1 ; Copyright (c) 2010-2011, Linaro Limited
3 ; SPDX-License-Identifier: BSD-2-Clause-Patent
7 ; Written by Dave Gilbert <david.gilbert@linaro.org>
9 ; This memchr routine is optimised on a Cortex-A9 and should work on
10 ; all ARMv7 processors. It has a fast past for short sizes, and has
11 ; an optimised path for large data sets; the worst case is finding the
12 ; match early in a large data set.
16 ; 2011-02-07 david.gilbert@linaro.org
17 ; Extracted from local git a5b438d861
18 ; 2011-07-14 david.gilbert@linaro.org
19 ; Import endianness fix from local git ea786f1b
20 ; 2011-12-07 david.gilbert@linaro.org
21 ; Removed unneeded cbz from align loop
23 ; this lets us check a flag in a 00/ff byte easily in either endianness
24 #define CHARTSTMASK(c) 1<<(c*8)
26 EXPORT InternalMemScanMem8
27 AREA ScanMem, CODE, READONLY
31 ; r0 = start of memory to scan
33 ; r2 = character to look for
34 ; returns r0 = pointer to character or NULL if not found
35 uxtb r2, r2 ; Don't think we can trust the caller to actually pass a char
37 cmp r1, #16 ; If it's short don't bother with anything clever
40 tst r0, #7 ; If it's already aligned skip the next bit
43 ; Work up to an aligned point
48 beq L50 ; If it matches exit found
50 bne L5 ; If not aligned yet then do next byte
53 ; At this point, we are aligned, we know we have at least 8 bytes to work with
55 orr r2, r2, r2, lsl #8 ; expand the match word across to all bytes
56 orr r2, r2, r2, lsl #16
57 bic r4, r1, #7 ; Number of double words to work with
64 eor r5, r5, r2 ; Get it so that r5,r6 have 00's where the bytes match the target
66 uadd8 r5, r5, r7 ; Parallel add 0xff - sets the GE bits for anything that wasn't 0
67 sel r5, r3, r7 ; bytes are 00 for none-00 bytes, or ff for 00 bytes - NOTE INVERSION
68 uadd8 r6, r6, r7 ; Parallel add 0xff - sets the GE bits for anything that wasn't 0
69 sel r6, r5, r7 ; chained....bytes are 00 for none-00 bytes, or ff for 00 bytes - NOTE INVERSION
71 bne L15 ; (Flags from the subs above) If not run out of bytes then go around again
74 and r2, r2, #0xff ; Get r2 back to a single character from the expansion above
75 and r1, r1, #7 ; Leave the count remaining as the number after the double words have been done
78 cbz r1, L40 ; 0 length or hit the end already then not found
80 L21 ; Post aligned section, or just a short call
83 eor r3, r3, r2 ; r3 = 0 if match - doesn't break flags from sub
88 movs r0, #0 ; not found
92 subs r0, r0, #1 ; found
95 L60 ; We're here because the fast path found a hit - now we have to track down exactly which word it was
96 ; r0 points to the start of the double word after the one that was tested
97 ; r5 has the 00/ff pattern for the first word, r6 has the chained value
100 moveq r5, r6 ; the end is in the 2nd word
101 subeq r0, r0, #3 ; Points to 2nd byte of 2nd word
102 subne r0, r0, #7 ; or 2nd byte of 1st word
104 ; r0 currently points to the 3rd byte of the word containing the hit
105 tst r5, #CHARTSTMASK(0) ; 1st character
108 tst r5, #CHARTSTMASK(1) ; 2nd character
111 tsteq r5, #(3 << 15) ; 2nd & 3rd character
112 ; If not the 3rd must be the last one