| 1 | #! /usr/bin/env perl
|
|---|
| 2 | # Copyright 2009-2020 The OpenSSL Project Authors. All Rights Reserved.
|
|---|
| 3 | #
|
|---|
| 4 | # Licensed under the OpenSSL license (the "License"). You may not use
|
|---|
| 5 | # this file except in compliance with the License. You can obtain a copy
|
|---|
| 6 | # in the file LICENSE in the source distribution or at
|
|---|
| 7 | # https://www.openssl.org/source/license.html
|
|---|
| 8 |
|
|---|
| 9 |
|
|---|
| 10 | # ====================================================================
|
|---|
| 11 | # Written by Andy Polyakov <[email protected]> for the OpenSSL
|
|---|
| 12 | # project. The module is, however, dual licensed under OpenSSL and
|
|---|
| 13 | # CRYPTOGAMS licenses depending on where you obtain it. For further
|
|---|
| 14 | # details see http://www.openssl.org/~appro/cryptogams/.
|
|---|
| 15 | # ====================================================================
|
|---|
| 16 |
|
|---|
| 17 | # On PA-7100LC this module performs ~90-50% better, less for longer
|
|---|
| 18 | # keys, than code generated by gcc 3.2 for PA-RISC 1.1. Latter means
|
|---|
| 19 | # that compiler utilized xmpyu instruction to perform 32x32=64-bit
|
|---|
| 20 | # multiplication, which in turn means that "baseline" performance was
|
|---|
| 21 | # optimal in respect to instruction set capabilities. Fair comparison
|
|---|
| 22 | # with vendor compiler is problematic, because OpenSSL doesn't define
|
|---|
| 23 | # BN_LLONG [presumably] for historical reasons, which drives compiler
|
|---|
| 24 | # toward 4 times 16x16=32-bit multiplications [plus complementary
|
|---|
| 25 | # shifts and additions] instead. This means that you should observe
|
|---|
| 26 | # several times improvement over code generated by vendor compiler
|
|---|
| 27 | # for PA-RISC 1.1, but the "baseline" is far from optimal. The actual
|
|---|
| 28 | # improvement coefficient was never collected on PA-7100LC, or any
|
|---|
| 29 | # other 1.1 CPU, because I don't have access to such machine with
|
|---|
| 30 | # vendor compiler. But to give you a taste, PA-RISC 1.1 code path
|
|---|
| 31 | # reportedly outperformed code generated by cc +DA1.1 +O3 by factor
|
|---|
| 32 | # of ~5x on PA-8600.
|
|---|
| 33 | #
|
|---|
| 34 | # On PA-RISC 2.0 it has to compete with pa-risc2[W].s, which is
|
|---|
| 35 | # reportedly ~2x faster than vendor compiler generated code [according
|
|---|
| 36 | # to comment in pa-risc2[W].s]. Here comes a catch. Execution core of
|
|---|
| 37 | # this implementation is actually 32-bit one, in the sense that it
|
|---|
| 38 | # operates on 32-bit values. But pa-risc2[W].s operates on arrays of
|
|---|
| 39 | # 64-bit BN_LONGs... How do they interoperate then? No problem. This
|
|---|
| 40 | # module picks halves of 64-bit values in reverse order and pretends
|
|---|
| 41 | # they were 32-bit BN_LONGs. But can 32-bit core compete with "pure"
|
|---|
| 42 | # 64-bit code such as pa-risc2[W].s then? Well, the thing is that
|
|---|
| 43 | # 32x32=64-bit multiplication is the best even PA-RISC 2.0 can do,
|
|---|
| 44 | # i.e. there is no "wider" multiplication like on most other 64-bit
|
|---|
| 45 | # platforms. This means that even being effectively 32-bit, this
|
|---|
| 46 | # implementation performs "64-bit" computational task in same amount
|
|---|
| 47 | # of arithmetic operations, most notably multiplications. It requires
|
|---|
| 48 | # more memory references, most notably to tp[num], but this doesn't
|
|---|
| 49 | # seem to exhaust memory port capacity. And indeed, dedicated PA-RISC
|
|---|
| 50 | # 2.0 code path provides virtually same performance as pa-risc2[W].s:
|
|---|
| 51 | # it's ~10% better for shortest key length and ~10% worse for longest
|
|---|
| 52 | # one.
|
|---|
| 53 | #
|
|---|
| 54 | # In case it wasn't clear. The module has two distinct code paths:
|
|---|
| 55 | # PA-RISC 1.1 and PA-RISC 2.0 ones. Latter features carry-free 64-bit
|
|---|
| 56 | # additions and 64-bit integer loads, not to mention specific
|
|---|
| 57 | # instruction scheduling. In 64-bit build naturally only 2.0 code path
|
|---|
| 58 | # is assembled. In 32-bit application context both code paths are
|
|---|
| 59 | # assembled, PA-RISC 2.0 CPU is detected at run-time and proper path
|
|---|
| 60 | # is taken automatically. Also, in 32-bit build the module imposes
|
|---|
| 61 | # couple of limitations: vector lengths has to be even and vector
|
|---|
| 62 | # addresses has to be 64-bit aligned. Normally neither is a problem:
|
|---|
| 63 | # most common key lengths are even and vectors are commonly malloc-ed,
|
|---|
| 64 | # which ensures alignment.
|
|---|
| 65 | #
|
|---|
| 66 | # Special thanks to polarhome.com for providing HP-UX account on
|
|---|
| 67 | # PA-RISC 1.1 machine, and to correspondent who chose to remain
|
|---|
| 68 | # anonymous for testing the code on PA-RISC 2.0 machine.
|
|---|
| 69 | |
|---|
| 70 |
|
|---|
| 71 | $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
|
|---|
| 72 |
|
|---|
| 73 | $flavour = shift;
|
|---|
| 74 | $output = shift;
|
|---|
| 75 |
|
|---|
| 76 | open STDOUT,">$output";
|
|---|
| 77 |
|
|---|
| 78 | if ($flavour =~ /64/) {
|
|---|
| 79 | $LEVEL ="2.0W";
|
|---|
| 80 | $SIZE_T =8;
|
|---|
| 81 | $FRAME_MARKER =80;
|
|---|
| 82 | $SAVED_RP =16;
|
|---|
| 83 | $PUSH ="std";
|
|---|
| 84 | $PUSHMA ="std,ma";
|
|---|
| 85 | $POP ="ldd";
|
|---|
| 86 | $POPMB ="ldd,mb";
|
|---|
| 87 | $BN_SZ =$SIZE_T;
|
|---|
| 88 | } else {
|
|---|
| 89 | $LEVEL ="1.1"; #$LEVEL.="\n\t.ALLOW\t2.0";
|
|---|
| 90 | $SIZE_T =4;
|
|---|
| 91 | $FRAME_MARKER =48;
|
|---|
| 92 | $SAVED_RP =20;
|
|---|
| 93 | $PUSH ="stw";
|
|---|
| 94 | $PUSHMA ="stwm";
|
|---|
| 95 | $POP ="ldw";
|
|---|
| 96 | $POPMB ="ldwm";
|
|---|
| 97 | $BN_SZ =$SIZE_T;
|
|---|
| 98 | if (open CONF,"<${dir}../../opensslconf.h") {
|
|---|
| 99 | while(<CONF>) {
|
|---|
| 100 | if (m/#\s*define\s+SIXTY_FOUR_BIT/) {
|
|---|
| 101 | $BN_SZ=8;
|
|---|
| 102 | $LEVEL="2.0";
|
|---|
| 103 | last;
|
|---|
| 104 | }
|
|---|
| 105 | }
|
|---|
| 106 | close CONF;
|
|---|
| 107 | }
|
|---|
| 108 | }
|
|---|
| 109 |
|
|---|
| 110 | $FRAME=8*$SIZE_T+$FRAME_MARKER; # 8 saved regs + frame marker
|
|---|
| 111 | # [+ argument transfer]
|
|---|
| 112 | $LOCALS=$FRAME-$FRAME_MARKER;
|
|---|
| 113 | $FRAME+=32; # local variables
|
|---|
| 114 |
|
|---|
| 115 | $tp="%r31";
|
|---|
| 116 | $ti1="%r29";
|
|---|
| 117 | $ti0="%r28";
|
|---|
| 118 |
|
|---|
| 119 | $rp="%r26";
|
|---|
| 120 | $ap="%r25";
|
|---|
| 121 | $bp="%r24";
|
|---|
| 122 | $np="%r23";
|
|---|
| 123 | $n0="%r22"; # passed through stack in 32-bit
|
|---|
| 124 | $num="%r21"; # passed through stack in 32-bit
|
|---|
| 125 | $idx="%r20";
|
|---|
| 126 | $arrsz="%r19";
|
|---|
| 127 |
|
|---|
| 128 | $nm1="%r7";
|
|---|
| 129 | $nm0="%r6";
|
|---|
| 130 | $ab1="%r5";
|
|---|
| 131 | $ab0="%r4";
|
|---|
| 132 |
|
|---|
| 133 | $fp="%r3";
|
|---|
| 134 | $hi1="%r2";
|
|---|
| 135 | $hi0="%r1";
|
|---|
| 136 |
|
|---|
| 137 | $xfer=$n0; # accommodates [-16..15] offset in fld[dw]s
|
|---|
| 138 |
|
|---|
| 139 | $fm0="%fr4"; $fti=$fm0;
|
|---|
| 140 | $fbi="%fr5L";
|
|---|
| 141 | $fn0="%fr5R";
|
|---|
| 142 | $fai="%fr6"; $fab0="%fr7"; $fab1="%fr8";
|
|---|
| 143 | $fni="%fr9"; $fnm0="%fr10"; $fnm1="%fr11";
|
|---|
| 144 |
|
|---|
| 145 | $code=<<___;
|
|---|
| 146 | .LEVEL $LEVEL
|
|---|
| 147 | .SPACE \$TEXT\$
|
|---|
| 148 | .SUBSPA \$CODE\$,QUAD=0,ALIGN=8,ACCESS=0x2C,CODE_ONLY
|
|---|
| 149 |
|
|---|
| 150 | .EXPORT bn_mul_mont,ENTRY,ARGW0=GR,ARGW1=GR,ARGW2=GR,ARGW3=GR
|
|---|
| 151 | .ALIGN 64
|
|---|
| 152 | bn_mul_mont
|
|---|
| 153 | .PROC
|
|---|
| 154 | .CALLINFO FRAME=`$FRAME-8*$SIZE_T`,NO_CALLS,SAVE_RP,SAVE_SP,ENTRY_GR=6
|
|---|
| 155 | .ENTRY
|
|---|
| 156 | $PUSH %r2,-$SAVED_RP(%sp) ; standard prologue
|
|---|
| 157 | $PUSHMA %r3,$FRAME(%sp)
|
|---|
| 158 | $PUSH %r4,`-$FRAME+1*$SIZE_T`(%sp)
|
|---|
| 159 | $PUSH %r5,`-$FRAME+2*$SIZE_T`(%sp)
|
|---|
| 160 | $PUSH %r6,`-$FRAME+3*$SIZE_T`(%sp)
|
|---|
| 161 | $PUSH %r7,`-$FRAME+4*$SIZE_T`(%sp)
|
|---|
| 162 | $PUSH %r8,`-$FRAME+5*$SIZE_T`(%sp)
|
|---|
| 163 | $PUSH %r9,`-$FRAME+6*$SIZE_T`(%sp)
|
|---|
| 164 | $PUSH %r10,`-$FRAME+7*$SIZE_T`(%sp)
|
|---|
| 165 | ldo -$FRAME(%sp),$fp
|
|---|
| 166 | ___
|
|---|
| 167 | $code.=<<___ if ($SIZE_T==4);
|
|---|
| 168 | ldw `-$FRAME_MARKER-4`($fp),$n0
|
|---|
| 169 | ldw `-$FRAME_MARKER-8`($fp),$num
|
|---|
| 170 | nop
|
|---|
| 171 | nop ; alignment
|
|---|
| 172 | ___
|
|---|
| 173 | $code.=<<___ if ($BN_SZ==4);
|
|---|
| 174 | comiclr,<= 6,$num,%r0 ; are vectors long enough?
|
|---|
| 175 | b L\$abort
|
|---|
| 176 | ldi 0,%r28 ; signal "unhandled"
|
|---|
| 177 | add,ev %r0,$num,$num ; is $num even?
|
|---|
| 178 | b L\$abort
|
|---|
| 179 | nop
|
|---|
| 180 | or $ap,$np,$ti1
|
|---|
| 181 | extru,= $ti1,31,3,%r0 ; are ap and np 64-bit aligned?
|
|---|
| 182 | b L\$abort
|
|---|
| 183 | nop
|
|---|
| 184 | nop ; alignment
|
|---|
| 185 | nop
|
|---|
| 186 |
|
|---|
| 187 | fldws 0($n0),${fn0}
|
|---|
| 188 | fldws,ma 4($bp),${fbi} ; bp[0]
|
|---|
| 189 | ___
|
|---|
| 190 | $code.=<<___ if ($BN_SZ==8);
|
|---|
| 191 | comib,> 3,$num,L\$abort ; are vectors long enough?
|
|---|
| 192 | ldi 0,%r28 ; signal "unhandled"
|
|---|
| 193 | addl $num,$num,$num ; I operate on 32-bit values
|
|---|
| 194 |
|
|---|
| 195 | fldws 4($n0),${fn0} ; only low part of n0
|
|---|
| 196 | fldws 4($bp),${fbi} ; bp[0] in flipped word order
|
|---|
| 197 | ___
|
|---|
| 198 | $code.=<<___;
|
|---|
| 199 | fldds 0($ap),${fai} ; ap[0,1]
|
|---|
| 200 | fldds 0($np),${fni} ; np[0,1]
|
|---|
| 201 |
|
|---|
| 202 | sh2addl $num,%r0,$arrsz
|
|---|
| 203 | ldi 31,$hi0
|
|---|
| 204 | ldo 36($arrsz),$hi1 ; space for tp[num+1]
|
|---|
| 205 | andcm $hi1,$hi0,$hi1 ; align
|
|---|
| 206 | addl $hi1,%sp,%sp
|
|---|
| 207 | $PUSH $fp,-$SIZE_T(%sp)
|
|---|
| 208 |
|
|---|
| 209 | ldo `$LOCALS+16`($fp),$xfer
|
|---|
| 210 | ldo `$LOCALS+32+4`($fp),$tp
|
|---|
| 211 |
|
|---|
| 212 | xmpyu ${fai}L,${fbi},${fab0} ; ap[0]*bp[0]
|
|---|
| 213 | xmpyu ${fai}R,${fbi},${fab1} ; ap[1]*bp[0]
|
|---|
| 214 | xmpyu ${fn0},${fab0}R,${fm0}
|
|---|
| 215 |
|
|---|
| 216 | addl $arrsz,$ap,$ap ; point at the end
|
|---|
| 217 | addl $arrsz,$np,$np
|
|---|
| 218 | subi 0,$arrsz,$idx ; j=0
|
|---|
| 219 | ldo 8($idx),$idx ; j++++
|
|---|
| 220 |
|
|---|
| 221 | xmpyu ${fni}L,${fm0}R,${fnm0} ; np[0]*m
|
|---|
| 222 | xmpyu ${fni}R,${fm0}R,${fnm1} ; np[1]*m
|
|---|
| 223 | fstds ${fab0},-16($xfer)
|
|---|
| 224 | fstds ${fnm0},-8($xfer)
|
|---|
| 225 | fstds ${fab1},0($xfer)
|
|---|
| 226 | fstds ${fnm1},8($xfer)
|
|---|
| 227 | flddx $idx($ap),${fai} ; ap[2,3]
|
|---|
| 228 | flddx $idx($np),${fni} ; np[2,3]
|
|---|
| 229 | ___
|
|---|
| 230 | $code.=<<___ if ($BN_SZ==4);
|
|---|
| 231 | mtctl $hi0,%cr11 ; $hi0 still holds 31
|
|---|
| 232 | extrd,u,*= $hi0,%sar,1,$hi0 ; executes on PA-RISC 1.0
|
|---|
| 233 | b L\$parisc11
|
|---|
| 234 | nop
|
|---|
| 235 | ___
|
|---|
| 236 | $code.=<<___; # PA-RISC 2.0 code-path
|
|---|
| 237 | xmpyu ${fai}L,${fbi},${fab0} ; ap[j]*bp[0]
|
|---|
| 238 | xmpyu ${fni}L,${fm0}R,${fnm0} ; np[j]*m
|
|---|
| 239 | ldd -16($xfer),$ab0
|
|---|
| 240 | fstds ${fab0},-16($xfer)
|
|---|
| 241 |
|
|---|
| 242 | extrd,u $ab0,31,32,$hi0
|
|---|
| 243 | extrd,u $ab0,63,32,$ab0
|
|---|
| 244 | ldd -8($xfer),$nm0
|
|---|
| 245 | fstds ${fnm0},-8($xfer)
|
|---|
| 246 | ldo 8($idx),$idx ; j++++
|
|---|
| 247 | addl $ab0,$nm0,$nm0 ; low part is discarded
|
|---|
| 248 | extrd,u $nm0,31,32,$hi1
|
|---|
| 249 | |
|---|
| 250 |
|
|---|
| 251 | L\$1st
|
|---|
| 252 | xmpyu ${fai}R,${fbi},${fab1} ; ap[j+1]*bp[0]
|
|---|
| 253 | xmpyu ${fni}R,${fm0}R,${fnm1} ; np[j+1]*m
|
|---|
| 254 | ldd 0($xfer),$ab1
|
|---|
| 255 | fstds ${fab1},0($xfer)
|
|---|
| 256 | addl $hi0,$ab1,$ab1
|
|---|
| 257 | extrd,u $ab1,31,32,$hi0
|
|---|
| 258 | ldd 8($xfer),$nm1
|
|---|
| 259 | fstds ${fnm1},8($xfer)
|
|---|
| 260 | extrd,u $ab1,63,32,$ab1
|
|---|
| 261 | addl $hi1,$nm1,$nm1
|
|---|
| 262 | flddx $idx($ap),${fai} ; ap[j,j+1]
|
|---|
| 263 | flddx $idx($np),${fni} ; np[j,j+1]
|
|---|
| 264 | addl $ab1,$nm1,$nm1
|
|---|
| 265 | extrd,u $nm1,31,32,$hi1
|
|---|
| 266 |
|
|---|
| 267 | xmpyu ${fai}L,${fbi},${fab0} ; ap[j]*bp[0]
|
|---|
| 268 | xmpyu ${fni}L,${fm0}R,${fnm0} ; np[j]*m
|
|---|
| 269 | ldd -16($xfer),$ab0
|
|---|
| 270 | fstds ${fab0},-16($xfer)
|
|---|
| 271 | addl $hi0,$ab0,$ab0
|
|---|
| 272 | extrd,u $ab0,31,32,$hi0
|
|---|
| 273 | ldd -8($xfer),$nm0
|
|---|
| 274 | fstds ${fnm0},-8($xfer)
|
|---|
| 275 | extrd,u $ab0,63,32,$ab0
|
|---|
| 276 | addl $hi1,$nm0,$nm0
|
|---|
| 277 | stw $nm1,-4($tp) ; tp[j-1]
|
|---|
| 278 | addl $ab0,$nm0,$nm0
|
|---|
| 279 | stw,ma $nm0,8($tp) ; tp[j-1]
|
|---|
| 280 | addib,<> 8,$idx,L\$1st ; j++++
|
|---|
| 281 | extrd,u $nm0,31,32,$hi1
|
|---|
| 282 |
|
|---|
| 283 | xmpyu ${fai}R,${fbi},${fab1} ; ap[j]*bp[0]
|
|---|
| 284 | xmpyu ${fni}R,${fm0}R,${fnm1} ; np[j]*m
|
|---|
| 285 | ldd 0($xfer),$ab1
|
|---|
| 286 | fstds ${fab1},0($xfer)
|
|---|
| 287 | addl $hi0,$ab1,$ab1
|
|---|
| 288 | extrd,u $ab1,31,32,$hi0
|
|---|
| 289 | ldd 8($xfer),$nm1
|
|---|
| 290 | fstds ${fnm1},8($xfer)
|
|---|
| 291 | extrd,u $ab1,63,32,$ab1
|
|---|
| 292 | addl $hi1,$nm1,$nm1
|
|---|
| 293 | ldd -16($xfer),$ab0
|
|---|
| 294 | addl $ab1,$nm1,$nm1
|
|---|
| 295 | ldd -8($xfer),$nm0
|
|---|
| 296 | extrd,u $nm1,31,32,$hi1
|
|---|
| 297 |
|
|---|
| 298 | addl $hi0,$ab0,$ab0
|
|---|
| 299 | extrd,u $ab0,31,32,$hi0
|
|---|
| 300 | stw $nm1,-4($tp) ; tp[j-1]
|
|---|
| 301 | extrd,u $ab0,63,32,$ab0
|
|---|
| 302 | addl $hi1,$nm0,$nm0
|
|---|
| 303 | ldd 0($xfer),$ab1
|
|---|
| 304 | addl $ab0,$nm0,$nm0
|
|---|
| 305 | ldd,mb 8($xfer),$nm1
|
|---|
| 306 | extrd,u $nm0,31,32,$hi1
|
|---|
| 307 | stw,ma $nm0,8($tp) ; tp[j-1]
|
|---|
| 308 |
|
|---|
| 309 | ldo -1($num),$num ; i--
|
|---|
| 310 | subi 0,$arrsz,$idx ; j=0
|
|---|
| 311 | ___
|
|---|
| 312 | $code.=<<___ if ($BN_SZ==4);
|
|---|
| 313 | fldws,ma 4($bp),${fbi} ; bp[1]
|
|---|
| 314 | ___
|
|---|
| 315 | $code.=<<___ if ($BN_SZ==8);
|
|---|
| 316 | fldws 0($bp),${fbi} ; bp[1] in flipped word order
|
|---|
| 317 | ___
|
|---|
| 318 | $code.=<<___;
|
|---|
| 319 | flddx $idx($ap),${fai} ; ap[0,1]
|
|---|
| 320 | flddx $idx($np),${fni} ; np[0,1]
|
|---|
| 321 | fldws 8($xfer),${fti}R ; tp[0]
|
|---|
| 322 | addl $hi0,$ab1,$ab1
|
|---|
| 323 | extrd,u $ab1,31,32,$hi0
|
|---|
| 324 | extrd,u $ab1,63,32,$ab1
|
|---|
| 325 | ldo 8($idx),$idx ; j++++
|
|---|
| 326 | xmpyu ${fai}L,${fbi},${fab0} ; ap[0]*bp[1]
|
|---|
| 327 | xmpyu ${fai}R,${fbi},${fab1} ; ap[1]*bp[1]
|
|---|
| 328 | addl $hi1,$nm1,$nm1
|
|---|
| 329 | addl $ab1,$nm1,$nm1
|
|---|
| 330 | extrd,u $nm1,31,32,$hi1
|
|---|
| 331 | fstws,mb ${fab0}L,-8($xfer) ; save high part
|
|---|
| 332 | stw $nm1,-4($tp) ; tp[j-1]
|
|---|
| 333 |
|
|---|
| 334 | fcpy,sgl %fr0,${fti}L ; zero high part
|
|---|
| 335 | fcpy,sgl %fr0,${fab0}L
|
|---|
| 336 | addl $hi1,$hi0,$hi0
|
|---|
| 337 | extrd,u $hi0,31,32,$hi1
|
|---|
| 338 | fcnvxf,dbl,dbl ${fti},${fti} ; 32-bit unsigned int -> double
|
|---|
| 339 | fcnvxf,dbl,dbl ${fab0},${fab0}
|
|---|
| 340 | stw $hi0,0($tp)
|
|---|
| 341 | stw $hi1,4($tp)
|
|---|
| 342 |
|
|---|
| 343 | fadd,dbl ${fti},${fab0},${fab0} ; add tp[0]
|
|---|
| 344 | fcnvfx,dbl,dbl ${fab0},${fab0} ; double -> 33-bit unsigned int
|
|---|
| 345 | xmpyu ${fn0},${fab0}R,${fm0}
|
|---|
| 346 | ldo `$LOCALS+32+4`($fp),$tp
|
|---|
| 347 | L\$outer
|
|---|
| 348 | xmpyu ${fni}L,${fm0}R,${fnm0} ; np[0]*m
|
|---|
| 349 | xmpyu ${fni}R,${fm0}R,${fnm1} ; np[1]*m
|
|---|
| 350 | fstds ${fab0},-16($xfer) ; 33-bit value
|
|---|
| 351 | fstds ${fnm0},-8($xfer)
|
|---|
| 352 | flddx $idx($ap),${fai} ; ap[2]
|
|---|
| 353 | flddx $idx($np),${fni} ; np[2]
|
|---|
| 354 | ldo 8($idx),$idx ; j++++
|
|---|
| 355 | ldd -16($xfer),$ab0 ; 33-bit value
|
|---|
| 356 | ldd -8($xfer),$nm0
|
|---|
| 357 | ldw 0($xfer),$hi0 ; high part
|
|---|
| 358 |
|
|---|
| 359 | xmpyu ${fai}L,${fbi},${fab0} ; ap[j]*bp[i]
|
|---|
| 360 | xmpyu ${fni}L,${fm0}R,${fnm0} ; np[j]*m
|
|---|
| 361 | extrd,u $ab0,31,32,$ti0 ; carry bit
|
|---|
| 362 | extrd,u $ab0,63,32,$ab0
|
|---|
| 363 | fstds ${fab1},0($xfer)
|
|---|
| 364 | addl $ti0,$hi0,$hi0 ; account carry bit
|
|---|
| 365 | fstds ${fnm1},8($xfer)
|
|---|
| 366 | addl $ab0,$nm0,$nm0 ; low part is discarded
|
|---|
| 367 | ldw 0($tp),$ti1 ; tp[1]
|
|---|
| 368 | extrd,u $nm0,31,32,$hi1
|
|---|
| 369 | fstds ${fab0},-16($xfer)
|
|---|
| 370 | fstds ${fnm0},-8($xfer)
|
|---|
| 371 | |
|---|
| 372 |
|
|---|
| 373 | L\$inner
|
|---|
| 374 | xmpyu ${fai}R,${fbi},${fab1} ; ap[j+1]*bp[i]
|
|---|
| 375 | xmpyu ${fni}R,${fm0}R,${fnm1} ; np[j+1]*m
|
|---|
| 376 | ldd 0($xfer),$ab1
|
|---|
| 377 | fstds ${fab1},0($xfer)
|
|---|
| 378 | addl $hi0,$ti1,$ti1
|
|---|
| 379 | addl $ti1,$ab1,$ab1
|
|---|
| 380 | ldd 8($xfer),$nm1
|
|---|
| 381 | fstds ${fnm1},8($xfer)
|
|---|
| 382 | extrd,u $ab1,31,32,$hi0
|
|---|
| 383 | extrd,u $ab1,63,32,$ab1
|
|---|
| 384 | flddx $idx($ap),${fai} ; ap[j,j+1]
|
|---|
| 385 | flddx $idx($np),${fni} ; np[j,j+1]
|
|---|
| 386 | addl $hi1,$nm1,$nm1
|
|---|
| 387 | addl $ab1,$nm1,$nm1
|
|---|
| 388 | ldw 4($tp),$ti0 ; tp[j]
|
|---|
| 389 | stw $nm1,-4($tp) ; tp[j-1]
|
|---|
| 390 |
|
|---|
| 391 | xmpyu ${fai}L,${fbi},${fab0} ; ap[j]*bp[i]
|
|---|
| 392 | xmpyu ${fni}L,${fm0}R,${fnm0} ; np[j]*m
|
|---|
| 393 | ldd -16($xfer),$ab0
|
|---|
| 394 | fstds ${fab0},-16($xfer)
|
|---|
| 395 | addl $hi0,$ti0,$ti0
|
|---|
| 396 | addl $ti0,$ab0,$ab0
|
|---|
| 397 | ldd -8($xfer),$nm0
|
|---|
| 398 | fstds ${fnm0},-8($xfer)
|
|---|
| 399 | extrd,u $ab0,31,32,$hi0
|
|---|
| 400 | extrd,u $nm1,31,32,$hi1
|
|---|
| 401 | ldw 8($tp),$ti1 ; tp[j]
|
|---|
| 402 | extrd,u $ab0,63,32,$ab0
|
|---|
| 403 | addl $hi1,$nm0,$nm0
|
|---|
| 404 | addl $ab0,$nm0,$nm0
|
|---|
| 405 | stw,ma $nm0,8($tp) ; tp[j-1]
|
|---|
| 406 | addib,<> 8,$idx,L\$inner ; j++++
|
|---|
| 407 | extrd,u $nm0,31,32,$hi1
|
|---|
| 408 |
|
|---|
| 409 | xmpyu ${fai}R,${fbi},${fab1} ; ap[j]*bp[i]
|
|---|
| 410 | xmpyu ${fni}R,${fm0}R,${fnm1} ; np[j]*m
|
|---|
| 411 | ldd 0($xfer),$ab1
|
|---|
| 412 | fstds ${fab1},0($xfer)
|
|---|
| 413 | addl $hi0,$ti1,$ti1
|
|---|
| 414 | addl $ti1,$ab1,$ab1
|
|---|
| 415 | ldd 8($xfer),$nm1
|
|---|
| 416 | fstds ${fnm1},8($xfer)
|
|---|
| 417 | extrd,u $ab1,31,32,$hi0
|
|---|
| 418 | extrd,u $ab1,63,32,$ab1
|
|---|
| 419 | ldw 4($tp),$ti0 ; tp[j]
|
|---|
| 420 | addl $hi1,$nm1,$nm1
|
|---|
| 421 | addl $ab1,$nm1,$nm1
|
|---|
| 422 | ldd -16($xfer),$ab0
|
|---|
| 423 | ldd -8($xfer),$nm0
|
|---|
| 424 | extrd,u $nm1,31,32,$hi1
|
|---|
| 425 |
|
|---|
| 426 | addl $hi0,$ab0,$ab0
|
|---|
| 427 | addl $ti0,$ab0,$ab0
|
|---|
| 428 | stw $nm1,-4($tp) ; tp[j-1]
|
|---|
| 429 | extrd,u $ab0,31,32,$hi0
|
|---|
| 430 | ldw 8($tp),$ti1 ; tp[j]
|
|---|
| 431 | extrd,u $ab0,63,32,$ab0
|
|---|
| 432 | addl $hi1,$nm0,$nm0
|
|---|
| 433 | ldd 0($xfer),$ab1
|
|---|
| 434 | addl $ab0,$nm0,$nm0
|
|---|
| 435 | ldd,mb 8($xfer),$nm1
|
|---|
| 436 | extrd,u $nm0,31,32,$hi1
|
|---|
| 437 | stw,ma $nm0,8($tp) ; tp[j-1]
|
|---|
| 438 |
|
|---|
| 439 | addib,= -1,$num,L\$outerdone ; i--
|
|---|
| 440 | subi 0,$arrsz,$idx ; j=0
|
|---|
| 441 | ___
|
|---|
| 442 | $code.=<<___ if ($BN_SZ==4);
|
|---|
| 443 | fldws,ma 4($bp),${fbi} ; bp[i]
|
|---|
| 444 | ___
|
|---|
| 445 | $code.=<<___ if ($BN_SZ==8);
|
|---|
| 446 | ldi 12,$ti0 ; bp[i] in flipped word order
|
|---|
| 447 | addl,ev %r0,$num,$num
|
|---|
| 448 | ldi -4,$ti0
|
|---|
| 449 | addl $ti0,$bp,$bp
|
|---|
| 450 | fldws 0($bp),${fbi}
|
|---|
| 451 | ___
|
|---|
| 452 | $code.=<<___;
|
|---|
| 453 | flddx $idx($ap),${fai} ; ap[0]
|
|---|
| 454 | addl $hi0,$ab1,$ab1
|
|---|
| 455 | flddx $idx($np),${fni} ; np[0]
|
|---|
| 456 | fldws 8($xfer),${fti}R ; tp[0]
|
|---|
| 457 | addl $ti1,$ab1,$ab1
|
|---|
| 458 | extrd,u $ab1,31,32,$hi0
|
|---|
| 459 | extrd,u $ab1,63,32,$ab1
|
|---|
| 460 |
|
|---|
| 461 | ldo 8($idx),$idx ; j++++
|
|---|
| 462 | xmpyu ${fai}L,${fbi},${fab0} ; ap[0]*bp[i]
|
|---|
| 463 | xmpyu ${fai}R,${fbi},${fab1} ; ap[1]*bp[i]
|
|---|
| 464 | ldw 4($tp),$ti0 ; tp[j]
|
|---|
| 465 |
|
|---|
| 466 | addl $hi1,$nm1,$nm1
|
|---|
| 467 | fstws,mb ${fab0}L,-8($xfer) ; save high part
|
|---|
| 468 | addl $ab1,$nm1,$nm1
|
|---|
| 469 | extrd,u $nm1,31,32,$hi1
|
|---|
| 470 | fcpy,sgl %fr0,${fti}L ; zero high part
|
|---|
| 471 | fcpy,sgl %fr0,${fab0}L
|
|---|
| 472 | stw $nm1,-4($tp) ; tp[j-1]
|
|---|
| 473 |
|
|---|
| 474 | fcnvxf,dbl,dbl ${fti},${fti} ; 32-bit unsigned int -> double
|
|---|
| 475 | fcnvxf,dbl,dbl ${fab0},${fab0}
|
|---|
| 476 | addl $hi1,$hi0,$hi0
|
|---|
| 477 | fadd,dbl ${fti},${fab0},${fab0} ; add tp[0]
|
|---|
| 478 | addl $ti0,$hi0,$hi0
|
|---|
| 479 | extrd,u $hi0,31,32,$hi1
|
|---|
| 480 | fcnvfx,dbl,dbl ${fab0},${fab0} ; double -> 33-bit unsigned int
|
|---|
| 481 | stw $hi0,0($tp)
|
|---|
| 482 | stw $hi1,4($tp)
|
|---|
| 483 | xmpyu ${fn0},${fab0}R,${fm0}
|
|---|
| 484 |
|
|---|
| 485 | b L\$outer
|
|---|
| 486 | ldo `$LOCALS+32+4`($fp),$tp
|
|---|
| 487 | |
|---|
| 488 |
|
|---|
| 489 | L\$outerdone
|
|---|
| 490 | addl $hi0,$ab1,$ab1
|
|---|
| 491 | addl $ti1,$ab1,$ab1
|
|---|
| 492 | extrd,u $ab1,31,32,$hi0
|
|---|
| 493 | extrd,u $ab1,63,32,$ab1
|
|---|
| 494 |
|
|---|
| 495 | ldw 4($tp),$ti0 ; tp[j]
|
|---|
| 496 |
|
|---|
| 497 | addl $hi1,$nm1,$nm1
|
|---|
| 498 | addl $ab1,$nm1,$nm1
|
|---|
| 499 | extrd,u $nm1,31,32,$hi1
|
|---|
| 500 | stw $nm1,-4($tp) ; tp[j-1]
|
|---|
| 501 |
|
|---|
| 502 | addl $hi1,$hi0,$hi0
|
|---|
| 503 | addl $ti0,$hi0,$hi0
|
|---|
| 504 | extrd,u $hi0,31,32,$hi1
|
|---|
| 505 | stw $hi0,0($tp)
|
|---|
| 506 | stw $hi1,4($tp)
|
|---|
| 507 |
|
|---|
| 508 | ldo `$LOCALS+32`($fp),$tp
|
|---|
| 509 | sub %r0,%r0,%r0 ; clear borrow
|
|---|
| 510 | ___
|
|---|
| 511 | $code.=<<___ if ($BN_SZ==4);
|
|---|
| 512 | ldws,ma 4($tp),$ti0
|
|---|
| 513 | extru,= $rp,31,3,%r0 ; is rp 64-bit aligned?
|
|---|
| 514 | b L\$sub_pa11
|
|---|
| 515 | addl $tp,$arrsz,$tp
|
|---|
| 516 | L\$sub
|
|---|
| 517 | ldwx $idx($np),$hi0
|
|---|
| 518 | subb $ti0,$hi0,$hi1
|
|---|
| 519 | ldwx $idx($tp),$ti0
|
|---|
| 520 | addib,<> 4,$idx,L\$sub
|
|---|
| 521 | stws,ma $hi1,4($rp)
|
|---|
| 522 |
|
|---|
| 523 | subb $ti0,%r0,$hi1
|
|---|
| 524 | ___
|
|---|
| 525 | $code.=<<___ if ($BN_SZ==8);
|
|---|
| 526 | ldd,ma 8($tp),$ti0
|
|---|
| 527 | L\$sub
|
|---|
| 528 | ldd $idx($np),$hi0
|
|---|
| 529 | shrpd $ti0,$ti0,32,$ti0 ; flip word order
|
|---|
| 530 | std $ti0,-8($tp) ; save flipped value
|
|---|
| 531 | sub,db $ti0,$hi0,$hi1
|
|---|
| 532 | ldd,ma 8($tp),$ti0
|
|---|
| 533 | addib,<> 8,$idx,L\$sub
|
|---|
| 534 | std,ma $hi1,8($rp)
|
|---|
| 535 |
|
|---|
| 536 | extrd,u $ti0,31,32,$ti0 ; carry in flipped word order
|
|---|
| 537 | sub,db $ti0,%r0,$hi1
|
|---|
| 538 | ___
|
|---|
| 539 | $code.=<<___;
|
|---|
| 540 | ldo `$LOCALS+32`($fp),$tp
|
|---|
| 541 | sub $rp,$arrsz,$rp ; rewind rp
|
|---|
| 542 | subi 0,$arrsz,$idx
|
|---|
| 543 | L\$copy
|
|---|
| 544 | ldd 0($tp),$ti0
|
|---|
| 545 | ldd 0($rp),$hi0
|
|---|
| 546 | std,ma %r0,8($tp)
|
|---|
| 547 | comiclr,= 0,$hi1,%r0
|
|---|
| 548 | copy $ti0,$hi0
|
|---|
| 549 | addib,<> 8,$idx,L\$copy
|
|---|
| 550 | std,ma $hi0,8($rp)
|
|---|
| 551 | ___
|
|---|
| 552 |
|
|---|
| 553 | if ($BN_SZ==4) { # PA-RISC 1.1 code-path
|
|---|
| 554 | $ablo=$ab0;
|
|---|
| 555 | $abhi=$ab1;
|
|---|
| 556 | $nmlo0=$nm0;
|
|---|
| 557 | $nmhi0=$nm1;
|
|---|
| 558 | $nmlo1="%r9";
|
|---|
| 559 | $nmhi1="%r8";
|
|---|
| 560 |
|
|---|
| 561 | $code.=<<___;
|
|---|
| 562 | b L\$done
|
|---|
| 563 | nop
|
|---|
| 564 |
|
|---|
| 565 | .ALIGN 8
|
|---|
| 566 | L\$parisc11
|
|---|
| 567 | xmpyu ${fai}L,${fbi},${fab0} ; ap[j]*bp[0]
|
|---|
| 568 | xmpyu ${fni}L,${fm0}R,${fnm0} ; np[j]*m
|
|---|
| 569 | ldw -12($xfer),$ablo
|
|---|
| 570 | ldw -16($xfer),$hi0
|
|---|
| 571 | ldw -4($xfer),$nmlo0
|
|---|
| 572 | ldw -8($xfer),$nmhi0
|
|---|
| 573 | fstds ${fab0},-16($xfer)
|
|---|
| 574 | fstds ${fnm0},-8($xfer)
|
|---|
| 575 |
|
|---|
| 576 | ldo 8($idx),$idx ; j++++
|
|---|
| 577 | add $ablo,$nmlo0,$nmlo0 ; discarded
|
|---|
| 578 | addc %r0,$nmhi0,$hi1
|
|---|
| 579 | ldw 4($xfer),$ablo
|
|---|
| 580 | ldw 0($xfer),$abhi
|
|---|
| 581 | nop
|
|---|
| 582 | |
|---|
| 583 |
|
|---|
| 584 | L\$1st_pa11
|
|---|
| 585 | xmpyu ${fai}R,${fbi},${fab1} ; ap[j+1]*bp[0]
|
|---|
| 586 | flddx $idx($ap),${fai} ; ap[j,j+1]
|
|---|
| 587 | xmpyu ${fni}R,${fm0}R,${fnm1} ; np[j+1]*m
|
|---|
| 588 | flddx $idx($np),${fni} ; np[j,j+1]
|
|---|
| 589 | add $hi0,$ablo,$ablo
|
|---|
| 590 | ldw 12($xfer),$nmlo1
|
|---|
| 591 | addc %r0,$abhi,$hi0
|
|---|
| 592 | ldw 8($xfer),$nmhi1
|
|---|
| 593 | add $ablo,$nmlo1,$nmlo1
|
|---|
| 594 | fstds ${fab1},0($xfer)
|
|---|
| 595 | addc %r0,$nmhi1,$nmhi1
|
|---|
| 596 | fstds ${fnm1},8($xfer)
|
|---|
| 597 | add $hi1,$nmlo1,$nmlo1
|
|---|
| 598 | ldw -12($xfer),$ablo
|
|---|
| 599 | addc %r0,$nmhi1,$hi1
|
|---|
| 600 | ldw -16($xfer),$abhi
|
|---|
| 601 |
|
|---|
| 602 | xmpyu ${fai}L,${fbi},${fab0} ; ap[j]*bp[0]
|
|---|
| 603 | ldw -4($xfer),$nmlo0
|
|---|
| 604 | xmpyu ${fni}L,${fm0}R,${fnm0} ; np[j]*m
|
|---|
| 605 | ldw -8($xfer),$nmhi0
|
|---|
| 606 | add $hi0,$ablo,$ablo
|
|---|
| 607 | stw $nmlo1,-4($tp) ; tp[j-1]
|
|---|
| 608 | addc %r0,$abhi,$hi0
|
|---|
| 609 | fstds ${fab0},-16($xfer)
|
|---|
| 610 | add $ablo,$nmlo0,$nmlo0
|
|---|
| 611 | fstds ${fnm0},-8($xfer)
|
|---|
| 612 | addc %r0,$nmhi0,$nmhi0
|
|---|
| 613 | ldw 0($xfer),$abhi
|
|---|
| 614 | add $hi1,$nmlo0,$nmlo0
|
|---|
| 615 | ldw 4($xfer),$ablo
|
|---|
| 616 | stws,ma $nmlo0,8($tp) ; tp[j-1]
|
|---|
| 617 | addib,<> 8,$idx,L\$1st_pa11 ; j++++
|
|---|
| 618 | addc %r0,$nmhi0,$hi1
|
|---|
| 619 |
|
|---|
| 620 | ldw 8($xfer),$nmhi1
|
|---|
| 621 | ldw 12($xfer),$nmlo1
|
|---|
| 622 | xmpyu ${fai}R,${fbi},${fab1} ; ap[j]*bp[0]
|
|---|
| 623 | xmpyu ${fni}R,${fm0}R,${fnm1} ; np[j]*m
|
|---|
| 624 | add $hi0,$ablo,$ablo
|
|---|
| 625 | fstds ${fab1},0($xfer)
|
|---|
| 626 | addc %r0,$abhi,$hi0
|
|---|
| 627 | fstds ${fnm1},8($xfer)
|
|---|
| 628 | add $ablo,$nmlo1,$nmlo1
|
|---|
| 629 | ldw -16($xfer),$abhi
|
|---|
| 630 | addc %r0,$nmhi1,$nmhi1
|
|---|
| 631 | ldw -12($xfer),$ablo
|
|---|
| 632 | add $hi1,$nmlo1,$nmlo1
|
|---|
| 633 | ldw -8($xfer),$nmhi0
|
|---|
| 634 | addc %r0,$nmhi1,$hi1
|
|---|
| 635 | ldw -4($xfer),$nmlo0
|
|---|
| 636 |
|
|---|
| 637 | add $hi0,$ablo,$ablo
|
|---|
| 638 | stw $nmlo1,-4($tp) ; tp[j-1]
|
|---|
| 639 | addc %r0,$abhi,$hi0
|
|---|
| 640 | ldw 0($xfer),$abhi
|
|---|
| 641 | add $ablo,$nmlo0,$nmlo0
|
|---|
| 642 | ldw 4($xfer),$ablo
|
|---|
| 643 | addc %r0,$nmhi0,$nmhi0
|
|---|
| 644 | ldws,mb 8($xfer),$nmhi1
|
|---|
| 645 | add $hi1,$nmlo0,$nmlo0
|
|---|
| 646 | ldw 4($xfer),$nmlo1
|
|---|
| 647 | addc %r0,$nmhi0,$hi1
|
|---|
| 648 | stws,ma $nmlo0,8($tp) ; tp[j-1]
|
|---|
| 649 |
|
|---|
| 650 | ldo -1($num),$num ; i--
|
|---|
| 651 | subi 0,$arrsz,$idx ; j=0
|
|---|
| 652 |
|
|---|
| 653 | fldws,ma 4($bp),${fbi} ; bp[1]
|
|---|
| 654 | flddx $idx($ap),${fai} ; ap[0,1]
|
|---|
| 655 | flddx $idx($np),${fni} ; np[0,1]
|
|---|
| 656 | fldws 8($xfer),${fti}R ; tp[0]
|
|---|
| 657 | add $hi0,$ablo,$ablo
|
|---|
| 658 | addc %r0,$abhi,$hi0
|
|---|
| 659 | ldo 8($idx),$idx ; j++++
|
|---|
| 660 | xmpyu ${fai}L,${fbi},${fab0} ; ap[0]*bp[1]
|
|---|
| 661 | xmpyu ${fai}R,${fbi},${fab1} ; ap[1]*bp[1]
|
|---|
| 662 | add $hi1,$nmlo1,$nmlo1
|
|---|
| 663 | addc %r0,$nmhi1,$nmhi1
|
|---|
| 664 | add $ablo,$nmlo1,$nmlo1
|
|---|
| 665 | addc %r0,$nmhi1,$hi1
|
|---|
| 666 | fstws,mb ${fab0}L,-8($xfer) ; save high part
|
|---|
| 667 | stw $nmlo1,-4($tp) ; tp[j-1]
|
|---|
| 668 |
|
|---|
| 669 | fcpy,sgl %fr0,${fti}L ; zero high part
|
|---|
| 670 | fcpy,sgl %fr0,${fab0}L
|
|---|
| 671 | add $hi1,$hi0,$hi0
|
|---|
| 672 | addc %r0,%r0,$hi1
|
|---|
| 673 | fcnvxf,dbl,dbl ${fti},${fti} ; 32-bit unsigned int -> double
|
|---|
| 674 | fcnvxf,dbl,dbl ${fab0},${fab0}
|
|---|
| 675 | stw $hi0,0($tp)
|
|---|
| 676 | stw $hi1,4($tp)
|
|---|
| 677 |
|
|---|
| 678 | fadd,dbl ${fti},${fab0},${fab0} ; add tp[0]
|
|---|
| 679 | fcnvfx,dbl,dbl ${fab0},${fab0} ; double -> 33-bit unsigned int
|
|---|
| 680 | xmpyu ${fn0},${fab0}R,${fm0}
|
|---|
| 681 | ldo `$LOCALS+32+4`($fp),$tp
|
|---|
| 682 | L\$outer_pa11
|
|---|
| 683 | xmpyu ${fni}L,${fm0}R,${fnm0} ; np[0]*m
|
|---|
| 684 | xmpyu ${fni}R,${fm0}R,${fnm1} ; np[1]*m
|
|---|
| 685 | fstds ${fab0},-16($xfer) ; 33-bit value
|
|---|
| 686 | fstds ${fnm0},-8($xfer)
|
|---|
| 687 | flddx $idx($ap),${fai} ; ap[2,3]
|
|---|
| 688 | flddx $idx($np),${fni} ; np[2,3]
|
|---|
| 689 | ldw -16($xfer),$abhi ; carry bit actually
|
|---|
| 690 | ldo 8($idx),$idx ; j++++
|
|---|
| 691 | ldw -12($xfer),$ablo
|
|---|
| 692 | ldw -8($xfer),$nmhi0
|
|---|
| 693 | ldw -4($xfer),$nmlo0
|
|---|
| 694 | ldw 0($xfer),$hi0 ; high part
|
|---|
| 695 |
|
|---|
| 696 | xmpyu ${fai}L,${fbi},${fab0} ; ap[j]*bp[i]
|
|---|
| 697 | xmpyu ${fni}L,${fm0}R,${fnm0} ; np[j]*m
|
|---|
| 698 | fstds ${fab1},0($xfer)
|
|---|
| 699 | addl $abhi,$hi0,$hi0 ; account carry bit
|
|---|
| 700 | fstds ${fnm1},8($xfer)
|
|---|
| 701 | add $ablo,$nmlo0,$nmlo0 ; discarded
|
|---|
| 702 | ldw 0($tp),$ti1 ; tp[1]
|
|---|
| 703 | addc %r0,$nmhi0,$hi1
|
|---|
| 704 | fstds ${fab0},-16($xfer)
|
|---|
| 705 | fstds ${fnm0},-8($xfer)
|
|---|
| 706 | ldw 4($xfer),$ablo
|
|---|
| 707 | ldw 0($xfer),$abhi
|
|---|
| 708 | |
|---|
| 709 |
|
|---|
| 710 | L\$inner_pa11
|
|---|
| 711 | xmpyu ${fai}R,${fbi},${fab1} ; ap[j+1]*bp[i]
|
|---|
| 712 | flddx $idx($ap),${fai} ; ap[j,j+1]
|
|---|
| 713 | xmpyu ${fni}R,${fm0}R,${fnm1} ; np[j+1]*m
|
|---|
| 714 | flddx $idx($np),${fni} ; np[j,j+1]
|
|---|
| 715 | add $hi0,$ablo,$ablo
|
|---|
| 716 | ldw 4($tp),$ti0 ; tp[j]
|
|---|
| 717 | addc %r0,$abhi,$abhi
|
|---|
| 718 | ldw 12($xfer),$nmlo1
|
|---|
| 719 | add $ti1,$ablo,$ablo
|
|---|
| 720 | ldw 8($xfer),$nmhi1
|
|---|
| 721 | addc %r0,$abhi,$hi0
|
|---|
| 722 | fstds ${fab1},0($xfer)
|
|---|
| 723 | add $ablo,$nmlo1,$nmlo1
|
|---|
| 724 | fstds ${fnm1},8($xfer)
|
|---|
| 725 | addc %r0,$nmhi1,$nmhi1
|
|---|
| 726 | ldw -12($xfer),$ablo
|
|---|
| 727 | add $hi1,$nmlo1,$nmlo1
|
|---|
| 728 | ldw -16($xfer),$abhi
|
|---|
| 729 | addc %r0,$nmhi1,$hi1
|
|---|
| 730 |
|
|---|
| 731 | xmpyu ${fai}L,${fbi},${fab0} ; ap[j]*bp[i]
|
|---|
| 732 | ldw 8($tp),$ti1 ; tp[j]
|
|---|
| 733 | xmpyu ${fni}L,${fm0}R,${fnm0} ; np[j]*m
|
|---|
| 734 | ldw -4($xfer),$nmlo0
|
|---|
| 735 | add $hi0,$ablo,$ablo
|
|---|
| 736 | ldw -8($xfer),$nmhi0
|
|---|
| 737 | addc %r0,$abhi,$abhi
|
|---|
| 738 | stw $nmlo1,-4($tp) ; tp[j-1]
|
|---|
| 739 | add $ti0,$ablo,$ablo
|
|---|
| 740 | fstds ${fab0},-16($xfer)
|
|---|
| 741 | addc %r0,$abhi,$hi0
|
|---|
| 742 | fstds ${fnm0},-8($xfer)
|
|---|
| 743 | add $ablo,$nmlo0,$nmlo0
|
|---|
| 744 | ldw 4($xfer),$ablo
|
|---|
| 745 | addc %r0,$nmhi0,$nmhi0
|
|---|
| 746 | ldw 0($xfer),$abhi
|
|---|
| 747 | add $hi1,$nmlo0,$nmlo0
|
|---|
| 748 | stws,ma $nmlo0,8($tp) ; tp[j-1]
|
|---|
| 749 | addib,<> 8,$idx,L\$inner_pa11 ; j++++
|
|---|
| 750 | addc %r0,$nmhi0,$hi1
|
|---|
| 751 |
|
|---|
| 752 | xmpyu ${fai}R,${fbi},${fab1} ; ap[j]*bp[i]
|
|---|
| 753 | ldw 12($xfer),$nmlo1
|
|---|
| 754 | xmpyu ${fni}R,${fm0}R,${fnm1} ; np[j]*m
|
|---|
| 755 | ldw 8($xfer),$nmhi1
|
|---|
| 756 | add $hi0,$ablo,$ablo
|
|---|
| 757 | ldw 4($tp),$ti0 ; tp[j]
|
|---|
| 758 | addc %r0,$abhi,$abhi
|
|---|
| 759 | fstds ${fab1},0($xfer)
|
|---|
| 760 | add $ti1,$ablo,$ablo
|
|---|
| 761 | fstds ${fnm1},8($xfer)
|
|---|
| 762 | addc %r0,$abhi,$hi0
|
|---|
| 763 | ldw -16($xfer),$abhi
|
|---|
| 764 | add $ablo,$nmlo1,$nmlo1
|
|---|
| 765 | ldw -12($xfer),$ablo
|
|---|
| 766 | addc %r0,$nmhi1,$nmhi1
|
|---|
| 767 | ldw -8($xfer),$nmhi0
|
|---|
| 768 | add $hi1,$nmlo1,$nmlo1
|
|---|
| 769 | ldw -4($xfer),$nmlo0
|
|---|
| 770 | addc %r0,$nmhi1,$hi1
|
|---|
| 771 |
|
|---|
| 772 | add $hi0,$ablo,$ablo
|
|---|
| 773 | stw $nmlo1,-4($tp) ; tp[j-1]
|
|---|
| 774 | addc %r0,$abhi,$abhi
|
|---|
| 775 | add $ti0,$ablo,$ablo
|
|---|
| 776 | ldw 8($tp),$ti1 ; tp[j]
|
|---|
| 777 | addc %r0,$abhi,$hi0
|
|---|
| 778 | ldw 0($xfer),$abhi
|
|---|
| 779 | add $ablo,$nmlo0,$nmlo0
|
|---|
| 780 | ldw 4($xfer),$ablo
|
|---|
| 781 | addc %r0,$nmhi0,$nmhi0
|
|---|
| 782 | ldws,mb 8($xfer),$nmhi1
|
|---|
| 783 | add $hi1,$nmlo0,$nmlo0
|
|---|
| 784 | ldw 4($xfer),$nmlo1
|
|---|
| 785 | addc %r0,$nmhi0,$hi1
|
|---|
| 786 | stws,ma $nmlo0,8($tp) ; tp[j-1]
|
|---|
| 787 |
|
|---|
| 788 | addib,= -1,$num,L\$outerdone_pa11; i--
|
|---|
| 789 | subi 0,$arrsz,$idx ; j=0
|
|---|
| 790 |
|
|---|
| 791 | fldws,ma 4($bp),${fbi} ; bp[i]
|
|---|
| 792 | flddx $idx($ap),${fai} ; ap[0]
|
|---|
| 793 | add $hi0,$ablo,$ablo
|
|---|
| 794 | addc %r0,$abhi,$abhi
|
|---|
| 795 | flddx $idx($np),${fni} ; np[0]
|
|---|
| 796 | fldws 8($xfer),${fti}R ; tp[0]
|
|---|
| 797 | add $ti1,$ablo,$ablo
|
|---|
| 798 | addc %r0,$abhi,$hi0
|
|---|
| 799 |
|
|---|
| 800 | ldo 8($idx),$idx ; j++++
|
|---|
| 801 | xmpyu ${fai}L,${fbi},${fab0} ; ap[0]*bp[i]
|
|---|
| 802 | xmpyu ${fai}R,${fbi},${fab1} ; ap[1]*bp[i]
|
|---|
| 803 | ldw 4($tp),$ti0 ; tp[j]
|
|---|
| 804 |
|
|---|
| 805 | add $hi1,$nmlo1,$nmlo1
|
|---|
| 806 | addc %r0,$nmhi1,$nmhi1
|
|---|
| 807 | fstws,mb ${fab0}L,-8($xfer) ; save high part
|
|---|
| 808 | add $ablo,$nmlo1,$nmlo1
|
|---|
| 809 | addc %r0,$nmhi1,$hi1
|
|---|
| 810 | fcpy,sgl %fr0,${fti}L ; zero high part
|
|---|
| 811 | fcpy,sgl %fr0,${fab0}L
|
|---|
| 812 | stw $nmlo1,-4($tp) ; tp[j-1]
|
|---|
| 813 |
|
|---|
| 814 | fcnvxf,dbl,dbl ${fti},${fti} ; 32-bit unsigned int -> double
|
|---|
| 815 | fcnvxf,dbl,dbl ${fab0},${fab0}
|
|---|
| 816 | add $hi1,$hi0,$hi0
|
|---|
| 817 | addc %r0,%r0,$hi1
|
|---|
| 818 | fadd,dbl ${fti},${fab0},${fab0} ; add tp[0]
|
|---|
| 819 | add $ti0,$hi0,$hi0
|
|---|
| 820 | addc %r0,$hi1,$hi1
|
|---|
| 821 | fcnvfx,dbl,dbl ${fab0},${fab0} ; double -> 33-bit unsigned int
|
|---|
| 822 | stw $hi0,0($tp)
|
|---|
| 823 | stw $hi1,4($tp)
|
|---|
| 824 | xmpyu ${fn0},${fab0}R,${fm0}
|
|---|
| 825 |
|
|---|
| 826 | b L\$outer_pa11
|
|---|
| 827 | ldo `$LOCALS+32+4`($fp),$tp
|
|---|
| 828 | |
|---|
| 829 |
|
|---|
| 830 | L\$outerdone_pa11
|
|---|
| 831 | add $hi0,$ablo,$ablo
|
|---|
| 832 | addc %r0,$abhi,$abhi
|
|---|
| 833 | add $ti1,$ablo,$ablo
|
|---|
| 834 | addc %r0,$abhi,$hi0
|
|---|
| 835 |
|
|---|
| 836 | ldw 4($tp),$ti0 ; tp[j]
|
|---|
| 837 |
|
|---|
| 838 | add $hi1,$nmlo1,$nmlo1
|
|---|
| 839 | addc %r0,$nmhi1,$nmhi1
|
|---|
| 840 | add $ablo,$nmlo1,$nmlo1
|
|---|
| 841 | addc %r0,$nmhi1,$hi1
|
|---|
| 842 | stw $nmlo1,-4($tp) ; tp[j-1]
|
|---|
| 843 |
|
|---|
| 844 | add $hi1,$hi0,$hi0
|
|---|
| 845 | addc %r0,%r0,$hi1
|
|---|
| 846 | add $ti0,$hi0,$hi0
|
|---|
| 847 | addc %r0,$hi1,$hi1
|
|---|
| 848 | stw $hi0,0($tp)
|
|---|
| 849 | stw $hi1,4($tp)
|
|---|
| 850 |
|
|---|
| 851 | ldo `$LOCALS+32+4`($fp),$tp
|
|---|
| 852 | sub %r0,%r0,%r0 ; clear borrow
|
|---|
| 853 | ldw -4($tp),$ti0
|
|---|
| 854 | addl $tp,$arrsz,$tp
|
|---|
| 855 | L\$sub_pa11
|
|---|
| 856 | ldwx $idx($np),$hi0
|
|---|
| 857 | subb $ti0,$hi0,$hi1
|
|---|
| 858 | ldwx $idx($tp),$ti0
|
|---|
| 859 | addib,<> 4,$idx,L\$sub_pa11
|
|---|
| 860 | stws,ma $hi1,4($rp)
|
|---|
| 861 |
|
|---|
| 862 | subb $ti0,%r0,$hi1
|
|---|
| 863 |
|
|---|
| 864 | ldo `$LOCALS+32`($fp),$tp
|
|---|
| 865 | sub $rp,$arrsz,$rp ; rewind rp
|
|---|
| 866 | subi 0,$arrsz,$idx
|
|---|
| 867 | L\$copy_pa11
|
|---|
| 868 | ldw 0($tp),$ti0
|
|---|
| 869 | ldw 0($rp),$hi0
|
|---|
| 870 | stws,ma %r0,4($tp)
|
|---|
| 871 | comiclr,= 0,$hi1,%r0
|
|---|
| 872 | copy $ti0,$hi0
|
|---|
| 873 | addib,<> 4,$idx,L\$copy_pa11
|
|---|
| 874 | stws,ma $hi0,4($rp)
|
|---|
| 875 |
|
|---|
| 876 | nop ; alignment
|
|---|
| 877 | L\$done
|
|---|
| 878 | ___
|
|---|
| 879 | }
|
|---|
| 880 | |
|---|
| 881 |
|
|---|
| 882 | $code.=<<___;
|
|---|
| 883 | ldi 1,%r28 ; signal "handled"
|
|---|
| 884 | ldo $FRAME($fp),%sp ; destroy tp[num+1]
|
|---|
| 885 |
|
|---|
| 886 | $POP `-$FRAME-$SAVED_RP`(%sp),%r2 ; standard epilogue
|
|---|
| 887 | $POP `-$FRAME+1*$SIZE_T`(%sp),%r4
|
|---|
| 888 | $POP `-$FRAME+2*$SIZE_T`(%sp),%r5
|
|---|
| 889 | $POP `-$FRAME+3*$SIZE_T`(%sp),%r6
|
|---|
| 890 | $POP `-$FRAME+4*$SIZE_T`(%sp),%r7
|
|---|
| 891 | $POP `-$FRAME+5*$SIZE_T`(%sp),%r8
|
|---|
| 892 | $POP `-$FRAME+6*$SIZE_T`(%sp),%r9
|
|---|
| 893 | $POP `-$FRAME+7*$SIZE_T`(%sp),%r10
|
|---|
| 894 | L\$abort
|
|---|
| 895 | bv (%r2)
|
|---|
| 896 | .EXIT
|
|---|
| 897 | $POPMB -$FRAME(%sp),%r3
|
|---|
| 898 | .PROCEND
|
|---|
| 899 | .STRINGZ "Montgomery Multiplication for PA-RISC, CRYPTOGAMS by <appro\@openssl.org>"
|
|---|
| 900 | ___
|
|---|
| 901 | |
|---|
| 902 |
|
|---|
| 903 | # Explicitly encode PA-RISC 2.0 instructions used in this module, so
|
|---|
| 904 | # that it can be compiled with .LEVEL 1.0. It should be noted that I
|
|---|
| 905 | # wouldn't have to do this, if GNU assembler understood .ALLOW 2.0
|
|---|
| 906 | # directive...
|
|---|
| 907 |
|
|---|
| 908 | my $ldd = sub {
|
|---|
| 909 | my ($mod,$args) = @_;
|
|---|
| 910 | my $orig = "ldd$mod\t$args";
|
|---|
| 911 |
|
|---|
| 912 | if ($args =~ /%r([0-9]+)\(%r([0-9]+)\),%r([0-9]+)/) # format 4
|
|---|
| 913 | { my $opcode=(0x03<<26)|($2<<21)|($1<<16)|(3<<6)|$3;
|
|---|
| 914 | sprintf "\t.WORD\t0x%08x\t; %s",$opcode,$orig;
|
|---|
| 915 | }
|
|---|
| 916 | elsif ($args =~ /(\-?[0-9]+)\(%r([0-9]+)\),%r([0-9]+)/) # format 5
|
|---|
| 917 | { my $opcode=(0x03<<26)|($2<<21)|(1<<12)|(3<<6)|$3;
|
|---|
| 918 | $opcode|=(($1&0xF)<<17)|(($1&0x10)<<12); # encode offset
|
|---|
| 919 | $opcode|=(1<<5) if ($mod =~ /^,m/);
|
|---|
| 920 | $opcode|=(1<<13) if ($mod =~ /^,mb/);
|
|---|
| 921 | sprintf "\t.WORD\t0x%08x\t; %s",$opcode,$orig;
|
|---|
| 922 | }
|
|---|
| 923 | else { "\t".$orig; }
|
|---|
| 924 | };
|
|---|
| 925 |
|
|---|
| 926 | my $std = sub {
|
|---|
| 927 | my ($mod,$args) = @_;
|
|---|
| 928 | my $orig = "std$mod\t$args";
|
|---|
| 929 |
|
|---|
| 930 | if ($args =~ /%r([0-9]+),(\-?[0-9]+)\(%r([0-9]+)\)/) # format 6
|
|---|
| 931 | { my $opcode=(0x03<<26)|($3<<21)|($1<<16)|(1<<12)|(0xB<<6);
|
|---|
| 932 | $opcode|=(($2&0xF)<<1)|(($2&0x10)>>4); # encode offset
|
|---|
| 933 | $opcode|=(1<<5) if ($mod =~ /^,m/);
|
|---|
| 934 | $opcode|=(1<<13) if ($mod =~ /^,mb/);
|
|---|
| 935 | sprintf "\t.WORD\t0x%08x\t; %s",$opcode,$orig;
|
|---|
| 936 | }
|
|---|
| 937 | else { "\t".$orig; }
|
|---|
| 938 | };
|
|---|
| 939 |
|
|---|
| 940 | my $extrd = sub {
|
|---|
| 941 | my ($mod,$args) = @_;
|
|---|
| 942 | my $orig = "extrd$mod\t$args";
|
|---|
| 943 |
|
|---|
| 944 | # I only have ",u" completer, it's implicitly encoded...
|
|---|
| 945 | if ($args =~ /%r([0-9]+),([0-9]+),([0-9]+),%r([0-9]+)/) # format 15
|
|---|
| 946 | { my $opcode=(0x36<<26)|($1<<21)|($4<<16);
|
|---|
| 947 | my $len=32-$3;
|
|---|
| 948 | $opcode |= (($2&0x20)<<6)|(($2&0x1f)<<5); # encode pos
|
|---|
| 949 | $opcode |= (($len&0x20)<<7)|($len&0x1f); # encode len
|
|---|
| 950 | sprintf "\t.WORD\t0x%08x\t; %s",$opcode,$orig;
|
|---|
| 951 | }
|
|---|
| 952 | elsif ($args =~ /%r([0-9]+),%sar,([0-9]+),%r([0-9]+)/) # format 12
|
|---|
| 953 | { my $opcode=(0x34<<26)|($1<<21)|($3<<16)|(2<<11)|(1<<9);
|
|---|
| 954 | my $len=32-$2;
|
|---|
| 955 | $opcode |= (($len&0x20)<<3)|($len&0x1f); # encode len
|
|---|
| 956 | $opcode |= (1<<13) if ($mod =~ /,\**=/);
|
|---|
| 957 | sprintf "\t.WORD\t0x%08x\t; %s",$opcode,$orig;
|
|---|
| 958 | }
|
|---|
| 959 | else { "\t".$orig; }
|
|---|
| 960 | };
|
|---|
| 961 |
|
|---|
| 962 | my $shrpd = sub {
|
|---|
| 963 | my ($mod,$args) = @_;
|
|---|
| 964 | my $orig = "shrpd$mod\t$args";
|
|---|
| 965 |
|
|---|
| 966 | if ($args =~ /%r([0-9]+),%r([0-9]+),([0-9]+),%r([0-9]+)/) # format 14
|
|---|
| 967 | { my $opcode=(0x34<<26)|($2<<21)|($1<<16)|(1<<10)|$4;
|
|---|
| 968 | my $cpos=63-$3;
|
|---|
| 969 | $opcode |= (($cpos&0x20)<<6)|(($cpos&0x1f)<<5); # encode sa
|
|---|
| 970 | sprintf "\t.WORD\t0x%08x\t; %s",$opcode,$orig;
|
|---|
| 971 | }
|
|---|
| 972 | else { "\t".$orig; }
|
|---|
| 973 | };
|
|---|
| 974 |
|
|---|
| 975 | my $sub = sub {
|
|---|
| 976 | my ($mod,$args) = @_;
|
|---|
| 977 | my $orig = "sub$mod\t$args";
|
|---|
| 978 |
|
|---|
| 979 | if ($mod eq ",db" && $args =~ /%r([0-9]+),%r([0-9]+),%r([0-9]+)/) {
|
|---|
| 980 | my $opcode=(0x02<<26)|($2<<21)|($1<<16)|$3;
|
|---|
| 981 | $opcode|=(1<<10); # e1
|
|---|
| 982 | $opcode|=(1<<8); # e2
|
|---|
| 983 | $opcode|=(1<<5); # d
|
|---|
| 984 | sprintf "\t.WORD\t0x%08x\t; %s",$opcode,$orig
|
|---|
| 985 | }
|
|---|
| 986 | else { "\t".$orig; }
|
|---|
| 987 | };
|
|---|
| 988 |
|
|---|
| 989 | sub assemble {
|
|---|
| 990 | my ($mnemonic,$mod,$args)=@_;
|
|---|
| 991 | my $opcode = eval("\$$mnemonic");
|
|---|
| 992 |
|
|---|
| 993 | ref($opcode) eq 'CODE' ? &$opcode($mod,$args) : "\t$mnemonic$mod\t$args";
|
|---|
| 994 | }
|
|---|
| 995 |
|
|---|
| 996 | if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
|
|---|
| 997 | =~ /GNU assembler/) {
|
|---|
| 998 | $gnuas = 1;
|
|---|
| 999 | }
|
|---|
| 1000 |
|
|---|
| 1001 | foreach (split("\n",$code)) {
|
|---|
| 1002 | s/\`([^\`]*)\`/eval $1/ge;
|
|---|
| 1003 | # flip word order in 64-bit mode...
|
|---|
| 1004 | s/(xmpyu\s+)($fai|$fni)([LR])/$1.$2.($3 eq "L"?"R":"L")/e if ($BN_SZ==8);
|
|---|
| 1005 | # assemble 2.0 instructions in 32-bit mode...
|
|---|
| 1006 | s/^\s+([a-z]+)([\S]*)\s+([\S]*)/&assemble($1,$2,$3)/e if ($BN_SZ==4);
|
|---|
| 1007 |
|
|---|
| 1008 | s/(\.LEVEL\s+2\.0)W/$1w/ if ($gnuas && $SIZE_T==8);
|
|---|
| 1009 | s/\.SPACE\s+\$TEXT\$/.text/ if ($gnuas && $SIZE_T==8);
|
|---|
| 1010 | s/\.SUBSPA.*// if ($gnuas && $SIZE_T==8);
|
|---|
| 1011 | s/\bbv\b/bve/ if ($SIZE_T==8);
|
|---|
| 1012 |
|
|---|
| 1013 | print $_,"\n";
|
|---|
| 1014 | }
|
|---|
| 1015 | close STDOUT or die "error closing STDOUT: $!";
|
|---|
