For unsignad integer division gcc used to emit a call to __udivsi3
"millicode" function that uses compiler-private ABI. Newer gcc uses
heavily tuned __udivsi3_i4i that is NOT compatible with __udivsi3
because it's expected to clobber different registers. We don't want
to link the kernel against libgcc and we don't have resources to write
heavily tuned version ourselves, so clone __udivsi3 but adjust it to
conform to the __udivsi3_i4i clobber spec.

Ditto for signed division.

You can make gcc use old routines with -mdiv=call-div1 to avoid few
extra instructions to save/restore the right registers in the signed
division funcion.

Move that majority of the 'SRCS+= foo.c' into the main Makefile.libkern.
Any .S files added by the arch/*/Makefile.inc cause the .c file to
be excluded.
Specific exclusions added using NO_SRCS to match previous files.
At least sparc, sparc64, i386, amd64 and vax GENERIC still build.
(There is a fubar with the naming of the byte-swap files ...)

Split the lines that add .c to SRCS from those that add .S
No changes (apart from the order) intended.

branches: 1.16.36;
Provide __movmemSI12_i4 (gcc4) a.k.a. __movstrSI12_i4 (gcc3) that gcc
emits for sh4 to copy 12 bytes at certain optimization levels.
Discovered by compiling landisk kernel with -Os.

Use gcc3 naming ("movstr") to match the rest of the "movstr" files.

branches: 1.15.24; 1.15.26; 1.15.30; 1.15.32;
merge ktrace-lwp.

branches: 1.14.16;
Use assembler version of bswap64.

Eliminate bzero.

Nuke bcopy. Move the code to memmove.S (which it is) make memmove the
primary entry point and comment out the bcopy entry point.

Kernels before and after the change are identical.

no need for bcmp.c

branches: 1.10.6;
Use assembly version of bzero() and memset().

add strtoul.c, it's now used in MI code.

Add __movstr_i4_{odd,even} for -m4.
Written by SHIMIZU Ryo.

Use assembly version of ffs(3).

branches: 1.6.2; 1.6.4; 1.6.16;
Add a strcasecmp(), modified from chopps's strncasecmp().

compile byte_swap_2.S and byte_swap_4.S

branches: 1.4.2;
new bcopy.S

branches: 1.3.2;
asm bcopy() for sh3

add strncasecmp.c

branches: 1.1.2; 1.1.4; 1.1.6;
Merge in NetBSD/sh3 from cvs.kame.net repository.

Tree structure:
- sys/arch/sh3: sh3 generic code
As commented, in-chip device drivers are put into sys/arch/sh3/dev.
- sys/arch/evbsh3: sh3 evaluation boards (pure sh3 CPU, no fancy external HW)
- sys/arch/mmeye: Brains mmEye, www.brains.co.jp
MI source code includes couple of #ifdef for sh3-coff support.
(sh3 uses coff or elf)

Needs some more improvements, especialy in sys/arch/sh3/conf/files.sh3,
to compile the tree (due to last minute tree structure change).

These copy of gcc millicode routines is now included in librump (that
sucks in all libkern sources), so mark them ".hidden". I'm not sure
if this is the best course of action (dropping millicode from librump
might be a better idea), but it's the quickest fix to get sh3 builds
going again for now.

branches: 1.6.2; 1.6.58; 1.6.62; 1.6.72;
G/c #ifdef PIC we inherited from the userland.

branches: 1.5.2; 1.5.4;
libc gets gcc millicode from libgcc, so we don't need to keep it
in src/common. Move it to sys/lib/libkern/arch/sh3.

branches: 1.4.4; 1.4.6; 1.4.8;
Use common sources with userland.

branches: 1.3.40;
don't profile (use NENTRY() and ALTENTRY())

_C_LABEL()

branches: 1.1.2;
Merge in NetBSD/sh3 from cvs.kame.net repository.

Tree structure:
- sys/arch/sh3: sh3 generic code
As commented, in-chip device drivers are put into sys/arch/sh3/dev.
- sys/arch/evbsh3: sh3 evaluation boards (pure sh3 CPU, no fancy external HW)
- sys/arch/mmeye: Brains mmEye, www.brains.co.jp
MI source code includes couple of #ifdef for sh3-coff support.
(sh3 uses coff or elf)

Needs some more improvements, especialy in sys/arch/sh3/conf/files.sh3,
to compile the tree (due to last minute tree structure change).

These copy of gcc millicode routines is now included in librump (that
sucks in all libkern sources), so mark them ".hidden". I'm not sure
if this is the best course of action (dropping millicode from librump
might be a better idea), but it's the quickest fix to get sh3 builds
going again for now.

branches: 1.5.2; 1.5.58; 1.5.62; 1.5.72;
G/c #ifdef PIC we inherited from the userland.

branches: 1.4.2; 1.4.4;
libc gets gcc millicode from libgcc, so we don't need to keep it
in src/common. Move it to sys/lib/libkern/arch/sh3.

branches: 1.3.4; 1.3.6; 1.3.8;
Use common sources with userland.

branches: 1.2.40;
_C_LABEL()

branches: 1.1.2;
Merge in NetBSD/sh3 from cvs.kame.net repository.

Tree structure:
- sys/arch/sh3: sh3 generic code
As commented, in-chip device drivers are put into sys/arch/sh3/dev.
- sys/arch/evbsh3: sh3 evaluation boards (pure sh3 CPU, no fancy external HW)
- sys/arch/mmeye: Brains mmEye, www.brains.co.jp
MI source code includes couple of #ifdef for sh3-coff support.
(sh3 uses coff or elf)

Needs some more improvements, especialy in sys/arch/sh3/conf/files.sh3,
to compile the tree (due to last minute tree structure change).

These copy of gcc millicode routines is now included in librump (that
sucks in all libkern sources), so mark them ".hidden". I'm not sure
if this is the best course of action (dropping millicode from librump
might be a better idea), but it's the quickest fix to get sh3 builds
going again for now.

branches: 1.5.2; 1.5.58; 1.5.62; 1.5.72;
G/c #ifdef PIC we inherited from the userland.

branches: 1.4.2; 1.4.4;
libc gets gcc millicode from libgcc, so we don't need to keep it
in src/common. Move it to sys/lib/libkern/arch/sh3.

branches: 1.3.4; 1.3.6; 1.3.8;
Use common sources with userland.

branches: 1.2.40;
_C_LABEL()

branches: 1.1.2;
Merge in NetBSD/sh3 from cvs.kame.net repository.

Tree structure:
- sys/arch/sh3: sh3 generic code
As commented, in-chip device drivers are put into sys/arch/sh3/dev.
- sys/arch/evbsh3: sh3 evaluation boards (pure sh3 CPU, no fancy external HW)
- sys/arch/mmeye: Brains mmEye, www.brains.co.jp
MI source code includes couple of #ifdef for sh3-coff support.
(sh3 uses coff or elf)

Needs some more improvements, especialy in sys/arch/sh3/conf/files.sh3,
to compile the tree (due to last minute tree structure change).

Nuke bcopy. Move the code to memmove.S (which it is) make memmove the
primary entry point and comment out the bcopy entry point.

Kernels before and after the change are identical.

branches: 1.3.6;
memcpy() and memmove() must return the first parameter.
Problem found by itohy, fixed by SHIMIZU Ryo.

branches: 1.2.4; 1.2.6;
new bcopy.S

branches: 1.1.2;
asm bcopy() for sh3

Always _LOCORE is defined for standalone program.

branches: 1.5.2; 1.5.4; 1.5.6;
Truncate the argument to 16 bit with extu.w

branches: 1.4.2; 1.4.4; 1.4.6;
merge ktrace-lwp.

branches: 1.3.16;
Compile identity ntoh*/hton* only for big endian case.

branches: 1.2.24;
support "big endian", too.

branches: 1.1.2; 1.1.12;
Merge in NetBSD/sh3 from cvs.kame.net repository.

Tree structure:
- sys/arch/sh3: sh3 generic code
As commented, in-chip device drivers are put into sys/arch/sh3/dev.
- sys/arch/evbsh3: sh3 evaluation boards (pure sh3 CPU, no fancy external HW)
- sys/arch/mmeye: Brains mmEye, www.brains.co.jp
MI source code includes couple of #ifdef for sh3-coff support.
(sh3 uses coff or elf)

Needs some more improvements, especialy in sys/arch/sh3/conf/files.sh3,
to compile the tree (due to last minute tree structure change).

Always _LOCORE is defined for standalone program.

branches: 1.4.4; 1.4.6; 1.4.8; 1.4.10; 1.4.12;
merge ktrace-lwp.

branches: 1.3.16;
Compile identity ntoh*/hton* only for big endian case.

branches: 1.2.24;
support "big endian", too.

branches: 1.1.2; 1.1.12;
Merge in NetBSD/sh3 from cvs.kame.net repository.

Tree structure:
- sys/arch/sh3: sh3 generic code
As commented, in-chip device drivers are put into sys/arch/sh3/dev.
- sys/arch/evbsh3: sh3 evaluation boards (pure sh3 CPU, no fancy external HW)
- sys/arch/mmeye: Brains mmEye, www.brains.co.jp
MI source code includes couple of #ifdef for sh3-coff support.
(sh3 uses coff or elf)

Needs some more improvements, especialy in sys/arch/sh3/conf/files.sh3,
to compile the tree (due to last minute tree structure change).

merge ktrace-lwp.

branches: 1.1.4;
bswap64 written in assembler.

Eliminate bzero.

branches: 1.1.2; 1.1.8;
Assembly version of bzero()/memset().
Written by SHIMIZU Ryo.

Use common sources with userland.

branches: 1.4.2; 1.4.24;
Save 1-4 instructions on all cases except for the ret=0 case.

This is probably the last version from me. :)
You are welcome to speed it up, of course. :)

Here's a benchmark on SH-4 200MHz.
9.2% faster if all the cases occur evenly.

return value C version previous vers this version speed ratio
of ffs() (ns/call) *1 (ns/call) (ns/call) *2 (*1/*2)

branches: 1.3.2;
Slightly improved version of ffs(3).
Partially from SHIMIZU Ryo <ryo@iij.ad.jp>. Thanks.
Some cases are slower, but other most cases are faster.

Here's a benchmark on SH-4 200MHz.

return value C version previous vers this version speed ratio
of ffs() (ns/call) *1 (ns/call) (ns/call) *2 (*1/*2)

branches: 1.2.2;
Oops, SYSLIBC_SCCS -> LIBC_SCCS

Assembly version of ffs(3).
Confirmed to return the same value as that of the C version.

The results of a simple benchmark on SH-4 200MHz, is shown below.
I think this shows acceptable performance.

return value C version this version speed
of ffs() (ns/call) (ns/call) ratio

These copy of gcc millicode routines is now included in librump (that
sucks in all libkern sources), so mark them ".hidden". I'm not sure
if this is the best course of action (dropping millicode from librump
might be a better idea), but it's the quickest fix to get sh3 builds
going again for now.

branches: 1.5.2; 1.5.58; 1.5.62; 1.5.72;
G/c #ifdef PIC we inherited from the userland.

branches: 1.4.2; 1.4.4;
libc gets gcc millicode from libgcc, so we don't need to keep it
in src/common. Move it to sys/lib/libkern/arch/sh3.

branches: 1.3.4; 1.3.6; 1.3.8;
Use common sources with userland.

branches: 1.2.40;
_C_LABEL()

branches: 1.1.2;
Merge in NetBSD/sh3 from cvs.kame.net repository.

Tree structure:
- sys/arch/sh3: sh3 generic code
As commented, in-chip device drivers are put into sys/arch/sh3/dev.
- sys/arch/evbsh3: sh3 evaluation boards (pure sh3 CPU, no fancy external HW)
- sys/arch/mmeye: Brains mmEye, www.brains.co.jp
MI source code includes couple of #ifdef for sh3-coff support.
(sh3 uses coff or elf)

Needs some more improvements, especialy in sys/arch/sh3/conf/files.sh3,
to compile the tree (due to last minute tree structure change).

Use common sources with userland.

branches: 1.2.4; 1.2.44;
new bcopy.S

branches: 1.1.2;
asm bcopy() for sh3

Use common sources with userland.

merge ktrace-lwp.

branches: 1.4.16;
Get rid of bcopy relics. Minimize diffs against the libc version.

Nuke bcopy. Move the code to memmove.S (which it is) make memmove the
primary entry point and comment out the bcopy entry point.

Kernels before and after the change are identical.

branches: 1.2.4; 1.2.28;
new bcopy.S

branches: 1.1.2;
asm bcopy() for sh3

Use common sources with userland.

branches: 1.1.2; 1.1.24;
Assembly version of bzero()/memset().
Written by SHIMIZU Ryo.

These copy of gcc millicode routines is now included in librump (that
sucks in all libkern sources), so mark them ".hidden". I'm not sure
if this is the best course of action (dropping millicode from librump
might be a better idea), but it's the quickest fix to get sh3 builds
going again for now.

branches: 1.8.2; 1.8.58; 1.8.62; 1.8.72;
Provide __movmem* aliases to __movstr* functions.
Gcc4 uses movmem, older versions use movstr.

G/c #ifdef PIC we inherited from the userland.

branches: 1.6.2; 1.6.4;
libc gets gcc millicode from libgcc, so we don't need to keep it
in src/common. Move it to sys/lib/libkern/arch/sh3.

branches: 1.5.4; 1.5.6; 1.5.8;
Use common sources with userland.

merge ktrace-lwp.

branches: 1.3.4;
nuke trailing whitespace

branches: 1.2.24; 1.2.32; 1.2.34;
_C_LABEL()

branches: 1.1.2;
Merge in NetBSD/sh3 from cvs.kame.net repository.

Tree structure:
- sys/arch/sh3: sh3 generic code
As commented, in-chip device drivers are put into sys/arch/sh3/dev.
- sys/arch/evbsh3: sh3 evaluation boards (pure sh3 CPU, no fancy external HW)
- sys/arch/mmeye: Brains mmEye, www.brains.co.jp
MI source code includes couple of #ifdef for sh3-coff support.
(sh3 uses coff or elf)

Needs some more improvements, especialy in sys/arch/sh3/conf/files.sh3,
to compile the tree (due to last minute tree structure change).

unused file (we have an asm version)

Merge in NetBSD/sh3 from cvs.kame.net repository.

Tree structure:
- sys/arch/sh3: sh3 generic code
As commented, in-chip device drivers are put into sys/arch/sh3/dev.
- sys/arch/evbsh3: sh3 evaluation boards (pure sh3 CPU, no fancy external HW)
- sys/arch/mmeye: Brains mmEye, www.brains.co.jp
MI source code includes couple of #ifdef for sh3-coff support.
(sh3 uses coff or elf)

Needs some more improvements, especialy in sys/arch/sh3/conf/files.sh3,
to compile the tree (due to last minute tree structure change).

These copy of gcc millicode routines is now included in librump (that
sucks in all libkern sources), so mark them ".hidden". I'm not sure
if this is the best course of action (dropping millicode from librump
might be a better idea), but it's the quickest fix to get sh3 builds
going again for now.

branches: 1.7.2; 1.7.58; 1.7.62; 1.7.72;
Provide __movmem* aliases to __movstr* functions.
Gcc4 uses movmem, older versions use movstr.

G/c #ifdef PIC we inherited from the userland.

branches: 1.5.2; 1.5.4;
libc gets gcc millicode from libgcc, so we don't need to keep it
in src/common. Move it to sys/lib/libkern/arch/sh3.

branches: 1.4.4; 1.4.6; 1.4.8;
Use common sources with userland.

branches: 1.3.40;
don't profile (use NENTRY() and ALTENTRY())

ENTRY()

branches: 1.1.4; 1.1.6;
movstrSI*() for gcc -Os

These copy of gcc millicode routines is now included in librump (that
sucks in all libkern sources), so mark them ".hidden". I'm not sure
if this is the best course of action (dropping millicode from librump
might be a better idea), but it's the quickest fix to get sh3 builds
going again for now.

branches: 1.1.4; 1.1.6; 1.1.10; 1.1.12; 1.1.22; 1.1.42; 1.1.46; 1.1.56;
Provide __movmemSI12_i4 (gcc4) a.k.a. __movstrSI12_i4 (gcc3) that gcc
emits for sh4 to copy 12 bytes at certain optimization levels.
Discovered by compiling landisk kernel with -Os.

Use gcc3 naming ("movstr") to match the rest of the "movstr" files.

These copy of gcc millicode routines is now included in librump (that
sucks in all libkern sources), so mark them ".hidden". I'm not sure
if this is the best course of action (dropping millicode from librump
might be a better idea), but it's the quickest fix to get sh3 builds
going again for now.

branches: 1.5.2; 1.5.58; 1.5.62; 1.5.72;
Provide __movmem* aliases to __movstr* functions.
Gcc4 uses movmem, older versions use movstr.

G/c #ifdef PIC we inherited from the userland.

branches: 1.3.2; 1.3.4;
libc gets gcc millicode from libgcc, so we don't need to keep it
in src/common. Move it to sys/lib/libkern/arch/sh3.

branches: 1.2.4; 1.2.6; 1.2.8;
Use common sources with userland.

branches: 1.1.2; 1.1.4; 1.1.26;
Add __movstr_i4_{odd,even} for -m4.
Written by SHIMIZU Ryo.

These copy of gcc millicode routines is now included in librump (that
sucks in all libkern sources), so mark them ".hidden". I'm not sure
if this is the best course of action (dropping millicode from librump
might be a better idea), but it's the quickest fix to get sh3 builds
going again for now.

branches: 1.5.2; 1.5.58; 1.5.62; 1.5.72;
G/c #ifdef PIC we inherited from the userland.

branches: 1.4.2; 1.4.4;
libc gets gcc millicode from libgcc, so we don't need to keep it
in src/common. Move it to sys/lib/libkern/arch/sh3.

branches: 1.3.4; 1.3.6; 1.3.8;
Use common sources with userland.

branches: 1.2.40;
_C_LABEL()

branches: 1.1.2;
Merge in NetBSD/sh3 from cvs.kame.net repository.

Tree structure:
- sys/arch/sh3: sh3 generic code
As commented, in-chip device drivers are put into sys/arch/sh3/dev.
- sys/arch/evbsh3: sh3 evaluation boards (pure sh3 CPU, no fancy external HW)
- sys/arch/mmeye: Brains mmEye, www.brains.co.jp
MI source code includes couple of #ifdef for sh3-coff support.
(sh3 uses coff or elf)

Needs some more improvements, especialy in sys/arch/sh3/conf/files.sh3,
to compile the tree (due to last minute tree structure change).

Reword the comment about "millicode" nature of these functions and add
it to sdivsi3.S too.

Remove !_KERNEL code left over from the days when we shared these
files with userland.

Use .L prefix for local label.

revert the _i4i aliases. they are not the right thing.

add weak aliases for __sdivsi3_i4i and __udivsi3_i4i.

These copy of gcc millicode routines is now included in librump (that
sucks in all libkern sources), so mark them ".hidden". I'm not sure
if this is the best course of action (dropping millicode from librump
might be a better idea), but it's the quickest fix to get sh3 builds
going again for now.

branches: 1.8.2; 1.8.58; 1.8.62; 1.8.72;
G/c #ifdef PIC we inherited from the userland.

branches: 1.7.2; 1.7.4;
libc gets gcc millicode from libgcc, so we don't need to keep it
in src/common. Move it to sys/lib/libkern/arch/sh3.

branches: 1.6.4; 1.6.6; 1.6.8;
Use common sources with userland.

merge ktrace-lwp.

branches: 1.4.16;
Move UCB-licensed code from 4-clause to 3-clause licence.

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

branches: 1.3.24;
check _STANDALONE macro, too.

branches: 1.2.4;
don't clobber r4 and r5

branches: 1.1.2;
Merge in NetBSD/sh3 from cvs.kame.net repository.

Tree structure:
- sys/arch/sh3: sh3 generic code
As commented, in-chip device drivers are put into sys/arch/sh3/dev.
- sys/arch/evbsh3: sh3 evaluation boards (pure sh3 CPU, no fancy external HW)
- sys/arch/mmeye: Brains mmEye, www.brains.co.jp
MI source code includes couple of #ifdef for sh3-coff support.
(sh3 uses coff or elf)

Needs some more improvements, especialy in sys/arch/sh3/conf/files.sh3,
to compile the tree (due to last minute tree structure change).

Reword the comment about "millicode" nature of these functions and add
it to sdivsi3_i4i.S too.

For unsignad integer division gcc used to emit a call to __udivsi3
"millicode" function that uses compiler-private ABI. Newer gcc uses
heavily tuned __udivsi3_i4i that is NOT compatible with __udivsi3
because it's expected to clobber different registers. We don't want
to link the kernel against libgcc and we don't have resources to write
heavily tuned version ourselves, so clone __udivsi3 but adjust it to
conform to the __udivsi3_i4i clobber spec.

Ditto for signed division.

You can make gcc use old routines with -mdiv=call-div1 to avoid few
extra instructions to save/restore the right registers in the signed
division funcion.

Reword the comment about "millicode" nature of these functions and add
it to sdivsi3.S too.

Remove !_KERNEL code left over from the days when we shared these
files with userland.

Use .L prefix for local label.

revert the _i4i aliases. they are not the right thing.

add weak aliases for __sdivsi3_i4i and __udivsi3_i4i.

These copy of gcc millicode routines is now included in librump (that
sucks in all libkern sources), so mark them ".hidden". I'm not sure
if this is the best course of action (dropping millicode from librump
might be a better idea), but it's the quickest fix to get sh3 builds
going again for now.

branches: 1.7.2; 1.7.58; 1.7.62; 1.7.72;
G/c #ifdef PIC we inherited from the userland.

branches: 1.6.2; 1.6.4;
libc gets gcc millicode from libgcc, so we don't need to keep it
in src/common. Move it to sys/lib/libkern/arch/sh3.

branches: 1.5.4; 1.5.6; 1.5.8;
Use common sources with userland.

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: 1.2.28;
check _STANDALONE

branches: 1.1.2; 1.1.8;
Merge in NetBSD/sh3 from cvs.kame.net repository.

Tree structure:
- sys/arch/sh3: sh3 generic code
As commented, in-chip device drivers are put into sys/arch/sh3/dev.
- sys/arch/evbsh3: sh3 evaluation boards (pure sh3 CPU, no fancy external HW)
- sys/arch/mmeye: Brains mmEye, www.brains.co.jp
MI source code includes couple of #ifdef for sh3-coff support.
(sh3 uses coff or elf)

Needs some more improvements, especialy in sys/arch/sh3/conf/files.sh3,
to compile the tree (due to last minute tree structure change).

Reword the comment about "millicode" nature of these functions and add
it to sdivsi3_i4i.S too.

For unsignad integer division gcc used to emit a call to __udivsi3
"millicode" function that uses compiler-private ABI. Newer gcc uses
heavily tuned __udivsi3_i4i that is NOT compatible with __udivsi3
because it's expected to clobber different registers. We don't want
to link the kernel against libgcc and we don't have resources to write
heavily tuned version ourselves, so clone __udivsi3 but adjust it to
conform to the __udivsi3_i4i clobber spec.

Ditto for signed division.

You can make gcc use old routines with -mdiv=call-div1 to avoid few
extra instructions to save/restore the right registers in the signed
division funcion.