Commit 2511d032 authored by Russell King (Oracle)'s avatar Russell King (Oracle)

ARM: findbit: operate by words

Convert the implementations to operate on words rather than bytes
which makes bitmap searching faster.
Signed-off-by: default avatarRussell King (Oracle) <rmk+kernel@armlinux.org.uk>
parent 2953a3e1
...@@ -761,6 +761,12 @@ THUMB( orr \reg , \reg , #PSR_T_BIT ) ...@@ -761,6 +761,12 @@ THUMB( orr \reg , \reg , #PSR_T_BIT )
.endif .endif
.endm .endm
.if __LINUX_ARM_ARCH__ < 6
.set .Lrev_l_uses_tmp, 1
.else
.set .Lrev_l_uses_tmp, 0
.endif
/* /*
* bl_r - branch and link to register * bl_r - branch and link to register
* *
......
...@@ -14,32 +14,32 @@ ...@@ -14,32 +14,32 @@
#include <asm/assembler.h> #include <asm/assembler.h>
.text .text
#ifdef __ARMEB__
#define SWAB_ENDIAN le
#else
#define SWAB_ENDIAN be
#endif
.macro find_first, endian, set, name .macro find_first, endian, set, name
ENTRY(_find_first_\name\()bit_\endian) ENTRY(_find_first_\name\()bit_\endian)
teq r1, #0 teq r1, #0
beq 3f beq 3f
mov r2, #0 mov r2, #0
1: 1: ldr r3, [r0], #4
.ifc \endian, be .ifeq \set
eor r3, r2, #0x18 mvns r3, r3 @ invert/test bits
ARM( ldrb r3, [r0, r3, lsr #3] )
THUMB( lsr r3, #3 )
THUMB( ldrb r3, [r0, r3] )
.else .else
ARM( ldrb r3, [r0, r2, lsr #3] ) movs r3, r3 @ test bits
THUMB( lsr r3, r2, #3 )
THUMB( ldrb r3, [r0, r3] )
.endif .endif
.ifeq \set .ifc \endian, SWAB_ENDIAN
eors r3, r3, #0xff @ invert bits bne .L_found_swab
.else .else
movs r3, r3 bne .L_found @ found the bit?
.endif .endif
bne .L_found @ any now set - found zero bit add r2, r2, #32 @ next index
add r2, r2, #8 @ next bit pointer
2: cmp r2, r1 @ any more? 2: cmp r2, r1 @ any more?
blo 1b blo 1b
3: mov r0, r1 @ no free bits 3: mov r0, r1 @ no more bits
ret lr ret lr
ENDPROC(_find_first_\name\()bit_\endian) ENDPROC(_find_first_\name\()bit_\endian)
.endm .endm
...@@ -48,24 +48,25 @@ ENDPROC(_find_first_\name\()bit_\endian) ...@@ -48,24 +48,25 @@ ENDPROC(_find_first_\name\()bit_\endian)
ENTRY(_find_next_\name\()bit_\endian) ENTRY(_find_next_\name\()bit_\endian)
cmp r2, r1 cmp r2, r1
bhs 3b bhs 3b
ands ip, r2, #7 mov ip, r2, lsr #5 @ word index
beq 1b @ If new byte, goto old routine add r0, r0, ip, lsl #2
.ifc \endian, be ands ip, r2, #31 @ bit position
eor r3, r2, #0x18 beq 1b
ARM( ldrb r3, [r0, r3, lsr #3] ) ldr r3, [r0], #4
THUMB( lsr r3, #3 )
THUMB( ldrb r3, [r0, r3] )
.else
ARM( ldrb r3, [r0, r2, lsr #3] )
THUMB( lsr r3, r2, #3 )
THUMB( ldrb r3, [r0, r3] )
.endif
.ifeq \set .ifeq \set
eor r3, r3, #0xff @ now looking for a 1 bit mvn r3, r3 @ invert bits
.endif
.ifc \endian, SWAB_ENDIAN
rev_l r3, ip
.if .Lrev_l_uses_tmp
@ we need to recompute ip because rev_l will have overwritten
@ it.
and ip, r2, #31 @ bit position
.endif
.endif .endif
movs r3, r3, lsr ip @ shift off unused bits movs r3, r3, lsr ip @ shift off unused bits
bne .L_found bne .L_found
orr r2, r2, #7 @ if zero, then no bits here orr r2, r2, #31 @ no zero bits
add r2, r2, #1 @ align bit pointer add r2, r2, #1 @ align bit pointer
b 2b @ loop for next bit b 2b @ loop for next bit
ENDPROC(_find_next_\name\()bit_\endian) ENDPROC(_find_next_\name\()bit_\endian)
...@@ -95,6 +96,8 @@ ENDPROC(_find_next_\name\()bit_\endian) ...@@ -95,6 +96,8 @@ ENDPROC(_find_next_\name\()bit_\endian)
/* /*
* One or more bits in the LSB of r3 are assumed to be set. * One or more bits in the LSB of r3 are assumed to be set.
*/ */
.L_found_swab:
rev_l r3, ip
.L_found: .L_found:
#if __LINUX_ARM_ARCH__ >= 7 #if __LINUX_ARM_ARCH__ >= 7
rbit r3, r3 @ reverse bits rbit r3, r3 @ reverse bits
...@@ -107,13 +110,20 @@ ENDPROC(_find_next_\name\()bit_\endian) ...@@ -107,13 +110,20 @@ ENDPROC(_find_next_\name\()bit_\endian)
rsb r3, r3, #31 @ offset of first set bit rsb r3, r3, #31 @ offset of first set bit
add r0, r2, r3 @ add offset of first set bit add r0, r2, r3 @ add offset of first set bit
#else #else
tst r3, #0x0f mov ip, #~0
tst r3, ip, lsr #16 @ test bits 0-15
addeq r2, r2, #16
moveq r3, r3, lsr #16
tst r3, #0x00ff
addeq r2, r2, #8
moveq r3, r3, lsr #8
tst r3, #0x000f
addeq r2, r2, #4 addeq r2, r2, #4
movne r3, r3, lsl #4 moveq r3, r3, lsr #4
tst r3, #0x30 tst r3, #0x0003
addeq r2, r2, #2 addeq r2, r2, #2
movne r3, r3, lsl #2 moveq r3, r3, lsr #2
tst r3, #0x40 tst r3, #0x0001
addeq r2, r2, #1 addeq r2, r2, #1
mov r0, r2 mov r0, r2
#endif #endif
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment