The optimization introduced in commit
f13c2a8dff, causes regressions in
sorting for languages that have digraphs that change sort order, like
cs_CZ which sorts ch between h and i.
My analysis shows the fast-forwarding optimization in STRCOLL advances
through a digraph while possibly stopping in the middle which results
in a subsequent skipping of the digraph and incorrect sorting. The
optimization is incorrect as implemented and because of that I'm
removing it for 2.23, and I will also commit this fix for 2.22 where
it was originally introduced.
This patch reverts the optimization, introduces a new bug-strcoll2.c
regression test that tests both cs_CZ.UTF-8 and da_DK.ISO-8859-1 and
ensures they sort one digraph each correctly. The optimization can't be
applied without regressing this test.
Checked on x86_64, bug-strcoll2.c fails without this patch and passes
after. This will also get a fix on 2.22 which has the same bug.
The dbl-64, ldbl-96 and ldbl-128 implementations of lround and llround
fail to produce "invalid" exceptions in cases where the rounded result
overflows the target type, but truncating the floating-point argument
to the next integer towards zero does not overflow it (so in
particular casts do not produce such exceptions). (This issue cannot
arise for float, or for double with 64-bit target type, or for ldbl-96
with 64-bit target type and negative arguments, because of
insufficient precision in the floating-point type for arguments with
the relevant property to exist.)
This patch fixes these problems by inserting checks for the special
cases that can occur in each implementation, and explicitly raising
FE_INVALID (and avoiding the cast if it might raise spurious
FE_INEXACT).
Tested for x86_64, x86 and mips64.
[BZ #19088]
* sysdeps/ieee754/dbl-64/s_lround.c: Include <fenv.h> and
<limits.h>.
(__lround) [FE_INVALID]: Force FE_INVALID exception when result
overflows but exception would not result from cast.
* sysdeps/ieee754/dbl-64/wordsize-64/s_lround.c: Include <fenv.h>
and <limits.h>.
(__lround) [FE_INVALID]: Force FE_INVALID exception when result
overflows but exception would not result from cast.
* sysdeps/ieee754/ldbl-128/s_llroundl.c: Include <fenv.h> and
<limits.h>.
(__llroundl) [FE_INVALID]: Force FE_INVALID exception when result
overflows but exception would not result from cast.
* sysdeps/ieee754/ldbl-128/s_lroundl.c: Include <fenv.h> and
<limits.h>.
(__lroundl) [FE_INVALID]: Force FE_INVALID exception when result
overflows but exception would not result from cast.
* sysdeps/ieee754/ldbl-96/s_llroundl.c: Include <fenv.h> and
<limits.h>.
(__llroundl) [FE_INVALID]: Force FE_INVALID exception when result
overflows but exception would not result from cast.
* sysdeps/ieee754/ldbl-96/s_lroundl.c: Include <fenv.h> and
<limits.h>.
(__lroundl) [FE_INVALID]: Force FE_INVALID exception when result
overflows but exception would not result from cast.
* math/libm-test.inc (lround_test_data): Add more tests.
(llround_test_data): Likewise.
The ldbl-128 implementations of lrintl and lroundl miss "invalid"
exceptions on systems with 32-bit long for arguments that overflow
long but have exponent below 48. This patch fixes this by rearranging
the sequence of tests in the code so the exponent < 48 case is only
used for exponents that don't overflow long.
Tested for mips64 (n32 and n64).
[BZ #19085]
* sysdeps/ieee754/ldbl-128/s_lrintl.c (__lrintl): Move test for
exponent below 48 inside case for non-overflowing exponent.
* sysdeps/ieee754/ldbl-128/s_lroundl.c (__lroundl): Likewise.
The implementation of lround in dbl-64/wordsize-64 as an alias or
wrapper for llround is always incorrect when long is not 64-bit,
because it misses required exceptions in overflow cases, as shown by
my recently added tests. This patch removes that alias / wrapper in
the non-LP64 case, together with the REGISTER_CAST_INT32_TO_INT64
macro, restoring the previous version of lround for dbl-64/wordsize-64
(newly conditioned on !_LP64).
Tested for x86_64, and for mips64 with use of dbl-64/wordsize-64
enabled.
[BZ #19079]
* sysdeps/ieee754/dbl-64/wordsize-64/s_lround.c: Restore previous
file, conditioned on [!_LP64].
* sysdeps/ieee754/dbl-64/wordsize-64/s_llround.c
[!_LP64] (__lround): Do not define as function or alias.
[!_LP64] (lround): Likewise.
[!_LP64] (__lroundl): Likewise.
[!_LP64] (lroundl): Likewise.
* sysdeps/tile/sysdep.h (REGISTER_CAST_INT32_TO_INT64): Remove
macro.
* sysdeps/x86_64/x32/sysdep.h (REGISTER_CAST_INT32_TO_INT64):
Likewise.
The ldbl-128ibm expl wrapper checks the argument to determine when to
call __kernel_standard_l, thereby overriding overflowing results from
__ieee754_expl that could otherwise (given appropriately patched
libgcc) be correct for the rounding mode. This patch changes it to
check the result of __ieee754_expl instead, as other versions of this
wrapper do.
Tested for powerpc.
[BZ #19078]
* sysdeps/ieee754/ldbl-128ibm/w_expl.c (o_thres): Remove variable.
(u_thres): Likewise.
(__expl): Determine whether to call __kernel_standard_l based on
value of result, not argument.
The ldbl-128ibm implementation of logl produces a zero with the wrong
sign for logl (1) in FE_DOWNWARD mode. This patch makes it explicitly
return 0.0L in that case, as in e.g. the ldbl-128 implementation.
Tested for powerpc.
[BZ #19077]
* sysdeps/ieee754/ldbl-128ibm/e_logl.c (__ieee754_logl): Return
0.0L for argument 1.0L.
The ldbl-128ibm implementation of log1pl produces an infinity with the
wrong sign for log1pl (-1) in FE_DOWNWARD mode. This patch fixes this
by changing a division (-1.0L / (x - x)) (incorrect in FE_DOWNWARD
mode) to (-1.0L / 0.0L) (correct in all rounding modes).
Tested for powerpc.
[BZ #19076]
* sysdeps/ieee754/ldbl-128ibm/s_log1pl.c (__log1pl): Divide by
constant 0.0L when computing infinite result.
The ldbl-96 version of lroundl is incorrect for systems with 64-bit
long when the argument's absolute value is just below a power of 2,
2^32 or more, and rounds up to the next integer; in such cases, it
returns 0. The problem is incrementing the high part of the mantissa
loses the high bit of the value (which is not an issue for any other
floating-point format, and is handled specially in lround when the bit
corresponding to 0.5 was in the high part rather than the low part).
This patch fixes this in a similar way to that used in llroundl:
storing the high part in an unsigned long variable before incrementing
it, so problems cannot occur in the case when this code is reachable.
I improved test coverage for both lround and llround by making them
use the same test inputs (appropriately conditioned on the size of
long in the lround case) - complete with the same comments, to make
comparison as easy as possible. (This test coverage improvement was
how I found the lroundl bug.)
Tested for x86_64 and x86.
[BZ #19071]
* sysdeps/ieee754/ldbl-96/s_lroundl.c (__lroundl): Use unsigned
long int variable to store possibly incremented high part of
mantissa.
* math/libm-test.inc (lround_test_data): Add tests used for
llround. Use [LONG_MAX > 0x7fffffff] consistently as condition
for tests requiring 64-bit long. Do not condition tests on
[TEST_FLOAT] unnecessarily.
(llround_test_data): Add tests used for lround. Add another
expectation for the "inexact" exception. Do not condition tests
on [TEST_FLOAT] unnecessarily.
On powerpc32 hard-float, older processors (ones where fcfid is not
available for 32-bit code), GCC generates conversions from integers to
floating point that wrongly convert integer 0 to -0 instead of +0 in
FE_DOWNWARD mode. This in turn results in logb and a few other
functions wrongly returning -0 when they should return +0.
This patch works around this issue in glibc as I proposed in
<https://sourceware.org/ml/libc-alpha/2015-09/msg00728.html>, so that
the affected functions can be correct and the affected tests pass in
the absence of a GCC fix for this longstanding issue (GCC bug 67771 -
if fixed, of course we can put in GCC version conditionals, and
eventually phase out the workarounds). A new macro
FIX_INT_FP_CONVERT_ZERO is added in a new sysdeps header
fix-int-fp-convert-zero.h, and the powerpc32/fpu version of that
header defines the macro based on the results of a configure test for
whether such conversions use the fcfid instruction.
Tested for x86_64 (that installed stripped shared libraries are
unchanged by the patch) and powerpc (that HAVE_PPC_FCFID comes out to
0 as expected and that the relevant tests are fixed). Also tested a
build with GCC configured for -mcpu=power4 and verified that
HAVE_PPC_FCFID comes out to 1 in that case.
There are still some other issues to fix to get test-float and
test-double passing cleanly for older powerpc32 processors (apart from
the need for an ulps regeneration for powerpc). (test-ldouble will be
harder to get passing cleanly, but with a combination of selected
fixes to ldbl-128ibm code that don't involve significant performance
issues, allowing spurious underflow and inexact exceptions for that
format, and lots of XFAILing for the default case of unpatched libgcc,
it should be doable.)
[BZ #887]
[BZ #19049]
[BZ #19050]
* sysdeps/generic/fix-int-fp-convert-zero.h: New file.
* sysdeps/ieee754/dbl-64/e_log10.c: Include
<fix-int-fp-convert-zero.h>.
(__ieee754_log10): Adjust signs as needed if FIX_INT_FP_CONVERT_ZERO.
* sysdeps/ieee754/dbl-64/e_log2.c: Include
<fix-int-fp-convert-zero.h>.
(__ieee754_log2): Adjust signs as needed if FIX_INT_FP_CONVERT_ZERO.
* sysdeps/ieee754/dbl-64/s_erf.c: Include
<fix-int-fp-convert-zero.h>.
(__erfc): Adjust signs as needed if FIX_INT_FP_CONVERT_ZERO.
* sysdeps/ieee754/dbl-64/s_logb.c: Include
<fix-int-fp-convert-zero.h>.
(__logb): Adjust signs as needed if FIX_INT_FP_CONVERT_ZERO.
* sysdeps/ieee754/flt-32/e_log10f.c: Include
<fix-int-fp-convert-zero.h>.
(__ieee754_log10f): Adjust signs as needed if FIX_INT_FP_CONVERT_ZERO.
* sysdeps/ieee754/flt-32/e_log2f.c: Include
<fix-int-fp-convert-zero.h>.
(__ieee754_log2f): Adjust signs as needed if FIX_INT_FP_CONVERT_ZERO.
* sysdeps/ieee754/flt-32/s_erff.c: Include
<fix-int-fp-convert-zero.h>.
(__erfcf): Adjust signs as needed if FIX_INT_FP_CONVERT_ZERO.
* sysdeps/ieee754/flt-32/s_logbf.c: Include
<fix-int-fp-convert-zero.h>.
(__logbf): Adjust signs as needed if FIX_INT_FP_CONVERT_ZERO.
* sysdeps/ieee754/ldbl-128ibm/s_erfl.c: Include
<fix-int-fp-convert-zero.h>.
(__erfcl): Adjust signs as needed if FIX_INT_FP_CONVERT_ZERO.
* sysdeps/ieee754/ldbl-128ibm/s_logbl.c: Include
<fix-int-fp-convert-zero.h>.
(__logbl): Adjust signs as needed if FIX_INT_FP_CONVERT_ZERO.
* sysdeps/powerpc/powerpc32/fpu/configure.ac: New file.
* sysdeps/powerpc/powerpc32/fpu/configure: New generated file.
* sysdeps/powerpc/powerpc32/fpu/fix-int-fp-convert-zero.h: New
file.
* config.h.in [_LIBC] (HAVE_PPC_FCFID): New macro.
ISO C requires overflowing results from nexttoward to be the
appropriate infinity independent of the rounding mode, but some
implementations use a rounding-mode-dependent result (this is the same
issue as was fixed for nextafter in bug 16677). This patch fixes the
problem by making the nexttoward implementations discard the result
from the floating-point computation that forced an overflow exception
and then return the infinity previously computed with integer
arithmetic.
Tested for x86_64, x86, mips64 and powerpc.
[BZ #19059]
* math/s_nexttowardf.c (__nexttowardf): Do not return value from
overflowing computation.
* sysdeps/i386/fpu/s_nexttoward.c (__nexttoward): Likewise.
* sysdeps/i386/fpu/s_nexttowardf.c (__nexttowardf): Likewise.
* sysdeps/ieee754/ldbl-128/s_nexttoward.c (__nexttoward):
Likewise.
* sysdeps/ieee754/ldbl-128/s_nexttowardf.c (__nexttowardf):
Likewise.
* sysdeps/ieee754/ldbl-128ibm/s_nexttoward.c (__nexttoward):
Likewise.
* sysdeps/ieee754/ldbl-128ibm/s_nexttowardf.c (__nexttowardf):
Likewise.
* sysdeps/ieee754/ldbl-96/s_nexttoward.c (__nexttoward): Likewise.
* sysdeps/ieee754/ldbl-96/s_nexttowardf.c (__nexttowardf):
Likewise.
* sysdeps/ieee754/ldbl-opt/s_nexttowardfd.c (__nldbl_nexttowardf):
Likewise.
* math/libm-test.inc (nexttoward_test_data): Add more tests.
This prevents injection of ':' and '\n' into output functions which
use the NSS files database syntax. Critical fields (user/group names
and file system paths) are checked strictly. For backwards
compatibility, the GECOS field is rewritten instead.
The getent program is adjusted to use the put*ent functions in libc,
instead of local copies. This changes the behavior of getent if user
names start with '-' or '+'.
The ldbl-128 / ldbl-128ibm implementation of lgamma has problems with
its handling of large arguments. It has an overflow threshold that is
correct only for ldbl-128, despite being used for both types - with
diagnostic control macros as a temporary measure to disable warnings
about that constant overflowing for ldbl-128ibm - and it has a
calculation that's roughly x * log(x) - x, resulting in overflows for
arguments that are roughly at most a factor 1/log(threshold) below the
overflow threshold.
This patch fixes both issues, using an overflow threshold appropriate
for the type in question and adding another case for large arguments
that avoids the possible intermediate overflow.
Tested for x86_64, x86, mips64 and powerpc.
[BZ #16347]
[BZ #19046]
* sysdeps/ieee754/ldbl-128/e_lgammal_r.c: Do not include
<libc-internal.h>.
(MAXLGM): Do not use diagnostic control macros.
[LDBL_MANT_DIG == 106] (MAXLGM): Change value to overflow
threshold for ldbl-128ibm.
(__ieee754_lgammal_r): For large arguments, multiply by log - 1
instead of multiplying by log then subtracting.
* math/auto-libm-test-in: Add more tests of lgamma.
* math/auto-libm-test-out: Regenerated.
The ldbl-128ibm implementation of exp10l uses a version of log(10)
split into high and low parts - but the low part is negative, so
causing spurious overflows from __ieee754_expl (exp_high) in cases
close to the overflow threshold (I added relevant tests close to the
overflow threshold to the testsuite earlier today). The same issue
applies close to the underflow threshold as well (except that spurious
underflows in IBM long double arithmetic are harder to fix than the
other deficiencies, so we might end up permitting those for IBM long
double in the libm testsuite, as permitted by ISO C).
This patch fixes it to use a low part rounded downward to 48 bits
instead. (The choice of 48 instead of 53 bits is to make it more
obviously safe even when the low part of the argument is negative.)
Tested for powerpc. (Note that because of libgcc bugs with
multiplication very close to LDBL_MAX, libgcc also needs patching for
all the problem cases to be fixed, but this patch is still safe and
correct in the absence of such libgcc fixes.)
[BZ #16620]
* sysdeps/ieee754/ldbl-128ibm/e_exp10l.c (log10_high): Use value
of log (10) rounded downward to 48 bits.
(log10_low): Use corresponding low part of log (10).
The i386 versions of acoshf and acosh raise a spurious "invalid"
exception for an argument that is a quiet NaN with the sign bit set.
The integer arithmetic to detect arguments < 1 also detects -NaN, and
then the computation 0 / 0 in that case raises the exception. This
patch fixes this by using (x - x) / (x - x) as the computation in that
case instead, which will always raise the exception for non-NaN
arguments reaching that code, but not for quiet NaN arguments.
Tested for x86_64 and x86.
[BZ #19032]
* sysdeps/i386/fpu/e_acosh.S (__ieee754_acosh): For arguments < 1,
compute result as (x - x) / (x - x) not as 0 / 0.
* sysdeps/i386/fpu/e_acoshf.S (__ieee754_acoshf): Likewise.
* math/libm-test.inc (acosh_test_data): Add another test of acosh.
For arguments with X^2 + Y^2 close to 1, clog and clog10 avoid large
errors from log(hypot) by computing X^2 + Y^2 - 1 in a way that avoids
cancellation error and then using log1p.
However, the thresholds for using that approach still result in log
being used on argument as large as sqrt(13/16) > 0.9, leading to
significant errors, in some cases above the 9ulp maximum allowed in
glibc libm. This patch arranges for the approach using log1p to be
used in any cases where |X|, |Y| < 1 and X^2 + Y^2 >= 0.5 (with the
existing allowance for cases where one of X and Y is very small),
adjusting the __x2y2m1 functions to work with the wider range of
inputs. This way, log only gets used on arguments below sqrt(1/2) (or
substantially above 1), where the error involved is much less.
Tested for x86_64, x86, mips64 and powerpc. For the ulps regeneration
I removed the existing clog and clog10 ulps before regenerating to
allow any reduced ulps to appear. Tests added include those found by
random test generation to produce large ulps either before or after
the patch, and some found by trying inputs close to the (0.75, 0.5)
threshold where the potential errors from using log are largest.
[BZ #19016]
* sysdeps/generic/math_private.h (__x2y2m1f): Update comment to
allow more cases with X^2 + Y^2 >= 0.5.
* sysdeps/ieee754/dbl-64/x2y2m1.c (__x2y2m1): Likewise. Add -1 as
normal element in sum instead of special-casing based on values of
arguments.
* sysdeps/ieee754/dbl-64/x2y2m1f.c (__x2y2m1f): Update comment.
* sysdeps/ieee754/ldbl-128/x2y2m1l.c (__x2y2m1l): Likewise. Add
-1 as normal element in sum instead of special-casing based on
values of arguments.
* sysdeps/ieee754/ldbl-128ibm/x2y2m1l.c (__x2y2m1l): Likewise.
* sysdeps/ieee754/ldbl-96/x2y2m1.c [FLT_EVAL_METHOD != 0]
(__x2y2m1): Update comment.
* sysdeps/ieee754/ldbl-96/x2y2m1l.c (__x2y2m1l): Likewise. Add -1
as normal element in sum instead of special-casing based on values
of arguments.
* math/s_clog.c (__clog): Handle more cases using log1p without
hypot.
* math/s_clog10.c (__clog10): Likewise.
* math/s_clog10f.c (__clog10f): Likewise.
* math/s_clog10l.c (__clog10l): Likewise.
* math/s_clogf.c (__clogf): Likewise.
* math/s_clogl.c (__clogl): Likewise.
* math/auto-libm-test-in: Add more tests of clog and clog10.
* math/auto-libm-test-out: Regenerated.
* sysdeps/i386/fpu/libm-test-ulps: Update.
* sysdeps/x86_64/fpu/libm-test-ulps: Likewise.
The flt-32 version of powf can be inaccurate because of bugs in the
extra-precision calculation of (x-1)/(x+1) or (x-1.5)/(x+1.5) as part
of calculating log(x) with extra precision: a constant used (as part
of adding 1 or 1.5 through integer arithmetic) is incorrect, and then
the code fails to mask a computed high part before using it in
arithmetic that relies on s_h*t_h being exactly representable. This
patch fixes these bugs.
Tested for x86_64 and x86. x86_64 ulps for powf removed and
regenerated to reflect reduced ulps from the increased accuracy for
existing tests.
[BZ #18956]
* sysdeps/ieee754/flt-32/e_powf.c (__ieee754_powf): Add 0x00400000
not 0x0040000 for high bit of mantissa. Mask with 0xfffff000 when
extracting high part.
* math/auto-libm-test-in: Add another test of pow.
* math/auto-libm-test-out: Regenerated.
* sysdeps/x86_64/fpu/libm-test-ulps: Update.
Similar to various other bugs in this area, pow functions can fail to
raise the underflow exception when the result is tiny and inexact but
one or more low bits of the intermediate result that is scaled down
(or, in the i386 case, converted from a wider evaluation format) are
zero. This patch forces the exception in a similar way to previous
fixes, thereby concluding the fixes for known bugs with missing
underflow exceptions currently filed in Bugzilla.
Tested for x86_64, x86, mips64 and powerpc.
[BZ #18825]
* sysdeps/i386/fpu/i386-math-asm.h (FLT_NARROW_EVAL_UFLOW_NONNAN):
New macro.
(DBL_NARROW_EVAL_UFLOW_NONNAN): Likewise.
(LDBL_CHECK_FORCE_UFLOW_NONNAN): Likewise.
* sysdeps/i386/fpu/e_pow.S: Use DEFINE_DBL_MIN.
(__ieee754_pow): Use DBL_NARROW_EVAL_UFLOW_NONNAN instead of
DBL_NARROW_EVAL, reloading the PIC register as needed.
* sysdeps/i386/fpu/e_powf.S: Use DEFINE_FLT_MIN.
(__ieee754_powf): Use FLT_NARROW_EVAL_UFLOW_NONNAN instead of
FLT_NARROW_EVAL. Use separate return path for case when first
argument is NaN.
* sysdeps/i386/fpu/e_powl.S: Include <i386-math-asm.h>. Use
DEFINE_LDBL_MIN.
(__ieee754_powl): Use LDBL_CHECK_FORCE_UFLOW_NONNAN, reloading the
PIC register.
* sysdeps/ieee754/dbl-64/e_pow.c (__ieee754_pow): Use
math_check_force_underflow_nonneg.
* sysdeps/ieee754/flt-32/e_powf.c (__ieee754_powf): Force
underflow for subnormal result.
* sysdeps/ieee754/ldbl-128/e_powl.c (__ieee754_powl): Likewise.
* sysdeps/ieee754/ldbl-128ibm/e_powl.c (__ieee754_powl): Use
math_check_force_underflow_nonneg.
* sysdeps/x86/fpu/powl_helper.c (__powl_helper): Use
math_check_force_underflow.
* sysdeps/x86_64/fpu/x86_64-math-asm.h
(LDBL_CHECK_FORCE_UFLOW_NONNAN): New macro.
* sysdeps/x86_64/fpu/e_powl.S: Include <x86_64-math-asm.h>. Use
DEFINE_LDBL_MIN.
(__ieee754_powl): Use LDBL_CHECK_FORCE_UFLOW_NONNAN.
* math/auto-libm-test-in: Add more tests of pow.
* math/auto-libm-test-out: Regenerated.
Fix a regression introduced with commit 0d23a5c1 [Static dlopen
correction fallout fixes] that caused the default library search path to
be ignored for modules loaded with dlopen from static executables.
[BZ #17250]
* elf/dl-support.c (_dl_main_map): Don't initialize l_flags_1
member.
Similar to various other bugs in this area, hypot functions can fail
to raise the underflow exception when the result is tiny and inexact
but one or more low bits of the intermediate result that is scaled
down (or, in the i386 case, converted from a wider evaluation format)
are zero. This patch forces the exception in a similar way to
previous fixes.
Note that this issue cannot arise for implementations of hypotf using
double (or wider) for intermediate evaluation (if hypotf should
underflow, that means the double square root is being computed of some
number of the form N*2^-298, for 0 < N < 2^46, which is exactly
represented as a double, and whatever the rounding mode such a square
root cannot have a mantissa with all zeroes after the initial 23
bits). Thus no changes are made to hypotf implementations in this
patch, only to hypot and hypotl.
Tested for x86_64, x86, mips64 and powerpc.
[BZ #18803]
* sysdeps/i386/fpu/e_hypot.S: Use DEFINE_DBL_MIN.
(MO): New macro.
(__ieee754_hypot) [PIC]: Load PIC register.
(__ieee754_hypot): Use DBL_NARROW_EVAL_UFLOW_NONNEG instead of
DBL_NARROW_EVAL.
* sysdeps/ieee754/dbl-64/e_hypot.c (__ieee754_hypot): Use
math_check_force_underflow_nonneg in case where result might be
tiny.
* sysdeps/ieee754/ldbl-128/e_hypotl.c (__ieee754_hypotl):
Likewise.
* sysdeps/ieee754/ldbl-128ibm/e_hypotl.c (__ieee754_hypotl):
Likewise.
* sysdeps/ieee754/ldbl-96/e_hypotl.c (__ieee754_hypotl): Likewise.
* sysdeps/powerpc/fpu/e_hypot.c (__ieee754_hypot): Likewise.
* math/auto-libm-test-in: Add more tests of hypot.
* math/auto-libm-test-out: Regenerated.
The x86_64 fma4 version of pow fails to disable contraction of
operations other than those explicitly intended to use fma
instructions, so resulting in large ulps errors on processors with
fma4 instructions, as in bug 18104 (165ulp for the test added for that
bug; error originally reported by "blaaa" on #glibc). This patch adds
$(config-cflags-nofma) for e_pow-fma4.c, corresponding to the use for
e_pow.c in sysdeps/ieee754/dbl-64/Makefile.
Tested for x86_64 on a processor with fma4.
[BZ #19003]
* sysdeps/x86_64/fpu/multiarch/Makefile (CFLAGS-e_pow-fma4.c): Add
$(config-cflags-nofma).
i386 exp, hypot and pow functions can return overflowing and
underflowing values with excess range and precision; ; Wilco
Dijkstra's patches to make isfinite etc. expand inline cause this
pre-existing issue to result in test failures.
This patch fixes those functions to avoid excess range and precision
in their return values. Appropriate macros are added for the repeated
code sequences; in future I'll add more such macros and refactor
existing code forcing underflow (with or without also eliminating
excess range and precision from the return value) to use such macros.
Tested for x86. If, after this patch, you still see x86 libm test
failures with excess range or precision, please file bugs in Bugzilla.
[BZ #18980]
* sysdeps/i386/fpu/i386-math-asm.h (DEFINE_FLT_MIN): New macro.
(DEFINE_DBL_MIN): Likewise.
(FLT_NARROW_EVAL_UFLOW_NONNEG_NAN): Likewise.
(DBL_NARROW_EVAL_UFLOW_NONNEG_NAN): Likewise.
(FLT_NARROW_EVAL_UFLOW_NONNEG): Likewise.
(DBL_NARROW_EVAL_UFLOW_NONNEG): Likewise.
* sysdeps/i386/fpu/e_exp.S: Include <i386-math-asm.h>.
(dbl_min): Replace with use of DEFINE_DBL_MIN.
(__ieee754_exp): Use DBL_NARROW_EVAL_UFLOW_NONNEG_NAN.
(__exp_finite): Use DBL_NARROW_EVAL_UFLOW_NONNEG.
* sysdeps/i386/fpu/e_exp10.S: Include <i386-math-asm.h>.
(dbl_min): Replace with use of DEFINE_DBL_MIN.
(__ieee754_exp10): Use DBL_NARROW_EVAL_UFLOW_NONNEG_NAN.
* sysdeps/i386/fpu/e_exp10f.S: Include <i386-math-asm.h>.
(flt_min): Replace with use of DEFINE_FLT_MIN.
(__ieee754_exp10f): Use FLT_NARROW_EVAL_UFLOW_NONNEG_NAN.
* sysdeps/i386/fpu/e_exp2.S: Include <i386-math-asm.h>.
(dbl_min): Replace with use of DEFINE_DBL_MIN.
(__ieee754_exp2): Use DBL_NARROW_EVAL_UFLOW_NONNEG_NAN.
* sysdeps/i386/fpu/e_exp2f.S: Include <i386-math-asm.h>.
(flt_min): Replace with use of DEFINE_FLT_MIN.
(__ieee754_exp2f): Use FLT_NARROW_EVAL_UFLOW_NONNEG_NAN.
* sysdeps/i386/fpu/e_expf.S: Include <i386-math-asm.h>.
(flt_min): Replace with use of DEFINE_FLT_MIN.
(__ieee754_expf): Use FLT_NARROW_EVAL_UFLOW_NONNEG_NAN.
(__expf_finite): Use FLT_NARROW_EVAL_UFLOW_NONNEG.
* sysdeps/i386/fpu/e_hypot.S: Include <i386-math-asm.h>.
(__ieee754_hypot): Use DBL_NARROW_EVAL.
* sysdeps/i386/fpu/e_hypotf.S: Include <i386-math-asm.h>.
(__ieee754_hypotf): Use FLT_NARROW_EVAL.
* sysdeps/i386/fpu/e_pow.S: Include <i386-math-asm.h>.
(__ieee754_pow): Use DBL_NARROW_EVAL.
* sysdeps/i386/fpu/e_powf.S: Include <i386-math-asm.h>.
(__ieee754_powf): Use FLT_NARROW_EVAL.
* sysdeps/i386/i686/fpu/multiarch/e_expf-sse2.S
(__ieee754_expf_sse2): Convert double-precision result to single
precision.
* sysdeps/i386/fpu/libm-test-ulps: Update.
i386 scalb / scalbn / scalbln (and thus ldexp) functions for float and
double can return results with excess range (and consequently excess
precision for subnormal results). As the results of these functions
are fully determined by reference to IEEE 754 operations, this is
unambiguously a bug, apart from the testsuite failures it causes.
This patch makes those functions store their results on the stack and
load them back to eliminate the excess range. Double rounding is not
a problem, as the only cases where it could occur are when the result
overflows or underflows for extended precision, and then the
double-rounded results are the same as the single-rounded results.
The new macros will be used for more functions, more such macros
added, and existing code refactored to use such macros, in subsequent
patches.
Tested for x86. Committed.
[BZ #18981]
* sysdeps/i386/fpu/i386-math-asm.h: New file.
* sysdeps/i386/fpu/e_scalb.S: Include <i386-math-asm.h>.
(__ieee754_scalb): Use DBL_NARROW_EVAL.
* sysdeps/i386/fpu/e_scalbf.S: Include <i386-math-asm.h>.
(__ieee754_scalbf): Use FLT_NARROW_EVAL.
* sysdeps/i386/fpu/s_scalbn.S: Include <i386-math-asm.h>.
(__scalbn): Use DBL_NARROW_EVAL.
* sysdeps/i386/fpu/s_scalbnf.S: Include <i386-math-asm.h>.
(__scalbnf): Use FLT_NARROW_EVAL.
built-ins when available. Since going through the PLT is expensive for these small functions,
inlining results in major speedups (about 7x on Cortex-A57 for isinf). The GCC built-ins are not
correct if signalling NaN support is required, and thus are turned off in that case (see GCC bug
66462). The test-snan.c tests sNaNs and so must be explicitly built with -fsignaling-nans.
2015-09-18 Wilco Dijkstra <wdijkstr@arm.com>
[BZ #15367]
[BZ #17441]
* math/Makefile: Build test-snan.c with -fsignaling-nans.
* math/math.h (fpclassify): Use __builtin_fpclassify when
available. (signbit): Use __builtin_signbit(f/l).
(isfinite): Use__builtin_isfinite. (isnormal): Use
__builtin_isnormal. (isnan): Use __builtin_isnan.
(isinf): Use __builtin_isinf_sign.
In ISO 8601, +03:30 is a valid time zone. Currently, strptime() only
parses it as a 2-digit time zone an believes this is +03:00. This change
makes it accept a single colon.
C99/C11 Annex G specifies the sign of the zero part of the result of
ctan (x +/- i * Inf) and ctanh (+/-Inf + i * y). This patch fixes glibc
to follow that specification, along the lines I described in my review
of Andreas's previous patch for this issue
<https://sourceware.org/ml/libc-alpha/2014-08/msg00142.html>.
Tested for x86_64.
2015-09-17 Joseph Myers <joseph@codesourcery.com>
Andreas Schwab <schwab@suse.de>
[BZ #17118]
* math/s_ctan.c (__ctan): Determine sign of zero real part of
result when imaginary part of argument is infinite using sine and
cosine.
* math/s_ctanf.c (__ctanf): Likewise.
* math/s_ctanl.c (__ctanl): Likewise.
* math/s_ctanh.c (__ctanh): Determine sign of zero imaginary part
of result when real part of argument is infinite using sine and
cosine.
* math/s_ctanhf.c (__ctanhf): Likewise.
* math/s_ctanhl.c (__ctanhl): Likewise.
* math/libm-test.inc (ctan_test_data): Add more tests of ctan.
(ctanh_test_data): Add more tests of ctanh.
Bug 15384 notes that in __finite, two different constants are used
that could be the same constant (the result only depends on the
exponent of the floating-point representation), and that using the
same constant is better for architectures where constants need loading
from a constant pool. This patch implements that change.
Tested for x86_64, mips64 and powerpc.
[BZ #15384]
* sysdeps/ieee754/dbl-64/s_finite.c (FINITE): Use same constant as
bit-mask as in subtraction.
* sysdeps/ieee754/dbl-64/wordsize-64/s_finite.c (__finite):
Likewise.
* sysdeps/ieee754/flt-32/s_finitef.c (FINITEF): Likewise.
* sysdeps/ieee754/ldbl-128/s_finitel.c (__finitel): Likewise.
* sysdeps/ieee754/ldbl-128ibm/s_finitel.c (__finitel): Likewise.
Similar to various other bugs in this area, tgamma functions can fail
to raise the underflow exception when the result is tiny and inexact
but one or more low bits of the intermediate result that is scaled
down are zero. This patch forces the exception in a similar way to
previous fixes.
Tested for x86_64, x86, mips64 and powerpc.
[BZ #18951]
* sysdeps/ieee754/dbl-64/e_gamma_r.c (__ieee754_gamma_r): Force
underflow exception for small results.
* sysdeps/ieee754/flt-32/e_gammaf_r.c (__ieee754_gammaf_r):
Likewise.
* sysdeps/ieee754/ldbl-128/e_gammal_r.c (__ieee754_gammal_r):
Likewise.
* sysdeps/ieee754/ldbl-128ibm/e_gammal_r.c (__ieee754_gammal_r):
Likewise.
* sysdeps/ieee754/ldbl-96/e_gammal_r.c (__ieee754_gammal_r):
Likewise.
* math/auto-libm-test-in: Add more tests of tgamma.
* math/auto-libm-test-out: Regenerated.
As noted in bug 6803, scalbn fails to set errno on overflow and
underflow. This patch fixes this by making scalbn an alias of ldexp,
which has exactly the same semantics (for floating-point types with
radix 2) and already has wrappers that deal with setting errno,
instead of an alias of the internal __scalbn (which ldexp calls).
Notes:
* Where compat symbols were defined for scalbn functions, I didn't
change what they point to (to keep the patch minimal), so such
compat symbols continue to go directly to the non-errno-setting
functions.
* Mike, I didn't do anything with the IA64 versions of these
functions, where I think both the ldexp and scalbn functions already
deal with setting errno. As a cleanup (not needed to fix this bug)
however you might want to make those functions into aliases for
IA64; there is no need for them to be separate function
implementations at all.
* This concludes the fix for bug 6803 since the scalb and scalbln
cases of that bug were fixed some time ago.
Tested for x86_64, x86, mips64 and powerpc.
[BZ #6803]
* math/s_ldexp.c (scalbn): Define as weak alias of __ldexp.
[NO_LONG_DOUBLE] (scalbnl): Define as weak alias of __ldexp.
* math/s_ldexpf.c (scalbnf): Define as weak alias of __ldexpf.
* math/s_ldexpl.c (scalbnl): Define as weak alias of __ldexpl.
* sysdeps/i386/fpu/s_scalbn.S (scalbn): Remove alias.
* sysdeps/i386/fpu/s_scalbnf.S (scalbnf): Likewise.
* sysdeps/i386/fpu/s_scalbnl.S (scalbnl): Likewise.
* sysdeps/ieee754/dbl-64/s_scalbn.c (scalbn): Likewise.
[NO_LONG_DOUBLE] (scalbnl): Likewise.
* sysdeps/ieee754/dbl-64/wordsize-64/s_scalbn.c (scalbn):
Likewise.
[NO_LONG_DOUBLE] (scalbnl): Likewise.
* sysdeps/ieee754/flt-32/s_scalbnf.c (scalbnf): Likewise.
* sysdeps/ieee754/ldbl-128/s_scalbnl.c (scalbnl): Likewise.
* sysdeps/ieee754/ldbl-128ibm/s_scalbnl.c (scalbnl): Remove
long_double_symbol calls.
* sysdeps/ieee754/ldbl-64-128/s_scalbnl.c (scalbnl): Likewise.
* sysdeps/ieee754/ldbl-opt/s_ldexpl.c (__ldexpl_2): Define as
strong alias of __ldexpl.
(scalbnl): Define using long_double_symbol.
* sysdeps/m68k/m680x0/fpu/s_scalbn.c (__CONCATX(scalbn,suffix)):
Remove alias.
* sysdeps/sparc/sparc64/soft-fp/s_scalbnl.c (scalbnl): Likewise.
* sysdeps/x86_64/fpu/s_scalbnl.S (scalbnl): Likewise.
* math/libm-test.inc (scalbn_test_data): Add errno expectations.
(scalbln_test_data): Add more errno expectations.
The ldbl-128 and ldbl-128ibm expm1l implementations have code to
handle +Inf and finite arguments above an overflow threshold. Since
they now use __expl for large positive arguments to fix other
problems, this code is unreachable; this patch removes it.
Tested for mips64 and powerpc.
[BZ #16415]
* sysdeps/ieee754/ldbl-128/s_expm1l.c (maxlog): Remove variable.
(__expm1l): Remove code to handle positive infinity and overflow.
* sysdeps/ieee754/ldbl-128ibm/s_expm1l.c (maxlog): Remove
variable.
(__expm1l): Remove code to handle positive infinity and overflow.
math.h incorrectly declares various functions for XSI POSIX 2001 and
2008 editions. gamma was removed in the 2001 edition but is still
declared, along with gammaf and gammal which were never standard
functions. isnan is still declared as a function, along with isnanf
and isnanl which were never standard functions, although in 2001 the
function was replaced by the type-generic macro. scalbf and scalbl
are declared although never standard, and scalb was removed in the
2008 edition but is still declared. The scalb type-generic macro in
tgmath.h shouldn't be present for any POSIX version, since POSIX never
had such a type-generic macro.
This patch disables all those declarations in the relevant cases (as a
minimal fix, it leaves them enabled for __USE_MISC). For the matter
of declaring scalb but not scalbf or scalbl for the 2001 edition, a
new macro __MATH_DECLARING_DOUBLE is added, defined by math.h around
includes of bits/mathcalls.h, for bits/mathcalls.h to use to test
which type's functions are being declared.
Tested for x86_64 and x86 (testsuite, and that installed stripped
shared libraries are unchanged by the patch).
[BZ #18967]
* math/math.h (__MATH_DECLARING_DOUBLE): New macro. Define and
undefine around includes of <bits/mathcalls.h>.
* math/bits/mathcalls.h [!__USE_MISC && __USE_XOPEN2K] (isnan): Do
not declare function.
[!__USE_MISC && __USE_XOPEN2K] (gamma): Likewise.
[!__USE_MISC && (!__MATH_DECLARING_DOUBLE || __USE_XOPEN2K8)]
(scalb): Likewise.
* math/tgmath.h [!__USE_MISC && __USE_XOPEN_EXTENDED] (scalb): Do
not define macro.
* conform/Makefile (test-xfail-XOPEN2K/math.h/conform): Remove
variable.
(test-xfail-XOPEN2K/tgmath.h/conform): Likewise.
(test-xfail-XOPEN2K8/math.h/conform): Likewise.
(test-xfail-XOPEN2K8/tgmath.h/conform): Likewise.
The ldbl-128ibm implementation of nearbyintl wrongly uses signaling
comparisons such as "if (fabs (u.d[0].d) < TWO52)" on arguments that
might be NaNs, when "invalid" exceptions should not be raised. (For
hard float, this issue may be hidden by
<https://gcc.gnu.org/bugzilla/show_bug.cgi?id=58684>, powerpc GCC
wrongly only using unordered comparison instructions.) This patch
fixes this by just returning the argument if it is not finite (because
of the arbitrary value of the low part of a NaN in IBM long double,
there are quite a lot of comparisons that could end up involving a NaN
when the argument to nearbyintl is a NaN, so excluding NaN arguments
at the start is the simplest and safest fix).
Tested for powerpc-nofpu, where it removes failures for spurious
"invalid" exceptions from nearbyintl.
[BZ #18857]
* sysdeps/ieee754/ldbl-128ibm/s_nearbyintl.c (__nearbyintl): Just
return non-finite argument without doing ordered comparisons on
it.
Bug 16296 notes that fegetround is a pure function and should be
marked as such in fenv.h. This patch implements that.
Tested for x86_64 and x86 (testsuite, and that installed stripped
shared libraries are unchanged by this patch).
[BZ #16296]
* math/fenv.h (fegetround): Use __attribute_pure__.
* include/fenv.h (__fegetround): Likewise.
Similar to various other bugs in this area, ctan and ctanh can fail to
raise the underflow exception for some cases of results that are tiny
and inexact. This patch forces the exception in a similar way to
previous fixes.
Tested for x86_64 and x86.
[BZ #18595]
* math/s_ctan.c (__ctan): Force underflow exception for results
whose real or imaginary part has small absolute value.
* 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: Do not allow missing underflow for ctan
and ctanh. Add more tests of ctan and ctanh.
Bug 15918 points out that the handling of infinities in hypotf can be
simplified: it's enough to return the absolute value of the infinite
argument without first comparing it to the other argument and possibly
returning that other argument's absolute value. This patch makes that
cleanup (which should not change how hypotf behaves on any input).
Tested for x86_64.
[BZ #15918]
* sysdeps/ieee754/flt-32/e_hypotf.c (__ieee754_hypotf): Simplify
handling of cases where one argument is an infinity.
On i386, the double version of exp10 can miss underflow exceptions if
the result is in the subnormal range for double but the last 11 bits
of the 64-bit extended-precision mantissa happen to be zero. This
patch forces the exception in a similar way to previous fixes.
As with the exp2 and exp fixes, the exp10f changes may in fact not be
needed to ensure underflow exceptions, but are included for
consistency and to fix the exp10 part of bug 18875 by ensuring that
excess range and precision is removed from underflowing return values.
Tested for x86_64 and x86.
[BZ #18875]
[BZ #18966]
* sysdeps/i386/fpu/e_exp10.S (dbl_min): New object.
(MO): New macro.
(__ieee754_exp10): For small results, force underflow exception
and remove excess range and precision from return value.
* sysdeps/i386/fpu/e_exp10f.S (flt_min): New object.
(MO): New macro.
(__ieee754_exp10f): For small results, force underflow exception
and remove excess range and precision from return value.
* math/auto-libm-test-in: Add more tests of exp10.
* math/auto-libm-test-out: Regenerated.
On i386, the double version of exp can miss underflow exceptions if
the result is in the subnormal range for double but the last 11 bits
of the 64-bit extended-precision mantissa happen to be zero. This
patch forces the exception in a similar way to previous fixes.
As with the exp2 fixes, the expf changes may in fact not be needed to
ensure underflow exceptions, but are included for consistency and to
fix the exp part of bug 18875 by ensuring that excess range and
precision is removed from underflowing return values.
Tested for x86_64 and x86.
[BZ #18875]
[BZ #18961]
* sysdeps/i386/fpu/e_exp.S (dbl_min): New object.
(MO): New macro.
(__ieee754_exp): For small results, force underflow exception and
remove excess range and precision from return value.
(__exp_finite): Likewise.
* sysdeps/i386/fpu/e_expf.S (flt_min): New object.
(MO): New macro.
(__ieee754_expf): For small results, force underflow exception and
remove excess range and precision from return value.
(__expf_finite): Likewise.
* math/auto-libm-test-in: Add more tests of exp.
* math/auto-libm-test-out: Regenerated.
Various exp2 implementations in glibc can miss underflow exceptions
when the scaling down part of the calculation is exact (or, in the x86
case, when the conversion from extended precision to the target
precision is exact). This patch forces the exception in a similar way
to previous fixes.
The x86 exp2f changes may in fact not be needed for this purpose -
it's likely to be the case that no argument of type float has an exp2
result so close to an exact subnormal float value that it equals that
value when rounded to 64 bits (even taking account of variation
between different x86 implementations). However, they are included
for consistency with the changes to exp2 and so as to fix the exp2f
part of bug 18875 by ensuring that excess range and precision is
removed from underflowing return values.
Tested for x86_64, x86 and mips64.
[BZ #16521]
[BZ #18875]
* math/e_exp2l.c (__ieee754_exp2l): Force underflow exception for
small results.
* sysdeps/i386/fpu/e_exp2.S (dbl_min): New object.
(MO): New macro.
(__ieee754_exp2): For small results, force underflow exception and
remove excess range and precision from return value.
* sysdeps/i386/fpu/e_exp2f.S (flt_min): New object.
(MO): New macro.
(__ieee754_exp2f): For small results, force underflow exception
and remove excess range and precision from return value.
* sysdeps/i386/fpu/e_exp2l.S (ldbl_min): New object.
(MO): New macro.
(__ieee754_exp2l): Force underflow exception for small results.
* sysdeps/ieee754/dbl-64/e_exp2.c (__ieee754_exp2): Likewise.
* sysdeps/ieee754/flt-32/e_exp2f.c (__ieee754_exp2f): Likewise.
* sysdeps/x86_64/fpu/e_exp2l.S (ldbl_min): New object.
(MO): New macro.
(__ieee754_exp2l): Force underflow exception for small results.
* math/auto-libm-test-in: Add more tests or exp2.
* math/auto-libm-test-out: Regenerated.
If you pass in a path that fails to be opened, then output_path is set to
NULL, and an error is flagged. Then at the end, we use both of those:
cannot write output files to `(null)': No such file or directory
Tweak the message to use the user's input when output_path is NULL.
It was noted in
<https://sourceware.org/ml/libc-alpha/2012-09/msg00305.html> that the
bits/*.h naming scheme should only be used for installed headers.
This patch renames bits/atomic.h to atomic-machine.h to follow that
convention.
This is the only change in this series that needs to change the
filename rather than simply removing a directory level (because both
atomic.h and bits/atomic.h exist at present).
Tested for x86_64 (testsuite, and that installed stripped shared
libraries are unchanged by the patch).
[BZ #14912]
* sysdeps/aarch64/bits/atomic.h: Move to ...
* sysdeps/aarch64/atomic-machine.h: ...here.
(_AARCH64_BITS_ATOMIC_H): Rename macro to
_AARCH64_ATOMIC_MACHINE_H.
* sysdeps/alpha/bits/atomic.h: Move to ...
* sysdeps/alpha/atomic-machine.h: ...here.
* sysdeps/arm/bits/atomic.h: Move to ...
* sysdeps/arm/atomic-machine.h: ...here. Update comments.
* bits/atomic.h: Move to ...
* sysdeps/generic/atomic-machine.h: ...here.
(_BITS_ATOMIC_H): Rename macro to _ATOMIC_MACHINE_H.
* sysdeps/i386/bits/atomic.h: Move to ...
* sysdeps/i386/atomic-machine.h: ...here.
* sysdeps/ia64/bits/atomic.h: Move to ...
* sysdeps/ia64/atomic-machine.h: ...here.
* sysdeps/m68k/coldfire/bits/atomic.h: Move to ...
* sysdeps/m68k/coldfire/atomic-machine.h: ...here.
(_BITS_ATOMIC_H): Rename macro to _ATOMIC_MACHINE_H.
* sysdeps/m68k/m680x0/m68020/bits/atomic.h: Move to ...
* sysdeps/m68k/m680x0/m68020/atomic-machine.h: ...here.
* sysdeps/microblaze/bits/atomic.h: Move to ...
* sysdeps/microblaze/atomic-machine.h: ...here.
* sysdeps/mips/bits/atomic.h: Move to ...
* sysdeps/mips/atomic-machine.h: ...here.
(_MIPS_BITS_ATOMIC_H): Rename macro to _MIPS_ATOMIC_MACHINE_H.
* sysdeps/powerpc/bits/atomic.h: Move to ...
* sysdeps/powerpc/atomic-machine.h: ...here. Update comments.
* sysdeps/powerpc/powerpc32/bits/atomic.h: Move to ...
* sysdeps/powerpc/powerpc32/atomic-machine.h: ...here. Update
comments. Include <atomic-machine.h> instead of <bits/atomic.h>.
* sysdeps/powerpc/powerpc64/bits/atomic.h: Move to ...
* sysdeps/powerpc/powerpc64/atomic-machine.h: ...here. Include
<atomic-machine.h> instead of <bits/atomic.h>.
* sysdeps/s390/bits/atomic.h: Move to ...
* sysdeps/s390/atomic-machine.h: ...here.
* sysdeps/sparc/sparc32/bits/atomic.h: Move to ...
* sysdeps/sparc/sparc32/atomic-machine.h: ...here.
(_BITS_ATOMIC_H): Rename macro to _ATOMIC_MACHINE_H.
* sysdeps/sparc/sparc32/sparcv9/bits/atomic.h: Move to ...
* sysdeps/sparc/sparc32/sparcv9/atomic-machine.h: ...here.
* sysdeps/sparc/sparc64/bits/atomic.h: Move to ...
* sysdeps/sparc/sparc64/atomic-machine.h: ...here.
* sysdeps/tile/bits/atomic.h: Move to ...
* sysdeps/tile/atomic-machine.h: ...here.
* sysdeps/tile/tilegx/bits/atomic.h: Move to ...
* sysdeps/tile/tilegx/atomic-machine.h: ...here. Include
<sysdeps/tile/atomic-machine.h> instead of
<sysdeps/tile/bits/atomic.h>.
(_BITS_ATOMIC_H): Rename macro to _ATOMIC_MACHINE_H.
* sysdeps/tile/tilepro/bits/atomic.h: Move to ...
* sysdeps/tile/tilepro/atomic-machine.h: ...here. Include
<sysdeps/tile/atomic-machine.h> instead of
<sysdeps/tile/bits/atomic.h>.
(_BITS_ATOMIC_H): Rename macro to _ATOMIC_MACHINE_H.
* sysdeps/unix/sysv/linux/arm/bits/atomic.h: Move to ...
* sysdeps/unix/sysv/linux/arm/atomic-machine.h: ...here. Include
<sysdeps/arm/atomic-machine.h> instead of
<sysdeps/arm/bits/atomic.h>.
* sysdeps/unix/sysv/linux/hppa/bits/atomic.h: Move to ...
* sysdeps/unix/sysv/linux/hppa/atomic-machine.h: ...here.
(_BITS_ATOMIC_H): Rename macro to _ATOMIC_MACHINE_H.
* sysdeps/unix/sysv/linux/m68k/coldfire/bits/atomic.h: Move to ...
* sysdeps/unix/sysv/linux/m68k/coldfire/atomic-machine.h: ...here.
(_BITS_ATOMIC_H): Rename macro to _ATOMIC_MACHINE_H.
* sysdeps/unix/sysv/linux/nios2/bits/atomic.h: Move to ...
* sysdeps/unix/sysv/linux/nios2/atomic-machine.h: ...here.
(_NIOS2_BITS_ATOMIC_H): Rename macro to _NIOS2_ATOMIC_MACHINE_H.
* sysdeps/unix/sysv/linux/sh/bits/atomic.h: Move to ...
* sysdeps/unix/sysv/linux/sh/atomic-machine.h: ...here.
* sysdeps/x86_64/bits/atomic.h: Move to ...
* sysdeps/x86_64/atomic-machine.h: ...here.
* include/atomic.h: Include <atomic-machine.h> instead of
<bits/atomic.h>.
The ldbl-128 / ldbl-128ibm implementation of lgammal converts (the
floor of minus) non-integer negative arguments to int to determine the
value of signgam. When those values are outside the range of int,
this produces spurious "invalid" exceptions and incorrect values of
signgam. This patch fixes this by instead determining signgam through
comparing half the integer in question to floor of half the integer.
Tested for mips64, x86_64 and x86.
[BZ #18952]
* sysdeps/ieee754/ldbl-128/e_lgammal_r.c (__ieee754_lgammal_r): Do
not convert non-integer negative arguments to int to determine the
value of signgam.
* math/auto-libm-test-in: Add more tests of lgamma.
* math/auto-libm-test-out: Regenerated.
The existing implementations of lgamma functions (except for the ia64
versions) use the reflection formula for negative arguments. This
suffers large inaccuracy from cancellation near zeros of lgamma (near
where the gamma function is +/- 1).
This patch fixes this inaccuracy. For arguments above -2, there are
no zeros and no large cancellation, while for sufficiently large
negative arguments the zeros are so close to integers that even for
integers +/- 1ulp the log(gamma(1-x)) term dominates and cancellation
is not significant. Thus, it is only necessary to take special care
about cancellation for arguments around a limited number of zeros.
Accordingly, this patch uses precomputed tables of relevant zeros,
expressed as the sum of two floating-point values. The log of the
ratio of two sines can be computed accurately using log1p in cases
where log would lose accuracy. The log of the ratio of two gamma(1-x)
values can be computed using Stirling's approximation (the difference
between two values of that approximation to lgamma being computable
without computing the two values and then subtracting), with
appropriate adjustments (which don't reduce accuracy too much) in
cases where 1-x is too small to use Stirling's approximation directly.
In the interval from -3 to -2, using the ratios of sines and of
gamma(1-x) can still produce too much cancellation between those two
parts of the computation (and that interval is also the worst interval
for computing the ratio between gamma(1-x) values, which computation
becomes more accurate, while being less critical for the final result,
for larger 1-x). Because this can result in errors slightly above
those accepted in glibc, this interval is instead dealt with by
polynomial approximations. Separate polynomial approximations to
(|gamma(x)|-1)(x-n)/(x-x0) are used for each interval of length 1/8
from -3 to -2, where n (-3 or -2) is the nearest integer to the
1/8-interval and x0 is the zero of lgamma in the relevant half-integer
interval (-3 to -2.5 or -2.5 to -2).
Together, the two approaches are intended to give sufficient accuracy
for all negative arguments in the problem range. Outside that range,
the previous implementation continues to be used.
Tested for x86_64, x86, mips64 and powerpc. The mips64 and powerpc
testing shows up pre-existing problems for ldbl-128 and ldbl-128ibm
with large negative arguments giving spurious "invalid" exceptions
(exposed by newly added tests for cases this patch doesn't affect the
logic for); I'll address those problems separately.
[BZ #2542]
[BZ #2543]
[BZ #2558]
* sysdeps/ieee754/dbl-64/e_lgamma_r.c (__ieee754_lgamma_r): Call
__lgamma_neg for arguments from -28.0 to -2.0.
* sysdeps/ieee754/flt-32/e_lgammaf_r.c (__ieee754_lgammaf_r): Call
__lgamma_negf for arguments from -15.0 to -2.0.
* sysdeps/ieee754/ldbl-128/e_lgammal_r.c (__ieee754_lgammal_r):
Call __lgamma_negl for arguments from -48.0 or -50.0 to -2.0.
* sysdeps/ieee754/ldbl-96/e_lgammal_r.c (__ieee754_lgammal_r):
Call __lgamma_negl for arguments from -33.0 to -2.0.
* sysdeps/ieee754/dbl-64/lgamma_neg.c: New file.
* sysdeps/ieee754/dbl-64/lgamma_product.c: Likewise.
* sysdeps/ieee754/flt-32/lgamma_negf.c: Likewise.
* sysdeps/ieee754/flt-32/lgamma_productf.c: Likewise.
* sysdeps/ieee754/ldbl-128/lgamma_negl.c: Likewise.
* sysdeps/ieee754/ldbl-128/lgamma_productl.c: Likewise.
* sysdeps/ieee754/ldbl-128ibm/lgamma_negl.c: Likewise.
* sysdeps/ieee754/ldbl-128ibm/lgamma_productl.c: Likewise.
* sysdeps/ieee754/ldbl-96/lgamma_negl.c: Likewise.
* sysdeps/ieee754/ldbl-96/lgamma_product.c: Likewise.
* sysdeps/ieee754/ldbl-96/lgamma_productl.c: Likewise.
* sysdeps/generic/math_private.h (__lgamma_negf): New prototype.
(__lgamma_neg): Likewise.
(__lgamma_negl): Likewise.
(__lgamma_product): Likewise.
(__lgamma_productl): Likewise.
* math/Makefile (libm-calls): Add lgamma_neg and lgamma_product.
* math/auto-libm-test-in: Add more tests of lgamma.
* math/auto-libm-test-out: Regenerated.
* sysdeps/i386/fpu/libm-test-ulps: Update.
* sysdeps/x86_64/fpu/libm-test-ulps: Likewise.
Topic: strptime supports a %z input field descriptor, which parses a
time zone offset from UTC time into the broken-out time field tm_gmtoff.
Problems:
1) In the current implementation, the minutes portion calculation is
correct only for minutes evenly divisible by 3. This is because the
minutes value is converted to decimal time, but inadequate precision
leads to rounding which calculates results that are too low for
some values.
For example, due to rounding, a +1159 offset string results in an
incorrect tm_gmtoff of 43128 (== 11 * 3600 + 58.8 * 60) seconds,
instead of 43140 (== 11 * 3600 + 59 * 60) seconds. In contrast,
a +1157 offset (minutes divisible by 3) does not cause the bug,
and results in a correct tm_gmtoff of 43020.
2) strptime's %z specifier will not parse time offsets less than
-1200 or greater than +1200, or if only hour digits are present, less
than -12 or greater than +12. It will return NULL for offsets outside
that range. These limits do not meet historical and modern use cases:
* Present day exceeds the +1200 limit:
- Pacific/Auckland (New Zealand) summer time is +1300.
- Pacific/Kiritimati (Christmas Island) is +1400.
- Pacific/Apia (Samoa) summer time is +1400.
* Historical offsets exceeded +1500/-1500.
* POSIX supports -2459 to +2559.
* Offsets up to +/-9959 may occasionally be useful.
* Paul Eggert's notes provide additional detail:
- https://sourceware.org/ml/libc-alpha/2014-12/msg00068.html
- https://sourceware.org/ml/libc-alpha/2014-12/msg00072.html
3) tst-strptime2, part of the 'make check' test suite, does not test
for the above problems.
Corrective actions:
1) In time/strptime_l.c, calculate the offset from the hour and
minute portions directly, without the rounding errors introduced by
decimal time.
2) Remove the +/-1200 range limit, permitting strptime to parse offsets
from -9959 through +9959.
3) Add zone offset values to time/tst-strptime2.c.
* Test minutes evenly divisible by three (+1157) and not evenly
divisible by three (+1158 and +1159).
* Test offsets near the old and new range limits (-1201, -1330, -2459,
-2500, -99, -9959, +1201, +1330, +1400, +1401, +2559, +2600, +99,
and +9959)
The revised strptime passes all old and new tst-strptime2 tests.
This patch fixes the default wordsize-32 mmap implementation offset
calculation for negative values. Current code uses signed shift
operation to calculate the multiple size to use with syscall and
it is implementation defined. Change it to use a division base
on mmap page size (default being as before, 4096).
Tested on armv7hf.
[BZ #18877]
* posix/Makefile (tests): Add tst-mmap-offset.
* posix/tst-mmap.c: New file.
* sysdeps/unix/sysv/linux/generic/wordsize-32/mmap.c (__mmap): Fix
offset calculation for negative values.
This patch set introduces optimized string, wcsmbs and memory functions for
S390/S390x. The functions are accelerated by the usage of the new z13 vector
instructions.
The Principles of Operations manual for IBM z13 is publically available:
http://publibfi.boulder.ibm.com/epubs/pdf/dz9zr010.pdf
The support for these instructions in assembler was introduced by commits:
-"[Committed] S/390: Add support for IBM z13."
(https://sourceware.org/ml/binutils/2015-01/msg00197.html)
-"[Committed] S/390: Add more IBM z13 instructions"
(https://sourceware.org/ml/binutils/2015-03/msg00088.html)
The first patches do preparation for the latter optimization patches.
The floating point exception handling - fetestexcept(), ... - is fixed and
the platform and hwcap strings are extended.
The current ifunc routines memset, memcpy and memcmp are refactored and the
ifunc test-framework is now enabled.
A S390 specific configure-check tests if the used binutils supports the new
vector instructions. The optimized functions are provided via ifunc if the
binutils supports the vector instructions. Otherwise a message is dumped to
configure output and only the currently used common code functions are
available.
The optimized functions are implemented in common for s390-32 and s390-64
and the few differences are handled via #ifdef.
The ifunc-resolvers are defined in files sysdeps/s390/multiarch/<func>.c,
which choose either the current implementation __<func>_c() or the vector
implementation __<func>_vx() depending on the HWCAP_S390_VX flag bit in
AT_HWCAP field. If the bit is set, the hardware and the kernel are supporting
vector registers and instructions. If the used binutils lacks vector-support,
then the default implementation in string or wcsmbs directory is included
here instead.
The file sysdeps/s390/multiarch/<func>-c.c includes the current implementation
and defines the function name __<func>_c.
The assembler files sysdeps/s390/multiarch/<func>-vx.S with the vector
instructions are using the directive '.machine "z13"' to allow building glibc
without option '-march=z13'. Additionally the directive '.machinemode
"zarch_nohighgprs"' is needed for the 31bit glibc. This mode does not set the
highgprs flag in ELF header, which would lead to an unloadable libc on a 31bit
kernel.
The most optimized string functions are structured in the same way:
The first 16 bytes of the string is loaded unaligned via vlbb - vector load
to block boundary (e.g. 4k). This instruction loads 16 bytes if possible.
In case of a page cross, it only loads the last bytes of the current page
without a segmentation fault.
Afterwards these first part of string is processed. If e.g. for strlen the end
of string is reached within this first part, the function returns. Otherwise
the pointer is aligned to 16 byte, so i can load a full vector register with vl
without checking for a page cross. Afterwards the first part of string is
processed. If e.g. for strlen the end of string is reached within this first
part, the function returns. Otherwise the pointer is aligned to 16 byte, so
a full vector register can be loaded with vl - vector load - without checking
for a page cross. The remaining string is processed in a four times unrolled
loop, because benchmark results measured improvements compared to a non
unrolled loop.
The optimized wide string functions can only handle 4byte aligned string
pointers. Although a wchar_t pointer should always be 4byte aligned, the most
current common code wide string functions can handle non aligned strings.
Thus the optimized functions will fall back to the common code functions in
case of a non aligned wide string to behave the same as before this patch.
Some string tests can test the string and the wide string version of a function.
The remaining ones are extended and new wide string tests are added.
This is the same in case of the benchtests.
ChangeLog:
* NEWS: New item for IBM z13 string optimizations.
The csqrt implementations in glibc can miss underflow exceptions when
the real or imaginary part of the result becomes tiny in the course of
scaling down (in particular, multiplication by 0.5) and that scaling
is exact although the relevant part of the mathematical result isn't.
This patch forces the exception in a similar way to previous fixes.
Tested for x86_64 and x86.
[BZ #18370]
* math/s_csqrt.c (__csqrt): Force underflow exception for results
whose real or imaginary part has small absolute value.
* 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.
lang_lib (which reflects ISO 639-2/B (bibliographic) codes) and
lang_term (which reflects ISO 639-2/T (terminology) codes) should be
identical except for those languages for which ISO 639-2 specifies
separate bibliographic/terminology values.
I used this Library of Congress page as the source:
http://www.loc.gov/standards/iso639-2/php/code_list.php
The csqrt functions scale up small arguments to avoid underflows when
calling hypot functions. However, even when hypot does not underflow,
a subsequent calculation of 0.5 * hypot can underflow. This patch
duly increases the threshold and scale factor to avoid such underflows
as well.
Tested for x86_64, x86 and mips64.
[BZ #18823]
* math/s_csqrt.c (__csqrt): Increase threshold and scale factor
for scaling up small arguments.
* 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.
I think the last clause of the conditional,
|| __n <= __bos (__dest)
may be backward. The code should call the runtime-checking function
if __n is not constant, or if __n is known to be LARGER than the size
of the destination.
Various fma implementations have logic that, when computing fma (x, y,
z) where z is large (so care needs taking to avoid internal overflow)
but x * y is small, scale x * y up instead of down to avoid internal
underflows resulting from scaling down. (In these cases, x * y is
small enough that only its sign actually matters rather than the exact
value.)
The threshold for scaling up instead of down was correct for "if the
unscaled values were multiplied, the low part of the multiplication
could underflow", and the scaling was sufficient to ensure that the
low part of the multiplication did not underflow (given that cases of
very small x * y - less than half the least subnormal - were
previously dealt with). However, the choice in the functions wasn't
between scaling up or no scaling, but between scaling up and scaling
down (scaling down actually being needed when x * y isn't so small
compared to z and so the exact value does matter). Thus a larger
threshold is needed to ensure that scaling down doesn't produce values
the multiplication of whose low parts underflows. This patch
increases the thresholds accordingly.
Tested for x86_64, x86 and mips64 (with the MIPS version of s_fmal.c
removed so that the ldbl-128 version gets tested instead of the
soft-fp one).
[BZ #18824]
* sysdeps/ieee754/dbl-64/s_fma.c (__fma): Increase threshold for
scaling x * y up instead of down.
* sysdeps/ieee754/ldbl-128/s_fmal.c (__fmal): Likewise.
* sysdeps/ieee754/ldbl-96/s_fmal.c (__fmal): Likewise.
* math/auto-libm-test-in: Add more tests of fma.
* math/auto-libm-test-out: Regenerated.
Similar to various other bugs in this area, some tanh 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 #16520]
* sysdeps/ieee754/dbl-64/s_tanh.c: Include <float.h>.
(__tanh): Force underflow exception for arguments with small
absolute value.
* sysdeps/ieee754/flt-32/s_tanhf.c: Include <float.h>.
(__tanhf): Force underflow exception for arguments with small
absolute value.
* sysdeps/ieee754/ldbl-128/s_tanhl.c: Include <float.h>.
(__tanhl): Force underflow exception for arguments with small
absolute value.
* sysdeps/ieee754/ldbl-128ibm/s_tanhl.c: Include <float.h>.
(__tanhl): Force underflow exception for arguments with small
absolute value.
* sysdeps/ieee754/ldbl-96/s_tanhl.c: Include <float.h>.
(__tanhl): Force underflow exception for arguments with small
absolute value.
* math/auto-libm-test-in: Add more tests of tanh.
* math/auto-libm-test-out: Regenerated.
* sysdeps/i386/fpu/libm-test-ulps: Update.
https://sourceware.org/bugzilla/show_bug.cgi?id=18778
If dlopen fails to load an object that has triggered loading libpthread it
causes ld.so to unload libpthread because its DF_1_NODELETE flags has been
forcefully cleared. The next call to __rtdl_unlock_lock_recursive will crash
since pthread_mutex_unlock no longer exists.
This patch moves l->l_flags_1 &= ~DF_1_NODELETE out of loop through all loaded
libraries and performs the action only on inconsistent one.
[BZ #18778]
* elf/Makefile (tests): Add Add tst-nodelete2.
(modules-names): Add tst-nodelete2mod.
(tst-nodelete2mod.so-no-z-defs): New.
($(objpfx)tst-nodelete2): Likewise.
($(objpfx)tst-nodelete2.out): Likewise.
(LDFLAGS-tst-nodelete2): Likewise.
* elf/dl-close.c (_dl_close_worker): Move DF_1_NODELETE clearing
out of loop through all loaded libraries.
* elf/tst-nodelete2.c: New file.
* elf/tst-nodelete2mod.c: Likewise.
ldbl-128ibm tanhl uses a too-small threshold to decide when to return
+/-1, resulting in large errors. This patch changes it to a more
appropriate threshold (the requirement is for 2*exp(-2|x|) to be small
in terms of ulps of 1).
Tested for x86_64, x86 and powerpc.
[BZ #18790]
* sysdeps/ieee754/ldbl-128ibm/s_tanhl.c (__tanhl): Increase
threshold for returning +/- 1.
* math/auto-libm-test-in: Add more tests of tanh.
* math/auto-libm-test-out: Regenerated.
* sysdeps/i386/fpu/libm-test-ulps: Update.
ldbl-128ibm sinhl uses a too-big threshold to decide when to return
the argument, resulting in large errors. This patch fixes it to use a
more appropriate threshold.
Tested for x86_64, x86 and powerpc.
[BZ #18789]
* sysdeps/ieee754/ldbl-128ibm/e_sinhl.c (__ieee754_sinhl): Use
smaller threshold for returning the argument.
* math/auto-libm-test-in: Add more tests of sinh.
* math/auto-libm-test-out: Regenerated.
* sysdeps/i386/fpu/libm-test-ulps: Update.
The attached change fixes the miscompilation of sched_setaffinity() on
hppa. This is an old problem that was fixed on other architectures using
a similar approach to the attached change. See:
https://sourceware.org/ml/libc-hacker/2004-04/msg00016.html
Build tested on trunk. Patch has been applied to debian glibc for some time.
As noted in the bug, the asm operands need to be copied to register
variables to avoid operand reloads in the principal asm of the macro.
See the arm implementation for reference. Otherwise we get:
../sysdeps/unix/sysv/linux/hppa/bits/atomic.h:68:6: error:
can't find a register in class 'R1_REGS' while reloading 'asm'
Build tested on trunk with gcc-4.8. Similar patch has been tested
with 2.19 on Debian hppa-unknown-linux-gnu.
Similar to various other bugs in this area, some tan 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 #16517]
* sysdeps/ieee754/dbl-64/s_tan.c: Include <float.h>.
(tan): Force underflow exception for arguments with small absolute
value.
* sysdeps/ieee754/flt-32/k_tanf.c: Include <float.h>.
(__kernel_tanf): Force underflow exception for arguments with
small absolute value.
* sysdeps/ieee754/ldbl-128/k_tanl.c: Include <float.h>.
(__kernel_tanl): Force underflow exception for arguments with
small absolute value.
* sysdeps/ieee754/ldbl-128ibm/k_tanl.c: Include <float.h>.
(__kernel_tanl): Force underflow exception for arguments with
small absolute value.
* sysdeps/ieee754/ldbl-96/k_tanl.c: Include <float.h>.
(__kernel_tanl): Force underflow exception for arguments with
small absolute value.
* math/auto-libm-test-in: Add more tests of tan.
* math/auto-libm-test-out: Regenerated.
Similar to various other bugs in this area, some sinh 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 #16519]
* sysdeps/ieee754/dbl-64/e_sinh.c: Include <float.h>.
(__ieee754_sinh): Force underflow exception for arguments with
small absolute value.
* sysdeps/ieee754/flt-32/e_sinhf.c: Include <float.h>.
(__ieee754_sinhf): Force underflow exception for arguments with
small absolute value.
* sysdeps/ieee754/ldbl-128/e_sinhl.c: Include <float.h>.
(__ieee754_sinhl): Force underflow exception for arguments with
small absolute value.
* sysdeps/ieee754/ldbl-128ibm/e_sinhl.c: Include <float.h>.
(__ieee754_sinhl): Force underflow exception for arguments with
small absolute value.
* sysdeps/ieee754/ldbl-96/e_sinhl.c: Include <float.h>.
(__ieee754_sinhl): Force underflow exception for arguments with
small absolute value.
* math/auto-libm-test-in: Add more tests of sinh.
* math/auto-libm-test-out: Regenerated.
* sysdeps/i386/fpu/libm-test-ulps: Update.
In the "Kill regexp.h" thread, Joseph dug up more accurate information
about exactly which editions of the Single Unix Standard included and
deprecated this header.
