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. The module is big-endian [which is
13 # not big deal as there're no little-endian targets left around].
16 # -m64 -m32 | -m64 -m32
17 # --------------------------------------+-----------------------
18 # PPC970,gcc-4.0.0 +50% +38% | +40% +410%(*)
19 # Power6,xlc-7 +150% +90% | +100% +430%(*)
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:-(
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.
40 if ($output =~ /64/) {
47 } elsif ($output =~ /32/) {
54 } else { die "nonsense $output"; }
56 ( defined shift || open STDOUT,"| $^X ../perlasm/ppc-xlate.pl $output" ) ||
57 die "can't call ../perlasm/ppc-xlate.pl: $!";
59 if ($output =~ /512/) {
60 $func="sha512_block_data_order";
72 $func="sha256_block_data_order";
89 $ctx="r3"; # zapped by $a0
90 $inp="r4"; # zapped by $a1
91 $num="r5"; # zapped by $t0
105 $F ="r13"; $F="r2" if ($SIZE_T==8);# reassigned to exempt TLS pointer
109 @V=($A,$B,$C,$D,$E,$F,$G,$H);
110 @X=("r16","r17","r18","r19","r20","r21","r22","r23",
111 "r24","r25","r26","r27","r28","r29","r30","r31");
113 $inp="r31"; # reassigned $inp! aliases with @X[15]
116 my ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_;
118 $LD $T,`$i*$SZ`($Tbl)
119 $ROR $a0,$e,$Sigma1[0]
120 $ROR $a1,$e,$Sigma1[1]
125 $ROR $a1,$a1,`$Sigma1[2]-$Sigma1[1]`
126 or $t0,$t0,$t1 ; Ch(e,f,g)
128 xor $a0,$a0,$a1 ; Sigma1(e)
132 $ROR $a0,$a,$Sigma0[0]
133 $ROR $a1,$a,$Sigma0[1]
137 $ROR $a1,$a1,`$Sigma0[2]-$Sigma0[1]`
140 xor $a0,$a0,$a1 ; Sigma0(a)
142 xor $t0,$t0,$t1 ; Maj(a,b,c)
150 my ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_;
153 $ROR $a0,@X[($i+1)%16],$sigma0[0]
154 $ROR $a1,@X[($i+1)%16],$sigma0[1]
155 $ROR $t0,@X[($i+14)%16],$sigma1[0]
156 $ROR $t1,@X[($i+14)%16],$sigma1[1]
158 $SHR $a1,@X[($i+1)%16],$sigma0[2]
160 $SHR $t1,@X[($i+14)%16],$sigma1[2]
161 add @X[$i],@X[$i],@X[($i+9)%16]
162 xor $a0,$a0,$a1 ; sigma0(X[(i+1)&0x0f])
163 xor $t0,$t0,$t1 ; sigma1(X[(i+14)&0x0f])
164 add @X[$i],@X[$i],$a0
165 add @X[$i],@X[$i],$t0
167 &ROUND_00_15($i,$a,$b,$c,$d,$e,$f,$g,$h);
177 $STU $sp,`-($FRAME+16*$SZ)`($sp)
178 $SHL $num,$num,`log(16*$SZ)/log(2)`
180 $PUSH $ctx,`$FRAME-$SIZE_T*22`($sp)
182 $PUSH r0,`$FRAME-$SIZE_T*21`($sp)
183 $PUSH $toc,`$FRAME-$SIZE_T*20`($sp)
184 $PUSH r13,`$FRAME-$SIZE_T*19`($sp)
185 $PUSH r14,`$FRAME-$SIZE_T*18`($sp)
186 $PUSH r15,`$FRAME-$SIZE_T*17`($sp)
187 $PUSH r16,`$FRAME-$SIZE_T*16`($sp)
188 $PUSH r17,`$FRAME-$SIZE_T*15`($sp)
189 $PUSH r18,`$FRAME-$SIZE_T*14`($sp)
190 $PUSH r19,`$FRAME-$SIZE_T*13`($sp)
191 $PUSH r20,`$FRAME-$SIZE_T*12`($sp)
192 $PUSH r21,`$FRAME-$SIZE_T*11`($sp)
193 $PUSH r22,`$FRAME-$SIZE_T*10`($sp)
194 $PUSH r23,`$FRAME-$SIZE_T*9`($sp)
195 $PUSH r24,`$FRAME-$SIZE_T*8`($sp)
196 $PUSH r25,`$FRAME-$SIZE_T*7`($sp)
197 $PUSH r26,`$FRAME-$SIZE_T*6`($sp)
198 $PUSH r27,`$FRAME-$SIZE_T*5`($sp)
199 $PUSH r28,`$FRAME-$SIZE_T*4`($sp)
200 $PUSH r29,`$FRAME-$SIZE_T*3`($sp)
201 $PUSH r30,`$FRAME-$SIZE_T*2`($sp)
202 $PUSH r31,`$FRAME-$SIZE_T*1`($sp)
205 mr $inp,r4 ; incarnate $inp
220 $PUSH $num,`$FRAME-$SIZE_T*24`($sp) ; end pointer
221 $PUSH $inp,`$FRAME-$SIZE_T*23`($sp) ; inp pointer
222 bl Lsha2_block_private
224 $POP r0,`$FRAME-$SIZE_T*21`($sp)
225 $POP $toc,`$FRAME-$SIZE_T*20`($sp)
226 $POP r13,`$FRAME-$SIZE_T*19`($sp)
227 $POP r14,`$FRAME-$SIZE_T*18`($sp)
228 $POP r15,`$FRAME-$SIZE_T*17`($sp)
229 $POP r16,`$FRAME-$SIZE_T*16`($sp)
230 $POP r17,`$FRAME-$SIZE_T*15`($sp)
231 $POP r18,`$FRAME-$SIZE_T*14`($sp)
232 $POP r19,`$FRAME-$SIZE_T*13`($sp)
233 $POP r20,`$FRAME-$SIZE_T*12`($sp)
234 $POP r21,`$FRAME-$SIZE_T*11`($sp)
235 $POP r22,`$FRAME-$SIZE_T*10`($sp)
236 $POP r23,`$FRAME-$SIZE_T*9`($sp)
237 $POP r24,`$FRAME-$SIZE_T*8`($sp)
238 $POP r25,`$FRAME-$SIZE_T*7`($sp)
239 $POP r26,`$FRAME-$SIZE_T*6`($sp)
240 $POP r27,`$FRAME-$SIZE_T*5`($sp)
241 $POP r28,`$FRAME-$SIZE_T*4`($sp)
242 $POP r29,`$FRAME-$SIZE_T*3`($sp)
243 $POP r30,`$FRAME-$SIZE_T*2`($sp)
244 $POP r31,`$FRAME-$SIZE_T*1`($sp)
246 addi $sp,$sp,`$FRAME+16*$SZ`
250 # PowerPC specification allows an implementation to be ill-behaved
251 # upon unaligned access which crosses page boundary. "Better safe
252 # than sorry" principle makes me treat it specially. But I don't
253 # look for particular offending word, but rather for the input
254 # block which crosses the boundary. Once found that block is aligned
255 # and hashed separately...
260 andi. $t1,$t1,`4096-16*$SZ` ; distance to closest page boundary
263 ble- Laligned ; didn't cross the page boundary
266 $PUSH $num,`$FRAME-$SIZE_T*25`($sp) ; save real remaining num
267 $PUSH $t1,`$FRAME-$SIZE_T*24`($sp) ; intermediate end pointer
268 $PUSH $inp,`$FRAME-$SIZE_T*23`($sp) ; inp pointer
269 bl Lsha2_block_private
270 ; $inp equals to the intermediate end pointer here
271 $POP $num,`$FRAME-$SIZE_T*25`($sp) ; restore real remaining num
275 addi r20,$sp,$FRAME ; aligned spot below the frame
289 $PUSH $inp,`$FRAME-$SIZE_T*26`($sp) ; save real inp
290 addi $t1,$sp,`$FRAME+16*$SZ` ; fictitious end pointer
291 addi $inp,$sp,$FRAME ; fictitious inp pointer
292 $PUSH $num,`$FRAME-$SIZE_T*25`($sp) ; save real num
293 $PUSH $t1,`$FRAME-$SIZE_T*24`($sp) ; end pointer
294 $PUSH $inp,`$FRAME-$SIZE_T*23`($sp) ; inp pointer
295 bl Lsha2_block_private
296 $POP $inp,`$FRAME-$SIZE_T*26`($sp) ; restore real inp
297 $POP $num,`$FRAME-$SIZE_T*25`($sp) ; restore real num
298 addic. $num,$num,`-16*$SZ` ; num--
307 for($i=0;$i<16;$i++) {
308 $code.=<<___ if ($SZ==4);
309 lwz @X[$i],`$i*$SZ`($inp)
311 # 64-bit loads are split to 2x32-bit ones, as CPU can't handle
312 # unaligned 64-bit loads, only 32-bit ones...
313 $code.=<<___ if ($SZ==8);
314 lwz $t0,`$i*$SZ`($inp)
315 lwz @X[$i],`$i*$SZ+4`($inp)
316 insrdi @X[$i],$t0,32,0
326 addi $Tbl,$Tbl,`16*$SZ`
335 $POP $ctx,`$FRAME-$SIZE_T*22`($sp)
336 $POP $inp,`$FRAME-$SIZE_T*23`($sp) ; inp pointer
337 $POP $num,`$FRAME-$SIZE_T*24`($sp) ; end pointer
338 subi $Tbl,$Tbl,`($rounds-16)*$SZ` ; rewind Tbl
340 $LD r16,`0*$SZ`($ctx)
341 $LD r17,`1*$SZ`($ctx)
342 $LD r18,`2*$SZ`($ctx)
343 $LD r19,`3*$SZ`($ctx)
344 $LD r20,`4*$SZ`($ctx)
345 $LD r21,`5*$SZ`($ctx)
346 $LD r22,`6*$SZ`($ctx)
347 addi $inp,$inp,`16*$SZ` ; advance inp
348 $LD r23,`7*$SZ`($ctx)
351 $PUSH $inp,`$FRAME-$SIZE_T*23`($sp)
367 bne Lsha2_block_private
371 # Ugly hack here, because PPC assembler syntax seem to vary too
372 # much from platforms to platform...
377 addi $Tbl,$Tbl,`64-4` ; "distance" between . and last nop
393 $code.=<<___ if ($SZ==8);
394 .long 0x428a2f98,0xd728ae22,0x71374491,0x23ef65cd
395 .long 0xb5c0fbcf,0xec4d3b2f,0xe9b5dba5,0x8189dbbc
396 .long 0x3956c25b,0xf348b538,0x59f111f1,0xb605d019
397 .long 0x923f82a4,0xaf194f9b,0xab1c5ed5,0xda6d8118
398 .long 0xd807aa98,0xa3030242,0x12835b01,0x45706fbe
399 .long 0x243185be,0x4ee4b28c,0x550c7dc3,0xd5ffb4e2
400 .long 0x72be5d74,0xf27b896f,0x80deb1fe,0x3b1696b1
401 .long 0x9bdc06a7,0x25c71235,0xc19bf174,0xcf692694
402 .long 0xe49b69c1,0x9ef14ad2,0xefbe4786,0x384f25e3
403 .long 0x0fc19dc6,0x8b8cd5b5,0x240ca1cc,0x77ac9c65
404 .long 0x2de92c6f,0x592b0275,0x4a7484aa,0x6ea6e483
405 .long 0x5cb0a9dc,0xbd41fbd4,0x76f988da,0x831153b5
406 .long 0x983e5152,0xee66dfab,0xa831c66d,0x2db43210
407 .long 0xb00327c8,0x98fb213f,0xbf597fc7,0xbeef0ee4
408 .long 0xc6e00bf3,0x3da88fc2,0xd5a79147,0x930aa725
409 .long 0x06ca6351,0xe003826f,0x14292967,0x0a0e6e70
410 .long 0x27b70a85,0x46d22ffc,0x2e1b2138,0x5c26c926
411 .long 0x4d2c6dfc,0x5ac42aed,0x53380d13,0x9d95b3df
412 .long 0x650a7354,0x8baf63de,0x766a0abb,0x3c77b2a8
413 .long 0x81c2c92e,0x47edaee6,0x92722c85,0x1482353b
414 .long 0xa2bfe8a1,0x4cf10364,0xa81a664b,0xbc423001
415 .long 0xc24b8b70,0xd0f89791,0xc76c51a3,0x0654be30
416 .long 0xd192e819,0xd6ef5218,0xd6990624,0x5565a910
417 .long 0xf40e3585,0x5771202a,0x106aa070,0x32bbd1b8
418 .long 0x19a4c116,0xb8d2d0c8,0x1e376c08,0x5141ab53
419 .long 0x2748774c,0xdf8eeb99,0x34b0bcb5,0xe19b48a8
420 .long 0x391c0cb3,0xc5c95a63,0x4ed8aa4a,0xe3418acb
421 .long 0x5b9cca4f,0x7763e373,0x682e6ff3,0xd6b2b8a3
422 .long 0x748f82ee,0x5defb2fc,0x78a5636f,0x43172f60
423 .long 0x84c87814,0xa1f0ab72,0x8cc70208,0x1a6439ec
424 .long 0x90befffa,0x23631e28,0xa4506ceb,0xde82bde9
425 .long 0xbef9a3f7,0xb2c67915,0xc67178f2,0xe372532b
426 .long 0xca273ece,0xea26619c,0xd186b8c7,0x21c0c207
427 .long 0xeada7dd6,0xcde0eb1e,0xf57d4f7f,0xee6ed178
428 .long 0x06f067aa,0x72176fba,0x0a637dc5,0xa2c898a6
429 .long 0x113f9804,0xbef90dae,0x1b710b35,0x131c471b
430 .long 0x28db77f5,0x23047d84,0x32caab7b,0x40c72493
431 .long 0x3c9ebe0a,0x15c9bebc,0x431d67c4,0x9c100d4c
432 .long 0x4cc5d4be,0xcb3e42b6,0x597f299c,0xfc657e2a
433 .long 0x5fcb6fab,0x3ad6faec,0x6c44198c,0x4a475817
435 $code.=<<___ if ($SZ==4);
436 .long 0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5
437 .long 0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5
438 .long 0xd807aa98,0x12835b01,0x243185be,0x550c7dc3
439 .long 0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174
440 .long 0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc
441 .long 0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da
442 .long 0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7
443 .long 0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967
444 .long 0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13
445 .long 0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85
446 .long 0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3
447 .long 0xd192e819,0xd6990624,0xf40e3585,0x106aa070
448 .long 0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5
449 .long 0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3
450 .long 0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208
451 .long 0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2
454 $code =~ s/\`([^\`]*)\`/eval $1/gem;