This patch adds ULPs for the float128 type, updates the abilist for libc
and libm, and adds the files bits/floatn.h and float128-abi.h, in order to
enable the new type for powerpc64le.
This patch also adds the implementation of sqrtf128 for powerpc64le, since
it is not implemented in libgcc. The sfp-machine.h header is taken from
libgcc.
Tested for powerpc64le (GCC 6.2 and GCC 7.1), powerpc64 and s390x.
* manual/math.texi (Mathematics): Mention the enabling of float128
for powerpc64le.
* sysdeps/powerpc/bits/floatn.h: New file.
* sysdeps/powerpc/fpu/libm-test-ulps: Regenerated.
* sysdeps/powerpc/fpu/math_private.h:
(__ieee754_sqrtf128): New inline override.
* sysdeps/powerpc/powerpc64le/Implies-before: New file.
* sysdeps/powerpc/powerpc64le/Makefile: New file.
* sysdeps/powerpc/powerpc64le/fpu/e_sqrtf128.c: New file.
* sysdeps/powerpc/powerpc64le/fpu/sfp-machine.h: New file.
* sysdeps/powerpc/powerpc64le/power9/fpu/e_sqrtf128.c: New file.
* sysdeps/unix/sysv/linux/powerpc/powerpc64/libc-le.abilist:
Updated.
* sysdeps/unix/sysv/linux/powerpc/powerpc64/libm-le.abilist:
Likewise.
* sysdeps/unix/sysv/linux/powerpc/powerpc64le/float128-abi.h:
New file.
Support for powerpc64le requires POWER8 or newer processors. Builds for
older processors are not tested. Require at least POWER8 to avoid
unintentional builds.
* sysdeps/powerpc/powerpc64le/configure.ac: Check for POWER8.
* sysdeps/powerpc/powerpc64le/configure: Update.
On powerpc64le, support for float128 will be enabled, which requires some
compiler features to be present. This patch adds a configure test to check
for such features, which are provided for powerpc64le since GCC 6.2.
Tested for powerpc64 and powerpc64le.
* INSTALL: Regenerate.
* manual/install.texi (Recommended Tools for Compilation): Mention
the powerpc64le-specific requirement in the manual.
* sysdeps/powerpc/powerpc64le/configure.ac: New file with checks
for the compiler features required for building float128.
* sysdeps/powerpc/powerpc64le/configure: New, auto-generated file.
According to ISO C11, section 6.5.3.3 "Unary arithmetic operators", the
result of the ~ operator is the bitwise complement of its (promoted)
operand.
This can lead to a comparison of a char with another integer type.
Tested on powerpc, powerpc64 and powerpc64le.
* malloc/tst-alloc_buffer.c (test_misaligned): Cast to char
before comparing with another char.
The .symver directive on common symbol just creates a new common symbol,
not an alias and the newer assembler with the bug fix for
https://sourceware.org/bugzilla/show_bug.cgi?id=21661
will issue an error. Before the fix, we got
$ readelf -sW libc.so | grep "loc[12s]"
5109: 00000000003a0608 8 OBJECT LOCAL DEFAULT 36 loc1
5188: 00000000003a0610 8 OBJECT LOCAL DEFAULT 36 loc2
5455: 00000000003a0618 8 OBJECT LOCAL DEFAULT 36 locs
6575: 00000000003a05f0 8 OBJECT GLOBAL DEFAULT 36 locs@GLIBC_2.2.5
7156: 00000000003a05f8 8 OBJECT GLOBAL DEFAULT 36 loc1@GLIBC_2.2.5
7312: 00000000003a0600 8 OBJECT GLOBAL DEFAULT 36 loc2@GLIBC_2.2.5
in libc.so. The versioned loc1, loc2 and locs have the wrong addresses.
After the fix, we got
$ readelf -sW libc.so | grep "loc[12s]"
6570: 000000000039e3b8 8 OBJECT GLOBAL DEFAULT 34 locs@GLIBC_2.2.5
7151: 000000000039e3c8 8 OBJECT GLOBAL DEFAULT 34 loc1@GLIBC_2.2.5
7307: 000000000039e3c0 8 OBJECT GLOBAL DEFAULT 34 loc2@GLIBC_2.2.5
[BZ #21666]
* misc/regexp.c (loc1): Add __attribute__ ((nocommon));
(loc2): Likewise.
(locs): Likewise.
Testing with GCC 7 for 32-bit x86 showed some ulps differences,
presumably from variation in when values with excess precision get
spilled to the stack and so lose that precision. This patch updates
the libm-test-ulps files accordingly.
* sysdeps/i386/fpu/libm-test-ulps: Update.
* sysdeps/i386/i686/fpu/multiarch/libm-test-ulps: Likewise.
The libm-test-support code clears exceptions at the end of
test_exceptions. However, it's possible that exceptions get raised as
part of the subsequent checks on the results of the function being
tested, and such exceptions then carry through to the next test run
and can cause that to fail spuriously. In particular, it's possible
for the ulps calculation to raise exceptions in some cases. This
patch puts exception clearing at the end of each of the check_*
functions, alongside errno clearing, to avoid one test affecting the
next in that way.
Tested for x86_64 (in conjunction with float128 patches).
* libm-test-support.c (check_float_internal): Clear exceptions at
end of function.
(check_int): Likewise.
(check_long): Likewise.
(check_bool): Likewise.
(check_longlong): Likewise.
(check_intmax_t): Likewise.
(check_uintmax_t): Likewise.
Building for x86_64 with float128 support, I get a localplt test
failure from lrintf128 calling feraiseexcept.
The problem is that an inline optimized version of feraiseexcept calls
__feraiseexcept_renamed in cases where it doesn't completely expand
inline, and that in turn is redirected to feraiseexcept for a library
call, so meaning the redirection of feraiseexcept to
__GI_feraiseexcept inside libm is lost for that call.
This patch fixes the problem by moving the redirect to an internal
header in the _LIBC case, with the internal header using
__GI_feraiseexcept where appropriate.
Tested for x86_64 (in conjunction with float128 patches).
* sysdeps/x86/fpu/bits/fenv.h [_LIBC] (__feraiseexcept_renamed):
Do not declare.
* sysdeps/x86/fpu/include/bits/fenv.h [_LIBC &&
__USE_EXTERN_INLINES] (__feraiseexcept_renamed): Declare here,
redirected to __GI_feraiseexcept if [SHARED && IS_IN (libm)].
This code:
L(between_2_3):
/* Load as big endian with overlapping loads and bswap to avoid
branches. */
movzwl -2(%rdi, %rdx), %eax
movzwl -2(%rsi, %rdx), %ecx
shll $16, %eax
shll $16, %ecx
movzwl (%rdi), %edi
movzwl (%rsi), %esi
orl %edi, %eax
orl %esi, %ecx
bswap %eax
bswap %ecx
subl %ecx, %eax
ret
needs a saturating subtract because the full register is used.
With this commit, only the lower 24 bits of the register are used,
so a regular subtraction suffices.
The test case change adds coverage for these kinds of bugs.
Similar to the other functions in the strfrom class, strfromf128 calls
__printf_fp in order to convert the floating-point value to characters.
This requires the value of IO_MTSAFE_IO.
Tested for powerpc64le and s390x.
* sysdeps/ieee754/float128/Makefile (CFLAGS-strfromf128.c): Add
$(libio-mtsafe) to get the value of IO_MTSAFE_IO.
When float128 support gets enabled for powerpc64le, the ULP errors for
float128 functions need to be presented in the manual. This patch adds
support for displaying them.
Tested for powerpc64le and s390x.
* manual/libm-err-tab.pl (@all_floats, %suffices, parse_ulps):
Enable generation of float128 entries on the error table.
The file include/float.h uses the macro __GLIBC_USE to test for TS 18661-3
support. Such macro is provided by bits/libc-header-start.h, so include it
to get the definition.
Tested for powerpc64le and s390x.
* include/float.h: Include libc-header-start.h to get the
definition of __GLIBC_USE.
Most significant changes are code simplification and use of doubles for
intermediate values. Also, some rearrangement to move early
non-dependent code later, out of the faster paths.
* sysdeps/ieee754/flt-32/e_powf.c: Optimized implementation utilizing
rearranged code and doubles float types.
sysdeps/powerpc/powerpc64/fpu/multiarch/s_sinf-ppc64.c should fall back to
sysdeps/powerpc/fpu/s_sinf.c not to sysdeps/ieee754/flt-32/s_sinf.c.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_sinf-ppc64.c: Change
s_sinf.c from sysdeps/ieee754/flt-32/ to sysdeps/powerpc/fpu/.
Commit cdfbe5037f added sysconf support
for cache geometries on powerpc, but mishandled errno. For valid input
parameters, sysconf() should not set errno.
* sysdeps/unix/sysv/linux/powerpc/sysconf.c: Remove references
to errno, and simplify remaining related code.
Linux commit ID a4700a26107241cc7b9ac8528b2c6714ff99983d reserved 2 more
bits for the instructions darn (Deliver a Random Number) and scv (System
Call Vectored).
Linux commit ID 6997e57d693b07289694239e52a10d2f02c3a46f reserved
another bit for internal usage.
* sysdeps/powerpc/bits/hwcap.h: Add PPC_FEATURE2_DARN and
PPC_FEATURE2_SCV.
* sysdeps/powerpc/dl-procinfo.c (_dl_powerpc_cap_flags): Add scv
and darn.
This patch adds _Float128 support to the ldbl-96 bits/iscanonical.h,
as needed for x86_64 / x86 / ia64 support of _Float128.
Tested for x86_64 (in conjunction with float128 patches).
* sysdeps/ieee754/ldbl-96/bits/iscanonical.h
[__HAVE_DISTINCT_FLOAT128] (__iscanonicalf128): New macro.
This patch makes math-tests.h, as used to describe support of given
floating-point types for sNaNs, rounding modes and exceptions, handle
distinguishing _Float128 from long double. This is needed for x86_64,
where if building with GCC 6 or earlier there is no __builtin_nansq,
so no way to get a signaling NaN of _Float128 type, so associated
tests cannot be run (although glibc itself works fine, as there is
never any need to create such an sNaN with a built-in function inside
glibc).
Tested for x86_64 (in conjunction with float128 patches).
* sysdeps/generic/math-tests.h: Include <bits/floatn.h>.
(MATH_TESTS_TG): New macro.
(SNAN_TESTS_float128): Likewise.
(ROUNDING_TESTS_float128): Likewise.
(EXCEPTION_TESTS_float128): Likewise.
(SNAN_TESTS): Define using MATH_TESTS_TG.
(ROUNDING_TESTS): Likewise.
(EXCEPTION_TESTS): Likewise.
As with other long double identifiers, float128_private.h has a
redefinition of SET_RESTORE_ROUNDL. However, that redefinition is
broken, since this is a macro with one argument being defined to take
no arguments. This patch fixes the redefinition. (x86_64 needs the
redefinition because SET_RESTORE_ROUNDL only changes the x87 rounding
mode, whereas _Float128 arithmetic uses the SSE rounding mode instead
on x86_64.)
Tested for x86_64 (in conjunction with float128 patches).
* sysdeps/ieee754/float128/float128_private.h
[SET_RESTORE_ROUNDF128] (SET_RESTORE_ROUNDL): Take an argument and
pass it to SET_RESTORE_ROUNDF128.
float128_private.h redefines ieee754.h identifiers ieee854_long_double
and IEEE854_LONG_DOUBLE_BIAS to map them to identifiers from
ieee754_float128.h.
This causes problems when ieee754.h is included after
float128_private.h and it's a version of ieee754.h that also defines
those identifiers; specifically, sysdeps/ieee754/ieee754.h, which
defines those identifiers for the x86 extended format. This patch
fixes this by ensuring an include of ieee754.h from float128_private.h
before the redefinitions.
Tested for x86_64 (in conjunction with float128 patches).
* sysdeps/ieee754/float128/float128_private.h: Include
<ieee754.h>.
The math_private.h macro min_of_type has broken _Float128 handling:
instead of passing its type argument to the key __EXPR_FLT128 macro,
it passes x, which is not a macro argument but whatever variable
called x happens to be visible in the calling function. If that
variable has the wrong type, the wrong one of long double and
_Float128 can get chosen. In particular, this applies to some
_Complex long double functions (where x happens to have type _Complex
long double, resulting in min_of_type returning a _Float128 value when
it should return a long double value). For some reason, this only
caused test failures for me on x86_64 with GCC 6 but not GCC 7 (I
suspect it triggers known bugs in conversions from x86 long double to
_Float128 that are present in GCC 6's soft-fp).
Tested for x86_64 (in conjunction with float128 patches).
* sysdeps/generic/math_private.h (__EXPR_FLT128): Do not apply
typeof to argument passed to __builtin_types_compatible_p.
(min_of_type): Pass type argument, not x, to __EXPR_FLT128.
Various type-generic libm wrapper templates, as used for float128, set
errno but do not include errno.h. I presume they must get an implicit
include from some internal header on powerpc64le; they don't get such
an implicit include on x86_64. This patch adds the missing includes
of errno.h to each such wrapper.
Tested for x86_64 (in conjunction with float128 patches).
* math/w_acos_template.c [__USE_WRAPPER_TEMPLATE]: Include
<errno.h>.
* math/w_acosh_template.c [__USE_WRAPPER_TEMPLATE]: Likewise.
* math/w_asin_template.c [__USE_WRAPPER_TEMPLATE]: Likewise.
* math/w_atanh_template.c [__USE_WRAPPER_TEMPLATE]: Likewise.
* math/w_cosh_template.c [__USE_WRAPPER_TEMPLATE]: Likewise.
* math/w_exp10_template.c [__USE_WRAPPER_TEMPLATE]: Likewise.
* math/w_exp2_template.c [__USE_WRAPPER_TEMPLATE]: Likewise.
* math/w_exp_template.c [__USE_WRAPPER_TEMPLATE]: Likewise.
* math/w_fmod_template.c [__USE_WRAPPER_TEMPLATE]: Likewise.
* math/w_hypot_template.c [__USE_WRAPPER_TEMPLATE]: Likewise.
* math/w_j0_template.c [__USE_WRAPPER_TEMPLATE]: Likewise.
* math/w_j1_template.c [__USE_WRAPPER_TEMPLATE]: Likewise.
* math/w_jn_template.c [__USE_WRAPPER_TEMPLATE]: Likewise.
* math/w_lgamma_r_template.c [__USE_WRAPPER_TEMPLATE]: Likewise.
* math/w_lgamma_template.c [__USE_WRAPPER_TEMPLATE]: Likewise.
* math/w_log10_template.c [__USE_WRAPPER_TEMPLATE]: Likewise.
* math/w_log2_template.c [__USE_WRAPPER_TEMPLATE]: Likewise.
* math/w_log_template.c [__USE_WRAPPER_TEMPLATE]: Likewise.
* math/w_pow_template.c [__USE_WRAPPER_TEMPLATE]: Likewise.
* math/w_remainder_template.c [__USE_WRAPPER_TEMPLATE]: Likewise.
* math/w_sinh_template.c [__USE_WRAPPER_TEMPLATE]: Likewise.
* math/w_sqrt_template.c [__USE_WRAPPER_TEMPLATE]: Likewise.
* math/w_tgamma_template.c [__USE_WRAPPER_TEMPLATE]: Likewise.
Three float128 files still include xlocale.h after it was removed. I
don't know why this didn't cause problems for powerpc64le float128
testing; it did cause problems for my x86_64 float128 testing. This
patch changes the includes to use bits/types/locale_t.h.
Tested for x86_64 (in conjunction with float128 patches).
* sysdeps/ieee754/float128/strtof128_l.c: Include
<bits/types/locale_t.h> instead of <xlocale.h>.
* sysdeps/ieee754/float128/wcstof128.c: Likewise.
* sysdeps/ieee754/float128/wcstof128_l.c: Likewise.
In preparation for the documentation of _FloatN and _FloatNx variants of
the remainder function, this patch changes the descriptions of remainder
and drem, so that remainder is described as primary and drem as an
alternative name for the same functionality.
* manual/arith.texi (Remainder Functions): Describe remainder as
primary and drem as an alternative name. Change the comment on
remainder to ISO, since it is defined in ISO C99.
The macro F128 in stdlib/tst-strtod.h is defined to provide the literal
suffix for _Float128 constants. It uses the macro __f128 (), which is
defined in bits/floatn.h to provide the correct literal suffix depending on
what is provided by the compiler.
However, F128 was not being expanded and only worked correctly, when
compiling with GCC 7 (or greater), since F128 is the literal suffix itself.
This patch adds an additional macro expansion so that the macro F128
expands to the correct literal suffix on older compilers.
* stdlib/tst-strtod.h (MMFUNC): New macro to provide an addition
macro expansion.
(GEN_TEST_STRTOD_FOREACH): Use MMFUNC for _Float128.
* Unicode 10.0.0 Support: Character encoding, character type info, and
transliteration tables are all updated to Unicode 10.0.0, using
generator scripts contributed by Mike FABIAN (Red Hat).
This commit adds fixed-size allocation buffers. The primary use
case is in NSS modules, where dynamically sized data is stored
in a fixed-size buffer provided by the caller.
Other uses include a replacement of mempcpy cascades (which is
safer due to the size checking inherent to allocation buffers).
Implement strcmp family IFUNC selectors in C.
All internal calls within libc.so can use IFUNC on x86-64 since unlike
x86, x86-64 supports PC-relative addressing to access the GOT entry so
that it can call via PLT without using an extra register. For libc.a,
we can't use IFUNC for functions which are called before IFUNC has been
initialized. Use IFUNC internally reduces the icache footprint since
libc.so and other codes in the process use the same implementations.
This patch uses IFUNC for strcmp family functions within libc.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strcmp-sse2, strcmp-sse4_2, strncmp-sse2, strncmp-sse4_2,
strcasecmp_l-sse2, strcasecmp_l-sse4_2, strcasecmp_l-avx,
strncase_l-sse2, strncase_l-sse4_2 and strncase_l-avx.
* sysdeps/x86_64/multiarch/ifunc-strcasecmp.h: New file.
* sysdeps/x86_64/multiarch/strcasecmp.c: Likewise.
* sysdeps/x86_64/multiarch/strcasecmp_l-avx.S: Likewise.
* sysdeps/x86_64/multiarch/strcasecmp_l-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strcasecmp_l-sse4_2.S: Likewise.
* sysdeps/x86_64/multiarch/strcasecmp_l.c: Likewise.
* sysdeps/x86_64/multiarch/strcmp-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strcmp-sse4_2.S: Likewise.
* sysdeps/x86_64/multiarch/strcmp.c: Likewise.
* sysdeps/x86_64/multiarch/strncase.c: Likewise.
* sysdeps/x86_64/multiarch/strncase_l-avx.S : Likewise.
* sysdeps/x86_64/multiarch/strncase_l-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strncase_l-sse4_2.S: Likewise.
* sysdeps/x86_64/multiarch/strncase_l.c: Likewise.
* sysdeps/x86_64/multiarch/strncmp-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strncmp-sse4_2.S: Likewise.
* sysdeps/x86_64/multiarch/strncmp.c: Likewise.
* sysdeps/x86_64/multiarch/strcasecmp_l.S: Removed.
* sysdeps/x86_64/multiarch/strcmp.S: Likewise.
* sysdeps/x86_64/multiarch/strncase_l.S: Likewise.
* sysdeps/x86_64/multiarch/strncmp.S: Likewise.
* sysdeps/x86_64/multiarch/strcmp-sse42.S: Include <sysdep.h>.
(STRCMP_SSE42): New. Defined to __strcmp_sse42 if not defined.
[USE_AS_STRCASECMP_L || USE_AS_STRNCASECMP_L]: Include
"locale-defines.h".
(UPDATE_STRNCMP_COUNTER): New.
(SECTION): Likewise.
(GLABEL): Likewise.
(LABEL): Likewise.
* sysdeps/x86_64/multiarch/strncmp-ssse3.S: Rewrite and enable
for libc.a.
As shown by conform/ tests once the remaining namespace issues are
fixed, the tile bits/sigaction.h fails to declare SA_RESETHAND,
SA_RESTART and SA_NODEFER for non-XSI POSIX.1:2008 as other versions
do. Those constants were moved from XSI to Base in the 2008 edition
of POSIX. This patch fixes the conditions to match other versions of
this header.
Tested (compilation only) for tilegx-linux-gnu with
build-many-glibcs.py.
[BZ #21622]
* sysdeps/unix/sysv/linux/tile/bits/sigaction.h (SA_RESTART):
Define for [__USE_UNIX98 || __USE_XOPEN2K8], not [__USE_UNIX98 ||
__USE_MISC].
(SA_NODEFER): Likewise.
(SA_RESETHAND): Likewise.
Rename glibc.tune.ifunc to glibc.tune.hwcaps and move it to
sysdeps/x86/dl-tunables.list since it is x86 specicifc. Also
change type of data_cache_size, data_cache_size and
non_temporal_threshold to unsigned long int to match size_t.
Remove usage DEFAULT_STRLEN from cpu-tunables.c.
* elf/dl-tunables.list (glibc.tune.ifunc): Removed.
* sysdeps/x86/dl-tunables.list (glibc.tune.hwcaps): New.
Remove security_level on all fields.
* manual/tunables.texi: Replace ifunc with hwcaps.
* sysdeps/x86/cpu-features.c (TUNABLE_CALLBACK (set_ifunc)):
Renamed to ..
(TUNABLE_CALLBACK (set_hwcaps)): This.
(init_cpu_features): Updated.
* sysdeps/x86/cpu-features.h (cpu_features): Change type of
data_cache_size, data_cache_size and non_temporal_threshold to
unsigned long int.
* sysdeps/x86/cpu-tunables.c (DEFAULT_STRLEN): Removed.
(TUNABLE_CALLBACK (set_ifunc)): Renamed to ...
(TUNABLE_CALLBACK (set_hwcaps)): This. Update comments. Don't
use DEFAULT_STRLEN.
Backtrace through _dl_tlsdesc_resolve_rela was broken because the offset
of x30 from cfa was not in the debug info.
Add enough annotation so backtracing from the dynamic linker through
tlsdesc entry points works and the debugger shows registers correctly.
This patch add an extra test for passing invalid flags and check its
expected failure. It shows an invalid LO_HI_LONG macro definition for
x86_64 with leads to passing invalid flags on some configurations.
The new tests fails on i686-linux-gnu and potentially on other 32 bits
architecture that uses the compat syscall definition due a kernel bug.
It is intended to be fixed upstream.
Checked on x86_64-linux-gnu
* misc/tst-preadvwritev2-common.c: New file.
* misc/tst-preadvwritev2.c (do_test): Add test for invalid flag.
* misc/tst-preadvwritev64v2.c (do_test): Likewise.
