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 # I let hardware handle unaligned input(*), except on page boundaries
11 # (see below for details). Otherwise straightforward implementation
12 # with X vector in register bank.
14 # (*) this means that this module is inappropriate for PPC403? Does
15 # anybody know if pre-POWER3 can sustain unaligned load?
18 # ----------------------------------
19 # PPC970,gcc-4.0.0 +76% +59%
20 # Power6,xlc-7 +68% +33%
24 if ($flavour =~ /64/) {
31 } elsif ($flavour =~ /32/) {
38 } else { die "nonsense $flavour"; }
40 # Define endianess based on flavour
42 $LITTLE_ENDIAN = ($flavour=~/le$/) ? $SIZE_T : 0;
44 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
45 ( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
46 ( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
47 die "can't locate ppc-xlate.pl";
49 open STDOUT,"| $^X $xlate $flavour ".shift || die "can't call $xlate: $!";
70 @V=($A,$B,$C,$D,$E,$T);
71 @X=("r16","r17","r18","r19","r20","r21","r22","r23",
72 "r24","r25","r26","r27","r28","r29","r30","r31");
75 my ($dst, $src, $temp_reg) = @_;
76 $code.=<<___ if (!$LITTLE_ENDIAN);
79 $code.=<<___ if ($LITTLE_ENDIAN);
81 rotlwi $dst,$temp_reg,8
82 rlwimi $dst,$temp_reg,24,0,7
83 rlwimi $dst,$temp_reg,24,16,23
88 my ($i,$a,$b,$c,$d,$e,$f)=@_;
91 # Since the last value of $f is discarded, we can use
92 # it as a temp reg to swap byte-order when needed.
93 loadbe("@X[$i]","`$i*4`($inp)",$f) if ($i==0);
94 loadbe("@X[$j]","`$j*4`($inp)",$f) if ($i<15);
95 $code.=<<___ if ($i<15);
106 $code.=<<___ if ($i>=15);
109 xor @X[$j%16],@X[$j%16],@X[($j+2)%16]
112 xor @X[$j%16],@X[$j%16],@X[($j+8)%16]
117 xor @X[$j%16],@X[$j%16],@X[($j+13)%16]
119 rotlwi @X[$j%16],@X[$j%16],1
124 my ($i,$a,$b,$c,$d,$e,$f)=@_;
126 $code.=<<___ if ($i<79);
129 xor @X[$j%16],@X[$j%16],@X[($j+2)%16]
132 xor @X[$j%16],@X[$j%16],@X[($j+8)%16]
136 xor @X[$j%16],@X[$j%16],@X[($j+13)%16]
138 rotlwi @X[$j%16],@X[$j%16],1
140 $code.=<<___ if ($i==79);
158 my ($i,$a,$b,$c,$d,$e,$f)=@_;
163 xor @X[$j%16],@X[$j%16],@X[($j+2)%16]
166 xor @X[$j%16],@X[$j%16],@X[($j+8)%16]
170 xor @X[$j%16],@X[$j%16],@X[($j+13)%16]
173 rotlwi @X[$j%16],@X[$j%16],1
182 .globl .sha1_block_data_order
184 .sha1_block_data_order:
185 $STU $sp,-$FRAME($sp)
187 $PUSH r15,`$FRAME-$SIZE_T*17`($sp)
188 $PUSH r16,`$FRAME-$SIZE_T*16`($sp)
189 $PUSH r17,`$FRAME-$SIZE_T*15`($sp)
190 $PUSH r18,`$FRAME-$SIZE_T*14`($sp)
191 $PUSH r19,`$FRAME-$SIZE_T*13`($sp)
192 $PUSH r20,`$FRAME-$SIZE_T*12`($sp)
193 $PUSH r21,`$FRAME-$SIZE_T*11`($sp)
194 $PUSH r22,`$FRAME-$SIZE_T*10`($sp)
195 $PUSH r23,`$FRAME-$SIZE_T*9`($sp)
196 $PUSH r24,`$FRAME-$SIZE_T*8`($sp)
197 $PUSH r25,`$FRAME-$SIZE_T*7`($sp)
198 $PUSH r26,`$FRAME-$SIZE_T*6`($sp)
199 $PUSH r27,`$FRAME-$SIZE_T*5`($sp)
200 $PUSH r28,`$FRAME-$SIZE_T*4`($sp)
201 $PUSH r29,`$FRAME-$SIZE_T*3`($sp)
202 $PUSH r30,`$FRAME-$SIZE_T*2`($sp)
203 $PUSH r31,`$FRAME-$SIZE_T*1`($sp)
204 $PUSH r0,`$FRAME+$LRSAVE`($sp)
214 bl Lsha1_block_private
217 ; PowerPC specification allows an implementation to be ill-behaved
218 ; upon unaligned access which crosses page boundary. "Better safe
219 ; than sorry" principle makes me treat it specially. But I don't
220 ; look for particular offending word, but rather for 64-byte input
221 ; block which crosses the boundary. Once found that block is aligned
222 ; and hashed separately...
226 andi. $t1,$t1,4095 ; distance to closest page boundary
227 srwi. $t1,$t1,6 ; t1/=64
230 ble- Laligned ; didn't cross the page boundary
233 bl Lsha1_block_private
237 addi r20,$sp,$LOCALS ; spot within the frame
251 $PUSH $inp,`$FRAME-$SIZE_T*18`($sp)
253 addi $inp,$sp,$LOCALS
255 bl Lsha1_block_private
256 $POP $inp,`$FRAME-$SIZE_T*18`($sp)
261 $POP r0,`$FRAME+$LRSAVE`($sp)
262 $POP r15,`$FRAME-$SIZE_T*17`($sp)
263 $POP r16,`$FRAME-$SIZE_T*16`($sp)
264 $POP r17,`$FRAME-$SIZE_T*15`($sp)
265 $POP r18,`$FRAME-$SIZE_T*14`($sp)
266 $POP r19,`$FRAME-$SIZE_T*13`($sp)
267 $POP r20,`$FRAME-$SIZE_T*12`($sp)
268 $POP r21,`$FRAME-$SIZE_T*11`($sp)
269 $POP r22,`$FRAME-$SIZE_T*10`($sp)
270 $POP r23,`$FRAME-$SIZE_T*9`($sp)
271 $POP r24,`$FRAME-$SIZE_T*8`($sp)
272 $POP r25,`$FRAME-$SIZE_T*7`($sp)
273 $POP r26,`$FRAME-$SIZE_T*6`($sp)
274 $POP r27,`$FRAME-$SIZE_T*5`($sp)
275 $POP r28,`$FRAME-$SIZE_T*4`($sp)
276 $POP r29,`$FRAME-$SIZE_T*3`($sp)
277 $POP r30,`$FRAME-$SIZE_T*2`($sp)
278 $POP r31,`$FRAME-$SIZE_T*1`($sp)
283 .byte 0,12,4,1,0x80,18,3,0
287 # This is private block function, which uses tailored calling
288 # interface, namely upon entry SHA_CTX is pre-loaded to given
289 # registers and counter register contains amount of chunks to
295 $code.=<<___; # load K_00_19
299 for($i=0;$i<20;$i++) { &BODY_00_19($i,@V); unshift(@V,pop(@V)); }
300 $code.=<<___; # load K_20_39
304 for(;$i<40;$i++) { &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
305 $code.=<<___; # load K_40_59
309 for(;$i<60;$i++) { &BODY_40_59($i,@V); unshift(@V,pop(@V)); }
310 $code.=<<___; # load K_60_79
314 for(;$i<80;$i++) { &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
331 addi $inp,$inp,`16*4`
332 bdnz- Lsha1_block_private
335 .byte 0,12,0x14,0,0,0,0,0
336 .size .sha1_block_data_order,.-.sha1_block_data_order
339 .asciz "SHA1 block transform for PPC, CRYPTOGAMS by <appro\@fy.chalmers.se>"
342 $code =~ s/\`([^\`]*)\`/eval $1/gem;