This patch arranges for the libm-test tests to be run for the
finite-math-only function variants, in addition to the existing runs
for out-of-line and bits/mathinline.h inline variants.
gen-libm-test.pl is made to add a flag to all tests with non-finite
inputs or outputs so that they can be skipped at runtime when the
finite-math-only variants are being tested (skipping is for all
rounding modes; that is, -ffinite-math-only is being treated as
excluding overflow cases even when the rounding mode is such that the
overflowed result is finite). errno setting is not tested for these
variants (in general they don't set it, and it's
implementation-defined in ISO C whether it's set on underflow, with
the glibc definition being that it may not be for -ffinite-math-only;
other cases where errno would normally be expected to be set are
mostly excluded as having non-finite or overflowing arguments or
results).
As with the inline function tests, these ones are built with
-D__FINITE_MATH_ONLY=1 to select the function variants, rather than
-ffinite-math-only. Use of -ffinite-math-only would not be suitable
for these tests because it would also affect libm-test.inc code that
e.g. tests whether results are infinities or NaNs - if these function
variants have bugs that incorrectly produce such results, we want them
to show up in the tests, which means the compiler should not be
optimizing the tests on the basis of results being finite.
The finite-math-only functions share the same ulps settings as the
main out-of-line functions.
These interfaces were one significant group of untested ABIs listed at
<https://sourceware.org/ml/libc-alpha/2013-07/msg00386.html>. I
haven't rerun the script to list untested interfaces, but I expect
that the vast bulk of interfaces there are still untested and could do
with testcases being added (or, if applicable, being made compat
symbols; in general, most symbols not starting '_' are safe bets to
add tests for, while _* need more investigation of what the actual
public API is, if any). I'd like to encourage people to help reduce
the accumulation of untested interfaces by adding more tests.
Tested for x86_64 and x86. Given my recent lgamma/gamma and log*
fixes, the new tests pass (before those fixes, the new tests showed up
those bugs, so illustrating the practical utility of having tests for
these function variants).
* math/libm-test.inc (NON_FINITE): New macro.
(enable_test): Do not run tests flagged NON_FINITE if TEST_FINITE.
* math/gen-libm-test.pl (show_exceptions): Add argument
$non_finite.
(parse_args): Update call to show_exceptions.
* math/test-math-finite.h: New file.
* math/test-math-no-finite.h: Likewise.
* math/test-double-finite.c: Likewise.
* math/test-float-finite.c: Likewise.
* math/test-ldouble-finite.c: Likewise.
* math/test-double.c: Include "test-math-no-finite.h".
* math/test-float.c: Include "test-math-no-finite.h".
* math/test-ldouble.c: Include "test-math-no-finite.h".
* math/test-math-inline.h (TEST_FINITE): New macro.
* math/test-math-vector.h (TEST_FINITE): Likewise.
* math/Makefile (test-longdouble-yes): Add test-ldouble-finite.
(libm-tests): Add test-float-finite and test-double-finite.
($(objpfx)test-float-finite.o): New dependency on
$(objpfx)libm-test.stmp.
($(objpfx)test-double-finite.o): Likewise.
($(objpfx)test-ldouble-finite.o): Likewise.
(libm-test-no-inline-cflags): New variable.
(libm-test-finite-cflags): Likewise.
(CFLAGS-test-float-finite.c): Likewise.
(CFLAGS-test-double-finite.c): Likewise.
(CFLAGS-test-ldouble-finite.c): Likewise.
(CFLAGS-test-float.c): Use $(libm-test-no-inline-cflags).
(CFLAGS-test-double.c): Likewise.
(CFLAGS-test-ldouble.c): Likewise.
This patch refactors how libm-test.inc handles disabling
errno/exception handling tests, and some other tests, for
__FAST_MATH__ inline function tests.
The macro TEST_INLINE is changed from being defined/undefined to being
defined to 1 or 0, so that it can be tested in "if" conditionals
instead of #if. For tests of errno and exception setting, separate
macros TEST_ERRNO and TEST_EXCEPTIONS are added, and TEST_ERRNO is
also used in the disabling of errno testing for vector function
tests. The relevant conditionals are moved up a function, so that
they take effect before the code that counts the number of tests, so
the inline function tests now accurately report that 0 tests for
exceptions and errno were executed (whereas they previously reported a
large number desipte not running any such tests).
Tested for x86_64 and x86.
* math/test-math-errno.h: New file.
* math/test-math-inline.h (TEST_INLINE): Define to 1 instead of
empty.
(TEST_ERRNO): New macro.
(TEST_EXCEPTIONS): Likewise.
* math/test-math-no-inline.h (TEST_INLINE): Likewise.
(TEST_EXCEPTIONS): Likewise.
* math/test-math-vector.h (TEST_ERRNO): Likewise.
* math/test-double.c: Include "test-math-errno.h".
* math/test-float.c: Likewise.
* math/test-ldouble.c: Likewise.
* math/libm-test.inc (test_single_exception) [!TEST_INLINE]: Make
code unconditional.
(test_exceptions): Only run code if TEST_EXCEPTIONS.
(test_single_errno) [!TEST_INLINE && !TEST_MATHVEC]: Make code
unconditional.
(test_errno): Only run code if TEST_ERRNO.
(enable_test): Use "if" conditional on TEST_INLINE, not #ifdef.
of vector math functions infrastructure and several x86_64 implementations.
This patch is preparatory change in libm-test.c - splitting of macros which
form name of tested functions for ability to use separate name for tested
functions and for functions used in test suite infrastructure.
2015-05-14 Andrew Senkevich <andrew.senkevich@intel.com>
* math/test-double.c (FUNC_TEST): New macro.
* math/test-float.c: Likewise.
* math/test-idouble.c: Likewise.
* math/test-ifloat.c: Likewise.
* math/test-ildoubl.c: Likewise.
* math/test-ldouble.c: Likewise.
* math/libm-test.inc: Use FUNC_TEST for name of tested functions.
2001-07-06 Paul Eggert <eggert@twinsun.com>
* manual/argp.texi: Remove ignored LGPL copyright notice; it's
not appropriate for documentation anyway.
* manual/libc-texinfo.sh: "Library General Public License" ->
"Lesser General Public License".
2001-07-06 Andreas Jaeger <aj@suse.de>
* All files under GPL/LGPL version 2: Place under LGPL version
2.1.
* sysdeps/i386/fpu/libm-test-ulps: Add ulps for new exp tests.
* math/libm-test.inc (exp_test): Add some more tests.
(pow_test): Run double only test also for long double.
* math/test-ldouble.c (TEST_LDOUBLE): Define.
* math/test-ildoubl.c (TEST_LDOUBLE): Likewise.
1999-10-18 Andreas Jaeger <aj@suse.de>
Added improved math testsuite:
* math/libm-test.inc: New file, contains all tests from
libm-test.c but in a different form.
* math/libm-test.c: Removed.
* sysdeps/generic/libm-test-ulps: New file: Fallback ulps file.
* math/gen-libm-test.pl: New file: Preprocessor.
* math/test-double.c: Adopted to new testsuite.
* math/test-idouble.c: Likewise.
* math/test-float.c: Likewise.
* math/test-ifloat.c: Likewise.
* math/test-ldouble.c: Likewise.
* math/test-ildoubl.c: Likewise.
* math/Makefile: Added rules to generate new suite.
1997-11-22 19:28 Ulrich Drepper <drepper@cygnus.com>
* iconv/gconv_simple.c: Fix lots of bugs.
* iconv/iconv_open.c: Normalize names before passing them to
__gconv_open.
* libio/iolibio.h: Add prototypes for _IO_{new,old}_fclose.
* libio/iopopen.c: Check for correct mode argument and punt otherwise.
* stdio-common/test-popen.c: Add test for above change.
* math/libm-test.c: Print test results also using %a format,
specified in PRINTF_XEXPR macro.
* math/test-double.c: Define PRINTF_XEXPR.
* math/test-float.c: Likewise.
* math/test-ldouble.c: Likewise.
* math/test-idouble.c: Likewise.
* math/test-ifloat.c: Likewise.
* math/test-ildoubl.c: Likewise.
* stdio-common/vfscanf.c: Recognize a in format specifier only as
modifier if used before s, S, or [. Otherwise it's the hexadecimal
FP number format.
* stdlib/strtod.c: Little code cleanup.
* sysdeps/i386/Makefile: Define __LIBC_INTERNAL_MATH_INLINES for
compiling in math subdir.
* sysdeps/m68k/fpu/e_acos.c: Define __LIBC_INTERNAL_MATH_INLINES
instead of __LIBC_M81_MATH_INLINES.
* sysdeps/m68k/fpu/e_atan2.c: Likewise.
* sysdeps/m68k/fpu/e_fmod.c: Likewise.
* sysdeps/m68k/fpu/e_pow.c: Likewise.
* sysdeps/m68k/fpu/e_scalb.c: Likewise.
* sysdeps/m68k/fpu/k_cos.c: Likewise.
* sysdeps/m68k/fpu/k_sin.c: Likewise.
* sysdeps/m68k/fpu/k_tan.c: Likewise.
* sysdeps/m68k/fpu/s_atan.c: Likewise.
* sysdeps/m68k/fpu/s_ccos.c: Likewise.
* sysdeps/m68k/fpu/s_ccosh.c: Likewise.
* sysdeps/m68k/fpu/s_cexp.c: Likewise.
* sysdeps/m68k/fpu/s_csin.c: Likewise.
* sysdeps/m68k/fpu/s_csinh.c: Likewise.
* sysdeps/m68k/fpu/s_frexp.c: Likewise.
* sysdeps/m68k/fpu/s_ilogb.c: Likewise.
* sysdeps/m68k/fpu/s_isinf.c: Likewise.
* sysdeps/m68k/fpu/s_llrint.c: Likewise.
* sysdeps/m68k/fpu/s_llrintf.c: Likewise.
* sysdeps/m68k/fpu/s_llrintl.c: Likewise.
* sysdeps/m68k/fpu/s_lrint.c: Likewise.
* sysdeps/m68k/fpu/s_modf.c: Likewise.
* sysdeps/m68k/fpu/s_remquo.c: Likewise.
* sysdeps/m68k/fpu/s_scalbn.c: Likewise.
* sysdeps/m68k/fpu/s_sincos.c: Likewise.
* sysdeps/m68k/fpu/bits/mathinline.h: Define optimizations if
__LIBC_INTERNAL_MATH_INLINES is defined.
* sysdeps/i386/fpu/bits/mathinline.h: Define special i686 versions
of compare macros from ISO C 9X. Optimize generic versions a bit.
Define __signbit inline functions and __ieee754_sqrt which can be
used if the ompiler is fixed.
* sysdeps/posix/sysv_signal.c: Define sysv_signal as weak alias.
1997-11-20 Andreas Schwab <schwab@issan.informatik.uni-dortmund.de>
* Makerules ($(common-objpfx)stub-$(subdir)): Look for
sysdeps/generic in the whole dep file, not only in the first line.
1997-11-20 Andreas Schwab <schwab@issan.informatik.uni-dortmund.de>
* sysdeps/ieee754/bits/nan.h (__nan) [!__GNUC__]: Fix type.
1997-11-19 Andreas Schwab <schwab@issan.informatik.uni-dortmund.de>
* manual/arith.texi (Imaginary Unit): Move @end deftypevr to right
place.
1997-11-19 Andreas Schwab <schwab@issan.informatik.uni-dortmund.de>
* elf/dl-error.c (_dl_signal_error): Only print colon if OBJNAME
is non-empty.
1997-11-20 16:51 Ulrich Drepper <drepper@cygnus.com>
* malloc/malloc.c (malloc_update_mallinfo): Initialize struct for
return value.
1997-11-19 21:36 Ulrich Drepper <drepper@cygnus.com>
* sysdeps/unix/sysv/linux/getcwd.c: Free allocated buffer if
readlink succeeds but the return value is too long.
* stdio-common/printf_fphex.c: Determine sign of number correctly.
1997-11-19 02:45 Ulrich Drepper <drepper@cygnus.com>
* libc.map: Export __divl, __reml, __divq, __remq, __divqu and
__remqu for Alpha.
1997-11-18 23:08 Ulrich Drepper <drepper@cygnus.com>
* sysdeps/generic/sys/ucontext.h: Stub context file.
* libio/iofprintf.c: Removed. We use the stdio-common version.
* manual/Makefile (distribute): Add stdio-fp.c.
* sysdeps/powerpc/Dist: Add ppc-mcount.S.
* db2/Makefile (distribute): Add include/os_func.h, README and
mutex/tsl_parisc.s.
* hesiod/Makefile (distribute): Add README.hesiod.
1997-11-18 22:14 Ulrich Drepper <drepper@cygnus.com>
* iconv/gconv.c: Correctly call transformation function.
* manual/maint.texi: Document need of binutils 2.8.1.0.17.
1997-11-13 Andreas Jaeger <aj@arthur.rhein-neckar.de>
* sysdeps/libm-ieee754/s_remquo.c (__remquo): Fix function (gave
wrong results for e.g. remquo (3,2, &x)).
* sysdeps/libm-ieee754/s_remquof.c (__remquof): Likewise.
* sysdeps/libm-ieee754/s_remquol.c (__remquol): Likewise.
1997-11-18 Andreas Jaeger <aj@arthur.rhein-neckar.de>
* configure.in (ld --version-script): Test if the same symbol
maybe defined in all versions (needed by recent changes to
libc.map and supported with binutils 2.8.1.0.17) and issue warning
if versioning is not supported/enabled.
(bash test): Correct test case.
* math/libm-test.c: Added counting/reporting of number of test cases.
* elf/Makefile ($(objpfx)rtldtbl.h): gawk is not longer required,
use awk instead.
* sysdeps/unix/sysv/linux/net/if.h: Add __BEGIN_DECLS/__END_DECLS.
1997-11-18 Andreas Jaeger <aj@arthur.rhein-neckar.de>
* elf/dl-load.c (_dl_init_paths): Correct __mempcpy logic.
1997-11-18 18:04 Ulrich Drepper <drepper@cygnus.com>
* sysdeps/generic/pty.c: Define REVOKE to use revoke function.
Replace revoke call by use of REVOKE.
* sysdeps/unix/sysv/linux/pty.c: New file. Define REVOKE as
empty and #include generic version.
Suggested by Zack Weinberg <zack@rabi.phys.columbia.edu>.
1997-11-18 16:02 Ulrich Drepper <drepper@cygnus.com>
* libc.map: Add socket functions with __libc_* names.
Reported by Andreas Jaeger.
1997-11-18 Andreas Jaeger <aj@arthur.rhein-neckar.de>
* libc.map: Tidy up (remove duplicates, sort entries).
1997-11-18 Andreas Jaeger <aj@arthur.rhein-neckar.de>
* iconv/gconv.h: Add __BEGIN_DECLS/__END_DECLS.
1997-11-18 06:45 Ulrich Drepper <drepper@cygnus.com>
* sysdeps/generic/pty.c: Make thread safe. Fix bug when calling
again after successful call.
Patch by Marc Lehmann <pcg@goof.com>.
1997-11-17 Paul Eggert <eggert@twinsun.com>
* time/mktime.c, time/strftime.c (_REENTRANT): #define,
as some hosts need this to declare localtime_r properly.
1997-11-18 06:08 Ulrich Drepper <drepper@cygnus.com>
* libc.map: Add iconv_open, iconv, and iconv_close.
1997-05-29 12:48 Ulrich Drepper <drepper@cygnus.com>
* io/ftw.c: Complete rewrite. Add implementation of `nftw'.
* io/ftw.h: Update for new implementation and XPG4.2.
* login/Makefile: Update for UTMP daemon implementation.
Update resolver code to bind-4.9.6-T1A.
* resolv/Banner: Update.
* nss/digits_dots.c: Adapt text address matching to T1A.
* nss/nss_files/files-hosts.c: Always use inet_pton.
* resolv/base64.c (b64_pton): Follow T1A but don't use this code since
it would lead to warnings.
* resolv/gethnamaddr.c (getanswer): Test host name for maximal length
at several places.
* resolv/inet_net_pton.c (inet_net_pton_ipv4): Correct typo in comment.
* resolv/res_comp.c (dn_expand): Check for overflow.
(dn_comp): Likewise.
* resolv/res_debug.c (precsize_aton): Better implementation.
* resolv/res_init.c (res_init): Make `buf' of size MAXDNAME.
* resolv/res_send.c (res_send): Check for overflow in descriptor set.
* resolv/nss_dns/dns-host.c (getanswer_r): Test host name for maximal
length at several places.
1997-05-29 12:51 Mark Kettenis <kettenis@phys.uva.nl>
* login/utmp-private.h (struct utfuncs): Add one more parameter
to updwtmp function.
Declare all three function jump tables.
* login/utmp.h: Declare __utmpname.
* login/getutent_r.c: Remove db backend and provide support for
utmpd backend.
* login/login.c: Use `updwtmp' function insteead of writing the
record ourself.
* login/logwtmp.c: Move `updwtmp' function to...
* login/updwtmp.c: ...here. New file.
* login/utmp_db.h: Removed.
* login/utmp_file.c: Add updwtmp function to write to file.
* login/utmp_daemon.c: New file. Daemon backend.
* login/utmpname.c: New file. Implementation of utmpname function.
* login/utmpdump.c: New file. Tool to dump utmp-like files.
* login/utmpd/connection.c: New file.
* login/utmpd/database.c: New file.
* login/utmpd/error.c: New file.
* login/utmpd/request.c: New file.
* login/utmpd/utmpd-private.h: New file.
* login/utmpd/utmpd.c: New file.
* login/utmpd/utmpd.h: New file.
* login/utmpd/xtmp.c: New file.
* login/utmpd/xtmp.h: New file.
1997-05-29 12:28 Jim Meyering <meyering@eng.ascend.com>
* time/strftime.c: Correct/normalize indentation in cpp directives.
1997-05-28 20:43 Philip Blundell <pjb27@cam.ac.uk>
* nis/nis_error.c: Include <string.h> to fix warning.
* nis/nis_print.c: Likewise.
* nis/nss_nisplus/nisplus-hosts.c: Arg 3 of map_v4v6_hostent
is int* not size_t*.
1997-05-28 21:56 Andreas Jaeger <aj@arthur.rhein-neckar.de>
* math/cmathcalls.h: Correct typo in comment.
* inet/netinet/icmp6.h: Include <netinet/in.h> for in6_addr.
* sysdeps/unix/sysv/linux/netinet/ip_fw.h: Include <net/if.h> for
IFNAMSIZ.
* sysdeps/unix/sysv/linux/net/ppp_defs.h: Include <time.h> for
time_t.
* login/pty.h: Include <ioctl-types.h> for definition of struct
winsize.
* misc/regexp.h (compile): Correct typo.
* argp/argp.h: Put extern before __const in defintion of
argp_program_bug_address.
1997-05-29 00:20 Ulrich Drepper <drepper@cygnus.com>
* sysdeps/wordsize-32/inttypes.h: Correct names of unsigned fast
and least types. Correct names of ?INT_FAST*_{MIN,MAX} macros.
* sysdeps/wordsize-64/inttypes.h: Likewise.
Reported by Andreas Jaeger <aj@arthur.rhein-neckar.de>.
1997-05-28 22:51 Ulrich Drepper <drepper@cygnus.com>
* sysdeps/unix/Makefile (make-ioctls-CFLAGS): Use generic
ttydefaults.h file instead of non-existing version in termios/sys.
Reported by Zack Weinberg <zack@rabi.phys.columbia.edu>.
* time/strptime.c (strptime_internal, case 'Y'): Restrict year
number to four digits and to representable range for 4 byte time_t
values.
Patch by H.J. Lu <hjl@lucon.org>.
1997-05-28 18:19 Philip Blundell <pjb27@cam.ac.uk>
* posix/execl.c: Include <alloca.h> to avoid warning.
1997-05-27 18:19 Andreas Jaeger <aj@arthur.rhein-neckar.de>
* math/libm-test.c: Implement testing of inlined functions, make
output nicer, update comments.
* math/test-idouble.c: New file. Frontend for double tests of
inlined functions.
* math/test-ildoubl.c: New file. Frontend for long double tests of
inlined functions.
* math/test-ifloat.c: New file. Frontend for float tests of
inlined functions.
* math/test-longdouble.c: Rename to...
* math/test-ldouble.c: ...this.
* math/Makefile: Add rules for new test programs, change rules for
renaming of longdouble test.
1997-05-20 15:50 H.J. Lu <hjl@gnu.ai.mit.edu>
* sunrpc/rpc/svc.h (__dispatch_fn_t): New.
(svc_register): Use __dispatch_fn_t in prototype.
1997-05-28 17:02 Ulrich Drepper <drepper@cygnus.com>
* sysdeps/generic/bzero.c (bzero): Fix typo.
Patch by Witek Wnuk <spider@pest.waw.ids.edu.pl>.
1997-05-27 12:00 Andreas Jaeger <aj@arthur.rhein-neckar.de>
* sysdeps/generic/vtimes.c: Use ISO C declaration style.
* sysdeps/unix/bsd/ualarm.c: Include <unistd.h> for prototype.
* sysdeps/generic/memccpy.c: Include <string.h> for prototype.
* signal/tst-signal.c (handler): Correct function declaration to
avoid warning.
* stdlib/testsort.c (compare): Likewise.
* string/tester.c: Likewise.
1997-05-27 14:16 Miles Bader <miles@gnu.ai.mit.edu>
* argp-help.c (argp_args_usage): Supply correct argp to filter_doc.
1997-05-27 17:51 Andreas Schwab <schwab@issan.informatik.uni-dortmund.de>
* db/hash/extern.h, db/hash/hash.c, db/hash/hash.h,
db/hash/hash_log2.c: Rename __log2 to __hash_log2 to avoid clash
with libm.
1997-05-27 14:47 Andreas Schwab <schwab@issan.informatik.uni-dortmund.de>
* sysdeps/m68k/fpu/e_atan2.c: Fix missing negate. Use __m81_test
instead of explicit comparisons.
1997-05-26 18:36 Andreas Schwab <schwab@issan.informatik.uni-dortmund.de>
* inet/netinet/icmp6.h: Remove use of <asm/bitops.h> which has no
place in a generic header and is no user include file.
Joseph Myers
Andrew Senkevich
Allan McRae
Paul Eggert
Andreas Jaeger
Ulrich Drepper