fix mainly same typos as in my previous commit but outside sys/dev/dm.

Update a list of implement/unimplement functions.

branches: 1.5.2; 1.5.12;
Update implemented/unimplemented FP functions.

Update from Ken Nakata.

Fixes from Ken Nakata. Mainly renaming debug_level to fpu_debug_level.

Add NetBSD id string and update instructions.

Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.

Introduce the CORDIC algorithm.
o sine and cosine (e.g., FSIN, FCOS and FSINCOS instructions) is now
calculated in the CORDIC instead of Taylor expansion.
o tangent (FTAN) is not touched from a viewpoint of the code size.
o The CORDIC is applicable for hyperbolic functions (e.g., FSINH,
FCOSH, FTANH instructions), but I didn't use it because its working
range is poor.
o The CORDIC is also usable for inverse trigonometric functions,
I will commit it at next phase.
o The code size becomes a bit big. I cannot evaluate speed on m68k
for some reasons, but in test on i386 the CORDIC is approximately
100 times faster in sin/cos.

branches: 1.2.204; 1.2.214;
Add in NetBSD id string.

Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.

u_int -> uint32_t

Remove 'register'.

branches: 1.7.12; 1.7.22;
Change about 4500 of the K&R function definitions to ANSI ones.
There are still about 1600 left, but they have ',' or /* ... */
in the actual variable definitions - which my awk script doesn't handle.
There are also many that need () -> (void).
(The script does handle misordered arguments.)

branches: 1.6.78; 1.6.86; 1.6.92;
merge ktrace-lwp.

Move UCB-licensed code from 4-clause to 3-clause licence.

Patches provided by Joel Baker in PR 22364, verified by myself.

__KERNEL_RCSID()

branches: 1.3.36;
PR 7220 from Ken Nakata:
I've fixed most (not all) m68k FPE bugs that give bogus
calculation results, esp. fsqrt instruction. Also, the internal FP
representation has been reduced from 115-bit mantissa to 67-bit
mantissa which reduced the required mantissa operation roughly by one
fourth. I've done an extensive (though not exhaustive - it's
impossible!) test on the internal routines by feeding them randomly
generated FP numbers, and found that the new code is more precise than
MC68040 FPU (it seems to have a rounding bug).

[ Only change was to keep fpu_calcea.c's name instead of renaming to
fpu_ea.c in Ken's patch. --akb ]

branches: 1.2.24; 1.2.26; 1.2.28;
Update for -Wall -Wstrict-prototypes -Wmissing-prototypes -Wno-uninitialized.

Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.

quad_t -> uint64_t

u_int -> uint32_t

Remove 'register'.

branches: 1.5.112; 1.5.122;
bare asm -> __asm

merge ktrace-lwp.

branches: 1.3.16;
Move UCB-licensed code from 4-clause to 3-clause licence.

Patches provided by Joel Baker in PR 22364, verified by myself.

branches: 1.2.36;
PR 7220 from Ken Nakata:
I've fixed most (not all) m68k FPE bugs that give bogus
calculation results, esp. fsqrt instruction. Also, the internal FP
representation has been reduced from 115-bit mantissa to 67-bit
mantissa which reduced the required mantissa operation roughly by one
fourth. I've done an extensive (though not exhaustive - it's
impossible!) test on the internal routines by feeding them randomly
generated FP numbers, and found that the new code is more precise than
MC68040 FPU (it seems to have a rounding bug).

[ Only change was to keep fpu_calcea.c's name instead of renaming to
fpu_ea.c in Ken's patch. --akb ]

branches: 1.1.24; 1.1.26; 1.1.28;
Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.

Remove all the __P() from sys (excluding sys/dist)
Diff checked with grep and MK1 eyeball.
i386 and amd64 GENERIC and sys still build.

branches: 1.1.170; 1.1.178; 1.1.184;
Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.

Fix indent.

branches: 1.27.36;
Overhaul the API used to fetch and store individual memory cells in
userspace. The old fetch(9) and store(9) APIs (fubyte(), fuword(),
subyte(), suword(), etc.) are retired and replaced with new ufetch(9)
and ustore(9) APIs that can return proper error codes, etc. and are
implemented consistently across all platforms. The interrupt-safe
variants are no longer supported (and several of the existing attempts
at fuswintr(), etc. were buggy and not actually interrupt-safe).

Also augmement the ucas(9) API, making it consistently available on
all plaforms, supporting uniprocessor and multiprocessor systems, even
those that do not have CAS or LL/SC primitives.

Welcome to NetBSD 8.99.37.

branches: 1.26.54;
fix indent again.
- "Second level indents are four spaces." pointed out by tsutsui@
- fold long line.

fix indent. no functional changes.

- consistently use #ifdef DEBUG_FPE and #ifdef DIAGNOSTIC
- use DPRINTF() style debug printfs
- use __func__ to print function names

KNF. No binary change.

branches: 1.22.2;
fix PR 6724 - convert m68k options to defflag's. this means that
M680[12346] are now available from opt_m68k_arch.h. FPSP meantioned
in the PR has already been fixed, and i could not find any more.

i built these kernels to ensure i did not break their builds:

amiga: GENERIC DRACO
atari: HADES FALCON MILAN-PCIIDE
mac68k: GENERIC
sun2: GENERIC
sun3: GENERIC GENERIC3X
cesfic: attempted GENERIC, does not build due to lack of machine/bus.h
hp300: GENERIC
luna68k: GENERIC
mvme68k: GENERIC
news68k: GENERIC
next68k: GENERIC
x68k: GENERIC

branches: 1.21.2; 1.21.4;
ANSIfy another 1261 function definitions.
The only ones left in sys are beyond by sed script!
(or in sys/dist or sys/external)
Mostly they have function pointer parameters.

Change about 4500 of the K&R function definitions to ANSI ones.
There are still about 1600 left, but they have ',' or /* ... */
in the actual variable definitions - which my awk script doesn't handle.
There are also many that need () -> (void).
(The script does handle misordered arguments.)

Remove all the __P() from sys (excluding sys/dist)
Diff checked with grep and MK1 eyeball.
i386 and amd64 GENERIC and sys still build.

branches: 1.18.44; 1.18.52; 1.18.58;
Don't disable EA_DIRECT in fpu_load_ea(), which seems disabled
accidentally in rev 1.11 about six years ago, otherwise
fmovel %dN,%fpN instructions cause unexpected memory access.

branches: 1.17.24; 1.17.26;
merge ktrace-lwp.

branches: 1.16.14; 1.16.16;
Uppercase CPU, plural is CPUs.

__KERNEL_RCSID()

branches: 1.14.6;
remove trailing \n in panic(). approved perry.

branches: 1.13.6;
bcopy -> memcpy

branches: 1.12.2;
Fix typo. s/extention/extension/

Don't use the frames' provided EA yet; something is wrong.

Make this actually sort-of-work for 68060 w. disabled FPU (tested), 68LC060,
and I suspect, 68LC040 and 68040V machines. Method used:

1. do NOT use the frames effective address at all
2. do NOT use the frames next instruction address, but
3. restore the "this instruction" pc into the pc field of the frame at the
beginning of the emulation, if we got a frame format 4.

TODO:
A. find the bug in the usage of the frame EA, and revert to using it.
B. find the bug in the usage of the frames next instruction address, and
consider whether it's worthwhile to use it.
C. make the emulation complete.... FEXP FETOM1, and the trigonometric functions
are missing.

Argument modreg of fpu_decode_ea() is actually an opecode.
That bug prevented awk(1) to work.

branches: 1.8.2; 1.8.10;
PR 7220 from Ken Nakata:
I've fixed most (not all) m68k FPE bugs that give bogus
calculation results, esp. fsqrt instruction. Also, the internal FP
representation has been reduced from 115-bit mantissa to 67-bit
mantissa which reduced the required mantissa operation roughly by one
fourth. I've done an extensive (though not exhaustive - it's
impossible!) test on the internal routines by feeding them randomly
generated FP numbers, and found that the new code is more precise than
MC68040 FPU (it seems to have a rounding bug).

[ Only change was to keep fpu_calcea.c's name instead of renaming to
fpu_ea.c in Ken's patch. --akb ]

branches: 1.7.22; 1.7.24; 1.7.26;
Eliminate fusword() prototype, as it's now in <sys/systm.h>

backout previous kprintf change

printf -> kprintf, sprintf -> ksprintf

Update for -Wall -Wstrict-prototypes -Wmissing-prototypes -Wno-uninitialized.

sys/types.h -> sys/param.h and remove include of stddef.h.

Fixes from Ken Nakata. Mainly renaming debug_level to fpu_debug_level.

Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.

Remove fpu_cordit2() and atanh_table[] completely.
Since cordit1 (for trigonometric functions) and cordit2 (for
hyperbolic functions) are very similar, so I implemented both
at first, but I didn't use cordit2 after all :(

Move fpu_cordit2() to #ifdef CORDIC_BOOTSTRAP section.
This reduces ~2KB text segment.
Reported by Krister Walfridsson on tech-kern two months ago.

branches: 1.2.4; 1.2.12; 1.2.16; 1.2.20;
s/tayler/taylor/. pointed out by christos@

Introduce the CORDIC algorithm.
o sine and cosine (e.g., FSIN, FCOS and FSINCOS instructions) is now
calculated in the CORDIC instead of Taylor expansion.
o tangent (FTAN) is not touched from a viewpoint of the code size.
o The CORDIC is applicable for hyperbolic functions (e.g., FSINH,
FCOSH, FTANH instructions), but I didn't use it because its working
range is poor.
o The CORDIC is also usable for inverse trigonometric functions,
I will commit it at next phase.
o The code size becomes a bit big. I cannot evaluate speed on m68k
for some reasons, but in test on i386 the CORDIC is approximately
100 times faster in sin/cos.

indent fix in a comment.

Fix a sign when a source or destination is either (plus/minus)zero
or (plus/minus)infinity. Found by XM6i.

branches: 1.8.4;
u_int -> uint32_t

Remove 'register'.

branches: 1.6.12; 1.6.22;
Change about 4500 of the K&R function definitions to ANSI ones.
There are still about 1600 left, but they have ',' or /* ... */
in the actual variable definitions - which my awk script doesn't handle.
There are also many that need () -> (void).
(The script does handle misordered arguments.)

branches: 1.5.78; 1.5.86; 1.5.92;
merge ktrace-lwp.

Move UCB-licensed code from 4-clause to 3-clause licence.

Patches provided by Joel Baker in PR 22364, verified by myself.

__KERNEL_RCSID()

branches: 1.2.36;
PR 7220 from Ken Nakata:
I've fixed most (not all) m68k FPE bugs that give bogus
calculation results, esp. fsqrt instruction. Also, the internal FP
representation has been reduced from 115-bit mantissa to 67-bit
mantissa which reduced the required mantissa operation roughly by one
fourth. I've done an extensive (though not exhaustive - it's
impossible!) test on the internal routines by feeding them randomly
generated FP numbers, and found that the new code is more precise than
MC68040 FPU (it seems to have a rounding bug).

[ Only change was to keep fpu_calcea.c's name instead of renaming to
fpu_ea.c in Ken's patch. --akb ]

branches: 1.1.24; 1.1.26; 1.1.28;
Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.

m68k/fpe: Avoid an illegal mod/reg before decoding it.
This also works for machines with FPU.
If the kernel defines FPU_EMULATE, even illegal FPU instructions in the real
6888x machines, the FPE will be executed by F line exception. So the FPE
must raise SIGILL appropriately for these.

m68k/fpe: Remove duplicated assignments.
These are already stored before. No functional changes intended.

m68k/fpe: Fix FTRAPcc emulation raise the correct trap.
Currently fpu_emul_type1() returns SIGFPE only if FTRAPcc's condition is met,
so this works. But I wish there was a better way..
fputest.x(0.24.10.08) about ftrapcc passed 100%.

m68k/fpe: Reimplement and simplify test_cc().
This logic was analized by Y.Sugahara (in 2016).
Finally fputest.x(0.24.10.08) about fbcc fdbcc fscc (40960 tests) passed 100%!

m68k/fpe: Two bugfixes in test_cc().
- Don't clear the exception byte in FPSR.
According to the manual, all bits except BSUN are "not affected".
- Accumulate AIOP bit always.
XXX Need to look this again later. brcc() looks to have similar one.

m68k/fpe: Several bugfixes in FDBcc and FTRAPcc emulation.
In fpu_emul_type1(),
- If test_cc() returns >0, it's an error, that is an illegal instruction in
this case. So it should just return without doing anything.
By this fix, FTRAPcc with illegal cc now causes SIGILL correctly.
- In result, branch can only be 0 or -1 after that. It makes code simple.
By this cleanup, FDBcc with illegal cc now causes SIGILL correctly.
- FTRAPcc must advance PC regardless of whether the condition is met.
(although FTRAPcc does not work at this point by anothor problem)

m68k/fpe: Fix several FMOVEM FPctl emulation.
- Fix the write order of FMOVEM FPctl,-(An).
For example, "FMOVEM FPCR/FPSR,-(An)" instruction first decrements An by
the total size (4*2 bytes), then writes FPCR and FPSR in this order.
Therefore, it's differs from "FMOVEM FPCR,-(An); FMOVEM FPSR,-(An)" sequence.
- Support multiple immediate case like "FMOVEM #imm/#imm,FPCR/FPIAR".
- Mask writes to FPSR and FPCR registers correctly.
- Support reglist=0 case too. It's not common case but makes code easier.

m68k/fpe: FMOVEM (both FPn and FPctl) must not update accrued byte in FPSR.

branches: 1.41.6;
word was changed to sval at rev 1.39, but DPRINTF values were not updated.

Fixes DEBUG_FPE enabled build for 68k FPE code.

s/opration/operation/

Overhaul the API used to fetch and store individual memory cells in
userspace. The old fetch(9) and store(9) APIs (fubyte(), fuword(),
subyte(), suword(), etc.) are retired and replaced with new ufetch(9)
and ustore(9) APIs that can return proper error codes, etc. and are
implemented consistently across all platforms. The interrupt-safe
variants are no longer supported (and several of the existing attempts
at fuswintr(), etc. were buggy and not actually interrupt-safe).

Also augmement the ucas(9) API, making it consistently available on
all plaforms, supporting uniprocessor and multiprocessor systems, even
those that do not have CAS or LL/SC primitives.

Welcome to NetBSD 8.99.37.

branches: 1.38.30;
Comment unused stuff (for documentation purposes)

branches: 1.37.4;
u_int -> uint32_t

branches: 1.36.2; 1.36.12;
Add hyperboric and trigonometric functions to m68k FPE, written by isaki@.
With these emulations (~4KB text) xeyes on XM6i works better.
Discussed with isaki@ at OSC 2011 Hiroshima.

fix indent again.
- "Second level indents are four spaces." pointed out by tsutsui@
- fold long line.

fix indent. no functional changes.

- use DPRINTF() style debug printfs
- use __func__ to print function names
- consistently use #ifdef DEBUG_FPE
- add some missing debug messages including \n in error paths

KNF, mostly indent. No binary change.

Fix botch in rev 1.28 that causes wrong results of fcmp and ftst in FPE.
fpu_upd_fpsr() should be called even in discard_result case if an emulated
instruction gets proper result without signal.

Fixes weird behavior of awk(1) seen on /etc/rc.d/postfix on XM6i and
TME emulating sun3 without 68881.

Should be pulled up to all netbsd-4 and netbsd-5 branches.

branches: 1.30.4; 1.30.6;
Change about 4500 of the K&R function definitions to ANSI ones.
There are still about 1600 left, but they have ',' or /* ... */
in the actual variable definitions - which my awk script doesn't handle.
There are also many that need () -> (void).
(The script does handle misordered arguments.)

Remove all the __P() from sys (excluding sys/dist)
Diff checked with grep and MK1 eyeball.
i386 and amd64 GENERIC and sys still build.

branches: 1.28.2;
In fpu_emul_arith(), check lower 7 bits in word1 rather than only 6 bits
to check 040/060 FP instructions, and don't call fpu_implode() and
fpu_upd_fpsr() if no vaild emulated result is set otherwise these
functions cause NULL pointer dereference.

Fixes panics triggered by 040/060's FDADD instruction
(which has the same lower 6 bits with fscale instruction)
on 020/030 machines (even with 68881/68882) running kernels
with options FPU_EMULATE.
Problem reported by John Carr on port-sun3.

Should be pulled up to netbsd-4-0, netbsd-4 and netbsd-5.

XXX: m68k 4.0 packages binaries on ftp have this FDADD instruction.

branches: 1.27.44; 1.27.52; 1.27.54;
Make these compile with options DEBUG_FPE.

branches: 1.26.24; 1.26.26;
merge ktrace-lwp.

branches: 1.25.14; 1.25.16;
SA_SIGINFO support for m68k

__KERNEL_RCSID()

branches: 1.23.24;
conditionalize the ddb hook on DDB && DEBUG_FPE rather than DDB && DEBUG,
since that's the condition under which we include the header file that
allows it to compile.

Make this actually sort-of-work for 68060 w. disabled FPU (tested), 68LC060,
and I suspect, 68LC040 and 68040V machines. Method used:

1. do NOT use the frames effective address at all
2. do NOT use the frames next instruction address, but
3. restore the "this instruction" pc into the pc field of the frame at the
beginning of the emulation, if we got a frame format 4.

TODO:
A. find the bug in the usage of the frame EA, and revert to using it.
B. find the bug in the usage of the frames next instruction address, and
consider whether it's worthwhile to use it.
C. make the emulation complete.... FEXP FETOM1, and the trigonometric functions
are missing.

branches: 1.21.2;
PR 7220 from Ken Nakata:
I've fixed most (not all) m68k FPE bugs that give bogus
calculation results, esp. fsqrt instruction. Also, the internal FP
representation has been reduced from 115-bit mantissa to 67-bit
mantissa which reduced the required mantissa operation roughly by one
fourth. I've done an extensive (though not exhaustive - it's
impossible!) test on the internal routines by feeding them randomly
generated FP numbers, and found that the new code is more precise than
MC68040 FPU (it seems to have a rounding bug).

[ Only change was to keep fpu_calcea.c's name instead of renaming to
fpu_ea.c in Ken's patch. --akb ]

branches: 1.20.8; 1.20.10; 1.20.12;
defopt DDB.

Back out last change, which was a thinko.

Add ftwotox emulation, by using the fscale emulation routine (preload
destination register bit pattern with 1.0), which automatically provides
corner case handling.
Missing ftwotox emulation originally reported by Norman Mackenzie in PR 4237,
but he proposed a different implementation.

Make us pass the full ieeefp except regression test.
XXX This is probably a horrible hack, and wrong. FPU signal handling needs
to be further evaluated.

Fix my own PR port-m68k/3871, nee port-amiga/3871:
- don't erase FPSR exception bits _after_ doing most of the operations in
fpu_implode(), erase them before doing arith and store operations. This fixes
losing the DZ bit.
- create FPSR_OVFL and FPSR_UNFL bits in fpu_implode(). This showed up when
the first error was fixed.
XXX some more work needs to be done. E.g., creating OPERR together with
OVFL looks bogus, but I'm too tired know to re-check docs; and at least we
pass our own regression tests know.

include <m68k/db_machdep.h> to get the prototype for kdb_trap if DDB and
DEBUG are defined and add a cast for the second argument to the
kdb_trap call.

Work around another uninitialized variable warning

Fix warnings.

Bug fix for FSINCOS: wouldn't handle FPc == FP1 earlier.

Eliminate fusword() prototype, as it's now in <sys/systm.h>

backout previous kprintf change

printf -> kprintf, sprintf -> ksprintf

The 68LC040 generates a format 4 stack frame for floating point
exceptions, which puts the address of the instruction we faulted
on in a different location. Copy it and handle as we normally would,
restoring the saved PC before returning.

The FPE should probably be reworked to take advantage of the 68LC040's
precalculated effective address, at some point.

Fix FMOVEM emulation; postincrement modes use different ordering than
predecrement modes, per 68K Programmer's Manual p. 5-88. Pointed out
by Takeshi Nakayama <takeshi@sakabe.nuie.nagoya-u.ac.jp>, verified by me.

branches: 1.6.4;
Fix "might be used uninitialized" warnings.

Update for -Wall -Wstrict-prototypes -Wmissing-prototypes -Wno-uninitialized.

Fixes from Ken Nakata. Mainly renaming debug_level to fpu_debug_level.

Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.

Add missing break in switch (not that it mattered 8^)

branches: 1.1.1;
Initial revision

Remove fpu_cordit2() and atanh_table[] completely.
Since cordit1 (for trigonometric functions) and cordit2 (for
hyperbolic functions) are very similar, so I implemented both
at first, but I didn't use cordit2 after all :(

Move fpu_cordit2() to #ifdef CORDIC_BOOTSTRAP section.
This reduces ~2KB text segment.
Reported by Krister Walfridsson on tech-kern two months ago.

branches: 1.24.12; 1.24.16;
Introduce the CORDIC algorithm.
o sine and cosine (e.g., FSIN, FCOS and FSINCOS instructions) is now
calculated in the CORDIC instead of Taylor expansion.
o tangent (FTAN) is not touched from a viewpoint of the code size.
o The CORDIC is applicable for hyperbolic functions (e.g., FSINH,
FCOSH, FTANH instructions), but I didn't use it because its working
range is poor.
o The CORDIC is also usable for inverse trigonometric functions,
I will commit it at next phase.
o The code size becomes a bit big. I cannot evaluate speed on m68k
for some reasons, but in test on i386 the CORDIC is approximately
100 times faster in sin/cos.

Introduce FPU_CONST_* constants to avoid a magic number.

u_int -> uint32_t

const-ify.

Remove 'register'.

branches: 1.19.2; 1.19.12;
- remove variable names from function declarations
- some KNF

Add hyperboric and trigonometric functions to m68k FPE, written by isaki@.
With these emulations (~4KB text) xeyes on XM6i works better.
Discussed with isaki@ at OSC 2011 Hiroshima.

Use static inline structure assignment for CPYFPN(). From isaki@.
Tested on XM6i.

fix indent. no functional changes.

Remove all the __P() from sys (excluding sys/dist)
Diff checked with grep and MK1 eyeball.
i386 and amd64 GENERIC and sys still build.

branches: 1.14.2;
fix include path in previous

Move FPSR_ register bit definitions into their own file

branches: 1.12.78; 1.12.86; 1.12.88;
merge ktrace-lwp.

Include <sys/time.h> before <sys/signalvar.h> so that copyinout_t is
defined before it is used in the second of these.

Fixes a build problem for several of our m68k ports. I've been
waiting for a better fix, but now my "proper fix timeout" triggered.

branches: 1.10.16;
SA_SIGINFO support for m68k

branches: 1.9.8; 1.9.24;
Fix typo. s/extention/extension/

rename round() to fpu_round() to avoid a name conflict with
the "round" assembly function in FPSP when building with ELF.

fpu_emulate.h:
the fmovecr constant table has the internal format
of the constants. So, when changing the mantissa size by a
non-multiple of 32 bits, we'd have to change this table, too. As
all other code changes just chopped of the least significand
32bit word of the mantissa, we correct the mantissa size instead
to (115 - 32 == 83) bits.
fpu_fmovecr.c:
put a safety belt in, to catch the next person who doesn't know this.
fpu_int.c:
in one place, the reduction of the mantissa size was overlooked.
fpu_log.c:
as the most significand 32bit word of the mantissa was changed back to the
old format, change back the table indexing code, too.

This should fix PR 11045.

Correct comments.

branches: 1.5.2; 1.5.12;
PR 7220 from Ken Nakata:
I've fixed most (not all) m68k FPE bugs that give bogus
calculation results, esp. fsqrt instruction. Also, the internal FP
representation has been reduced from 115-bit mantissa to 67-bit
mantissa which reduced the required mantissa operation roughly by one
fourth. I've done an extensive (though not exhaustive - it's
impossible!) test on the internal routines by feeding them randomly
generated FP numbers, and found that the new code is more precise than
MC68040 FPU (it seems to have a rounding bug).

[ Only change was to keep fpu_calcea.c's name instead of renaming to
fpu_ea.c in Ken's patch. --akb ]

branches: 1.4.24; 1.4.26; 1.4.28;
Update for -Wall -Wstrict-prototypes -Wmissing-prototypes -Wno-uninitialized.

Remove explicit #define of DEBUG, per PR #1931.

Fixes from Ken Nakata. Mainly renaming debug_level to fpu_debug_level.

Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.

exp(>11356) is +inf even if extended precision.
exp(<-11401) is 0 even if extended precision.

Fix sign of zero in case of x > -(2^18).
# By the way, I will modify this case later.

Improve the exponential and hyperbolic function's performance
10..100 times faster.
PR port-m68k/51645 from rin@ (and modified by me)

branches: 1.8.12; 1.8.16;
Terminate a loop in EXT_FRACBITS(64bits) instead of FP_NMANT(83bits).
I don't know why the mantissa of the accumulator is 83bits, but 64bits
or more are not affected.

Break a loop off to avoid a long loop even if the precision is not enough.

Implement exponential and power functions.
o FETOX .. exp(x)
o FETOXM1 .. exp(x) - 1
o FTENTOX .. 10^x
o FTWOTOX .. 2^x
At last all mathematics functions of FPE were implemented.
Thanks to Yosuke Sugahara.

branches: 1.5.2; 1.5.12;
fix indent. no functional changes.

Change about 4500 of the K&R function definitions to ANSI ones.
There are still about 1600 left, but they have ',' or /* ... */
in the actual variable definitions - which my awk script doesn't handle.
There are also many that need () -> (void).
(The script does handle misordered arguments.)

branches: 1.3.78; 1.3.86; 1.3.92;
merge ktrace-lwp.

__KERNEL_RCSID()

branches: 1.1.64;
Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.

Remove incorrect byte and word conversions from fpu_explode.
The correct operation here is arithmetic right shift, but nobody calls it.

branches: 1.15.30;
For the extended precision, the MSB of the mantissa is an integer
part, and this bit must be ignored at Infinity.
found by tests/lib/libc/stdlib/t_strtod.c::strtold_inf.

branches: 1.14.12;
u_int -> uint32_t

const-ify.

Remove 'register'.

branches: 1.11.2; 1.11.12;
fix indent. no functional changes.

ANSIfy another 1261 function definitions.
The only ones left in sys are beyond by sed script!
(or in sys/dist or sys/external)
Mostly they have function pointer parameters.

Change about 4500 of the K&R function definitions to ANSI ones.
There are still about 1600 left, but they have ',' or /* ... */
in the actual variable definitions - which my awk script doesn't handle.
There are also many that need () -> (void).
(The script does handle misordered arguments.)

Remove all the __P() from sys (excluding sys/dist)
Diff checked with grep and MK1 eyeball.
i386 and amd64 GENERIC and sys still build.

branches: 1.7.78; 1.7.86; 1.7.92;
merge ktrace-lwp.

While I'm here, retire the FPE code's own copy of ieee.h.

Move UCB-licensed code from 4-clause to 3-clause licence.

Patches provided by Joel Baker in PR 22364, verified by myself.

__KERNEL_RCSID()

branches: 1.3.36;
PR 7220 from Ken Nakata:
I've fixed most (not all) m68k FPE bugs that give bogus
calculation results, esp. fsqrt instruction. Also, the internal FP
representation has been reduced from 115-bit mantissa to 67-bit
mantissa which reduced the required mantissa operation roughly by one
fourth. I've done an extensive (though not exhaustive - it's
impossible!) test on the internal routines by feeding them randomly
generated FP numbers, and found that the new code is more precise than
MC68040 FPU (it seems to have a rounding bug).

[ Only change was to keep fpu_calcea.c's name instead of renaming to
fpu_ea.c in Ken's patch. --akb ]

branches: 1.2.24; 1.2.26; 1.2.28;
Update for -Wall -Wstrict-prototypes -Wmissing-prototypes -Wno-uninitialized.

Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.

u_int -> uint32_t

Use the #error directive.

branches: 1.14.2; 1.14.12;
fix indent again.
- "Second level indents are four spaces." pointed out by tsutsui@
- fold long line.

fix indent. no functional changes.

Change about 4500 of the K&R function definitions to ANSI ones.
There are still about 1600 left, but they have ',' or /* ... */
in the actual variable definitions - which my awk script doesn't handle.
There are also many that need () -> (void).
(The script does handle misordered arguments.)

branches: 1.11.78; 1.11.86; 1.11.92;
merge ktrace-lwp.

__KERNEL_RCSID()

branches: 1.9.6;
remove trailing \n in panic(). approved perry.

branches: 1.8.4; 1.8.8;
fpu_emulate.h:
the fmovecr constant table has the internal format
of the constants. So, when changing the mantissa size by a
non-multiple of 32 bits, we'd have to change this table, too. As
all other code changes just chopped of the least significand
32bit word of the mantissa, we correct the mantissa size instead
to (115 - 32 == 83) bits.
fpu_fmovecr.c:
put a safety belt in, to catch the next person who doesn't know this.
fpu_int.c:
in one place, the reduction of the mantissa size was overlooked.
fpu_log.c:
as the most significand 32bit word of the mantissa was changed back to the
old format, change back the table indexing code, too.

This should fix PR 11045.

branches: 1.7.2; 1.7.12;
PR 7220 from Ken Nakata:
I've fixed most (not all) m68k FPE bugs that give bogus
calculation results, esp. fsqrt instruction. Also, the internal FP
representation has been reduced from 115-bit mantissa to 67-bit
mantissa which reduced the required mantissa operation roughly by one
fourth. I've done an extensive (though not exhaustive - it's
impossible!) test on the internal routines by feeding them randomly
generated FP numbers, and found that the new code is more precise than
MC68040 FPU (it seems to have a rounding bug).

[ Only change was to keep fpu_calcea.c's name instead of renaming to
fpu_ea.c in Ken's patch. --akb ]

branches: 1.6.22; 1.6.24; 1.6.26;
backout previous kprintf change

printf -> kprintf, sprintf -> ksprintf

Update for -Wall -Wstrict-prototypes -Wmissing-prototypes -Wno-uninitialized.

sys/types.h -> sys/param.h and remove include of stddef.h.

Fixes from Ken Nakata. Mainly renaming debug_level to fpu_debug_level.

Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.

m68k/fpe: Avoid an illegal mod/reg before decoding it.
This also works for machines with FPU.
If the kernel defines FPU_EMULATE, even illegal FPU instructions in the real
6888x machines, the FPE will be executed by F line exception. So the FPE
must raise SIGILL appropriately for these.

branches: 1.16.72;
u_int -> uint32_t

branches: 1.15.2; 1.15.12;
fix indent again.
- "Second level indents are four spaces." pointed out by tsutsui@
- fold long line.

fix indent. no functional changes.

Change about 4500 of the K&R function definitions to ANSI ones.
There are still about 1600 left, but they have ',' or /* ... */
in the actual variable definitions - which my awk script doesn't handle.
There are also many that need () -> (void).
(The script does handle misordered arguments.)

branches: 1.12.78; 1.12.86; 1.12.92;
merge ktrace-lwp.

__KERNEL_RCSID()

branches: 1.10.36;
PR 7220 from Ken Nakata:
I've fixed most (not all) m68k FPE bugs that give bogus
calculation results, esp. fsqrt instruction. Also, the internal FP
representation has been reduced from 115-bit mantissa to 67-bit
mantissa which reduced the required mantissa operation roughly by one
fourth. I've done an extensive (though not exhaustive - it's
impossible!) test on the internal routines by feeding them randomly
generated FP numbers, and found that the new code is more precise than
MC68040 FPU (it seems to have a rounding bug).

[ Only change was to keep fpu_calcea.c's name instead of renaming to
fpu_ea.c in Ken's patch. --akb ]

branches: 1.9.8; 1.9.10; 1.9.12;
Back out last change, which was a thinko.

Add ftwotox emulation, by using the fscale emulation routine (preload
destination register bit pattern with 1.0), which automatically provides
corner case handling.
Missing ftwotox emulation originally reported by Norman Mackenzie in PR 4237,
but he proposed a different implementation.

Don't forget to set the scale variable even if we dont need to sign-extend it.
From PR 4236 by Norman Mackenzie.

branches: 1.6.10;
backout previous kprintf change

printf -> kprintf, sprintf -> ksprintf

Fix "might be used uninitialized" warnings.

Update for -Wall -Wstrict-prototypes -Wmissing-prototypes -Wno-uninitialized.

Fixes from Ken Nakata. Mainly renaming debug_level to fpu_debug_level.

Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.

m68k/fpe: Remove redundant code.
This was assigned before the function was called.

m68k/fpe: Check an illegal mod/reg before decoding it.
This avoids a kernel panic if an instruction has illegal mod/reg bits
like FMOVE.X FPn,#imm (Of course normally assemblers never emit these).
XXX Other instructions probably need such treatment...

m68k/fpe: Add a missing return.
FMOVE.P FPn,<ea> now raises SIGFPE. (Our FPE doesn't support .P)

branches: 1.14.72;
u_int -> uint32_t

branches: 1.13.2; 1.13.12;
fix indent again.
- "Second level indents are four spaces." pointed out by tsutsui@
- fold long line.

fix indent. no functional changes.

Change about 4500 of the K&R function definitions to ANSI ones.
There are still about 1600 left, but they have ',' or /* ... */
in the actual variable definitions - which my awk script doesn't handle.
There are also many that need () -> (void).
(The script does handle misordered arguments.)

branches: 1.10.44; 1.10.52; 1.10.58;
Make these compile with options DEBUG_FPE.

branches: 1.9.24; 1.9.26;
merge ktrace-lwp.

branches: 1.8.14; 1.8.16;
__KERNEL_RCSID()

branches: 1.7.36;
PR 7220 from Ken Nakata:
I've fixed most (not all) m68k FPE bugs that give bogus
calculation results, esp. fsqrt instruction. Also, the internal FP
representation has been reduced from 115-bit mantissa to 67-bit
mantissa which reduced the required mantissa operation roughly by one
fourth. I've done an extensive (though not exhaustive - it's
impossible!) test on the internal routines by feeding them randomly
generated FP numbers, and found that the new code is more precise than
MC68040 FPU (it seems to have a rounding bug).

[ Only change was to keep fpu_calcea.c's name instead of renaming to
fpu_ea.c in Ken's patch. --akb ]

branches: 1.6.14; 1.6.16; 1.6.18;
Fix my own PR port-m68k/3871, nee port-amiga/3871:
- don't erase FPSR exception bits _after_ doing most of the operations in
fpu_implode(), erase them before doing arith and store operations. This fixes
losing the DZ bit.
- create FPSR_OVFL and FPSR_UNFL bits in fpu_implode(). This showed up when
the first error was fixed.
XXX some more work needs to be done. E.g., creating OPERR together with
OVFL looks bogus, but I'm too tired know to re-check docs; and at least we
pass our own regression tests know.

backout previous kprintf change

printf -> kprintf, sprintf -> ksprintf

Update for -Wall -Wstrict-prototypes -Wmissing-prototypes -Wno-uninitialized.

Fixes from Ken Nakata. Mainly renaming debug_level to fpu_debug_level.

Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.

fix indent. no functional changes.

Change about 4500 of the K&R function definitions to ANSI ones.
There are still about 1600 left, but they have ',' or /* ... */
in the actual variable definitions - which my awk script doesn't handle.
There are also many that need () -> (void).
(The script does handle misordered arguments.)

branches: 1.3.78; 1.3.86; 1.3.92;
merge ktrace-lwp.

__KERNEL_RCSID()

branches: 1.1.64;
Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.

Improve the exponential and hyperbolic function's performance
10..100 times faster.
PR port-m68k/51645 from rin@ (and modified by me)

branches: 1.16.6; 1.16.10;
FATANH(-0) is -0, not +0.
This bug was introduced by the mistake of my test program.
Last week, I talked about the probelm (my FATANH(-0) returns +0 ?)
by NetBSD seminar at Open Source Conference 2013 Hiroshima,
and tsutsui@ confirmed that returns -0.

branches: 1.15.4;
Support sinh(-0.0) and tanh(-0.0).

Clean up some useless codes.

Fix typo in comment.

Terminate a loop in EXT_FRACBITS(64bits) instead of FP_NMANT(83bits).
I don't know why the mantissa of the accumulator is 83bits, but 64bits
or more are not affected.

Break a loop off to avoid a long loop even if the precision is not enough.

Implement a hyperbolic arctangent (FATANH).

Introduce the CORDIC algorithm.
o sine and cosine (e.g., FSIN, FCOS and FSINCOS instructions) is now
calculated in the CORDIC instead of Taylor expansion.
o tangent (FTAN) is not touched from a viewpoint of the code size.
o The CORDIC is applicable for hyperbolic functions (e.g., FSINH,
FCOSH, FTANH instructions), but I didn't use it because its working
range is poor.
o The CORDIC is also usable for inverse trigonometric functions,
I will commit it at next phase.
o The code size becomes a bit big. I cannot evaluate speed on m68k
for some reasons, but in test on i386 the CORDIC is approximately
100 times faster in sin/cos.

Introduce FPU_CONST_* constants to avoid a magic number.

Remove about updating fpsr. It was introduced by me but obviously
duplicated with fpu_emul_arith().

branches: 1.6.2; 1.6.12;
Add hyperboric and trigonometric functions to m68k FPE, written by isaki@.
With these emulations (~4KB text) xeyes on XM6i works better.
Discussed with isaki@ at OSC 2011 Hiroshima.

fix indent. no functional changes.

Change about 4500 of the K&R function definitions to ANSI ones.
There are still about 1600 left, but they have ',' or /* ... */
in the actual variable definitions - which my awk script doesn't handle.
There are also many that need () -> (void).
(The script does handle misordered arguments.)

branches: 1.3.78; 1.3.86; 1.3.92;
merge ktrace-lwp.

__KERNEL_RCSID()

branches: 1.1.64;
Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.

u_int -> uint32_t

Fix fpu_ftox().
Update not only exponential but also mantissa when an integer part
becomes 2 by rounding up. Without this fix, the extended precision
value becomes 0.0 because mantissa (including explicit integer bit)
is all-zero.

Remove 'register'.

branches: 1.12.2; 1.12.12;
fix indent. no functional changes.

Change about 4500 of the K&R function definitions to ANSI ones.
There are still about 1600 left, but they have ',' or /* ... */
in the actual variable definitions - which my awk script doesn't handle.
There are also many that need () -> (void).
(The script does handle misordered arguments.)

Remove all the __P() from sys (excluding sys/dist)
Diff checked with grep and MK1 eyeball.
i386 and amd64 GENERIC and sys still build.

branches: 1.9.78; 1.9.86; 1.9.92;
merge ktrace-lwp.

While I'm here, retire the FPE code's own copy of ieee.h.

Move UCB-licensed code from 4-clause to 3-clause licence.

Patches provided by Joel Baker in PR 22364, verified by myself.

__KERNEL_RCSID()

branches: 1.5.24;
rename round() to fpu_round() to avoid a name conflict with
the "round" assembly function in FPSP when building with ELF.

branches: 1.4.2;
PR 7220 from Ken Nakata:
I've fixed most (not all) m68k FPE bugs that give bogus
calculation results, esp. fsqrt instruction. Also, the internal FP
representation has been reduced from 115-bit mantissa to 67-bit
mantissa which reduced the required mantissa operation roughly by one
fourth. I've done an extensive (though not exhaustive - it's
impossible!) test on the internal routines by feeding them randomly
generated FP numbers, and found that the new code is more precise than
MC68040 FPU (it seems to have a rounding bug).

[ Only change was to keep fpu_calcea.c's name instead of renaming to
fpu_ea.c in Ken's patch. --akb ]

branches: 1.3.14; 1.3.16; 1.3.18;
Fix my own PR port-m68k/3871, nee port-amiga/3871:
- don't erase FPSR exception bits _after_ doing most of the operations in
fpu_implode(), erase them before doing arith and store operations. This fixes
losing the DZ bit.
- create FPSR_OVFL and FPSR_UNFL bits in fpu_implode(). This showed up when
the first error was fixed.
XXX some more work needs to be done. E.g., creating OPERR together with
OVFL looks bogus, but I'm too tired know to re-check docs; and at least we
pass our own regression tests know.

Update for -Wall -Wstrict-prototypes -Wmissing-prototypes -Wno-uninitialized.

Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.

Remove 'register'.

branches: 1.11.2;
Rewrite fpu_int().
Especially, remove the special treatment when |x| < 1
because it forgets to consider FPCR round mode.
See PR/46627 for the detail. Thanks Y.Sugahara for advice.

branches: 1.10.2; 1.10.8;
fix indent again.
- "Second level indents are four spaces." pointed out by tsutsui@
- fold long line.

fix indent. no functional changes.

Change about 4500 of the K&R function definitions to ANSI ones.
There are still about 1600 left, but they have ',' or /* ... */
in the actual variable definitions - which my awk script doesn't handle.
There are also many that need () -> (void).
(The script does handle misordered arguments.)

branches: 1.7.78; 1.7.86; 1.7.92;
merge ktrace-lwp.

__KERNEL_RCSID()

branches: 1.5.6;
Fix undefined operation (= is not a sequence point).

branches: 1.4.4; 1.4.8; 1.4.16;
rename round() to fpu_round() to avoid a name conflict with
the "round" assembly function in FPSP when building with ELF.

fpu_emulate.h:
the fmovecr constant table has the internal format
of the constants. So, when changing the mantissa size by a
non-multiple of 32 bits, we'd have to change this table, too. As
all other code changes just chopped of the least significand
32bit word of the mantissa, we correct the mantissa size instead
to (115 - 32 == 83) bits.
fpu_fmovecr.c:
put a safety belt in, to catch the next person who doesn't know this.
fpu_int.c:
in one place, the reduction of the mantissa size was overlooked.
fpu_log.c:
as the most significand 32bit word of the mantissa was changed back to the
old format, change back the table indexing code, too.

This should fix PR 11045.

branches: 1.2.2; 1.2.12;
PR 7220 from Ken Nakata:
I've fixed most (not all) m68k FPE bugs that give bogus
calculation results, esp. fsqrt instruction. Also, the internal FP
representation has been reduced from 115-bit mantissa to 67-bit
mantissa which reduced the required mantissa operation roughly by one
fourth. I've done an extensive (though not exhaustive - it's
impossible!) test on the internal routines by feeding them randomly
generated FP numbers, and found that the new code is more precise than
MC68040 FPU (it seems to have a rounding bug).

[ Only change was to keep fpu_calcea.c's name instead of renaming to
fpu_ea.c in Ken's patch. --akb ]

branches: 1.1.24; 1.1.26; 1.1.28;
Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.

FLOGNP1(-0.0) is -0.0, not +0.0. Found by XM6i.

branches: 1.17.4;
Clean up some useless codes.

Introduce FPU_CONST_* constants to avoid a magic number.

u_int -> uint32_t

branches: 1.14.2; 1.14.12;
fix indent again.
- "Second level indents are four spaces." pointed out by tsutsui@
- fold long line.

fix indent. no functional changes.

ANSIfy another 1261 function definitions.
The only ones left in sys are beyond by sed script!
(or in sys/dist or sys/external)
Mostly they have function pointer parameters.

Change about 4500 of the K&R function definitions to ANSI ones.
There are still about 1600 left, but they have ',' or /* ... */
in the actual variable definitions - which my awk script doesn't handle.
There are also many that need () -> (void).
(The script does handle misordered arguments.)

Remove all the __P() from sys (excluding sys/dist)
Diff checked with grep and MK1 eyeball.
i386 and amd64 GENERIC and sys still build.

branches: 1.9.78; 1.9.86; 1.9.92;
merge ktrace-lwp.

__KERNEL_RCSID()

branches: 1.7.24;
fpu_emulate.h:
the fmovecr constant table has the internal format
of the constants. So, when changing the mantissa size by a
non-multiple of 32 bits, we'd have to change this table, too. As
all other code changes just chopped of the least significand
32bit word of the mantissa, we correct the mantissa size instead
to (115 - 32 == 83) bits.
fpu_fmovecr.c:
put a safety belt in, to catch the next person who doesn't know this.
fpu_int.c:
in one place, the reduction of the mantissa size was overlooked.
fpu_log.c:
as the most significand 32bit word of the mantissa was changed back to the
old format, change back the table indexing code, too.

This should fix PR 11045.

branches: 1.6.2; 1.6.12;
PR 7220 from Ken Nakata:
I've fixed most (not all) m68k FPE bugs that give bogus
calculation results, esp. fsqrt instruction. Also, the internal FP
representation has been reduced from 115-bit mantissa to 67-bit
mantissa which reduced the required mantissa operation roughly by one
fourth. I've done an extensive (though not exhaustive - it's
impossible!) test on the internal routines by feeding them randomly
generated FP numbers, and found that the new code is more precise than
MC68040 FPU (it seems to have a rounding bug).

[ Only change was to keep fpu_calcea.c's name instead of renaming to
fpu_ea.c in Ken's patch. --akb ]

branches: 1.5.22; 1.5.24; 1.5.26;
backout previous kprintf change

printf -> kprintf, sprintf -> ksprintf

Update for -Wall -Wstrict-prototypes -Wmissing-prototypes -Wno-uninitialized.

Fixes from Ken Nakata. Mainly renaming debug_level to fpu_debug_level.

Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.

Fix sign of NAN. Found by XM6i.

branches: 1.8.12; 1.8.16;
u_int -> uint32_t

Remove 'register'.

branches: 1.6.12; 1.6.22;
Change about 4500 of the K&R function definitions to ANSI ones.
There are still about 1600 left, but they have ',' or /* ... */
in the actual variable definitions - which my awk script doesn't handle.
There are also many that need () -> (void).
(The script does handle misordered arguments.)

branches: 1.5.78; 1.5.86; 1.5.92;
merge ktrace-lwp.

Move UCB-licensed code from 4-clause to 3-clause licence.

Patches provided by Joel Baker in PR 22364, verified by myself.

__KERNEL_RCSID()

branches: 1.2.36;
PR 7220 from Ken Nakata:
I've fixed most (not all) m68k FPE bugs that give bogus
calculation results, esp. fsqrt instruction. Also, the internal FP
representation has been reduced from 115-bit mantissa to 67-bit
mantissa which reduced the required mantissa operation roughly by one
fourth. I've done an extensive (though not exhaustive - it's
impossible!) test on the internal routines by feeding them randomly
generated FP numbers, and found that the new code is more precise than
MC68040 FPU (it seems to have a rounding bug).

[ Only change was to keep fpu_calcea.c's name instead of renaming to
fpu_ea.c in Ken's patch. --akb ]

branches: 1.1.24; 1.1.26; 1.1.28;
Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.

m68k: Remove an unused variable since rev 1.1.
Detected by clang15 (nono emulator has imported and used this FPE).

Fix typo in comment. pointed out by Y.Sugahara.

branches: 1.16.10;
Revise the algorithm after Step3.
almost written by Y.Sugahara and minor modify by me.
This works for all input of FMOD/FREM and of course solves PR/47810.

R := X even if L < 0 in Step2, according to algorithm of the comment.
This solves many cases of |X| < |Y|.

Rename modrem -> is_mod for a readability.

u_int -> uint32_t

Fix a wrong "raw" comparison in step3.2.
It should solve a PR kern/47692.

branches: 1.11.2; 1.11.12;
fix indent again.
- "Second level indents are four spaces." pointed out by tsutsui@
- fold long line.

fix indent. no functional changes.

ANSIfy another 1261 function definitions.
The only ones left in sys are beyond by sed script!
(or in sys/dist or sys/external)
Mostly they have function pointer parameters.

Change about 4500 of the K&R function definitions to ANSI ones.
There are still about 1600 left, but they have ',' or /* ... */
in the actual variable definitions - which my awk script doesn't handle.
There are also many that need () -> (void).
(The script does handle misordered arguments.)

Remove all the __P() from sys (excluding sys/dist)
Diff checked with grep and MK1 eyeball.
i386 and amd64 GENERIC and sys still build.

branches: 1.6.78; 1.6.86; 1.6.92;
merge ktrace-lwp.

__KERNEL_RCSID()

branches: 1.4.36;
PR 7220 from Ken Nakata:
I've fixed most (not all) m68k FPE bugs that give bogus
calculation results, esp. fsqrt instruction. Also, the internal FP
representation has been reduced from 115-bit mantissa to 67-bit
mantissa which reduced the required mantissa operation roughly by one
fourth. I've done an extensive (though not exhaustive - it's
impossible!) test on the internal routines by feeding them randomly
generated FP numbers, and found that the new code is more precise than
MC68040 FPU (it seems to have a rounding bug).

[ Only change was to keep fpu_calcea.c's name instead of renaming to
fpu_ea.c in Ken's patch. --akb ]

branches: 1.3.24; 1.3.26; 1.3.28;
Bring back Step9 and fix an oversight from Ken Nakata <kenn@remus.rutgers.edu>

Update for -Wall -Wstrict-prototypes -Wmissing-prototypes -Wno-uninitialized.

Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.

s/equivalant/equivalent/ in copy pasted comment.

u_int -> uint32_t

Remove 'register'.

branches: 1.6.12; 1.6.22;
Change about 4500 of the K&R function definitions to ANSI ones.
There are still about 1600 left, but they have ',' or /* ... */
in the actual variable definitions - which my awk script doesn't handle.
There are also many that need () -> (void).
(The script does handle misordered arguments.)

branches: 1.5.78; 1.5.86; 1.5.92;
merge ktrace-lwp.

Move UCB-licensed code from 4-clause to 3-clause licence.

Patches provided by Joel Baker in PR 22364, verified by myself.

__KERNEL_RCSID()

branches: 1.2.36;
PR 7220 from Ken Nakata:
I've fixed most (not all) m68k FPE bugs that give bogus
calculation results, esp. fsqrt instruction. Also, the internal FP
representation has been reduced from 115-bit mantissa to 67-bit
mantissa which reduced the required mantissa operation roughly by one
fourth. I've done an extensive (though not exhaustive - it's
impossible!) test on the internal routines by feeding them randomly
generated FP numbers, and found that the new code is more precise than
MC68040 FPU (it seems to have a rounding bug).

[ Only change was to keep fpu_calcea.c's name instead of renaming to
fpu_ea.c in Ken's patch. --akb ]

branches: 1.1.24; 1.1.26; 1.1.28;
Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.

M68010 -> __mc68010__, pointed out by tsutsui@

Rewrite around BFFFO inline asm.
o Prepare C version of BFFFO (from XM6i).
It is helpful in running FPE on other platforms (for example, for a
test). It is also helpful in porting to non-m68k 3rd party :)
o A BFFFO is supported on 68020 or later (though I'm not sure whether
sun2/68010 uses this FPE correctly or not).

u_int -> uint32_t

Remove 'register'.

branches: 1.8.112; 1.8.122;
Remove leading __ from __(const|inline|signed|volatile) -- it is obsolete.

merge ktrace-lwp.

branches: 1.6.16;
Move UCB-licensed code from 4-clause to 3-clause licence.

Patches provided by Joel Baker in PR 22364, verified by myself.

__KERNEL_RCSID()

branches: 1.4.2;
Use "__asm __volatile" instead of "asm volatile" to pacify lint.

branches: 1.3.20;
PR 7220 from Ken Nakata:
I've fixed most (not all) m68k FPE bugs that give bogus
calculation results, esp. fsqrt instruction. Also, the internal FP
representation has been reduced from 115-bit mantissa to 67-bit
mantissa which reduced the required mantissa operation roughly by one
fourth. I've done an extensive (though not exhaustive - it's
impossible!) test on the internal routines by feeding them randomly
generated FP numbers, and found that the new code is more precise than
MC68040 FPU (it seems to have a rounding bug).

[ Only change was to keep fpu_calcea.c's name instead of renaming to
fpu_ea.c in Ken's patch. --akb ]

branches: 1.2.24; 1.2.26; 1.2.28;
Update for -Wall -Wstrict-prototypes -Wmissing-prototypes -Wno-uninitialized.

Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.

FSINCOS: Fix register address which writes cosine value back.

branches: 1.17.2;
Modify fpu_sin()'s logic to avoid GCC's warning that has been
pointed out in the previous commit.
For fpu_cos() there is no such problem, but sync with fpu_sin().

branches: 1.16.2;
avoid a GCC warning with this:

+#if defined(__GNUC__) && (__GNUC__ >= 5) && defined(__OPTIMIZE__)
+ x.fp_sign = 0;
+#endif

(ridiculous, but seems better than disabling the warning entirely.)

branches: 1.15.12;
Support sin(-0.0).

Clean up some useless codes.

Support tan(-0.0).

Implement inverse trigonometric functions (i.e., FACOS, FASIN, FATAN
instructions).
o arccos is calculated using arcsin.
o arcsin is calculated using arctan.
o arctan is calculated by the CORDIC.

Introduce the CORDIC algorithm.
o sine and cosine (e.g., FSIN, FCOS and FSINCOS instructions) is now
calculated in the CORDIC instead of Taylor expansion.
o tangent (FTAN) is not touched from a viewpoint of the code size.
o The CORDIC is applicable for hyperbolic functions (e.g., FSINH,
FCOSH, FTANH instructions), but I didn't use it because its working
range is poor.
o The CORDIC is also usable for inverse trigonometric functions,
I will commit it at next phase.
o The code size becomes a bit big. I cannot evaluate speed on m68k
for some reasons, but in test on i386 the CORDIC is approximately
100 times faster in sin/cos.

Improve how to use cmp/sub.

Introduce FPU_CONST_* constants to avoid a magic number.

u_int -> uint32_t

Remove about updating fpsr. It was introduced by me but obviously
duplicated with fpu_emul_arith().

branches: 1.6.2; 1.6.12;
Add hyperboric and trigonometric functions to m68k FPE, written by isaki@.
With these emulations (~4KB text) xeyes on XM6i works better.
Discussed with isaki@ at OSC 2011 Hiroshima.

fix indent. no functional changes.

Change about 4500 of the K&R function definitions to ANSI ones.
There are still about 1600 left, but they have ',' or /* ... */
in the actual variable definitions - which my awk script doesn't handle.
There are also many that need () -> (void).
(The script does handle misordered arguments.)

branches: 1.3.78; 1.3.86; 1.3.92;
merge ktrace-lwp.

__KERNEL_RCSID()

branches: 1.1.64;
Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.

While I'm here, retire the FPE code's own copy of ieee.h.

Move UCB-licensed code from 4-clause to 3-clause licence.

Patches provided by Joel Baker in PR 22364, verified by myself.

branches: 1.1.64;
Still incomplete, but much more complete FPE from Ken Nakata
<kenn@remus.rutgers.edu>. This emulator does not yet emulate
the following functions:
FSINH, FETOXM1, FTANH, FATAN, FASIN, FATANH, FSIN, FTAN,
FETOX, FTWOTOX, FTENTOX, FCOSH, FACOS, FCOS, FSINCOS
It is sufficient, however, to allow programs like df, w, and newfs,
to run to completion with correct results.
Portions of this code were based on the sparc fpe and on initial
work by gwr.