Text file
src/crypto/sha256/sha256block_arm64.s
1 // Copyright 2017 The Go Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style
3 // license that can be found in the LICENSE file.
4
5 #include "textflag.h"
6
7 #define HASHUPDATE \
8 SHA256H V9.S4, V3, V2 \
9 SHA256H2 V9.S4, V8, V3 \
10 VMOV V2.B16, V8.B16
11
12 // func sha256block(h []uint32, p []byte, k []uint32)
13 TEXT ·sha256block(SB),NOSPLIT,$0
14 MOVD h_base+0(FP), R0 // Hash value first address
15 MOVD p_base+24(FP), R1 // message first address
16 MOVD k_base+48(FP), R2 // k constants first address
17 MOVD p_len+32(FP), R3 // message length
18 VLD1 (R0), [V0.S4, V1.S4] // load h(a,b,c,d,e,f,g,h)
19 VLD1.P 64(R2), [V16.S4, V17.S4, V18.S4, V19.S4]
20 VLD1.P 64(R2), [V20.S4, V21.S4, V22.S4, V23.S4]
21 VLD1.P 64(R2), [V24.S4, V25.S4, V26.S4, V27.S4]
22 VLD1 (R2), [V28.S4, V29.S4, V30.S4, V31.S4] //load 64*4bytes K constant(K0-K63)
23
24 blockloop:
25
26 VLD1.P 16(R1), [V4.B16] // load 16bytes message
27 VLD1.P 16(R1), [V5.B16] // load 16bytes message
28 VLD1.P 16(R1), [V6.B16] // load 16bytes message
29 VLD1.P 16(R1), [V7.B16] // load 16bytes message
30 VMOV V0.B16, V2.B16 // backup: VO h(dcba)
31 VMOV V1.B16, V3.B16 // backup: V1 h(hgfe)
32 VMOV V2.B16, V8.B16
33 VREV32 V4.B16, V4.B16 // prepare for using message in Byte format
34 VREV32 V5.B16, V5.B16
35 VREV32 V6.B16, V6.B16
36 VREV32 V7.B16, V7.B16
37
38 VADD V16.S4, V4.S4, V9.S4 // V18(W0+K0...W3+K3)
39 SHA256SU0 V5.S4, V4.S4 // V4: (su0(W1)+W0,...,su0(W4)+W3)
40 HASHUPDATE // H4
41
42 VADD V17.S4, V5.S4, V9.S4 // V18(W4+K4...W7+K7)
43 SHA256SU0 V6.S4, V5.S4 // V5: (su0(W5)+W4,...,su0(W8)+W7)
44 SHA256SU1 V7.S4, V6.S4, V4.S4 // V4: W16-W19
45 HASHUPDATE // H8
46
47 VADD V18.S4, V6.S4, V9.S4 // V18(W8+K8...W11+K11)
48 SHA256SU0 V7.S4, V6.S4 // V6: (su0(W9)+W8,...,su0(W12)+W11)
49 SHA256SU1 V4.S4, V7.S4, V5.S4 // V5: W20-W23
50 HASHUPDATE // H12
51
52 VADD V19.S4, V7.S4, V9.S4 // V18(W12+K12...W15+K15)
53 SHA256SU0 V4.S4, V7.S4 // V7: (su0(W13)+W12,...,su0(W16)+W15)
54 SHA256SU1 V5.S4, V4.S4, V6.S4 // V6: W24-W27
55 HASHUPDATE // H16
56
57 VADD V20.S4, V4.S4, V9.S4 // V18(W16+K16...W19+K19)
58 SHA256SU0 V5.S4, V4.S4 // V4: (su0(W17)+W16,...,su0(W20)+W19)
59 SHA256SU1 V6.S4, V5.S4, V7.S4 // V7: W28-W31
60 HASHUPDATE // H20
61
62 VADD V21.S4, V5.S4, V9.S4 // V18(W20+K20...W23+K23)
63 SHA256SU0 V6.S4, V5.S4 // V5: (su0(W21)+W20,...,su0(W24)+W23)
64 SHA256SU1 V7.S4, V6.S4, V4.S4 // V4: W32-W35
65 HASHUPDATE // H24
66
67 VADD V22.S4, V6.S4, V9.S4 // V18(W24+K24...W27+K27)
68 SHA256SU0 V7.S4, V6.S4 // V6: (su0(W25)+W24,...,su0(W28)+W27)
69 SHA256SU1 V4.S4, V7.S4, V5.S4 // V5: W36-W39
70 HASHUPDATE // H28
71
72 VADD V23.S4, V7.S4, V9.S4 // V18(W28+K28...W31+K31)
73 SHA256SU0 V4.S4, V7.S4 // V7: (su0(W29)+W28,...,su0(W32)+W31)
74 SHA256SU1 V5.S4, V4.S4, V6.S4 // V6: W40-W43
75 HASHUPDATE // H32
76
77 VADD V24.S4, V4.S4, V9.S4 // V18(W32+K32...W35+K35)
78 SHA256SU0 V5.S4, V4.S4 // V4: (su0(W33)+W32,...,su0(W36)+W35)
79 SHA256SU1 V6.S4, V5.S4, V7.S4 // V7: W44-W47
80 HASHUPDATE // H36
81
82 VADD V25.S4, V5.S4, V9.S4 // V18(W36+K36...W39+K39)
83 SHA256SU0 V6.S4, V5.S4 // V5: (su0(W37)+W36,...,su0(W40)+W39)
84 SHA256SU1 V7.S4, V6.S4, V4.S4 // V4: W48-W51
85 HASHUPDATE // H40
86
87 VADD V26.S4, V6.S4, V9.S4 // V18(W40+K40...W43+K43)
88 SHA256SU0 V7.S4, V6.S4 // V6: (su0(W41)+W40,...,su0(W44)+W43)
89 SHA256SU1 V4.S4, V7.S4, V5.S4 // V5: W52-W55
90 HASHUPDATE // H44
91
92 VADD V27.S4, V7.S4, V9.S4 // V18(W44+K44...W47+K47)
93 SHA256SU0 V4.S4, V7.S4 // V7: (su0(W45)+W44,...,su0(W48)+W47)
94 SHA256SU1 V5.S4, V4.S4, V6.S4 // V6: W56-W59
95 HASHUPDATE // H48
96
97 VADD V28.S4, V4.S4, V9.S4 // V18(W48+K48,...,W51+K51)
98 HASHUPDATE // H52
99 SHA256SU1 V6.S4, V5.S4, V7.S4 // V7: W60-W63
100
101 VADD V29.S4, V5.S4, V9.S4 // V18(W52+K52,...,W55+K55)
102 HASHUPDATE // H56
103
104 VADD V30.S4, V6.S4, V9.S4 // V18(W59+K59,...,W59+K59)
105 HASHUPDATE // H60
106
107 VADD V31.S4, V7.S4, V9.S4 // V18(W60+K60,...,W63+K63)
108 HASHUPDATE // H64
109
110 SUB $64, R3, R3 // message length - 64bytes, then compare with 64bytes
111 VADD V2.S4, V0.S4, V0.S4
112 VADD V3.S4, V1.S4, V1.S4
113 CBNZ R3, blockloop
114
115 sha256ret:
116
117 VST1 [V0.S4, V1.S4], (R0) // store hash value H
118 RET
119
120
View as plain text