3 # ====================================================================
4 # Written by Andy Polyakov <appro@fy.chalmers.se> 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 # ====================================================================
10 # sha1_block procedure for ARMv4.
14 # Size/performance trade-off
15 # ====================================================================
16 # impl size in bytes comp cycles[*] measured performance
17 # ====================================================================
19 # armv4-small 392/+29% 1958/+64% 2250/+96%
20 # armv4-compact 740/+89% 1552/+26% 1840/+22%
21 # armv4-large 1420/+92% 1307/+19% 1500/+23%
22 # full unroll ~5100/+260% ~1260/+4% ~1500/+0%
23 # ====================================================================
24 # thumb = same as 'small' but in Thumb instructions[**] and
25 # with recurring code in two private functions;
26 # small = detached Xload/update, loops are folded;
27 # compact = detached Xload/update, 5x unroll;
28 # large = interleaved Xload/update, 5x unroll;
29 # full unroll = interleaved Xload/update, full unroll, estimated[!];
31 # [*] Manually counted instructions in "grand" loop body. Measured
32 # performance is affected by prologue and epilogue overhead,
33 # i-cache availability, branch penalties, etc.
34 # [**] While each Thumb instruction is twice smaller, they are not as
35 # diverse as ARM ones: e.g., there are only two arithmetic
36 # instructions with 3 arguments, no [fixed] rotate, addressing
37 # modes are limited. As result it takes more instructions to do
38 # the same job in Thumb, therefore the code is never twice as
39 # small and always slower.
42 open STDOUT,">$output";
59 # One can optimize this for aligned access on big-endian architecture,
60 # but code's endian neutrality makes it too pretty:-)
62 my ($a,$b,$c,$d,$e)=@_;
67 add $e,$K,$e,ror#2 @ E+=K_00_19
71 add $e,$e,$a,ror#27 @ E+=ROR(A,27)
73 add $e,$e,$t0 @ E+=X[i]
74 eor $t1,$c,$d @ F_xx_xx
79 my ($a,$b,$c,$d,$e,$flag)=@_;
84 add $e,$K,$e,ror#2 @ E+=K_xx_xx
87 add $e,$e,$a,ror#27 @ E+=ROR(A,27)
91 $code.=<<___ if (!defined($flag));
92 eor $t1,$c,$d @ F_xx_xx, but not in 40_59
96 add $e,$e,$t0 @ E+=X[i]
102 my ($a,$b,$c,$d,$e)=@_;
106 eor $t1,$t1,$d,ror#2 @ F_00_19(B,C,D)
107 add $e,$e,$t1 @ E+=F_00_19(B,C,D)
112 my ($a,$b,$c,$d,$e)=@_;
116 eor $t1,$t1,$d,ror#2 @ F_00_19(B,C,D)
117 add $e,$e,$t1 @ E+=F_00_19(B,C,D)
122 my ($a,$b,$c,$d,$e)=@_;
125 eor $t1,$b,$t1,ror#2 @ F_20_39(B,C,D)
126 add $e,$e,$t1 @ E+=F_20_39(B,C,D)
131 my ($a,$b,$c,$d,$e)=@_;
137 orr $t1,$t1,$t2 @ F_40_59(B,C,D)
138 add $e,$e,$t1 @ E+=F_40_59(B,C,D)
145 .global sha1_block_data_order
146 .type sha1_block_data_order,%function
149 sha1_block_data_order:
150 stmdb sp!,{r4-r12,lr}
151 add $len,$inp,$len,lsl#6 @ $len to point at the end of $inp
152 ldmia $ctx,{$a,$b,$c,$d,$e}
159 mov $e,$e,ror#30 @ [6]
162 for($i=0;$i<5;$i++) {
163 &BODY_00_15(@V); unshift(@V,pop(@V));
167 bne .L_00_15 @ [((11+4)*5+2)*3]
169 &BODY_00_15(@V); unshift(@V,pop(@V));
170 &BODY_16_19(@V); unshift(@V,pop(@V));
171 &BODY_16_19(@V); unshift(@V,pop(@V));
172 &BODY_16_19(@V); unshift(@V,pop(@V));
173 &BODY_16_19(@V); unshift(@V,pop(@V));
176 ldr $K,.LK_20_39 @ [+15+16*4]
178 cmn sp,#0 @ [+3], clear carry to denote 20_39
181 for($i=0;$i<5;$i++) {
182 &BODY_20_39(@V); unshift(@V,pop(@V));
185 teq $Xi,sp @ preserve carry
186 bne .L_20_39_or_60_79 @ [+((12+3)*5+2)*4]
187 bcs .L_done @ [+((12+3)*5+2)*4], spare 300 bytes
190 sub sp,sp,#20*4 @ [+2]
193 for($i=0;$i<5;$i++) {
194 &BODY_40_59(@V); unshift(@V,pop(@V));
198 bne .L_40_59 @ [+((12+5)*5+2)*4]
202 cmp sp,#0 @ set carry to denote 60_79
203 b .L_20_39_or_60_79 @ [+4], spare 300 bytes
205 add sp,sp,#80*4 @ "deallocate" stack frame
206 ldmia $ctx,{$K,$t0,$t1,$t2,$Xi}
212 stmia $ctx,{$a,$b,$c,$d,$e}
214 bne .Lloop @ [+18], total 1307
216 ldmia sp!,{r4-r12,lr}
218 moveq pc,lr @ be binary compatible with V4, yet
219 bx lr @ interoperable with Thumb ISA:-)
221 .LK_00_19: .word 0x5a827999
222 .LK_20_39: .word 0x6ed9eba1
223 .LK_40_59: .word 0x8f1bbcdc
224 .LK_60_79: .word 0xca62c1d6
225 .size sha1_block_data_order,.-sha1_block_data_order
226 .asciz "SHA1 block transform for ARMv4, CRYPTOGAMS by <appro\@openssl.org>"
230 close STDOUT; # enforce flush