7da9cf96ce7e81388f25406619f8a538e8cd66ac
[openssl.git] / crypto / sha / asm / sha512-ppc.pl
1 #!/usr/bin/env perl
2
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 # ====================================================================
9
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. The module is big-endian [which is
13 # not big deal as there're no little-endian targets left around].
14
15 #                       sha256          |       sha512
16 #                       -m64    -m32    |       -m64    -m32
17 # --------------------------------------+-----------------------
18 # PPC970,gcc-4.0.0      +50%    +38%    |       +40%    +410%(*)
19 # Power6,xlc-7          +150%   +90%    |       +100%   +430%(*)
20 #
21 # (*)   64-bit code in 32-bit application context, which actually is
22 #       on TODO list. It should be noted that for safe deployment in
23 #       32-bit *mutli-threaded* context asyncronous signals should be
24 #       blocked upon entry to SHA512 block routine. This is because
25 #       32-bit signaling procedure invalidates upper halves of GPRs.
26 #       Context switch procedure preserves them, but not signaling:-(
27
28 # Second version is true multi-thread safe. Trouble with the original
29 # version was that it was using thread local storage pointer register.
30 # Well, it scrupulously preserved it, but the problem would arise the
31 # moment asynchronous signal was delivered and signal handler would
32 # dereference the TLS pointer. While it's never the case in openssl
33 # application or test suite, we have to respect this scenario and not
34 # use TLS pointer register. Alternative would be to require caller to
35 # block signals prior calling this routine. For the record, in 32-bit
36 # context R2 serves as TLS pointer, while in 64-bit context - R13.
37
38 $flavour=shift;
39 $output =shift;
40
41 if ($flavour =~ /64/) {
42         $SIZE_T=8;
43         $STU="stdu";
44         $UCMP="cmpld";
45         $SHL="sldi";
46         $POP="ld";
47         $PUSH="std";
48 } elsif ($flavour =~ /32/) {
49         $SIZE_T=4;
50         $STU="stwu";
51         $UCMP="cmplw";
52         $SHL="slwi";
53         $POP="lwz";
54         $PUSH="stw";
55 } else { die "nonsense $flavour"; }
56
57 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
58 ( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
59 ( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
60 die "can't locate ppc-xlate.pl";
61
62 open STDOUT,"| $^X $xlate $flavour $output" || die "can't call $xlate: $!";
63
64 if ($output =~ /512/) {
65         $func="sha512_block_data_order";
66         $SZ=8;
67         @Sigma0=(28,34,39);
68         @Sigma1=(14,18,41);
69         @sigma0=(1,  8, 7);
70         @sigma1=(19,61, 6);
71         $rounds=80;
72         $LD="ld";
73         $ST="std";
74         $ROR="rotrdi";
75         $SHR="srdi";
76 } else {
77         $func="sha256_block_data_order";
78         $SZ=4;
79         @Sigma0=( 2,13,22);
80         @Sigma1=( 6,11,25);
81         @sigma0=( 7,18, 3);
82         @sigma1=(17,19,10);
83         $rounds=64;
84         $LD="lwz";
85         $ST="stw";
86         $ROR="rotrwi";
87         $SHR="srwi";
88 }
89
90 $FRAME=32*$SIZE_T;
91
92 $sp ="r1";
93 $toc="r2";
94 $ctx="r3";      # zapped by $a0
95 $inp="r4";      # zapped by $a1
96 $num="r5";      # zapped by $t0
97
98 $T  ="r0";
99 $a0 ="r3";
100 $a1 ="r4";
101 $t0 ="r5";
102 $t1 ="r6";
103 $Tbl="r7";
104
105 $A  ="r8";
106 $B  ="r9";
107 $C  ="r10";
108 $D  ="r11";
109 $E  ="r12";
110 $F  ="r13";     $F="r2" if ($SIZE_T==8);# reassigned to exempt TLS pointer
111 $G  ="r14";
112 $H  ="r15";
113
114 @V=($A,$B,$C,$D,$E,$F,$G,$H);
115 @X=("r16","r17","r18","r19","r20","r21","r22","r23",
116     "r24","r25","r26","r27","r28","r29","r30","r31");
117
118 $inp="r31";     # reassigned $inp! aliases with @X[15]
119
120 sub ROUND_00_15 {
121 my ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_;
122 $code.=<<___;
123         $LD     $T,`$i*$SZ`($Tbl)
124         $ROR    $a0,$e,$Sigma1[0]
125         $ROR    $a1,$e,$Sigma1[1]
126         and     $t0,$f,$e
127         andc    $t1,$g,$e
128         add     $T,$T,$h
129         xor     $a0,$a0,$a1
130         $ROR    $a1,$a1,`$Sigma1[2]-$Sigma1[1]`
131         or      $t0,$t0,$t1             ; Ch(e,f,g)
132         add     $T,$T,@X[$i]
133         xor     $a0,$a0,$a1             ; Sigma1(e)
134         add     $T,$T,$t0
135         add     $T,$T,$a0
136
137         $ROR    $a0,$a,$Sigma0[0]
138         $ROR    $a1,$a,$Sigma0[1]
139         and     $t0,$a,$b
140         and     $t1,$a,$c
141         xor     $a0,$a0,$a1
142         $ROR    $a1,$a1,`$Sigma0[2]-$Sigma0[1]`
143         xor     $t0,$t0,$t1
144         and     $t1,$b,$c
145         xor     $a0,$a0,$a1             ; Sigma0(a)
146         add     $d,$d,$T
147         xor     $t0,$t0,$t1             ; Maj(a,b,c)
148         add     $h,$T,$a0
149         add     $h,$h,$t0
150
151 ___
152 }
153
154 sub ROUND_16_xx {
155 my ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_;
156 $i-=16;
157 $code.=<<___;
158         $ROR    $a0,@X[($i+1)%16],$sigma0[0]
159         $ROR    $a1,@X[($i+1)%16],$sigma0[1]
160         $ROR    $t0,@X[($i+14)%16],$sigma1[0]
161         $ROR    $t1,@X[($i+14)%16],$sigma1[1]
162         xor     $a0,$a0,$a1
163         $SHR    $a1,@X[($i+1)%16],$sigma0[2]
164         xor     $t0,$t0,$t1
165         $SHR    $t1,@X[($i+14)%16],$sigma1[2]
166         add     @X[$i],@X[$i],@X[($i+9)%16]
167         xor     $a0,$a0,$a1             ; sigma0(X[(i+1)&0x0f])
168         xor     $t0,$t0,$t1             ; sigma1(X[(i+14)&0x0f])
169         add     @X[$i],@X[$i],$a0
170         add     @X[$i],@X[$i],$t0
171 ___
172 &ROUND_00_15($i,$a,$b,$c,$d,$e,$f,$g,$h);
173 }
174
175 $code=<<___;
176 .text
177
178 .globl  $func
179 .align  6
180 $func:
181         mflr    r0
182         $STU    $sp,`-($FRAME+16*$SZ)`($sp)
183         $SHL    $num,$num,`log(16*$SZ)/log(2)`
184
185         $PUSH   $ctx,`$FRAME-$SIZE_T*22`($sp)
186
187         $PUSH   r0,`$FRAME-$SIZE_T*21`($sp)
188         $PUSH   $toc,`$FRAME-$SIZE_T*20`($sp)
189         $PUSH   r13,`$FRAME-$SIZE_T*19`($sp)
190         $PUSH   r14,`$FRAME-$SIZE_T*18`($sp)
191         $PUSH   r15,`$FRAME-$SIZE_T*17`($sp)
192         $PUSH   r16,`$FRAME-$SIZE_T*16`($sp)
193         $PUSH   r17,`$FRAME-$SIZE_T*15`($sp)
194         $PUSH   r18,`$FRAME-$SIZE_T*14`($sp)
195         $PUSH   r19,`$FRAME-$SIZE_T*13`($sp)
196         $PUSH   r20,`$FRAME-$SIZE_T*12`($sp)
197         $PUSH   r21,`$FRAME-$SIZE_T*11`($sp)
198         $PUSH   r22,`$FRAME-$SIZE_T*10`($sp)
199         $PUSH   r23,`$FRAME-$SIZE_T*9`($sp)
200         $PUSH   r24,`$FRAME-$SIZE_T*8`($sp)
201         $PUSH   r25,`$FRAME-$SIZE_T*7`($sp)
202         $PUSH   r26,`$FRAME-$SIZE_T*6`($sp)
203         $PUSH   r27,`$FRAME-$SIZE_T*5`($sp)
204         $PUSH   r28,`$FRAME-$SIZE_T*4`($sp)
205         $PUSH   r29,`$FRAME-$SIZE_T*3`($sp)
206         $PUSH   r30,`$FRAME-$SIZE_T*2`($sp)
207         $PUSH   r31,`$FRAME-$SIZE_T*1`($sp)
208
209         $LD     $A,`0*$SZ`($ctx)
210         mr      $inp,r4                         ; incarnate $inp
211         $LD     $B,`1*$SZ`($ctx)
212         $LD     $C,`2*$SZ`($ctx)
213         $LD     $D,`3*$SZ`($ctx)
214         $LD     $E,`4*$SZ`($ctx)
215         $LD     $F,`5*$SZ`($ctx)
216         $LD     $G,`6*$SZ`($ctx)
217         $LD     $H,`7*$SZ`($ctx)
218
219         b       LPICmeup
220 LPICedup:
221         andi.   r0,$inp,3
222         bne     Lunaligned
223 Laligned:
224         add     $num,$inp,$num
225         $PUSH   $num,`$FRAME-$SIZE_T*24`($sp)   ; end pointer
226         $PUSH   $inp,`$FRAME-$SIZE_T*23`($sp)   ; inp pointer
227         bl      Lsha2_block_private
228 Ldone:
229         $POP    r0,`$FRAME-$SIZE_T*21`($sp)
230         $POP    $toc,`$FRAME-$SIZE_T*20`($sp)
231         $POP    r13,`$FRAME-$SIZE_T*19`($sp)
232         $POP    r14,`$FRAME-$SIZE_T*18`($sp)
233         $POP    r15,`$FRAME-$SIZE_T*17`($sp)
234         $POP    r16,`$FRAME-$SIZE_T*16`($sp)
235         $POP    r17,`$FRAME-$SIZE_T*15`($sp)
236         $POP    r18,`$FRAME-$SIZE_T*14`($sp)
237         $POP    r19,`$FRAME-$SIZE_T*13`($sp)
238         $POP    r20,`$FRAME-$SIZE_T*12`($sp)
239         $POP    r21,`$FRAME-$SIZE_T*11`($sp)
240         $POP    r22,`$FRAME-$SIZE_T*10`($sp)
241         $POP    r23,`$FRAME-$SIZE_T*9`($sp)
242         $POP    r24,`$FRAME-$SIZE_T*8`($sp)
243         $POP    r25,`$FRAME-$SIZE_T*7`($sp)
244         $POP    r26,`$FRAME-$SIZE_T*6`($sp)
245         $POP    r27,`$FRAME-$SIZE_T*5`($sp)
246         $POP    r28,`$FRAME-$SIZE_T*4`($sp)
247         $POP    r29,`$FRAME-$SIZE_T*3`($sp)
248         $POP    r30,`$FRAME-$SIZE_T*2`($sp)
249         $POP    r31,`$FRAME-$SIZE_T*1`($sp)
250         mtlr    r0
251         addi    $sp,$sp,`$FRAME+16*$SZ`
252         blr
253 ___
254
255 # PowerPC specification allows an implementation to be ill-behaved
256 # upon unaligned access which crosses page boundary. "Better safe
257 # than sorry" principle makes me treat it specially. But I don't
258 # look for particular offending word, but rather for the input
259 # block which crosses the boundary. Once found that block is aligned
260 # and hashed separately...
261 $code.=<<___;
262 .align  4
263 Lunaligned:
264         subfic  $t1,$inp,4096
265         andi.   $t1,$t1,`4096-16*$SZ`   ; distance to closest page boundary
266         beq     Lcross_page
267         $UCMP   $num,$t1
268         ble-    Laligned                ; didn't cross the page boundary
269         subfc   $num,$t1,$num
270         add     $t1,$inp,$t1
271         $PUSH   $num,`$FRAME-$SIZE_T*25`($sp)   ; save real remaining num
272         $PUSH   $t1,`$FRAME-$SIZE_T*24`($sp)    ; intermediate end pointer
273         $PUSH   $inp,`$FRAME-$SIZE_T*23`($sp)   ; inp pointer
274         bl      Lsha2_block_private
275         ; $inp equals to the intermediate end pointer here
276         $POP    $num,`$FRAME-$SIZE_T*25`($sp)   ; restore real remaining num
277 Lcross_page:
278         li      $t1,`16*$SZ/4`
279         mtctr   $t1
280         addi    r20,$sp,$FRAME                  ; aligned spot below the frame
281 Lmemcpy:
282         lbz     r16,0($inp)
283         lbz     r17,1($inp)
284         lbz     r18,2($inp)
285         lbz     r19,3($inp)
286         addi    $inp,$inp,4
287         stb     r16,0(r20)
288         stb     r17,1(r20)
289         stb     r18,2(r20)
290         stb     r19,3(r20)
291         addi    r20,r20,4
292         bdnz    Lmemcpy
293
294         $PUSH   $inp,`$FRAME-$SIZE_T*26`($sp)   ; save real inp
295         addi    $t1,$sp,`$FRAME+16*$SZ`         ; fictitious end pointer
296         addi    $inp,$sp,$FRAME                 ; fictitious inp pointer
297         $PUSH   $num,`$FRAME-$SIZE_T*25`($sp)   ; save real num
298         $PUSH   $t1,`$FRAME-$SIZE_T*24`($sp)    ; end pointer
299         $PUSH   $inp,`$FRAME-$SIZE_T*23`($sp)   ; inp pointer
300         bl      Lsha2_block_private
301         $POP    $inp,`$FRAME-$SIZE_T*26`($sp)   ; restore real inp
302         $POP    $num,`$FRAME-$SIZE_T*25`($sp)   ; restore real num
303         addic.  $num,$num,`-16*$SZ`             ; num--
304         bne-    Lunaligned
305         b       Ldone
306 ___
307
308 $code.=<<___;
309 .align  4
310 Lsha2_block_private:
311 ___
312 for($i=0;$i<16;$i++) {
313 $code.=<<___ if ($SZ==4);
314         lwz     @X[$i],`$i*$SZ`($inp)
315 ___
316 # 64-bit loads are split to 2x32-bit ones, as CPU can't handle
317 # unaligned 64-bit loads, only 32-bit ones...
318 $code.=<<___ if ($SZ==8);
319         lwz     $t0,`$i*$SZ`($inp)
320         lwz     @X[$i],`$i*$SZ+4`($inp)
321         insrdi  @X[$i],$t0,32,0
322 ___
323         &ROUND_00_15($i,@V);
324         unshift(@V,pop(@V));
325 }
326 $code.=<<___;
327         li      $T,`$rounds/16-1`
328         mtctr   $T
329 .align  4
330 Lrounds:
331         addi    $Tbl,$Tbl,`16*$SZ`
332 ___
333 for(;$i<32;$i++) {
334         &ROUND_16_xx($i,@V);
335         unshift(@V,pop(@V));
336 }
337 $code.=<<___;
338         bdnz-   Lrounds
339
340         $POP    $ctx,`$FRAME-$SIZE_T*22`($sp)
341         $POP    $inp,`$FRAME-$SIZE_T*23`($sp)   ; inp pointer
342         $POP    $num,`$FRAME-$SIZE_T*24`($sp)   ; end pointer
343         subi    $Tbl,$Tbl,`($rounds-16)*$SZ`    ; rewind Tbl
344
345         $LD     r16,`0*$SZ`($ctx)
346         $LD     r17,`1*$SZ`($ctx)
347         $LD     r18,`2*$SZ`($ctx)
348         $LD     r19,`3*$SZ`($ctx)
349         $LD     r20,`4*$SZ`($ctx)
350         $LD     r21,`5*$SZ`($ctx)
351         $LD     r22,`6*$SZ`($ctx)
352         addi    $inp,$inp,`16*$SZ`              ; advance inp
353         $LD     r23,`7*$SZ`($ctx)
354         add     $A,$A,r16
355         add     $B,$B,r17
356         $PUSH   $inp,`$FRAME-$SIZE_T*23`($sp)
357         add     $C,$C,r18
358         $ST     $A,`0*$SZ`($ctx)
359         add     $D,$D,r19
360         $ST     $B,`1*$SZ`($ctx)
361         add     $E,$E,r20
362         $ST     $C,`2*$SZ`($ctx)
363         add     $F,$F,r21
364         $ST     $D,`3*$SZ`($ctx)
365         add     $G,$G,r22
366         $ST     $E,`4*$SZ`($ctx)
367         add     $H,$H,r23
368         $ST     $F,`5*$SZ`($ctx)
369         $ST     $G,`6*$SZ`($ctx)
370         $UCMP   $inp,$num
371         $ST     $H,`7*$SZ`($ctx)
372         bne     Lsha2_block_private
373         blr
374 ___
375
376 # Ugly hack here, because PPC assembler syntax seem to vary too
377 # much from platforms to platform...
378 $code.=<<___;
379 .align  6
380 LPICmeup:
381         bl      LPIC
382         addi    $Tbl,$Tbl,`64-4`        ; "distance" between . and last nop
383         b       LPICedup
384         nop
385         nop
386         nop
387         nop
388         nop
389 LPIC:   mflr    $Tbl
390         blr
391         nop
392         nop
393         nop
394         nop
395         nop
396         nop
397 ___
398 $code.=<<___ if ($SZ==8);
399         .long   0x428a2f98,0xd728ae22,0x71374491,0x23ef65cd
400         .long   0xb5c0fbcf,0xec4d3b2f,0xe9b5dba5,0x8189dbbc
401         .long   0x3956c25b,0xf348b538,0x59f111f1,0xb605d019
402         .long   0x923f82a4,0xaf194f9b,0xab1c5ed5,0xda6d8118
403         .long   0xd807aa98,0xa3030242,0x12835b01,0x45706fbe
404         .long   0x243185be,0x4ee4b28c,0x550c7dc3,0xd5ffb4e2
405         .long   0x72be5d74,0xf27b896f,0x80deb1fe,0x3b1696b1
406         .long   0x9bdc06a7,0x25c71235,0xc19bf174,0xcf692694
407         .long   0xe49b69c1,0x9ef14ad2,0xefbe4786,0x384f25e3
408         .long   0x0fc19dc6,0x8b8cd5b5,0x240ca1cc,0x77ac9c65
409         .long   0x2de92c6f,0x592b0275,0x4a7484aa,0x6ea6e483
410         .long   0x5cb0a9dc,0xbd41fbd4,0x76f988da,0x831153b5
411         .long   0x983e5152,0xee66dfab,0xa831c66d,0x2db43210
412         .long   0xb00327c8,0x98fb213f,0xbf597fc7,0xbeef0ee4
413         .long   0xc6e00bf3,0x3da88fc2,0xd5a79147,0x930aa725
414         .long   0x06ca6351,0xe003826f,0x14292967,0x0a0e6e70
415         .long   0x27b70a85,0x46d22ffc,0x2e1b2138,0x5c26c926
416         .long   0x4d2c6dfc,0x5ac42aed,0x53380d13,0x9d95b3df
417         .long   0x650a7354,0x8baf63de,0x766a0abb,0x3c77b2a8
418         .long   0x81c2c92e,0x47edaee6,0x92722c85,0x1482353b
419         .long   0xa2bfe8a1,0x4cf10364,0xa81a664b,0xbc423001
420         .long   0xc24b8b70,0xd0f89791,0xc76c51a3,0x0654be30
421         .long   0xd192e819,0xd6ef5218,0xd6990624,0x5565a910
422         .long   0xf40e3585,0x5771202a,0x106aa070,0x32bbd1b8
423         .long   0x19a4c116,0xb8d2d0c8,0x1e376c08,0x5141ab53
424         .long   0x2748774c,0xdf8eeb99,0x34b0bcb5,0xe19b48a8
425         .long   0x391c0cb3,0xc5c95a63,0x4ed8aa4a,0xe3418acb
426         .long   0x5b9cca4f,0x7763e373,0x682e6ff3,0xd6b2b8a3
427         .long   0x748f82ee,0x5defb2fc,0x78a5636f,0x43172f60
428         .long   0x84c87814,0xa1f0ab72,0x8cc70208,0x1a6439ec
429         .long   0x90befffa,0x23631e28,0xa4506ceb,0xde82bde9
430         .long   0xbef9a3f7,0xb2c67915,0xc67178f2,0xe372532b
431         .long   0xca273ece,0xea26619c,0xd186b8c7,0x21c0c207
432         .long   0xeada7dd6,0xcde0eb1e,0xf57d4f7f,0xee6ed178
433         .long   0x06f067aa,0x72176fba,0x0a637dc5,0xa2c898a6
434         .long   0x113f9804,0xbef90dae,0x1b710b35,0x131c471b
435         .long   0x28db77f5,0x23047d84,0x32caab7b,0x40c72493
436         .long   0x3c9ebe0a,0x15c9bebc,0x431d67c4,0x9c100d4c
437         .long   0x4cc5d4be,0xcb3e42b6,0x597f299c,0xfc657e2a
438         .long   0x5fcb6fab,0x3ad6faec,0x6c44198c,0x4a475817
439 ___
440 $code.=<<___ if ($SZ==4);
441         .long   0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5
442         .long   0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5
443         .long   0xd807aa98,0x12835b01,0x243185be,0x550c7dc3
444         .long   0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174
445         .long   0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc
446         .long   0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da
447         .long   0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7
448         .long   0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967
449         .long   0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13
450         .long   0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85
451         .long   0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3
452         .long   0xd192e819,0xd6990624,0xf40e3585,0x106aa070
453         .long   0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5
454         .long   0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3
455         .long   0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208
456         .long   0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2
457 ___
458
459 $code =~ s/\`([^\`]*)\`/eval $1/gem;
460 print $code;
461 close STDOUT;