1 /* SPDX-License-Identifier: GPL-2.0 */ 1 /* SPDX-License-Identifier: GPL-2.0 */
2 /* 2 /*
3 * arch/alpha/lib/ev67-strchr.S 3 * arch/alpha/lib/ev67-strchr.S
4 * 21264 version contributed by Rick Gorton <ri 4 * 21264 version contributed by Rick Gorton <rick.gorton@alpha-processor.com>
5 * 5 *
6 * Return the address of a given character wit 6 * Return the address of a given character within a null-terminated
7 * string, or null if it is not found. 7 * string, or null if it is not found.
8 * 8 *
9 * Much of the information about 21264 schedul 9 * Much of the information about 21264 scheduling/coding comes from:
10 * Compiler Writer's Guide for the Alpha 10 * Compiler Writer's Guide for the Alpha 21264
11 * abbreviated as 'CWG' in other comments 11 * abbreviated as 'CWG' in other comments here
12 * ftp.digital.com/pub/Digital/info/semic 12 * ftp.digital.com/pub/Digital/info/semiconductor/literature/dsc-library.html
13 * Scheduling notation: 13 * Scheduling notation:
14 * E - either cluster 14 * E - either cluster
15 * U - upper subcluster; U0 - subcl 15 * U - upper subcluster; U0 - subcluster U0; U1 - subcluster U1
16 * L - lower subcluster; L0 - subcl 16 * L - lower subcluster; L0 - subcluster L0; L1 - subcluster L1
17 * Try not to change the actual algorithm if p 17 * Try not to change the actual algorithm if possible for consistency.
18 */ 18 */
19 #include <linux/export.h> 19 #include <linux/export.h>
20 #include <asm/regdef.h> 20 #include <asm/regdef.h>
21 21
22 .set noreorder 22 .set noreorder
23 .set noat 23 .set noat
24 24
25 .align 4 25 .align 4
26 .globl strchr 26 .globl strchr
27 .ent strchr 27 .ent strchr
28 strchr: 28 strchr:
29 .frame sp, 0, ra 29 .frame sp, 0, ra
30 .prologue 0 30 .prologue 0
31 31
32 ldq_u t0, 0(a0) # L : load fir 32 ldq_u t0, 0(a0) # L : load first quadword Latency=3
33 and a1, 0xff, t3 # E : 00000000 33 and a1, 0xff, t3 # E : 00000000000000ch
34 insbl a1, 1, t5 # U : 00000000 34 insbl a1, 1, t5 # U : 000000000000ch00
35 insbl a1, 7, a2 # U : ch000000 35 insbl a1, 7, a2 # U : ch00000000000000
36 36
37 insbl t3, 6, a3 # U : 00ch0000 37 insbl t3, 6, a3 # U : 00ch000000000000
38 or t5, t3, a1 # E : 00000000 38 or t5, t3, a1 # E : 000000000000chch
39 andnot a0, 7, v0 # E : align ou 39 andnot a0, 7, v0 # E : align our loop pointer
40 lda t4, -1 # E : build ga 40 lda t4, -1 # E : build garbage mask
41 41
42 mskqh t4, a0, t4 # U : only wan 42 mskqh t4, a0, t4 # U : only want relevant part of first quad
43 or a2, a3, a2 # E : chch0000 43 or a2, a3, a2 # E : chch000000000000
44 inswl a1, 2, t5 # E : 00000000 44 inswl a1, 2, t5 # E : 00000000chch0000
45 inswl a1, 4, a3 # E : 0000chch 45 inswl a1, 4, a3 # E : 0000chch00000000
46 46
47 or a1, a2, a1 # E : chch0000 47 or a1, a2, a1 # E : chch00000000chch
48 or a3, t5, t5 # E : 0000chch 48 or a3, t5, t5 # E : 0000chchchch0000
49 cmpbge zero, t0, t2 # E : bits set 49 cmpbge zero, t0, t2 # E : bits set iff byte == zero
50 cmpbge zero, t4, t4 # E : bits set 50 cmpbge zero, t4, t4 # E : bits set iff byte is garbage
51 51
52 /* This quad is _very_ serialized. Lo 52 /* This quad is _very_ serialized. Lots of stalling happens */
53 or t5, a1, a1 # E : chchchch 53 or t5, a1, a1 # E : chchchchchchchch
54 xor t0, a1, t1 # E : make byt 54 xor t0, a1, t1 # E : make bytes == c zero
55 cmpbge zero, t1, t3 # E : bits set 55 cmpbge zero, t1, t3 # E : bits set iff byte == c
56 or t2, t3, t0 # E : bits set 56 or t2, t3, t0 # E : bits set iff char match or zero match
57 57
58 andnot t0, t4, t0 # E : clear ga 58 andnot t0, t4, t0 # E : clear garbage bits
59 cttz t0, a2 # U0 : specula 59 cttz t0, a2 # U0 : speculative (in case we get a match)
60 nop # E : 60 nop # E :
61 bne t0, $found # U : 61 bne t0, $found # U :
62 62
63 /* 63 /*
64 * Yuk. This loop is going to stall l 64 * Yuk. This loop is going to stall like crazy waiting for the
65 * data to be loaded. Not much can be 65 * data to be loaded. Not much can be done about it unless it's
66 * unrolled multiple times - is that s 66 * unrolled multiple times - is that safe to do in kernel space?
67 * Or would exception handling recover 67 * Or would exception handling recovery code do the trick here?
68 */ 68 */
69 $loop: ldq t0, 8(v0) # L : Latency= 69 $loop: ldq t0, 8(v0) # L : Latency=3
70 addq v0, 8, v0 # E : 70 addq v0, 8, v0 # E :
71 xor t0, a1, t1 # E : 71 xor t0, a1, t1 # E :
72 cmpbge zero, t0, t2 # E : bits set 72 cmpbge zero, t0, t2 # E : bits set iff byte == 0
73 73
74 cmpbge zero, t1, t3 # E : bits set 74 cmpbge zero, t1, t3 # E : bits set iff byte == c
75 or t2, t3, t0 # E : 75 or t2, t3, t0 # E :
76 cttz t3, a2 # U0 : specula 76 cttz t3, a2 # U0 : speculative (in case we get a match)
77 beq t0, $loop # U : 77 beq t0, $loop # U :
78 78
79 $found: negq t0, t1 # E : clear al 79 $found: negq t0, t1 # E : clear all but least set bit
80 and t0, t1, t0 # E : 80 and t0, t1, t0 # E :
81 and t0, t3, t1 # E : bit set 81 and t0, t3, t1 # E : bit set iff byte was the char
82 addq v0, a2, v0 # E : Add in t 82 addq v0, a2, v0 # E : Add in the bit number from above
83 83
84 cmoveq t1, $31, v0 # E : Two mapp 84 cmoveq t1, $31, v0 # E : Two mapping slots, latency = 2
85 nop 85 nop
86 nop 86 nop
87 ret # L0 : 87 ret # L0 :
88 88
89 .end strchr 89 .end strchr
90 EXPORT_SYMBOL(strchr) 90 EXPORT_SYMBOL(strchr)
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.