1
1
This commit adds assembly support for aarch64.

Signed-off-by: Nathan Hjelm <hjelmn@me.com>
Этот коммит содержится в:
Nathan Hjelm 2016-05-22 00:13:45 -06:00 коммит произвёл Nathan Hjelm
родитель 60519c2b4e
Коммит 0084ad0d1b
9 изменённых файлов: 386 добавлений и 0 удалений

Просмотреть файл

@ -934,6 +934,14 @@ AC_DEFUN([OPAL_CONFIG_ASM],[
OPAL_ASM_SUPPORT_64BIT=1
OPAL_GCC_INLINE_ASSIGN='"mov %0=r0\n;;\n" : "=&r"(ret)'
;;
aarch64*)
opal_cv_asm_arch="ARM64"
OPAL_ASM_SUPPORT_64BIT=1
OPAL_ASM_ARM_VERSION=8
AC_DEFINE_UNQUOTED([OPAL_ASM_ARM_VERSION], [$OPAL_ASM_ARM_VERSION],
[What ARM assembly version to use])
OPAL_GCC_INLINE_ASSIGN='"mov %0, #0" : "=&r"(ret)'
;;
armv7*)
opal_cv_asm_arch="ARM"

Просмотреть файл

@ -11,6 +11,8 @@
# All rights reserved.
# Copyright (c) 2010 Cisco Systems, Inc. All rights reserved.
# Copyright (c) 2011 Sandia National Laboratories. All rights reserved.
# Copyright (c) 2016 Los Alamos National Security, LLC. All rights
# reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
@ -29,6 +31,7 @@ headers += \
include opal/sys/amd64/Makefile.am
include opal/sys/arm/Makefile.am
include opal/sys/arm64/Makefile.am
include opal/sys/ia32/Makefile.am
include opal/sys/ia64/Makefile.am
include opal/sys/mips/Makefile.am

Просмотреть файл

@ -11,6 +11,8 @@
* All rights reserved.
* Copyright (c) 2011 Sandia National Laboratories. All rights reserved.
* Copyright (c) 2014 Intel, Inc. All rights reserved
* Copyright (c) 2016 Los Alamos National Security, LLC. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -37,6 +39,7 @@
#define OPAL_SPARCV9_64 0062
#define OPAL_MIPS 0070
#define OPAL_ARM 0100
#define OPAL_ARM64 0101
#define OPAL_BUILTIN_SYNC 0200
#define OPAL_BUILTIN_OSX 0201
#define OPAL_BUILTIN_NO 0202

24
opal/include/opal/sys/arm64/Makefile.am Обычный файл
Просмотреть файл

@ -0,0 +1,24 @@
#
# Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
# University Research and Technology
# Corporation. All rights reserved.
# Copyright (c) 2004-2008 The University of Tennessee and The University
# of Tennessee Research Foundation. All rights
# reserved.
# Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
# University of Stuttgart. All rights reserved.
# Copyright (c) 2004-2005 The Regents of the University of California.
# All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
# This makefile.am does not stand on its own - it is included from opal/include/Makefile.am
headers += \
opal/sys/arm64/atomic.h \
opal/sys/arm64/timer.h

270
opal/include/opal/sys/arm64/atomic.h Обычный файл
Просмотреть файл

@ -0,0 +1,270 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2005 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
* University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* Copyright (c) 2010 IBM Corporation. All rights reserved.
* Copyright (c) 2010 ARM ltd. All rights reserved.
* Copyright (c) 2016 Los Alamos National Security, LLC. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#if !defined(OPAL_SYS_ARCH_ATOMIC_H)
#define OPAL_SYS_ARCH_ATOMIC_H 1
#if OPAL_GCC_INLINE_ASSEMBLY
#define OPAL_HAVE_ATOMIC_MEM_BARRIER 1
#define OPAL_HAVE_ATOMIC_LLSC_32 1
#define OPAL_HAVE_ATOMIC_CMPSET_32 1
#define OPAL_HAVE_ATOMIC_MATH_32 1
#define OPAL_HAVE_ATOMIC_CMPSET_64 1
#define OPAL_HAVE_ATOMIC_LLSC_64 1
#define OPAL_HAVE_ATOMIC_ADD_32 1
#define OPAL_HAVE_ATOMIC_SUB_32 1
#define OPAL_HAVE_ATOMIC_ADD_64 1
#define OPAL_HAVE_ATOMIC_SUB_64 1
#define MB() __asm__ __volatile__ ("dmb sy" : : : "memory")
#define RMB() __asm__ __volatile__ ("dmb ld" : : : "memory")
#define WMB() __asm__ __volatile__ ("dmb st" : : : "memory")
/**********************************************************************
*
* Memory Barriers
*
*********************************************************************/
static inline void opal_atomic_mb (void)
{
MB();
}
static inline void opal_atomic_rmb (void)
{
RMB();
}
static inline void opal_atomic_wmb (void)
{
WMB();
}
static inline void opal_atomic_isync (void)
{
__asm__ __volatile__ ("isb");
}
/**********************************************************************
*
* Atomic math operations
*
*********************************************************************/
static inline int opal_atomic_cmpset_32(volatile int32_t *addr,
int32_t oldval, int32_t newval)
{
int32_t ret, tmp;
__asm__ __volatile__ ("1: ldaxr %w0, [%2] \n"
" cmp %w0, %w3 \n"
" bne 2f \n"
" stxr %w1, %w4, [%2] \n"
" cbnz %w1, 1b \n"
"2: \n"
: "=&r" (ret), "=&r" (tmp)
: "r" (addr), "r" (oldval), "r" (newval)
: "cc", "memory");
return (ret == oldval);
}
/* these two functions aren't inlined in the non-gcc case because then
there would be two function calls (since neither cmpset_32 nor
atomic_?mb can be inlined). Instead, we "inline" them by hand in
the assembly, meaning there is one function call overhead instead
of two */
static inline int opal_atomic_cmpset_acq_32(volatile int32_t *addr,
int32_t oldval, int32_t newval)
{
int32_t ret, tmp;
__asm__ __volatile__ ("1: ldaxr %w0, [%2] \n"
" cmp %w0, %w3 \n"
" bne 2f \n"
" stxr %w1, %w4, [%2] \n"
" cbnz %w1, 1b \n"
"2: \n"
: "=&r" (ret), "=&r" (tmp)
: "r" (addr), "r" (oldval), "r" (newval)
: "cc", "memory");
return (ret == oldval);
}
static inline int opal_atomic_cmpset_rel_32(volatile int32_t *addr,
int32_t oldval, int32_t newval)
{
int32_t ret, tmp;
__asm__ __volatile__ ("1: ldxr %w0, [%2] \n"
" cmp %w0, %w3 \n"
" bne 2f \n"
" stlxr %w1, %w4, [%2] \n"
" cbnz %w1, 1b \n"
"2: \n"
: "=&r" (ret), "=&r" (tmp)
: "r" (addr), "r" (oldval), "r" (newval)
: "cc", "memory");
return (ret == oldval);
}
static inline int32_t opal_atomic_ll_32 (volatile int32_t *addr)
{
int32_t ret;
__asm__ __volatile__ ("ldaxr %w0, [%1] \n"
: "=&r" (ret)
: "r" (addr));
return ret;
}
static inline int opal_atomic_sc_32 (volatile int32_t *addr, int32_t newval)
{
int ret;
__asm__ __volatile__ ("stlxr %w0, %w2, [%1] \n"
: "=&r" (ret)
: "r" (addr), "r" (newval)
: "cc", "memory");
return ret == 0;
}
static inline int opal_atomic_cmpset_64(volatile int64_t *addr,
int64_t oldval, int64_t newval)
{
int64_t ret;
int tmp;
__asm__ __volatile__ ("1: ldaxr %0, [%2] \n"
" cmp %0, %3 \n"
" bne 2f \n"
" stxr %w1, %4, [%2] \n"
" cbnz %w1, 1b \n"
"2: \n"
: "=&r" (ret), "=&r" (tmp)
: "r" (addr), "r" (oldval), "r" (newval)
: "cc", "memory");
return (ret == oldval);
}
/* these two functions aren't inlined in the non-gcc case because then
there would be two function calls (since neither cmpset_64 nor
atomic_?mb can be inlined). Instead, we "inline" them by hand in
the assembly, meaning there is one function call overhead instead
of two */
static inline int opal_atomic_cmpset_acq_64(volatile int64_t *addr,
int64_t oldval, int64_t newval)
{
int64_t ret;
int tmp;
__asm__ __volatile__ ("1: ldaxr %0, [%2] \n"
" cmp %0, %3 \n"
" bne 2f \n"
" stxr %w1, %4, [%2] \n"
" cbnz %w1, 1b \n"
"2: \n"
: "=&r" (ret), "=&r" (tmp)
: "r" (addr), "r" (oldval), "r" (newval)
: "cc", "memory");
return (ret == oldval);
}
static inline int opal_atomic_cmpset_rel_64(volatile int64_t *addr,
int64_t oldval, int64_t newval)
{
int64_t ret;
int tmp;
__asm__ __volatile__ ("1: ldxr %0, [%2] \n"
" cmp %0, %3 \n"
" bne 2f \n"
" stlxr %w1, %4, [%2] \n"
" cbnz %w1, 1b \n"
"2: \n"
: "=&r" (ret), "=&r" (tmp)
: "r" (addr), "r" (oldval), "r" (newval)
: "cc", "memory");
return (ret == oldval);
}
static inline int64_t opal_atomic_ll_64 (volatile int64_t *addr)
{
int64_t ret;
__asm__ __volatile__ ("ldaxr %0, [%1] \n"
: "=&r" (ret)
: "r" (addr));
return ret;
}
static inline int opal_atomic_sc_64 (volatile int64_t *addr, int64_t newval)
{
int ret;
__asm__ __volatile__ ("stlxr %w0, %2, [%1] \n"
: "=&r" (ret)
: "r" (addr), "r" (newval)
: "cc", "memory");
return ret == 0;
}
#define OPAL_ASM_MAKE_ATOMIC(type, bits, name, inst, reg) \
static inline type opal_atomic_ ## name ## _ ## bits (volatile type *addr, type value) \
{ \
type newval; \
int32_t tmp; \
\
__asm__ __volatile__("1: ldxr %" reg "0, [%2] \n" \
" " inst " %" reg "0, %" reg "0, %" reg "3 \n" \
" stxr %w1, %" reg "0, [%2] \n" \
" cbnz %w1, 1b \n" \
: "=&r" (newval), "=&r" (tmp) \
: "r" (addr), "r" (value) \
: "cc", "memory"); \
\
return newval; \
}
OPAL_ASM_MAKE_ATOMIC(int32_t, 32, add, "add", "w")
OPAL_ASM_MAKE_ATOMIC(int32_t, 32, sub, "sub", "w")
OPAL_ASM_MAKE_ATOMIC(int64_t, 64, add, "add", "")
OPAL_ASM_MAKE_ATOMIC(int64_t, 64, sub, "sub", "")
#endif /* OPAL_GCC_INLINE_ASSEMBLY */
#endif /* ! OPAL_SYS_ARCH_ATOMIC_H */

33
opal/include/opal/sys/arm64/timer.h Обычный файл
Просмотреть файл

@ -0,0 +1,33 @@
/*
* Copyright (c) 2008 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#ifndef OPAL_SYS_ARCH_TIMER_H
#define OPAL_SYS_ARCH_TIMER_H 1
#include <sys/times.h>
typedef uint64_t opal_timer_t;
static inline opal_timer_t
opal_sys_timer_get_cycles(void)
{
opal_timer_t ret;
struct tms accurate_clock;
times(&accurate_clock);
ret = accurate_clock.tms_utime + accurate_clock.tms_stime;
return ret;
}
#define OPAL_HAVE_SYS_TIMER_GET_CYCLES 1
#endif /* ! OPAL_SYS_ARCH_TIMER_H */

36
opal/include/opal/sys/arm64/update.sh Обычный файл
Просмотреть файл

@ -0,0 +1,36 @@
#!/bin/sh
#
# Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
# University Research and Technology
# Corporation. All rights reserved.
# Copyright (c) 2004-2005 The University of Tennessee and The University
# of Tennessee Research Foundation. All rights
# reserved.
# Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
# University of Stuttgart. All rights reserved.
# Copyright (c) 2004-2005 The Regents of the University of California.
# All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
CFILE=/tmp/opal_atomic_$$.c
trap "/bin/rm -f $CFILE; exit 0" 0 1 2 15
echo Updating atomic.s from atomic.h using gcc
cat > $CFILE<<EOF
#include <stdlib.h>
#include <inttypes.h>
#define static
#define inline
#define OPAL_GCC_INLINE_ASSEMBLY 1
#include "../architecture.h"
#include "atomic.h"
EOF
gcc -O1 -I. -S $CFILE -o atomic.s

Просмотреть файл

@ -147,6 +147,8 @@ typedef struct opal_atomic_lock_t opal_atomic_lock_t;
#include "opal/sys/amd64/atomic.h"
#elif OPAL_ASSEMBLY_ARCH == OPAL_ARM
#include "opal/sys/arm/atomic.h"
#elif OPAL_ASSEMBLY_ARCH == OPAL_ARM64
#include "opal/sys/arm64/atomic.h"
#elif OPAL_ASSEMBLY_ARCH == OPAL_IA32
#include "opal/sys/ia32/atomic.h"
#elif OPAL_ASSEMBLY_ARCH == OPAL_IA64

Просмотреть файл

@ -52,6 +52,13 @@
#define __NR_process_vm_readv 376
#define __NR_process_vm_writev 377
#elif OPAL_ASSEMBLY_ARCH == OPAL_ARM64
/* ARM64 uses the asm-generic syscall numbers */
#define __NR_process_vm_readv 270
#define __NR_process_vm_writev 271
#elif OPAL_ASSEMBLY_ARCH == OPAL_MIPS
#if _MIPS_SIM == _MIPS_SIM_ABI32