3 # ====================================================================
4 # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
5 # project. The module is, however, dual licensed under OpenSSL and
6 # CRYPTOGAMS licenses depending on where you obtain it. For further
7 # details see http://www.openssl.org/~appro/cryptogams/.
8 # ====================================================================
14 # If compared to compiler-generated code with similar characteristics,
15 # i.e. compiled with OPENSSL_SMALL_FOOTPRINT and utilizing SPLOOPs,
16 # this implementation is 25% smaller and >2x faster. In absolute terms
17 # performance is (quite impressive) ~6.5 cycles per processed byte.
18 # Fully unrolled assembler would be ~5x larger and is likely to be
19 # ~15% faster. It would be free from references to intermediate ring
20 # buffer, but put more pressure on L1P [both because the code would be
21 # larger and won't be using SPLOOP buffer]. There are no plans to
22 # realize fully unrolled variant though...
24 # !!! Note that this module uses AMR, which means that all interrupt
25 # service routines are expected to preserve it and for own well-being
28 while (($output=shift) && ($output!~/\w[\w\-]*\.\w+$/)) {}
29 open STDOUT,">$output";
31 ($CTX,$INP,$NUM) = ("A4","B4","A6"); # arguments
33 ($A,$B,$C,$D,$E, $Arot,$F,$F0,$T,$K) = map("A$_",(16..20, 21..25));
34 ($X0,$X2,$X8,$X13) = ("A26","B26","A27","B27");
35 ($TX0,$TX1,$TX2,$TX3) = map("B$_",(28..31));
36 ($XPA,$XPB) = ("A5","B5"); # X circular buffer
37 ($Actx,$Bctx,$Cctx,$Dctx,$Ectx) = map("A$_",(3,6..9)); # zaps $NUM
51 .global _sha1_block_data_order
52 _sha1_block_data_order:
53 .asmfunc stack_usage(64)
54 MV $NUM,A0 ; reassign $NUM
56 [!A0] BNOP RA ; if ($NUM==0) return;
57 || [A0] STW FP,*SP--[16] ; save frame pointer and alloca(64)
59 [A0] LDW *${CTX}[0],$A ; load A-E...
60 || [A0] AND B0,SP,SP ; align stack at 64 bytes
61 [A0] LDW *${CTX}[1],$B
62 || [A0] SUBAW SP,2,SP ; reserve two words above buffer
63 [A0] LDW *${CTX}[2],$C
64 || [A0] MVK 0x00404,B0
65 [A0] LDW *${CTX}[3],$D
66 || [A0] MVKH 0x50000,B0 ; 0x050404, 64 bytes for $XP[AB]
67 [A0] LDW *${CTX}[4],$E
68 || [A0] MVC B0,AMR ; setup circular addressing
69 LDNW *${INP}++,$TX1 ; pre-fetch input
77 MVKH 0x5a820000,$K ; K_00_19
81 ;;==================================================
82 SPLOOPD 5 ; BODY_00_13
91 || ADD $K,$E,$T ; T=E+K
93 XOR $F0,$F,$F ; F_00_19(B,C,D)
97 || LDNW *${INP}++,$TX1
99 ADD $F,$T,$T ; T+=F_00_19(B,C,D)
100 || ROTL $B,30,$C ; C=ROL(B,30)
101 || SWAP4 $TX2,$TX3 ; byte swap
103 ADD $Arot,$T,$T ; T+=ROL(A,5)
106 ADD $TX3,$T,$A ; A=T+Xi
107 || STW $TX3,*${XPB}++
109 ;;==================================================
110 ROTL $A,5,$Arot ; BODY_14
113 || ADD $K,$E,$T ; T=E+K
115 XOR $F0,$F,$F ; F_00_19(B,C,D)
119 || LDNW *${INP}++,$TX1
121 ADD $F,$T,$T ; T+=F_00_19(B,C,D)
122 || ROTL $B,30,$C ; C=ROL(B,30)
123 || SWAP4 $TX2,$TX2 ; byte swap
124 || LDW *${XPA}++,$X0 ; fetches from X ring buffer are
125 || LDW *${XPB}[4],$X2 ; 2 iterations ahead
127 ADD $Arot,$T,$T ; T+=ROL(A,5)
129 || LDW *${XPA}[7],$X8
130 || MV $TX3,$X13 ; || LDW *${XPB}[15],$X13
133 ADD $TX2,$T,$A ; A=T+Xi
134 || STW $TX2,*${XPB}++
135 ;;==================================================
136 ROTL $A,5,$Arot ; BODY_15
139 || ADD $K,$E,$T ; T=E+K
141 XOR $F0,$F,$F ; F_00_19(B,C,D)
146 ADD $F,$T,$T ; T+=F_00_19(B,C,D)
147 || ROTL $B,30,$C ; C=ROL(B,30)
148 || SWAP4 $TX2,$TX2 ; byte swap
149 || XOR $X0,$X2,$TX0 ; Xupdate XORs are 1 iteration ahead
151 || LDW *${XPB}[4],$X2
153 ADD $Arot,$T,$T ; T+=ROL(A,5)
156 || LDW *${XPA}[7],$X8
157 || MV $TX3,$X13 ; || LDW *${XPB}[15],$X13
160 ADD $TX2,$T,$A ; A=T+Xi
161 || STW $TX2,*${XPB}++
162 || XOR $TX0,$TX1,$TX1
164 ;;==================================================
165 SPLOOPD 5 ; BODY_16_19
171 || ADD $K,$E,$T ; T=E+K
172 || ROTL $TX1,1,$TX2 ; Xupdate output
174 XOR $F0,$F,$F ; F_00_19(B,C,D)
178 ADD $F,$T,$T ; T+=F_00_19(B,C,D)
179 || ROTL $B,30,$C ; C=ROL(B,30)
182 || LDW *${XPB}[4],$X2
184 ADD $Arot,$T,$T ; T+=ROL(A,5)
187 || LDW *${XPA}[7],$X8
188 || MV $TX3,$X13 ; || LDW *${XPB}[15],$X13
191 ADD $TX2,$T,$A ; A=T+Xi
192 || STW $TX2,*${XPB}++
193 || XOR $TX0,$TX1,$TX1
198 MVKH 0x6ed90000,$K ; K_20_39
202 ;;==================================================
203 SPLOOPD 5 ; BODY_20_39
208 || ADD $K,$E,$T ; T=E+K
209 || ROTL $TX1,1,$TX2 ; Xupdate output
211 XOR $D,$F,$F ; F_20_39(B,C,D)
215 ADD $F,$T,$T ; T+=F_20_39(B,C,D)
216 || ROTL $B,30,$C ; C=ROL(B,30)
219 || LDW *${XPB}[4],$X2
221 ADD $Arot,$T,$T ; T+=ROL(A,5)
224 || LDW *${XPA}[7],$X8
225 || MV $TX3,$X13 ; || LDW *${XPB}[15],$X13
228 ADD $TX2,$T,$A ; A=T+Xi
229 || STW $TX2,*${XPB}++ ; last one is redundant
230 || XOR $TX0,$TX1,$TX1
233 $code.=<<___ if (!shift);
235 MVKH 0x8f1b0000,$K ; K_40_59
239 ;;==================================================
240 SPLOOPD 5 ; BODY_40_59
248 || ADD $K,$E,$T ; T=E+K
249 || ROTL $TX1,1,$TX2 ; Xupdate output
251 XOR $F0,$F,$F ; F_40_59(B,C,D)
255 ADD $F,$T,$T ; T+=F_40_59(B,C,D)
256 || ROTL $B,30,$C ; C=ROL(B,30)
259 || LDW *${XPB}[4],$X2
261 ADD $Arot,$T,$T ; T+=ROL(A,5)
264 || LDW *${XPA}[7],$X8
265 || MV $TX3,$X13 ; || LDW *${XPB}[15],$X13
268 ADD $TX2,$T,$A ; A=T+Xi
269 || STW $TX2,*${XPB}++
270 || XOR $TX0,$TX1,$TX1
277 MVKH 0xca620000,$K ; K_60_79
279 &BODY_20_39(-1); # BODY_60_78
281 ;;==================================================
283 || ROTL $A,5,$Arot ; BODY_79
285 || ROTL $TX1,1,$TX2 ; Xupdate output
287 [A0] LDNW *${INP}++,$TX1 ; pre-fetch input
288 || ADD $K,$E,$T ; T=E+K
289 || XOR $D,$F,$F ; F_20_39(B,C,D)
291 ADD $F,$T,$T ; T+=F_20_39(B,C,D)
292 || ADD $Ectx,$D,$E ; E=D,E+=Ectx
293 || ADD $Dctx,$C,$D ; D=C,D+=Dctx
294 || ROTL $B,30,$C ; C=ROL(B,30)
296 ADD $Arot,$T,$T ; T+=ROL(A,5)
297 || ADD $Bctx,$A,$B ; B=A,B+=Bctx
299 ADD $TX2,$T,$A ; A=T+Xi
301 ADD $Actx,$A,$A ; A+=Actx
302 || ADD $Cctx,$C,$C ; C+=Cctx
306 || MV FP,SP ; restore stack pointer
307 || LDW *FP[0],FP ; restore frame pointer
308 STW $A,*${CTX}[0] ; emit A-E...
311 || MVC B0,AMR ; clear AMR
318 .cstring "SHA1 block transform for C64x+, CRYPTOGAMS by <appro\@openssl.org>"