Fixed another precision bug in powf(). This one is in the computation
[t=p_l+p_h High]. We multiply t by lg2_h, and want the result to be
exact. For the bogus float case of the high-low decomposition trick, we
normally discard the lowest 12 bits of the fraction for the high part,
keeping 12 bits of precision. That was used for t here, but it doesnt't
work because for some reason we only discard the lowest 9 bits in the
fraction for lg2_h. Discard another 3 bits of the fraction for t to
compensate.
This bug gave wrong results like:
powf(0.9999999, -2.9999995) = 1.0000002 (should be 1.0000001)
hex values: 3F7FFFFF C03FFFFE 3F800002 3F800001
As explained in the log for the previous commit, the bug is normally
masked by doing float calculations in extra precision on i386's, but is
easily detected by ucbtest on systems that don't have accidental extra
precision.
Reference: 5f20e5ce7f
Original Author: Bruce Evans
(1) The bit for the 1.0 part of bp[k] was right shifted by 4. This
seems to have been caused by a typo in converting e_pow.c to
e_powf.c.
(2) The lower 12 bits of ax+bp[k] were not discarded, so t_h was
actually plain ax+bp[k]. This seems to have been caused by a logic
error in the conversion.
These bugs gave wrong results like:
powf(-1.1, 101.0) = -15158.703 (should be -15158.707)
hex values: BF8CCCCD 42CA0000 C66CDAD0 C66CDAD4
Fixing (1) gives a result wrong in the opposite direction
(hex C66CDAD8), and fixing (2) gives the correct result.
ucbtest has been reporting this particular wrong result on i386 systems
with unpatched libraries for 9 years. I finally figured out the extent
of the bugs. On i386's they are normally hidden by extra precision.
We use the trick of representing floats as a sum of 2 floats (one much
smaller) to get extra precision in intermediate calculations without
explicitly using more than float precision. This trick is just a
pessimization when extra precision is available naturally (as it always
is when dealing with IEEE single precision, so the float precision part
of the library is mostly misimplemented). (1) and (2) break the trick
in different ways, except on i386's it turns out that the intermediate
calculations are done in enough precision to mask both the bugs and
the limited precision of the float variables (as far as ucbtest can
check).
ucbtest detects the bugs because it forces float precision, but this
is not a normal mode of operation so the bug normally has little effect
on i386's.
On systems that do float arithmetic in float precision, e.g., amd64's,
there is no accidental extra precision and the bugs just give wrong
results.
Reference: 12be4e0d5a
Original Author: Bruce Evans
The decomposition needs to be into 12+24 bits of precision for extra-
precision multiplication, but was into 13+24 bits. On i386 with -O1 the
bug was hidden by accidental extra precision, but on amd64, in 2^32
trials the bug caused about 200000 errors of more than 1 ulp, with a
maximum error of about 80 ulps. Now the maximum error in 2^32 trials
on amd64 is 0.8573 ulps. It is still 0.8316 ulps on i386 with -O1.
The nearby decomposition of 1/ln2 and the decomposition of 2/(3ln2) in
the double precision version seem to be sub-optimal but not broken.
Reference: b4437c3d32
Original Author: Bruce Evans
nl_langinfo_l accesses lc_messages_T::codeset as soon as
__HAVE_LOCALE_INFO__ is defined, but codeset only exists
if __HAVE_LOCALE_INFO_EXTENDED__ is defined.
Fix this by defining lc_messages_T::codeset depending on
__HAVE_LOCALE_INFO__.
Fixes: ac7f1d5e931c ("Get rid of LCID, reformat type definitions in setlocale.h")
Reported-by: Inglis <Brian.Inglis@SystematicSW.ab.ca>
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
The nl_ext array contains offsets into the extended info of
locale structures, the index being equivalent to the nl_item
values in langinfo.h.
For the lc_messages_T struct, nl_ext erroneusly contains an
entry for the codeset member, which is in fact not part of the
extended info in nl_item. However, due to that, the offsets for
subsequent entries are off by one.
Fix this by dropping the messages::codeset entry from nl_ext.
Fixes: d47d5b850bed ("Extend locale support to maintain wide char values of native strings")
Reported-by: Brian Inglis <Brian.Inglis@systematicsw.ab.ca>
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
This avoids a `__utoa undefined' warning when building newlib
for Cygwin. We still need to export the symbols for backward
compatibility.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
If specified, use __builtin_dynamic_object_size() instead of
__builtin_object_size() if supported (GCC 12.0 or later).
This enables buffer overflow checks if the buffer size is non-const
but known during runtime.
Use new macro __ssp_bos_known() instead of the (bos(p) != (size_t)-1)
checks. The latter is no longer a compile time constant in all cases.
This avoids the generation of unused code.
Signed-off-by: Christian Franke <christian.franke@t-online.de>
When creating a split manual with one-node-per-page, the main index.html
ends up getting clobbered by the page for the index() function because
it uses "@node index" which, for html, also creates an index.html page.
To remedy this, add "Function " to every function node so now we output
"Function-index.html" and avoid clobbering. It also namespaces every
other function and helps make sure we don't clobber anything else.
Otherwise, there isn't really much rendering difference as @node text
is mostly internal. Node title text comes from @section instead.
The generated function documentation makes sure to include entries for
every function in the function index via @findex, but then the manuals
forget to actually print the index.
The local vars dst and src are unsigned pointers, but two assignments
cast their value to signed explicitely. This results in the warning
"pointer targets in assignment from ‘char *’ to ‘unsigned char *’
differ in signedness [-Wpointer-sign]" in case of -Wall.
Fix the cast.
Fixes: d254189b38bb ("2002-07-23 Jeff Johnston <jjohnstn@redhat.com>")
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
The warning "right-hand operand of comma expression has no effect
[-Wunused-value]" crops up with -Wall in cases where ORIENT is called
without checking or assigning its value.
Explicitely void the expression in these cases to avoid the warning.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
Compiling with -Wall uncovered a bug in _fputwc_r introduced in
commit 09119463a1445 ("stdio: split byte- and wide-char-oriented
low-level output functions"). The underlying function __fputwc
has been accidentally called without fetching its return value.
So the return value of _fputwc_r (and thus fputwc) was undefined.
Fixes: 09119463a1445 ("stdio: split byte- and wide-char-oriented low-level output functions"
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
This function closes or sets the close-on-exec flag for a specified
range of file descriptors. It is available on FreeBSD and Linux.
Signed-off-by: Christian Franke <christian.franke@t-online.de>
From: Rich Felker <dalias@aerifal.cx>
Date: Mon, 7 Nov 2022 22:17:55 -0500
the rule that longest digit sequence not beginning with a zero is
greater only applies when both sequences being compared are
non-degenerate. this is spelled out explicitly in the man page, which
may be deemed authoritative for this nonstandard function: "If one or
both of these is empty, then return what strcmp(3) would have
returned..."
we were wrongly treating any sequence of digits not beginning with a
zero as greater than a non-digit in the other string.
Signed-off-by: Brian Inglis <Brian.Inglis@SystematicSW.ab.ca>
Basically the startup code needs to clear memory from _edata to _end. In the
past it's been done with a fairly naive copy loop. This changes the code to
just call memset and let memset figure out a sensible way to handle the
operation given the size and alignment requirements.
I don't have performance data on this. I cobbled it together some time ago in
response to seeing some of the GCC tests with larger .bss sections taking an
insane amount of time to just get from _start to main. With the fixes to the
H8 decoder in the simulator it may not matter nearly as much anymore.
This has been in my tester for months. Naturally it does not cause any
regressions in the H8 port.
Without this, when building with recent gcc, we'll see errors when
compiling for --target mmix the first being:
CC libc/sys/mmixware/libc_a-chmod.o
In file included from /x/newlib/libc/sys/mmixware/chmod.c:17:
/x/newlib/libc/sys/mmixware/chmod.c: In function 'chmod':
/x/newlib/libc/sys/mmixware/sys/syscall.h:139:6: error: implicit declaration \
of function 'sprintf' [-Wimplicit-function-declaration]
139 | sprintf (buf, "UNIMPLEMENTED %s in %s\n", __FUNCTION__,
__FILE__); \
Other warnings also quelled.
* libc/sys/mmixware/sys/syscall.h: Include stdio.h, string.h
and unistd.h.
* libc/sys/mmixware/_exit.c: Call __unreachable after simulator exit.
* libc/sys/mmixware/chown.c (chown): Match declaration in unistd.h.
* libc/sys/mmixware/getpid.c (_getpid): Ditto.
* libc/sys/mmixware/kill.c (_kill): Ditto.
* libc/sys/mmixware/link.c (_link): Ditto.
* libc/sys/mmixware/read.c (_read): Ditto.
* libc/sys/mmixware/sbrk.c (_sbrk): Ditto.
* libc/sys/mmixware/unlink.c (_unlink): Ditto.
* libc/sys/mmixware/write.c (_write): Ditto.
... so that all of 'exit', '_exit', '_Exit' work. 'exit' thus becomes the
standard 'newlib/libc/stdlib/exit.c'. (Getting 'atexit' functional needs
further work elsewhere.)
See also commit 5841b2f6a4208682264d03e3edfa0a09881932a6
"nvptx: Implement '_exit' instead of 'exit'".
Similar to other patches. This adds a missing prototype and #include to some
H8/300 specific code in newlib. Pushed to the trunk given Jeff J's
pre-approval for these kinds of changes.
RISC-V newlib fails to build for soft float multilibs since long double
support was enabled in:
commit 04798b7bb69571452d2cfc7e0b052a9bbd3b619d
Author: Kito Cheng <kito.cheng@sifive.com>
Date: Mon Dec 4 15:41:39 2023 +0800
RISC-V: Support long double math
Long double for RISC-V is using 128 bit IEEE 754 format like Aarch64,
so we reference AArch64 to support that.
The RISC-V soft floating point environment only supports the
FE_TONEAREST rounding mode and does not support exceptions. Guard long
double rounding and exception support with ifdefs based on the presence
of the relevant rounding modes and exceptions.
Tested on gcc/g++ testsuite using RISC-V GNU Newlib Toolchain built by
riscv-gnu-toolchain with multilibs:
riscv-sim/-march=rv32i/-mabi=ilp32/-mcmodel=medlow
riscv-sim/-march=rv32iac/-mabi=ilp32/-mcmodel=medlow
riscv-sim/-march=rv32im/-mabi=ilp32/-mcmodel=medlow
riscv-sim/-march=rv32imac/-mabi=ilp32/-mcmodel=medlow
riscv-sim/-march=rv32imafc/-mabi=ilp32f/-mcmodel=medlow
riscv-sim/-march=rv64imac/-mabi=lp64/-mcmodel=medlow
Co-authored-by: Simon Cook <simon.cook@embecosm.com>
These fixes fall into a few different buckets. First c99 doesn't allow a
parameter without a type. So in cases where the type had previously been an
implicit int, make it an explicit int. Second, for return values, don't allow
them to be implicit ints either, make them explicit. In a few cases change c89
function definitions to c99 function definitions. Lastly include <stdlib.h>
in sbrk.c to get the prototype for abort () which we call when we detect a
heap/stack collision.
This reverts commit 17ac400c11bab30ac2c0bef12cbf7788f0b6f954.
The build failures were due to incorrectly using $(INCLUDES) when
running $(AS). Let's roll this back and drop $(INCLUDES) from the
$(AS) invocations.
For the newlib part, warnings are all from lack of sync between
libc/machine/cris/sys/signal.h and libc/include/sys/signal.h. This
commit gets them sufficiently in sync again, functionality-wise and
declaration-wise. Still, nothing is declared that isn't supported at
the system level (i.e. in libgloss system calls and handled by the
CRIS simulator in the gdb project).
Upstream GCC changed -Wimplicit-function-declaration warning into an
error. The build break about missing fpclassifyf function prototype
exposed a bug in the PRU port of libm.
The fix is to use the fpclassify macro for both double and float types.
Signed-off-by: Dimitar Dimitrov <dimitar@dinux.eu>
In cases where malloc fails for the 'g->matchjump' allocation, the code
path does not handle the failure gracefully, potentially leading to a
memory leak. This fix ensures proper cleanup by freeing the allocated
memory for 'pmatches' before returning.
Signed-off-by: Kuan-Wei Chiu <visitorckw@gmail.com>
We've been manually editing newlib.hin after generating it with
autoheader to drop the version defines that we keep in the separate
_newlib_version.h header. This is confusing for people, and is an
easy source of mistakes/errors.
Since we're already running sed on newlib.h during configure to
filter out defines we don't want to expose, add the version macros
there too. This way we don't have to manually edit newlib.hin.
This simplifies the autoreconf step in exchange for a slightly more
complicated configure+sed step, but seems worth the trade-off.
Previously, e.g. nl_langinfo(_NL_TIME_WMONTH_1) returns "February"
due to the bug. Similarly, nl_langinfo(_NL_TIME_WWDAY_1) returns
"Mon". This occurs because wide char month and weekday arrays are
pointed off-by-one (e.g. the array wmon[12] is reffered as wmon[1-12]
rather than wmon[0-11]). This patch fixes that.
Fixes: d47d5b850bed ("Extend locale support to maintain wide char values of native strings")
Reviewed-by: Corinna Vinschen <corinna@vinschen.de>
Signed-off-by: Takashi Yano <takashi.yano@nifty.ne.jp>
Now that the low-level functions set and test stream orientation,
a few calls in API functions are redundant.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
This allows to set and test orientation correctly if input is only
using macros from stdio.h. Wide-char-oriented functions must call
__srefill_r directly.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
Introduce function __swbufw_r and macros/inline-functions called
__swputc_r. Call these functions/macros exclusively from wide-char
functions.
This allows to set and test the stream orientation correctly even if
output is only performed using byte-oriented macros from stdio.h.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
The commit 3d94e07c49b5 has a few bugs which cause testsuite failure
in libstdc++. This is due to excess orientation check in __srefill_r()
and _ungetc_r(). Further, sscanf() family also calls ssvfscanf() family
with fp->_file == -1. This causes undesired orientation set/check for
sscanf() family. This patch fixes these problems.
Also, as in GLibC, do not set orientation in ungetc, and only set, but
do not check orientation in ungetwc.
Fixes: 3d94e07c49b5 ("newlib: libc: Fix crash on fprintf to a wide-oriented stream.")
Reported-by: Christophe Lyon <christophe.lyon@linaro.org>
Reported-by: Joel Sherrill <joel@rtems.org>
Co-developed-by: Corinna Vinschen <corinna@vinschen.de>
Signed-off-by: Takashi Yano <takashi.yano@nifty.ne.jp>
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
The _REENT_MP_P5S blocks are allocated using Balloc via i2b and linked in the
pow5mult call. As a result these blocks are not on the freelist managed by the
Bfree call. This change fixes a memory leak in threads that clean up using
_reclaim_reent.
RTEMS: Closes#4967
__sfvwrite_r is called under lock. There's no reason to call
putc, locking the file recursively. Add a comment that locking
is required when calling __sfvwrite_r.
Fixes: 49d64538cd20 ("* libc/include/stdio.h (FILE): define __SCLE for "convert line endings" for Cygwin.")
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
So far, the printf family of functions has two output helper functions
called __sprint_r and __sfputs_r. Both are called from all variants of
vfprintf as well as vfwprintf. There are also analogue helper functions
for the string-creating functions vsprintf/vswprintf called __ssprint_r
and __ssputs_r.
However, the helpers are built once when building vfprintf/vsprintf with
the INTEGER_ONLY flag, and then they are part of the vfiprintf.c and
vsiprintf.c files.
The problem is this:
Even if a process only calls vfwprintf or the non-INTEGER_ONLY vfprintf
it will always have to include the INTEGER_ONLY vfiprintf. Otherwise the
helper functions are undefined. Analogue for the string-creating
functions.
That's a useless waste of space by including one (or two) big, unused
function, if newlib is linked in statically.
Create new files to define the printf output helpers separately and
split them into byte-oriented and wide-char-oriented functions. This
allows to link only the required functions.
Also, simplify the string output helpers and fix a potential (but
unlikely) buffer overflow in __ssprint_r.
Fixes: 8a0efa53e449 ("import newlib-2000-02-17 snapshot")
Fixes: 6121968b198d ("* libc/include/stdio.h (__VALIST): Guard against multiple definition.")
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
The last two patches to configure.ac missed regeneration of newlib.hin.
Fixes: 64a11fded15b ("Fix problem with _newlib_version.h not being filled in correctly")
Fixes: 39f734a857e2 ("newlib: fix AC_DEFINE quoting style with version macros")
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
Previously, fprintf() on a wide-oriented stream crashes or outputs
garbage. This is because a narrow char string which can be odd bytes
in length is cast into a wide char string which should be even
bytes in length in __sprint_r/__sfputs_r based on the __SWID flag.
As a result, if the length is odd bytes, the reading buffer runs over
the buffer length, which causes a crash. If the length is even bytes,
garbage is printed.
With this patch, any output to the stream which is set to different
orientation fails with error just like glibc. Note that it behaves
differently from other libc implementations such as BSD, musl and
Solaris.
Reviewed-by: Corinna Vinschen <corinna@vinschen.de>
Signed-off-by: Takashi Yano <takashi.yano@nifty.ne.jp>
