Difference between revisions of "ARM: Count leading zeros"
From ScienceZero
| (3 intermediate revisions by 2 users not shown) | |||
| Line 1: | Line 1: | ||
| + | This is using the old 32 bit instruction set, modern chips have a single cycle CLZ instruction. | ||
| + | |||
==Minimum size== | ==Minimum size== | ||
;R0 - value | ;R0 - value | ||
;R1 = count of leading zeros | ;R1 = count of leading zeros | ||
| − | ;Uses R0-R1 | + | ;Uses R0-R1 |
;Worst case 159 cycles, best case 4 cycles | ;Worst case 159 cycles, best case 4 cycles | ||
mov r1,#0 | mov r1,#0 | ||
| Line 39: | Line 41: | ||
mov pc,r14 | mov pc,r14 | ||
| + | |||
| + | ==Maximum performance with normalising== | ||
| + | ;R0 - value | ||
| + | ;R1 = count of leading zeros | ||
| + | ;Uses R0-R1 | ||
| + | ;17 cycles constant | ||
| + | movs r1,r0,lsr #16 | ||
| + | mov r1,#0 | ||
| + | addeq r1,r1,#16 | ||
| + | moveqs r0,r0,lsl #16 | ||
| + | addeq r1,r1,#1 | ||
| + | |||
| + | tst r0,#0xff000000 | ||
| + | addeq r1,r1,#8 | ||
| + | moveq r0,r0,lsl #8 | ||
| + | |||
| + | tst r0,#0xf0000000 | ||
| + | addeq r1,r1,#4 | ||
| + | moveq r0,r0,lsl #4 | ||
| + | |||
| + | tst r0,#0xc0000000 | ||
| + | addeq r1,r1,#2 | ||
| + | moveq r0,r0,lsl #2 | ||
| + | |||
| + | tst r0,#0x80000000 | ||
| + | addeq r1,r1,#1 | ||
| + | moveq r0,r0,lsl #1 | ||
| + | |||
| + | mov pc,r14 | ||
[[Category:Computing]] | [[Category:Computing]] | ||
Latest revision as of 05:26, 5 June 2011
This is using the old 32 bit instruction set, modern chips have a single cycle CLZ instruction.
Minimum size
;R0 - value
;R1 = count of leading zeros
;Uses R0-R1
;Worst case 159 cycles, best case 4 cycles
mov r1,#0
loop movs r0,r0,lsl #1
addcc r1,r1,#1
bcc loop
mov pc,r14
Maximum performance
;R0 - value
;R1 = count of leading zeros
;Uses R0-R1
;16 cycles constant
movs r1,r0,lsr #16
mov r1,#0
addeq r1,r1,#16
moveqs r0,r0,lsl #16
addeq r1,r1,#1
tst r0,#0xff000000
addeq r1,r1,#8
moveq r0,r0,lsl #8
tst r0,#0xf0000000
addeq r1,r1,#4
moveq r0,r0,lsl #4
tst r0,#0xc0000000
addeq r1,r1,#2
moveq r0,r0,lsl #2
tst r0,#0x80000000
addeq r1,r1,#1
mov pc,r14
Maximum performance with normalising
;R0 - value
;R1 = count of leading zeros
;Uses R0-R1
;17 cycles constant
movs r1,r0,lsr #16
mov r1,#0
addeq r1,r1,#16
moveqs r0,r0,lsl #16
addeq r1,r1,#1
tst r0,#0xff000000
addeq r1,r1,#8
moveq r0,r0,lsl #8
tst r0,#0xf0000000
addeq r1,r1,#4
moveq r0,r0,lsl #4
tst r0,#0xc0000000
addeq r1,r1,#2
moveq r0,r0,lsl #2
tst r0,#0x80000000
addeq r1,r1,#1
moveq r0,r0,lsl #1
mov pc,r14
