Nagy
0 siblings, 1 reply; 10+ messages in thread
From: Corinna Vinschen @ 2017-09-06 6:58 UTC (permalink / raw)
To: newlib
[-- Attachment #1: Type: text/plain, Size: 2829 bytes --]
On Sep 5 00:18, Wilco Dijkstra wrote:
> Hi,
>
> > Any comment on the difference in code and data size? Just curious
> > if we are considering it on all targets.
>
> When statically linking with GLIBC, I noticed text size of binaries using
> Szabolc's patch was around 1200 bytes smaller on AArch64 when using
> expf, logf and powf. Newlib uses an almost identical fdlibm version so the
> savings should be similar. Compile it on your favourite target, YMMV.
>
> > What capabilities does it require of the target architecture? You
> > mentioned VFP. Does it require vector operations. If something
> > is assumed, if the minimum capabilities are not met, how well
> > does this code perform?
>
> Very little, if double precision operations have similar performance as
> single precision, you will get a huge speedup. The source is standard C99
> so it doesn't require any vector extensions. What is required is decent
> support for uint64_t, so it works best on 32/64-bit ISAs.
>
> For 8/16-bit ISAs or CPUs without hardware floating point you generally
> have to use integer arithmetic using a fast 32x32->64 bit multiplier, but
> that requires a different algorithm. If your CPU doesn't have a multiplier(!?!)
> then CORDIC is your only option. However newlib isn't optimizing for these
> targets today, so this patch isn't aiming to either.
>
> Mixed-endian IEEE formats could be supported but the various asuint()
> functions would need porting (same for issignaling). Non-IEEE formats would
> be harder given the use of integer masking to manipulate floating point values.
>
> Wilco
Ok, I retract my reservations against OBSOLETE_MATH as such. However,
there are a few problems with the current patch:
- Rename to __OBSOLETE_MATH (two underscores) so as not to spoil the
namespace.
- The current implementation disallows a developer to override the
setting from, e. g., sys/config.h or the command line. As a minimum,
__OBSOLETE_MATH should only be set if it isn't already set, and the
checks for __OBSOLETE_MATH should check for being != 0:
- #define __TGT_OBSOLETE_MATH 1 in arm and aarch cases
- At the end of the file
#ifndef __OBSOLETE_MATH
#define __OBSOLETE_MATH __TGT_OBSOLETE_MATH
#endif
- Further checks in the code:
#if __OBSOLETE_MATH
rather than
#ifdef __OBSOLETE_MATH
- The header of ieeefp.h explains the options. Add __TGT_OBSOLETE_MATH /
__OBSOLETE_MATH and add an informative comment along the lines of the
above text, so developers have enough info to decide if they want to
enable this for their target.
Not sure if __TGT_OBSOLETE_MATH is a good name, though. I'm open for
better suggestions.
Thanks,
Corinna
--
Corinna Vinschen
Cygwin Maintainer
Red Hat
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]
^ permalink raw reply [flat|nested] 10+ messages in thread
* [PATCH v2] New expf, exp2f, logf, log2f and powf implementations
2017-09-06 6:58 ` Corinna Vinschen
@ 2017-09-29 18:45 ` Szabolcs Nagy
2017-10-11 11:57 ` Szabolcs Nagy
2017-10-11 12:58 ` Corinna Vinschen
0 siblings, 2 replies; 10+ messages in thread
From: Szabolcs Nagy @ 2017-09-29 18:45 UTC (permalink / raw)
To: newlib; +Cc: nd
[-- Attachment #1: Type: text/plain, Size: 1531 bytes --]
On 05/09/17 09:38, Corinna Vinschen wrote:
> Ok, I retract my reservations against OBSOLETE_MATH as such. However,
> there are a few problems with the current patch:
>
> - Rename to __OBSOLETE_MATH (two underscores) so as not to spoil the
> namespace.
>
> - The current implementation disallows a developer to override the
> setting from, e. g., sys/config.h or the command line. As a minimum,
> __OBSOLETE_MATH should only be set if it isn't already set, and the
> checks for __OBSOLETE_MATH should check for being != 0:
>
> - #define __TGT_OBSOLETE_MATH 1 in arm and aarch cases
>
> - At the end of the file
>
> #ifndef __OBSOLETE_MATH
> #define __OBSOLETE_MATH __TGT_OBSOLETE_MATH
> #endif
>
> - Further checks in the code:
>
> #if __OBSOLETE_MATH
>
> rather than
>
> #ifdef __OBSOLETE_MATH
>
> - The header of ieeefp.h explains the options. Add __TGT_OBSOLETE_MATH /
> __OBSOLETE_MATH and add an informative comment along the lines of the
> above text, so developers have enough info to decide if they want to
> enable this for their target.
>
> Not sure if __TGT_OBSOLETE_MATH is a good name, though. I'm open for
> better suggestions.
>
i used __OBSOLETE_MATH_DEFAULT.
v2:
- Use __OBSOLETE_MATH and __OBSOLETE_MATH_DEFAULT.
- Upstream change: whitespace code style fixes.
- Upstream change: HIDDEN attribute for data symbols.
- Upstream change: __logf_data.poly array size is reduced.
- Upstream change: issignalingf_inline now has code for non-standard SNaN.
[-- Attachment #2: 0001-New-expf-exp2f-logf-log2f-and-powf-implementations.patch --]
[-- Type: text/x-patch, Size: 68212 bytes --]
From 6072e2726d4eabaf121b6f5221d7014d1ce826e8 Mon Sep 17 00:00:00 2001
From: Szabolcs Nagy <szabolcs.nagy@arm.com>
Date: Thu, 25 May 2017 16:41:38 +0100
Subject: [PATCH] New expf, exp2f, logf, log2f and powf implementations
Based on code from https://github.com/ARM-software/optimized-routines/
This patch adds a highly optimized generic implementation of expf,
exp2f, logf, log2f and powf. The new functions are not only
faster (6x for powf!), but are also smaller and more accurate.
In order to achieve this, the algorithm uses double precision
arithmetic for accuracy, avoids divisions and uses small table
lookups to minimize the polynomials. Special cases are handled
inline to avoid the unnecessary overhead of wrapper functions and
set errno to POSIX requirements.
The new functions are added under newlib/libm/common, but the old
implementations are kept (in newlib/libm/math) for non-IEEE or
pre-C99 systems. Targets can enable the new math code by defining
__OBSOLETE_MATH_DEFAULT to 0 in newlib/libc/include/machine/ieeefp.h,
users can override the default by defining __OBSOLETE_MATH.
Currently the new code is enabled for AArch64 and AArch32 with VFP.
Targets with a single precision FPU may still prefer the old
implementation.
libm.a size changes:
arm: -1692
arm/thumb/v7-a/nofp: -878
arm/thumb/v7-a+fp/hard: -864
arm/thumb/v7-a+fp/softfp: -908
aarch64: -1476
---
newlib/libc/include/machine/ieeefp.h | 27 ++++
newlib/libm/common/Makefile.am | 7 +-
newlib/libm/common/Makefile.in | 148 +++++++++++++++-------
newlib/libm/common/math_config.h | 179 ++++++++++++++++++++++++++
newlib/libm/common/math_errf.c | 89 +++++++++++++
newlib/libm/common/sf_exp.c | 113 +++++++++++++++++
newlib/libm/common/sf_exp2.c | 99 +++++++++++++++
newlib/libm/common/sf_exp2_data.c | 56 +++++++++
newlib/libm/common/sf_log.c | 98 +++++++++++++++
newlib/libm/common/sf_log2.c | 131 +++++++++++++------
newlib/libm/common/sf_log2_data.c | 56 +++++++++
newlib/libm/common/sf_log_data.c | 56 +++++++++
newlib/libm/common/sf_pow.c | 230 ++++++++++++++++++++++++++++++++++
newlib/libm/common/sf_pow_log2_data.c | 57 +++++++++
newlib/libm/math/Makefile.am | 3 +-
newlib/libm/math/Makefile.in | 14 ++-
newlib/libm/math/ef_exp.c | 2 +
newlib/libm/math/ef_log.c | 2 +
newlib/libm/math/ef_pow.c | 2 +
newlib/libm/math/wf_exp.c | 2 +
newlib/libm/math/wf_exp2.c | 2 +
newlib/libm/math/wf_log.c | 2 +
newlib/libm/math/wf_log2.c | 50 ++++++++
newlib/libm/math/wf_pow.c | 2 +
24 files changed, 1339 insertions(+), 88 deletions(-)
create mode 100644 newlib/libm/common/math_config.h
create mode 100644 newlib/libm/common/math_errf.c
create mode 100644 newlib/libm/common/sf_exp.c
create mode 100644 newlib/libm/common/sf_exp2.c
create mode 100644 newlib/libm/common/sf_exp2_data.c
create mode 100644 newlib/libm/common/sf_log.c
create mode 100644 newlib/libm/common/sf_log2_data.c
create mode 100644 newlib/libm/common/sf_log_data.c
create mode 100644 newlib/libm/common/sf_pow.c
create mode 100644 newlib/libm/common/sf_pow_log2_data.c
create mode 100644 newlib/libm/math/wf_log2.c
diff --git a/newlib/libc/include/machine/ieeefp.h b/newlib/libc/include/machine/ieeefp.h
index 29c08fa43..b1b44665e 100644
--- a/newlib/libc/include/machine/ieeefp.h
+++ b/newlib/libc/include/machine/ieeefp.h
@@ -48,6 +48,23 @@
This represents what type a float arg is passed as. It is used when the type is
not promoted to double.
+
+ __OBSOLETE_MATH_DEFAULT
+
+ Default value for __OBSOLETE_MATH if that's not set by the user.
+ It should be set here based on predefined feature macros.
+
+ __OBSOLETE_MATH
+
+ If set to 1 then some new math code will be disabled and older libm
+ code will be used instead. This is necessary because the new math
+ code does not support all targets, it assumes that the toolchain has
+ ISO C99 support (hexfloat literals, standard fenv semantics), the
+ target has IEEE-754 conforming binary32 float and binary64 double
+ (not mixed endian) representation, standard SNaN representation,
+ double and single precision arithmetics has similar latency and it
+ has no legacy SVID matherr support, only POSIX errno and fenv
+ exception based error handling.
*/
#if (defined(__arm__) || defined(__thumb__)) && !defined(__MAVERICK__)
@@ -61,6 +78,7 @@
# else
# define __IEEE_BIG_ENDIAN
# endif
+# define __OBSOLETE_MATH_DEFAULT 0
#else
# define __IEEE_BIG_ENDIAN
# ifdef __ARMEL__
@@ -75,6 +93,7 @@
#else
#define __IEEE_BIG_ENDIAN
#endif
+#define __OBSOLETE_MATH_DEFAULT 0
#endif
#ifdef __epiphany__
@@ -427,6 +446,14 @@
#define __IEEE_BIG_ENDIAN
#endif
+#ifndef __OBSOLETE_MATH_DEFAULT
+/* Use old math code by default. */
+#define __OBSOLETE_MATH_DEFAULT 1
+#endif
+#ifndef __OBSOLETE_MATH
+#define __OBSOLETE_MATH __OBSOLETE_MATH_DEFAULT
+#endif
+
#ifndef __IEEE_BIG_ENDIAN
#ifndef __IEEE_LITTLE_ENDIAN
#error Endianess not declared!!
diff --git a/newlib/libm/common/Makefile.am b/newlib/libm/common/Makefile.am
index 1268e17db..3e2620c7e 100644
--- a/newlib/libm/common/Makefile.am
+++ b/newlib/libm/common/Makefile.am
@@ -18,11 +18,14 @@ fsrc = sf_finite.c sf_copysign.c sf_modf.c sf_scalbn.c \
sf_cbrt.c sf_exp10.c sf_expm1.c sf_ilogb.c \
sf_infinity.c sf_isinf.c sf_isinff.c sf_isnan.c sf_isnanf.c \
sf_log1p.c sf_nan.c sf_nextafter.c sf_pow10.c \
- sf_rint.c sf_logb.c sf_log2.c \
+ sf_rint.c sf_logb.c \
sf_fdim.c sf_fma.c sf_fmax.c sf_fmin.c sf_fpclassify.c \
sf_lrint.c sf_llrint.c \
sf_lround.c sf_llround.c sf_nearbyint.c sf_remquo.c sf_round.c \
- sf_scalbln.c sf_trunc.c
+ sf_scalbln.c sf_trunc.c \
+ sf_exp.c sf_exp2.c sf_exp2_data.c sf_log.c sf_log_data.c \
+ sf_log2.c sf_log2_data.c sf_pow_log2_data.c sf_pow.c \
+ math_errf.c
lsrc = atanl.c cosl.c sinl.c tanl.c tanhl.c frexpl.c modfl.c ceill.c fabsl.c \
floorl.c log1pl.c expm1l.c acosl.c asinl.c atan2l.c coshl.c sinhl.c \
diff --git a/newlib/libm/common/Makefile.in b/newlib/libm/common/Makefile.in
index 9b367653d..c603be361 100644
--- a/newlib/libm/common/Makefile.in
+++ b/newlib/libm/common/Makefile.in
@@ -78,21 +78,21 @@ am__objects_1 = lib_a-s_finite.$(OBJEXT) lib_a-s_copysign.$(OBJEXT) \
lib_a-s_modf.$(OBJEXT) lib_a-s_scalbn.$(OBJEXT) \
lib_a-s_cbrt.$(OBJEXT) lib_a-s_exp10.$(OBJEXT) \
lib_a-s_expm1.$(OBJEXT) lib_a-s_ilogb.$(OBJEXT) \
- lib_a-s_infinity.$(OBJEXT) \
- lib_a-s_isinf.$(OBJEXT) lib_a-s_isinfd.$(OBJEXT) \
- lib_a-s_isnan.$(OBJEXT) lib_a-s_isnand.$(OBJEXT) \
- lib_a-s_log1p.$(OBJEXT) lib_a-s_nan.$(OBJEXT) \
- lib_a-s_nextafter.$(OBJEXT) lib_a-s_pow10.$(OBJEXT) \
- lib_a-s_rint.$(OBJEXT) lib_a-s_logb.$(OBJEXT) \
- lib_a-s_log2.$(OBJEXT) lib_a-s_matherr.$(OBJEXT) \
- lib_a-s_lib_ver.$(OBJEXT) lib_a-s_fdim.$(OBJEXT) \
- lib_a-s_fma.$(OBJEXT) lib_a-s_fmax.$(OBJEXT) \
- lib_a-s_fmin.$(OBJEXT) lib_a-s_fpclassify.$(OBJEXT) \
- lib_a-s_lrint.$(OBJEXT) lib_a-s_llrint.$(OBJEXT) \
- lib_a-s_lround.$(OBJEXT) lib_a-s_llround.$(OBJEXT) \
- lib_a-s_nearbyint.$(OBJEXT) lib_a-s_remquo.$(OBJEXT) \
- lib_a-s_round.$(OBJEXT) lib_a-s_scalbln.$(OBJEXT) \
- lib_a-s_signbit.$(OBJEXT) lib_a-s_trunc.$(OBJEXT)
+ lib_a-s_infinity.$(OBJEXT) lib_a-s_isinf.$(OBJEXT) \
+ lib_a-s_isinfd.$(OBJEXT) lib_a-s_isnan.$(OBJEXT) \
+ lib_a-s_isnand.$(OBJEXT) lib_a-s_log1p.$(OBJEXT) \
+ lib_a-s_nan.$(OBJEXT) lib_a-s_nextafter.$(OBJEXT) \
+ lib_a-s_pow10.$(OBJEXT) lib_a-s_rint.$(OBJEXT) \
+ lib_a-s_logb.$(OBJEXT) lib_a-s_log2.$(OBJEXT) \
+ lib_a-s_matherr.$(OBJEXT) lib_a-s_lib_ver.$(OBJEXT) \
+ lib_a-s_fdim.$(OBJEXT) lib_a-s_fma.$(OBJEXT) \
+ lib_a-s_fmax.$(OBJEXT) lib_a-s_fmin.$(OBJEXT) \
+ lib_a-s_fpclassify.$(OBJEXT) lib_a-s_lrint.$(OBJEXT) \
+ lib_a-s_llrint.$(OBJEXT) lib_a-s_lround.$(OBJEXT) \
+ lib_a-s_llround.$(OBJEXT) lib_a-s_nearbyint.$(OBJEXT) \
+ lib_a-s_remquo.$(OBJEXT) lib_a-s_round.$(OBJEXT) \
+ lib_a-s_scalbln.$(OBJEXT) lib_a-s_signbit.$(OBJEXT) \
+ lib_a-s_trunc.$(OBJEXT)
am__objects_2 = lib_a-sf_finite.$(OBJEXT) lib_a-sf_copysign.$(OBJEXT) \
lib_a-sf_modf.$(OBJEXT) lib_a-sf_scalbn.$(OBJEXT) \
lib_a-sf_cbrt.$(OBJEXT) lib_a-sf_exp10.$(OBJEXT) \
@@ -102,14 +102,19 @@ am__objects_2 = lib_a-sf_finite.$(OBJEXT) lib_a-sf_copysign.$(OBJEXT) \
lib_a-sf_isnanf.$(OBJEXT) lib_a-sf_log1p.$(OBJEXT) \
lib_a-sf_nan.$(OBJEXT) lib_a-sf_nextafter.$(OBJEXT) \
lib_a-sf_pow10.$(OBJEXT) lib_a-sf_rint.$(OBJEXT) \
- lib_a-sf_logb.$(OBJEXT) lib_a-sf_log2.$(OBJEXT) \
- lib_a-sf_fdim.$(OBJEXT) lib_a-sf_fma.$(OBJEXT) \
- lib_a-sf_fmax.$(OBJEXT) lib_a-sf_fmin.$(OBJEXT) \
- lib_a-sf_fpclassify.$(OBJEXT) lib_a-sf_lrint.$(OBJEXT) \
- lib_a-sf_llrint.$(OBJEXT) lib_a-sf_lround.$(OBJEXT) \
- lib_a-sf_llround.$(OBJEXT) lib_a-sf_nearbyint.$(OBJEXT) \
- lib_a-sf_remquo.$(OBJEXT) lib_a-sf_round.$(OBJEXT) \
- lib_a-sf_scalbln.$(OBJEXT) lib_a-sf_trunc.$(OBJEXT)
+ lib_a-sf_logb.$(OBJEXT) lib_a-sf_fdim.$(OBJEXT) \
+ lib_a-sf_fma.$(OBJEXT) lib_a-sf_fmax.$(OBJEXT) \
+ lib_a-sf_fmin.$(OBJEXT) lib_a-sf_fpclassify.$(OBJEXT) \
+ lib_a-sf_lrint.$(OBJEXT) lib_a-sf_llrint.$(OBJEXT) \
+ lib_a-sf_lround.$(OBJEXT) lib_a-sf_llround.$(OBJEXT) \
+ lib_a-sf_nearbyint.$(OBJEXT) lib_a-sf_remquo.$(OBJEXT) \
+ lib_a-sf_round.$(OBJEXT) lib_a-sf_scalbln.$(OBJEXT) \
+ lib_a-sf_trunc.$(OBJEXT) lib_a-sf_exp.$(OBJEXT) \
+ lib_a-sf_exp2.$(OBJEXT) lib_a-sf_exp2_data.$(OBJEXT) \
+ lib_a-sf_log.$(OBJEXT) lib_a-sf_log_data.$(OBJEXT) \
+ lib_a-sf_log2.$(OBJEXT) lib_a-sf_log2_data.$(OBJEXT) \
+ lib_a-sf_pow_log2_data.$(OBJEXT) lib_a-sf_pow.$(OBJEXT) \
+ lib_a-math_errf.$(OBJEXT)
am__objects_3 = lib_a-atanl.$(OBJEXT) lib_a-cosl.$(OBJEXT) \
lib_a-sinl.$(OBJEXT) lib_a-tanl.$(OBJEXT) \
lib_a-tanhl.$(OBJEXT) lib_a-frexpl.$(OBJEXT) \
@@ -148,21 +153,23 @@ lib_a_OBJECTS = $(am_lib_a_OBJECTS)
LTLIBRARIES = $(noinst_LTLIBRARIES)
libcommon_la_LIBADD =
am__objects_5 = s_finite.lo s_copysign.lo s_modf.lo s_scalbn.lo \
- s_cbrt.lo s_exp10.lo s_expm1.lo s_ilogb.lo \
- s_infinity.lo s_isinf.lo s_isinfd.lo s_isnan.lo s_isnand.lo \
- s_log1p.lo s_nan.lo s_nextafter.lo s_pow10.lo s_rint.lo \
- s_logb.lo s_log2.lo s_matherr.lo s_lib_ver.lo s_fdim.lo \
- s_fma.lo s_fmax.lo s_fmin.lo s_fpclassify.lo s_lrint.lo \
- s_llrint.lo s_lround.lo s_llround.lo s_nearbyint.lo \
- s_remquo.lo s_round.lo s_scalbln.lo s_signbit.lo s_trunc.lo
+ s_cbrt.lo s_exp10.lo s_expm1.lo s_ilogb.lo s_infinity.lo \
+ s_isinf.lo s_isinfd.lo s_isnan.lo s_isnand.lo s_log1p.lo \
+ s_nan.lo s_nextafter.lo s_pow10.lo s_rint.lo s_logb.lo \
+ s_log2.lo s_matherr.lo s_lib_ver.lo s_fdim.lo s_fma.lo \
+ s_fmax.lo s_fmin.lo s_fpclassify.lo s_lrint.lo s_llrint.lo \
+ s_lround.lo s_llround.lo s_nearbyint.lo s_remquo.lo s_round.lo \
+ s_scalbln.lo s_signbit.lo s_trunc.lo
am__objects_6 = sf_finite.lo sf_copysign.lo sf_modf.lo sf_scalbn.lo \
sf_cbrt.lo sf_exp10.lo sf_expm1.lo sf_ilogb.lo sf_infinity.lo \
sf_isinf.lo sf_isinff.lo sf_isnan.lo sf_isnanf.lo sf_log1p.lo \
sf_nan.lo sf_nextafter.lo sf_pow10.lo sf_rint.lo sf_logb.lo \
- sf_log2.lo sf_fdim.lo sf_fma.lo sf_fmax.lo sf_fmin.lo \
- sf_fpclassify.lo sf_lrint.lo sf_llrint.lo sf_lround.lo \
- sf_llround.lo sf_nearbyint.lo sf_remquo.lo sf_round.lo \
- sf_scalbln.lo sf_trunc.lo
+ sf_fdim.lo sf_fma.lo sf_fmax.lo sf_fmin.lo sf_fpclassify.lo \
+ sf_lrint.lo sf_llrint.lo sf_lround.lo sf_llround.lo \
+ sf_nearbyint.lo sf_remquo.lo sf_round.lo sf_scalbln.lo \
+ sf_trunc.lo sf_exp.lo sf_exp2.lo sf_exp2_data.lo sf_log.lo \
+ sf_log_data.lo sf_log2.lo sf_log2_data.lo sf_pow_log2_data.lo \
+ sf_pow.lo math_errf.lo
am__objects_7 = atanl.lo cosl.lo sinl.lo tanl.lo tanhl.lo frexpl.lo \
modfl.lo ceill.lo fabsl.lo floorl.lo log1pl.lo expm1l.lo \
acosl.lo asinl.lo atan2l.lo coshl.lo sinhl.lo expl.lo \
@@ -347,11 +354,14 @@ fsrc = sf_finite.c sf_copysign.c sf_modf.c sf_scalbn.c \
sf_cbrt.c sf_exp10.c sf_expm1.c sf_ilogb.c \
sf_infinity.c sf_isinf.c sf_isinff.c sf_isnan.c sf_isnanf.c \
sf_log1p.c sf_nan.c sf_nextafter.c sf_pow10.c \
- sf_rint.c sf_logb.c sf_log2.c \
+ sf_rint.c sf_logb.c \
sf_fdim.c sf_fma.c sf_fmax.c sf_fmin.c sf_fpclassify.c \
sf_lrint.c sf_llrint.c \
sf_lround.c sf_llround.c sf_nearbyint.c sf_remquo.c sf_round.c \
- sf_scalbln.c sf_trunc.c
+ sf_scalbln.c sf_trunc.c \
+ sf_exp.c sf_exp2.c sf_exp2_data.c sf_log.c sf_log_data.c \
+ sf_log2.c sf_log2_data.c sf_pow_log2_data.c sf_pow.c \
+ math_errf.c
lsrc = atanl.c cosl.c sinl.c tanl.c tanhl.c frexpl.c modfl.c ceill.c fabsl.c \
floorl.c log1pl.c expm1l.c acosl.c asinl.c atan2l.c coshl.c sinhl.c \
@@ -796,12 +806,6 @@ lib_a-sf_logb.o: sf_logb.c
lib_a-sf_logb.obj: sf_logb.c
$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-sf_logb.obj `if test -f 'sf_logb.c'; then $(CYGPATH_W) 'sf_logb.c'; else $(CYGPATH_W) '$(srcdir)/sf_logb.c'; fi`
-lib_a-sf_log2.o: sf_log2.c
- $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-sf_log2.o `test -f 'sf_log2.c' || echo '$(srcdir)/'`sf_log2.c
-
-lib_a-sf_log2.obj: sf_log2.c
- $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-sf_log2.obj `if test -f 'sf_log2.c'; then $(CYGPATH_W) 'sf_log2.c'; else $(CYGPATH_W) '$(srcdir)/sf_log2.c'; fi`
-
lib_a-sf_fdim.o: sf_fdim.c
$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-sf_fdim.o `test -f 'sf_fdim.c' || echo '$(srcdir)/'`sf_fdim.c
@@ -886,6 +890,66 @@ lib_a-sf_trunc.o: sf_trunc.c
lib_a-sf_trunc.obj: sf_trunc.c
$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-sf_trunc.obj `if test -f 'sf_trunc.c'; then $(CYGPATH_W) 'sf_trunc.c'; else $(CYGPATH_W) '$(srcdir)/sf_trunc.c'; fi`
+lib_a-sf_exp.o: sf_exp.c
+ $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-sf_exp.o `test -f 'sf_exp.c' || echo '$(srcdir)/'`sf_exp.c
+
+lib_a-sf_exp.obj: sf_exp.c
+ $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-sf_exp.obj `if test -f 'sf_exp.c'; then $(CYGPATH_W) 'sf_exp.c'; else $(CYGPATH_W) '$(srcdir)/sf_exp.c'; fi`
+
+lib_a-sf_exp2.o: sf_exp2.c
+ $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-sf_exp2.o `test -f 'sf_exp2.c' || echo '$(srcdir)/'`sf_exp2.c
+
+lib_a-sf_exp2.obj: sf_exp2.c
+ $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-sf_exp2.obj `if test -f 'sf_exp2.c'; then $(CYGPATH_W) 'sf_exp2.c'; else $(CYGPATH_W) '$(srcdir)/sf_exp2.c'; fi`
+
+lib_a-sf_exp2_data.o: sf_exp2_data.c
+ $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-sf_exp2_data.o `test -f 'sf_exp2_data.c' || echo '$(srcdir)/'`sf_exp2_data.c
+
+lib_a-sf_exp2_data.obj: sf_exp2_data.c
+ $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-sf_exp2_data.obj `if test -f 'sf_exp2_data.c'; then $(CYGPATH_W) 'sf_exp2_data.c'; else $(CYGPATH_W) '$(srcdir)/sf_exp2_data.c'; fi`
+
+lib_a-sf_log.o: sf_log.c
+ $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-sf_log.o `test -f 'sf_log.c' || echo '$(srcdir)/'`sf_log.c
+
+lib_a-sf_log.obj: sf_log.c
+ $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-sf_log.obj `if test -f 'sf_log.c'; then $(CYGPATH_W) 'sf_log.c'; else $(CYGPATH_W) '$(srcdir)/sf_log.c'; fi`
+
+lib_a-sf_log_data.o: sf_log_data.c
+ $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-sf_log_data.o `test -f 'sf_log_data.c' || echo '$(srcdir)/'`sf_log_data.c
+
+lib_a-sf_log_data.obj: sf_log_data.c
+ $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-sf_log_data.obj `if test -f 'sf_log_data.c'; then $(CYGPATH_W) 'sf_log_data.c'; else $(CYGPATH_W) '$(srcdir)/sf_log_data.c'; fi`
+
+lib_a-sf_log2.o: sf_log2.c
+ $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-sf_log2.o `test -f 'sf_log2.c' || echo '$(srcdir)/'`sf_log2.c
+
+lib_a-sf_log2.obj: sf_log2.c
+ $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-sf_log2.obj `if test -f 'sf_log2.c'; then $(CYGPATH_W) 'sf_log2.c'; else $(CYGPATH_W) '$(srcdir)/sf_log2.c'; fi`
+
+lib_a-sf_log2_data.o: sf_log2_data.c
+ $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-sf_log2_data.o `test -f 'sf_log2_data.c' || echo '$(srcdir)/'`sf_log2_data.c
+
+lib_a-sf_log2_data.obj: sf_log2_data.c
+ $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-sf_log2_data.obj `if test -f 'sf_log2_data.c'; then $(CYGPATH_W) 'sf_log2_data.c'; else $(CYGPATH_W) '$(srcdir)/sf_log2_data.c'; fi`
+
+lib_a-sf_pow_log2_data.o: sf_pow_log2_data.c
+ $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-sf_pow_log2_data.o `test -f 'sf_pow_log2_data.c' || echo '$(srcdir)/'`sf_pow_log2_data.c
+
+lib_a-sf_pow_log2_data.obj: sf_pow_log2_data.c
+ $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-sf_pow_log2_data.obj `if test -f 'sf_pow_log2_data.c'; then $(CYGPATH_W) 'sf_pow_log2_data.c'; else $(CYGPATH_W) '$(srcdir)/sf_pow_log2_data.c'; fi`
+
+lib_a-sf_pow.o: sf_pow.c
+ $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-sf_pow.o `test -f 'sf_pow.c' || echo '$(srcdir)/'`sf_pow.c
+
+lib_a-sf_pow.obj: sf_pow.c
+ $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-sf_pow.obj `if test -f 'sf_pow.c'; then $(CYGPATH_W) 'sf_pow.c'; else $(CYGPATH_W) '$(srcdir)/sf_pow.c'; fi`
+
+lib_a-math_errf.o: math_errf.c
+ $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-math_errf.o `test -f 'math_errf.c' || echo '$(srcdir)/'`math_errf.c
+
+lib_a-math_errf.obj: math_errf.c
+ $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-math_errf.obj `if test -f 'math_errf.c'; then $(CYGPATH_W) 'math_errf.c'; else $(CYGPATH_W) '$(srcdir)/math_errf.c'; fi`
+
lib_a-atanl.o: atanl.c
$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-atanl.o `test -f 'atanl.c' || echo '$(srcdir)/'`atanl.c
diff --git a/newlib/libm/common/math_config.h b/newlib/libm/common/math_config.h
new file mode 100644
index 000000000..c5667844a
--- /dev/null
+++ b/newlib/libm/common/math_config.h
@@ -0,0 +1,179 @@
+/* Configuration for math routines.
+ Copyright (c) 2017 ARM Ltd. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without
+ modification, are permitted provided that the following conditions
+ are met:
+ 1. Redistributions of source code must retain the above copyright
+ notice, this list of conditions and the following disclaimer.
+ 2. Redistributions in binary form must reproduce the above copyright
+ notice, this list of conditions and the following disclaimer in the
+ documentation and/or other materials provided with the distribution.
+ 3. The name of the company may not be used to endorse or promote
+ products derived from this software without specific prior written
+ permission.
+
+ THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS AND ANY EXPRESS OR IMPLIED
+ WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+ TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */
+
+#ifndef _MATH_CONFIG_H
+#define _MATH_CONFIG_H
+
+#include <math.h>
+#include <stdint.h>
+
+#ifndef WANT_ROUNDING
+/* Correct special case results in non-nearest rounding modes. */
+# define WANT_ROUNDING 1
+#endif
+#ifndef WANT_ERRNO
+/* Set errno according to ISO C with (math_errhandling & MATH_ERRNO) != 0. */
+# define WANT_ERRNO 1
+#endif
+#ifndef WANT_ERRNO_UFLOW
+/* Set errno to ERANGE if result underflows to 0 (in all rounding modes). */
+# define WANT_ERRNO_UFLOW (WANT_ROUNDING && WANT_ERRNO)
+#endif
+
+#ifndef TOINT_INTRINSICS
+# define TOINT_INTRINSICS 0
+#endif
+#ifndef TOINT_RINT
+# define TOINT_RINT 0
+#endif
+#ifndef TOINT_SHIFT
+# define TOINT_SHIFT 1
+#endif
+
+static inline uint32_t
+asuint (float f)
+{
+ union
+ {
+ float f;
+ uint32_t i;
+ } u = {f};
+ return u.i;
+}
+
+static inline float
+asfloat (uint32_t i)
+{
+ union
+ {
+ uint32_t i;
+ float f;
+ } u = {i};
+ return u.f;
+}
+
+static inline uint64_t
+asuint64 (double f)
+{
+ union
+ {
+ double f;
+ uint64_t i;
+ } u = {f};
+ return u.i;
+}
+
+static inline double
+asdouble (uint64_t i)
+{
+ union
+ {
+ uint64_t i;
+ double f;
+ } u = {i};
+ return u.f;
+}
+
+#ifndef IEEE_754_2008_SNAN
+# define IEEE_754_2008_SNAN 1
+#endif
+static inline int
+issignalingf_inline (float x)
+{
+ uint32_t ix = asuint (x);
+ if (!IEEE_754_2008_SNAN)
+ return (ix & 0x7fc00000) == 0x7fc00000;
+ return 2 * (ix ^ 0x00400000) > 2u * 0x7fc00000;
+}
+
+#ifdef __GNUC__
+# define HIDDEN __attribute__ ((__visibility__ ("hidden")))
+# define NOINLINE __attribute__ ((noinline))
+#else
+# define HIDDEN
+# define NOINLINE
+#endif
+
+HIDDEN float __math_oflowf (unsigned long);
+HIDDEN float __math_uflowf (unsigned long);
+HIDDEN float __math_may_uflowf (unsigned long);
+HIDDEN float __math_divzerof (unsigned long);
+HIDDEN float __math_invalidf (float);
+
+/* Shared between expf, exp2f and powf. */
+#define EXP2F_TABLE_BITS 5
+#define EXP2F_POLY_ORDER 3
+extern const struct exp2f_data
+{
+ uint64_t tab[1 << EXP2F_TABLE_BITS];
+ double shift_scaled;
+ double poly[EXP2F_POLY_ORDER];
+ double shift;
+ double invln2_scaled;
+ double poly_scaled[EXP2F_POLY_ORDER];
+} __exp2f_data HIDDEN;
+
+#define LOGF_TABLE_BITS 4
+#define LOGF_POLY_ORDER 4
+extern const struct logf_data
+{
+ struct
+ {
+ double invc, logc;
+ } tab[1 << LOGF_TABLE_BITS];
+ double ln2;
+ double poly[LOGF_POLY_ORDER - 1]; /* First order coefficient is 1. */
+} __logf_data HIDDEN;
+
+#define LOG2F_TABLE_BITS 4
+#define LOG2F_POLY_ORDER 4
+extern const struct log2f_data
+{
+ struct
+ {
+ double invc, logc;
+ } tab[1 << LOG2F_TABLE_BITS];
+ double poly[LOG2F_POLY_ORDER];
+} __log2f_data HIDDEN;
+
+#define POWF_LOG2_TABLE_BITS 4
+#define POWF_LOG2_POLY_ORDER 5
+#if TOINT_INTRINSICS
+# define POWF_SCALE_BITS EXP2F_TABLE_BITS
+#else
+# define POWF_SCALE_BITS 0
+#endif
+#define POWF_SCALE ((double) (1 << POWF_SCALE_BITS))
+extern const struct powf_log2_data
+{
+ struct
+ {
+ double invc, logc;
+ } tab[1 << POWF_LOG2_TABLE_BITS];
+ double poly[POWF_LOG2_POLY_ORDER];
+} __powf_log2_data HIDDEN;
+
+#endif
diff --git a/newlib/libm/common/math_errf.c b/newlib/libm/common/math_errf.c
new file mode 100644
index 000000000..60b1bb3f2
--- /dev/null
+++ b/newlib/libm/common/math_errf.c
@@ -0,0 +1,89 @@
+/* Single-precision math error handling.
+ Copyright (c) 2017 ARM Ltd. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without
+ modification, are permitted provided that the following conditions
+ are met:
+ 1. Redistributions of source code must retain the above copyright
+ notice, this list of conditions and the following disclaimer.
+ 2. Redistributions in binary form must reproduce the above copyright
+ notice, this list of conditions and the following disclaimer in the
+ documentation and/or other materials provided with the distribution.
+ 3. The name of the company may not be used to endorse or promote
+ products derived from this software without specific prior written
+ permission.
+
+ THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS AND ANY EXPRESS OR IMPLIED
+ WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+ TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */
+
+#include "fdlibm.h"
+#if !__OBSOLETE_MATH
+
+#include "math_config.h"
+
+#if WANT_ERRNO
+#include <errno.h>
+/* NOINLINE reduces code size and avoids making math functions non-leaf
+ when the error handling is inlined. */
+NOINLINE static float
+with_errnof (float y, int e)
+{
+ errno = e;
+ return y;
+}
+#else
+#define with_errnof(x, e) (x)
+#endif
+
+/* NOINLINE prevents fenv semantics breaking optimizations. */
+NOINLINE static float
+xflowf (unsigned long sign, float y)
+{
+ y = (sign ? -y : y) * y;
+ return with_errnof (y, ERANGE);
+}
+
+HIDDEN float
+__math_uflowf (unsigned long sign)
+{
+ return xflowf (sign, 0x1p-95f);
+}
+
+#if WANT_ERRNO_UFLOW
+/* Underflows to zero in some non-nearest rounding mode, setting errno
+ is valid even if the result is non-zero, but in the subnormal range. */
+HIDDEN float
+__math_may_uflowf (unsigned long sign)
+{
+ return xflowf (sign, 0x1.4p-75f);
+}
+#endif
+
+HIDDEN float
+__math_oflowf (unsigned long sign)
+{
+ return xflowf (sign, 0x1p97f);
+}
+
+HIDDEN float
+__math_divzerof (unsigned long sign)
+{
+ float y = 0;
+ return with_errnof ((sign ? -1 : 1) / y, ERANGE);
+}
+
+HIDDEN float
+__math_invalidf (float x)
+{
+ float y = (x - x) / (x - x);
+ return isnan (x) ? y : with_errnof (y, EDOM);
+}
+#endif /* !__OBSOLETE_MATH */
diff --git a/newlib/libm/common/sf_exp.c b/newlib/libm/common/sf_exp.c
new file mode 100644
index 000000000..79ec62bf5
--- /dev/null
+++ b/newlib/libm/common/sf_exp.c
@@ -0,0 +1,113 @@
+/* Single-precision e^x function.
+ Copyright (c) 2017 ARM Ltd. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without
+ modification, are permitted provided that the following conditions
+ are met:
+ 1. Redistributions of source code must retain the above copyright
+ notice, this list of conditions and the following disclaimer.
+ 2. Redistributions in binary form must reproduce the above copyright
+ notice, this list of conditions and the following disclaimer in the
+ documentation and/or other materials provided with the distribution.
+ 3. The name of the company may not be used to endorse or promote
+ products derived from this software without specific prior written
+ permission.
+
+ THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS AND ANY EXPRESS OR IMPLIED
+ WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+ TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */
+
+#include "fdlibm.h"
+#if !__OBSOLETE_MATH
+
+#include <math.h>
+#include <stdint.h>
+#include "math_config.h"
+
+/*
+EXP2F_TABLE_BITS = 5
+EXP2F_POLY_ORDER = 3
+
+ULP error: 0.502 (nearest rounding.)
+Relative error: 1.69 * 2^-34 in [-ln2/64, ln2/64] (before rounding.)
+Wrong count: 170635 (all nearest rounding wrong results with fma.)
+Non-nearest ULP error: 1 (rounded ULP error)
+*/
+
+#define N (1 << EXP2F_TABLE_BITS)
+#define InvLn2N __exp2f_data.invln2_scaled
+#define T __exp2f_data.tab
+#define C __exp2f_data.poly_scaled
+
+static inline uint32_t
+top12 (float x)
+{
+ return asuint (x) >> 20;
+}
+
+float
+expf (float x)
+{
+ uint32_t abstop;
+ uint64_t ki, t;
+ /* double_t for better performance on targets with FLT_EVAL_METHOD==2. */
+ double_t kd, xd, z, r, r2, y, s;
+
+ xd = (double_t) x;
+ abstop = top12 (x) & 0x7ff;
+ if (__builtin_expect (abstop >= top12 (88.0f), 0))
+ {
+ /* |x| >= 88 or x is nan. */
+ if (asuint (x) == asuint (-INFINITY))
+ return 0.0f;
+ if (abstop >= top12 (INFINITY))
+ return x + x;
+ if (x > 0x1.62e42ep6f) /* x > log(0x1p128) ~= 88.72 */
+ return __math_oflowf (0);
+ if (x < -0x1.9fe368p6f) /* x < log(0x1p-150) ~= -103.97 */
+ return __math_uflowf (0);
+#if WANT_ERRNO_UFLOW
+ if (x < -0x1.9d1d9ep6f) /* x < log(0x1p-149) ~= -103.28 */
+ return __math_may_uflowf (0);
+#endif
+ }
+
+ /* x*N/Ln2 = k + r with r in [-1/2, 1/2] and int k. */
+ z = InvLn2N * xd;
+
+ /* Round and convert z to int, the result is in [-150*N, 128*N] and
+ ideally ties-to-even rule is used, otherwise the magnitude of r
+ can be bigger which gives larger approximation error. */
+#if TOINT_INTRINSICS
+ kd = roundtoint (z);
+ ki = converttoint (z);
+#elif TOINT_RINT
+ kd = rint (z);
+ ki = (long) kd;
+#elif TOINT_SHIFT
+# define SHIFT __exp2f_data.shift
+ kd = (double) (z + SHIFT); /* Rounding to double precision is required. */
+ ki = asuint64 (kd);
+ kd -= SHIFT;
+#endif
+ r = z - kd;
+
+ /* exp(x) = 2^(k/N) * 2^(r/N) ~= s * (C0*r^3 + C1*r^2 + C2*r + 1) */
+ t = T[ki % N];
+ t += ki << (52 - EXP2F_TABLE_BITS);
+ s = asdouble (t);
+ z = C[0] * r + C[1];
+ r2 = r * r;
+ y = C[2] * r + 1;
+ y = z * r2 + y;
+ y = y * s;
+ return (float) y;
+}
+#endif /* !__OBSOLETE_MATH */
diff --git a/newlib/libm/common/sf_exp2.c b/newlib/libm/common/sf_exp2.c
new file mode 100644
index 000000000..75f2551c1
--- /dev/null
+++ b/newlib/libm/common/sf_exp2.c
@@ -0,0 +1,99 @@
+/* Single-precision 2^x function.
+ Copyright (c) 2017 ARM Ltd. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without
+ modification, are permitted provided that the following conditions
+ are met:
+ 1. Redistributions of source code must retain the above copyright
+ notice, this list of conditions and the following disclaimer.
+ 2. Redistributions in binary form must reproduce the above copyright
+ notice, this list of conditions and the following disclaimer in the
+ documentation and/or other materials provided with the distribution.
+ 3. The name of the company may not be used to endorse or promote
+ products derived from this software without specific prior written
+ permission.
+
+ THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS AND ANY EXPRESS OR IMPLIED
+ WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+ TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */
+
+#include "fdlibm.h"
+#if !__OBSOLETE_MATH
+
+#include <math.h>
+#include <stdint.h>
+#include "math_config.h"
+
+/*
+EXP2F_TABLE_BITS = 5
+EXP2F_POLY_ORDER = 3
+
+ULP error: 0.502 (nearest rounding.)
+Relative error: 1.69 * 2^-34 in [-1/64, 1/64] (before rounding.)
+Wrong count: 168353 (all nearest rounding wrong results with fma.)
+Non-nearest ULP error: 1 (rounded ULP error)
+*/
+
+#define N (1 << EXP2F_TABLE_BITS)
+#define T __exp2f_data.tab
+#define C __exp2f_data.poly
+#define SHIFT __exp2f_data.shift_scaled
+
+static inline uint32_t
+top12 (float x)
+{
+ return asuint (x) >> 20;
+}
+
+float
+exp2f (float x)
+{
+ uint32_t abstop;
+ uint64_t ki, t;
+ /* double_t for better performance on targets with FLT_EVAL_METHOD==2. */
+ double_t kd, xd, z, r, r2, y, s;
+
+ xd = (double_t) x;
+ abstop = top12 (x) & 0x7ff;
+ if (__builtin_expect (abstop >= top12 (128.0f), 0))
+ {
+ /* |x| >= 128 or x is nan. */
+ if (asuint (x) == asuint (-INFINITY))
+ return 0.0f;
+ if (abstop >= top12 (INFINITY))
+ return x + x;
+ if (x > 0.0f)
+ return __math_oflowf (0);
+ if (x <= -150.0f)
+ return __math_uflowf (0);
+#if WANT_ERRNO_UFLOW
+ if (x < -149.0f)
+ return __math_may_uflowf (0);
+#endif
+ }
+
+ /* x = k/N + r with r in [-1/(2N), 1/(2N)] and int k. */
+ kd = (double) (xd + SHIFT); /* Rounding to double precision is required. */
+ ki = asuint64 (kd);
+ kd -= SHIFT; /* k/N for int k. */
+ r = xd - kd;
+
+ /* exp2(x) = 2^(k/N) * 2^r ~= s * (C0*r^3 + C1*r^2 + C2*r + 1) */
+ t = T[ki % N];
+ t += ki << (52 - EXP2F_TABLE_BITS);
+ s = asdouble (t);
+ z = C[0] * r + C[1];
+ r2 = r * r;
+ y = C[2] * r + 1;
+ y = z * r2 + y;
+ y = y * s;
+ return (float) y;
+}
+#endif /* !__OBSOLETE_MATH */
diff --git a/newlib/libm/common/sf_exp2_data.c b/newlib/libm/common/sf_exp2_data.c
new file mode 100644
index 000000000..63b6f391d
--- /dev/null
+++ b/newlib/libm/common/sf_exp2_data.c
@@ -0,0 +1,56 @@
+/* Shared data between expf, exp2f and powf.
+ Copyright (c) 2017 ARM Ltd. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without
+ modification, are permitted provided that the following conditions
+ are met:
+ 1. Redistributions of source code must retain the above copyright
+ notice, this list of conditions and the following disclaimer.
+ 2. Redistributions in binary form must reproduce the above copyright
+ notice, this list of conditions and the following disclaimer in the
+ documentation and/or other materials provided with the distribution.
+ 3. The name of the company may not be used to endorse or promote
+ products derived from this software without specific prior written
+ permission.
+
+ THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS AND ANY EXPRESS OR IMPLIED
+ WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+ TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */
+
+#include "fdlibm.h"
+#if !__OBSOLETE_MATH
+
+#include "math_config.h"
+
+#define N (1 << EXP2F_TABLE_BITS)
+
+const struct exp2f_data __exp2f_data = {
+ /* tab[i] = uint(2^(i/N)) - (i << 52-BITS)
+ used for computing 2^(k/N) for an int |k| < 150 N as
+ double(tab[k%N] + (k << 52-BITS)) */
+ .tab = {
+0x3ff0000000000000, 0x3fefd9b0d3158574, 0x3fefb5586cf9890f, 0x3fef9301d0125b51,
+0x3fef72b83c7d517b, 0x3fef54873168b9aa, 0x3fef387a6e756238, 0x3fef1e9df51fdee1,
+0x3fef06fe0a31b715, 0x3feef1a7373aa9cb, 0x3feedea64c123422, 0x3feece086061892d,
+0x3feebfdad5362a27, 0x3feeb42b569d4f82, 0x3feeab07dd485429, 0x3feea47eb03a5585,
+0x3feea09e667f3bcd, 0x3fee9f75e8ec5f74, 0x3feea11473eb0187, 0x3feea589994cce13,
+0x3feeace5422aa0db, 0x3feeb737b0cdc5e5, 0x3feec49182a3f090, 0x3feed503b23e255d,
+0x3feee89f995ad3ad, 0x3feeff76f2fb5e47, 0x3fef199bdd85529c, 0x3fef3720dcef9069,
+0x3fef5818dcfba487, 0x3fef7c97337b9b5f, 0x3fefa4afa2a490da, 0x3fefd0765b6e4540,
+ },
+ .shift_scaled = 0x1.8p+52 / N,
+ .poly = { 0x1.c6af84b912394p-5, 0x1.ebfce50fac4f3p-3, 0x1.62e42ff0c52d6p-1 },
+ .shift = 0x1.8p+52,
+ .invln2_scaled = 0x1.71547652b82fep+0 * N,
+ .poly_scaled = {
+0x1.c6af84b912394p-5/N/N/N, 0x1.ebfce50fac4f3p-3/N/N, 0x1.62e42ff0c52d6p-1/N
+ },
+};
+#endif /* !__OBSOLETE_MATH */
diff --git a/newlib/libm/common/sf_log.c b/newlib/libm/common/sf_log.c
new file mode 100644
index 000000000..ba5558a54
--- /dev/null
+++ b/newlib/libm/common/sf_log.c
@@ -0,0 +1,98 @@
+/* Single-precision log function.
+ Copyright (c) 2017 ARM Ltd. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without
+ modification, are permitted provided that the following conditions
+ are met:
+ 1. Redistributions of source code must retain the above copyright
+ notice, this list of conditions and the following disclaimer.
+ 2. Redistributions in binary form must reproduce the above copyright
+ notice, this list of conditions and the following disclaimer in the
+ documentation and/or other materials provided with the distribution.
+ 3. The name of the company may not be used to endorse or promote
+ products derived from this software without specific prior written
+ permission.
+
+ THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS AND ANY EXPRESS OR IMPLIED
+ WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+ TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */
+
+#include "fdlibm.h"
+#if !__OBSOLETE_MATH
+
+#include <math.h>
+#include <stdint.h>
+#include "math_config.h"
+
+/*
+LOGF_TABLE_BITS = 4
+LOGF_POLY_ORDER = 4
+
+ULP error: 0.818 (nearest rounding.)
+Relative error: 1.957 * 2^-26 (before rounding.)
+*/
+
+#define T __logf_data.tab
+#define A __logf_data.poly
+#define Ln2 __logf_data.ln2
+#define N (1 << LOGF_TABLE_BITS)
+#define OFF 0x3f330000
+
+float
+logf (float x)
+{
+ /* double_t for better performance on targets with FLT_EVAL_METHOD==2. */
+ double_t z, r, r2, y, y0, invc, logc;
+ uint32_t ix, iz, tmp;
+ int k, i;
+
+ ix = asuint (x);
+#if WANT_ROUNDING
+ /* Fix sign of zero with downward rounding when x==1. */
+ if (__builtin_expect (ix == 0x3f800000, 0))
+ return 0;
+#endif
+ if (__builtin_expect (ix - 0x00800000 >= 0x7f800000 - 0x00800000, 0))
+ {
+ /* x < 0x1p-126 or inf or nan. */
+ if (ix * 2 == 0)
+ return __math_divzerof (1);
+ if (ix == 0x7f800000) /* log(inf) == inf. */
+ return x;
+ if ((ix & 0x80000000) || ix * 2 >= 0xff000000)
+ return __math_invalidf (x);
+ /* x is subnormal, normalize it. */
+ ix = asuint (x * 0x1p23f);
+ ix -= 23 << 23;
+ }
+
+ /* x = 2^k z; where z is in range [OFF,2*OFF] and exact.
+ The range is split into N subintervals.
+ The ith subinterval contains z and c is near its center. */
+ tmp = ix - OFF;
+ i = (tmp >> (23 - LOGF_TABLE_BITS)) % N;
+ k = (int32_t) tmp >> 23; /* arithmetic shift */
+ iz = ix - (tmp & 0x1ff << 23);
+ invc = T[i].invc;
+ logc = T[i].logc;
+ z = (double_t) asfloat (iz);
+
+ /* log(x) = log1p(z/c-1) + log(c) + k*Ln2 */
+ r = z * invc - 1;
+ y0 = logc + (double_t) k * Ln2;
+
+ /* Pipelined polynomial evaluation to approximate log1p(r). */
+ r2 = r * r;
+ y = A[1] * r + A[2];
+ y = A[0] * r2 + y;
+ y = y * r2 + (y0 + r);
+ return (float) y;
+}
+#endif /* !__OBSOLETE_MATH */
diff --git a/newlib/libm/common/sf_log2.c b/newlib/libm/common/sf_log2.c
index 9fbaaca72..f52cd385d 100644
--- a/newlib/libm/common/sf_log2.c
+++ b/newlib/libm/common/sf_log2.c
@@ -1,48 +1,99 @@
-/* sf_log2.c -- float version of s_log2.c.
- * Modification of sf_exp10.c by Yaakov Selkowitz 2009.
- */
+/* Single-precision log2 function.
+ Copyright (c) 2017 ARM Ltd. All rights reserved.
-/*
- * ====================================================
- * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
- *
- * Developed at SunPro, a Sun Microsystems, Inc. business.
- * Permission to use, copy, modify, and distribute this
- * software is freely granted, provided that this notice
- * is preserved.
- * ====================================================
- */
-
-/*
- * wrapper log2f(x)
- */
+ Redistribution and use in source and binary forms, with or without
+ modification, are permitted provided that the following conditions
+ are met:
+ 1. Redistributions of source code must retain the above copyright
+ notice, this list of conditions and the following disclaimer.
+ 2. Redistributions in binary form must reproduce the above copyright
+ notice, this list of conditions and the following disclaimer in the
+ documentation and/or other materials provided with the distribution.
+ 3. The name of the company may not be used to endorse or promote
+ products derived from this software without specific prior written
+ permission.
+
+ THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS AND ANY EXPRESS OR IMPLIED
+ WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+ TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */
#include "fdlibm.h"
-#include <errno.h>
+#if !__OBSOLETE_MATH
+
#include <math.h>
-#undef log2
-#undef log2f
-
-#ifdef __STDC__
- float log2f(float x) /* wrapper log2f */
-#else
- float log2f(x) /* wrapper log2f */
- float x;
-#endif
-{
- return (logf(x) / (float_t) M_LN2);
-}
+#include <stdint.h>
+#include "math_config.h"
-#ifdef _DOUBLE_IS_32BITS
+/*
+LOG2F_TABLE_BITS = 4
+LOG2F_POLY_ORDER = 4
-#ifdef __STDC__
- double log2(double x)
-#else
- double log2(x)
- double x;
-#endif
+ULP error: 0.752 (nearest rounding.)
+Relative error: 1.9 * 2^-26 (before rounding.)
+*/
+
+#define N (1 << LOG2F_TABLE_BITS)
+#define T __log2f_data.tab
+#define A __log2f_data.poly
+#define OFF 0x3f330000
+
+float
+log2f (float x)
{
- return (double) log2f((float) x);
-}
+ /* double_t for better performance on targets with FLT_EVAL_METHOD==2. */
+ double_t z, r, r2, p, y, y0, invc, logc;
+ uint32_t ix, iz, top, tmp;
+ int k, i;
+
+ ix = asuint (x);
+#if WANT_ROUNDING
+ /* Fix sign of zero with downward rounding when x==1. */
+ if (__builtin_expect (ix == 0x3f800000, 0))
+ return 0;
+#endif
+ if (__builtin_expect (ix - 0x00800000 >= 0x7f800000 - 0x00800000, 0))
+ {
+ /* x < 0x1p-126 or inf or nan. */
+ if (ix * 2 == 0)
+ return __math_divzerof (1);
+ if (ix == 0x7f800000) /* log2(inf) == inf. */
+ return x;
+ if ((ix & 0x80000000) || ix * 2 >= 0xff000000)
+ return __math_invalidf (x);
+ /* x is subnormal, normalize it. */
+ ix = asuint (x * 0x1p23f);
+ ix -= 23 << 23;
+ }
+
+ /* x = 2^k z; where z is in range [OFF,2*OFF] and exact.
+ The range is split into N subintervals.
+ The ith subinterval contains z and c is near its center. */
+ tmp = ix - OFF;
+ i = (tmp >> (23 - LOG2F_TABLE_BITS)) % N;
+ top = tmp & 0xff800000;
+ iz = ix - top;
+ k = (int32_t) tmp >> 23; /* arithmetic shift */
+ invc = T[i].invc;
+ logc = T[i].logc;
+ z = (double_t) asfloat (iz);
-#endif /* defined(_DOUBLE_IS_32BITS) */
+ /* log2(x) = log1p(z/c-1)/ln2 + log2(c) + k */
+ r = z * invc - 1;
+ y0 = logc + (double_t) k;
+
+ /* Pipelined polynomial evaluation to approximate log1p(r)/ln2. */
+ r2 = r * r;
+ y = A[1] * r + A[2];
+ y = A[0] * r2 + y;
+ p = A[3] * r + y0;
+ y = y * r2 + p;
+ return (float) y;
+}
+#endif /* !__OBSOLETE_MATH */
diff --git a/newlib/libm/common/sf_log2_data.c b/newlib/libm/common/sf_log2_data.c
new file mode 100644
index 000000000..f0e80947c
--- /dev/null
+++ b/newlib/libm/common/sf_log2_data.c
@@ -0,0 +1,56 @@
+/* Data definition for log2f.
+ Copyright (c) 2017 ARM Ltd. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without
+ modification, are permitted provided that the following conditions
+ are met:
+ 1. Redistributions of source code must retain the above copyright
+ notice, this list of conditions and the following disclaimer.
+ 2. Redistributions in binary form must reproduce the above copyright
+ notice, this list of conditions and the following disclaimer in the
+ documentation and/or other materials provided with the distribution.
+ 3. The name of the company may not be used to endorse or promote
+ products derived from this software without specific prior written
+ permission.
+
+ THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS AND ANY EXPRESS OR IMPLIED
+ WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+ TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */
+
+#include "fdlibm.h"
+#if !__OBSOLETE_MATH
+
+#include "math_config.h"
+
+const struct log2f_data __log2f_data = {
+ .tab = {
+ { 0x1.661ec79f8f3bep+0, -0x1.efec65b963019p-2 },
+ { 0x1.571ed4aaf883dp+0, -0x1.b0b6832d4fca4p-2 },
+ { 0x1.49539f0f010bp+0, -0x1.7418b0a1fb77bp-2 },
+ { 0x1.3c995b0b80385p+0, -0x1.39de91a6dcf7bp-2 },
+ { 0x1.30d190c8864a5p+0, -0x1.01d9bf3f2b631p-2 },
+ { 0x1.25e227b0b8eap+0, -0x1.97c1d1b3b7afp-3 },
+ { 0x1.1bb4a4a1a343fp+0, -0x1.2f9e393af3c9fp-3 },
+ { 0x1.12358f08ae5bap+0, -0x1.960cbbf788d5cp-4 },
+ { 0x1.0953f419900a7p+0, -0x1.a6f9db6475fcep-5 },
+ { 0x1p+0, 0x0p+0 },
+ { 0x1.e608cfd9a47acp-1, 0x1.338ca9f24f53dp-4 },
+ { 0x1.ca4b31f026aap-1, 0x1.476a9543891bap-3 },
+ { 0x1.b2036576afce6p-1, 0x1.e840b4ac4e4d2p-3 },
+ { 0x1.9c2d163a1aa2dp-1, 0x1.40645f0c6651cp-2 },
+ { 0x1.886e6037841edp-1, 0x1.88e9c2c1b9ff8p-2 },
+ { 0x1.767dcf5534862p-1, 0x1.ce0a44eb17bccp-2 },
+ },
+ .poly = {
+ -0x1.712b6f70a7e4dp-2, 0x1.ecabf496832ep-2, -0x1.715479ffae3dep-1,
+ 0x1.715475f35c8b8p0,
+ }
+};
+#endif /* !__OBSOLETE_MATH */
diff --git a/newlib/libm/common/sf_log_data.c b/newlib/libm/common/sf_log_data.c
new file mode 100644
index 000000000..28231b4dc
--- /dev/null
+++ b/newlib/libm/common/sf_log_data.c
@@ -0,0 +1,56 @@
+/* Data definition for logf.
+ Copyright (c) 2017 ARM Ltd. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without
+ modification, are permitted provided that the following conditions
+ are met:
+ 1. Redistributions of source code must retain the above copyright
+ notice, this list of conditions and the following disclaimer.
+ 2. Redistributions in binary form must reproduce the above copyright
+ notice, this list of conditions and the following disclaimer in the
+ documentation and/or other materials provided with the distribution.
+ 3. The name of the company may not be used to endorse or promote
+ products derived from this software without specific prior written
+ permission.
+
+ THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS AND ANY EXPRESS OR IMPLIED
+ WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+ TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */
+
+#include "fdlibm.h"
+#if !__OBSOLETE_MATH
+
+#include "math_config.h"
+
+const struct logf_data __logf_data = {
+ .tab = {
+ { 0x1.661ec79f8f3bep+0, -0x1.57bf7808caadep-2 },
+ { 0x1.571ed4aaf883dp+0, -0x1.2bef0a7c06ddbp-2 },
+ { 0x1.49539f0f010bp+0, -0x1.01eae7f513a67p-2 },
+ { 0x1.3c995b0b80385p+0, -0x1.b31d8a68224e9p-3 },
+ { 0x1.30d190c8864a5p+0, -0x1.6574f0ac07758p-3 },
+ { 0x1.25e227b0b8eap+0, -0x1.1aa2bc79c81p-3 },
+ { 0x1.1bb4a4a1a343fp+0, -0x1.a4e76ce8c0e5ep-4 },
+ { 0x1.12358f08ae5bap+0, -0x1.1973c5a611cccp-4 },
+ { 0x1.0953f419900a7p+0, -0x1.252f438e10c1ep-5 },
+ { 0x1p+0, 0x0p+0 },
+ { 0x1.e608cfd9a47acp-1, 0x1.aa5aa5df25984p-5 },
+ { 0x1.ca4b31f026aap-1, 0x1.c5e53aa362eb4p-4 },
+ { 0x1.b2036576afce6p-1, 0x1.526e57720db08p-3 },
+ { 0x1.9c2d163a1aa2dp-1, 0x1.bc2860d22477p-3 },
+ { 0x1.886e6037841edp-1, 0x1.1058bc8a07ee1p-2 },
+ { 0x1.767dcf5534862p-1, 0x1.4043057b6ee09p-2 },
+ },
+ .ln2 = 0x1.62e42fefa39efp-1,
+ .poly = {
+ -0x1.00ea348b88334p-2, 0x1.5575b0be00b6ap-2, -0x1.ffffef20a4123p-2,
+ }
+};
+#endif /* !__OBSOLETE_MATH */
diff --git a/newlib/libm/common/sf_pow.c b/newlib/libm/common/sf_pow.c
new file mode 100644
index 000000000..6741a349b
--- /dev/null
+++ b/newlib/libm/common/sf_pow.c
@@ -0,0 +1,230 @@
+/* Single-precision pow function.
+ Copyright (c) 2017 ARM Ltd. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without
+ modification, are permitted provided that the following conditions
+ are met:
+ 1. Redistributions of source code must retain the above copyright
+ notice, this list of conditions and the following disclaimer.
+ 2. Redistributions in binary form must reproduce the above copyright
+ notice, this list of conditions and the following disclaimer in the
+ documentation and/or other materials provided with the distribution.
+ 3. The name of the company may not be used to endorse or promote
+ products derived from this software without specific prior written
+ permission.
+
+ THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS AND ANY EXPRESS OR IMPLIED
+ WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+ TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */
+
+#include "fdlibm.h"
+#if !__OBSOLETE_MATH
+
+#include <math.h>
+#include <stdint.h>
+#include "math_config.h"
+
+/*
+POWF_LOG2_POLY_ORDER = 5
+EXP2F_TABLE_BITS = 5
+
+ULP error: 0.82 (~ 0.5 + relerr*2^24)
+relerr: 1.27 * 2^-26 (Relative error ~= 128*Ln2*relerr_log2 + relerr_exp2)
+relerr_log2: 1.83 * 2^-33 (Relative error of logx.)
+relerr_exp2: 1.69 * 2^-34 (Relative error of exp2(ylogx).)
+*/
+
+#define N (1 << POWF_LOG2_TABLE_BITS)
+#define T __powf_log2_data.tab
+#define A __powf_log2_data.poly
+#define OFF 0x3f330000
+
+/* Subnormal input is normalized so ix has negative biased exponent.
+ Output is multiplied by N (POWF_SCALE) if TOINT_INTRINICS is set. */
+static inline double_t
+log2_inline (uint32_t ix)
+{
+ /* double_t for better performance on targets with FLT_EVAL_METHOD==2. */
+ double_t z, r, r2, r4, p, q, y, y0, invc, logc;
+ uint32_t iz, top, tmp;
+ int k, i;
+
+ /* x = 2^k z; where z is in range [OFF,2*OFF] and exact.
+ The range is split into N subintervals.
+ The ith subinterval contains z and c is near its center. */
+ tmp = ix - OFF;
+ i = (tmp >> (23 - POWF_LOG2_TABLE_BITS)) % N;
+ top = tmp & 0xff800000;
+ iz = ix - top;
+ k = (int32_t) top >> (23 - POWF_SCALE_BITS); /* arithmetic shift */
+ invc = T[i].invc;
+ logc = T[i].logc;
+ z = (double_t) asfloat (iz);
+
+ /* log2(x) = log1p(z/c-1)/ln2 + log2(c) + k */
+ r = z * invc - 1;
+ y0 = logc + (double_t) k;
+
+ /* Pipelined polynomial evaluation to approximate log1p(r)/ln2. */
+ r2 = r * r;
+ y = A[0] * r + A[1];
+ p = A[2] * r + A[3];
+ r4 = r2 * r2;
+ q = A[4] * r + y0;
+ q = p * r2 + q;
+ y = y * r4 + q;
+ return y;
+}
+
+#undef N
+#undef T
+#define N (1 << EXP2F_TABLE_BITS)
+#define T __exp2f_data.tab
+#define SIGN_BIAS (1 << (EXP2F_TABLE_BITS + 11))
+
+/* The output of log2 and thus the input of exp2 is either scaled by N
+ (in case of fast toint intrinsics) or not. The unscaled xd must be
+ in [-1021,1023], sign_bias sets the sign of the result. */
+static inline double_t
+exp2_inline (double_t xd, unsigned long sign_bias)
+{
+ uint64_t ki, ski, t;
+ /* double_t for better performance on targets with FLT_EVAL_METHOD==2. */
+ double_t kd, z, r, r2, y, s;
+
+#if TOINT_INTRINSICS
+# define C __exp2f_data.poly_scaled
+ /* N*x = k + r with r in [-1/2, 1/2] */
+ kd = roundtoint (xd); /* k */
+ ki = converttoint (xd);
+#else
+# define C __exp2f_data.poly
+# define SHIFT __exp2f_data.shift_scaled
+ /* x = k/N + r with r in [-1/(2N), 1/(2N)] */
+ kd = (double) (xd + SHIFT); /* Rounding to double precision is required. */
+ ki = asuint64 (kd);
+ kd -= SHIFT; /* k/N */
+#endif
+ r = xd - kd;
+
+ /* exp2(x) = 2^(k/N) * 2^r ~= s * (C0*r^3 + C1*r^2 + C2*r + 1) */
+ t = T[ki % N];
+ ski = ki + sign_bias;
+ t += ski << (52 - EXP2F_TABLE_BITS);
+ s = asdouble (t);
+ z = C[0] * r + C[1];
+ r2 = r * r;
+ y = C[2] * r + 1;
+ y = z * r2 + y;
+ y = y * s;
+ return y;
+}
+
+/* Returns 0 if not int, 1 if odd int, 2 if even int. */
+static inline int
+checkint (uint32_t iy)
+{
+ int e = iy >> 23 & 0xff;
+ if (e < 0x7f)
+ return 0;
+ if (e > 0x7f + 23)
+ return 2;
+ if (iy & ((1 << (0x7f + 23 - e)) - 1))
+ return 0;
+ if (iy & (1 << (0x7f + 23 - e)))
+ return 1;
+ return 2;
+}
+
+static inline int
+zeroinfnan (uint32_t ix)
+{
+ return 2 * ix - 1 >= 2u * 0x7f800000 - 1;
+}
+
+float
+powf (float x, float y)
+{
+ unsigned long sign_bias = 0;
+ uint32_t ix, iy;
+
+ ix = asuint (x);
+ iy = asuint (y);
+ if (__builtin_expect (ix - 0x00800000 >= 0x7f800000 - 0x00800000
+ || zeroinfnan (iy),
+ 0))
+ {
+ /* Either (x < 0x1p-126 or inf or nan) or (y is 0 or inf or nan). */
+ if (__builtin_expect (zeroinfnan (iy), 0))
+ {
+ if (2 * iy == 0)
+ return issignalingf_inline (x) ? x + y : 1.0f;
+ if (ix == 0x3f800000)
+ return issignalingf_inline (y) ? x + y : 1.0f;
+ if (2 * ix > 2u * 0x7f800000 || 2 * iy > 2u * 0x7f800000)
+ return x + y;
+ if (2 * ix == 2 * 0x3f800000)
+ return 1.0f;
+ if ((2 * ix < 2 * 0x3f800000) == !(iy & 0x80000000))
+ return 0.0f; /* |x|<1 && y==inf or |x|>1 && y==-inf. */
+ return y * y;
+ }
+ if (__builtin_expect (zeroinfnan (ix), 0))
+ {
+ float_t x2 = x * x;
+ if (ix & 0x80000000 && checkint (iy) == 1)
+ {
+ x2 = -x2;
+ sign_bias = 1;
+ }
+#if WANT_ERRNO
+ if (2 * ix == 0 && iy & 0x80000000)
+ return __math_divzerof (sign_bias);
+#endif
+ return iy & 0x80000000 ? 1 / x2 : x2;
+ }
+ /* x and y are non-zero finite. */
+ if (ix & 0x80000000)
+ {
+ /* Finite x < 0. */
+ int yint = checkint (iy);
+ if (yint == 0)
+ return __math_invalidf (x);
+ if (yint == 1)
+ sign_bias = SIGN_BIAS;
+ ix &= 0x7fffffff;
+ }
+ if (ix < 0x00800000)
+ {
+ /* Normalize subnormal x so exponent becomes negative. */
+ ix = asuint (x * 0x1p23f);
+ ix &= 0x7fffffff;
+ ix -= 23 << 23;
+ }
+ }
+ double_t logx = log2_inline (ix);
+ double_t ylogx = y * logx; /* Note: cannot overflow, y is single prec. */
+ if (__builtin_expect ((asuint64 (ylogx) >> 47 & 0xffff)
+ >= asuint64 (126.0 * POWF_SCALE) >> 47,
+ 0))
+ {
+ /* |y*log(x)| >= 126. */
+ if (ylogx > 0x1.fffffffd1d571p+6 * POWF_SCALE)
+ return __math_oflowf (sign_bias);
+ if (ylogx <= -150.0 * POWF_SCALE)
+ return __math_uflowf (sign_bias);
+#if WANT_ERRNO_UFLOW
+ if (ylogx < -149.0 * POWF_SCALE)
+ return __math_may_uflowf (sign_bias);
+#endif
+ }
+ return (float) exp2_inline (ylogx, sign_bias);
+}
+#endif /* !__OBSOLETE_MATH */
diff --git a/newlib/libm/common/sf_pow_log2_data.c b/newlib/libm/common/sf_pow_log2_data.c
new file mode 100644
index 000000000..24f638c19
--- /dev/null
+++ b/newlib/libm/common/sf_pow_log2_data.c
@@ -0,0 +1,57 @@
+/* Data definition for powf.
+ Copyright (c) 2017 ARM Ltd. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without
+ modification, are permitted provided that the following conditions
+ are met:
+ 1. Redistributions of source code must retain the above copyright
+ notice, this list of conditions and the following disclaimer.
+ 2. Redistributions in binary form must reproduce the above copyright
+ notice, this list of conditions and the following disclaimer in the
+ documentation and/or other materials provided with the distribution.
+ 3. The name of the company may not be used to endorse or promote
+ products derived from this software without specific prior written
+ permission.
+
+ THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS AND ANY EXPRESS OR IMPLIED
+ WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+ TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */
+
+#include "fdlibm.h"
+#if !__OBSOLETE_MATH
+
+#include "math_config.h"
+
+const struct powf_log2_data __powf_log2_data = {
+ .tab = {
+ { 0x1.661ec79f8f3bep+0, -0x1.efec65b963019p-2 * POWF_SCALE },
+ { 0x1.571ed4aaf883dp+0, -0x1.b0b6832d4fca4p-2 * POWF_SCALE },
+ { 0x1.49539f0f010bp+0, -0x1.7418b0a1fb77bp-2 * POWF_SCALE },
+ { 0x1.3c995b0b80385p+0, -0x1.39de91a6dcf7bp-2 * POWF_SCALE },
+ { 0x1.30d190c8864a5p+0, -0x1.01d9bf3f2b631p-2 * POWF_SCALE },
+ { 0x1.25e227b0b8eap+0, -0x1.97c1d1b3b7afp-3 * POWF_SCALE },
+ { 0x1.1bb4a4a1a343fp+0, -0x1.2f9e393af3c9fp-3 * POWF_SCALE },
+ { 0x1.12358f08ae5bap+0, -0x1.960cbbf788d5cp-4 * POWF_SCALE },
+ { 0x1.0953f419900a7p+0, -0x1.a6f9db6475fcep-5 * POWF_SCALE },
+ { 0x1p+0, 0x0p+0 * POWF_SCALE },
+ { 0x1.e608cfd9a47acp-1, 0x1.338ca9f24f53dp-4 * POWF_SCALE },
+ { 0x1.ca4b31f026aap-1, 0x1.476a9543891bap-3 * POWF_SCALE },
+ { 0x1.b2036576afce6p-1, 0x1.e840b4ac4e4d2p-3 * POWF_SCALE },
+ { 0x1.9c2d163a1aa2dp-1, 0x1.40645f0c6651cp-2 * POWF_SCALE },
+ { 0x1.886e6037841edp-1, 0x1.88e9c2c1b9ff8p-2 * POWF_SCALE },
+ { 0x1.767dcf5534862p-1, 0x1.ce0a44eb17bccp-2 * POWF_SCALE },
+ },
+ .poly = {
+ 0x1.27616c9496e0bp-2 * POWF_SCALE, -0x1.71969a075c67ap-2 * POWF_SCALE,
+ 0x1.ec70a6ca7baddp-2 * POWF_SCALE, -0x1.7154748bef6c8p-1 * POWF_SCALE,
+ 0x1.71547652ab82bp0 * POWF_SCALE,
+ }
+};
+#endif /* !__OBSOLETE_MATH */
diff --git a/newlib/libm/math/Makefile.am b/newlib/libm/math/Makefile.am
index 9e8e92840..e745159ae 100644
--- a/newlib/libm/math/Makefile.am
+++ b/newlib/libm/math/Makefile.am
@@ -48,7 +48,8 @@ fsrc = kf_rem_pio2.c \
sf_frexp.c sf_ldexp.c \
sf_signif.c sf_sin.c \
sf_tan.c sf_tanh.c \
- wf_exp2.c wf_tgamma.c
+ wf_exp2.c wf_tgamma.c \
+ wf_log2.c
lsrc = el_hypot.c
diff --git a/newlib/libm/math/Makefile.in b/newlib/libm/math/Makefile.in
index 8f76aad4e..1f5a69406 100644
--- a/newlib/libm/math/Makefile.in
+++ b/newlib/libm/math/Makefile.in
@@ -137,7 +137,8 @@ am__objects_2 = lib_a-kf_rem_pio2.$(OBJEXT) lib_a-kf_cos.$(OBJEXT) \
lib_a-sf_frexp.$(OBJEXT) lib_a-sf_ldexp.$(OBJEXT) \
lib_a-sf_signif.$(OBJEXT) lib_a-sf_sin.$(OBJEXT) \
lib_a-sf_tan.$(OBJEXT) lib_a-sf_tanh.$(OBJEXT) \
- lib_a-wf_exp2.$(OBJEXT) lib_a-wf_tgamma.$(OBJEXT)
+ lib_a-wf_exp2.$(OBJEXT) lib_a-wf_tgamma.$(OBJEXT) \
+ lib_a-wf_log2.$(OBJEXT)
am__objects_3 = lib_a-el_hypot.$(OBJEXT)
@USE_LIBTOOL_FALSE@am_lib_a_OBJECTS = $(am__objects_1) \
@USE_LIBTOOL_FALSE@ $(am__objects_2) $(am__objects_3)
@@ -169,7 +170,7 @@ am__objects_5 = kf_rem_pio2.lo kf_cos.lo kf_sin.lo kf_tan.lo \
wf_sincos.lo wf_drem.lo sf_asinh.lo sf_atan.lo sf_ceil.lo \
sf_cos.lo sf_erf.lo sf_fabs.lo sf_floor.lo sf_frexp.lo \
sf_ldexp.lo sf_signif.lo sf_sin.lo sf_tan.lo sf_tanh.lo \
- wf_exp2.lo wf_tgamma.lo
+ wf_exp2.lo wf_tgamma.lo wf_log2.lo
am__objects_6 = el_hypot.lo
@USE_LIBTOOL_TRUE@am_libmath_la_OBJECTS = $(am__objects_4) \
@USE_LIBTOOL_TRUE@ $(am__objects_5) $(am__objects_6)
@@ -372,7 +373,8 @@ fsrc = kf_rem_pio2.c \
sf_frexp.c sf_ldexp.c \
sf_signif.c sf_sin.c \
sf_tan.c sf_tanh.c \
- wf_exp2.c wf_tgamma.c
+ wf_exp2.c wf_tgamma.c \
+ wf_log2.c
lsrc = el_hypot.c
libmath_la_LDFLAGS = -Xcompiler -nostdlib
@@ -1270,6 +1272,12 @@ lib_a-wf_tgamma.o: wf_tgamma.c
lib_a-wf_tgamma.obj: wf_tgamma.c
$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-wf_tgamma.obj `if test -f 'wf_tgamma.c'; then $(CYGPATH_W) 'wf_tgamma.c'; else $(CYGPATH_W) '$(srcdir)/wf_tgamma.c'; fi`
+lib_a-wf_log2.o: wf_log2.c
+ $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-wf_log2.o `test -f 'wf_log2.c' || echo '$(srcdir)/'`wf_log2.c
+
+lib_a-wf_log2.obj: wf_log2.c
+ $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-wf_log2.obj `if test -f 'wf_log2.c'; then $(CYGPATH_W) 'wf_log2.c'; else $(CYGPATH_W) '$(srcdir)/wf_log2.c'; fi`
+
lib_a-el_hypot.o: el_hypot.c
$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-el_hypot.o `test -f 'el_hypot.c' || echo '$(srcdir)/'`el_hypot.c
diff --git a/newlib/libm/math/ef_exp.c b/newlib/libm/math/ef_exp.c
index 6824f991a..e817370ac 100644
--- a/newlib/libm/math/ef_exp.c
+++ b/newlib/libm/math/ef_exp.c
@@ -15,6 +15,7 @@
#include "fdlibm.h"
+#if __OBSOLETE_MATH
#ifdef __v810__
#define const
#endif
@@ -97,3 +98,4 @@ P5 = 4.1381369442e-08; /* 0x3331bb4c */
return y*twom100;
}
}
+#endif /* __OBSOLETE_MATH */
diff --git a/newlib/libm/math/ef_log.c b/newlib/libm/math/ef_log.c
index 619fe9090..8dc902478 100644
--- a/newlib/libm/math/ef_log.c
+++ b/newlib/libm/math/ef_log.c
@@ -15,6 +15,7 @@
#include "fdlibm.h"
+#if __OBSOLETE_MATH
#ifdef __STDC__
static const float
#else
@@ -90,3 +91,4 @@ static float zero = 0.0;
return dk*ln2_hi-((s*(f-R)-dk*ln2_lo)-f);
}
}
+#endif /* __OBSOLETE_MATH */
diff --git a/newlib/libm/math/ef_pow.c b/newlib/libm/math/ef_pow.c
index b3041dbdc..524e3f9b0 100644
--- a/newlib/libm/math/ef_pow.c
+++ b/newlib/libm/math/ef_pow.c
@@ -15,6 +15,7 @@
#include "fdlibm.h"
+#if __OBSOLETE_MATH
#ifdef __v810__
#define const
#endif
@@ -253,3 +254,4 @@ ivln2_l = 7.0526075433e-06; /* 0x36eca570 =1/ln2 tail*/
else SET_FLOAT_WORD(z,j);
return s*z;
}
+#endif /* __OBSOLETE_MATH */
diff --git a/newlib/libm/math/wf_exp.c b/newlib/libm/math/wf_exp.c
index 70f4459b4..f16af1d1b 100644
--- a/newlib/libm/math/wf_exp.c
+++ b/newlib/libm/math/wf_exp.c
@@ -18,6 +18,7 @@
*/
#include "fdlibm.h"
+#if __OBSOLETE_MATH
#include <errno.h>
#ifdef __STDC__
@@ -101,3 +102,4 @@ u_threshold= -1.0397208405e+02; /* 0xc2cff1b5 */
}
#endif /* defined(_DOUBLE_IS_32BITS) */
+#endif /* __OBSOLETE_MATH */
diff --git a/newlib/libm/math/wf_exp2.c b/newlib/libm/math/wf_exp2.c
index 944031405..5dce9c832 100644
--- a/newlib/libm/math/wf_exp2.c
+++ b/newlib/libm/math/wf_exp2.c
@@ -18,6 +18,7 @@
*/
#include "fdlibm.h"
+#if __OBSOLETE_MATH
#include <errno.h>
#include <math.h>
@@ -44,3 +45,4 @@
}
#endif /* defined(_DOUBLE_IS_32BITS) */
+#endif /* __OBSOLETE_MATH */
diff --git a/newlib/libm/math/wf_log.c b/newlib/libm/math/wf_log.c
index 4518b863b..07be8d63c 100644
--- a/newlib/libm/math/wf_log.c
+++ b/newlib/libm/math/wf_log.c
@@ -18,6 +18,7 @@
*/
#include "fdlibm.h"
+#if __OBSOLETE_MATH
#include <errno.h>
#ifdef __STDC__
@@ -84,3 +85,4 @@
}
#endif /* defined(_DOUBLE_IS_32BITS) */
+#endif /* __OBSOLETE_MATH */
diff --git a/newlib/libm/math/wf_log2.c b/newlib/libm/math/wf_log2.c
new file mode 100644
index 000000000..9785a4fea
--- /dev/null
+++ b/newlib/libm/math/wf_log2.c
@@ -0,0 +1,50 @@
+/* wf_log2.c -- float version of s_log2.c.
+ * Modification of sf_exp10.c by Yaakov Selkowitz 2009.
+ */
+
+/*
+ * ====================================================
+ * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
+ *
+ * Developed at SunPro, a Sun Microsystems, Inc. business.
+ * Permission to use, copy, modify, and distribute this
+ * software is freely granted, provided that this notice
+ * is preserved.
+ * ====================================================
+ */
+
+/*
+ * wrapper log2f(x)
+ */
+
+#include "fdlibm.h"
+#if __OBSOLETE_MATH
+#include <errno.h>
+#include <math.h>
+#undef log2
+#undef log2f
+
+#ifdef __STDC__
+ float log2f(float x) /* wrapper log2f */
+#else
+ float log2f(x) /* wrapper log2f */
+ float x;
+#endif
+{
+ return (logf(x) / (float_t) M_LN2);
+}
+
+#ifdef _DOUBLE_IS_32BITS
+
+#ifdef __STDC__
+ double log2(double x)
+#else
+ double log2(x)
+ double x;
+#endif
+{
+ return (double) log2f((float) x);
+}
+
+#endif /* defined(_DOUBLE_IS_32BITS) */
+#endif /* __OBSOLETE_MATH */
diff --git a/newlib/libm/math/wf_pow.c b/newlib/libm/math/wf_pow.c
index a30f8808e..be453558b 100644
--- a/newlib/libm/math/wf_pow.c
+++ b/newlib/libm/math/wf_pow.c
@@ -18,6 +18,7 @@
*/
#include "fdlibm.h"
+#if __OBSOLETE_MATH
#include <errno.h>
#ifdef __STDC__
@@ -177,3 +178,4 @@
}
#endif /* defined(_DOUBLE_IS_32BITS) */
+#endif /* __OBSOLETE_MATH */
--
2.11.0
^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [PATCH v2] New expf, exp2f, logf, log2f and powf implementations
2017-09-29 18:45 ` [PATCH v2] " Szabolcs Nagy
@ 2017-10-11 11:57 ` Szabolcs Nagy
2017-10-11 12:58 ` Corinna Vinschen
1 sibling, 0 replies; 10+ messages in thread
From: Szabolcs Nagy @ 2017-10-11 11:57 UTC (permalink / raw)
To: newlib; +Cc: nd
On 29/09/17 11:41, Szabolcs Nagy wrote:
> v2:
> - Use __OBSOLETE_MATH and __OBSOLETE_MATH_DEFAULT.
> - Upstream change: whitespace code style fixes.
> - Upstream change: HIDDEN attribute for data symbols.
> - Upstream change: __logf_data.poly array size is reduced.
> - Upstream change: issignalingf_inline now has code for non-standard SNaN.
ping.
^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [PATCH v2] New expf, exp2f, logf, log2f and powf implementations
2017-09-29 18:45 ` [PATCH v2] " Szabolcs Nagy
2017-10-11 11:57 ` Szabolcs Nagy
@ 2017-10-11 12:58 ` Corinna Vinschen
2017-10-13 1:31 ` Jeff Johnston
1 sibling, 1 reply; 10+ messages in thread
From: Corinna Vinschen @ 2017-10-11 12:58 UTC (permalink / raw)
To: Szabolcs Nagy, Jeff Johnston; +Cc: newlib, nd
[-- Attachment #1: Type: text/plain, Size: 1894 bytes --]
On Sep 29 11:41, Szabolcs Nagy wrote:
> On 05/09/17 09:38, Corinna Vinschen wrote:
> > Ok, I retract my reservations against OBSOLETE_MATH as such. However,
> > there are a few problems with the current patch:
> >
> > - Rename to __OBSOLETE_MATH (two underscores) so as not to spoil the
> > namespace.
> >
> > - The current implementation disallows a developer to override the
> > setting from, e. g., sys/config.h or the command line. As a minimum,
> > __OBSOLETE_MATH should only be set if it isn't already set, and the
> > checks for __OBSOLETE_MATH should check for being != 0:
> >
> > - #define __TGT_OBSOLETE_MATH 1 in arm and aarch cases
> >
> > - At the end of the file
> >
> > #ifndef __OBSOLETE_MATH
> > #define __OBSOLETE_MATH __TGT_OBSOLETE_MATH
> > #endif
> >
> > - Further checks in the code:
> >
> > #if __OBSOLETE_MATH
> >
> > rather than
> >
> > #ifdef __OBSOLETE_MATH
> >
> > - The header of ieeefp.h explains the options. Add __TGT_OBSOLETE_MATH /
> > __OBSOLETE_MATH and add an informative comment along the lines of the
> > above text, so developers have enough info to decide if they want to
> > enable this for their target.
> >
> > Not sure if __TGT_OBSOLETE_MATH is a good name, though. I'm open for
> > better suggestions.
> >
>
> i used __OBSOLETE_MATH_DEFAULT.
>
> v2:
> - Use __OBSOLETE_MATH and __OBSOLETE_MATH_DEFAULT.
> - Upstream change: whitespace code style fixes.
> - Upstream change: HIDDEN attribute for data symbols.
> - Upstream change: __logf_data.poly array size is reduced.
> - Upstream change: issignalingf_inline now has code for non-standard SNaN.
The above points have been fixed, so I'm ok with the change if
it makes sense from an FP point of view. Jeff?
Thanks,
Corinna
--
Corinna Vinschen
Cygwin Maintainer
Red Hat
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]
^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [PATCH v2] New expf, exp2f, logf, log2f and powf implementations
2017-10-11 12:58 ` Corinna Vinschen
@ 2017-10-13 1:31 ` Jeff Johnston
2017-10-13 9:15 ` Corinna Vinschen
0 siblings, 1 reply; 10+ messages in thread
From: Jeff Johnston @ 2017-10-13 1:31 UTC (permalink / raw)
To: Newlib, Szabolcs Nagy, Jeff Johnston, nd
On Wed, Oct 11, 2017 at 8:05 AM, Corinna Vinschen <vinschen@redhat.com>
wrote:
> On Sep 29 11:41, Szabolcs Nagy wrote:
> > On 05/09/17 09:38, Corinna Vinschen wrote:
> > > Ok, I retract my reservations against OBSOLETE_MATH as such. However,
> > > there are a few problems with the current patch:
> > >
> > > - Rename to __OBSOLETE_MATH (two underscores) so as not to spoil the
> > > namespace.
> > >
> > > - The current implementation disallows a developer to override the
> > > setting from, e. g., sys/config.h or the command line. As a minimum,
> > > __OBSOLETE_MATH should only be set if it isn't already set, and the
> > > checks for __OBSOLETE_MATH should check for being != 0:
> > >
> > > - #define __TGT_OBSOLETE_MATH 1 in arm and aarch cases
> > >
> > > - At the end of the file
> > >
> > > #ifndef __OBSOLETE_MATH
> > > #define __OBSOLETE_MATH __TGT_OBSOLETE_MATH
> > > #endif
> > >
> > > - Further checks in the code:
> > >
> > > #if __OBSOLETE_MATH
> > >
> > > rather than
> > >
> > > #ifdef __OBSOLETE_MATH
> > >
> > > - The header of ieeefp.h explains the options. Add
> __TGT_OBSOLETE_MATH /
> > > __OBSOLETE_MATH and add an informative comment along the lines of the
> > > above text, so developers have enough info to decide if they want to
> > > enable this for their target.
> > >
> > > Not sure if __TGT_OBSOLETE_MATH is a good name, though. I'm open for
> > > better suggestions.
> > >
> >
> > i used __OBSOLETE_MATH_DEFAULT.
> >
> > v2:
> > - Use __OBSOLETE_MATH and __OBSOLETE_MATH_DEFAULT.
> > - Upstream change: whitespace code style fixes.
> > - Upstream change: HIDDEN attribute for data symbols.
> > - Upstream change: __logf_data.poly array size is reduced.
> > - Upstream change: issignalingf_inline now has code for non-standard
> SNaN.
>
> The above points have been fixed, so I'm ok with the change if
> it makes sense from an FP point of view. Jeff?
>
I am fine with it. Please go ahead.
>
> Thanks,
> Corinna
>
> --
> Corinna Vinschen
> Cygwin Maintainer
> Red Hat
>
^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [PATCH v2] New expf, exp2f, logf, log2f and powf implementations
2017-10-13 1:31 ` Jeff Johnston
@ 2017-10-13 9:15 ` Corinna Vinschen
2017-10-17 13:01 ` Szabolcs Nagy
0 siblings, 1 reply; 10+ messages in thread
From: Corinna Vinschen @ 2017-10-13 9:15 UTC (permalink / raw)
To: Jeff Johnston; +Cc: Newlib, Szabolcs Nagy, nd
[-- Attachment #1: Type: text/plain, Size: 2302 bytes --]
On Oct 11 12:28, Jeff Johnston wrote:
> On Wed, Oct 11, 2017 at 8:05 AM, Corinna Vinschen <vinschen@redhat.com>
> wrote:
>
> > On Sep 29 11:41, Szabolcs Nagy wrote:
> > > On 05/09/17 09:38, Corinna Vinschen wrote:
> > > > Ok, I retract my reservations against OBSOLETE_MATH as such. However,
> > > > there are a few problems with the current patch:
> > > >
> > > > - Rename to __OBSOLETE_MATH (two underscores) so as not to spoil the
> > > > namespace.
> > > >
> > > > - The current implementation disallows a developer to override the
> > > > setting from, e. g., sys/config.h or the command line. As a minimum,
> > > > __OBSOLETE_MATH should only be set if it isn't already set, and the
> > > > checks for __OBSOLETE_MATH should check for being != 0:
> > > >
> > > > - #define __TGT_OBSOLETE_MATH 1 in arm and aarch cases
> > > >
> > > > - At the end of the file
> > > >
> > > > #ifndef __OBSOLETE_MATH
> > > > #define __OBSOLETE_MATH __TGT_OBSOLETE_MATH
> > > > #endif
> > > >
> > > > - Further checks in the code:
> > > >
> > > > #if __OBSOLETE_MATH
> > > >
> > > > rather than
> > > >
> > > > #ifdef __OBSOLETE_MATH
> > > >
> > > > - The header of ieeefp.h explains the options. Add
> > __TGT_OBSOLETE_MATH /
> > > > __OBSOLETE_MATH and add an informative comment along the lines of the
> > > > above text, so developers have enough info to decide if they want to
> > > > enable this for their target.
> > > >
> > > > Not sure if __TGT_OBSOLETE_MATH is a good name, though. I'm open for
> > > > better suggestions.
> > > >
> > >
> > > i used __OBSOLETE_MATH_DEFAULT.
> > >
> > > v2:
> > > - Use __OBSOLETE_MATH and __OBSOLETE_MATH_DEFAULT.
> > > - Upstream change: whitespace code style fixes.
> > > - Upstream change: HIDDEN attribute for data symbols.
> > > - Upstream change: __logf_data.poly array size is reduced.
> > > - Upstream change: issignalingf_inline now has code for non-standard
> > SNaN.
> >
> > The above points have been fixed, so I'm ok with the change if
> > it makes sense from an FP point of view. Jeff?
> >
>
> I am fine with it. Please go ahead.
You have push perms, too ;)
Pushed.
Thanks,
Corinna
--
Corinna Vinschen
Cygwin Maintainer
Red Hat
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]
^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [PATCH v2] New expf, exp2f, logf, log2f and powf implementations
2017-10-13 9:15 ` Corinna Vinschen
@ 2017-10-17 13:01 ` Szabolcs Nagy
2017-10-19 1:01 ` Szabolcs Nagy
0 siblings, 1 reply; 10+ messages in thread
From: Szabolcs Nagy @ 2017-10-17 13:01 UTC (permalink / raw)
To: Jeff Johnston, Newlib; +Cc: nd
On 13/10/17 10:13, Corinna Vinschen wrote:
>
> Pushed.
>
sorry, i did not realize that there are internal calls
to __ieee754_logf and __ieee754_expf from various math
functions.
in the new math code i only define logf and expf symbols.
(same for powf, but __ieee754_powf is not used in newlib)
i can reintroduce the __ieee754* symbols as trivial wrappers
or weak aliases or macro redirect.
i think a macro is probably the best since there is no
requirement to make these symbols extern linkable, they are
purely implementation internal so fdlibm.h could have:
#if !__OBSOLETE_MATH
# define __ieee754_expf(x) expf(x)
# define __ieee754_logf(x) logf(x)
# define __ieee754_powf(x,y) powf(x,y)
#endif
this changes the semantics a bit: __ieee754_* symbols didn't
do error handling previously, however the error code paths
should not really matter for internal calls, if they are ever
reached then at worst errno is set spuriously in the new code.
does this sound ok?
^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [PATCH v2] New expf, exp2f, logf, log2f and powf implementations
2017-10-17 13:01 ` Szabolcs Nagy
@ 2017-10-19 1:01 ` Szabolcs Nagy
2017-10-23 17:16 ` Corinna Vinschen
0 siblings, 1 reply; 10+ messages in thread
From: Szabolcs Nagy @ 2017-10-19 1:01 UTC (permalink / raw)
To: Jeff Johnston, Newlib; +Cc: nd
[-- Attachment #1: Type: text/plain, Size: 1382 bytes --]
On 17/10/17 12:40, Szabolcs Nagy wrote:
> On 13/10/17 10:13, Corinna Vinschen wrote:
>>
>> Pushed.
>>
>
> sorry, i did not realize that there are internal calls
> to __ieee754_logf and __ieee754_expf from various math
> functions.
>
> in the new math code i only define logf and expf symbols.
> (same for powf, but __ieee754_powf is not used in newlib)
>
> i can reintroduce the __ieee754* symbols as trivial wrappers
> or weak aliases or macro redirect.
>
> i think a macro is probably the best since there is no
> requirement to make these symbols extern linkable, they are
> purely implementation internal so fdlibm.h could have:
>
> #if !__OBSOLETE_MATH
> # define __ieee754_expf(x) expf(x)
> # define __ieee754_logf(x) logf(x)
> # define __ieee754_powf(x,y) powf(x,y)
> #endif
>
> this changes the semantics a bit: __ieee754_* symbols didn't
> do error handling previously, however the error code paths
> should not really matter for internal calls, if they are ever
> reached then at worst errno is set spuriously in the new code.
>
> does this sound ok?
>
implemented this approach in the attached patch, verified
that there are no longer undefined __ieee754_* symbols.
adding __ieee754* symbol aliases is safer in principle
(e.g. user defined expf would not override internal calls),
but i'm not sure if i can use weak alias in newlib, so
i went with the macros.
[-- Attachment #2: 0001-fix-internal-__ieee754_expf-and-__ieee754_logf-calls.patch --]
[-- Type: text/x-patch, Size: 1970 bytes --]
From 23753d87bfcf048c5fa0a297d7cce48b733b696b Mon Sep 17 00:00:00 2001
From: Szabolcs Nagy <szabolcs.nagy@arm.com>
Date: Tue, 17 Oct 2017 12:41:20 +0100
Subject: [PATCH] fix internal __ieee754_expf and __ieee754_logf calls
The recently added new math code inlines error handling instead of using
error handling wrappers around __ieee754* internal symbols, and thus the
__ieee754* symbols are no longer provided.
However __ieee754_expf and __ieee754_logf are used in the implementation
of a number of other math functions. These symbols are safe to redirect
to the external expf and logf symbols, because those names are always
reserved when single precision math functions are reserved and the
additional error handling code is either not reached or there will be
an error in the final result that will override an internal spurious
errno setting.
For consistency all of __ieee754_expf, __ieee754_logf and __ieee754_powf
are redirected using a macro.
---
newlib/libm/common/fdlibm.h | 11 +++++++++++
1 file changed, 11 insertions(+)
diff --git a/newlib/libm/common/fdlibm.h b/newlib/libm/common/fdlibm.h
index 821e4dedb..4523e8b2a 100644
--- a/newlib/libm/common/fdlibm.h
+++ b/newlib/libm/common/fdlibm.h
@@ -225,6 +225,17 @@ extern float __ieee754_scalbf __P((float,int));
extern float __ieee754_scalbf __P((float,float));
#endif
+#if !__OBSOLETE_MATH
+/* The new math code does not provide separate wrapper function
+ for error handling, so the extern symbol is called directly.
+ This is valid as long as there are no namespace issues (the
+ extern symbol is reserved whenever the caller is reserved)
+ and there are no observable error handling side effects. */
+# define __ieee754_expf(x) expf(x)
+# define __ieee754_logf(x) logf(x)
+# define __ieee754_powf(x,y) powf(x,y)
+#endif
+
/* float versions of fdlibm kernel functions */
extern float __kernel_sinf __P((float,float,int));
extern float __kernel_cosf __P((float,float));
--
2.11.0
^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [PATCH v2] New expf, exp2f, logf, log2f and powf implementations
2017-10-19 1:01 ` Szabolcs Nagy
@ 2017-10-23 17:16 ` Corinna Vinschen
0 siblings, 0 replies; 10+ messages in thread
From: Corinna Vinschen @ 2017-10-23 17:16 UTC (permalink / raw)
To: newlib
[-- Attachment #1: Type: text/plain, Size: 1140 bytes --]
On Oct 17 14:01, Szabolcs Nagy wrote:
> >From 23753d87bfcf048c5fa0a297d7cce48b733b696b Mon Sep 17 00:00:00 2001
> From: Szabolcs Nagy <szabolcs.nagy@arm.com>
> Date: Tue, 17 Oct 2017 12:41:20 +0100
> Subject: [PATCH] fix internal __ieee754_expf and __ieee754_logf calls
>
> The recently added new math code inlines error handling instead of using
> error handling wrappers around __ieee754* internal symbols, and thus the
> __ieee754* symbols are no longer provided.
>
> However __ieee754_expf and __ieee754_logf are used in the implementation
> of a number of other math functions. These symbols are safe to redirect
> to the external expf and logf symbols, because those names are always
> reserved when single precision math functions are reserved and the
> additional error handling code is either not reached or there will be
> an error in the final result that will override an internal spurious
> errno setting.
>
> For consistency all of __ieee754_expf, __ieee754_logf and __ieee754_powf
> are redirected using a macro.
Pushed.
Thanks,
Corinna
--
Corinna Vinschen
Cygwin Maintainer
Red Hat
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]
^ permalink raw reply [flat|nested] 10+ messages in thread
end of thread, other threads:[~2017-10-20 9:19 UTC | newest]
Thread overview: 10+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-09-05 8:39 [RFC][PATCH] New expf, exp2f, logf, log2f and powf implementations Wilco Dijkstra
2017-09-06 6:58 ` Corinna Vinschen
2017-09-29 18:45 ` [PATCH v2] " Szabolcs Nagy
2017-10-11 11:57 ` Szabolcs Nagy
2017-10-11 12:58 ` Corinna Vinschen
2017-10-13 1:31 ` Jeff Johnston
2017-10-13 9:15 ` Corinna Vinschen
2017-10-17 13:01 ` Szabolcs Nagy
2017-10-19 1:01 ` Szabolcs Nagy
2017-10-23 17:16 ` Corinna Vinschen
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).