glibc/sysdeps/i386/fpu
Patrick McGehearty 6fd0a3c6a8 Improve __ieee754_exp() performance by greater than 5x on sparc/x86.
These changes will be active for all platforms that don't provide
their own exp() routines. They will also be active for ieee754
versions of ccos, ccosh, cosh, csin, csinh, sinh, exp10, gamma, and
erf.

Typical performance gains is typically around 5x when measured on
Sparc s7 for common values between exp(1) and exp(40).

Using the glibc perf tests on sparc,
      sparc (nsec)    x86 (nsec)
      old     new     old     new
max   17629   395    5173     144
min     399    54      15      13
mean   5317   200    1349      23

The extreme max times for the old (ieee754) exp are due to the
multiprecision computation in the old algorithm when the true value is
very near 0.5 ulp away from an value representable in double
precision. The new algorithm does not take special measures for those
cases. The current glibc exp perf tests overrepresent those values.
Informal testing suggests approximately one in 200 cases might
invoke the high cost computation. The performance advantage of the new
algorithm for other values is still large but not as large as indicated
by the chart above.

Glibc correctness tests for exp() and expf() were run. Within the
test suite 3 input values were found to cause 1 bit differences (ulp)
when "FE_TONEAREST" rounding mode is set. No differences in exp() were
seen for the tested values for the other rounding modes.
Typical example:
exp(-0x1.760cd2p+0)  (-1.46113312244415283203125)
 new code:    2.31973271630014299393707e-01   0x1.db14cd799387ap-3
 old code:    2.31973271630014271638132e-01   0x1.db14cd7993879p-3
    exp    =  2.31973271630014285508337 (high precision)
Old delta: off by 0.49 ulp
New delta: off by 0.51 ulp

In addition, because ieee754_exp() is used by other routines, cexp()
showed test results with very small imaginary input values where the
imaginary portion of the result was off by 3 ulp when in upward
rounding mode, but not in the other rounding modes.  For x86, tgamma
showed a few values where the ulp increased to 6 (max ulp for tgamma
is 5). Sparc tgamma did not show these failures.  I presume the tgamma
differences are due to compiler optimization differences within the
gamma function.The gamma function is known to be difficult to compute
accurately.

	* sysdeps/ieee754/dbl-64/e_exp.c: Include <math-svid-compat.h> and
	<errno.h>.  Include "eexp.tbl".
	(half): New constant.
	(one): Likewise.
	(__ieee754_exp): Rewrite.
	(__slowexp): Remove prototype.
	* sysdeps/ieee754/dbl-64/eexp.tbl: New file.
	* sysdeps/ieee754/dbl-64/slowexp.c: Remove file.
	* sysdeps/i386/fpu/slowexp.c: Likewise.
	* sysdeps/ia64/fpu/slowexp.c: Likewise.
	* sysdeps/m68k/m680x0/fpu/slowexp.c: Likewise.
	* sysdeps/x86_64/fpu/multiarch/slowexp-avx.c: Likewise.
	* sysdeps/x86_64/fpu/multiarch/slowexp-fma.c: Likewise.
	* sysdeps/x86_64/fpu/multiarch/slowexp-fma4.c: Likewise.
	* sysdeps/generic/math_private.h (__slowexp): Remove prototype.
	* sysdeps/ieee754/dbl-64/e_pow.c: Remove mention of slowexp.c in
	comment.
	* sysdeps/powerpc/power4/fpu/Makefile [$(subdir) = math]
	(CPPFLAGS-slowexp.c): Remove variable.
	* sysdeps/x86_64/fpu/multiarch/Makefile (libm-sysdep_routines):
	Remove slowexp-fma, slowexp-fma4 and slowexp-avx.
	(CFLAGS-slowexp-fma.c): Remove variable.
	(CFLAGS-slowexp-fma4.c): Likewise.
	(CFLAGS-slowexp-avx.c): Likewise.
	* sysdeps/x86_64/fpu/multiarch/e_exp-avx.c (__slowexp): Do not
	define as macro.
	* sysdeps/x86_64/fpu/multiarch/e_exp-fma.c (__slowexp): Likewise.
	* sysdeps/x86_64/fpu/multiarch/e_exp-fma4.c (__slowexp): Likewise.
	* math/Makefile (type-double-routines): Remove slowexp.
	* manual/probes.texi (slowexp_p6): Remove.
	(slowexp_p32): Likewise.
2017-12-19 17:27:31 +00:00
..
doasin.c
e_acos.S
e_acosf.S
e_acosh.S Update copyright dates with scripts/update-copyrights. 2017-01-01 00:14:16 +00:00
e_acoshf.S Update copyright dates with scripts/update-copyrights. 2017-01-01 00:14:16 +00:00
e_acoshl.S Update copyright dates with scripts/update-copyrights. 2017-01-01 00:14:16 +00:00
e_acosl.c
e_asin.S Refactor i386 libm code forcing underflow exceptions. 2015-09-24 21:41:00 +00:00
e_asinf.S Refactor i386 libm code forcing underflow exceptions. 2015-09-24 21:41:00 +00:00
e_atan2.S Refactor i386 libm code forcing underflow exceptions. 2015-09-24 21:41:00 +00:00
e_atan2f.S Refactor i386 libm code forcing underflow exceptions. 2015-09-24 21:41:00 +00:00
e_atan2l.c
e_atanh.S Update copyright dates with scripts/update-copyrights. 2017-01-01 00:14:16 +00:00
e_atanhf.S Update copyright dates with scripts/update-copyrights. 2017-01-01 00:14:16 +00:00
e_atanhl.S Update copyright dates with scripts/update-copyrights. 2017-01-01 00:14:16 +00:00
e_exp2.S Avoid excess range in results from i386 exp, hypot, pow functions (bug 18980). 2015-09-18 21:53:22 +00:00
e_exp2l.S Refactor i386 libm code forcing underflow exceptions. 2015-09-24 21:41:00 +00:00
e_exp10.S Avoid excess range in results from i386 exp, hypot, pow functions (bug 18980). 2015-09-18 21:53:22 +00:00
e_exp10f.S Avoid excess range in results from i386 exp, hypot, pow functions (bug 18980). 2015-09-18 21:53:22 +00:00
e_exp10l.S
e_exp.S Avoid excess range in results from i386 exp, hypot, pow functions (bug 18980). 2015-09-18 21:53:22 +00:00
e_expl.S Use libm_alias_ldouble in sysdeps/i386/fpu. 2017-11-17 22:54:39 +00:00
e_fmod.S
e_fmodf.S
e_fmodl.c
e_hypot.S Update copyright dates with scripts/update-copyrights. 2017-01-01 00:14:16 +00:00
e_hypotf.S Update copyright dates with scripts/update-copyrights. 2017-01-01 00:14:16 +00:00
e_ilogb.S
e_ilogbf.S
e_ilogbl.S
e_log2.S
e_log2l.S Fix i386/x86_64 log2l (sNaN) (bug 20235). 2016-06-09 18:04:30 +00:00
e_log10.S
e_log10f.S
e_log10l.S Fix i386/x86_64 log10l (sNaN) (bug 20228). 2016-06-08 22:59:18 +00:00
e_log.S Fix i386/x86_64 log* (1) zero sign for -ffinite-math-only (bug 19213). 2015-11-05 21:56:31 +00:00
e_logl.S Fix i386/x86_64 logl (sNaN) (bug 20227). 2016-06-08 22:24:06 +00:00
e_pow.S Update copyright dates with scripts/update-copyrights. 2017-01-01 00:14:16 +00:00
e_powl.S Update copyright dates with scripts/update-copyrights. 2017-01-01 00:14:16 +00:00
e_rem_pio2.c
e_remainder.S
e_remainderf.S
e_remainderl.S
e_scalb.S Avoid excess range in results from i386 scalb functions (bug 18981). 2015-09-18 20:34:59 +00:00
e_scalbf.S Avoid excess range in results from i386 scalb functions (bug 18981). 2015-09-18 20:34:59 +00:00
e_scalbl.S Fix i386/x86_64 scalbl with sNaN input (bug 20296). 2016-06-23 22:17:41 +00:00
e_sqrt.S
e_sqrtf.S
e_sqrtl.c
fclrexcpt.c Check if SSE is available with HAS_CPU_FEATURE 2017-04-07 07:44:59 -07:00
fedisblxcpt.c Check if SSE is available with HAS_CPU_FEATURE 2017-04-07 07:44:59 -07:00
feenablxcpt.c Check if SSE is available with HAS_CPU_FEATURE 2017-04-07 07:44:59 -07:00
fegetenv.c Check if SSE is available with HAS_CPU_FEATURE 2017-04-07 07:44:59 -07:00
fegetexcept.c Update copyright dates with scripts/update-copyrights. 2017-01-01 00:14:16 +00:00
fegetmode.c Check if SSE is available with HAS_CPU_FEATURE 2017-04-07 07:44:59 -07:00
fegetround.c Update copyright dates with scripts/update-copyrights. 2017-01-01 00:14:16 +00:00
feholdexcpt.c Check if SSE is available with HAS_CPU_FEATURE 2017-04-07 07:44:59 -07:00
fenv_private.h Add float128 support for x86_64, x86. 2017-06-26 22:02:24 +00:00
fesetenv.c Check if SSE is available with HAS_CPU_FEATURE 2017-04-07 07:44:59 -07:00
fesetexcept.c Update copyright dates with scripts/update-copyrights. 2017-01-01 00:14:16 +00:00
fesetmode.c Check if SSE is available with HAS_CPU_FEATURE 2017-04-07 07:44:59 -07:00
fesetround.c Check if SSE is available with HAS_CPU_FEATURE 2017-04-07 07:44:59 -07:00
feupdateenv.c Check if SSE is available with HAS_CPU_FEATURE 2017-04-07 07:44:59 -07:00
fgetexcptflg.c Check if SSE is available with HAS_CPU_FEATURE 2017-04-07 07:44:59 -07:00
fraiseexcpt.c Update copyright dates with scripts/update-copyrights. 2017-01-01 00:14:16 +00:00
fsetexcptflg.c Check if SSE is available with HAS_CPU_FEATURE 2017-04-07 07:44:59 -07:00
ftestexcept.c Check if SSE is available with HAS_CPU_FEATURE 2017-04-07 07:44:59 -07:00
halfulp.c
i386-math-asm.h Update copyright dates with scripts/update-copyrights. 2017-01-01 00:14:16 +00:00
Implies
libm-test-ulps Add _Float32 function aliases. 2017-12-07 00:48:31 +00:00
libm-test-ulps-name Do not hardcode platform names in manual/libm-err-tab.pl (bug 14139). 2016-11-04 16:49:06 +00:00
math_private.h Fix math_private.h multiple include guards. 2015-11-20 23:46:23 +00:00
math-tests.h Update copyright dates with scripts/update-copyrights. 2017-01-01 00:14:16 +00:00
mpatan2.c
mpatan.c
mpexp.c
mplog.c
mpsqrt.c
s_asinh.S Use libm_alias_double for i386. 2017-11-28 18:15:44 +00:00
s_asinhf.S Use libm_alias_float for i386. 2017-11-29 00:10:35 +00:00
s_asinhl.S Use libm_alias_ldouble in sysdeps/i386/fpu. 2017-11-17 22:54:39 +00:00
s_atan.S Use libm_alias_double for i386. 2017-11-28 18:15:44 +00:00
s_atanf.S Use libm_alias_float for i386. 2017-11-29 00:10:35 +00:00
s_atanl.c Use libm_alias_ldouble in sysdeps/i386/fpu. 2017-11-17 22:54:39 +00:00
s_cbrt.S Use libm_alias_double for i386. 2017-11-28 18:15:44 +00:00
s_cbrtf.S Use libm_alias_float for i386. 2017-11-29 00:10:35 +00:00
s_cbrtl.S Use libm_alias_ldouble in sysdeps/i386/fpu. 2017-11-17 22:54:39 +00:00
s_ceil.S Use libm_alias_double for i386. 2017-11-28 18:15:44 +00:00
s_ceilf.S Use libm_alias_float for i386. 2017-11-29 00:10:35 +00:00
s_ceill.S Use libm_alias_ldouble in sysdeps/i386/fpu. 2017-11-17 22:54:39 +00:00
s_copysign.S Use libm_alias_double for i386. 2017-11-28 18:15:44 +00:00
s_copysignf.S Use libm_alias_float for i386. 2017-11-29 00:10:35 +00:00
s_copysignl.S Use libm_alias_ldouble in sysdeps/i386/fpu. 2017-11-17 22:54:39 +00:00
s_expm1.S Use libm_alias_double for i386. 2017-11-28 18:15:44 +00:00
s_expm1f.S Use libm_alias_float for i386. 2017-11-29 00:10:35 +00:00
s_expm1l.S
s_fabs.S Use libm_alias_double for i386. 2017-11-28 18:15:44 +00:00
s_fabsf.S Use libm_alias_float for i386. 2017-11-29 00:10:35 +00:00
s_fabsl.S Use libm_alias_ldouble in sysdeps/i386/fpu. 2017-11-17 22:54:39 +00:00
s_fdim.c Use libm_alias_double for i386. 2017-11-28 18:15:44 +00:00
s_finite.S
s_finitef.S
s_finitel.S
s_floor.S Use libm_alias_double for i386. 2017-11-28 18:15:44 +00:00
s_floorf.S Use libm_alias_float for i386. 2017-11-29 00:10:35 +00:00
s_floorl.S Use libm_alias_ldouble in sysdeps/i386/fpu. 2017-11-17 22:54:39 +00:00
s_fmax.S Use libm_alias_double for i386. 2017-11-28 18:15:44 +00:00
s_fmaxf.S Use libm_alias_float for i386. 2017-11-29 00:10:35 +00:00
s_fmaxl.S Use libm_alias_ldouble in sysdeps/i386/fpu. 2017-11-17 22:54:39 +00:00
s_fmin.S Use libm_alias_double for i386. 2017-11-28 18:15:44 +00:00
s_fminf.S Use libm_alias_float for i386. 2017-11-29 00:10:35 +00:00
s_fminl.S Use libm_alias_ldouble in sysdeps/i386/fpu. 2017-11-17 22:54:39 +00:00
s_fpclassifyl.c Consistently use uintN_t not u_intN_t in libm. 2017-08-03 19:55:04 +00:00
s_frexp.S Use libm_alias_double for i386. 2017-11-28 18:15:44 +00:00
s_frexpf.S Use libm_alias_float for i386. 2017-11-29 00:10:35 +00:00
s_frexpl.S Use libm_alias_ldouble in sysdeps/i386/fpu. 2017-11-17 22:54:39 +00:00
s_isinfl.c
s_isnanl.c Consistently use uintN_t not u_intN_t in libm. 2017-08-03 19:55:04 +00:00
s_llrint.S Use libm_alias_double for i386. 2017-11-28 18:15:44 +00:00
s_llrintf.S Use libm_alias_float for i386. 2017-11-29 00:10:35 +00:00
s_llrintl.S Use libm_alias_ldouble in sysdeps/i386/fpu. 2017-11-17 22:54:39 +00:00
s_log1p.S Refactor i386 libm code forcing underflow exceptions. 2015-09-24 21:41:00 +00:00
s_log1pf.S Refactor i386 libm code forcing underflow exceptions. 2015-09-24 21:41:00 +00:00
s_log1pl.S Fix i386/x86_64 log1pl (sNaN) (bug 20229). 2016-06-08 23:11:42 +00:00
s_logb.S Use libm_alias_double for i386. 2017-11-28 18:15:44 +00:00
s_logbf.S Use libm_alias_float for i386. 2017-11-29 00:10:35 +00:00
s_logbl.c Use libm_alias_ldouble in sysdeps/i386/fpu. 2017-11-17 22:54:39 +00:00
s_lrint.S Use libm_alias_double for i386. 2017-11-28 18:15:44 +00:00
s_lrintf.S Use libm_alias_float for i386. 2017-11-29 00:10:35 +00:00
s_lrintl.S Use libm_alias_ldouble in sysdeps/i386/fpu. 2017-11-17 22:54:39 +00:00
s_nearbyint.S Use libm_alias_double for i386. 2017-11-28 18:15:44 +00:00
s_nearbyintf.S Use libm_alias_float for i386. 2017-11-29 00:10:35 +00:00
s_nearbyintl.S Use libm_alias_ldouble in sysdeps/i386/fpu. 2017-11-17 22:54:39 +00:00
s_nextafterl.c Use libm_alias_ldouble in sysdeps/i386/fpu. 2017-11-17 22:54:39 +00:00
s_nexttoward.c Consistently use uintN_t not u_intN_t in libm. 2017-08-03 19:55:04 +00:00
s_nexttowardf.c Consistently use uintN_t not u_intN_t in libm. 2017-08-03 19:55:04 +00:00
s_remquo.S Use libm_alias_double for i386. 2017-11-28 18:15:44 +00:00
s_remquof.S Use libm_alias_float for i386. 2017-11-29 00:10:35 +00:00
s_remquol.S Use libm_alias_ldouble in sysdeps/i386/fpu. 2017-11-17 22:54:39 +00:00
s_rint.S Use libm_alias_double for i386. 2017-11-28 18:15:44 +00:00
s_rintf.S Use libm_alias_float for i386. 2017-11-29 00:10:35 +00:00
s_rintl.c Use libm_alias_ldouble in sysdeps/i386/fpu. 2017-11-17 22:54:39 +00:00
s_scalbln.c
s_scalblnf.c
s_scalblnl.c
s_scalbn.S Avoid excess range in results from i386 scalb functions (bug 18981). 2015-09-18 20:34:59 +00:00
s_scalbnf.S Avoid excess range in results from i386 scalb functions (bug 18981). 2015-09-18 20:34:59 +00:00
s_scalbnl.S Make scalbn set errno (bug 6803). 2015-09-16 21:11:00 +00:00
s_significand.S
s_significandf.S
s_significandl.c
s_trunc.S Use libm_alias_double for i386. 2017-11-28 18:15:44 +00:00
s_truncf.S Use libm_alias_float for i386. 2017-11-29 00:10:35 +00:00
s_truncl.S Use libm_alias_ldouble in sysdeps/i386/fpu. 2017-11-17 22:54:39 +00:00
slowpow.c
t_exp.c
Versions
w_sqrt_compat.c Move wrappers to libm-compat-calls-auto 2017-01-04 16:25:04 -02:00
w_sqrt.c Prefer new libm function wrappers for !LIBM_SVID_COMPAT. 2017-09-05 23:35:55 +00:00