2 # Copyright 2012-2016 The OpenSSL Project Authors. All Rights Reserved.
4 # Licensed under the Apache License 2.0 (the "License"). You may not use
5 # this file except in compliance with the License. You can obtain a copy
6 # in the file LICENSE in the source distribution or at
7 # https://www.openssl.org/source/license.html
10 # ====================================================================
11 # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
12 # project. The module is, however, dual licensed under OpenSSL and
13 # CRYPTOGAMS licenses depending on where you obtain it. For further
14 # details see http://www.openssl.org/~appro/cryptogams/.
15 # ====================================================================
21 # If compared to compiler-generated code with similar characteristics,
22 # i.e. compiled with OPENSSL_SMALL_FOOTPRINT and utilizing SPLOOPs,
23 # this implementation is 25% smaller and >2x faster. In absolute terms
24 # performance is (quite impressive) ~6.5 cycles per processed byte.
25 # Fully unrolled assembler would be ~5x larger and is likely to be
26 # ~15% faster. It would be free from references to intermediate ring
27 # buffer, but put more pressure on L1P [both because the code would be
28 # larger and won't be using SPLOOP buffer]. There are no plans to
29 # realize fully unrolled variant though...
31 # !!! Note that this module uses AMR, which means that all interrupt
32 # service routines are expected to preserve it and for own well-being
35 $output = pop and open STDOUT,">$output";
37 ($CTX,$INP,$NUM) = ("A4","B4","A6"); # arguments
39 ($A,$B,$C,$D,$E, $Arot,$F,$F0,$T,$K) = map("A$_",(16..20, 21..25));
40 ($X0,$X2,$X8,$X13) = ("A26","B26","A27","B27");
41 ($TX0,$TX1,$TX2,$TX3) = map("B$_",(28..31));
42 ($XPA,$XPB) = ("A5","B5"); # X circular buffer
43 ($Actx,$Bctx,$Cctx,$Dctx,$Ectx) = map("A$_",(3,6..9)); # zaps $NUM
48 .if .ASSEMBLER_VERSION<7000000
52 .asg sha1_block_data_order,_sha1_block_data_order
64 .global _sha1_block_data_order
65 _sha1_block_data_order:
66 .asmfunc stack_usage(64)
67 MV $NUM,A0 ; reassign $NUM
69 [!A0] BNOP RA ; if ($NUM==0) return;
70 || [A0] STW FP,*SP--[16] ; save frame pointer and alloca(64)
72 [A0] LDW *${CTX}[0],$A ; load A-E...
73 || [A0] AND B0,SP,SP ; align stack at 64 bytes
74 [A0] LDW *${CTX}[1],$B
75 || [A0] SUBAW SP,2,SP ; reserve two words above buffer
76 [A0] LDW *${CTX}[2],$C
77 || [A0] MVK 0x00404,B0
78 [A0] LDW *${CTX}[3],$D
79 || [A0] MVKH 0x50000,B0 ; 0x050404, 64 bytes for $XP[AB]
80 [A0] LDW *${CTX}[4],$E
81 || [A0] MVC B0,AMR ; setup circular addressing
82 LDNW *${INP}++,$TX1 ; pre-fetch input
90 MVKH 0x5a820000,$K ; K_00_19
94 ;;==================================================
95 SPLOOPD 5 ; BODY_00_13
104 || ADD $K,$E,$T ; T=E+K
106 XOR $F0,$F,$F ; F_00_19(B,C,D)
110 || LDNW *${INP}++,$TX1
112 ADD $F,$T,$T ; T+=F_00_19(B,C,D)
113 || ROTL $B,30,$C ; C=ROL(B,30)
114 || SWAP4 $TX2,$TX3 ; byte swap
116 ADD $Arot,$T,$T ; T+=ROL(A,5)
119 ADD $TX3,$T,$A ; A=T+Xi
120 || STW $TX3,*${XPB}++
122 ;;==================================================
123 ROTL $A,5,$Arot ; BODY_14
126 || ADD $K,$E,$T ; T=E+K
128 XOR $F0,$F,$F ; F_00_19(B,C,D)
132 || LDNW *${INP}++,$TX1
134 ADD $F,$T,$T ; T+=F_00_19(B,C,D)
135 || ROTL $B,30,$C ; C=ROL(B,30)
136 || SWAP4 $TX2,$TX2 ; byte swap
137 || LDW *${XPA}++,$X0 ; fetches from X ring buffer are
138 || LDW *${XPB}[4],$X2 ; 2 iterations ahead
140 ADD $Arot,$T,$T ; T+=ROL(A,5)
142 || LDW *${XPA}[7],$X8
143 || MV $TX3,$X13 ; || LDW *${XPB}[15],$X13
146 ADD $TX2,$T,$A ; A=T+Xi
147 || STW $TX2,*${XPB}++
148 ;;==================================================
149 ROTL $A,5,$Arot ; BODY_15
152 || ADD $K,$E,$T ; T=E+K
154 XOR $F0,$F,$F ; F_00_19(B,C,D)
159 ADD $F,$T,$T ; T+=F_00_19(B,C,D)
160 || ROTL $B,30,$C ; C=ROL(B,30)
161 || SWAP4 $TX2,$TX2 ; byte swap
162 || XOR $X0,$X2,$TX0 ; Xupdate XORs are 1 iteration ahead
164 || LDW *${XPB}[4],$X2
166 ADD $Arot,$T,$T ; T+=ROL(A,5)
169 || LDW *${XPA}[7],$X8
170 || MV $TX3,$X13 ; || LDW *${XPB}[15],$X13
173 ADD $TX2,$T,$A ; A=T+Xi
174 || STW $TX2,*${XPB}++
175 || XOR $TX0,$TX1,$TX1
177 ;;==================================================
178 SPLOOPD 5 ; BODY_16_19
184 || ADD $K,$E,$T ; T=E+K
185 || ROTL $TX1,1,$TX2 ; Xupdate output
187 XOR $F0,$F,$F ; F_00_19(B,C,D)
191 ADD $F,$T,$T ; T+=F_00_19(B,C,D)
192 || ROTL $B,30,$C ; C=ROL(B,30)
195 || LDW *${XPB}[4],$X2
197 ADD $Arot,$T,$T ; T+=ROL(A,5)
200 || LDW *${XPA}[7],$X8
201 || MV $TX3,$X13 ; || LDW *${XPB}[15],$X13
204 ADD $TX2,$T,$A ; A=T+Xi
205 || STW $TX2,*${XPB}++
206 || XOR $TX0,$TX1,$TX1
211 MVKH 0x6ed90000,$K ; K_20_39
215 ;;==================================================
216 SPLOOPD 5 ; BODY_20_39
221 || ADD $K,$E,$T ; T=E+K
222 || ROTL $TX1,1,$TX2 ; Xupdate output
224 XOR $D,$F,$F ; F_20_39(B,C,D)
228 ADD $F,$T,$T ; T+=F_20_39(B,C,D)
229 || ROTL $B,30,$C ; C=ROL(B,30)
232 || LDW *${XPB}[4],$X2
234 ADD $Arot,$T,$T ; T+=ROL(A,5)
237 || LDW *${XPA}[7],$X8
238 || MV $TX3,$X13 ; || LDW *${XPB}[15],$X13
241 ADD $TX2,$T,$A ; A=T+Xi
242 || STW $TX2,*${XPB}++ ; last one is redundant
243 || XOR $TX0,$TX1,$TX1
246 $code.=<<___ if (!shift);
248 MVKH 0x8f1b0000,$K ; K_40_59
252 ;;==================================================
253 SPLOOPD 5 ; BODY_40_59
261 || ADD $K,$E,$T ; T=E+K
262 || ROTL $TX1,1,$TX2 ; Xupdate output
264 XOR $F0,$F,$F ; F_40_59(B,C,D)
268 ADD $F,$T,$T ; T+=F_40_59(B,C,D)
269 || ROTL $B,30,$C ; C=ROL(B,30)
272 || LDW *${XPB}[4],$X2
274 ADD $Arot,$T,$T ; T+=ROL(A,5)
277 || LDW *${XPA}[7],$X8
278 || MV $TX3,$X13 ; || LDW *${XPB}[15],$X13
281 ADD $TX2,$T,$A ; A=T+Xi
282 || STW $TX2,*${XPB}++
283 || XOR $TX0,$TX1,$TX1
290 MVKH 0xca620000,$K ; K_60_79
292 &BODY_20_39(-1); # BODY_60_78
294 ;;==================================================
296 || ROTL $A,5,$Arot ; BODY_79
298 || ROTL $TX1,1,$TX2 ; Xupdate output
300 [A0] LDNW *${INP}++,$TX1 ; pre-fetch input
301 || ADD $K,$E,$T ; T=E+K
302 || XOR $D,$F,$F ; F_20_39(B,C,D)
304 ADD $F,$T,$T ; T+=F_20_39(B,C,D)
305 || ADD $Ectx,$D,$E ; E=D,E+=Ectx
306 || ADD $Dctx,$C,$D ; D=C,D+=Dctx
307 || ROTL $B,30,$C ; C=ROL(B,30)
309 ADD $Arot,$T,$T ; T+=ROL(A,5)
310 || ADD $Bctx,$A,$B ; B=A,B+=Bctx
312 ADD $TX2,$T,$A ; A=T+Xi
314 ADD $Actx,$A,$A ; A+=Actx
315 || ADD $Cctx,$C,$C ; C+=Cctx
319 || MV FP,SP ; restore stack pointer
320 || LDW *FP[0],FP ; restore frame pointer
321 STW $A,*${CTX}[0] ; emit A-E...
324 || MVC B0,AMR ; clear AMR
331 .cstring "SHA1 block transform for C64x+, CRYPTOGAMS by <appro\@openssl.org>"
336 close STDOUT or die "error closing STDOUT";