For now this builds the C version, for completeness, so
that e.g. lang/ocaml can be configured & built. However,
googling reveals that powerpc does have a "fused multiply add"
instruction, ref.
https://www.ibm.com/docs/en/aix/7.1?topic=set-fmadd-fma-floating-multiply-add-instruction
so this could probably be taken advantage of for a more
optimized version.
Alpha has no hardware-assisted fused multiply-add, so the fallback
C implementation must be used. A basic smoke test shows the
implementation works for
"x * x - 1.0f" vs "fmaf(x, x, -1.0f)" with "volatile float x = 1.0008f;"
Fix ABI mismatch for armhf runtime routines for floating-point arithmetics;
For hard-float arm variants, provide
(1) generic runtime routines with correct calling convention, and
(2) EABI runtime routines at the same time.
I've confirmed that no binary changes for kernels.
LGTM by skrll
- Don't touch the trap flags (though on all ARMv8 I know they have no
effect anyway).
- Don't clear any existing raised exception flags; just add to them.
XXX atf test
XXX pullup-9
- fesetenv is supposed to set the stored rounding mode (and stored trap
settings, but they have no effect on any ARMv8 I know).
- feupdateenv is supposed to re-raise the exceptions that were raised
in the environment when it was called.
XXX atf test
XXX pullup-9
These were removed from the build for being wrong, but the implementation
stayed around. This is confusing, we have the attic for old code,
let's delete the unused implementations.
The long-standing GCC default is to not respect rounding mode.
it looks like GCC 7 optimizes rint to a builtin, causing our few
rounding mode tests to fail.
Fixes PR port-amd64/54000: FP tests failing on amd64 since gcc7 import
Similarly, when hardware NaN propagation is available, disable
default-NaN substitution.
This enables IEEE 754 semantics on any hardware that supports it by
default. Programs that want flush-to-zero or default-NaN substitution
can enable them explicitly.
ok ryo@
The x87 hardware uses a bad approximation to pi for argument
reduction, and consequently yields bad answers for inputs near pi or
pi/2.
Tweak one tanf test whose doubly rounded output is a little too far
from the correct answer on the doubly rounded input.
The sanitizers are designed to use pristine OS versions of these libraries,
install interceptors for public symbols in them and link with them in every
sanitized executable.
Having unmodified base libraries (c, m, rt, pthread) also makes the system
usable with programs linking with them and without sanitization, as there
are no fatal issues during the resolving of missing symbols.
The remaining libraries like libutil, libintl etc are prebuilt with
sanitization and all their users in the MKSANITIZER=yes distribution are
required to use the same sanitizer.
libc and libpthread are already marked with the NOSANITIZER flag.
As a component of atan2(y, x), the case of x == 1.0 is farmed out to
atan(y). The current implementation of this comparison is vulnerable
to signed integer underflow (that is, undefined behavior), and it's
performed in a somewhat more complicated way than it need be. Change
it to not be quite so cute, rather directly comparing the high/low
bits of x to the specific IEEE-754 bit pattern that encodes 1.0.
ok martin@ pgoyette@ maya@
obtained from FreeBSD
martin