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
42 .asg sha1_block_data_order,_sha1_block_data_order
54 .global _sha1_block_data_order
55 _sha1_block_data_order:
56 .asmfunc stack_usage(64)
57 MV $NUM,A0 ; reassign $NUM
59 [!A0] BNOP RA ; if ($NUM==0) return;
60 || [A0] STW FP,*SP--[16] ; save frame pointer and alloca(64)
62 [A0] LDW *${CTX}[0],$A ; load A-E...
63 || [A0] AND B0,SP,SP ; align stack at 64 bytes
64 [A0] LDW *${CTX}[1],$B
65 || [A0] SUBAW SP,2,SP ; reserve two words above buffer
66 [A0] LDW *${CTX}[2],$C
67 || [A0] MVK 0x00404,B0
68 [A0] LDW *${CTX}[3],$D
69 || [A0] MVKH 0x50000,B0 ; 0x050404, 64 bytes for $XP[AB]
70 [A0] LDW *${CTX}[4],$E
71 || [A0] MVC B0,AMR ; setup circular addressing
72 LDNW *${INP}++,$TX1 ; pre-fetch input
80 MVKH 0x5a820000,$K ; K_00_19
84 ;;==================================================
85 SPLOOPD 5 ; BODY_00_13
94 || ADD $K,$E,$T ; T=E+K
96 XOR $F0,$F,$F ; F_00_19(B,C,D)
100 || LDNW *${INP}++,$TX1
102 ADD $F,$T,$T ; T+=F_00_19(B,C,D)
103 || ROTL $B,30,$C ; C=ROL(B,30)
104 || SWAP4 $TX2,$TX3 ; byte swap
106 ADD $Arot,$T,$T ; T+=ROL(A,5)
109 ADD $TX3,$T,$A ; A=T+Xi
110 || STW $TX3,*${XPB}++
112 ;;==================================================
113 ROTL $A,5,$Arot ; BODY_14
116 || ADD $K,$E,$T ; T=E+K
118 XOR $F0,$F,$F ; F_00_19(B,C,D)
122 || LDNW *${INP}++,$TX1
124 ADD $F,$T,$T ; T+=F_00_19(B,C,D)
125 || ROTL $B,30,$C ; C=ROL(B,30)
126 || SWAP4 $TX2,$TX2 ; byte swap
127 || LDW *${XPA}++,$X0 ; fetches from X ring buffer are
128 || LDW *${XPB}[4],$X2 ; 2 iterations ahead
130 ADD $Arot,$T,$T ; T+=ROL(A,5)
132 || LDW *${XPA}[7],$X8
133 || MV $TX3,$X13 ; || LDW *${XPB}[15],$X13
136 ADD $TX2,$T,$A ; A=T+Xi
137 || STW $TX2,*${XPB}++
138 ;;==================================================
139 ROTL $A,5,$Arot ; BODY_15
142 || ADD $K,$E,$T ; T=E+K
144 XOR $F0,$F,$F ; F_00_19(B,C,D)
149 ADD $F,$T,$T ; T+=F_00_19(B,C,D)
150 || ROTL $B,30,$C ; C=ROL(B,30)
151 || SWAP4 $TX2,$TX2 ; byte swap
152 || XOR $X0,$X2,$TX0 ; Xupdate XORs are 1 iteration ahead
154 || LDW *${XPB}[4],$X2
156 ADD $Arot,$T,$T ; T+=ROL(A,5)
159 || LDW *${XPA}[7],$X8
160 || MV $TX3,$X13 ; || LDW *${XPB}[15],$X13
163 ADD $TX2,$T,$A ; A=T+Xi
164 || STW $TX2,*${XPB}++
165 || XOR $TX0,$TX1,$TX1
167 ;;==================================================
168 SPLOOPD 5 ; BODY_16_19
174 || ADD $K,$E,$T ; T=E+K
175 || ROTL $TX1,1,$TX2 ; Xupdate output
177 XOR $F0,$F,$F ; F_00_19(B,C,D)
181 ADD $F,$T,$T ; T+=F_00_19(B,C,D)
182 || ROTL $B,30,$C ; C=ROL(B,30)
185 || LDW *${XPB}[4],$X2
187 ADD $Arot,$T,$T ; T+=ROL(A,5)
190 || LDW *${XPA}[7],$X8
191 || MV $TX3,$X13 ; || LDW *${XPB}[15],$X13
194 ADD $TX2,$T,$A ; A=T+Xi
195 || STW $TX2,*${XPB}++
196 || XOR $TX0,$TX1,$TX1
201 MVKH 0x6ed90000,$K ; K_20_39
205 ;;==================================================
206 SPLOOPD 5 ; BODY_20_39
211 || ADD $K,$E,$T ; T=E+K
212 || ROTL $TX1,1,$TX2 ; Xupdate output
214 XOR $D,$F,$F ; F_20_39(B,C,D)
218 ADD $F,$T,$T ; T+=F_20_39(B,C,D)
219 || ROTL $B,30,$C ; C=ROL(B,30)
222 || LDW *${XPB}[4],$X2
224 ADD $Arot,$T,$T ; T+=ROL(A,5)
227 || LDW *${XPA}[7],$X8
228 || MV $TX3,$X13 ; || LDW *${XPB}[15],$X13
231 ADD $TX2,$T,$A ; A=T+Xi
232 || STW $TX2,*${XPB}++ ; last one is redundant
233 || XOR $TX0,$TX1,$TX1
236 $code.=<<___ if (!shift);
238 MVKH 0x8f1b0000,$K ; K_40_59
242 ;;==================================================
243 SPLOOPD 5 ; BODY_40_59
251 || ADD $K,$E,$T ; T=E+K
252 || ROTL $TX1,1,$TX2 ; Xupdate output
254 XOR $F0,$F,$F ; F_40_59(B,C,D)
258 ADD $F,$T,$T ; T+=F_40_59(B,C,D)
259 || ROTL $B,30,$C ; C=ROL(B,30)
262 || LDW *${XPB}[4],$X2
264 ADD $Arot,$T,$T ; T+=ROL(A,5)
267 || LDW *${XPA}[7],$X8
268 || MV $TX3,$X13 ; || LDW *${XPB}[15],$X13
271 ADD $TX2,$T,$A ; A=T+Xi
272 || STW $TX2,*${XPB}++
273 || XOR $TX0,$TX1,$TX1
280 MVKH 0xca620000,$K ; K_60_79
282 &BODY_20_39(-1); # BODY_60_78
284 ;;==================================================
286 || ROTL $A,5,$Arot ; BODY_79
288 || ROTL $TX1,1,$TX2 ; Xupdate output
290 [A0] LDNW *${INP}++,$TX1 ; pre-fetch input
291 || ADD $K,$E,$T ; T=E+K
292 || XOR $D,$F,$F ; F_20_39(B,C,D)
294 ADD $F,$T,$T ; T+=F_20_39(B,C,D)
295 || ADD $Ectx,$D,$E ; E=D,E+=Ectx
296 || ADD $Dctx,$C,$D ; D=C,D+=Dctx
297 || ROTL $B,30,$C ; C=ROL(B,30)
299 ADD $Arot,$T,$T ; T+=ROL(A,5)
300 || ADD $Bctx,$A,$B ; B=A,B+=Bctx
302 ADD $TX2,$T,$A ; A=T+Xi
304 ADD $Actx,$A,$A ; A+=Actx
305 || ADD $Cctx,$C,$C ; C+=Cctx
309 || MV FP,SP ; restore stack pointer
310 || LDW *FP[0],FP ; restore frame pointer
311 STW $A,*${CTX}[0] ; emit A-E...
314 || MVC B0,AMR ; clear AMR
321 .cstring "SHA1 block transform for C64x+, CRYPTOGAMS by <appro\@openssl.org>"