Commit c2959956 authored by Mao Minkai's avatar Mao Minkai Committed by guzitao
Browse files

sw64: choose lib at compile time 

Sunway inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/IB73UR



--------------------------------

Compile both SISD and SIMD version of memcpy/memset like libs and let
the compiler optimize out the unused version based on kernel config.

Signed-off-by: default avatarMao Minkai <maominkai@wxiat.com>
Reviewed-by: default avatarHe Sheng <hesheng@wxiat.com>
Signed-off-by: default avatarGu Zitao <guzitao@wxiat.com>
parent bb2c1f34

arch/sw_64/lib/Makefile

+4 −8
Original line number Diff line number Diff line
@@ -22,20 +22,16 @@ lib-y = __divlu.o __remlu.o __divwu.o __remwu.o \ 
lib-clear_page-y := clear_page.o

lib-clear_page-$(CONFIG_DEEP_CLEAR_PAGE) := deep-clear_page.o



lib-clear_user-y := clear_user.o

lib-clear_user-$(CONFIG_DEEP_CLEAR_USER) := deep-clear_user-hw_una.o deep-clear_user-sw_una.o

lib-clear_user-y := clear_user.o deep-clear_user-hw_una.o deep-clear_user-sw_una.o



lib-copy_page-y := copy_page.o

lib-copy_page-$(CONFIG_DEEP_COPY_PAGE) := deep-copy_page.o



lib-copy_user-y := copy_user.o

lib-copy_user-$(CONFIG_DEEP_COPY_USER) := deep-copy_user-hw_una.o deep-copy_user-sw_una.o

lib-copy_user-y := copy_user.o deep-copy_user-hw_una.o deep-copy_user-sw_una.o



lib-memcpy-y := memcpy.o

lib-memcpy-$(CONFIG_DEEP_MEMCPY) := deep-memcpy-hw_una.o deep-memcpy-sw_una.o

lib-memcpy-y := memcpy.o deep-memcpy-hw_una.o deep-memcpy-sw_una.o



lib-memset-y := memset.o

lib-memset-$(CONFIG_DEEP_MEMSET) := deep-memset-hw_una.o deep-memset-sw_una.o

lib-memset-y := memset.o deep-memset-hw_una.o deep-memset-sw_una.o



lib-$(CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE) += uaccess_flushcache.o

arch/sw_64/lib/clear_user.S

+4 −6
Original line number Diff line number Diff line
@@ -10,7 +10,6 @@ 
 * stuff.

 *

 */

#include <asm/export.h>

/* Allow an exception for an insn; exit if we get one.  */

#define EX(x,y...)			\

	99: x,##y;			\
@@ -23,11 +22,11 @@ 
	.set noreorder

	.align 4



	.globl __clear_user

	.ent __clear_user

	.globl ____clear_user_sisd

	.ent ____clear_user_sisd

	.frame $30, 0, $26

	.prologue 0

__clear_user:

____clear_user_sisd:

	and	$17, $17, $0

	and	$16, 7, $4

	beq	$0, $zerolength
@@ -98,5 +97,4 @@ $tail: 
	clr	$0

	ret	$31, ($26), 1



	.end __clear_user

	EXPORT_SYMBOL(__clear_user)
 
	.end ____clear_user_sisd

arch/sw_64/lib/copy_user.S

+4 −6
Original line number Diff line number Diff line
@@ -9,7 +9,6 @@ 
 * only _after_ a successful copy). There is also some rather minor

 * exception setup stuff..

 */

#include <asm/export.h>

/* Allow an exception for an insn; exit if we get one.  */

#define EXI(x,y...)			\

	99: x,##y;			\
@@ -27,9 +26,9 @@ 


	.set noat

	.align 4

	.globl __copy_user

	.ent __copy_user

__copy_user:

	.globl ____copy_user_sisd

	.ent ____copy_user_sisd

____copy_user_sisd:

	.prologue 0

	and	$18, $18, $0

	and	$16, 7, $3
@@ -102,5 +101,4 @@ $exitin: 
$exitout:

	ret	$31, ($26), 1



	.end __copy_user

	EXPORT_SYMBOL(__copy_user)
 
	.end ____copy_user_sisd

arch/sw_64/lib/memcpy.S

+5 −10
Original line number Diff line number Diff line 
/* SPDX-License-Identifier: GPL-2.0 */

/*

 * Reasonably optimized memcpy() routine for the sw64

 * Reasonably optimized ____memcpy_sisd() routine for the sw64

 *

 *	- memory accessed as aligned quadwords only

 *	- uses bcmpge to compare 8 bytes in parallel
@@ -8,14 +8,13 @@ 
 * Temp usage notes:

 *	$1, $2, - scratch

 */

#include <asm/export.h>

	.set noreorder

	.set noat



	.align 4

	.globl memcpy

	.ent memcpy

memcpy:

	.globl ____memcpy_sisd

	.ent ____memcpy_sisd

____memcpy_sisd:

	.frame $30, 0, $26, 0

	.prologue 0
@@ -194,8 +193,4 @@ $misalign_byte: 
$nomoredata:

	ret	$31, ($26), 1



	.end memcpy

	EXPORT_SYMBOL(memcpy)

/* For backwards module compatibility.  */

__memcpy = memcpy

.globl __memcpy
 
	.end ____memcpy_sisd

arch/sw_64/lib/memset.S

+4 −46
Original line number Diff line number Diff line
@@ -13,35 +13,16 @@ 
 * hand, so they might well be incorrect, please do tell me about it..)

 */



#include <asm/export.h>



	.set noat

	.set noreorder

.text

	.globl memset

	.globl __memset

	.globl ___memset

	.globl __memsetw

	.globl __constant_c_memset

	.globl ____constant_c_memset_sisd



	.ent ___memset

.align 5

___memset:

	.ent ____constant_c_memset_sisd

	.frame $30, 0, $26, 0

	.prologue 0



	and	$17, 255, $1

	inslb	$17, 1, $17

	bis	$17, $1, $17

	sll	$17, 16, $1



	bis	$17, $1, $17

	sll	$17, 32, $1

	bis	$17, $1, $17

	ldl_u	$31, 0($30)



.align 5

__constant_c_memset:

____constant_c_memset_sisd:

	addl	$18, $16, $6

	bis	$16, $16, $0

	xor	$16, $6, $1
@@ -127,27 +108,4 @@ within_one_quad_loop: 


end:

	ret	$31, ($26), 1

	.end ___memset

	EXPORT_SYMBOL(___memset)



	.align 5

	.ent __memsetw

__memsetw:

	.prologue 0



	inslh	$17, 0, $1

	inslh	$17, 2, $2

	inslh	$17, 4, $3

	or	$1, $2, $1

	inslh	$17, 6, $4

	or	$1, $3, $1

	or	$1, $4, $17

	br __constant_c_memset



	.end __memsetw

	EXPORT_SYMBOL(__memsetw)



memset = ___memset

EXPORT_SYMBOL(memset)

__memset = ___memset

EXPORT_SYMBOL(__memset)
 
	.end ____constant_c_memset_sisd