Calculate len separately for icache & dcache in case each has different
cacheline widths. Make the code for both loops the same except for the
dcbst/icbi. Deal with sizes >=2GB properly (like that'll happen but ...)
- Add alignment-safe double and float unions.
- Use the above for the __infinity and __nan constants on all
architectures that use the standard ieee754 representation of
those constants.
- Add a single copy of various ieee754 math functions (frexp, isinf,
isnan, ldexp and modf) that had numerous duplicates among the
arch-specific directories.
- Use the above functions on all architectures where the generic C
versions where used. Architectures that had local assembly
routines are untouched (for those functions only).
for most of this. I just integrated the build).
This adds proper functions for the _Qp* parts of the sparc64 ABI (which
handle quad softfloat). The routines work but need additional testing.
However if the compiler calls these currently bad things (core dumps)
will happen as gcc 2.95.3 generates bad calls for these.
Until that problem is fixed in the toolchain -msoft-quad-float cannot be
enabled as the default option for all builds (but the routines need to be
in libc as libgcc contains some references to these and anything including
it via --whole-archive will need the symbols to at least resolve).
of 3rd party software expects this.
XXX TODO: consider libm-fast for alpha that is not -mieee. Possibly
determine -mieee or not at run-time based on a bit in the executable
headers.
it (kernel and libc).
The current version of the gas assembler in the tree (2.11.2) already
defaults to generating object files for "-Av9 -64", supporting V9
instructions in ELF64 object format. "-Av9a" is only needed for specific
parts of the NetBSD base sources, and not for all third-party code.
Important fixes:
> ldexp(denormal, exp>1023) would generate the wrong result in all non-overflow
> cases. Totally rework this code to fix this bug *and* make it faster.
> Don't ever return -0.0.
