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 # Eternal question is what's wrong with compiler generated code? The
11 # trick is that it's possible to reduce the number of shifts required
12 # to perform rotations by maintaining copy of 32-bit value in upper
13 # bits of 64-bit register. Just follow mux2 and shrp instructions...
14 # Performance under big-endian OS such as HP-UX is 179MBps*1GHz, which
15 # is >50% better than HP C and >2x better than gcc.
18 .ident \"sha1-ia64.s, version 1.3\"
19 .ident \"IA-64 ISA artwork by Andy Polyakov <appro\@fy.chalmers.se>\"
27 for (@ARGV) { $ADDP="add" if (/[\+DD|\-mlp]64/); }
28 } else { $ADDP="add"; }
31 if ($human) { # useful for visual code auditing...
32 ($A,$B,$C,$D,$E) = ("A","B","C","D","E");
33 ($h0,$h1,$h2,$h3,$h4) = ("h0","h1","h2","h3","h4");
34 ($K_00_19, $K_20_39, $K_40_59, $K_60_79) =
35 ( "K_00_19","K_20_39","K_40_59","K_60_79" );
36 @X= ( "X0", "X1", "X2", "X3", "X4", "X5", "X6", "X7",
37 "X8", "X9","X10","X11","X12","X13","X14","X15" );
40 ($A,$B,$C,$D,$E) = ("loc0","loc1","loc2","loc3","loc4");
41 ($h0,$h1,$h2,$h3,$h4) = ("loc5","loc6","loc7","loc8","loc9");
42 ($K_00_19, $K_20_39, $K_40_59, $K_60_79) =
43 ( "r14", "r15", "loc10", "loc11" );
44 @X= ( "r16", "r17", "r18", "r19", "r20", "r21", "r22", "r23",
45 "r24", "r25", "r26", "r27", "r28", "r29", "r30", "r31" );
50 my ($i,$a,$b,$c,$d,$e)=@_;
54 $code.=<<___ if ($i==0);
55 { .mmi; ld1 $X[$i]=[inp],2 // MSB
57 { .mmi; ld1 tmp0=[inp],2
58 ld1 tmp4=[tmp3],2 // LSB
59 dep $X[$i]=$X[$i],tmp2,8,8 };;
63 { .mmi; ld1 $Xn=[inp],2 // forward Xload
65 dep tmp1=tmp0,tmp4,8,8 };;
66 { .mmi; ld1 tmp2=[tmp3],2 // forward Xload
68 dep $X[$i]=$X[$i],tmp1,16,16} //;;
69 { .mmi; add $e=$e,$K_00_19 // e+=K_00_19
71 dep.z tmp5=$a,5,27 };; // a<<5
72 { .mmi; add $e=$e,$X[$i] // e+=Xload
73 or tmp4=tmp4,tmp1 // F_00_19(b,c,d)=(b&c)|(~b&d)
74 extr.u tmp1=$a,27,5 };; // a>>27
75 { .mmi; ld1 tmp0=[inp],2 // forward Xload
76 add $e=$e,tmp4 // e+=F_00_19(b,c,d)
77 shrp $b=tmp6,tmp6,2 } // b=ROTATE(b,30)
78 { .mmi; ld1 tmp4=[tmp3],2 // forward Xload
79 or tmp5=tmp1,tmp5 // ROTATE(a,5)
80 mux2 tmp6=$a,0x44 };; // see b in next iteration
81 { .mii; add $e=$e,tmp5 // e+=ROTATE(a,5)
82 dep $Xn=$Xn,tmp2,8,8 // forward Xload
83 mux2 $X[$i]=$X[$i],0x44 } //;;
89 { .mii; and tmp3=$c,$b
90 dep tmp1=tmp0,tmp4,8,8;;
91 dep $X[$i]=$X[$i],tmp1,16,16} //;;
92 { .mmi; add $e=$e,$K_00_19 // e+=K_00_19
94 dep.z tmp5=$a,5,27 };; // a<<5
95 { .mmi; add $e=$e,$X[$i] // e+=Xupdate
96 or tmp4=tmp3,tmp1 // F_00_19(b,c,d)=(b&c)|(~b&d)
97 extr.u tmp1=$a,27,5 } // a>>27
98 { .mmi; xor $Xn=$Xn,$X[($j+2)%16] // forward Xupdate
99 xor tmp3=$X[($j+8)%16],$X[($j+13)%16] // forward Xupdate
101 { .mmi; add $e=$e,tmp4 // e+=F_00_19(b,c,d)
102 xor $Xn=$Xn,tmp3 // forward Xupdate
103 shrp $b=tmp6,tmp6,2 } // b=ROTATE(b,30)
104 { .mmi; or tmp1=tmp1,tmp5 // ROTATE(a,5)
105 mux2 tmp6=$a,0x44 };; // see b in next iteration
106 { .mii; add $e=$e,tmp1 // e+=ROTATE(a,5)
107 shrp $Xn=$Xn,$Xn,31 // ROTATE(x[0]^x[2]^x[8]^x[13],1)
108 mux2 $X[$i]=$X[$i],0x44 };;
116 my ($i,$a,$b,$c,$d,$e)=@_;
121 { .mib; add $e=$e,$K_00_19 // e+=K_00_19
122 dep.z tmp5=$a,5,27 } // a<<5
123 { .mib; andcm tmp1=$d,$b
125 { .mmi; add $e=$e,$X[$i%16] // e+=Xupdate
126 or tmp0=tmp0,tmp1 // F_00_19(b,c,d)=(b&c)|(~b&d)
127 extr.u tmp1=$a,27,5 } // a>>27
128 { .mmi; xor $Xn=$Xn,$X[($j+2)%16] // forward Xupdate
129 xor tmp3=$X[($j+8)%16],$X[($j+13)%16] // forward Xupdate
131 { .mmi; add $e=$e,tmp0 // f+=F_00_19(b,c,d)
132 xor $Xn=$Xn,tmp3 // forward Xupdate
133 shrp $b=tmp6,tmp6,2 } // b=ROTATE(b,30)
134 { .mmi; or tmp1=tmp1,tmp5 // ROTATE(a,5)
135 mux2 tmp6=$a,0x44 };; // see b in next iteration
136 { .mii; add $e=$e,tmp1 // e+=ROTATE(a,5)
137 shrp $Xn=$Xn,$Xn,31 // ROTATE(x[0]^x[2]^x[8]^x[13],1)
145 my ($i,$a,$b,$c,$d,$e,$Konst)=@_;
146 $Konst = $K_20_39 if (!defined($Konst));
152 { .mib; add $e=$e,$Konst // e+=K_XX_XX
153 dep.z tmp5=$a,5,27 } // a<<5
154 { .mib; xor tmp0=$c,$b
155 xor $Xn=$Xn,$X[($j+2)%16] };; // forward Xupdate
156 { .mib; add $e=$e,$X[$i%16] // e+=Xupdate
157 extr.u tmp1=$a,27,5 } // a>>27
158 { .mib; xor tmp0=tmp0,$d // F_20_39(b,c,d)=b^c^d
159 xor $Xn=$Xn,$X[($j+8)%16] };; // forward Xupdate
160 { .mmi; add $e=$e,tmp0 // e+=F_20_39(b,c,d)
161 xor $Xn=$Xn,$X[($j+13)%16] // forward Xupdate
162 shrp $b=tmp6,tmp6,2 } // b=ROTATE(b,30)
163 { .mmi; or tmp1=tmp1,tmp5 // ROTATE(a,5)
164 mux2 tmp6=$a,0x44 };; // see b in next iteration
165 { .mii; add $e=$e,tmp1 // e+=ROTATE(a,5)
166 shrp $Xn=$Xn,$Xn,31 // ROTATE(x[0]^x[2]^x[8]^x[13],1)
173 { .mib; add $e=$e,$Konst // e+=K_60_79
174 dep.z tmp5=$a,5,27 } // a<<5
175 { .mib; xor tmp0=$c,$b
176 add $h1=$h1,$a };; // wrap up
177 { .mib; add $e=$e,$X[$i%16] // e+=Xupdate
178 extr.u tmp1=$a,27,5 } // a>>27
179 { .mib; xor tmp0=tmp0,$d // F_20_39(b,c,d)=b^c^d
180 add $h3=$h3,$c };; // wrap up
181 { .mmi; add $e=$e,tmp0 // e+=F_20_39(b,c,d)
182 or tmp1=tmp1,tmp5 // ROTATE(a,5)
183 shrp $b=tmp6,tmp6,2 };; // b=ROTATE(b,30) ;;?
184 { .mmi; add $e=$e,tmp1 // e+=ROTATE(a,5)
185 add tmp3=1,inp // used in unaligned codepath
186 add $h4=$h4,$d };; // wrap up
194 my ($i,$a,$b,$c,$d,$e)=@_;
199 { .mib; add $e=$e,$K_40_59 // e+=K_40_59
200 dep.z tmp5=$a,5,27 } // a<<5
201 { .mib; and tmp1=$c,$d
203 { .mmi; add $e=$e,$X[$i%16] // e+=Xupdate
204 add tmp5=tmp5,tmp1 // a<<5+(c&d)
205 extr.u tmp1=$a,27,5 } // a>>27
206 { .mmi; and tmp0=tmp0,$b
207 xor $Xn=$Xn,$X[($j+2)%16] // forward Xupdate
208 xor tmp3=$X[($j+8)%16],$X[($j+13)%16] };; // forward Xupdate
209 { .mmi; add $e=$e,tmp0 // e+=b&(c^d)
210 add tmp5=tmp5,tmp1 // ROTATE(a,5)+(c&d)
211 shrp $b=tmp6,tmp6,2 } // b=ROTATE(b,30)
212 { .mmi; xor $Xn=$Xn,tmp3
213 mux2 tmp6=$a,0x44 };; // see b in next iteration
214 { .mii; add $e=$e,tmp5 // e+=ROTATE(a,5)+(c&d)
215 shrp $Xn=$Xn,$Xn,31 // ROTATE(x[0]^x[2]^x[8]^x[13],1)
220 sub BODY_60_79 { &BODY_20_39(@_,$K_60_79); }
232 // void sha1_block_data_order(SHA_CTX *c,const void *p,size_t num);
233 .global sha1_block_data_order#
234 .proc sha1_block_data_order#
236 sha1_block_data_order:
238 { .mmi; alloc tmp1=ar.pfs,3,14,0,0
242 { .mmi; $ADDP ctx=0,ctx
249 { .mlx; ld4 $h0=[ctx],8
250 movl $K_00_19=0x5a827999 }
251 { .mlx; ld4 $h1=[tmp0],8
252 movl $K_20_39=0x6ed9eba1 };;
253 { .mlx; ld4 $h2=[ctx],8
254 movl $K_40_59=0x8f1bbcdc }
255 { .mlx; ld4 $h3=[tmp0]
256 movl $K_60_79=0xca62c1d6 };;
257 { .mmi; ld4 $h4=[ctx],-16
258 add in2=-1,in2 // adjust num for ar.lc
262 mov ar.lc=in2 };; // brp.loop.imp: too far
275 my @V=($A,$B,$C,$D,$E);
277 for($i=0;$i<16;$i++) { &BODY_00_15(\$code,$i,@V); unshift(@V,pop(@V)); }
278 for(;$i<20;$i++) { &BODY_16_19(\$code,$i,@V); unshift(@V,pop(@V)); }
279 for(;$i<40;$i++) { &BODY_20_39(\$code,$i,@V); unshift(@V,pop(@V)); }
280 for(;$i<60;$i++) { &BODY_40_59(\$code,$i,@V); unshift(@V,pop(@V)); }
281 for(;$i<80;$i++) { &BODY_60_79(\$code,$i,@V); unshift(@V,pop(@V)); }
283 (($V[0] eq $A) and ($V[4] eq $E)) or die; # double-check
287 { .mmb; add $h0=$h0,$A
289 br.ctop.dptk.many .Ldtop };;
291 { .mmi; add tmp0=4,ctx
293 { .mmi; st4 [ctx]=$h0,8
295 { .mmi; st4 [ctx]=$h2,8
297 { .mib; st4 [ctx]=$h4,-16
299 br.ret.sptk.many b0 };;
300 .endp sha1_block_data_order#
301 stringz "SHA1 block transform for IA64, CRYPTOGAMS by <appro\@openssl.org>"
304 $output=shift and open STDOUT,">$output";