There is quiet truncation to double arithmetic in several
files. I noticed them when building ldbl-128 in a
soft-fp context. This did not change any test results.
sparc64 passes floating point values in the floating point registers.
As the the generic ceil, floor and trunc functions use integer
instructions, it makes sense to provide a VIS3 version consisting in
the the generic version compiled with -mvis3. GCC will then use
movdtox, movxtod, movwtos and movstow instructions.
sparc32 passes the floating point values in the integer registers, so it
doesn't make sense to do the same.
Changelog:
* sysdeps/ieee754/dbl-64/s_trunc.c: Avoid alias renamed.
* sysdeps/ieee754/dbl-64/wordsize-64/s_trunc.c: Likewise.
* sysdeps/ieee754/flt-32/s_truncf.c: Likewise.
* sysdeps/sparc/sparc64/fpu/multiarch/Makefile
[$(subdir) = math && $(have-as-vis3) = yes] (libm-sysdep_routines):
Add s_ceilf-vis3, s_ceil-vis3, s_floorf-vis3, s_floor-vis3,
s_truncf-vis3, s_trunc-vis3.
(CFLAGS-s_ceilf-vis3.c): New. Set to -Wa,-Av9d -mvis3.
(CFLAGS-s_ceil-vis3.c): Likewise.
(CFLAGS-s_floorf-vis3.c): Likewise.
(CFLAGS-s_floor-vis3.c): Likewise.
(CFLAGS-s_truncf-vis3.c): Likewise.
(CFLAGS-s_trunc-vis3.c): Likewise.
* sysdeps/sparc/sparc64/fpu/multiarch/s_ceil-vis3.c: New file.
* sysdeps/sparc/sparc64/fpu/multiarch/s_ceil.c: Likewise.
* sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf-vis3.c: Likewise.
* sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf.c: Likewise.
* sysdeps/sparc/sparc64/fpu/multiarch/s_floor-vis3.c: Likewise.
* sysdeps/sparc/sparc64/fpu/multiarch/s_floor.c: Likewise.
* sysdeps/sparc/sparc64/fpu/multiarch/s_floorf-vis3.c: Likewise.
* sysdeps/sparc/sparc64/fpu/multiarch/s_floorf.c: Likewise.
* sysdeps/sparc/sparc64/fpu/multiarch/s_trunc-vis3.c: Likewise.
* sysdeps/sparc/sparc64/fpu/multiarch/s_trunc.c: Likewise.
* sysdeps/sparc/sparc64/fpu/multiarch/s_truncf-vis3.c: Likewise.
* sysdeps/sparc/sparc64/fpu/multiarch/s_truncf.c: Likewise.
During the sincos consolidation I made two mistakes, one was a logical
error due to which cos(0x1.8475e5afd4481p+0) returned
sin(0x1.8475e5afd4481p+0) instead.
The second issue was an error in negating inputs for the correct
quadrants for sine. I could not find a suitable test case for this
despite running a program to search for such an input for a couple of
hours.
Following patch fixes both issues. Tested on x86_64. Thanks to Matt
Clay for identifying the issue.
[BZ #20357]
* sysdeps/ieee754/dbl-64/s_sin.c (sloww): Fix up condition
to call __mpsin/__mpcos and to negate values.
* math/auto-libm-test-in: Add test.
* math/auto-libm-test-out: Regenerate.
TS 18661 adds nextup and nextdown functions alongside nextafter to provide
support for float128 equivalent to it. This patch adds nextupl, nextup,
nextupf, nextdownl, nextdown and nextdownf to libm before float128 support.
The nextup functions return the next representable value in the direction of
positive infinity and the nextdown functions return the next representable
value in the direction of negative infinity. These are currently enabled
as GNU extensions.
The dbl-64 implementation of atan2, passed arguments (sNaN, qNaN),
fails to raise the "invalid" exception. This patch fixes it to add
both arguments, rather than just adding the second argument to itself,
in the case where the second argument is a NaN (which is checked for
before checking for the first argument being a NaN). sNaN tests for
atan2 are added, along with some qNaN tests I noticed were missing but
should have been there by analogy with other tests present.
Tested for x86_64 and x86.
[BZ #20252]
* sysdeps/ieee754/dbl-64/e_atan2.c (__ieee754_atan2): Add both
arguments when second argument is a NaN.
* math/libm-test.inc (atan2_test_data): Add sNaN tests and more
qNaN tests.
Various implementations of frexp functions return sNaN for sNaN
input. This patch fixes them to add such arguments to themselves so
that qNaN is returned.
Tested for x86_64, x86, mips64 and powerpc.
[BZ #20250]
* sysdeps/i386/fpu/s_frexpl.S (__frexpl): Add non-finite input to
itself.
* sysdeps/ieee754/dbl-64/s_frexp.c (__frexp): Add non-finite or
zero input to itself.
* sysdeps/ieee754/dbl-64/wordsize-64/s_frexp.c (__frexp):
Likewise.
* sysdeps/ieee754/flt-32/s_frexpf.c (__frexpf): Likewise.
* sysdeps/ieee754/ldbl-128/s_frexpl.c (__frexpl): Likewise.
* sysdeps/ieee754/ldbl-128ibm/s_frexpl.c (__frexpl): Likewise.
* sysdeps/ieee754/ldbl-96/s_frexpl.c (__frexpl): Likewise.
* math/libm-test.inc (frexp_test_data): Add sNaN tests.
The ldbl-128ibm version of log1pl returns sNaN for sNaN input. This
patch fixes it to add such inputs to themselves so that qNaN is
returned in this case.
Tested for powerpc.
[BZ #20234]
* sysdeps/ieee754/ldbl-128ibm/s_log1pl.c (__log1pl): Add positive
infinity or NaN input to itself.
The ldbl-128ibm version of expm1l returns sNaN for sNaN input. This
patch fixes it to add such inputs to themselves so that qNaN is
returned in this case.
Tested for powerpc.
[BZ #20233]
* sysdeps/ieee754/ldbl-128ibm/s_expm1l.c (__expm1l): Add NaN input
to itself.
The ldbl-128 version of expm1l returns sNaN for sNaN input. This
patch fixes it to add such inputs to themselves so that qNaN is
returned in this case.
Tested for mips64.
[BZ #20232]
* sysdeps/ieee754/ldbl-128/s_expm1l.c (__expm1l): Add NaN input to
itself.
The dbl-64 version of asin returns sNaN for sNaN arguments. This
patch fixes it to add NaN arguments to themselves so that qNaN is
returned in this case.
Tested for x86_64 and x86.
[BZ #20213]
* sysdeps/ieee754/dbl-64/e_asin.c (__ieee754_asin): Add NaN
argument to itself.
* math/libm-test.inc (asin_test_data): Add sNaN tests.
The dbl-64 version of acos returns sNaN for sNaN arguments. This
patch fixes it to add NaN arguments to themselves so that qNaN is
returned in this case.
Tested for x86_64 and x86.
[BZ #20212]
* sysdeps/ieee754/dbl-64/e_asin.c (__ieee754_acos): Add NaN
argument to itself.
* math/libm-test.inc (acos_test_data): Add sNaN tests.
The ldbl-128ibm implementations of ceill, floorl, roundl, truncl,
rintl and nearbyintl wrongly return an sNaN when given an sNaN
argument. This patch fixes them to add such an argument to itself to
turn it into a quiet NaN. (The code structure means this "else" case
applies to any argument which is zero or not finite; it's OK to do
this in all such cases.)
Tested for powerpc.
[BZ #20156]
* sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Add high part
to itself when zero or not finite.
* sysdeps/ieee754/ldbl-128ibm/s_floorl.c (__floorl): Likewise.
* sysdeps/ieee754/ldbl-128ibm/s_rintl.c (__rintl): Likewise.
* sysdeps/ieee754/ldbl-128ibm/s_roundl.c (__roundl): Likewise.
* sysdeps/ieee754/ldbl-128ibm/s_truncl.c (__truncl): Likewise.
The ldbl-128ibm implementation of sqrtl wrongly returns an sNaN for
signaling NaN arguments. This patch fixes it to quiet its argument,
using the same x * x + x return for infinities and NaNs as the dbl-64
implementation uses to ensure that +Inf maps to +Inf while -Inf and
NaN map to NaN.
Tested for powerpc.
[BZ #20153]
* sysdeps/ieee754/ldbl-128ibm/e_sqrtl.c (__ieee754_sqrtl): Return
x * x + x for infinities and NaNs.
The ldbl-128 implementations of j0l, j1l, y0l, y1l (also used for
ldbl-128ibm) return an sNaN argument unchanged. This patch fixes them
to add a NaN argument to itself to quiet it before return.
Tested for mips64.
[BZ #20151]
* sysdeps/ieee754/ldbl-128/e_j0l.c (__ieee754_j0l): Add NaN
argument to itself before returning result.
(__ieee754_y0l): Likewise.
* sysdeps/ieee754/ldbl-128/e_j1l.c (__ieee754_j1l): Likewise.
(__ieee754_y1l).
C99 and C11 allow but do not require ceil, floor, round and trunc to
raise the "inexact" exception for noninteger arguments. TS 18661-1
requires that this exception not be raised by these functions. This
aligns them with general IEEE semantics, where "inexact" is only
raised if the final step of rounding the infinite-precision result to
the result type is inexact; for these functions, the
infinite-precision integer result is always representable in the
result type, so "inexact" should never be raised.
The generic implementations of ceil, floor and round functions contain
code to force "inexact" to be raised. This patch removes it for round
functions to align them with TS 18661-1 in this regard. The tests
*are* updated by this patch; there are fewer architecture-specific
versions than for ceil and floor, and I fixed the powerpc ones some
time ago. If any others still have the issue, as shown by tests for
round failing with spurious exceptions, they can be fixed separately
by architecture maintainers or others.
Tested for x86_64, x86 and mips64.
[BZ #15479]
* sysdeps/ieee754/dbl-64/s_round.c (huge): Remove variable.
(__round): Do not force "inexact" exception.
* sysdeps/ieee754/dbl-64/wordsize-64/s_round.c (huge): Remove
variable.
(__round): Do not force "inexact" exception.
* sysdeps/ieee754/flt-32/s_roundf.c (huge): Remove variable.
(__roundf): Do not force "inexact" exception.
* sysdeps/ieee754/ldbl-128/s_roundl.c (huge): Remove variable.
(__roundl): Do not force "inexact" exception.
* sysdeps/ieee754/ldbl-96/s_roundl.c (huge): Remove variable.
(__roundl): Do not force "inexact" exception.
* math/libm-test.inc (round_test_data): Do not allow spurious
"inexact" exceptions.
C99 and C11 allow but do not require ceil, floor, round and trunc to
raise the "inexact" exception for noninteger arguments. TS 18661-1
requires that this exception not be raised by these functions. This
aligns them with general IEEE semantics, where "inexact" is only
raised if the final step of rounding the infinite-precision result to
the result type is inexact; for these functions, the
infinite-precision integer result is always representable in the
result type, so "inexact" should never be raised.
The generic implementations of ceil, floor and round functions contain
code to force "inexact" to be raised. This patch removes it for floor
functions to align them with TS 18661-1 in this regard. Note that
some architecture-specific versions may still raise "inexact", so the
tests are not updated and the bug is not yet fixed.
Tested for x86_64, x86 and mips64.
[BZ #15479]
* sysdeps/ieee754/dbl-64/s_floor.c: Do not mention "inexact"
exception in comment.
(huge): Remove variable.
(__floor): Do not force "inexact" exception.
* sysdeps/ieee754/dbl-64/wordsize-64/s_floor.c: Do not mention
"inexact" exception in comment.
(huge): Remove variable.
(__floor): Do not force "inexact" exception.
* sysdeps/ieee754/flt-32/s_floorf.c: Do not mention "inexact"
exception in comment.
(huge): Remove variable.
(__floorf): Do not force "inexact" exception.
* sysdeps/ieee754/ldbl-128/s_floorl.c: Do not mention "inexact"
exception in comment.
(huge): Remove variable.
(__floorl): Do not force "inexact" exception.
C99 and C11 allow but do not require ceil, floor, round and trunc to
raise the "inexact" exception for noninteger arguments. TS 18661-1
requires that this exception not be raised by these functions. This
aligns them with general IEEE semantics, where "inexact" is only
raised if the final step of rounding the infinite-precision result to
the result type is inexact; for these functions, the
infinite-precision integer result is always representable in the
result type, so "inexact" should never be raised.
The generic implementations of ceil, floor and round functions contain
code to force "inexact" to be raised. This patch removes it for ceil
functions to align them with TS 18661-1 in this regard. Note that
some architecture-specific versions may still raise "inexact", so the
tests are not updated and the bug is not yet fixed.
Tested for x86_64, x86 and mips64.
[BZ #15479]
* sysdeps/ieee754/dbl-64/s_ceil.c: Do not mention "inexact"
exception in comment.
(huge): Remove variable.
(__ceil): Do not force "inexact" exception.
* sysdeps/ieee754/dbl-64/wordsize-64/s_ceil.c: Do not mention
"inexact" exception in comment.
(huge): Remove variable.
(__ceil): Do not force "inexact" exception.
* sysdeps/ieee754/flt-32/s_ceilf.c (huge): Remove variable.
(__ceilf): Do not force "inexact" exception.
* sysdeps/ieee754/ldbl-128/s_ceill.c: Do not mention "inexact"
exception in comment.
(huge): Remove variable.
(__ceill): Do not force "inexact" exception.
ldbl-128ibm had an implementation of fmal that just did (x * y) + z in
most cases, with no attempt at actually being a fused operation.
This patch replaces it with a genuine fused operation. It is not
necessarily correctly rounding, but should produce a result at least
as accurate as the long double arithmetic operations in libgcc, which
I think is all that can reasonably be expected for such a non-IEEE
format where arithmetic is approximate rather than rounded according
to any particular rule for determining the exact result. Like the
libgcc arithmetic, it may produce spurious overflow and underflow
results, and it falls back to the libgcc multiplication in the case of
(finite, finite, zero).
This concludes the fixes for bug 13304; any subsequently found fma
issues should go in separate Bugzilla bugs. Various other pieces of
bug 13304 were fixed in past releases over the past several years.
Tested for powerpc.
[BZ #13304]
* sysdeps/ieee754/ldbl-128ibm/s_fmal.c: Include <fenv.h>,
<float.h>, <math_private.h> and <stdlib.h>.
(add_split): New function.
(mul_split): Likewise.
(ext_val): New typedef.
(store_ext_val): New function.
(mul_ext_val): New function.
(compare): New function.
(add_split_ext): New function.
(__fmal): After checking for Inf, NaN and zero, compute result as
an exact sum of scaled double values in round-to-nearest before
adding those up and adjusting for other rounding modes.
* math/auto-libm-test-in: Remove xfail-rounding:ldbl-128ibm from
tests of fma.
* math/auto-libm-test-out: Regenerated.
When the signs differ, the precision of the conversion sometimes
drops below 106 bits. This strategy is identical to the
hexadecimal variant.
I've refactored tst-sprintf3 to enable testing a value with more
than 30 significant digits in order to demonstrate this failure
and its solution.
Additionally, this implicitly fixes a typo in the shift
quantities when subtracting from the high mantissa to compute
the difference.
The ldbl-128ibm implementation of nearbyintl uses logic that only
works in round-to-nearest mode. This contrasts with rintl, which
works in all rounding modes.
Now, arguably nearbyintl could simply be aliased to rintl, given that
spurious "inexact" is generally allowed for ldbl-128ibm, even for the
underlying arithmetic operations. But given that the only point of
nearbyintl is to avoid "inexact", this patch follows the more
conservative approach of adding conditionals to the rintl
implementation to make it suitable for use to implement nearbyintl,
then builds it for nearbyintl with USE_AS_NEARBYINTL defined. The
test test-nearbyint-except-2 shows up issues when traps on "inexact"
are enabled, which turn out to be problems with the powerpc
fenv_private.h implementation (two functions that should disable
exception traps potentially failing to do so in some cases); this
patch duly fixes that as well (I don't see any other existing cases
where this would be user-visible; there isn't much use of *_NOEX,
*hold* etc. in libm that requires exceptions to be discarded and not
trapped on).
Tested for powerpc.
[BZ #19790]
* sysdeps/ieee754/ldbl-128ibm/s_rintl.c [USE_AS_NEARBYINTL]
(rintl): Define as macro.
[USE_AS_NEARBYINTL] (__rintl): Likewise.
(__rintl) [USE_AS_NEARBYINTL]: Use SET_RESTORE_ROUND_NOEX instead
of fesetround. Ensure results are evaluated before end of scope.
* sysdeps/ieee754/ldbl-128ibm/s_nearbyintl.c: Define
USE_AS_NEARBYINTL and include s_rintl.c.
* sysdeps/powerpc/fpu/fenv_private.h (libc_feholdsetround_ppc):
Disable exception traps in new environment.
(libc_feholdsetround_ppc_ctx): Likewise.
The ldbl-128ibm implementation of remainderl has logic resulting in
incorrect tests for equality of the absolute values of the arguments
in the case of zero low parts. If the low parts are both zero but
with different signs, this can wrongly cause equal arguments to be
treated as different, resulting in turn in incorrect signs of zero
result in nondefault rounding modes arising from the subtractions done
when the arguments are not equal.
This patch fixes the logic to convert -0 low parts into +0 before the
comparison (remquo already has separate logic to deal with signs of
zero results, so doesn't need such a change). Tests are added for
remainderl and remquol similar to that for fmodl, and based on a
refactoring of it, since the bug depends on low parts which should not
be relied upon in tests not setting the representation explicitly
(although in fact the bug shows up in test-ldouble with current GCC).
Tested for powerpc.
[BZ #19677]
* sysdeps/ieee754/ldbl-128ibm/e_remainderl.c
(__ieee754_remainderl): Put zero low parts in canonical form.
* sysdeps/ieee754/ldbl-128ibm/test-fmodrem-ldbl-128ibm.c: New
file. Based on
sysdeps/ieee754/ldbl-128ibm/test-fmodl-ldbl-128ibm.c.
* sysdeps/ieee754/ldbl-128ibm/test-fmodl-ldbl-128ibm.c: Replace
with wrapper round test-fmodrem-ldbl-128ibm.c.
* sysdeps/ieee754/ldbl-128ibm/test-remainderl-ldbl-128ibm.c: New
file.
* sysdeps/ieee754/ldbl-128ibm/test-remquol-ldbl-128ibm.c:
Likewise.
* sysdeps/ieee754/ldbl-128ibm/Makefile (tests): Add
test-remainderl-ldbl-128ibm and test-remquol-ldbl-128ibm.
The ldbl-128ibm implementation of nextafterl / nexttowardl returns -0
in FE_DOWNWARD mode when taking the next value below the least
positive subnormal, when it should return +0. This patch fixes it to
check explicitly for this case.
Tested for powerpc.
[BZ #19678]
* sysdeps/ieee754/ldbl-128ibm/s_nextafterl.c (__nextafterl):
Ensure +0.0 is returned when taking the next value below the least
positive value.
The ldbl-128ibm implementation of powl has some problems in the case
of overflow or underflow, which are mainly visible in non-default
rounding modes.
* When overflow or underflow is detected early, the correct sign of an
overflowing or underflowing result is not allowed for. This is
mostly hidden in the default rounding mode by the errno-setting
wrappers recomputing the result (except in non-default
error-handling modes such as -lieee), but visible in other rounding
modes where a result that is not zero or infinity causes the
wrappers not to do the recomputation.
* The final scaling is done before the sign is incorporated in the
result, but should be done afterwards for correct overflowing and
underflowing results in directed rounding modes.
This patch fixes those problems. Tested for powerpc.
[BZ #19674]
* sysdeps/ieee754/ldbl-128ibm/e_powl.c (__ieee754_powl): Include
sign in overflowing and underflowing results when overflow or
underflow is detected early. Include sign in result before rather
than after scaling.
The ldbl-128ibm implementations of remainderl and remquol have logic
resulting in incorrect tests for equality of the absolute values of
the arguments. Equality is tested based on the integer
representations of the high and low parts, with the sign bit masked
off the high part - but when this changes the sign of the high part,
the sign of the low part needs to be changed as well, and failure to
do this means arguments are wrongly treated as equal when they are
not.
This patch fixes the logic to adjust signs of low parts as needed.
Tested for powerpc.
[BZ #19603]
* sysdeps/ieee754/ldbl-128ibm/e_remainderl.c
(__ieee754_remainderl): Adjust sign of integer version of low part
when taking absolute value of high part.
* sysdeps/ieee754/ldbl-128ibm/s_remquol.c (__remquol): Likewise.
* math/libm-test.inc (remainder_test_data): Add another test.
(remquo_test_data): Likewise.
The ldbl-128ibm implementation of fmodl has logic to detect when the
first argument has absolute value less than or equal to the second.
This logic is only correct for nonzero low parts; if the high parts
are equal and the low parts are zero, then the signs of the low parts
(which have no semantic effect on the value of the long double number)
can result in equal values being wrongly treated as unequal, and an
incorrect result being returned from fmodl. This patch fixes this by
checking for the case of zero low parts.
Although this does show up in tests from libm-test.inc (both tests of
fmodl, and, indirectly, of remainderl / dreml), the dependence on
non-semantic zero low parts means that test shouldn't be expected to
reproduce it reliably; thus, this patch adds a standalone test that
sets up affected values using unions.
Tested for powerpc.
[BZ #19602]
* sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Handle
equal high parts and both low parts zero specially.
* sysdeps/ieee754/ldbl-128ibm/test-fmodl-ldbl-128ibm.c: New test.
* sysdeps/ieee754/ldbl-128ibm/Makefile [$(subdir) = math] (tests):
Add test-fmodl-ldbl-128ibm.
The ldbl-128ibm implementation of fmodl has completely bogus logic for
subnormal results (in this context, that means results for which the
result is in the subnormal range for double, not results with absolute
value below LDBL_MIN), based on code used for ldbl-128 that is correct
in that case but incorrect in the ldbl-128ibm use. This patch fixes
it to convert the mantissa into the correct form expected by
ldbl_insert_mantissa, removing the other cases of the code that were
incorrect and in one case unreachable for ldbl-128ibm. A correct
exponent value is then passed to ldbl_insert_mantissa to reflect the
shifted result.
Tested for powerpc.
[BZ #19595]
* sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Use
common logic for all cases of shifting subnormal results. Do not
insert sign bit in shifted mantissa. Always pass -1023 as biased
exponent to ldbl_insert_mantissa in subnormal case.
The ldbl-128ibm implementation of roundl is only correct in
round-to-nearest mode (in other modes, there are incorrect results and
overflow exceptions in some cases). This patch reimplements it along
the lines used for floorl, ceill and truncl, using __round on the high
part, and on the low part if the high part is an integer, and then
adjusting in the cases where this is incorrect.
Tested for powerpc.
[BZ #19594]
* sysdeps/ieee754/ldbl-128ibm/s_roundl.c (__roundl): Use __round
on high and low parts then adjust result and use
ldbl_canonicalize_int if needed.
The ldbl-128ibm implementation of truncl is only correct in
round-to-nearest mode (in other modes, there are incorrect results and
overflow exceptions in some cases). It is also unnecessarily
complicated, rounding both high and low parts to the nearest integer
and then adjusting for the semantics of trunc, when it seems more
natural to take the truncation of the high part (__trunc optimized
inline versions can be used), and the floor or ceiling of the low part
(depending on the sign of the high part) if the high part is an
integer, as was done for floorl and ceill. This patch makes it use
that simpler approach.
Tested for powerpc.
[BZ #19593]
* sysdeps/ieee754/ldbl-128ibm/s_truncl.c (__truncl): Use __trunc
on high part and __floor or __ceil on low part then use
ldbl_canonicalize_int if needed.
The ldbl-128ibm implementation of ceill is only correct in
round-to-nearest mode (in other modes, there are incorrect results and
overflow exceptions in some cases). It is also unnecessarily
complicated, rounding both high and low parts to the nearest integer
and then adjusting for the semantics of ceil, when it seems more
natural to take the ceiling of the high part (__ceil optimized inline
versions can be used), and that of the low part if the high part is an
integer, as was done for floorl. This patch makes it use that simpler
approach.
Tested for powerpc.
[BZ #19592]
* sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Use __ceil on
high and low parts then use ldbl_canonicalize_int if needed.
The ldbl-128ibm implementation of floorl is only correct in
round-to-nearest mode (in other modes, there are incorrect results and
overflow exceptions in some cases going beyond the incorrect signs of
zero results noted in bug 17899). It is also unnecessarily
complicated, rounding both high and low parts to the nearest integer
and then adjusting for the semantics of floor, when it seems more
natural to take the floor of the high part (__floor optimized inline
versions can be used), and that of the low part if the high part is an
integer. This patch makes it use that simpler approach, with a
canonicalization that works in all rounding modes (given that the only
way the result can be noncanonical is if taking the floor of a
negative noninteger low part increased its exponent).
Tested for powerpc, where over a thousand failures are removed from
test-ldouble.out (floorl problems affect many powl tests).
[BZ #17899]
* sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_canonicalize_int):
New function.
* sysdeps/ieee754/ldbl-128ibm/s_floorl.c (__floorl): Use __floor
on high and low parts then use ldbl_canonicalize_int if needed.
The changes to restrict implementation-namespace symbol aliases such
as __finitel to compat symbols used code for __finitel in libm
analogous to that for __finitel in libc. However, the versions for
the two symbols are actually different, GLIBC_2.0 in libc and
GLIBC_2.1 in libm. This patch fixes the handling of the libm compat
symbol.
Tested for mips (o32), where it fixes an ABI test failure.
* sysdeps/ieee754/dbl-64/s_finite.c
[NO_LONG_DOUBLE && LDBL_CLASSIFY_COMPAT] (__finitel): Define
compat symbol at version GLIBC_2_1 and use GLIBC_2_1 in
SHLIB_COMPAT condition for libm, not GLIBC_2_0.
* sysdeps/ieee754/dbl-64/wordsize-64/s_finite.c
[NO_LONG_DOUBLE && LDBL_CLASSIFY_COMPAT] (__finitel): Likewise.
I get some math test-failures on s390 for float/double/ldouble for
various lrint/lround functions like:
lrint (0x1p64): Exception "Inexact" set
lrint (-0x1p64): Exception "Inexact" set
lround (0x1p64): Exception "Inexact" set
lround (-0x1p64): Exception "Inexact" set
...
GCC emits "convert to fixed" instructions for casting floating point
values to integer values. These instructions raise invalid and inexact
exceptions if the floating point value exceeds the integer type ranges.
This patch enables the various FIX_DBL_LONG_CONVERT_OVERFLOW macros in
order to avoid a cast from floating point to integer type and raise the
invalid exception with feraiseexcept.
The ldbl-128 rint/round functions are now using the same logic.
ChangeLog:
[BZ #19486]
* sysdeps/s390/fix-fp-int-convert-overflow.h: New File.
* sysdeps/generic/fix-fp-int-convert-overflow.h
(FIX_LDBL_LONG_CONVERT_OVERFLOW,
FIX_LDBL_LLONG_CONVERT_OVERFLOW): New define.
* sysdeps/arm/fix-fp-int-convert-overflow.h: Likewise.
* sysdeps/mips/mips32/fpu/fix-fp-int-convert-overflow.h:
Likewise.
* sysdeps/ieee754/ldbl-128/s_lrintl.c (__lrintl):
Avoid conversions to long int where inexact exceptions
could be raised.
* sysdeps/ieee754/ldbl-128/s_lroundl.c (__lroundl):
Likewise.
* sysdeps/ieee754/ldbl-128/s_llrintl.c (__llrintl):
Avoid conversions to long long int where inexact exceptions
could be raised.
* sysdeps/ieee754/ldbl-128/s_llroundl.c (__llroundl):
Likewise.
When looking at the code generated for pow() on ppc64 I noticed quite
a few sign extensions. Making the array indices unsigned reduces the
number of sign extensions from 24 to 7.
Tested for powerpc64le and x86_64.
Like the previous change, exploit the fact that computation for sin
and cos is identical except that it is apart by a quadrant. Also
remove csloww, csloww1 and csloww2 since they can easily be expressed
in terms of sloww, sloww1 and sloww2.
The sin and cos computation for this range of input is identical
except for a difference in quadrants by 1. Exploit that fact and the
common argument reduction to reduce computations for sincos.
Range reduction needs to be done only once for sin and cos, so copy
over all of the relevant functions (__sin, __cos, reduce_and_compute)
and consolidate common code.
The ldbl-128ibm implementation of logl is inaccurate for arguments
near 1, because when deciding whether to bypass a series expansion for
log(1+z), where z = x-1, it compares the square of z rather than z
itself with an epsilon value. This patch fixes that comparison, so
eliminating the test failures for inaccuracy of logl in such cases.
Tested for powerpc.
[BZ #19351]
* sysdeps/ieee754/ldbl-128ibm/e_logl.c (__ieee754_logl): When
expanding log(1+z), compare z rather than its square with epsilon
to determine when to avoid evaluating the expansion.
The ldbl-128ibm implementation of sinhl uses a slightly too small
overflow threshold (similar to bug 16407 for coshl). This patch fixes
it to use a safe threshold (so that values whose high part is above
the value compared with definitely result in an overflow in all
rounding modes).
Tested for powerpc.
[BZ #19350]
* sysdeps/ieee754/ldbl-128ibm/e_sinhl.c (__ieee754_sinhl):
Increase overflow threshold.
The ldbl-128ibm implementation of tanhl is inaccurate for small
arguments, because it returns x*(1+x) (maybe in an attempt to raise
"inexact") when x itself would be the accurate return value but
multiplying by 1+x introduces large errors. This patch fixes it to
return x in that case (when the mathematical result is x plus a
negligible remainder on the order of x^3) to avoid those errors.
Tested for powerpc.
[BZ #19349]
* sysdeps/ieee754/ldbl-128ibm/s_tanhl.c (__tanhl): Return argument
when small.
If a platform does not define "long-double-fcts = yes" in its
Makefiles and it does define __NO_LONG_DOUBLE_MATH in its installed
headers, it will currently create exported symbols for __finitel,
__isinfl, and __isnanl that can't be reached from userspace by
correct use of the finite(), isinf(), or isnan() macros in <math.h>.
To avoid this situation, by default for such platforms we now no
longer export these symbols, thus causing appropriate link-time
errors. However, for platforms that previously exported these
symbols, we continue to do so as compat symbols; this is enabled
by adding LDBL_CLASSIFY_COMPAT to math_private.h for the platform.
For tile, remove the now-unnecessary exports of those functions from
libc and libm.
Various sysdeps/ieee754/dbl-64 functions use double constants defined
using a union between a double and two ints, with separate big-endian
and little-endian definitions of the constants.
With modern C, this is unnecessary complication; hex float constants
(or __builtin_inf etc.) suffice to specify the exact value desired,
and so can avoid separate versions for each endianness. Having this
complication also complicates cleanups such as removing slow paths
from these library functions, as they need to make sure to remove both
copies of variables that are no longer used after such a cleanup (and
in at least one case, proper removal of a slow path will also involve
removing slow-path-only values from the middle of an array - an array
with both big-endian and little-endian copies - and adjusting other
references to that array).
So it makes sense to clean up the code to define these constants using
hex floats and so eliminate the endianness conditional. This patch
does so in the case of sqrt, where the two constants are such that it
makes sense just to put them directly in the code using them and
eliminate the names for them altogether.
Tested for arm (the code generated for sqrt does change, though not in
any significant way).
* sysdeps/ieee754/dbl-64/e_sqrt.c: Do not include uroot.h.
(__ieee754_sqrt): Use hex float constants instead of tm256.x and
t512.x.
* sysdeps/ieee754/dbl-64/uroot.h: Remove file.
The nan* functions handle their string argument by constructing a
NAN(...) string on the stack as a VLA and passing it to strtod
functions.
This approach has problems discussed in bug 16961 and bug 16962: the
stack usage is unbounded, and it gives incorrect results in certain
cases where the argument is not a valid n-char-sequence.
The natural fix for both issues is to refactor the NaN payload parsing
out of strtod into a separate function that the nan* functions can
call directly, so that no temporary string needs constructing on the
stack at all. This patch does that refactoring in preparation for
fixing those bugs (but without actually using the new functions from
nan* - which will also require exporting them from libc at version
GLIBC_PRIVATE). This patch is not intended to change any user-visible
behavior, so no tests are added (fixes for the above bugs will of
course add tests for them).
This patch builds on my recent fixes for strtol and strtod issues in
Turkish locales. Given those fixes, the parsing of NaN payloads is
locale-independent; thus, the new functions do not need to take a
locale_t argument.
Tested for x86_64, x86, mips64 and powerpc.
* stdlib/strtod_nan.c: New file.
* stdlib/strtod_nan_double.h: Likewise.
* stdlib/strtod_nan_float.h: Likewise.
* stdlib/strtod_nan_main.c: Likewise.
* stdlib/strtod_nan_narrow.h: Likewise.
* stdlib/strtod_nan_wide.h: Likewise.
* stdlib/strtof_nan.c: Likewise.
* stdlib/strtold_nan.c: Likewise.
* sysdeps/ieee754/ldbl-128/strtod_nan_ldouble.h: Likewise.
* sysdeps/ieee754/ldbl-128ibm/strtod_nan_ldouble.h: Likewise.
* sysdeps/ieee754/ldbl-96/strtod_nan_ldouble.h: Likewise.
* wcsmbs/wcstod_nan.c: Likewise.
* wcsmbs/wcstof_nan.c: Likewise.
* wcsmbs/wcstold_nan.c: Likewise.
* stdlib/Makefile (routines): Add strtof_nan, strtod_nan and
strtold_nan.
* wcsmbs/Makefile (routines): Add wcstod_nan, wcstold_nan and
wcstof_nan.
* include/stdlib.h (__strtof_nan): Declare and use
libc_hidden_proto.
(__strtod_nan): Likewise.
(__strtold_nan): Likewise.
(__wcstof_nan): Likewise.
(__wcstod_nan): Likewise.
(__wcstold_nan): Likewise.
* include/wchar.h (____wcstoull_l_internal): Declare.
* stdlib/strtod_l.c: Do not include <ieee754.h>.
(____strtoull_l_internal): Remove declaration.
(STRTOF_NAN): Define macro.
(SET_MANTISSA): Remove macro.
(STRTOULL): Likewise.
(____STRTOF_INTERNAL): Use STRTOF_NAN to parse NaN payload.
* stdlib/strtof_l.c (____strtoull_l_internal): Remove declaration.
(STRTOF_NAN): Define macro.
(SET_MANTISSA): Remove macro.
* sysdeps/ieee754/ldbl-128/strtold_l.c (STRTOF_NAN): Define macro.
(SET_MANTISSA): Remove macro.
* sysdeps/ieee754/ldbl-128ibm/strtold_l.c (STRTOF_NAN): Define
macro.
(SET_MANTISSA): Remove macro.
* sysdeps/ieee754/ldbl-64-128/strtold_l.c (STRTOF_NAN): Define
macro.
(SET_MANTISSA): Remove macro.
* sysdeps/ieee754/ldbl-96/strtold_l.c (STRTOF_NAN): Define macro.
(SET_MANTISSA): Remove macro.
* wcsmbs/wcstod_l.c (____wcstoull_l_internal): Remove declaration.
* wcsmbs/wcstof_l.c (____wcstoull_l_internal): Likewise.
* wcsmbs/wcstold_l.c (____wcstoull_l_internal): Likewise.
The lgamma (and likewise lgammaf, lgammal) function wrongly sets the
signgam variable even when building for strict ISO C conformance
(-std=c99 / -std=c11), although the user may define such a variable
and it's only in the implementation namespace for POSIX with XSI
extensions enabled.
Following discussions starting at
<https://sourceware.org/ml/libc-alpha/2013-04/msg00767.html> and
<https://sourceware.org/ml/libc-alpha/2015-10/msg00844.html>, it seems
that the safest approach for fixing this particular issue is for
signgam to become a weak alias for a newly exported symbol __signgam,
with the library functions only setting __signgam, at which point
static linker magic will preserve the alias for newly linked binaries
that refer to the library's signgam rather than defining their own,
while breaking the alias for programs that define their own signgam,
with new symbol versions for lgamma functions and with compat symbols
for existing binaries that set both signgam and __signgam.
This patch implements that approach for the fix. signgam is made into
a weak alias. The four symbols __signgam, lgamma, lgammaf, lgammal
get new symbol versions at version GLIBC_2.23, with the existing
versions of lgamma, lgammaf and lgammal becoming compat symbols.
When the compat versions are built, gamma, gammaf and gammal are
aliases for the compat versions (i.e. always set signgam); this is OK
as they are not ISO C functions, and avoids adding new symbol versions
for them unnecessarily. When the compat versions are not built
(i.e. for static linking and for future glibc ports), gamma, gammaf
and gammal are aliases for the new versions that set __signgam. The
ldbl-opt versions are updated accordingly.
The lgamma wrappers are adjusted so that the same source files,
included from different files with different definitions of
USE_AS_COMPAT, can build either the new versions or the compat
versions. Similar changes are made to the ia64 versions (untested).
Tests are added that the lgamma functions do not interfere with a user
variable called signgam for ISO C, with various choices for the size
of that variable, whether it is initialized, and for static and
dynamic linking. The conformtest whitelist entry is removed as well.
Tested for x86_64, x86, mips64 and powerpc, including looking at
objdump --dynamic-syms output to make sure the expected sets of
symbols were aliases. Also spot-tested that a binary built with old
glibc works properly (i.e. gets signgam set) when run with new glibc.
[BZ #15421]
* sysdeps/ieee754/s_signgam.c (signgam): Rename to __signgam,
initialize with 0 and define as weak alias of __signgam.
* include/math.h [!_ISOMAC] (__signgam): Declare.
* math/Makefile (libm-calls): Add w_lgamma_compat.
(tests): Add test-signgam-uchar, test-signgam-uchar-init,
test-signgam-uint, test-signgam-uint-init, test-signgam-ullong and
test-signgam-ullong-init.
(tests-static): Add test-signgam-uchar-static,
test-signgam-uchar-init-static, test-signgam-uint-static,
test-signgam-uint-init-static, test-signgam-ullong-static and
test-signgam-ullong-init-static.
(CFLAGS-test-signgam-uchar.c): New variable.
(CFLAGS-test-signgam-uchar-init.c): Likewise.
(CFLAGS-test-signgam-uchar-static.c): Likewise.
(CFLAGS-test-signgam-uchar-init-static.c): Likewise.
(CFLAGS-test-signgam-uint.c): Likewise.
(CFLAGS-test-signgam-uint-init.c): Likewise.
(CFLAGS-test-signgam-uint-static.c): Likewise.
(CFLAGS-test-signgam-uint-init-static.c): Likewise.
(CFLAGS-test-signgam-ullong.c): Likewise.
(CFLAGS-test-signgam-ullong-init.c): Likewise.
(CFLAGS-test-signgam-ullong-static.c): Likewise.
(CFLAGS-test-signgam-ullong-init-static.c): Likewise.
* math/Versions (libm): Add GLIBC_2.23.
* math/lgamma-compat.h: New file.
* math/test-signgam-main.c: Likewise.
* math/test-signgam-uchar-init-static.c: Likewise.
* math/test-signgam-uchar-init.c: Likewise.
* math/test-signgam-uchar-static.c: Likewise.
* math/test-signgam-uchar.c: Likewise.
* math/test-signgam-uint-init-static.c: Likewise.
* math/test-signgam-uint-init.c: Likewise.
* math/test-signgam-uint-static.c: Likewise.
* math/test-signgam-uint.c: Likewise.
* math/test-signgam-ullong-init-static.c: Likewise.
* math/test-signgam-ullong-init.c: Likewise.
* math/test-signgam-ullong-static.c: Likewise.
* math/test-signgam-ullong.c: Likewise.
* math/w_lgamma.c: Rename to w_lgamma_main.c and replace by
wrapper of w_lgamma_main.c.
* math/w_lgamma_compat.c: New file.
* math/w_lgamma_compatf.c: Likewise.
* math/w_lgamma_compatl.c: Likewise.
* math/w_lgamma_main.c: New file. Based on w_lgamma.c. Include
<lgamma-compat.h>. Condition contents on [BUILD_LGAMMA]. Support
defining compatibility symbols.
(__lgamma): Change to LGFUNC (__lgamma). Use CALL_LGAMMA.
* math/w_lgammaf.c: Rename to w_lgammaf_main.c and replace by
wrapper of w_lgammaf_main.c.
* math/w_lgammaf_main.c: New file. Based on w_lgammaf.c. Include
<lgamma-compat.h>. Condition contents on [BUILD_LGAMMA]. Support
defining compatibility symbols.
(__lgammaf): Change to LGFUNC (__lgammaf). Use CALL_LGAMMA.
* math/w_lgammal.c: Rename to w_lgammal_main.c and replace by
wrapper of w_lgammal_main.c.
* math/w_lgammal_main.c: New file. Based on w_lgammal.c. Include
<lgamma-compat.h>. Condition contents on [BUILD_LGAMMA]. Support
defining compatibility symbols.
(__lgammal): Change to LGFUNC (__lgammal). Use CALL_LGAMMA.
* sysdeps/ia64/fpu/lgamma-compat.h: New file.
* sysdeps/ia64/fpu/w_lgamma.c: Move to ....
* sysdeps/ia64/fpu/w_lgamma_main.c: ...here. Include
<lgamma-compat.h>.
(__ieee754_lgamma): Change to LGFUNC (lgamma). Use CALL_LGAMMA.
(__ieee754_gamma): Define as alias.
* sysdeps/ia64/fpu/w_lgammaf.c: Move to ....
* sysdeps/ia64/fpu/w_lgammaf_main.c: ...here. Include
<lgamma-compat.h>.
(__ieee754_lgammaf): Change to LGFUNC (lgammaf). Use CALL_LGAMMA.
(__ieee754_gammaf): Define as alias.
* sysdeps/ia64/fpu/w_lgammal.c: Move to ....
* sysdeps/ia64/fpu/w_lgammal_main.c: ...here. Include
<lgamma-compat.h>.
(__ieee754_lgammal): Change to LGFUNC (lgammal). Use CALL_LGAMMA.
(__ieee754_gammal): Define as alias.
* sysdeps/ieee754/ldbl-opt/w_lgamma.c: Move to ....
* sysdeps/ieee754/ldbl-opt/w_lgamma_compat.c: ...here. Include
<math/w_lgamma_compat.c>.
[LONG_DOUBLE_COMPAT(libm, GLIBC_2_0)] (__lgammal_dbl_compat):
Define as alias of __lgamma_compat and use in defining lgammal.
* sysdeps/ieee754/ldbl-opt/w_lgammal.c: Move to ....
* sysdeps/ieee754/ldbl-opt/w_lgamma_compatl.c: ...here. Include
<math/lgamma-compat.h> and <math/w_lgamma_compatl.c>.
(USE_AS_COMPAT): New macro.
(LGAMMA_OLD_VER): Undefine and redefine.
(lgammal): Do not define here.
(gammal): Only define here if [GAMMA_ALIAS].
* conform/linknamespace.pl (@whitelist): Remove signgam.
* sysdeps/nacl/libm.abilist: Update.
* sysdeps/unix/sysv/linux/aarch64/libm.abilist: Likewise.
* sysdeps/unix/sysv/linux/alpha/libm.abilist: Likewise.
* sysdeps/unix/sysv/linux/arm/libm.abilist: Likewise.
* sysdeps/unix/sysv/linux/hppa/libm.abilist: Likewise.
* sysdeps/unix/sysv/linux/i386/libm.abilist: Likewise.
* sysdeps/unix/sysv/linux/ia64/libm.abilist: Likewise.
* sysdeps/unix/sysv/linux/m68k/coldfire/libm.abilist: Likewise.
* sysdeps/unix/sysv/linux/m68k/m680x0/libm.abilist: Likewise.
* sysdeps/unix/sysv/linux/microblaze/libm.abilist: Likewise.
* sysdeps/unix/sysv/linux/mips/mips32/libm.abilist: Likewise.
* sysdeps/unix/sysv/linux/mips/mips64/libm.abilist: Likewise.
* sysdeps/unix/sysv/linux/nios2/libm.abilist: Likewise.
* sysdeps/unix/sysv/linux/powerpc/powerpc32/fpu/libm.abilist:
Likewise.
* sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/libm.abilist:
Likewise.
* sysdeps/unix/sysv/linux/powerpc/powerpc64/libm-le.abilist:
Likewise.
* sysdeps/unix/sysv/linux/powerpc/powerpc64/libm.abilist:
Likewise.
* sysdeps/unix/sysv/linux/s390/s390-32/libm.abilist: Likewise.
* sysdeps/unix/sysv/linux/s390/s390-64/libm.abilist: Likewise.
* sysdeps/unix/sysv/linux/sh/libm.abilist: Likewise.
* sysdeps/unix/sysv/linux/sparc/sparc32/libm.abilist: Likewise.
* sysdeps/unix/sysv/linux/sparc/sparc64/libm.abilist: Likewise.
* sysdeps/unix/sysv/linux/tile/tilegx/tilegx32/libm.abilist:
Likewise.
* sysdeps/unix/sysv/linux/tile/tilegx/tilegx64/libm.abilist:
Likewise.
* sysdeps/unix/sysv/linux/tile/tilepro/libm.abilist: Likewise.
* sysdeps/unix/sysv/linux/x86_64/64/libm.abilist: Likewise.
* sysdeps/unix/sysv/linux/x86_64/x32/libm.abilist: Likewise.
Kind of hokey, but errno.h drags in misc/sys/param.h which
defines MIN/MAX causing an error. Include system headers
first to grab MIN/MAX definition in param.h, and define
HAVE_ALLOCA to preserve existing behavior.
* sysdeps/ieee754/ldbl-128ibm/mpn2ldl.c: Include gmp headers
after system headers to prevent MIN/MAX redefinition. Define
HAVE_ALLOCA to preserve builtin alloca usage.
Include the __sin and __cos functions as local static copies to allow
deper optimization of the functions. This change shows an improvement
of about 17% in the min case and 12.5% in the mean case for the sincos
microbenchmark on x86_64.
* sysdeps/ieee754/dbl-64/s_sin.c (__sin)[IN_SINCOS]: Mark function
static and don't set or restore rounding.
(__cos)[IN_SINCOS]: Likewise.
* sysdeps/ieee754/dbl-64/s_sincos.c: Include s_sin.c.
(__sincos): Set and restore rounding mode. Remove check for infinite
or NaN input.
