x86-defs.m4

a very popular packet of cryptography tools,it encloses the most common used algorithm and protocols

divert(-1)

dnl  m4 macros for x86 assembler.


dnl  Copyright 1999, 2000, 2001, 2002 Free Software Foundation, Inc.
dnl 
dnl  This file is part of the GNU MP Library.
dnl
dnl  The GNU MP Library is free software; you can redistribute it and/or
dnl  modify it under the terms of the GNU Lesser General Public License as
dnl  published by the Free Software Foundation; either version 2.1 of the
dnl  License, or (at your option) any later version.
dnl
dnl  The GNU MP Library is distributed in the hope that it will be useful,
dnl  but WITHOUT ANY WARRANTY; without even the implied warranty of
dnl  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
dnl  Lesser General Public License for more details.
dnl
dnl  You should have received a copy of the GNU Lesser General Public
dnl  License along with the GNU MP Library; see the file COPYING.LIB.  If
dnl  not, write to the Free Software Foundation, Inc., 59 Temple Place -
dnl  Suite 330, Boston, MA 02111-1307, USA.


dnl  Notes:
dnl
dnl  m4 isn't perfect for processing BSD style x86 assembler code, the main
dnl  problems are,
dnl
dnl  1. Doing define(foo,123) and then using foo in an addressing mode like
dnl     foo(%ebx) expands as a macro rather than a constant.  This is worked
dnl     around by using deflit() from asm-defs.m4, instead of define().
dnl
dnl  2. Immediates in macro definitions need a space or `' to stop the $
dnl     looking like a macro parameter.  For example,
dnl
dnl  	        define(foo, `mov $ 123, %eax')
dnl
dnl     This is only a problem in macro definitions, not in ordinary text,
dnl     and not in macro parameters like text passed to forloop() or ifdef().


deflit(BYTES_PER_MP_LIMB, 4)


dnl  Libtool gives -DPIC -DDLL_EXPORT to indicate a cygwin or mingw DLL.  We
dnl  undefine PIC since we don't need to be position independent in this
dnl  case and definitely don't want the ELF style _GLOBAL_OFFSET_TABLE_ etc.

ifdef(`DLL_EXPORT',`undefine(`PIC')')


dnl  Called: PROLOGUE_cpu(GSYM_PREFIX`'foo)
dnl
dnl  In the x86 code we use explicit TEXT and ALIGN() calls in the code,
dnl  since different alignments are wanted in various circumstances.  So for
dnl  instance,
dnl
dnl                  TEXT
dnl                  ALIGN(16)
dnl          PROLOGUE(mpn_add_n)
dnl          ...
dnl          EPILOGUE()

define(`PROLOGUE_cpu',
m4_assert_numargs(1)
	`GLOBL	$1
	TYPE($1,`function')
$1:
ifelse(WANT_PROFILING,`no',,`call_mcount
')')


dnl  Usage: call_mcount
dnl
dnl  For `gprof' style profiling, %ebp is setup as a frame pointer.  None of
dnl  the assembler routines use %ebp this way, so it's done only for the
dnl  benefit of mcount.  glibc sysdeps/i386/i386-mcount.S shows how mcount
dnl  gets the current function from (%esp) and the parent from 4(%ebp).
dnl
dnl  For `prof' style profiling gcc generates mcount calls without setting
dnl  up %ebp, and the same is done here.

define(`call_mcount',
m4_assert_numargs(-1)
m4_assert_defined(`WANT_PROFILING')
m4_assert_defined(`MCOUNT_PIC_REG')
m4_assert_defined(`MCOUNT_NONPIC_REG')
m4_assert_defined(`MCOUNT_PIC_CALL')
m4_assert_defined(`MCOUNT_NONPIC_CALL')
`ifelse(ifdef(`PIC',`MCOUNT_PIC_REG',`MCOUNT_NONPIC_REG'),,,
`	DATA
	ALIGN(4)
L(mcount_data_`'mcount_data_counter):
	W32	0
	TEXT
')dnl
ifelse(WANT_PROFILING,`gprof',
`	pushl	%ebp
	movl	%esp, %ebp
')dnl
ifdef(`PIC',
`	pushl	%ebx
	mcount_movl_GOT_ebx
ifelse(MCOUNT_PIC_REG,,,
`	leal	L(mcount_data_`'mcount_data_counter)@GOTOFF(%ebx), MCOUNT_PIC_REG')
MCOUNT_PIC_CALL
	popl	%ebx
',`dnl non-PIC
ifelse(MCOUNT_NONPIC_REG,,,
`	movl	`$'L(mcount_data_`'mcount_data_counter), MCOUNT_NONPIC_REG
')dnl
MCOUNT_NONPIC_CALL
')dnl
ifelse(WANT_PROFILING,`gprof',
`	popl	%ebp
')
define(`mcount_data_counter',eval(mcount_data_counter+1))')

define(mcount_data_counter,1)

dnl  Called: mcount_movl_GOT_ebx
dnl  Label H is "here", the %eip obtained from the call.  C is the called
dnl  subroutine.  J is the jump across that subroutine.  A fetch and "ret"
dnl  is always done so calls and returns are balanced for the benefit of the
dnl  various x86s that have return stack branch prediction.
define(mcount_movl_GOT_ebx,
m4_assert_numargs(-1)
`	call	L(mcount_movl_GOT_ebx_C`'mcount_movl_GOT_ebx_counter)
L(mcount_movl_GOT_ebx_H`'mcount_movl_GOT_ebx_counter):
	jmp	L(mcount_movl_GOT_ebx_J`'mcount_movl_GOT_ebx_counter)
L(mcount_movl_GOT_ebx_C`'mcount_movl_GOT_ebx_counter):
	movl	(%esp), %ebx
	ret
L(mcount_movl_GOT_ebx_J`'mcount_movl_GOT_ebx_counter):
	addl	$_GLOBAL_OFFSET_TABLE_+[.-L(mcount_movl_GOT_ebx_H`'mcount_movl_GOT_ebx_counter)], %ebx
define(`mcount_movl_GOT_ebx_counter',incr(mcount_movl_GOT_ebx_counter))')

define(mcount_movl_GOT_ebx_counter,1)


dnl  --------------------------------------------------------------------------
dnl  Various x86 macros.
dnl


dnl  Usage: ALIGN_OFFSET(bytes,offset)
dnl
dnl  Align to `offset' away from a multiple of `bytes'.
dnl
dnl  This is useful for testing, for example align to something very strict
dnl  and see what effect offsets from it have, "ALIGN_OFFSET(256,32)".
dnl
dnl  Generally you wouldn't execute across the padding, but it's done with
dnl  nop's so it'll work.

define(ALIGN_OFFSET,
m4_assert_numargs(2)
`ALIGN($1)
forloop(`i',1,$2,`	nop
')')


dnl  Usage: defframe(name,offset)
dnl
dnl  Make a definition like the following with which to access a parameter
dnl  or variable on the stack.
dnl
dnl         define(name,`FRAME+offset(%esp)')
dnl
dnl  Actually m4_empty_if_zero(FRAME+offset) is used, which will save one
dnl  byte if FRAME+offset is zero, by putting (%esp) rather than 0(%esp).
dnl  Use define(`defframe_empty_if_zero_disabled',1) if for some reason the
dnl  zero offset is wanted.
dnl
dnl  The new macro also gets a check that when it's used FRAME is actually
dnl  defined, and that the final %esp offset isn't negative, which would
dnl  mean an attempt to access something below the current %esp.
dnl
dnl  deflit() is used rather than a plain define(), so the new macro won't
dnl  delete any following parenthesized expression.  name(%edi) will come
dnl  out say as 16(%esp)(%edi).  This isn't valid assembler and should
dnl  provoke an error, which is better than silently giving just 16(%esp).
dnl
dnl  See README for more on the suggested way to access the stack frame.

define(defframe,
m4_assert_numargs(2)
`deflit(`$1',
m4_assert_defined(`FRAME')
`defframe_check_notbelow(`$1',$2,FRAME)dnl
defframe_empty_if_zero(FRAME+($2))(%esp)')')

dnl  Called: defframe_empty_if_zero(expression)
define(defframe_empty_if_zero,
m4_assert_numargs(1)
`ifelse(defframe_empty_if_zero_disabled,1,
`eval($1)',
`m4_empty_if_zero($1)')')

dnl  Called: defframe_check_notbelow(`name',offset,FRAME)
define(defframe_check_notbelow,
m4_assert_numargs(3)
`ifelse(eval(($3)+($2)<0),1,
`m4_error(`$1 at frame offset $2 used when FRAME is only $3 bytes
')')')


dnl  Usage: FRAME_pushl()
dnl         FRAME_popl()
dnl         FRAME_addl_esp(n)
dnl         FRAME_subl_esp(n)
dnl
dnl  Adjust FRAME appropriately for a pushl or popl, or for an addl or subl
dnl  %esp of n bytes.
dnl
dnl  Using these macros is completely optional.  Sometimes it makes more
dnl  sense to put explicit deflit(`FRAME',N) forms, especially when there's
dnl  jumps and different sequences of FRAME values need to be used in
dnl  different places.

define(FRAME_pushl,
m4_assert_numargs(0)
m4_assert_defined(`FRAME')
`deflit(`FRAME',eval(FRAME+4))')

define(FRAME_popl,
m4_assert_numargs(0)
m4_assert_defined(`FRAME')
`deflit(`FRAME',eval(FRAME-4))')

define(FRAME_addl_esp,
m4_assert_numargs(1)
m4_assert_defined(`FRAME')
`deflit(`FRAME',eval(FRAME-($1)))')

define(FRAME_subl_esp,
m4_assert_numargs(1)
m4_assert_defined(`FRAME')
`deflit(`FRAME',eval(FRAME+($1)))')


dnl  Usage: defframe_pushl(name)
dnl
dnl  Do a combination FRAME_pushl() and a defframe() to name the stack
dnl  location just pushed.  This should come after a pushl instruction.
dnl  Putting it on the same line works and avoids lengthening the code.  For
dnl  example,
dnl
dnl         pushl   %eax     defframe_pushl(VAR_COUNTER)
dnl
dnl  Notice the defframe() is done with an unquoted -FRAME thus giving its
dnl  current value without tracking future changes.

define(defframe_pushl,
m4_assert_numargs(1)
`FRAME_pushl()defframe(`$1',-FRAME)')


dnl  --------------------------------------------------------------------------
dnl  Assembler instruction macros.
dnl


dnl  Usage: emms_or_femms
dnl         femms_available_p
dnl
dnl  femms_available_p expands to 1 or 0 according to whether the AMD 3DNow
dnl  femms instruction is available.  emms_or_femms expands to femms if
dnl  available, or emms if not.
dnl
dnl  emms_or_femms is meant for use in the K6 directory where plain K6
dnl  (without femms) and K6-2 and K6-3 (with a slightly faster femms) are
dnl  supported together.
dnl
dnl  On K7 femms is no longer faster and is just an alias for emms, so plain
dnl  emms may as well be used.

define(femms_available_p,
m4_assert_numargs(-1)
`m4_ifdef_anyof_p(
	`HAVE_HOST_CPU_k62',
	`HAVE_HOST_CPU_k63',
	`HAVE_HOST_CPU_athlon')')

define(emms_or_femms,
m4_assert_numargs(-1)
`ifelse(femms_available_p,1,`femms',`emms')')


dnl  Usage: femms
dnl
dnl  Gas 2.9.1 which comes with FreeBSD 3.4 doesn't support femms, so the
dnl  following is a replacement using .byte.
dnl
dnl  If femms isn't available, an emms is generated instead, for convenience
dnl  when testing on a machine without femms.

define(femms,
m4_assert_numargs(-1)
`ifelse(femms_available_p,1,
`.byte	15,14	C AMD 3DNow femms',
`emms`'dnl
m4_warning(`warning, using emms in place of femms, use for testing only
')')')


dnl  Usage: jadcl0(op)
dnl
dnl  Generate a jnc/incl as a substitute for adcl $0,op.  Note this isn't an
dnl  exact replacement, since it doesn't set the flags like adcl does.
dnl
dnl  This finds a use in K6 mpn_addmul_1, mpn_submul_1, mpn_mul_basecase and
dnl  mpn_sqr_basecase because on K6 an adcl is slow, the branch
dnl  misprediction penalty is small, and the multiply algorithm used leads
dnl  to a carry bit on average only 1/4 of the time.
dnl
dnl  jadcl0_disabled can be set to 1 to instead generate an ordinary adcl
dnl  for comparison.  For example,
dnl
dnl		define(`jadcl0_disabled',1)
dnl
dnl  When using a register operand, eg. "jadcl0(%edx)", the jnc/incl code is
dnl  the same size as an adcl.  This makes it possible to use the exact same
dnl  computed jump code when testing the relative speed of the two.

define(jadcl0,
m4_assert_numargs(1)
`ifelse(jadcl0_disabled,1,
	`adcl	$`'0, $1',
	`jnc	L(jadcl0_`'jadcl0_counter)
	incl	$1
L(jadcl0_`'jadcl0_counter):
define(`jadcl0_counter',incr(jadcl0_counter))')')

define(jadcl0_counter,1)


dnl  Usage: cmov_available_p
dnl
dnl  Expand to 1 if cmov is available, 0 if not.

define(cmov_available_p,
m4_assert_numargs(-1)
`m4_ifdef_anyof_p(
	`HAVE_HOST_CPU_pentiumpro',
	`HAVE_HOST_CPU_pentium2',
	`HAVE_HOST_CPU_pentium3',
	`HAVE_HOST_CPU_pentium4',
	`HAVE_HOST_CPU_athlon')')


dnl  Usage: x86_lookup(target, key,value, key,value, ...)
dnl         x86_lookup_p(target, key,value, key,value, ...)
dnl
dnl  Look for `target' among the `key' parameters.
dnl
dnl  x86_lookup expands to the corresponding `value', or generates an error
dnl  if `target' isn't found.
dnl
dnl  x86_lookup_p expands to 1 if `target' is found, or 0 if not.

define(x86_lookup,
m4_assert_numargs_range(1,999)
`ifelse(eval($#<3),1,
`m4_error(`unrecognised part of x86 instruction: $1
')',
`ifelse(`$1',`$2', `$3',
`x86_lookup(`$1',shift(shift(shift($@))))')')')

define(x86_lookup_p,
m4_assert_numargs_range(1,999)
`ifelse(eval($#<3),1, `0',
`ifelse(`$1',`$2',    `1',
`x86_lookup_p(`$1',shift(shift(shift($@))))')')')


dnl  Usage: x86_opcode_reg32(reg)
dnl         x86_opcode_reg32_p(reg)
dnl
dnl  x86_opcode_reg32 expands to the standard 3 bit encoding for the given
dnl  32-bit register, eg. `%ebp' turns into 5.
dnl
dnl  x86_opcode_reg32_p expands to 1 if reg is a valid 32-bit register, or 0
dnl  if not.

define(x86_opcode_reg32,
m4_assert_numargs(1)
`x86_lookup(`$1',x86_opcode_reg32_list)')