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 1.2\"
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"; }
29 for (@ARGV) { $big_endian=1 if (/\-DB_ENDIAN/);
30 $big_endian=0 if (/\-DL_ENDIAN/); }
31 if (!defined($big_endian))
32 { $big_endian=(unpack('L',pack('N',1))==1); }
35 if ($human) { # useful for visual code auditing...
36 ($A,$B,$C,$D,$E,$T) = ("A","B","C","D","E","T");
37 ($h0,$h1,$h2,$h3,$h4) = ("h0","h1","h2","h3","h4");
38 ($K_00_19, $K_20_39, $K_40_59, $K_60_79) =
39 ( "K_00_19","K_20_39","K_40_59","K_60_79" );
40 @X= ( "X0", "X1", "X2", "X3", "X4", "X5", "X6", "X7",
41 "X8", "X9","X10","X11","X12","X13","X14","X15" );
44 ($A,$B,$C,$D,$E,$T) = ("loc0","loc1","loc2","loc3","loc4","loc5");
45 ($h0,$h1,$h2,$h3,$h4) = ("loc6","loc7","loc8","loc9","loc10");
46 ($K_00_19, $K_20_39, $K_40_59, $K_60_79) =
47 ( "r14", "r15", "loc11", "loc12" );
48 @X= ( "r16", "r17", "r18", "r19", "r20", "r21", "r22", "r23",
49 "r24", "r25", "r26", "r27", "r28", "r29", "r30", "r31" );
54 local ($i,$a,$b,$c,$d,$e,$f)=@_;
56 $code.=<<___ if ($i==0);
57 { .mmi; ld1 $X[$i&0xf]=[inp],2 // MSB
59 { .mmi; ld1 tmp0=[inp],2
60 ld1 tmp4=[tmp3],2 // LSB
61 dep $X[$i&0xf]=$X[$i&0xf],tmp2,8,8 };;
65 { .mmi; ld1 $X[($i+1)&0xf]=[inp],2 // +1
66 dep tmp1=tmp0,tmp4,8,8 };;
67 { .mmi; ld1 tmp2=[tmp3],2 // +1
69 dep $X[$i&0xf]=$X[$i&0xf],tmp1,16,16 } //;;
70 { .mmi; andcm tmp1=$d,$b
72 dep.z tmp5=$a,5,27 };; // a<<5
73 { .mmi; or tmp4=tmp4,tmp1 // F_00_19(b,c,d)=(b&c)|(~b&d)
74 add $f=tmp0,$X[$i&0xf] // f=xi+e+K_00_19
75 extr.u tmp1=$a,27,5 };; // a>>27
76 { .mmi; ld1 tmp0=[inp],2 // +1
77 add $f=$f,tmp4 // f+=F_00_19(b,c,d)
78 shrp $b=tmp6,tmp6,2 } // b=ROTATE(b,30)
79 { .mmi; ld1 tmp4=[tmp3],2 // +1
80 or tmp5=tmp1,tmp5 // ROTATE(a,5)
81 mux2 tmp6=$a,0x44 };; // see b in next iteration
82 { .mii; add $f=$f,tmp5 // f+=ROTATE(a,5)
83 dep $X[($i+1)&0xf]=$X[($i+1)&0xf],tmp2,8,8 // +1
84 mux2 $X[$i&0xf]=$X[$i&0xf],0x44 } //;;
90 { .mii; and tmp3=$c,$b
91 dep tmp1=tmp0,tmp4,8,8;;
92 dep $X[$i&0xf]=$X[$i&0xf],tmp1,16,16 } //;;
93 { .mmi; andcm tmp1=$d,$b
95 dep.z tmp5=$a,5,27 };; // a<<5
96 { .mmi; or tmp4=tmp3,tmp1 // F_00_19(b,c,d)=(b&c)|(~b&d)
97 add $f=tmp0,$X[$i&0xf] // f=xi+e+K_00_19
98 extr.u tmp1=$a,27,5 } // a>>27
99 { .mmi; xor tmp2=$X[($i+0+1)&0xf],$X[($i+2+1)&0xf] // +1
100 xor tmp3=$X[($i+8+1)&0xf],$X[($i+13+1)&0xf] // +1
102 { .mmi; add $f=$f,tmp4 // f+=F_00_19(b,c,d)
103 xor tmp2=tmp2,tmp3 // +1
104 shrp $b=tmp6,tmp6,2 } // b=ROTATE(b,30)
105 { .mmi; or tmp1=tmp1,tmp5 // ROTATE(a,5)
106 mux2 tmp6=$a,0x44 };; // see b in next iteration
107 { .mii; add $f=$f,tmp1 // f+=ROTATE(a,5)
108 shrp $e=tmp2,tmp2,31 // f+1=ROTATE(x[0]^x[2]^x[8]^x[13],1)
109 mux2 $X[$i&0xf]=$X[$i&0xf],0x44 };;
117 local ($i,$a,$b,$c,$d,$e,$f)=@_;
120 { .mmi; mov $X[$i&0xf]=$f // Xupdate
122 dep.z tmp5=$a,5,27 } // a<<5
123 { .mmi; andcm tmp1=$d,$b
124 add tmp4=$e,$K_00_19 };;
125 { .mmi; or tmp0=tmp0,tmp1 // F_00_19(b,c,d)=(b&c)|(~b&d)
126 add $f=$f,tmp4 // f+=e+K_00_19
127 extr.u tmp1=$a,27,5 } // a>>27
128 { .mmi; xor tmp2=$X[($i+0+1)&0xf],$X[($i+2+1)&0xf] // +1
129 xor tmp3=$X[($i+8+1)&0xf],$X[($i+13+1)&0xf] // +1
131 { .mmi; add $f=$f,tmp0 // f+=F_00_19(b,c,d)
132 xor tmp2=tmp2,tmp3 // +1
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 $f=$f,tmp1 // f+=ROTATE(a,5)
137 shrp $e=tmp2,tmp2,31 // f+1=ROTATE(x[0]^x[2]^x[8]^x[13],1)
145 local ($i,$a,$b,$c,$d,$e,$f,$Konst)=@_;
146 $Konst = $K_20_39 if (!defined($Konst));
150 { .mib; mov $X[$i&0xf]=$f // Xupdate
151 dep.z tmp5=$a,5,27 } // a<<5
152 { .mib; xor tmp0=$c,$b
153 add tmp4=$e,$Konst };;
154 { .mmi; xor tmp0=tmp0,$d // F_20_39(b,c,d)=b^c^d
155 add $f=$f,tmp4 // f+=e+K_20_39
156 extr.u tmp1=$a,27,5 } // a>>27
157 { .mmi; xor tmp2=$X[($i+0+1)&0xf],$X[($i+2+1)&0xf] // +1
158 xor tmp3=$X[($i+8+1)&0xf],$X[($i+13+1)&0xf] // +1
160 { .mmi; add $f=$f,tmp0 // f+=F_20_39(b,c,d)
161 xor tmp2=tmp2,tmp3 // +1
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 $f=$f,tmp1 // f+=ROTATE(a,5)
166 shrp $e=tmp2,tmp2,31 // f+1=ROTATE(x[0]^x[2]^x[8]^x[13],1)
173 { .mib; mov $X[$i&0xf]=$f // Xupdate
174 dep.z tmp5=$a,5,27 } // a<<5
175 { .mib; xor tmp0=$c,$b
176 add tmp4=$e,$Konst };;
177 { .mib; xor tmp0=tmp0,$d // F_20_39(b,c,d)=b^c^d
178 extr.u tmp1=$a,27,5 } // a>>27
179 { .mib; add $f=$f,tmp4 // f+=e+K_20_39
180 add $h1=$h1,$a };; // wrap up
181 { .mmi; add $f=$f,tmp0 // f+=F_20_39(b,c,d)
182 shrp $b=tmp6,tmp6,2 } // b=ROTATE(b,30) ;;?
183 { .mmi; or tmp1=tmp1,tmp5 // ROTATE(a,5)
184 add $h3=$h3,$c };; // wrap up
185 { .mib; add tmp3=1,inp // used in unaligned codepath
186 add $f=$f,tmp1 } // f+=ROTATE(a,5)
187 { .mib; add $h2=$h2,$b // wrap up
188 add $h4=$h4,$d };; // wrap up
196 local ($i,$a,$b,$c,$d,$e,$f)=@_;
199 { .mmi; mov $X[$i&0xf]=$f // Xupdate
201 dep.z tmp5=$a,5,27 } // a<<5
202 { .mmi; and tmp1=$d,$b
203 add tmp4=$e,$K_40_59 };;
204 { .mmi; or tmp0=tmp0,tmp1 // (b&c)|(b&d)
205 add $f=$f,tmp4 // f+=e+K_40_59
206 extr.u tmp1=$a,27,5 } // a>>27
207 { .mmi; and tmp4=$c,$d
208 xor tmp2=$X[($i+0+1)&0xf],$X[($i+2+1)&0xf] // +1
209 xor tmp3=$X[($i+8+1)&0xf],$X[($i+13+1)&0xf] // +1
211 { .mmi; or tmp1=tmp1,tmp5 // ROTATE(a,5)
212 xor tmp2=tmp2,tmp3 // +1
213 shrp $b=tmp6,tmp6,2 } // b=ROTATE(b,30)
214 { .mmi; or tmp0=tmp0,tmp4 // F_40_59(b,c,d)=(b&c)|(b&d)|(c&d)
215 mux2 tmp6=$a,0x44 };; // see b in next iteration
216 { .mii; add $f=$f,tmp0 // f+=F_40_59(b,c,d)
217 shrp $e=tmp2,tmp2,31;; // f+1=ROTATE(x[0]^x[2]^x[8]^x[13],1)
218 add $f=$f,tmp1 };; // f+=ROTATE(a,5)
222 sub BODY_60_79 { &BODY_20_39(@_,$K_60_79); }
234 // void sha1_block_data_order(SHA_CTX *c,const void *p,size_t num);
235 .global sha1_block_data_order#
236 .proc sha1_block_data_order#
238 sha1_block_data_order:
240 { .mmi; alloc tmp1=ar.pfs,3,15,0,0
244 { .mmi; $ADDP ctx=0,ctx
251 { .mlx; ld4 $h0=[ctx],8
252 movl $K_00_19=0x5a827999 }
253 { .mlx; ld4 $h1=[tmp0],8
254 movl $K_20_39=0x6ed9eba1 };;
255 { .mlx; ld4 $h2=[ctx],8
256 movl $K_40_59=0x8f1bbcdc }
257 { .mlx; ld4 $h3=[tmp0]
258 movl $K_60_79=0xca62c1d6 };;
259 { .mmi; ld4 $h4=[ctx],-16
260 add in2=-1,in2 // adjust num for ar.lc
264 mov ar.lc=in2 };; // brp.loop.imp: too far
276 { my $i,@V=($A,$B,$C,$D,$E,$T);
278 for($i=0;$i<16;$i++) { &BODY_00_15(\$code,$i,@V); unshift(@V,pop(@V)); }
279 for(;$i<20;$i++) { &BODY_16_19(\$code,$i,@V); unshift(@V,pop(@V)); }
280 for(;$i<40;$i++) { &BODY_20_39(\$code,$i,@V); unshift(@V,pop(@V)); }
281 for(;$i<60;$i++) { &BODY_40_59(\$code,$i,@V); unshift(@V,pop(@V)); }
282 for(;$i<80;$i++) { &BODY_60_79(\$code,$i,@V); unshift(@V,pop(@V)); }
284 (($V[5] eq $D) and ($V[0] eq $E)) or die; # double-check
288 { .mmb; add $h0=$h0,$E
290 br.ctop.dptk.many .Ldtop };;
292 { .mmi; add tmp0=4,ctx
294 { .mmi; st4 [ctx]=$h0,8
296 { .mmi; st4 [ctx]=$h2,8
298 { .mib; st4 [ctx]=$h4,-16
300 br.ret.sptk.many b0 };;
301 .endp sha1_block_data_order#
302 stringz "SHA1 block transform for IA64, CRYPTOGAMS by <appro\@openssl.org>"
305 $output=shift and open STDOUT,">$output";