In non-default rounding modes, tgamma can be slightly less accurate
than permitted by glibc's accuracy goals.
Part of the problem is error accumulation, addressed in this patch by
setting round-to-nearest for internal computations. However, there
was also a bug in the code dealing with computing pow (x + n, x + n)
where x + n is not exactly representable, providing another source of
error even in round-to-nearest mode; it was necessary to address both
bugs to get errors for all testcases within glibc's accuracy goals.
Given this second fix, accuracy in round-to-nearest mode is also
improved (hence regeneration of ulps for tgamma should be from scratch
- truncate libm-test-ulps or at least remove existing tgamma entries -
so that the expected ulps can be reduced).
Some additional complications also arose. Certain tgamma tests should
strictly, according to IEEE semantics, overflow or not depending on
the rounding mode; this is beyond the scope of glibc's accuracy goals
for any function without exactly-determined results, but
gen-auto-libm-tests doesn't handle being lax there as it does for
underflow. (libm-test.inc also doesn't handle being lax about whether
the result in cases very close to the overflow threshold is infinity
or a finite value close to overflow, but that doesn't cause problems
in this case though I've seen it cause problems with random test
generation for some functions.) Thus, spurious-overflow markings,
with a comment, are added to auto-libm-test-in (no bug in Bugzilla
because the issue is with the testsuite, not a user-visible bug in
glibc). And on x86, after the patch I saw ERANGE issues as previously
reported by Carlos (see my commentary in
<https://sourceware.org/ml/libc-alpha/2015-01/msg00485.html>), which
needed addressing by ensuring excess range and precision were
eliminated at various points if FLT_EVAL_METHOD != 0.
I also noticed and fixed a cosmetic issue where 1.0f was used in long
double functions and should have been 1.0L.
This completes the move of all functions to testing in all rounding
modes with ALL_RM_TEST, so gen-libm-have-vector-test.sh is updated to
remove the workaround for some functions not using ALL_RM_TEST.
Tested for x86_64, x86, mips64 and powerpc.
[BZ #18613]
* sysdeps/ieee754/dbl-64/e_gamma_r.c (gamma_positive): Take log of
X_ADJ not X when adjusting exponent.
(__ieee754_gamma_r): Do intermediate computations in
round-to-nearest then adjust overflowing and underflowing results
as needed.
* sysdeps/ieee754/flt-32/e_gammaf_r.c (gammaf_positive): Take log
of X_ADJ not X when adjusting exponent.
(__ieee754_gammaf_r): Do intermediate computations in
round-to-nearest then adjust overflowing and underflowing results
as needed.
* sysdeps/ieee754/ldbl-128/e_gammal_r.c (gammal_positive): Take
log of X_ADJ not X when adjusting exponent.
(__ieee754_gammal_r): Do intermediate computations in
round-to-nearest then adjust overflowing and underflowing results
as needed. Use 1.0L not 1.0f as numerator of division.
* sysdeps/ieee754/ldbl-128ibm/e_gammal_r.c (gammal_positive): Take
log of X_ADJ not X when adjusting exponent.
(__ieee754_gammal_r): Do intermediate computations in
round-to-nearest then adjust overflowing and underflowing results
as needed. Use 1.0L not 1.0f as numerator of division.
* sysdeps/ieee754/ldbl-96/e_gammal_r.c (gammal_positive): Take log
of X_ADJ not X when adjusting exponent.
(__ieee754_gammal_r): Do intermediate computations in
round-to-nearest then adjust overflowing and underflowing results
as needed. Use 1.0L not 1.0f as numerator of division.
* math/libm-test.inc (tgamma_test_data): Remove one test. Moved
to auto-libm-test-in.
(tgamma_test): Use ALL_RM_TEST.
* math/auto-libm-test-in: Add one test of tgamma. Mark some other
tests of tgamma with spurious-overflow.
* math/auto-libm-test-out: Regenerated.
* math/gen-libm-have-vector-test.sh: Do not check for START.
* sysdeps/i386/fpu/libm-test-ulps: Update.
* sysdeps/x86_64/fpu/libm-test-ulps: Likewise.
The ldbl-128 implementation of j1l produces spurious underflow
exceptions for some small arguments, as a result of squaring the
argument. This patch fixes it just to use a linear approximation for
sufficiently small arguments, and then to force an underflow exception
only in the cases where it is required.
Tested for mips64.
[BZ #18612]
* sysdeps/ieee754/ldbl-128/e_j1l.c (__ieee754_j1l): For small
arguments, just return 0.5 times the argument, with underflow
forced as needed.
* math/auto-libm-test-in: Add more tests of j1.
* math/auto-libm-test-out: Regenerated.
Similar to various other bugs in this area, j1 and jn implementations
can fail to raise the underflow exception when the internal
computation is exact although the actual function is inexact. This
patch forces the exception in a similar way to other such fixes. (The
ldbl-128 / ldbl-128ibm j1l implementation is different and doesn't
need a change for this until spurious underflows in it are fixed.)
Tested for x86_64, x86, mips64 and powerpc.
[BZ #16559]
* sysdeps/ieee754/dbl-64/e_j1.c: Include <float.h>.
(__ieee754_j1): Force underflow exception for small results.
* sysdeps/ieee754/dbl-64/e_jn.c (__ieee754_jn): Likewise.
* sysdeps/ieee754/flt-32/e_j1f.c: Include <float.h>.
(__ieee754_j1f): Force underflow exception for small results.
* sysdeps/ieee754/flt-32/e_jnf.c (__ieee754_jnf): Likewise.
* sysdeps/ieee754/ldbl-128/e_jnl.c (__ieee754_jnl): Likewise.
* sysdeps/ieee754/ldbl-128ibm/e_jnl.c (__ieee754_jnl): Likewise.
* sysdeps/ieee754/ldbl-96/e_j1l.c: Include <float.h>.
(__ieee754_j1l): Force underflow exception for small results.
* sysdeps/ieee754/ldbl-96/e_jnl.c (__ieee754_jnl): Likewise.
* math/auto-libm-test-in: Add more tests of j1 and jn.
* math/auto-libm-test-out: Regenerated.
Some existing jn tests, if run in non-default rounding modes, produce
errors above those accepted in glibc, which causes problems for moving
tests of jn to use ALL_RM_TEST. This patch makes jn set rounding
to-nearest internally, as was done for yn some time ago, then computes
the appropriate underflowing value for results that underflowed to
zero in to-nearest, and moves the tests to ALL_RM_TEST. It does
nothing about the general inaccuracy of Bessel function
implementations in glibc, though it should make jn more accurate on
average in non-default rounding modes through reduced error
accumulation. The recomputation of results that underflowed to zero
should as a side-effect fix some cases of bug 16559, where jn just
used an exact zero, but that is *not* the goal of this patch and other
cases of that bug remain unfixed.
(Most of the changes in the patch are reindentation to add new scopes
for SET_RESTORE_ROUND*.)
Tested for x86_64, x86, powerpc and mips64.
[BZ #16559]
[BZ #18602]
* sysdeps/ieee754/dbl-64/e_jn.c (__ieee754_jn): Set
round-to-nearest internally then recompute results that
underflowed to zero in the original rounding mode.
* sysdeps/ieee754/flt-32/e_jnf.c (__ieee754_jnf): Likewise.
* sysdeps/ieee754/ldbl-128/e_jnl.c (__ieee754_jnl): Likewise.
* sysdeps/ieee754/ldbl-128ibm/e_jnl.c (__ieee754_jnl): Likewise.
* sysdeps/ieee754/ldbl-96/e_jnl.c (__ieee754_jnl): Likewise
* math/libm-test.inc (jn_test): Use ALL_RM_TEST.
* sysdeps/i386/fpu/libm-test-ulps: Update.
* sysdeps/x86_64/fpu/libm-test-ulps: Likewise.
cexp, ccos, ccosh, csin and csinh have spurious underflows in cases
where they compute sin of the smallest normal, that produces an
underflow exception (depending on which sin implementation is in use)
but the final result does not underflow. ctan and ctanh may also have
such underflows, or they may be latent (the issue there is that
e.g. ctan (DBL_MIN) should, rounded upwards, be the next double value
above DBL_MIN, which under glibc's accuracy goals may not have an
underflow exception, but the intermediate computation of sin (DBL_MIN)
would legitimately underflow on before-rounding architectures).
This patch fixes all those functions so they use plain comparisons (>
DBL_MIN etc.) instead of comparing the result of fpclassify with
FP_SUBNORMAL (in all these cases, we already know the number being
compared is finite). Note that in the case of csin / csinf / csinl,
there is no need for fabs calls in the comparison because the real
part has already been reduced to its absolute value.
As the patch fixes the failures that previously obstructed moving
tests of cexp to use ALL_RM_TEST, those tests are moved to ALL_RM_TEST
by the patch (two functions remain yet to be converted).
Tested for x86_64 and x86 and ulps updated accordingly.
[BZ #18594]
* math/s_ccosh.c (__ccosh): Compare with least normal value
instead of comparing class with FP_SUBNORMAL.
* math/s_ccoshf.c (__ccoshf): Likewise.
* math/s_ccoshl.c (__ccoshl): Likewise.
* math/s_cexp.c (__cexp): Likewise.
* math/s_cexpf.c (__cexpf): Likewise.
* math/s_cexpl.c (__cexpl): Likewise.
* math/s_csin.c (__csin): Likewise.
* math/s_csinf.c (__csinf): Likewise.
* math/s_csinh.c (__csinh): Likewise.
* math/s_csinhf.c (__csinhf): Likewise.
* math/s_csinhl.c (__csinhl): Likewise.
* math/s_csinl.c (__csinl): Likewise.
* math/s_ctan.c (__ctan): Likewise.
* math/s_ctanf.c (__ctanf): Likewise.
* math/s_ctanh.c (__ctanh): Likewise.
* math/s_ctanhf.c (__ctanhf): Likewise.
* math/s_ctanhl.c (__ctanhl): Likewise.
* math/s_ctanl.c (__ctanl): Likewise.
* math/auto-libm-test-in: Add more tests of ccos, ccosh, cexp,
csin, csinh, ctan and ctanh.
* math/auto-libm-test-out: Regenerated.
* math/libm-test.inc (cexp_test): Use ALL_RM_TEST.
* sysdeps/i386/fpu/libm-test-ulps: Update.
* sysdeps/x86_64/fpu/libm-test-ulps: Likewise.
This patch moves most tests of csin and csinh with finite inputs from
libm-test.inc to auto-libm-test-in. The remaining two tests of each
function with small arguments are not moved because moving them causes
the time required by gen-auto-libm-tests to go up from under 8 seconds
to over 11 minutes for me. (The current development version of MPC
has had speed improvements for mpc_sin for some time, but there hasn't
been a release containing those improvements yet.)
Tested for x86_64 and x86.
* math/auto-libm-test-in: Add more tests of csin and csinh.
* math/auto-libm-test-out: Regenerated.
* math/libm-test.inc (csin_test_data): Remove tests moved to
auto-libm-test-in.
(csinh_test_data): Likewise.
csin and csinh can produce bad results when overflowing in directed
rounding modes, because a multiplication that can overflow is followed
by a possible negation. This patch fixes this by negating one of the
arguments of the multiplication before the multiplication instead of
negating the result.
The new tests for this issue are added to auto-libm-test-in, starting
use of that file for csin and csinh. The issue was found in the
course of moving existing tests for csin and csinh (existing tests, by
being enabled in more cases than previously, showed the issue for
float and double but not for long double); that move will now be done
separately.
Tested for x86_64 and x86 and ulps updated accordingly.
[BZ #18593]
* math/s_csin.c (__csin): Negate before rather than after possibly
overflowing multiplication.
* math/s_csinf.c (__csinf): Likewise.
* math/s_csinh.c (__csinh): Likewise.
* math/s_csinhf.c (__csinhf): Likewise.
* math/s_csinhl.c (__csinhl): Likewise.
* math/s_csinl.c (__csinl): Likewise.
* math/auto-libm-test-in: Add some tests of csin and csinh.
* math/auto-libm-test-out: Regenerated.
* math/libm-test.inc (csin_test_data): Use AUTO_TESTS_c_c.
(csinh_test_data): Likewise.
* sysdeps/x86_64/fpu/libm-test-ulps: Update.
Similar to various other bugs in this area, some sin and sincos
implementations do not raise the underflow exception for subnormal
arguments, when the result is tiny and inexact. This patch forces the
exception in a similar way to previous fixes.
Tested for x86_64, x86, mips64 and powerpc.
[BZ #16526]
[BZ #16538]
* sysdeps/ieee754/dbl-64/s_sin.c: Include <float.h>.
(__sin): Force underflow exception for arguments with small
absolute value.
* sysdeps/ieee754/flt-32/k_sinf.c: Include <float.h>.
(__kernel_sinf): Force underflow exception for arguments with
small absolute value.
* sysdeps/ieee754/ldbl-128/k_sincosl.c: Include <float.h>.
(__kernel_sincosl): Force underflow exception for arguments with
small absolute value.
* sysdeps/ieee754/ldbl-128/k_sinl.c: Include <float.h>.
(__kernel_sinl): Force underflow exception for arguments with
small absolute value.
* sysdeps/ieee754/ldbl-128ibm/k_sincosl.c: Include <float.h>.
(__kernel_sincosl): Force underflow exception for arguments with
small absolute value.
* sysdeps/ieee754/ldbl-128ibm/k_sinl.c: Include <float.h>.
(__kernel_sinl): Force underflow exception for arguments with
small absolute value.
* sysdeps/ieee754/ldbl-96/k_sinl.c: Include <float.h>.
(__kernel_sinl): Force underflow exception for arguments with
small absolute value.
* sysdeps/powerpc/fpu/k_sinf.c: Include <float.h>.
(__kernel_sinf): Force underflow exception for arguments with
small absolute value.
* math/auto-libm-test-in: Add more tests of sin and sincos.
* math/auto-libm-test-out: Regenerated.
__kernel_standard_l converts long double arguments to double for use
in SVID "struct exception". This has special-case handling for when
that conversion would overflow or underflow but the original long
double function wouldn't. However, it turns out that "inexact"
exceptions can be spurious here as well, when the function is exactly
determined and __kernel_standard_l is being called for a domain error.
This patch fixes this by using feholdexcept / fesetenv to avoid
exceptions from the conversion, replacing the previous special-case
logic for overflow and underflow (this covers all functions using
__kernel_standard_l, not just those that actually need a change, since
there doesn't seem to be much point in restricting things just to the
functions that mustn't get "inexact" here).
Tested for x86_64 and x86.
[BZ #18245]
[BZ #18583]
* sysdeps/ieee754/k_standardl.c: Include <fenv.h>.
(__kernel_standard_l): Use feholdexcept and fesetenv around
conversion to double instead of special-casing overflow and
underflow.
* math/libm-test.inc (fmod_test_data): Add more tests.
(remainder_test_data): Likewise.
(sqrt_test_data): Likewise.
Since the libmvec tests went in I've noticed build failures from
parallel testing in math/, when those tests start building before
libm-test.c has been fully generated. (This only applies if libm test
sources have been modified after the original glibc build, because
otherwise libm-test.stmp was generated during the original build and
doesn't get regenerated during testing.)
Those tests depend on libm-test.stmp, but the dependency uses
$(libmvec-tests), which is set in the sysdeps Makefile fragments, and
appears before the inclusion of ../Rules, which is what includes those
fragments; thus, the dependency does not work and parallel make can
start building the vector tests too soon. This patch moves the
dependency further down so that the required variable is defined when
the dependency is.
Tested for x86_64.
* math/Makefile [$(PERL) != no]
($(addprefix $(objpfx), $(addsuffix .o, $(libm-vec-tests)))): Move
dependency on libm-test.stmp below the inclusion of Rules.
The csqrt implementations in glibc can cause spurious underflows in
some cases as a side-effect of the scaling for large arguments (when
underflow is correct for the square root of the argument that was
scaled down to avoid overflow, but not for the original argument).
This patch arranges to avoid the underflowing intermediate computation
(eliminating a multiplication in 0.5 in the problem cases where a
subsequent scaling by 2 would follow).
Tested for x86_64 and x86 and ulps updated accordingly (only needed
for x86).
[BZ #18371]
* math/s_csqrt.c (__csqrt): Avoid multiplication by 0.5 where
intermediate but not final result might underflow.
* math/s_csqrtf.c (__csqrtf): Likewise.
* math/s_csqrtl.c (__csqrtl): Likewise.
* math/auto-libm-test-in: Add more tests of csqrt.
* math/auto-libm-test-out: Regenerated.
* sysdeps/i386/fpu/libm-test-ulps: Update.
The dbl-64 and flt-32 implementations of exp2 functions produce
spurious underflow exceptions. The underlying reason is the same in
both cases: the computation works as (2^a - 1)*2^b + 2^b for suitably
chosen a and b, where a has small magnitude so 2^a - 1 can be computed
with a low-degree polynomial approximation, and (2^a - 1)*2^b can
underflow even when the final result does not. This patch fixes this
by adjusting the threshold for when scaling is used to avoid
intermediate underflow so it works for any possible value of a where
the final result would not underflow.
Tested for x86_64 and x86.
[BZ #18219]
* sysdeps/ieee754/dbl-64/e_exp2.c (__ieee754_exp2): Reduce
threshold on absolute value of exponent for which scaling is used.
* sysdeps/ieee754/flt-32/e_exp2f.c (__ieee754_exp2f): Likewise.
* math/auto-libm-test-in: Add more tests of exp2.
* math/auto-libm-test-out: Regenerated.
Similar to various other bugs in this area, some expm1 implementations
do not raise the underflow exception for subnormal arguments, when the
result is tiny and inexact. This patch forces the exception in a
similar way to previous fixes.
(The issue does not apply to the ldbl-* implementations or to those
for x86 / x86_64 long double. The change to
sysdeps/ieee754/dbl-64/wordsize-64/e_cosh.c is one I missed when
previously fixing bug 16354; the bug in that implementation was
previously latent, but the expm1 fixes stopped it being latent and so
required it to be fixed to avoid spurious underflows from cosh.)
Tested for x86_64 and x86.
[BZ #16353]
* sysdeps/i386/fpu/s_expm1.S (dbl_min): New object.
(__expm1): Force underflow exception for arguments with small
absolute value.
* sysdeps/i386/fpu/s_expm1f.S (flt_min): New object.
(__expm1f): Force underflow exception for arguments with small
absolute value.
* sysdeps/ieee754/dbl-64/s_expm1.c: Include <float.h>.
(__expm1): Force underflow exception for arguments with small
absolute value.
* sysdeps/ieee754/flt-32/s_expm1f.c: Include <float.h>.
(__expm1f): Force underflow exception for arguments with small
absolute value.
* sysdeps/ieee754/dbl-64/wordsize-64/e_cosh.c (__ieee754_cosh):
Check for small arguments before calling __expm1.
* math/auto-libm-test-in: Do not mark underflow exceptions as
possibly missing for bug 16353.
* math/auto-libm-test-out: Regenerated.
In the x86 / x86_64 implementations of expm1l, when expm1l's result
should underflow to 0 (argument minus the least subnormal, in some
rounding modes), it can be a zero of the wrong sign. This patch fixes
this by returning the argument with underflow forced in that case
(this is a 1ulp error relative to the correctly rounded result of -0,
which is OK in terms of the documented accuracy goals, whereas a
result with the wrong sign never is).
Tested for x86_64 and x86.
[BZ #18569]
* sysdeps/i386/fpu/e_expl.S (IEEE754_EXPL) [USE_AS_EXPM1L]: Force
underflow and return argument in case of subnormal argument.
* sysdeps/x86_64/fpu/e_expl.S (IEEE754_EXPL) [USE_AS_EXPM1L]:
Likewise.
* math/auto-libm-test-in: Add more tests of expm1.
* math/auto-libm-test-out: Regenerated.
Similar to various other bugs in this area, the x86 and x86_64
implementations of expl / exp10l can fail to produce underflow
exceptions when the unscaled result has trailing 0 bits so the scaling
down to subnormal precision is exact. This patch fixes this by
forcing the exception in the case of tiny results.
Tested for x86_64 and x86.
[BZ #16361]
* sysdeps/i386/fpu/e_expl.S [!USE_AS_EXPM1L] (cmin): New object.
[!USE_AS_EXPM1L] (IEEE754_EXPL): Force underflow exception for
tiny results.
* sysdeps/x86_64/fpu/e_expl.S [!USE_AS_EXPM1L] (cmin): New object.
[!USE_AS_EXPM1L] (IEEE754_EXPL): Force underflow exception for
tiny results.
* math/auto-libm-test-in: Add more tests of exp and exp10. Do not
mark underflow exceptions as possibly missing for bug 16361.
* math/auto-libm-test-out: Regenerated.
Similar to various other bugs in this area, some asinh implementations
do not raise the underflow exception for subnormal arguments, when the
result is tiny and inexact. This patch forces the exception in a
similar way to previous fixes.
Tested for x86_64, x86 and mips64.
[BZ #16350]
* sysdeps/i386/fpu/s_asinh.S (__asinh): Force underflow exception
for arguments with small absolute value.
* sysdeps/i386/fpu/s_asinhf.S (__asinhf): Likewise.
* sysdeps/i386/fpu/s_asinhl.S (__asinhl): Likewise.
* sysdeps/ieee754/dbl-64/s_asinh.c: Include <float.h>.
(__asinh): Force underflow exception for arguments with small
absolute value.
* sysdeps/ieee754/flt-32/s_asinhf.c: Include <float.h>.
(__asinhf): Force underflow exception for arguments with small
absolute value.
* sysdeps/ieee754/ldbl-128/s_asinhl.c: Include <float.h>.
(__asinhl): Force underflow exception for arguments with small
absolute value.
* sysdeps/ieee754/ldbl-128ibm/s_asinhl.c: Include <float.h>.
(__asinhl): Force underflow exception for arguments with small
absolute value.
* sysdeps/ieee754/ldbl-96/s_asinhl.c: Include <float.h>.
(__asinhl): Force underflow exception for arguments with small
absolute value.
* math/auto-libm-test-in: Do not mark underflow exceptions as
possibly missing for bug 16350.
* math/auto-libm-test-out: Regenerated.
I noticed that I'd left a spurious-underflow allowance behind in
auto-libm-test-in for a bug that was fixed some time ago. This patch
removes it.
Tested for x86_64 and x86.
* math/auto-libm-test-in: Remove spurious underflow allowance for
tests of cexp.
* math/auto-libm-test-out: Regenerated.
sysdeps/ieee754/ldbl-128ibm has its own versions of cprojl, ctanhl and
ctanl.
Having its own versions, where otherwise the math/ copies are
generally used for all floating-point formats, means they are liable
to get out of sync and not benefit from bug fixes to the generic
versions. The substantive differences (not arising from getting out
of sync and slightly different fixes for the same issues) are: long
double compat handling (also done in the ldbl-opt versions, so doesn't
require special versions for ldbl-128ibm); handling of LDBL_EPSILON
(conditionally undefined and redefined in other math/ implementations,
so doesn't justify a special version), and:
/* __gcc_qmul does not respect -0.0 so we need the following fixup. */
if ((__real__ res == 0.0L) && (__real__ x == 0.0L))
__real__ res = __real__ x;
if ((__real__ res == 0.0L) && (__imag__ x == 0.0L))
__imag__ res = __imag__ x;
But if that statement about __gcc_qmul was ever true for an old
version of that libgcc function, it's not the case for any GCC version
now supported to build glibc; there's explicit logic early in that
function (and similarly in __gcc_qdiv) to return an appropriately
signed zero if the product of the high parts is zero. So this patch
adds the special LDBL_EPSILON handling to the generic functions and
removes the ldbl-128ibm versions.
Tested for powerpc32 (compared test-ldouble.out before and after the
changes; there are slight changes to results for ctanl / ctanhl,
arising from divergence of the implementations, but nothing that
affects the overall nature of the issues shown by the testsuite, and
in particular nothing related to signs of zero resutls).
* math/s_ctanhl.c [LDBL_MANT_DIG == 106] (LDBL_EPSILON): Undefine
and redefine.
* math/s_ctanl.c [LDBL_MANT_DIG == 106] (LDBL_EPSILON): Undefine
and redefine.
* sysdeps/ieee754/ldbl-128ibm/s_cprojl.c: Remove file.
* sysdeps/ieee754/ldbl-128ibm/s_ctanhl.c: Likewise.
* sysdeps/ieee754/ldbl-128ibm/s_ctanl.c: Likewise.
Here is implementation of vectorized sin containing SSE, AVX,
AVX2 and AVX512 versions according to Vector ABI
<https://groups.google.com/forum/#!topic/x86-64-abi/LmppCfN1rZ4>.
* bits/libm-simd-decl-stubs.h: Added stubs for sin.
* math/bits/mathcalls.h: Added sin declaration with __MATHCALL_VEC.
* sysdeps/unix/sysv/linux/x86_64/libmvec.abilist: New versions added.
* sysdeps/x86/fpu/bits/math-vector.h: SIMD declaration for sin.
* sysdeps/x86_64/fpu/Makefile (libmvec-support): Added new files.
* sysdeps/x86_64/fpu/Versions: New versions added.
* sysdeps/x86_64/fpu/libm-test-ulps: Regenerated.
* sysdeps/x86_64/fpu/multiarch/Makefile (libmvec-sysdep_routines): Added
build of SSE, AVX2 and AVX512 IFUNC versions.
* sysdeps/x86_64/fpu/multiarch/svml_d_sin2_core.S: New file.
* sysdeps/x86_64/fpu/multiarch/svml_d_sin2_core_sse4.S: New file.
* sysdeps/x86_64/fpu/multiarch/svml_d_sin4_core.S: New file.
* sysdeps/x86_64/fpu/multiarch/svml_d_sin4_core_avx2.S: New file.
* sysdeps/x86_64/fpu/multiarch/svml_d_sin8_core.S: New file.
* sysdeps/x86_64/fpu/multiarch/svml_d_sin8_core_avx512.S: New file.
* sysdeps/x86_64/fpu/svml_d_sin2_core.S: New file.
* sysdeps/x86_64/fpu/svml_d_sin4_core.S: New file.
* sysdeps/x86_64/fpu/svml_d_sin4_core_avx.S: New file.
* sysdeps/x86_64/fpu/svml_d_sin8_core.S: New file.
* sysdeps/x86_64/fpu/svml_d_sin_data.S: New file.
* sysdeps/x86_64/fpu/svml_d_sin_data.h: New file.
* sysdeps/x86_64/fpu/test-double-vlen2-wrappers.c: Added vector sin test.
* sysdeps/x86_64/fpu/test-double-vlen2.c: Likewise.
* sysdeps/x86_64/fpu/test-double-vlen4-avx2-wrappers.c: Likewise.
* sysdeps/x86_64/fpu/test-double-vlen4-avx2.c: Likewise.
* sysdeps/x86_64/fpu/test-double-vlen4-wrappers.c: Likewise.
* sysdeps/x86_64/fpu/test-double-vlen4.c: Likewise.
* sysdeps/x86_64/fpu/test-double-vlen8-wrappers.c: Likewise.
* sysdeps/x86_64/fpu/test-double-vlen8.c: Likewise.
* NEWS: Mention addition of x86_64 vector sin.
* math/Makefile: Added CFLAGS for new tests.
* math/test-float-vlen16.h: New file.
* math/test-float-vlen4.h: New file.
* math/test-float-vlen8.h: New file.
* math/test-double-vlen2.h: Fixed 2 argument macro and comment.
* sysdeps/x86_64/fpu/Makefile: Added new tests and variables.
* sysdeps/x86_64/fpu/libm-test-ulps: Regenerated.
* sysdeps/x86_64/fpu/test-float-vlen16-wrappers.c: New file.
* sysdeps/x86_64/fpu/test-float-vlen16.c: New file.
* sysdeps/x86_64/fpu/test-float-vlen4-wrappers.c: New file.
* sysdeps/x86_64/fpu/test-float-vlen4.c: New file.
* sysdeps/x86_64/fpu/test-float-vlen8-avx2-wrappers.c: New file.
* sysdeps/x86_64/fpu/test-float-vlen8-avx2.c: New file.
* sysdeps/x86_64/fpu/test-float-vlen8-wrappers.c: New file.
* sysdeps/x86_64/fpu/test-float-vlen8.c: New file.
We test vector math functions using scalar tests infrastructure with
help of special wrappers from scalar versions to vector ones. Wrapper
implemented using platform specific vector types and placed in separate
file for compilation with architecture specific options, main part of
test has no such options. With help of system of definitions unfolding
of which is drived from test code we have wrapper called in individual
testing function instead of scalar function. Also system of definitions
includes generated during make check header math/libm-have-vector-test.h
with series of conditional definitions which help to avoid build fails
for functions having no vector versions; runtime architecture check
to prevent runtime fails of test run on inappropriate hardware.
* math/Makefile: Added rules for vector tests.
* math/gen-libm-have-vector-test.sh: Added generation of wrapper
declaration under condition.
* math/test-double-vlen2.h: New file.
* math/test-double-vlen4.h: New file.
* math/test-double-vlen8.h: New file.
* math/test-vec-loop.h: Added initialization macro.
* sysdeps/x86_64/fpu/Makefile: Added variables for vector tests.
* sysdeps/x86_64/fpu/libm-test-ulps: Regenarated.
* sysdeps/x86_64/fpu/math-tests-arch.h: New file.
* sysdeps/x86_64/fpu/test-double-vlen2-wrappers.c: New file.
* sysdeps/x86_64/fpu/test-double-vlen2.c: New file.
* sysdeps/x86_64/fpu/test-double-vlen4-avx2-wrappers.c: New file.
* sysdeps/x86_64/fpu/test-double-vlen4-avx2.c: New file.
* sysdeps/x86_64/fpu/test-double-vlen4-wrappers.c: New file.
* sysdeps/x86_64/fpu/test-double-vlen4.c: New file.
* sysdeps/x86_64/fpu/test-double-vlen8-wrappers.c: New file.
* sysdeps/x86_64/fpu/test-double-vlen8.c: New file.
Here is implementation of cos containing SSE, AVX, AVX2 and AVX512
versions according to Vector ABI which had been discussed in
<https://groups.google.com/forum/#!topic/x86-64-abi/LmppCfN1rZ4>.
Vector math library build and ABI testing enabled by default for x86_64.
* sysdeps/x86_64/fpu/Makefile: New file.
* sysdeps/x86_64/fpu/Versions: New file.
* sysdeps/x86_64/fpu/svml_d_cos_data.S: New file.
* sysdeps/x86_64/fpu/svml_d_cos_data.h: New file.
* sysdeps/x86_64/fpu/svml_d_cos2_core.S: New file.
* sysdeps/x86_64/fpu/svml_d_cos4_core.S: New file.
* sysdeps/x86_64/fpu/svml_d_cos4_core_avx.S: New file.
* sysdeps/x86_64/fpu/svml_d_cos8_core.S: New file.
* sysdeps/x86_64/fpu/svml_d_wrapper_impl.h: New file.
* sysdeps/x86_64/fpu/multiarch/svml_d_cos2_core.S: New file.
* sysdeps/x86_64/fpu/multiarch/svml_d_cos2_core_sse4.S: New file.
* sysdeps/x86_64/fpu/multiarch/svml_d_cos4_core.S: New file.
* sysdeps/x86_64/fpu/multiarch/svml_d_cos4_core_avx2.S: New file.
* sysdeps/x86_64/fpu/multiarch/svml_d_cos8_core.S: New file.
* sysdeps/x86_64/fpu/multiarch/svml_d_cos8_core_avx512.S: New file.
* sysdeps/x86_64/fpu/multiarch/Makefile (libmvec-sysdep_routines): Added
build of SSE, AVX2 and AVX512 IFUNC versions.
* sysdeps/x86/fpu/bits/math-vector.h: Added SIMD declaration for cos.
* math/bits/mathcalls.h: Added cos declaration with __MATHCALL_VEC.
* sysdeps/x86_64/configure.ac: Options for libmvec build.
* sysdeps/x86_64/configure: Regenerated.
* sysdeps/x86_64/sysdep.h (cfi_offset_rel_rsp): New macro.
* sysdeps/unix/sysv/linux/x86_64/libmvec.abilist: New file.
* manual/install.texi (Configuring and compiling): Document
--disable-mathvec.
* INSTALL: Regenerated.
* NEWS: Mention addition of libmvec and x86_64 vector cos.
ldbl-96 remquol wrongly handles the case where the first argument is
finite and the second infinite, because the check for the second
argument being a NaN fails to disregard the explicit high mantissa bit
and so wrongly interprets an infinity as being a NaN. This patch
fixes this by masking off that bit, and improves test coverage for
both remainder and remquo (various cases were missing tests, or, as in
the case of the bug, were tested only for one of the two functions).
Tested for x86_64 and x86.
[BZ #18244]
* sysdeps/ieee754/ldbl-96/s_remquol.c (__remquol): Ignore explicit
high mantissa bit when testing whether P is a NaN.
* math/libm-test.inc (remainder_test_data): Add more tests.
(remquo_test_data): Likewise.
The i386 implementation of atanhl, for small arguments, does a
calculation that involves computing twice the square of the argument,
resulting in spurious underflows for some arguments. This patch fixes
this by just returning the argument when its exponent is below -32,
with underflow being forced as needed for subnormal arguments.
Tested for x86 and x86_64.
[BZ #18049]
* sysdeps/i386/fpu/e_atanhl.S (__ieee754_atanhl): For exponents
below -32, return the argument, with underflow if subnormal.
* math/auto-libm-test-in: Add more tests of atanh.
* math/auto-libm-test-out: Regenerated.
Similar to various other bugs in this area, some atanh implementations
do not raise the underflow exception for subnormal arguments, when the
result is tiny and inexact. This patch forces the exception in a
similar way to previous fixes. (No change in this regard is needed
for the i386 implementation; special handling to force underflows in
these cases will only be needed there when the spurious underflows,
bug 18049, get fixed.)
Tested for x86_64, x86, powerpc and mips64.
[BZ #16352]
* sysdeps/i386/fpu/e_atanh.S (dbl_min): New object.
(__ieee754_atanh): Force underflow exception for results with
small absolute value.
* sysdeps/i386/fpu/e_atanhf.S (flt_min): New object.
(__ieee754_atanhf): Force underflow exception for results with
small absolute value.
* sysdeps/ieee754/dbl-64/e_atanh.c: Include <float.h>.
(__ieee754_atanh): Force underflow exception for results with
small absolute value.
* sysdeps/ieee754/flt-32/e_atanhf.c: Include <float.h>.
(__ieee754_atanhf): Force underflow exception for results with
small absolute value.
* sysdeps/ieee754/ldbl-128/e_atanhl.c: Include <float.h>.
(__ieee754_atanhl): Force underflow exception for results with
small absolute value.
* sysdeps/ieee754/ldbl-128ibm/e_atanhl.c: Include <float.h>.
(__ieee754_atanhl): Force underflow exception for results with
small absolute value.
* sysdeps/ieee754/ldbl-96/e_atanhl.c: Include <float.h>.
(__ieee754_atanhl): Force underflow exception for results with
small absolute value.
* math/auto-libm-test-in: Do not allow missing underflow
exceptions from atanh.
* math/auto-libm-test-out: Regenerated.
The flt-32 implementation of tanf produces spurious underflow
exceptions for some small arguments, through computing values on the
order of x^5. This patch fixes this by adjusting the threshold for
returning x (or, as applicable, +/- 1/x) to 2**-13 (the next term in
the power series being x^3/3).
Tested for x86_64 and x86.
[BZ #18221]
* sysdeps/ieee754/flt-32/k_tanf.c (__kernel_tanf): Use 2**-13 not
2**-28 as threshold for returning x or +/- 1/x.
* math/auto-libm-test-in: Add more tests of tan.
* math/auto-libm-test-out: Regenerated.
The flt-32 implementation of lgammaf produces spurious underflow
exceptions for some large arguments, because of calculations involving
x^-2 multiplied by small constants. This patch fixes this by
adjusting the threshold for a simpler computation to 2**26 (the error
in the simpler computation is on the order of 0.5 * log (x), for a
result on the order of x * log (x)).
Tested for x86_64 and x86.
[BZ #18220]
* sysdeps/ieee754/flt-32/e_lgammaf_r.c (__ieee754_lgammaf_r): Use
2**26 not 2**58 as threshold for returning x * (log (x) - 1).
* math/auto-libm-test-in: Add another test of lgamma.
* math/auto-libm-test-out: Regenerated.
The flt-32 implementation of erfcf produces spurious underflow
exceptions for some arguments close to 0, because of calculations
squaring the argument and then multiplying by small constants. This
patch fixes this by adjusting the threshold for arguments for which
the result is so close to 1 that 1 - x will give the right result from
2**-56 to 2**-26. (If 1 - x * 2/sqrt(pi) were used, the errors would be
on the order of x^3 and a much larger threshold could be used.)
Tested for x86_64 and x86.
[BZ #18217]
* sysdeps/ieee754/flt-32/s_erff.c (__erfcf): Use 2**-26 not 2**-56
as threshold for returning 1 - x.
* math/auto-libm-test-in: Add more tests of erfc.
* math/auto-libm-test-out: Regenerated.
The sysdeps/ieee754/flt-32 version of atanf produces spurious
underflow exceptions for some large arguments, because of computations
that compute x^-4. This patch fixes this by adjusting the threshold
for large arguments (for which +/- pi/2 can just be returned, the
correct result being roughly +/- pi/2 - 1/x) from 2^34 to 2^25.
Tested for x86_64 and x86.
[BZ #18196]
* sysdeps/ieee754/flt-32/s_atanf.c (__atanf): Use 2^25 not 2^34 as
threshold for large arguments.
* math/auto-libm-test-in: Add another test of atan.
* math/auto-libm-test-out: Regenerated.
Similar to various other bugs in this area, some log1p implementations
do not raise the underflow exception for subnormal arguments, when the
result is tiny and inexact. This patch forces the exception in a
similar way to previous fixes. (The ldbl-128ibm implementation
doesn't currently need any change as it already generates this
exception, albeit through code that would generate spurious exceptions
in other cases; special code for this issue will only be needed there
when fixing the spurious exceptions.)
Tested for x86_64, x86, powerpc and mips64.
[BZ #16339]
* sysdeps/i386/fpu/s_log1p.S (dbl_min): New object.
(__log1p): Force underflow exception for results with small
absolute value.
* sysdeps/i386/fpu/s_log1pf.S (flt_min): New object.
(__log1pf): Force underflow exception for results with small
absolute value.
* sysdeps/ieee754/dbl-64/s_log1p.c: Include <float.h>.
(__log1p): Force underflow exception for results with small
absolute value.
* sysdeps/ieee754/flt-32/s_log1pf.c: Include <float.h>.
(__log1pf): Force underflow exception for results with small
absolute value.
* sysdeps/ieee754/ldbl-128/s_log1pl.c: Include <float.h>.
(__log1pl): Force underflow exception for results with small
absolute value.
* math/auto-libm-test-in: Do not allow missing underflow
exceptions from log1p.
* math/auto-libm-test-out: Regenerated.
Installation of libm.so as linker script only in case of libmvec.so build.
2015-05-14 Andrew Senkevich <andrew.n.senkevich@gmail.com>
* Makeconfig (rpath-dirs, all-subdirs): Added mathvec folder.
(libmvec): New variable.
* configure.ac: Added option for mathvec build.
* configure: Regenerated.
* mathvec/Depend: New file.
* mathvec/Makefile: New file.
* shlib-versions: Added libmvec.
* math/Makefile: Added rule for libm.so installation.
declarations for math functions in math.h. Added new headers math-vector.h
(only generic version for now) and libm-simd-decl-stubs.h with empty
definitions required for proper unfolding of new macros __MATHCALL_VEC which
will be used for declaration of vector math functions.
2015-05-14 Andrew Senkevich <andrew.senkevich@intel.com>
* bits/math-vector.h: New file.
* bits/libm-simd-decl-stubs.h: New header.
* math/Makefile (headers): Added new header libm-simd-decl-stubs.h.
* math/math.h (__MATHCALL_VEC): New macro.
of method for separation which exactly testing function needed to run with
help of generated during make check header with series of conditional
definitions.
2015-05-14 Andrew Senkevich <andrew.senkevich@intel.com>
* math/gen-libm-have-vector-test.sh: Script generates series of macros
for conditions in testing functions.
* math/Makefile: Added call of libm-have-vector-test.sh.
* math/libm-test.inc (HAVE_VECTOR): New macros.
of vector math functions infrastructure and several x86_64 implementations.
This patch is preparatory change in libm-test.c - splitting of macros which
form name of tested functions for ability to use separate name for tested
functions and for functions used in test suite infrastructure.
2015-05-14 Andrew Senkevich <andrew.senkevich@intel.com>
* math/test-double.c (FUNC_TEST): New macro.
* math/test-float.c: Likewise.
* math/test-idouble.c: Likewise.
* math/test-ifloat.c: Likewise.
* math/test-ildoubl.c: Likewise.
* math/test-ldouble.c: Likewise.
* math/libm-test.inc: Use FUNC_TEST for name of tested functions.
This patch adds more randomly-generated tests of various libm
functions that are observed to increase ulps on x86_64.
Tested for x86_64 and x86 and ulps updated accordingly.
* math/auto-libm-test-in: Add more tests of csqrt, lgamma, log10
and sinh.
* math/auto-libm-test-out: Regenerated.
* sysdeps/i386/fpu/libm-test-ulps: Update.
* sysdeps/x86_64/fpu/libm-test-ulps: Likewise.
This patch adds more randomly-generated tests of various libm
functions that are observed to increase ulps on x86_64.
Tested for x86_64 and x86 and ulps updated accordingly.
* math/auto-libm-test-in: Add more tests of acosh, atanh, cos,
csqrt, erfc, sin and sincos.
* math/auto-libm-test-out: Regenerated.
* sysdeps/i386/fpu/libm-test-ulps: Update.
* sysdeps/x86_64/fpu/libm-test-ulps: Likewise.
This patch adds more randomly-generated tests of various libm
functions that are observed to increase ulps on x86_64. (This process
must eventually converge, when my random test generation stops finding
inputs that increase the listed ulps, except maybe for any cases
uncovered where the errors exceed the maximum allowed 9ulp error and
so indicate actual libm bugs needing fixing.)
Tested for x86_64 and x86 and ulps updated accordingly.
* math/auto-libm-test-in: Add more tests of acosh, atanh, clog,
clog10, csqrt, erfc, exp2, expm1, log10, log2 and sinh.
* math/auto-libm-test-out: Regenerated.
* sysdeps/i386/fpu/libm-test-ulps: Update.
* sysdeps/x86_64/fpu/libm-test-ulps: Likewise.
This patch adds more randomly-generated tests of various libm
functions that are observed to increase ulps on x86_64.
Tested for x86_64 and x86 and ulps updated accordingly.
* math/auto-libm-test-in: Add more tests of atan, clog, clog10,
cos, csqrt, erf, erfc, exp2, lgamma, log1p, sin, sincos, tanh and
tgamma.
* math/auto-libm-test-out: Regenerated.
* sysdeps/i386/fpu/libm-test-ulps: Update.
* sysdeps/x86_64/fpu/libm-test-ulps: Likewise.
This patch adds some randomly-generated tests of tgamma that are
observed to increase ulps on x86_64.
Tested for x86_64 and x86 and ulps updated accordingly.
* math/auto-libm-test-in: Add more tests of tgamma.
* math/auto-libm-test-out: Regenerated.
* sysdeps/i386/fpu/libm-test-ulps: Update.
* sysdeps/x86_64/fpu/libm-test-ulps: Likewise.
This patch adds some randomly-generated tests of tanh that are
observed to increase ulps on x86_64.
Tested for x86_64 and x86 and ulps updated accordingly.
* math/auto-libm-test-in: Add more tests of tanh.
* math/auto-libm-test-out: Regenerated.
* sysdeps/i386/fpu/libm-test-ulps: Update.
* sysdeps/x86_64/fpu/libm-test-ulps: Likewise.
