view src/alloca.s @ 2367:ecf1ebac70d8

[xemacs-hg @ 2004-11-04 23:05:23 by ben] commit mega-patch configure.in: Turn off -Winline and -Wchar-subscripts. Use the right set of cflags when compiling modules. Rewrite ldap configuration to separate the inclusion of lber (needed in recent Cygwin) from the basic checks for the needed libraries. add a function for MAKE_JUNK_C; initially code was added to generate xemacs.def using this, but it will need to be rewritten. add an rm -f for junk.c to avoid weird Cygwin bug with cp -f onto an existing file. Sort list of auto-detected functions and eliminate unused checks for stpcpy, setlocale and getwd. Add autodetection of Cygwin scanf problems BETA: Rewrite section on configure to indicate what flags are important and what not. digest-doc.c, make-dump-id.c, profile.c, sorted-doc.c: Add proper decls for main(). make-msgfile.c: Document that this is old junk. Move proposal to text.c. make-msgfile.lex: Move proposal to text.c. make-mswin-unicode.pl: Convert error-generating code so that the entire message will be seen as a single unrecognized token. mule/mule-ccl.el: Update docs. lispref/mule.texi: Update CCL docs. ldap/eldap.c: Mule-ize. Use EXTERNAL_LIST_LOOP_2 instead of deleted EXTERNAL_LIST_LOOP. * XEmacs 21.5.18 "chestnut" is released. --------------------------------------------------------------- MULE-RELATED WORK: --------------------------------------------------------------- --------------------------- byte-char conversion --------------------------- buffer.c, buffer.h, insdel.c, text.c: Port FSF algorithm for byte-char conversion, replacing broken previous version. Track the char position of the gap. Add functions to do char-byte conversion downwards as well as upwards. Move comments about algorithm workings to internals manual. --------------------------- work on types --------------------------- alloc.c, console-x-impl.h, dump-data.c, dump-data.h, dumper.c, dialog-msw.c, dired-msw.c, doc.c, editfns.c, esd.c, event-gtk.h, event-msw.c, events.c, file-coding.c, file-coding.h, fns.c, glyphs-eimage.c, glyphs-gtk.c, glyphs-msw.c, glyphs-shared.c, glyphs-x.c, glyphs.c, glyphs.h, gui.c, hpplay.c, imgproc.c, intl-win32.c, lrecord.h, lstream.c, keymap.c, lisp.h, libsst.c, linuxplay.c, miscplay.c, miscplay.h, mule-coding.c, nas.c, nt.c, ntheap.c, ntplay.c, objects-msw.c, objects-tty.c, objects-x.c, print.c, process-nt.c, process.c, redisplay.h, select-common.h, select-gtk.c, select-x.c, sgiplay.c, sound.c, sound.h, sunplay.c, sysfile.h, sysdep.c, syswindows.h, text.c, unexnt.c, win32.c, xgccache.c: Further work on types. This creates a full set of types for all the basic semantics of `char' that I have so far identified, so that its semantics can always be identified for the purposes of proper Mule-safe code, and the raw use of `char' always avoided. (1) More type renaming, for consistency of naming. Char_ASCII -> Ascbyte UChar_ASCII -> UAscbyte Char_Binary -> CBinbyte UChar_Binary -> Binbyte SChar_Binary -> SBinbyte (2) Introduce Rawbyte, CRawbyte, Boolbyte, Chbyte, UChbyte, and Bitbyte and use them. (3) New types Itext, Wexttext and Textcount for separating out the concepts of bytes and textual units (different under UTF-16 and UTF-32, which are potential internal encodings). (4) qxestr*_c -> qxestr*_ascii. lisp.h: New; goes with other qxe() functions. #### Maybe goes in a different section. lisp.h: Group generic int-type defs together with EMACS_INT defs. lisp.h: * lisp.h (WEXTTEXT_IS_WIDE) New defns. lisp.h: New type to replace places where int occurs as a boolean. It's signed because occasionally people may want to use -1 as an error value, and because unsigned ints are viral -- see comments in the internals manual against using them. dynarr.c: int -> Bytecount. --------------------------- Mule-izing --------------------------- device-x.c: Partially Mule-ize. dumper.c, dumper.h: Mule-ize. Use Rawbyte. Use stderr_out not printf. Use wext_*(). sysdep.c, syswindows.h, text.c: New Wexttext API for manipulation of external text that may be Unicode (e.g. startup code under Windows). emacs.c: Mule-ize. Properly deal with argv in external encoding. Use wext_*() and Wexttext. Use Rawbyte. #if 0 some old junk on SCO that is unlikely to be correct. Rewrite allocation code in run-temacs. emacs.c, symsinit.h, win32.c: Rename win32 init function and call it even earlier, to initialize mswindows_9x_p even earlier, for use in startup code (XEUNICODE_P). process.c: Use _wenviron not environ under Windows, to get Unicode environment variables. event-Xt.c: Mule-ize drag-n-drop related stuff. dragdrop.c, dragdrop.h, frame-x.c: Mule-ize. text.h: Add some more stand-in defines for particular kinds of conversion; use in Mule-ization work in frame-x.c etc. --------------------------- Freshening --------------------------- intl-auto-encap-win32.c, intl-auto-encap-win32.h: Regenerate. --------------------------- Unicode-work --------------------------- intl-win32.c, syswindows.h: Factor out common options to MultiByteToWideChar and WideCharToMultiByte. Add convert_unicode_to_multibyte_malloc() and convert_unicode_to_multibyte_dynarr() and use. Add stuff for alloca() conversion of multibyte/unicode. alloc.c: Use dfc_external_data_len() in case of unicode coding system. alloc.c, mule-charset.c: Don't zero out and reinit charset Unicode tables. This fucks up dump-time loading. Anyway, either we load them at dump time or run time, never both. unicode.c: Dump the blank tables as well. --------------------------------------------------------------- DOCUMENTATION, MOSTLY MULE-RELATED: --------------------------------------------------------------- EmacsFrame.c, emodules.c, event-Xt.c, fileio.c, input-method-xlib.c, mule-wnnfns.c, redisplay-gtk.c, redisplay-tty.c, redisplay-x.c, regex.c, sysdep.c: Add comment about Mule work needed. text.h: Add more documentation describing why DFC routines were not written to return their value. Add some other DFC documentation. console-msw.c, console-msw.h: Add pointer to docs in win32.c. emacs.c: Add comments on sources of doc info. text.c, charset.h, unicode.c, intl-win32.c, intl-encap-win32.c, text.h, file-coding.c, mule-coding.c: Collect background comments and related to text matters and internationalization, and proposals for work to be done, in text.c or Internals manual, stuff related to specific textual API's in text.h, and stuff related to internal implementation of Unicode conversion in unicode.c. Put lots of pointers to the comments to make them easier to find. s/mingw32.h, s/win32-common.h, s/win32-native.h, s/windowsnt.h, win32.c: Add bunches of new documentation on the different kinds of builds and environments under Windows and how they work. Collect this info in win32.c. Add pointers to these docs in the relevant s/* files. emacs.c: Document places with long comments. Remove comment about exiting, move to internals manual, put in pointer. event-stream.c: Move docs about event queues and focus to internals manual, put in pointer. events.h: Move docs about event stream callbacks to internals manual, put in pointer. profile.c, redisplay.c, signal.c: Move documentation to the Internals manual. process-nt.c: Add pointer to comment in win32-native.el. lisp.h: Add comments about some comment conventions. lisp.h: Add comment about the second argument. device-msw.c, redisplay-msw.c: @@#### comments are out-of-date. --------------------------------------------------------------- PDUMP WORK (MOTIVATED BY UNICODE CHANGES) --------------------------------------------------------------- alloc.c, buffer.c, bytecode.c, console-impl.h, console.c, device.c, dumper.c, lrecord.h, elhash.c, emodules.h, events.c, extents.c, frame.c, glyphs.c, glyphs.h, mule-charset.c, mule-coding.c, objects.c, profile.c, rangetab.c, redisplay.c, specifier.c, specifier.h, window.c, lstream.c, file-coding.h, file-coding.c: PDUMP: Properly implement dump_add_root_block(), which never worked before, and is necessary for dumping Unicode tables. Pdump name changes for accuracy: XD_STRUCT_PTR -> XD_BLOCK_PTR. XD_STRUCT_ARRAY -> XD_BLOCK_ARRAY. XD_C_STRING -> XD_ASCII_STRING. *_structure_* -> *_block_*. lrecord.h: some comments added about dump_add_root_block() vs dump_add_root_block_ptr(). extents.c: remove incorrect comment about pdump problems with gap array. --------------------------------------------------------------- ALLOCATION --------------------------------------------------------------- abbrev.c, alloc.c, bytecode.c, casefiddle.c, device-msw.c, device-x.c, dired-msw.c, doc.c, doprnt.c, dragdrop.c, editfns.c, emodules.c, file-coding.c, fileio.c, filelock.c, fns.c, glyphs-eimage.c, glyphs-gtk.c, glyphs-msw.c, glyphs-x.c, gui-msw.c, gui-x.c, imgproc.c, intl-win32.c, lread.c, menubar-gtk.c, menubar.c, nt.c, objects-msw.c, objects-x.c, print.c, process-nt.c, process-unix.c, process.c, realpath.c, redisplay.c, search.c, select-common.c, symbols.c, sysdep.c, syswindows.h, text.c, text.h, ui-byhand.c: New macros {alloca,xnew}_{itext,{i,ext,raw,bin,asc}bytes} for more convenient allocation of these commonly requested items. Modify functions to use alloca_ibytes, alloca_array, alloca_extbytes, xnew_ibytes, etc. also XREALLOC_ARRAY, xnew. alloc.c: Rewrite the allocation functions to factor out repeated code. Add assertions for freeing dumped data. lisp.h: Moved down and consolidated with other allocation stuff. lisp.h, dynarr.c: New functions for allocation that's very efficient when mostly in LIFO order. lisp.h, text.c, text.h: Factor out some stuff for general use by alloca()-conversion funs. text.h, lisp.h: Fill out convenience routines for allocating various kinds of bytes and put them in lisp.h. Use them in place of xmalloc(), ALLOCA(). text.h: Fill out the convenience functions so the _MALLOC() kinds match the alloca() kinds. --------------------------------------------------------------- ERROR-CHECKING --------------------------------------------------------------- text.h: Create ASSERT_ASCTEXT_ASCII() and ASSERT_ASCTEXT_ASCII_LEN() from similar Eistring checkers and change the Eistring checkers to use them instead. --------------------------------------------------------------- MACROS IN LISP.H --------------------------------------------------------------- lisp.h: Redo GCPRO declarations. Create a "base" set of functions that can be used to generate any kind of gcpro sets -- regular, ngcpro, nngcpro, private ones used in GC_EXTERNAL_LIST_LOOP_2. buffer.c, callint.c, chartab.c, console-msw.c, device-x.c, dialog-msw.c, dired.c, extents.c, ui-gtk.c, rangetab.c, nt.c, mule-coding.c, minibuf.c, menubar-msw.c, menubar.c, menubar-gtk.c, lread.c, lisp.h, gutter.c, glyphs.c, glyphs-widget.c, fns.c, fileio.c, file-coding.c, specifier.c: Eliminate EXTERNAL_LIST_LOOP, which does not check for circularities. Use EXTERNAL_LIST_LOOP_2 instead or EXTERNAL_LIST_LOOP_3 or EXTERNAL_PROPERTY_LIST_LOOP_3 or GC_EXTERNAL_LIST_LOOP_2 (new macro). Removed/redid comments on EXTERNAL_LIST_LOOP. --------------------------------------------------------------- SPACING FIXES --------------------------------------------------------------- callint.c, hftctl.c, number-gmp.c, process-unix.c: Spacing fixes. --------------------------------------------------------------- FIX FOR GEOMETRY PROBLEM IN FIRST FRAME --------------------------------------------------------------- unicode.c: Add workaround for newlib bug in sscanf() [should be fixed by release 1.5.12 of Cygwin]. toolbar.c: bug fix for problem of initial frame being 77 chars wide on Windows. will be overridden by my other ws. --------------------------------------------------------------- FIX FOR LEAKING PROCESS HANDLES: --------------------------------------------------------------- process-nt.c: Fixes for leaking handles. Inspired by work done by Adrian Aichner <adrian@xemacs.org>. --------------------------------------------------------------- FIX FOR CYGWIN BUG (Unicode-related): --------------------------------------------------------------- unicode.c: Add workaround for newlib bug in sscanf() [should be fixed by release 1.5.12 of Cygwin]. --------------------------------------------------------------- WARNING FIXES: --------------------------------------------------------------- console-stream.c: `reinit' is unused. compiler.h, event-msw.c, frame-msw.c, intl-encap-win32.c, text.h: Add stuff to deal with ANSI-aliasing warnings I got. regex.c: Gather includes together to avoid warning. --------------------------------------------------------------- CHANGES TO INITIALIZATION ROUTINES: --------------------------------------------------------------- buffer.c, emacs.c, console.c, debug.c, device-x.c, device.c, dragdrop.c, emodules.c, eval.c, event-Xt.c, event-gtk.c, event-msw.c, event-stream.c, event-tty.c, events.c, extents.c, faces.c, file-coding.c, fileio.c, font-lock.c, frame-msw.c, glyphs-widget.c, glyphs.c, gui-x.c, insdel.c, lread.c, lstream.c, menubar-gtk.c, menubar-x.c, minibuf.c, mule-wnnfns.c, objects-msw.c, objects.c, print.c, scrollbar-x.c, search.c, select-x.c, text.c, undo.c, unicode.c, window.c, symsinit.h: Call reinit_*() functions directly from emacs.c, for clarity. Factor out some redundant init code. Move disallowed stuff that had crept into vars_of_glyphs() into complex_vars_of_glyphs(). Call init_eval_semi_early() from eval.c not in the middle of vars_of_() in emacs.c since there should be no order dependency in the latter calls. --------------------------------------------------------------- ARMAGEDDON: --------------------------------------------------------------- alloc.c, emacs.c, lisp.h, print.c: Rename inhibit_non_essential_printing_operations to inhibit_non_essential_conversion_operations. text.c: Assert on !inhibit_non_essential_conversion_operations. console-msw.c, print.c: Don't do conversion in SetConsoleTitle or FindWindow to avoid problems during armageddon. Put #errors for NON_ASCII_INTERNAL_FORMAT in places where problems would arise. --------------------------------------------------------------- CHANGES TO THE BUILD PROCEDURE: --------------------------------------------------------------- config.h.in, s/cxux.h, s/usg5-4-2.h, m/powerpc.h: Add comment about correct ordering of this file. Rearrange everything to follow this -- put all #undefs together and before the s&m files. Add undefs for HAVE_ALLOCA, C_ALLOCA, BROKEN_ALLOCA_IN_FUNCTION_CALLS, STACK_DIRECTION. Remove unused HAVE_STPCPY, HAVE_GETWD, HAVE_SETLOCALE. m/gec63.h: Deleted; totally broken, not used at all, not in FSF. m/7300.h, m/acorn.h, m/alliant-2800.h, m/alliant.h, m/altos.h, m/amdahl.h, m/apollo.h, m/att3b.h, m/aviion.h, m/celerity.h, m/clipper.h, m/cnvrgnt.h, m/convex.h, m/cydra5.h, m/delta.h, m/delta88k.h, m/dpx2.h, m/elxsi.h, m/ews4800r.h, m/gould.h, m/hp300bsd.h, m/hp800.h, m/hp9000s300.h, m/i860.h, m/ibmps2-aix.h, m/ibmrs6000.h, m/ibmrt-aix.h, m/ibmrt.h, m/intel386.h, m/iris4d.h, m/iris5d.h, m/iris6d.h, m/irist.h, m/isi-ov.h, m/luna88k.h, m/m68k.h, m/masscomp.h, m/mg1.h, m/mips-nec.h, m/mips-siemens.h, m/mips.h, m/news.h, m/nh3000.h, m/nh4000.h, m/ns32000.h, m/orion105.h, m/pfa50.h, m/plexus.h, m/pmax.h, m/powerpc.h, m/pyrmips.h, m/sequent-ptx.h, m/sequent.h, m/sgi-challenge.h, m/symmetry.h, m/tad68k.h, m/tahoe.h, m/targon31.h, m/tekxd88.h, m/template.h, m/tower32.h, m/tower32v3.h, m/ustation.h, m/vax.h, m/wicat.h, m/xps100.h: Delete C_ALLOCA, HAVE_ALLOCA, STACK_DIRECTION, BROKEN_ALLOCA_IN_FUNCTION_CALLS. All of this is auto-detected. When in doubt, I followed recent FSF sources, which also have these things deleted.
author ben
date Thu, 04 Nov 2004 23:08:28 +0000
parents 3ecd8885ac67
children aa5ed11f473b
line wrap: on
line source

/* `alloca' standard 4.2 subroutine for 68000's and 16000's and others.
   Also has _setjmp and _longjmp for pyramids.
   Copyright (C) 1985, 1986, 1988 Free Software Foundation, Inc.

This file is part of XEmacs.

XEmacs is free software; you can redistribute it and/or modify it
under the terms of the GNU General Public License as published by the
Free Software Foundation; either version 2, or (at your option) any
later version.

XEmacs is distributed in the hope that it will be useful, but WITHOUT
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
for more details.

You should have received a copy of the GNU General Public License
along with XEmacs; see the file COPYING.  If not, write to
the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
Boston, MA 02111-1307, USA.  */

/* Synched up with: FSF 19.30. */

/* Both 68000 systems I have run this on have had broken versions of alloca.
   Also, I am told that non-berkeley systems do not have it at all.
   So replace whatever system-provided alloca there may be
   on all 68000 systems.  */

#define NOT_C_CODE
#ifdef emacs
#include <config.h>
#else
#include "config.h"
#endif

#ifndef HAVE_ALLOCA  /* define this to use system's alloca */

#ifndef hp9000s300
#ifndef m68k
#ifndef m68000
#ifndef WICAT
#ifndef ns32000
#ifndef ns16000
#ifndef sequent
#ifndef pyramid
#ifndef ATT3B5
#ifndef XENIX
you
lose!!
#endif /* XENIX */
#endif /* ATT3B5 */
#endif /* pyramid */
#endif /* sequent */
#endif /* ns16000 */
#endif /* ns32000 */
#endif /* WICAT */
#endif /* m68000 */
#endif /* m68k */
#endif /* hp9000s300 */


#ifdef hp9000s300
#ifdef OLD_HP_ASSEMBLER
	data
	text
	globl	_alloca
_alloca
	move.l	(sp)+,a0	; pop return addr from top of stack
	move.l	(sp)+,d0	; pop size in bytes from top of stack
	add.l	#ROUND,d0	; round size up to long word
	and.l	#MASK,d0	; mask out lower two bits of size
	sub.l	d0,sp		; allocate by moving stack pointer
	tst.b	PROBE(sp)	; stack probe to allocate pages
	move.l	sp,d0		; return pointer
	add.l	#-4,sp		; new top of stack
	jmp	(a0)		; not a normal return
MASK	equ	-4		; Longword alignment
ROUND	equ	3		; ditto
PROBE	equ	-128		; safety buffer for C compiler scratch
	data
#else /* new hp assembler syntax */
/*
  The new compiler does "move.m <registers> (%sp)" to save registers,
    so we must copy the saved registers when we mung the sp.
  The old compiler did "move.m <register> <offset>(%a6)", which
    gave us no trouble
 */
	text
	set	PROBE,-128	# safety for C frame temporaries
	set	MAXREG,22	# d2-d7, a2-a5, fp2-fp7 may have been saved
	global	_alloca
_alloca:
	mov.l	(%sp)+,%a0	# return address
	mov.l	(%sp)+,%d0	# number of bytes to allocate
	mov.l	%sp,%a1		# save old sp for register copy
	mov.l	%sp,%d1		# compute new sp
	sub.l	%d0,%d1		# space requested
	and.l	&-4,%d1		# round down to longword
	sub.l	&MAXREG*4,%d1	# space for saving registers
	mov.l	%d1,%sp		# save new value of sp
	tst.b	PROBE(%sp)	# create pages (sigh)
	mov.l	%a2,%d1		# save reg a2
	mov.l	%sp,%a2
	move.w	&MAXREG-1,%d0
copy_regs_loop:			/* save caller's saved registers */
	mov.l	(%a1)+,(%a2)+
	dbra	%d0,copy_regs_loop
	mov.l	%a2,%d0		# return value
	mov.l	%d1,%a2		# restore a2
	add.l	&-4,%sp		# adjust tos
	jmp	(%a0)		# rts
#endif /* new hp assembler */
#else
#ifdef m68k			/* SGS assembler totally different */
	file	"alloca.s"
	global	alloca
alloca:
#ifdef MOTOROLA_DELTA
/* slightly modified version of alloca to motorola sysV/68 pcc - based
   compiler.
   this compiler saves used registers relative to %sp instead of %fp.
   alright, just make new copy of saved register set whenever we allocate
   new space from stack..
   this is true at last until SVR3V7 . bug has reported to Motorola. */
	set	MAXREG,10	# max no of registers to save (d2-d7, a2-a5)
        mov.l   (%sp)+,%a1	# pop return addr from top of stack
        mov.l   (%sp)+,%d0	# pop size in bytes from top of stack
	mov.l	%sp,%a0		# save stack pointer for register copy
        addq.l  &3,%d0		# round size up to long word
        andi.l  &-4,%d0		# mask out lower two bits of size
	mov.l	%sp,%d1		# compute new value of sp to d1
        sub.l	%d0,%d1		# pseudo-allocate by moving stack pointer
	sub.l	&MAXREG*4,%d1	# allocate more space for saved regs.
	mov.l	%d1,%sp		# actual allocation.
	move.w	&MAXREG-1,%d0	# d0 counts saved regs.
	mov.l	%a2,%d1		# preserve a2.
	mov.l	%sp,%a2		# make pointer to new reg save area.
copy_regs_loop: 		# copy stuff from old save area.
	mov.l	(%a0)+,(%a2)+	# save saved register
	dbra	%d0,copy_regs_loop
        mov.l   %a2,%a0		# now a2 is start of allocated space.
	mov.l	%a2,%d0		# return it in both a0 and d0 to play safe.
	mov.l	%d1,%a2		# restore a2.
        subq.l  &4,%sp		# new top of stack
        jmp     (%a1)		# far below normal return
#else /* not MOTOROLA_DELTA */
	mov.l	(%sp)+,%a1	# pop return addr from top of stack
	mov.l	(%sp)+,%d0	# pop size in bytes from top of stack
	add.l	&R%1,%d0	# round size up to long word
	and.l	&-4,%d0		# mask out lower two bits of size
	sub.l	%d0,%sp		# allocate by moving stack pointer
	tst.b	P%1(%sp)	# stack probe to allocate pages
	mov.l	%sp,%a0		# return pointer as pointer
	mov.l	%sp,%d0		# return pointer as int to avoid disaster
	add.l	&-4,%sp		# new top of stack
	jmp	(%a1)		# not a normal return
	set	S%1,64		# safety factor for C compiler scratch
	set	R%1,3+S%1	# add to size for rounding
	set	P%1,-132	# probe this far below current top of stack
#endif /* not MOTOROLA_DELTA */

#else /* not m68k */

#ifdef m68000

#ifdef WICAT
/*
 * Registers are saved after the corresponding link so we have to explicitly
 * move them to the top of the stack where they are expected to be.
 * Since we do not know how many registers were saved in the calling function
 * we must assume the maximum possible (d2-d7,a2-a5).  Hence, we end up
 * wasting some space on the stack.
 *
 * The large probe (tst.b) attempts to make up for the fact that we have
 * potentially used up the space that the caller probed for its own needs.
 */
	.procss m0
	.config "68000 1"
	.module	_alloca
MAXREG:	.const	10
	.sect	text
	.global	_alloca
_alloca:
	move.l	(sp)+,a1	; pop return address
	move.l	(sp)+,d0	; pop allocation size
	move.l	sp,d1		; get current SP value
	sub.l	d0,d1		; adjust to reflect required size...
	sub.l	#MAXREG*4,d1	; ...and space needed for registers
	and.l	#-4,d1		; backup to longword boundary
	move.l	sp,a0		; save old SP value for register copy
	move.l	d1,sp		; set the new SP value
	tst.b	-4096(sp)	; grab an extra page (to cover caller)
	move.l	a2,d1		; save callers register
	move.l	sp,a2
	move.w	#MAXREG-1,d0	; # of longwords to copy
loop:	move.l	(a0)+,(a2)+	; copy registers...
	dbra	d0,loop		; ...til there are no more
	move.l	a2,d0		; end of register area is addr for new space
	move.l	d1,a2		; restore saved a2.
	addq.l	#4,sp		; caller will increment sp by 4 after return.
	move.l	d0,a0		; return value in both a0 and d0.
	jmp	(a1)
	.end	_alloca
#else

/* Some systems want the _, some do not.  Win with both kinds.  */
.globl	_alloca
_alloca:
.globl	alloca
alloca:
	movl	sp@+,a0
	movl	a7,d0
	subl	sp@,d0
	andl	#~3,d0
	movl	d0,sp
	tstb	sp@(0)		/* Make stack pages exist  */
				/* Needed on certain systems
				   that lack true demand paging */
	addql	#4,d0
	jmp	a0@

#endif /* not WICAT */
#endif /* m68000 */
#endif /* not m68k */
#endif /* not hp9000s300 */

#if defined (ns16000) || defined (ns32000)

	.text
	.align	2
/* Some systems want the _, some do not.  Win with both kinds.  */
.globl	_alloca
_alloca:
.globl	alloca
alloca:

/* Two different assembler syntaxes are used for the same code
	on different systems.  */

#ifdef sequent
#define IM
#define REGISTER(x) x
#else
#ifdef NS5   /* ns SysV assembler */
#define IM $
#define REGISTER(x) x
#else
#define IM $
#define REGISTER(x) 0(x)
#endif
#endif

/*
 * The ns16000 is a little more difficult, need to copy regs.
 * Also the code assumes direct linkage call sequence (no mod table crap).
 * We have to copy registers, and therefore waste 32 bytes.
 *
 * Stack layout:
 * new	sp ->	junk
 *	 	registers (copy)
 *	r0 ->	new data
 *		 | 	  (orig retval)
 *		 |	  (orig arg)
 * old  sp ->	regs	  (orig)
 *		local data
 *	fp ->	old fp
 */

	movd	tos,r1		/*  pop return addr */
	negd	tos,r0		/*  pop amount to allocate */
	sprd	sp,r2
	addd	r2,r0
	bicb	IM/**/3,r0	/*  4-byte align */
	lprd	sp,r0
	adjspb	IM/**/36	/*  space for regs, +4 for caller to pop */
	movmd	0(r2),4(sp),IM/**/4	/*  copy regs */
	movmd	0x10(r2),0x14(sp),IM/**/4
	jump	REGISTER(r1)	/* funky return */
#endif /* ns16000 or ns32000 */

#ifdef pyramid

.globl _alloca

_alloca: addw $3,pr0	# add 3 (dec) to first argument
	bicw $3,pr0	# then clear its last 2 bits
	subw pr0,sp	# subtract from SP the val in PR0
	andw $-32,sp	# keep sp aligned on multiple of 32.
	movw sp,pr0	# ret. current SP
	ret

#ifdef PYRAMID_OLD /* This isn't needed in system version 4.  */
.globl __longjmp
.globl _longjmp
.globl __setjmp
.globl _setjmp

__longjmp: jump _longjmp
__setjmp:  jump _setjmp
#endif

#endif /* pyramid */

#ifdef ATT3B5

	.align 4
	.globl alloca

alloca:
	movw %ap, %r8
	subw2 $9*4, %r8
	movw 0(%r8), %r1    /* pc */
	movw 4(%r8), %fp
	movw 8(%r8), %sp
	addw2 %r0, %sp /* make room */
	movw %sp, %r0 /* return value */
	jmp (%r1) /* continue... */

#endif /* ATT3B5 */

#ifdef XENIX

.386

_TEXT segment dword use32 public 'CODE'
assume   cs:_TEXT

;-------------------------------------------------------------------------

public _alloca
_alloca proc near

	pop	ecx		; return address
	pop	eax		; amount to alloc
	add	eax,3		; round it to 32-bit boundary
	and	al,11111100B	;
	mov	edx,esp		; current sp in edx
	sub	edx,eax		; lower the stack
	xchg	esp,edx		; start of allocation in esp, old sp in edx
	mov	eax,esp		; return ptr to base in eax
	push	[edx+8]		; save poss. stored reg. values (esi,edi,ebx)
	push	[edx+4]		;  on lowered stack
	push	[edx]		;
	sub	esp,4		; allow for 'add esp, 4'
	jmp	ecx		; jump to return address

_alloca endp

_TEXT	ends

end

#endif /* XENIX */

#endif /* not HAVE_ALLOCA */