The simple newlib hypotl for real long double architectures is too
simple at this point. It's implemented as a real call to sqrtl(x^2+y^2).
This has a fatal tendency to overflow for big input numbers. Hypotl
isn't supposed to do that if the result would still be valid in range of
long double.
Given the complexity of implementing hypotl for various architectures,
we just take the hypotl function from Mingw-w64, which is in the public
domain.
Even though this hypotl is an architecture-independent implementation,
we can't use it for newlib yet, unfortunately, because it requires logbl
under the hood. Logbl is yet another function missing in newlib for
real long double architectures.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
The return value is now -NaN.
This fixes a bug in the mingw-w64 code that was imported into Cygwin.
The fix is consistent with Posix and Linux. It is also consistent
with the current mingw-w64 code, with one exception: The mingw-w64
code sets errno to EDOM if the input is -NaN, but this appears to
differ from Posix and Linux.
Addresses: https://cygwin.com/pipermail/cygwin/2020-October/246606.html
On 32 bit x86, clang seems to miss loading input parameters based
on asm constraints for inline assembly that uses the x87 floating
registers, unless the snippet has got the volatile keyword.
Signed-off-by: Martin Storsjö <martin@martin.st>
Drop Cygwin-specific nanl in favor of a generic implementation
in newlib. Requires GCC 3.3 or later.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
Mingw-w64 (where the code has been taken from) has 4 byte longs
independently of the architecture but x86_64 Cygwin has 64 bit longs.
So use fistpll instead of fistpl on x86_64 Cygwin.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
previous commit 4c90db7bc8 introduced
a compile time error because libm/common/s_infconst.c used the remove
__fmath, __dmath, and __ldmath union types.
Since this is very old, and unused for a very long time, just drop the
file and thus the __infinity constants entirely.
Exception: Cygwin exports __infinity from the beginning. There's a very,
VERY low probability that any existing executable or lib still uses this
constant, but we just keep it in for backward compat, nevertheless.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
This routine makes a call to fabs instead of fabsl(), causing truncation.
Clang complains (warning: absolute value function 'fabs' given an argument of type 'long double' but has parameter of type 'double' which may cause truncation of value).
Signed-off-by: David Wohlferd <dw@LimeGreenSocks.com>
The R language has some hacks specifically for mingw-w64 that
were caused by our handling of NaNs in sqrt(x). R uses a
special valued NaN to mean 'Not Available' and expects it to
be retained through various calculations. Our sqrt(x) doesn't
do this, instead it normalises such a NaN (retaining sign).
From:
http://wwwf.imperial.ac.uk/~drmii/M3SC_2016/IEEE_2008_4610935.pdf
"6.2.3 NaN propagation
An operation that propagates a NaN operand to its result and
has a single NaN as an input should produce a NaN with the
payload of the input NaN if representable in the destination
format."
There might even be a slight speed-up from this too.
Thanks to Duncan Murdoch for finding the reference.
Mingw-w64, which is the source of this code, uses different
definitions of the rounding bits FE_TONEAREST and friends.
They immediately reflect the bit values in the FPU control word,
while on Cygwin they are shifted down to become the values 0-3.
Fix the bit computing expression to account for the difference.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
This patch adds the long double functions missing in newlib to Cygwin.
Apart from some self-written additions (exp10l, finite{f,l}, isinf{f,l},
isnan{f,l}, pow10l) the files are taken from the Mingw-w64 math lib.
Minor changes were required, e.g. substitue _WIN64 with __x86_64__ and
fixing __FLT_RPT_DOMAIN/__FLT_RPT_ERANGE for Cygwin.
Cygwin:
* math: New subdir with math functions.
* Makefile.in (VPATH): Add math subdir.
(MATH_OFILES): List of object files collected from building files in
math subdir.
(DLL_OFILES): Add $(MATH_OFILES).
${CURDIR}/libm.a: Add $(MATH_OFILES) to build.
* common.din: Add new functions from math subdir.
* i686.din: Align to new math subdir. Remove functions now commonly
available.
* x86_64.din: Ditto.
* math.h: math.h wrapper to define mingw structs used in some files in
math subdir.
* include/cygwin/version.h: Bump API minor version.
newlib:
* libc/include/complex.h: Add prototypes for complex long double
functions. Only define for Cygwin.
* libc/include/math.h: Additionally enable prototypes of long double
functions for Cygwin. Add Cygwin-only prototypes for dreml, sincosl,
exp10l and pow10l. Explain why we don't add them to newlib.
* libc/include/tgmath.h: Enable long double handling on Cygwin.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
Corinna Vinschen
Ken Brown
Martin Storsjö
Liu Hao
Jozef Lawrynowicz
David Wohlferd
Ray Donnelly
Yaakov Selkowitz