The new implementations are provided under !__OBSOLETE_MATH, it uses
ISO C99 code. With default settings the worst case error in nearest
rounding mode is 0.519 ULP with inlined fma and fma contraction. It uses
a 2 KB lookup table, on aarch64 .text+.rodata size of libm.a is increased
by 1703 bytes. The w_log.c wrapper is disabled since error handling is
inline in the new code.
New __HAVE_FAST_FMA and __HAVE_FAST_FMA_DEFAULT feature macros were
added to enable selecting between the code path that uses fma and the
one that does not. Targets supposed to set __HAVE_FAST_FMA_DEFAULT
if they have single instruction fma and the compiler can actually
inline it (gcc has __FP_FAST_FMA macro but that does not guarantee
inlining with -fno-builtin-fma).
Improvements on Cortex-A72:
latency: 1.9x
thruput: 2.3x
* libm/mathfp/s_logarithm.c: Fix error introduced by previous
fix when handling negative input values. Make function
consistent with math directory and glibc version such that
inf and nan values return inf and nan respectively with no
errno setting.
* libm/mathfp/sf_logarithm.c: Ditto.
* libm/math/w_log.c: Set errno to ERANGE when input is 0.0.
* libm/math/wf_log.c: Ditto.
* libm/math/w_log10.c: Ditto.
* libm/math/wf_log10.c: Ditto.
* libm/math/e_pow.c: Fix to be consistent with glibc with regards
to treatment of NaN and +-inf arguments.
* libm/math/ef_pow.c: Ditto.
* libm/math/w_pow.c: Ditto.
* libm/math/wf_pow.c: Ditto.
* libm/math/w_acos.c: Fix domain errors to return NaN.
* libm/math/w_asin.c: Ditto.
* libm/math/wf_acos.c: Ditto.
* libm/math/wf_asin.c: Ditto.
* libm/math/w_log.c: Fix to return NaN for negative number inputs.
* libm/math/wf_log.c: Ditto.
* libm/math/wf_log10.c: Ditto.
* libm/math/w_log10.c: Ditto.
Szabolcs Nagy