dnl x86 fat binary entrypoints.
dnl Contributed to the GNU project by Kevin Ryde (original x86_32 code) and
dnl Torbjorn Granlund (port to x86_64)
dnl Copyright 2003, 2009, 2011-2014, 2016 Free Software Foundation, Inc.
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 modify
dnl it under the terms of either:
dnl
dnl * the GNU Lesser General Public License as published by the Free
dnl Software Foundation; either version 3 of the License, or (at your
dnl option) any later version.
dnl
dnl or
dnl
dnl * the GNU General Public License as published by the Free Software
dnl Foundation; either version 2 of the License, or (at your option) any
dnl later version.
dnl
dnl or both in parallel, as here.
dnl
dnl The GNU MP Library is distributed in the hope that it will be useful, but
dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
dnl or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
dnl for more details.
dnl
dnl You should have received copies of the GNU General Public License and the
dnl GNU Lesser General Public License along with the GNU MP Library. If not,
dnl see https://www.gnu.org/licenses/.
include(`../config.m4')
dnl Forcibly disable profiling.
dnl
dnl The entrypoints and inits are small enough not to worry about, the real
dnl routines arrived at will have any profiling. Also, the way the code
dnl here ends with a jump means we won't work properly with the
dnl "instrument" profiling scheme anyway.
define(`WANT_PROFILING',no)
dnl We define PRETEND_PIC as a helper symbol, the use it for suppressing
dnl normal, fast call code, since that triggers problems on Darwin, OpenBSD
dnl and some versions of GNU/Linux. This will go away when symbol hiding is
dnl finished.
ifdef(`DARWIN',
`define(`PRETEND_PIC')')
ifdef(`OPENBSD',
`define(`PRETEND_PIC')')
ifdef(`LINUX',
`define(`PRETEND_PIC')')
ifdef(`PIC',
`define(`PRETEND_PIC')')
ABI_SUPPORT(DOS64)
ABI_SUPPORT(STD64)
TEXT
dnl Usage: FAT_ENTRY(name, offset)
dnl
dnl Emit a fat binary entrypoint function of the given name. This is the
dnl normal entry for applications, eg. __gmpn_add_n.
dnl
dnl The code simply jumps through the function pointer in __gmpn_cpuvec at
dnl the given "offset" (in bytes).
dnl
dnl For non-PIC, the jumps are 5 bytes each, aligning them to 8 should be
dnl fine for all x86s.
dnl
dnl For ELF/DARWIN PIC, the jumps are 20 bytes each, and are best aligned to
dnl 16 to ensure at least the first two instructions don't cross a cache line
dnl boundary.
dnl
dnl For DOS64, the jumps are 6 bytes. The same form works also for GNU/Linux
dnl (at least with certain assembler/linkers) but FreeBSD 8.2 crashes. Not
dnl tested on Darwin, Slowaris, NetBSD, etc.
dnl
dnl Note the extra `' ahead of PROLOGUE obscures it from the HAVE_NATIVE
dnl grepping in configure, stopping that code trying to eval something with
dnl $1 in it.
define(FAT_ENTRY,
m4_assert_numargs(2)
`ifdef(`HOST_DOS64',
` ALIGN(8)
`'PROLOGUE($1)
jmp *$2+GSYM_PREFIX`'__gmpn_cpuvec(%rip)
EPILOGUE()
',
` ALIGN(ifdef(`PIC',16,8))
`'PROLOGUE($1)
ifdef(`PRETEND_PIC',
` LEA( GSYM_PREFIX`'__gmpn_cpuvec, %rax)
jmp *$2(%rax)
',`dnl non-PIC
jmp *GSYM_PREFIX`'__gmpn_cpuvec+$2
')
EPILOGUE()
')')
dnl FAT_ENTRY for each CPUVEC_FUNCS_LIST
dnl
define(`CPUVEC_offset',0)
foreach(i,
`FAT_ENTRY(MPN(i),CPUVEC_offset)
define(`CPUVEC_offset',eval(CPUVEC_offset + 8))',
CPUVEC_FUNCS_LIST)
dnl Usage: FAT_INIT(name, offset)
dnl
dnl Emit a fat binary initializer function of the given name. These
dnl functions are the initial values for the pointers in __gmpn_cpuvec.
dnl
dnl The code simply calls __gmpn_cpuvec_init, and then jumps back through
dnl the __gmpn_cpuvec pointer, at the given "offset" (in bytes).
dnl __gmpn_cpuvec_init will have stored the address of the selected
dnl implementation there.
dnl
dnl Only one of these routines will be executed, and only once, since after
dnl that all the __gmpn_cpuvec pointers go to real routines. So there's no
dnl need for anything special here, just something small and simple. To
dnl keep code size down, "fat_init" is a shared bit of code, arrived at
dnl with the offset in %al. %al is used since the movb instruction is 2
dnl bytes where %eax would be 4.
dnl
dnl Note having `PROLOGUE in FAT_INIT obscures that PROLOGUE from the
dnl HAVE_NATIVE grepping in configure, preventing that code trying to eval
dnl something with $1 in it.
dnl
dnl We need to preserve parameter registers over the __gmpn_cpuvec_init call
define(FAT_INIT,
m4_assert_numargs(2)
`PROLOGUE($1)
mov $`'$2, %al
jmp L(fat_init)
EPILOGUE()
')
dnl FAT_INIT for each CPUVEC_FUNCS_LIST
dnl
define(`CPUVEC_offset',0)
foreach(i,
`FAT_INIT(MPN(i`'_init),CPUVEC_offset)
define(`CPUVEC_offset',eval(CPUVEC_offset + 1))',
CPUVEC_FUNCS_LIST)
L(fat_init):
C al __gmpn_cpuvec byte offset
movzbl %al, %eax
IFSTD(` push %rdi ')
IFSTD(` push %rsi ')
push %rdx
push %rcx
push %r8
push %r9
push %rax
CALL( __gmpn_cpuvec_init)
pop %rax
pop %r9
pop %r8
pop %rcx
pop %rdx
IFSTD(` pop %rsi ')
IFSTD(` pop %rdi ')
ifdef(`PRETEND_PIC',`
LEA( GSYM_PREFIX`'__gmpn_cpuvec, %r10)
jmp *(%r10,%rax,8)
',`dnl non-PIC
jmp *GSYM_PREFIX`'__gmpn_cpuvec(,%rax,8)
')
C long __gmpn_cpuid (char dst[12], int id);
C
C This is called only 3 times, so just something simple and compact is fine.
C
C The rcx/ecx zeroing here is needed for the BMI2 check.
define(`rp', `%rdi')
define(`idx', `%rsi')
PROLOGUE(__gmpn_cpuid)
FUNC_ENTRY(2)
mov %rbx, %r8
mov R32(idx), R32(%rax)
xor %ecx, %ecx
cpuid
mov %ebx, (rp)
mov %edx, 4(rp)
mov %ecx, 8(rp)
mov %r8, %rbx
FUNC_EXIT()
ret
EPILOGUE()