PA-RISC assembler pack update from HEAD.
authorAndy Polyakov <appro@openssl.org>
Mon, 14 Nov 2011 20:50:15 +0000 (20:50 +0000)
committerAndy Polyakov <appro@openssl.org>
Mon, 14 Nov 2011 20:50:15 +0000 (20:50 +0000)
crypto/aes/asm/aes-parisc.pl [new file with mode: 0644]
crypto/bn/asm/parisc-mont.pl [new file with mode: 0644]
crypto/pariscid.pl [new file with mode: 0644]
crypto/rc4/asm/rc4-parisc.pl [new file with mode: 0644]
crypto/sha/asm/sha1-parisc.pl [new file with mode: 0644]
crypto/sha/asm/sha512-parisc.pl [new file with mode: 0755]

diff --git a/crypto/aes/asm/aes-parisc.pl b/crypto/aes/asm/aes-parisc.pl
new file mode 100644 (file)
index 0000000..c36b6a2
--- /dev/null
@@ -0,0 +1,1021 @@
+#!/usr/bin/env perl
+
+# ====================================================================
+# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+
+# AES for PA-RISC.
+#
+# June 2009.
+#
+# The module is mechanical transliteration of aes-sparcv9.pl, but with
+# a twist: S-boxes are compressed even further down to 1K+256B. On
+# PA-7100LC performance is ~40% better than gcc 3.2 generated code and
+# is about 33 cycles per byte processed with 128-bit key. Newer CPUs
+# perform at 16 cycles per byte. It's not faster than code generated
+# by vendor compiler, but recall that it has compressed S-boxes, which
+# requires extra processing.
+#
+# Special thanks to polarhome.com for providing HP-UX account.
+
+$flavour = shift;
+$output = shift;
+open STDOUT,">$output";
+
+if ($flavour =~ /64/) {
+       $LEVEL          ="2.0W";
+       $SIZE_T         =8;
+       $FRAME_MARKER   =80;
+       $SAVED_RP       =16;
+       $PUSH           ="std";
+       $PUSHMA         ="std,ma";
+       $POP            ="ldd";
+       $POPMB          ="ldd,mb";
+} else {
+       $LEVEL          ="1.0";
+       $SIZE_T         =4;
+       $FRAME_MARKER   =48;
+       $SAVED_RP       =20;
+       $PUSH           ="stw";
+       $PUSHMA         ="stwm";
+       $POP            ="ldw";
+       $POPMB          ="ldwm";
+}
+
+$FRAME=16*$SIZE_T+$FRAME_MARKER;# 16 saved regs + frame marker
+                               #                 [+ argument transfer]
+$inp="%r26";   # arg0
+$out="%r25";   # arg1
+$key="%r24";   # arg2
+
+($s0,$s1,$s2,$s3) = ("%r1","%r2","%r3","%r4");
+($t0,$t1,$t2,$t3) = ("%r5","%r6","%r7","%r8");
+
+($acc0, $acc1, $acc2, $acc3, $acc4, $acc5, $acc6, $acc7,
+ $acc8, $acc9,$acc10,$acc11,$acc12,$acc13,$acc14,$acc15) =
+("%r9","%r10","%r11","%r12","%r13","%r14","%r15","%r16",
+"%r17","%r18","%r19","%r20","%r21","%r22","%r23","%r26");
+
+$tbl="%r28";
+$rounds="%r29";
+
+$code=<<___;
+       .LEVEL  $LEVEL
+       .SPACE  \$TEXT\$
+       .SUBSPA \$CODE\$,QUAD=0,ALIGN=8,ACCESS=0x2C,CODE_ONLY
+
+       .EXPORT AES_encrypt,ENTRY,ARGW0=GR,ARGW1=GR,ARGW2=GR
+       .ALIGN  64
+AES_encrypt
+       .PROC
+       .CALLINFO       FRAME=`$FRAME-16*$SIZE_T`,NO_CALLS,SAVE_RP,ENTRY_GR=18
+       .ENTRY
+       $PUSH   %r2,-$SAVED_RP(%sp)     ; standard prologue
+       $PUSHMA %r3,$FRAME(%sp)
+       $PUSH   %r4,`-$FRAME+1*$SIZE_T`(%sp)
+       $PUSH   %r5,`-$FRAME+2*$SIZE_T`(%sp)
+       $PUSH   %r6,`-$FRAME+3*$SIZE_T`(%sp)
+       $PUSH   %r7,`-$FRAME+4*$SIZE_T`(%sp)
+       $PUSH   %r8,`-$FRAME+5*$SIZE_T`(%sp)
+       $PUSH   %r9,`-$FRAME+6*$SIZE_T`(%sp)
+       $PUSH   %r10,`-$FRAME+7*$SIZE_T`(%sp)
+       $PUSH   %r11,`-$FRAME+8*$SIZE_T`(%sp)
+       $PUSH   %r12,`-$FRAME+9*$SIZE_T`(%sp)
+       $PUSH   %r13,`-$FRAME+10*$SIZE_T`(%sp)
+       $PUSH   %r14,`-$FRAME+11*$SIZE_T`(%sp)
+       $PUSH   %r15,`-$FRAME+12*$SIZE_T`(%sp)
+       $PUSH   %r16,`-$FRAME+13*$SIZE_T`(%sp)
+       $PUSH   %r17,`-$FRAME+14*$SIZE_T`(%sp)
+       $PUSH   %r18,`-$FRAME+15*$SIZE_T`(%sp)
+
+       blr     %r0,$tbl
+       ldi     3,$t0
+L\$enc_pic
+       andcm   $tbl,$t0,$tbl
+       ldo     L\$AES_Te-L\$enc_pic($tbl),$tbl
+
+       and     $inp,$t0,$t0
+       sub     $inp,$t0,$inp
+       ldw     0($inp),$s0
+       ldw     4($inp),$s1
+       ldw     8($inp),$s2
+       comib,= 0,$t0,L\$enc_inp_aligned
+       ldw     12($inp),$s3
+
+       sh3addl $t0,%r0,$t0
+       subi    32,$t0,$t0
+       mtctl   $t0,%cr11
+       ldw     16($inp),$t1
+       vshd    $s0,$s1,$s0
+       vshd    $s1,$s2,$s1
+       vshd    $s2,$s3,$s2
+       vshd    $s3,$t1,$s3
+
+L\$enc_inp_aligned
+       bl      _parisc_AES_encrypt,%r31
+       nop
+
+       extru,<> $out,31,2,%r0
+       b       L\$enc_out_aligned
+       nop
+
+       _srm    $s0,24,$acc0
+       _srm    $s0,16,$acc1
+       stb     $acc0,0($out)
+       _srm    $s0,8,$acc2
+       stb     $acc1,1($out)
+       _srm    $s1,24,$acc4
+       stb     $acc2,2($out)
+       _srm    $s1,16,$acc5
+       stb     $s0,3($out)
+       _srm    $s1,8,$acc6
+       stb     $acc4,4($out)
+       _srm    $s2,24,$acc0
+       stb     $acc5,5($out)
+       _srm    $s2,16,$acc1
+       stb     $acc6,6($out)
+       _srm    $s2,8,$acc2
+       stb     $s1,7($out)
+       _srm    $s3,24,$acc4
+       stb     $acc0,8($out)
+       _srm    $s3,16,$acc5
+       stb     $acc1,9($out)
+       _srm    $s3,8,$acc6
+       stb     $acc2,10($out)
+       stb     $s2,11($out)
+       stb     $acc4,12($out)
+       stb     $acc5,13($out)
+       stb     $acc6,14($out)
+       b       L\$enc_done
+       stb     $s3,15($out)
+
+L\$enc_out_aligned
+       stw     $s0,0($out)
+       stw     $s1,4($out)
+       stw     $s2,8($out)
+       stw     $s3,12($out)
+
+L\$enc_done
+       $POP    `-$FRAME-$SAVED_RP`(%sp),%r2    ; standard epilogue
+       $POP    `-$FRAME+1*$SIZE_T`(%sp),%r4
+       $POP    `-$FRAME+2*$SIZE_T`(%sp),%r5
+       $POP    `-$FRAME+3*$SIZE_T`(%sp),%r6
+       $POP    `-$FRAME+4*$SIZE_T`(%sp),%r7
+       $POP    `-$FRAME+5*$SIZE_T`(%sp),%r8
+       $POP    `-$FRAME+6*$SIZE_T`(%sp),%r9
+       $POP    `-$FRAME+7*$SIZE_T`(%sp),%r10
+       $POP    `-$FRAME+8*$SIZE_T`(%sp),%r11
+       $POP    `-$FRAME+9*$SIZE_T`(%sp),%r12
+       $POP    `-$FRAME+10*$SIZE_T`(%sp),%r13
+       $POP    `-$FRAME+11*$SIZE_T`(%sp),%r14
+       $POP    `-$FRAME+12*$SIZE_T`(%sp),%r15
+       $POP    `-$FRAME+13*$SIZE_T`(%sp),%r16
+       $POP    `-$FRAME+14*$SIZE_T`(%sp),%r17
+       $POP    `-$FRAME+15*$SIZE_T`(%sp),%r18
+       bv      (%r2)
+       .EXIT
+       $POPMB  -$FRAME(%sp),%r3
+       .PROCEND
+
+       .ALIGN  16
+_parisc_AES_encrypt
+       .PROC
+       .CALLINFO       MILLICODE
+       .ENTRY
+       ldw     240($key),$rounds
+       ldw     0($key),$t0
+       ldw     4($key),$t1
+       ldw     8($key),$t2
+       _srm    $rounds,1,$rounds
+       xor     $t0,$s0,$s0
+       ldw     12($key),$t3
+       _srm    $s0,24,$acc0
+       xor     $t1,$s1,$s1
+       ldw     16($key),$t0
+       _srm    $s1,16,$acc1
+       xor     $t2,$s2,$s2
+       ldw     20($key),$t1
+       xor     $t3,$s3,$s3
+       ldw     24($key),$t2
+       ldw     28($key),$t3
+L\$enc_loop
+       _srm    $s2,8,$acc2
+       ldwx,s  $acc0($tbl),$acc0
+       _srm    $s3,0,$acc3
+       ldwx,s  $acc1($tbl),$acc1
+       _srm    $s1,24,$acc4
+       ldwx,s  $acc2($tbl),$acc2
+       _srm    $s2,16,$acc5
+       ldwx,s  $acc3($tbl),$acc3
+       _srm    $s3,8,$acc6
+       ldwx,s  $acc4($tbl),$acc4
+       _srm    $s0,0,$acc7
+       ldwx,s  $acc5($tbl),$acc5
+       _srm    $s2,24,$acc8
+       ldwx,s  $acc6($tbl),$acc6
+       _srm    $s3,16,$acc9
+       ldwx,s  $acc7($tbl),$acc7
+       _srm    $s0,8,$acc10
+       ldwx,s  $acc8($tbl),$acc8
+       _srm    $s1,0,$acc11
+       ldwx,s  $acc9($tbl),$acc9
+       _srm    $s3,24,$acc12
+       ldwx,s  $acc10($tbl),$acc10
+       _srm    $s0,16,$acc13
+       ldwx,s  $acc11($tbl),$acc11
+       _srm    $s1,8,$acc14
+       ldwx,s  $acc12($tbl),$acc12
+       _srm    $s2,0,$acc15
+       ldwx,s  $acc13($tbl),$acc13
+       ldwx,s  $acc14($tbl),$acc14
+       ldwx,s  $acc15($tbl),$acc15
+       addib,= -1,$rounds,L\$enc_last
+       ldo     32($key),$key
+
+               _ror    $acc1,8,$acc1
+               xor     $acc0,$t0,$t0
+       ldw     0($key),$s0
+               _ror    $acc2,16,$acc2
+               xor     $acc1,$t0,$t0
+       ldw     4($key),$s1
+               _ror    $acc3,24,$acc3
+               xor     $acc2,$t0,$t0
+       ldw     8($key),$s2
+               _ror    $acc5,8,$acc5
+               xor     $acc3,$t0,$t0
+       ldw     12($key),$s3
+               _ror    $acc6,16,$acc6
+               xor     $acc4,$t1,$t1
+               _ror    $acc7,24,$acc7
+               xor     $acc5,$t1,$t1
+               _ror    $acc9,8,$acc9
+               xor     $acc6,$t1,$t1
+               _ror    $acc10,16,$acc10
+               xor     $acc7,$t1,$t1
+               _ror    $acc11,24,$acc11
+               xor     $acc8,$t2,$t2
+               _ror    $acc13,8,$acc13
+               xor     $acc9,$t2,$t2
+               _ror    $acc14,16,$acc14
+               xor     $acc10,$t2,$t2
+               _ror    $acc15,24,$acc15
+               xor     $acc11,$t2,$t2
+               xor     $acc12,$acc14,$acc14
+               xor     $acc13,$t3,$t3
+       _srm    $t0,24,$acc0
+               xor     $acc14,$t3,$t3
+       _srm    $t1,16,$acc1
+               xor     $acc15,$t3,$t3
+
+       _srm    $t2,8,$acc2
+       ldwx,s  $acc0($tbl),$acc0
+       _srm    $t3,0,$acc3
+       ldwx,s  $acc1($tbl),$acc1
+       _srm    $t1,24,$acc4
+       ldwx,s  $acc2($tbl),$acc2
+       _srm    $t2,16,$acc5
+       ldwx,s  $acc3($tbl),$acc3
+       _srm    $t3,8,$acc6
+       ldwx,s  $acc4($tbl),$acc4
+       _srm    $t0,0,$acc7
+       ldwx,s  $acc5($tbl),$acc5
+       _srm    $t2,24,$acc8
+       ldwx,s  $acc6($tbl),$acc6
+       _srm    $t3,16,$acc9
+       ldwx,s  $acc7($tbl),$acc7
+       _srm    $t0,8,$acc10
+       ldwx,s  $acc8($tbl),$acc8
+       _srm    $t1,0,$acc11
+       ldwx,s  $acc9($tbl),$acc9
+       _srm    $t3,24,$acc12
+       ldwx,s  $acc10($tbl),$acc10
+       _srm    $t0,16,$acc13
+       ldwx,s  $acc11($tbl),$acc11
+       _srm    $t1,8,$acc14
+       ldwx,s  $acc12($tbl),$acc12
+       _srm    $t2,0,$acc15
+       ldwx,s  $acc13($tbl),$acc13
+               _ror    $acc1,8,$acc1
+       ldwx,s  $acc14($tbl),$acc14
+
+               _ror    $acc2,16,$acc2
+               xor     $acc0,$s0,$s0
+       ldwx,s  $acc15($tbl),$acc15
+               _ror    $acc3,24,$acc3
+               xor     $acc1,$s0,$s0
+       ldw     16($key),$t0
+               _ror    $acc5,8,$acc5
+               xor     $acc2,$s0,$s0
+       ldw     20($key),$t1
+               _ror    $acc6,16,$acc6
+               xor     $acc3,$s0,$s0
+       ldw     24($key),$t2
+               _ror    $acc7,24,$acc7
+               xor     $acc4,$s1,$s1
+       ldw     28($key),$t3
+               _ror    $acc9,8,$acc9
+               xor     $acc5,$s1,$s1
+       ldw     1024+0($tbl),%r0                ; prefetch te4
+               _ror    $acc10,16,$acc10
+               xor     $acc6,$s1,$s1
+       ldw     1024+32($tbl),%r0               ; prefetch te4
+               _ror    $acc11,24,$acc11
+               xor     $acc7,$s1,$s1
+       ldw     1024+64($tbl),%r0               ; prefetch te4
+               _ror    $acc13,8,$acc13
+               xor     $acc8,$s2,$s2
+       ldw     1024+96($tbl),%r0               ; prefetch te4
+               _ror    $acc14,16,$acc14
+               xor     $acc9,$s2,$s2
+       ldw     1024+128($tbl),%r0              ; prefetch te4
+               _ror    $acc15,24,$acc15
+               xor     $acc10,$s2,$s2
+       ldw     1024+160($tbl),%r0              ; prefetch te4
+       _srm    $s0,24,$acc0
+               xor     $acc11,$s2,$s2
+       ldw     1024+192($tbl),%r0              ; prefetch te4
+               xor     $acc12,$acc14,$acc14
+               xor     $acc13,$s3,$s3
+       ldw     1024+224($tbl),%r0              ; prefetch te4
+       _srm    $s1,16,$acc1
+               xor     $acc14,$s3,$s3
+       b       L\$enc_loop
+               xor     $acc15,$s3,$s3
+
+       .ALIGN  16
+L\$enc_last
+       ldo     1024($tbl),$rounds
+               _ror    $acc1,8,$acc1
+               xor     $acc0,$t0,$t0
+       ldw     0($key),$s0
+               _ror    $acc2,16,$acc2
+               xor     $acc1,$t0,$t0
+       ldw     4($key),$s1
+               _ror    $acc3,24,$acc3
+               xor     $acc2,$t0,$t0
+       ldw     8($key),$s2
+               _ror    $acc5,8,$acc5
+               xor     $acc3,$t0,$t0
+       ldw     12($key),$s3
+               _ror    $acc6,16,$acc6
+               xor     $acc4,$t1,$t1
+               _ror    $acc7,24,$acc7
+               xor     $acc5,$t1,$t1
+               _ror    $acc9,8,$acc9
+               xor     $acc6,$t1,$t1
+               _ror    $acc10,16,$acc10
+               xor     $acc7,$t1,$t1
+               _ror    $acc11,24,$acc11
+               xor     $acc8,$t2,$t2
+               _ror    $acc13,8,$acc13
+               xor     $acc9,$t2,$t2
+               _ror    $acc14,16,$acc14
+               xor     $acc10,$t2,$t2
+               _ror    $acc15,24,$acc15
+               xor     $acc11,$t2,$t2
+               xor     $acc12,$acc14,$acc14
+               xor     $acc13,$t3,$t3
+       _srm    $t0,24,$acc0
+               xor     $acc14,$t3,$t3
+       _srm    $t1,16,$acc1
+               xor     $acc15,$t3,$t3
+
+       _srm    $t2,8,$acc2
+       ldbx    $acc0($rounds),$acc0
+       _srm    $t1,24,$acc4
+       ldbx    $acc1($rounds),$acc1
+       _srm    $t2,16,$acc5
+       _srm    $t3,0,$acc3
+       ldbx    $acc2($rounds),$acc2
+       ldbx    $acc3($rounds),$acc3
+       _srm    $t3,8,$acc6
+       ldbx    $acc4($rounds),$acc4
+       _srm    $t2,24,$acc8
+       ldbx    $acc5($rounds),$acc5
+       _srm    $t3,16,$acc9
+       _srm    $t0,0,$acc7
+       ldbx    $acc6($rounds),$acc6
+       ldbx    $acc7($rounds),$acc7
+       _srm    $t0,8,$acc10
+       ldbx    $acc8($rounds),$acc8
+       _srm    $t3,24,$acc12
+       ldbx    $acc9($rounds),$acc9
+       _srm    $t0,16,$acc13
+       _srm    $t1,0,$acc11
+       ldbx    $acc10($rounds),$acc10
+       _srm    $t1,8,$acc14
+       ldbx    $acc11($rounds),$acc11
+       ldbx    $acc12($rounds),$acc12
+       ldbx    $acc13($rounds),$acc13
+       _srm    $t2,0,$acc15
+       ldbx    $acc14($rounds),$acc14
+
+               dep     $acc0,7,8,$acc3
+       ldbx    $acc15($rounds),$acc15
+               dep     $acc4,7,8,$acc7
+               dep     $acc1,15,8,$acc3
+               dep     $acc5,15,8,$acc7
+               dep     $acc2,23,8,$acc3
+               dep     $acc6,23,8,$acc7
+               xor     $acc3,$s0,$s0
+               xor     $acc7,$s1,$s1
+               dep     $acc8,7,8,$acc11
+               dep     $acc12,7,8,$acc15
+               dep     $acc9,15,8,$acc11
+               dep     $acc13,15,8,$acc15
+               dep     $acc10,23,8,$acc11
+               dep     $acc14,23,8,$acc15
+               xor     $acc11,$s2,$s2
+
+       bv      (%r31)
+       .EXIT
+               xor     $acc15,$s3,$s3
+       .PROCEND
+
+       .ALIGN  64
+L\$AES_Te
+       .WORD   0xc66363a5, 0xf87c7c84, 0xee777799, 0xf67b7b8d
+       .WORD   0xfff2f20d, 0xd66b6bbd, 0xde6f6fb1, 0x91c5c554
+       .WORD   0x60303050, 0x02010103, 0xce6767a9, 0x562b2b7d
+       .WORD   0xe7fefe19, 0xb5d7d762, 0x4dababe6, 0xec76769a
+       .WORD   0x8fcaca45, 0x1f82829d, 0x89c9c940, 0xfa7d7d87
+       .WORD   0xeffafa15, 0xb25959eb, 0x8e4747c9, 0xfbf0f00b
+       .WORD   0x41adadec, 0xb3d4d467, 0x5fa2a2fd, 0x45afafea
+       .WORD   0x239c9cbf, 0x53a4a4f7, 0xe4727296, 0x9bc0c05b
+       .WORD   0x75b7b7c2, 0xe1fdfd1c, 0x3d9393ae, 0x4c26266a
+       .WORD   0x6c36365a, 0x7e3f3f41, 0xf5f7f702, 0x83cccc4f
+       .WORD   0x6834345c, 0x51a5a5f4, 0xd1e5e534, 0xf9f1f108
+       .WORD   0xe2717193, 0xabd8d873, 0x62313153, 0x2a15153f
+       .WORD   0x0804040c, 0x95c7c752, 0x46232365, 0x9dc3c35e
+       .WORD   0x30181828, 0x379696a1, 0x0a05050f, 0x2f9a9ab5
+       .WORD   0x0e070709, 0x24121236, 0x1b80809b, 0xdfe2e23d
+       .WORD   0xcdebeb26, 0x4e272769, 0x7fb2b2cd, 0xea75759f
+       .WORD   0x1209091b, 0x1d83839e, 0x582c2c74, 0x341a1a2e
+       .WORD   0x361b1b2d, 0xdc6e6eb2, 0xb45a5aee, 0x5ba0a0fb
+       .WORD   0xa45252f6, 0x763b3b4d, 0xb7d6d661, 0x7db3b3ce
+       .WORD   0x5229297b, 0xdde3e33e, 0x5e2f2f71, 0x13848497
+       .WORD   0xa65353f5, 0xb9d1d168, 0x00000000, 0xc1eded2c
+       .WORD   0x40202060, 0xe3fcfc1f, 0x79b1b1c8, 0xb65b5bed
+       .WORD   0xd46a6abe, 0x8dcbcb46, 0x67bebed9, 0x7239394b
+       .WORD   0x944a4ade, 0x984c4cd4, 0xb05858e8, 0x85cfcf4a
+       .WORD   0xbbd0d06b, 0xc5efef2a, 0x4faaaae5, 0xedfbfb16
+       .WORD   0x864343c5, 0x9a4d4dd7, 0x66333355, 0x11858594
+       .WORD   0x8a4545cf, 0xe9f9f910, 0x04020206, 0xfe7f7f81
+       .WORD   0xa05050f0, 0x783c3c44, 0x259f9fba, 0x4ba8a8e3
+       .WORD   0xa25151f3, 0x5da3a3fe, 0x804040c0, 0x058f8f8a
+       .WORD   0x3f9292ad, 0x219d9dbc, 0x70383848, 0xf1f5f504
+       .WORD   0x63bcbcdf, 0x77b6b6c1, 0xafdada75, 0x42212163
+       .WORD   0x20101030, 0xe5ffff1a, 0xfdf3f30e, 0xbfd2d26d
+       .WORD   0x81cdcd4c, 0x180c0c14, 0x26131335, 0xc3ecec2f
+       .WORD   0xbe5f5fe1, 0x359797a2, 0x884444cc, 0x2e171739
+       .WORD   0x93c4c457, 0x55a7a7f2, 0xfc7e7e82, 0x7a3d3d47
+       .WORD   0xc86464ac, 0xba5d5de7, 0x3219192b, 0xe6737395
+       .WORD   0xc06060a0, 0x19818198, 0x9e4f4fd1, 0xa3dcdc7f
+       .WORD   0x44222266, 0x542a2a7e, 0x3b9090ab, 0x0b888883
+       .WORD   0x8c4646ca, 0xc7eeee29, 0x6bb8b8d3, 0x2814143c
+       .WORD   0xa7dede79, 0xbc5e5ee2, 0x160b0b1d, 0xaddbdb76
+       .WORD   0xdbe0e03b, 0x64323256, 0x743a3a4e, 0x140a0a1e
+       .WORD   0x924949db, 0x0c06060a, 0x4824246c, 0xb85c5ce4
+       .WORD   0x9fc2c25d, 0xbdd3d36e, 0x43acacef, 0xc46262a6
+       .WORD   0x399191a8, 0x319595a4, 0xd3e4e437, 0xf279798b
+       .WORD   0xd5e7e732, 0x8bc8c843, 0x6e373759, 0xda6d6db7
+       .WORD   0x018d8d8c, 0xb1d5d564, 0x9c4e4ed2, 0x49a9a9e0
+       .WORD   0xd86c6cb4, 0xac5656fa, 0xf3f4f407, 0xcfeaea25
+       .WORD   0xca6565af, 0xf47a7a8e, 0x47aeaee9, 0x10080818
+       .WORD   0x6fbabad5, 0xf0787888, 0x4a25256f, 0x5c2e2e72
+       .WORD   0x381c1c24, 0x57a6a6f1, 0x73b4b4c7, 0x97c6c651
+       .WORD   0xcbe8e823, 0xa1dddd7c, 0xe874749c, 0x3e1f1f21
+       .WORD   0x964b4bdd, 0x61bdbddc, 0x0d8b8b86, 0x0f8a8a85
+       .WORD   0xe0707090, 0x7c3e3e42, 0x71b5b5c4, 0xcc6666aa
+       .WORD   0x904848d8, 0x06030305, 0xf7f6f601, 0x1c0e0e12
+       .WORD   0xc26161a3, 0x6a35355f, 0xae5757f9, 0x69b9b9d0
+       .WORD   0x17868691, 0x99c1c158, 0x3a1d1d27, 0x279e9eb9
+       .WORD   0xd9e1e138, 0xebf8f813, 0x2b9898b3, 0x22111133
+       .WORD   0xd26969bb, 0xa9d9d970, 0x078e8e89, 0x339494a7
+       .WORD   0x2d9b9bb6, 0x3c1e1e22, 0x15878792, 0xc9e9e920
+       .WORD   0x87cece49, 0xaa5555ff, 0x50282878, 0xa5dfdf7a
+       .WORD   0x038c8c8f, 0x59a1a1f8, 0x09898980, 0x1a0d0d17
+       .WORD   0x65bfbfda, 0xd7e6e631, 0x844242c6, 0xd06868b8
+       .WORD   0x824141c3, 0x299999b0, 0x5a2d2d77, 0x1e0f0f11
+       .WORD   0x7bb0b0cb, 0xa85454fc, 0x6dbbbbd6, 0x2c16163a
+       .BYTE   0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5
+       .BYTE   0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76
+       .BYTE   0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0
+       .BYTE   0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0
+       .BYTE   0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc
+       .BYTE   0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15
+       .BYTE   0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a
+       .BYTE   0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75
+       .BYTE   0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0
+       .BYTE   0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84
+       .BYTE   0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b
+       .BYTE   0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf
+       .BYTE   0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85
+       .BYTE   0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8
+       .BYTE   0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5
+       .BYTE   0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2
+       .BYTE   0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17
+       .BYTE   0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73
+       .BYTE   0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88
+       .BYTE   0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb
+       .BYTE   0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c
+       .BYTE   0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79
+       .BYTE   0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9
+       .BYTE   0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08
+       .BYTE   0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6
+       .BYTE   0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a
+       .BYTE   0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e
+       .BYTE   0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e
+       .BYTE   0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94
+       .BYTE   0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf
+       .BYTE   0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68
+       .BYTE   0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16
+___
+
+$code.=<<___;
+       .EXPORT AES_decrypt,ENTRY,ARGW0=GR,ARGW1=GR,ARGW2=GR
+       .ALIGN  16
+AES_decrypt
+       .PROC
+       .CALLINFO       FRAME=`$FRAME-16*$SIZE_T`,NO_CALLS,SAVE_RP,ENTRY_GR=18
+       .ENTRY
+       $PUSH   %r2,-$SAVED_RP(%sp)     ; standard prologue
+       $PUSHMA %r3,$FRAME(%sp)
+       $PUSH   %r4,`-$FRAME+1*$SIZE_T`(%sp)
+       $PUSH   %r5,`-$FRAME+2*$SIZE_T`(%sp)
+       $PUSH   %r6,`-$FRAME+3*$SIZE_T`(%sp)
+       $PUSH   %r7,`-$FRAME+4*$SIZE_T`(%sp)
+       $PUSH   %r8,`-$FRAME+5*$SIZE_T`(%sp)
+       $PUSH   %r9,`-$FRAME+6*$SIZE_T`(%sp)
+       $PUSH   %r10,`-$FRAME+7*$SIZE_T`(%sp)
+       $PUSH   %r11,`-$FRAME+8*$SIZE_T`(%sp)
+       $PUSH   %r12,`-$FRAME+9*$SIZE_T`(%sp)
+       $PUSH   %r13,`-$FRAME+10*$SIZE_T`(%sp)
+       $PUSH   %r14,`-$FRAME+11*$SIZE_T`(%sp)
+       $PUSH   %r15,`-$FRAME+12*$SIZE_T`(%sp)
+       $PUSH   %r16,`-$FRAME+13*$SIZE_T`(%sp)
+       $PUSH   %r17,`-$FRAME+14*$SIZE_T`(%sp)
+       $PUSH   %r18,`-$FRAME+15*$SIZE_T`(%sp)
+
+       blr     %r0,$tbl
+       ldi     3,$t0
+L\$dec_pic
+       andcm   $tbl,$t0,$tbl
+       ldo     L\$AES_Td-L\$dec_pic($tbl),$tbl
+
+       and     $inp,$t0,$t0
+       sub     $inp,$t0,$inp
+       ldw     0($inp),$s0
+       ldw     4($inp),$s1
+       ldw     8($inp),$s2
+       comib,= 0,$t0,L\$dec_inp_aligned
+       ldw     12($inp),$s3
+
+       sh3addl $t0,%r0,$t0
+       subi    32,$t0,$t0
+       mtctl   $t0,%cr11
+       ldw     16($inp),$t1
+       vshd    $s0,$s1,$s0
+       vshd    $s1,$s2,$s1
+       vshd    $s2,$s3,$s2
+       vshd    $s3,$t1,$s3
+
+L\$dec_inp_aligned
+       bl      _parisc_AES_decrypt,%r31
+       nop
+
+       extru,<> $out,31,2,%r0
+       b       L\$dec_out_aligned
+       nop
+
+       _srm    $s0,24,$acc0
+       _srm    $s0,16,$acc1
+       stb     $acc0,0($out)
+       _srm    $s0,8,$acc2
+       stb     $acc1,1($out)
+       _srm    $s1,24,$acc4
+       stb     $acc2,2($out)
+       _srm    $s1,16,$acc5
+       stb     $s0,3($out)
+       _srm    $s1,8,$acc6
+       stb     $acc4,4($out)
+       _srm    $s2,24,$acc0
+       stb     $acc5,5($out)
+       _srm    $s2,16,$acc1
+       stb     $acc6,6($out)
+       _srm    $s2,8,$acc2
+       stb     $s1,7($out)
+       _srm    $s3,24,$acc4
+       stb     $acc0,8($out)
+       _srm    $s3,16,$acc5
+       stb     $acc1,9($out)
+       _srm    $s3,8,$acc6
+       stb     $acc2,10($out)
+       stb     $s2,11($out)
+       stb     $acc4,12($out)
+       stb     $acc5,13($out)
+       stb     $acc6,14($out)
+       b       L\$dec_done
+       stb     $s3,15($out)
+
+L\$dec_out_aligned
+       stw     $s0,0($out)
+       stw     $s1,4($out)
+       stw     $s2,8($out)
+       stw     $s3,12($out)
+
+L\$dec_done
+       $POP    `-$FRAME-$SAVED_RP`(%sp),%r2    ; standard epilogue
+       $POP    `-$FRAME+1*$SIZE_T`(%sp),%r4
+       $POP    `-$FRAME+2*$SIZE_T`(%sp),%r5
+       $POP    `-$FRAME+3*$SIZE_T`(%sp),%r6
+       $POP    `-$FRAME+4*$SIZE_T`(%sp),%r7
+       $POP    `-$FRAME+5*$SIZE_T`(%sp),%r8
+       $POP    `-$FRAME+6*$SIZE_T`(%sp),%r9
+       $POP    `-$FRAME+7*$SIZE_T`(%sp),%r10
+       $POP    `-$FRAME+8*$SIZE_T`(%sp),%r11
+       $POP    `-$FRAME+9*$SIZE_T`(%sp),%r12
+       $POP    `-$FRAME+10*$SIZE_T`(%sp),%r13
+       $POP    `-$FRAME+11*$SIZE_T`(%sp),%r14
+       $POP    `-$FRAME+12*$SIZE_T`(%sp),%r15
+       $POP    `-$FRAME+13*$SIZE_T`(%sp),%r16
+       $POP    `-$FRAME+14*$SIZE_T`(%sp),%r17
+       $POP    `-$FRAME+15*$SIZE_T`(%sp),%r18
+       bv      (%r2)
+       .EXIT
+       $POPMB  -$FRAME(%sp),%r3
+       .PROCEND
+
+       .ALIGN  16
+_parisc_AES_decrypt
+       .PROC
+       .CALLINFO       MILLICODE
+       .ENTRY
+       ldw     240($key),$rounds
+       ldw     0($key),$t0
+       ldw     4($key),$t1
+       ldw     8($key),$t2
+       ldw     12($key),$t3
+       _srm    $rounds,1,$rounds
+       xor     $t0,$s0,$s0
+       ldw     16($key),$t0
+       xor     $t1,$s1,$s1
+       ldw     20($key),$t1
+       _srm    $s0,24,$acc0
+       xor     $t2,$s2,$s2
+       ldw     24($key),$t2
+       xor     $t3,$s3,$s3
+       ldw     28($key),$t3
+       _srm    $s3,16,$acc1
+L\$dec_loop
+       _srm    $s2,8,$acc2
+       ldwx,s  $acc0($tbl),$acc0
+       _srm    $s1,0,$acc3
+       ldwx,s  $acc1($tbl),$acc1
+       _srm    $s1,24,$acc4
+       ldwx,s  $acc2($tbl),$acc2
+       _srm    $s0,16,$acc5
+       ldwx,s  $acc3($tbl),$acc3
+       _srm    $s3,8,$acc6
+       ldwx,s  $acc4($tbl),$acc4
+       _srm    $s2,0,$acc7
+       ldwx,s  $acc5($tbl),$acc5
+       _srm    $s2,24,$acc8
+       ldwx,s  $acc6($tbl),$acc6
+       _srm    $s1,16,$acc9
+       ldwx,s  $acc7($tbl),$acc7
+       _srm    $s0,8,$acc10
+       ldwx,s  $acc8($tbl),$acc8
+       _srm    $s3,0,$acc11
+       ldwx,s  $acc9($tbl),$acc9
+       _srm    $s3,24,$acc12
+       ldwx,s  $acc10($tbl),$acc10
+       _srm    $s2,16,$acc13
+       ldwx,s  $acc11($tbl),$acc11
+       _srm    $s1,8,$acc14
+       ldwx,s  $acc12($tbl),$acc12
+       _srm    $s0,0,$acc15
+       ldwx,s  $acc13($tbl),$acc13
+       ldwx,s  $acc14($tbl),$acc14
+       ldwx,s  $acc15($tbl),$acc15
+       addib,= -1,$rounds,L\$dec_last
+       ldo     32($key),$key
+
+               _ror    $acc1,8,$acc1
+               xor     $acc0,$t0,$t0
+       ldw     0($key),$s0
+               _ror    $acc2,16,$acc2
+               xor     $acc1,$t0,$t0
+       ldw     4($key),$s1
+               _ror    $acc3,24,$acc3
+               xor     $acc2,$t0,$t0
+       ldw     8($key),$s2
+               _ror    $acc5,8,$acc5
+               xor     $acc3,$t0,$t0
+       ldw     12($key),$s3
+               _ror    $acc6,16,$acc6
+               xor     $acc4,$t1,$t1
+               _ror    $acc7,24,$acc7
+               xor     $acc5,$t1,$t1
+               _ror    $acc9,8,$acc9
+               xor     $acc6,$t1,$t1
+               _ror    $acc10,16,$acc10
+               xor     $acc7,$t1,$t1
+               _ror    $acc11,24,$acc11
+               xor     $acc8,$t2,$t2
+               _ror    $acc13,8,$acc13
+               xor     $acc9,$t2,$t2
+               _ror    $acc14,16,$acc14
+               xor     $acc10,$t2,$t2
+               _ror    $acc15,24,$acc15
+               xor     $acc11,$t2,$t2
+               xor     $acc12,$acc14,$acc14
+               xor     $acc13,$t3,$t3
+       _srm    $t0,24,$acc0
+               xor     $acc14,$t3,$t3
+               xor     $acc15,$t3,$t3
+       _srm    $t3,16,$acc1
+
+       _srm    $t2,8,$acc2
+       ldwx,s  $acc0($tbl),$acc0
+       _srm    $t1,0,$acc3
+       ldwx,s  $acc1($tbl),$acc1
+       _srm    $t1,24,$acc4
+       ldwx,s  $acc2($tbl),$acc2
+       _srm    $t0,16,$acc5
+       ldwx,s  $acc3($tbl),$acc3
+       _srm    $t3,8,$acc6
+       ldwx,s  $acc4($tbl),$acc4
+       _srm    $t2,0,$acc7
+       ldwx,s  $acc5($tbl),$acc5
+       _srm    $t2,24,$acc8
+       ldwx,s  $acc6($tbl),$acc6
+       _srm    $t1,16,$acc9
+       ldwx,s  $acc7($tbl),$acc7
+       _srm    $t0,8,$acc10
+       ldwx,s  $acc8($tbl),$acc8
+       _srm    $t3,0,$acc11
+       ldwx,s  $acc9($tbl),$acc9
+       _srm    $t3,24,$acc12
+       ldwx,s  $acc10($tbl),$acc10
+       _srm    $t2,16,$acc13
+       ldwx,s  $acc11($tbl),$acc11
+       _srm    $t1,8,$acc14
+       ldwx,s  $acc12($tbl),$acc12
+       _srm    $t0,0,$acc15
+       ldwx,s  $acc13($tbl),$acc13
+               _ror    $acc1,8,$acc1
+       ldwx,s  $acc14($tbl),$acc14
+
+               _ror    $acc2,16,$acc2
+               xor     $acc0,$s0,$s0
+       ldwx,s  $acc15($tbl),$acc15
+               _ror    $acc3,24,$acc3
+               xor     $acc1,$s0,$s0
+       ldw     16($key),$t0
+               _ror    $acc5,8,$acc5
+               xor     $acc2,$s0,$s0
+       ldw     20($key),$t1
+               _ror    $acc6,16,$acc6
+               xor     $acc3,$s0,$s0
+       ldw     24($key),$t2
+               _ror    $acc7,24,$acc7
+               xor     $acc4,$s1,$s1
+       ldw     28($key),$t3
+               _ror    $acc9,8,$acc9
+               xor     $acc5,$s1,$s1
+       ldw     1024+0($tbl),%r0                ; prefetch td4
+               _ror    $acc10,16,$acc10
+               xor     $acc6,$s1,$s1
+       ldw     1024+32($tbl),%r0               ; prefetch td4
+               _ror    $acc11,24,$acc11
+               xor     $acc7,$s1,$s1
+       ldw     1024+64($tbl),%r0               ; prefetch td4
+               _ror    $acc13,8,$acc13
+               xor     $acc8,$s2,$s2
+       ldw     1024+96($tbl),%r0               ; prefetch td4
+               _ror    $acc14,16,$acc14
+               xor     $acc9,$s2,$s2
+       ldw     1024+128($tbl),%r0              ; prefetch td4
+               _ror    $acc15,24,$acc15
+               xor     $acc10,$s2,$s2
+       ldw     1024+160($tbl),%r0              ; prefetch td4
+       _srm    $s0,24,$acc0
+               xor     $acc11,$s2,$s2
+       ldw     1024+192($tbl),%r0              ; prefetch td4
+               xor     $acc12,$acc14,$acc14
+               xor     $acc13,$s3,$s3
+       ldw     1024+224($tbl),%r0              ; prefetch td4
+               xor     $acc14,$s3,$s3
+               xor     $acc15,$s3,$s3
+       b       L\$dec_loop
+       _srm    $s3,16,$acc1
+
+       .ALIGN  16
+L\$dec_last
+       ldo     1024($tbl),$rounds
+               _ror    $acc1,8,$acc1
+               xor     $acc0,$t0,$t0
+       ldw     0($key),$s0
+               _ror    $acc2,16,$acc2
+               xor     $acc1,$t0,$t0
+       ldw     4($key),$s1
+               _ror    $acc3,24,$acc3
+               xor     $acc2,$t0,$t0
+       ldw     8($key),$s2
+               _ror    $acc5,8,$acc5
+               xor     $acc3,$t0,$t0
+       ldw     12($key),$s3
+               _ror    $acc6,16,$acc6
+               xor     $acc4,$t1,$t1
+               _ror    $acc7,24,$acc7
+               xor     $acc5,$t1,$t1
+               _ror    $acc9,8,$acc9
+               xor     $acc6,$t1,$t1
+               _ror    $acc10,16,$acc10
+               xor     $acc7,$t1,$t1
+               _ror    $acc11,24,$acc11
+               xor     $acc8,$t2,$t2
+               _ror    $acc13,8,$acc13
+               xor     $acc9,$t2,$t2
+               _ror    $acc14,16,$acc14
+               xor     $acc10,$t2,$t2
+               _ror    $acc15,24,$acc15
+               xor     $acc11,$t2,$t2
+               xor     $acc12,$acc14,$acc14
+               xor     $acc13,$t3,$t3
+       _srm    $t0,24,$acc0
+               xor     $acc14,$t3,$t3
+               xor     $acc15,$t3,$t3
+       _srm    $t3,16,$acc1
+
+       _srm    $t2,8,$acc2
+       ldbx    $acc0($rounds),$acc0
+       _srm    $t1,24,$acc4
+       ldbx    $acc1($rounds),$acc1
+       _srm    $t0,16,$acc5
+       _srm    $t1,0,$acc3
+       ldbx    $acc2($rounds),$acc2
+       ldbx    $acc3($rounds),$acc3
+       _srm    $t3,8,$acc6
+       ldbx    $acc4($rounds),$acc4
+       _srm    $t2,24,$acc8
+       ldbx    $acc5($rounds),$acc5
+       _srm    $t1,16,$acc9
+       _srm    $t2,0,$acc7
+       ldbx    $acc6($rounds),$acc6
+       ldbx    $acc7($rounds),$acc7
+       _srm    $t0,8,$acc10
+       ldbx    $acc8($rounds),$acc8
+       _srm    $t3,24,$acc12
+       ldbx    $acc9($rounds),$acc9
+       _srm    $t2,16,$acc13
+       _srm    $t3,0,$acc11
+       ldbx    $acc10($rounds),$acc10
+       _srm    $t1,8,$acc14
+       ldbx    $acc11($rounds),$acc11
+       ldbx    $acc12($rounds),$acc12
+       ldbx    $acc13($rounds),$acc13
+       _srm    $t0,0,$acc15
+       ldbx    $acc14($rounds),$acc14
+
+               dep     $acc0,7,8,$acc3
+       ldbx    $acc15($rounds),$acc15
+               dep     $acc4,7,8,$acc7
+               dep     $acc1,15,8,$acc3
+               dep     $acc5,15,8,$acc7
+               dep     $acc2,23,8,$acc3
+               dep     $acc6,23,8,$acc7
+               xor     $acc3,$s0,$s0
+               xor     $acc7,$s1,$s1
+               dep     $acc8,7,8,$acc11
+               dep     $acc12,7,8,$acc15
+               dep     $acc9,15,8,$acc11
+               dep     $acc13,15,8,$acc15
+               dep     $acc10,23,8,$acc11
+               dep     $acc14,23,8,$acc15
+               xor     $acc11,$s2,$s2
+
+       bv      (%r31)
+       .EXIT
+               xor     $acc15,$s3,$s3
+       .PROCEND
+
+       .ALIGN  64
+L\$AES_Td
+       .WORD   0x51f4a750, 0x7e416553, 0x1a17a4c3, 0x3a275e96
+       .WORD   0x3bab6bcb, 0x1f9d45f1, 0xacfa58ab, 0x4be30393
+       .WORD   0x2030fa55, 0xad766df6, 0x88cc7691, 0xf5024c25
+       .WORD   0x4fe5d7fc, 0xc52acbd7, 0x26354480, 0xb562a38f
+       .WORD   0xdeb15a49, 0x25ba1b67, 0x45ea0e98, 0x5dfec0e1
+       .WORD   0xc32f7502, 0x814cf012, 0x8d4697a3, 0x6bd3f9c6
+       .WORD   0x038f5fe7, 0x15929c95, 0xbf6d7aeb, 0x955259da
+       .WORD   0xd4be832d, 0x587421d3, 0x49e06929, 0x8ec9c844
+       .WORD   0x75c2896a, 0xf48e7978, 0x99583e6b, 0x27b971dd
+       .WORD   0xbee14fb6, 0xf088ad17, 0xc920ac66, 0x7dce3ab4
+       .WORD   0x63df4a18, 0xe51a3182, 0x97513360, 0x62537f45
+       .WORD   0xb16477e0, 0xbb6bae84, 0xfe81a01c, 0xf9082b94
+       .WORD   0x70486858, 0x8f45fd19, 0x94de6c87, 0x527bf8b7
+       .WORD   0xab73d323, 0x724b02e2, 0xe31f8f57, 0x6655ab2a
+       .WORD   0xb2eb2807, 0x2fb5c203, 0x86c57b9a, 0xd33708a5
+       .WORD   0x302887f2, 0x23bfa5b2, 0x02036aba, 0xed16825c
+       .WORD   0x8acf1c2b, 0xa779b492, 0xf307f2f0, 0x4e69e2a1
+       .WORD   0x65daf4cd, 0x0605bed5, 0xd134621f, 0xc4a6fe8a
+       .WORD   0x342e539d, 0xa2f355a0, 0x058ae132, 0xa4f6eb75
+       .WORD   0x0b83ec39, 0x4060efaa, 0x5e719f06, 0xbd6e1051
+       .WORD   0x3e218af9, 0x96dd063d, 0xdd3e05ae, 0x4de6bd46
+       .WORD   0x91548db5, 0x71c45d05, 0x0406d46f, 0x605015ff
+       .WORD   0x1998fb24, 0xd6bde997, 0x894043cc, 0x67d99e77
+       .WORD   0xb0e842bd, 0x07898b88, 0xe7195b38, 0x79c8eedb
+       .WORD   0xa17c0a47, 0x7c420fe9, 0xf8841ec9, 0x00000000
+       .WORD   0x09808683, 0x322bed48, 0x1e1170ac, 0x6c5a724e
+       .WORD   0xfd0efffb, 0x0f853856, 0x3daed51e, 0x362d3927
+       .WORD   0x0a0fd964, 0x685ca621, 0x9b5b54d1, 0x24362e3a
+       .WORD   0x0c0a67b1, 0x9357e70f, 0xb4ee96d2, 0x1b9b919e
+       .WORD   0x80c0c54f, 0x61dc20a2, 0x5a774b69, 0x1c121a16
+       .WORD   0xe293ba0a, 0xc0a02ae5, 0x3c22e043, 0x121b171d
+       .WORD   0x0e090d0b, 0xf28bc7ad, 0x2db6a8b9, 0x141ea9c8
+       .WORD   0x57f11985, 0xaf75074c, 0xee99ddbb, 0xa37f60fd
+       .WORD   0xf701269f, 0x5c72f5bc, 0x44663bc5, 0x5bfb7e34
+       .WORD   0x8b432976, 0xcb23c6dc, 0xb6edfc68, 0xb8e4f163
+       .WORD   0xd731dcca, 0x42638510, 0x13972240, 0x84c61120
+       .WORD   0x854a247d, 0xd2bb3df8, 0xaef93211, 0xc729a16d
+       .WORD   0x1d9e2f4b, 0xdcb230f3, 0x0d8652ec, 0x77c1e3d0
+       .WORD   0x2bb3166c, 0xa970b999, 0x119448fa, 0x47e96422
+       .WORD   0xa8fc8cc4, 0xa0f03f1a, 0x567d2cd8, 0x223390ef
+       .WORD   0x87494ec7, 0xd938d1c1, 0x8ccaa2fe, 0x98d40b36
+       .WORD   0xa6f581cf, 0xa57ade28, 0xdab78e26, 0x3fadbfa4
+       .WORD   0x2c3a9de4, 0x5078920d, 0x6a5fcc9b, 0x547e4662
+       .WORD   0xf68d13c2, 0x90d8b8e8, 0x2e39f75e, 0x82c3aff5
+       .WORD   0x9f5d80be, 0x69d0937c, 0x6fd52da9, 0xcf2512b3
+       .WORD   0xc8ac993b, 0x10187da7, 0xe89c636e, 0xdb3bbb7b
+       .WORD   0xcd267809, 0x6e5918f4, 0xec9ab701, 0x834f9aa8
+       .WORD   0xe6956e65, 0xaaffe67e, 0x21bccf08, 0xef15e8e6
+       .WORD   0xbae79bd9, 0x4a6f36ce, 0xea9f09d4, 0x29b07cd6
+       .WORD   0x31a4b2af, 0x2a3f2331, 0xc6a59430, 0x35a266c0
+       .WORD   0x744ebc37, 0xfc82caa6, 0xe090d0b0, 0x33a7d815
+       .WORD   0xf104984a, 0x41ecdaf7, 0x7fcd500e, 0x1791f62f
+       .WORD   0x764dd68d, 0x43efb04d, 0xccaa4d54, 0xe49604df
+       .WORD   0x9ed1b5e3, 0x4c6a881b, 0xc12c1fb8, 0x4665517f
+       .WORD   0x9d5eea04, 0x018c355d, 0xfa877473, 0xfb0b412e
+       .WORD   0xb3671d5a, 0x92dbd252, 0xe9105633, 0x6dd64713
+       .WORD   0x9ad7618c, 0x37a10c7a, 0x59f8148e, 0xeb133c89
+       .WORD   0xcea927ee, 0xb761c935, 0xe11ce5ed, 0x7a47b13c
+       .WORD   0x9cd2df59, 0x55f2733f, 0x1814ce79, 0x73c737bf
+       .WORD   0x53f7cdea, 0x5ffdaa5b, 0xdf3d6f14, 0x7844db86
+       .WORD   0xcaaff381, 0xb968c43e, 0x3824342c, 0xc2a3405f
+       .WORD   0x161dc372, 0xbce2250c, 0x283c498b, 0xff0d9541
+       .WORD   0x39a80171, 0x080cb3de, 0xd8b4e49c, 0x6456c190
+       .WORD   0x7bcb8461, 0xd532b670, 0x486c5c74, 0xd0b85742
+       .BYTE   0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38
+       .BYTE   0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb
+       .BYTE   0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87
+       .BYTE   0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb
+       .BYTE   0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d
+       .BYTE   0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e
+       .BYTE   0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2
+       .BYTE   0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25
+       .BYTE   0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16
+       .BYTE   0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92
+       .BYTE   0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda
+       .BYTE   0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84
+       .BYTE   0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a
+       .BYTE   0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06
+       .BYTE   0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02
+       .BYTE   0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b
+       .BYTE   0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea
+       .BYTE   0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73
+       .BYTE   0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85
+       .BYTE   0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e
+       .BYTE   0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89
+       .BYTE   0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b
+       .BYTE   0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20
+       .BYTE   0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4
+       .BYTE   0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31
+       .BYTE   0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f
+       .BYTE   0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d
+       .BYTE   0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef
+       .BYTE   0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0
+       .BYTE   0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61
+       .BYTE   0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26
+       .BYTE   0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d
+       .STRINGZ "AES for PA-RISC, CRYPTOGAMS by <appro\@openssl.org>"
+___
+
+foreach (split("\n",$code)) {
+       s/\`([^\`]*)\`/eval $1/ge;
+
+       # translate made up instructons: _ror, _srm
+       s/_ror(\s+)(%r[0-9]+),/shd$1$2,$2,/                             or
+
+       s/_srm(\s+%r[0-9]+),([0-9]+),/
+               $SIZE_T==4 ? sprintf("extru%s,%d,8,",$1,31-$2)
+               :            sprintf("extrd,u%s,%d,8,",$1,63-$2)/e;
+
+       s/,\*/,/ if ($SIZE_T==4);
+       print $_,"\n";
+}
+close STDOUT;
diff --git a/crypto/bn/asm/parisc-mont.pl b/crypto/bn/asm/parisc-mont.pl
new file mode 100644 (file)
index 0000000..4a766a8
--- /dev/null
@@ -0,0 +1,993 @@
+#!/usr/bin/env perl
+
+# ====================================================================
+# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+
+# On PA-7100LC this module performs ~90-50% better, less for longer
+# keys, than code generated by gcc 3.2 for PA-RISC 1.1. Latter means
+# that compiler utilized xmpyu instruction to perform 32x32=64-bit
+# multiplication, which in turn means that "baseline" performance was
+# optimal in respect to instruction set capabilities. Fair comparison
+# with vendor compiler is problematic, because OpenSSL doesn't define
+# BN_LLONG [presumably] for historical reasons, which drives compiler
+# toward 4 times 16x16=32-bit multiplicatons [plus complementary
+# shifts and additions] instead. This means that you should observe
+# several times improvement over code generated by vendor compiler
+# for PA-RISC 1.1, but the "baseline" is far from optimal. The actual
+# improvement coefficient was never collected on PA-7100LC, or any
+# other 1.1 CPU, because I don't have access to such machine with
+# vendor compiler. But to give you a taste, PA-RISC 1.1 code path
+# reportedly outperformed code generated by cc +DA1.1 +O3 by factor
+# of ~5x on PA-8600.
+#
+# On PA-RISC 2.0 it has to compete with pa-risc2[W].s, which is
+# reportedly ~2x faster than vendor compiler generated code [according
+# to comment in pa-risc2[W].s]. Here comes a catch. Execution core of
+# this implementation is actually 32-bit one, in the sense that it
+# operates on 32-bit values. But pa-risc2[W].s operates on arrays of
+# 64-bit BN_LONGs... How do they interoperate then? No problem. This
+# module picks halves of 64-bit values in reverse order and pretends
+# they were 32-bit BN_LONGs. But can 32-bit core compete with "pure"
+# 64-bit code such as pa-risc2[W].s then? Well, the thing is that
+# 32x32=64-bit multiplication is the best even PA-RISC 2.0 can do,
+# i.e. there is no "wider" multiplication like on most other 64-bit
+# platforms. This means that even being effectively 32-bit, this
+# implementation performs "64-bit" computational task in same amount
+# of arithmetic operations, most notably multiplications. It requires
+# more memory references, most notably to tp[num], but this doesn't
+# seem to exhaust memory port capacity. And indeed, dedicated PA-RISC
+# 2.0 code path, provides virtually same performance as pa-risc2[W].s:
+# it's ~10% better for shortest key length and ~10% worse for longest
+# one.
+#
+# In case it wasn't clear. The module has two distinct code paths:
+# PA-RISC 1.1 and PA-RISC 2.0 ones. Latter features carry-free 64-bit
+# additions and 64-bit integer loads, not to mention specific
+# instruction scheduling. In 64-bit build naturally only 2.0 code path
+# is assembled. In 32-bit application context both code paths are
+# assembled, PA-RISC 2.0 CPU is detected at run-time and proper path
+# is taken automatically. Also, in 32-bit build the module imposes
+# couple of limitations: vector lengths has to be even and vector
+# addresses has to be 64-bit aligned. Normally neither is a problem:
+# most common key lengths are even and vectors are commonly malloc-ed,
+# which ensures alignment.
+#
+# Special thanks to polarhome.com for providing HP-UX account on
+# PA-RISC 1.1 machine, and to correspondent who chose to remain
+# anonymous for testing the code on PA-RISC 2.0 machine.
+\f
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+
+$flavour = shift;
+$output = shift;
+
+open STDOUT,">$output";
+
+if ($flavour =~ /64/) {
+       $LEVEL          ="2.0W";
+       $SIZE_T         =8;
+       $FRAME_MARKER   =80;
+       $SAVED_RP       =16;
+       $PUSH           ="std";
+       $PUSHMA         ="std,ma";
+       $POP            ="ldd";
+       $POPMB          ="ldd,mb";
+       $BN_SZ          =$SIZE_T;
+} else {
+       $LEVEL          ="1.1"; #$LEVEL.="\n\t.ALLOW\t2.0";
+       $SIZE_T         =4;
+       $FRAME_MARKER   =48;
+       $SAVED_RP       =20;
+       $PUSH           ="stw";
+       $PUSHMA         ="stwm";
+       $POP            ="ldw";
+       $POPMB          ="ldwm";
+       $BN_SZ          =$SIZE_T;
+       if (open CONF,"<${dir}../../opensslconf.h") {
+           while(<CONF>) {
+               if (m/#\s*define\s+SIXTY_FOUR_BIT/) {
+                   $BN_SZ=8;
+                   $LEVEL="2.0";
+                   last;
+               }
+           }
+           close CONF;
+       }
+}
+
+$FRAME=8*$SIZE_T+$FRAME_MARKER;        # 8 saved regs + frame marker
+                               #                [+ argument transfer]
+$LOCALS=$FRAME-$FRAME_MARKER;
+$FRAME+=32;                    # local variables
+
+$tp="%r31";
+$ti1="%r29";
+$ti0="%r28";
+
+$rp="%r26";
+$ap="%r25";
+$bp="%r24";
+$np="%r23";
+$n0="%r22";    # passed through stack in 32-bit
+$num="%r21";   # passed through stack in 32-bit
+$idx="%r20";
+$arrsz="%r19";
+
+$nm1="%r7";
+$nm0="%r6";
+$ab1="%r5";
+$ab0="%r4";
+
+$fp="%r3";
+$hi1="%r2";
+$hi0="%r1";
+
+$xfer=$n0;     # accomodates [-16..15] offset in fld[dw]s
+
+$fm0="%fr4";   $fti=$fm0;
+$fbi="%fr5L";
+$fn0="%fr5R";
+$fai="%fr6";   $fab0="%fr7";   $fab1="%fr8";
+$fni="%fr9";   $fnm0="%fr10";  $fnm1="%fr11";
+
+$code=<<___;
+       .LEVEL  $LEVEL
+       .SPACE  \$TEXT\$
+       .SUBSPA \$CODE\$,QUAD=0,ALIGN=8,ACCESS=0x2C,CODE_ONLY
+
+       .EXPORT bn_mul_mont,ENTRY,ARGW0=GR,ARGW1=GR,ARGW2=GR,ARGW3=GR
+       .ALIGN  64
+bn_mul_mont
+       .PROC
+       .CALLINFO       FRAME=`$FRAME-8*$SIZE_T`,NO_CALLS,SAVE_RP,SAVE_SP,ENTRY_GR=6
+       .ENTRY
+       $PUSH   %r2,-$SAVED_RP(%sp)             ; standard prologue
+       $PUSHMA %r3,$FRAME(%sp)
+       $PUSH   %r4,`-$FRAME+1*$SIZE_T`(%sp)
+       $PUSH   %r5,`-$FRAME+2*$SIZE_T`(%sp)
+       $PUSH   %r6,`-$FRAME+3*$SIZE_T`(%sp)
+       $PUSH   %r7,`-$FRAME+4*$SIZE_T`(%sp)
+       $PUSH   %r8,`-$FRAME+5*$SIZE_T`(%sp)
+       $PUSH   %r9,`-$FRAME+6*$SIZE_T`(%sp)
+       $PUSH   %r10,`-$FRAME+7*$SIZE_T`(%sp)
+       ldo     -$FRAME(%sp),$fp
+___
+$code.=<<___ if ($SIZE_T==4);
+       ldw     `-$FRAME_MARKER-4`($fp),$n0
+       ldw     `-$FRAME_MARKER-8`($fp),$num
+       nop
+       nop                                     ; alignment
+___
+$code.=<<___ if ($BN_SZ==4);
+       comiclr,<=      6,$num,%r0              ; are vectors long enough?
+       b               L\$abort
+       ldi             0,%r28                  ; signal "unhandled"
+       add,ev          %r0,$num,$num           ; is $num even?
+       b               L\$abort
+       nop
+       or              $ap,$np,$ti1
+       extru,=         $ti1,31,3,%r0           ; are ap and np 64-bit aligned?
+       b               L\$abort
+       nop
+       nop                                     ; alignment
+       nop
+
+       fldws           0($n0),${fn0}
+       fldws,ma        4($bp),${fbi}           ; bp[0]
+___
+$code.=<<___ if ($BN_SZ==8);
+       comib,>         3,$num,L\$abort         ; are vectors long enough?
+       ldi             0,%r28                  ; signal "unhandled"
+       addl            $num,$num,$num          ; I operate on 32-bit values
+
+       fldws           4($n0),${fn0}           ; only low part of n0
+       fldws           4($bp),${fbi}           ; bp[0] in flipped word order
+___
+$code.=<<___;
+       fldds           0($ap),${fai}           ; ap[0,1]
+       fldds           0($np),${fni}           ; np[0,1]
+
+       sh2addl         $num,%r0,$arrsz
+       ldi             31,$hi0
+       ldo             36($arrsz),$hi1         ; space for tp[num+1]
+       andcm           $hi1,$hi0,$hi1          ; align
+       addl            $hi1,%sp,%sp
+       $PUSH           $fp,-$SIZE_T(%sp)
+
+       ldo             `$LOCALS+16`($fp),$xfer
+       ldo             `$LOCALS+32+4`($fp),$tp
+
+       xmpyu           ${fai}L,${fbi},${fab0}  ; ap[0]*bp[0]
+       xmpyu           ${fai}R,${fbi},${fab1}  ; ap[1]*bp[0]
+       xmpyu           ${fn0},${fab0}R,${fm0}
+
+       addl            $arrsz,$ap,$ap          ; point at the end
+       addl            $arrsz,$np,$np
+       subi            0,$arrsz,$idx           ; j=0
+       ldo             8($idx),$idx            ; j++++
+
+       xmpyu           ${fni}L,${fm0}R,${fnm0} ; np[0]*m
+       xmpyu           ${fni}R,${fm0}R,${fnm1} ; np[1]*m
+       fstds           ${fab0},-16($xfer)
+       fstds           ${fnm0},-8($xfer)
+       fstds           ${fab1},0($xfer)
+       fstds           ${fnm1},8($xfer)
+        flddx          $idx($ap),${fai}        ; ap[2,3]
+        flddx          $idx($np),${fni}        ; np[2,3]
+___
+$code.=<<___ if ($BN_SZ==4);
+       mtctl           $hi0,%cr11              ; $hi0 still holds 31
+       extrd,u,*=      $hi0,%sar,1,$hi0        ; executes on PA-RISC 1.0
+       b               L\$parisc11
+       nop
+___
+$code.=<<___;                                  # PA-RISC 2.0 code-path
+       xmpyu           ${fai}L,${fbi},${fab0}  ; ap[j]*bp[0]
+       xmpyu           ${fni}L,${fm0}R,${fnm0} ; np[j]*m
+       ldd             -16($xfer),$ab0
+       fstds           ${fab0},-16($xfer)
+
+       extrd,u         $ab0,31,32,$hi0
+       extrd,u         $ab0,63,32,$ab0
+       ldd             -8($xfer),$nm0
+       fstds           ${fnm0},-8($xfer)
+        ldo            8($idx),$idx            ; j++++
+        addl           $ab0,$nm0,$nm0          ; low part is discarded
+        extrd,u        $nm0,31,32,$hi1
+\f
+L\$1st
+       xmpyu           ${fai}R,${fbi},${fab1}  ; ap[j+1]*bp[0]
+       xmpyu           ${fni}R,${fm0}R,${fnm1} ; np[j+1]*m
+       ldd             0($xfer),$ab1
+       fstds           ${fab1},0($xfer)
+        addl           $hi0,$ab1,$ab1
+        extrd,u        $ab1,31,32,$hi0
+       ldd             8($xfer),$nm1
+       fstds           ${fnm1},8($xfer)
+        extrd,u        $ab1,63,32,$ab1
+        addl           $hi1,$nm1,$nm1
+       flddx           $idx($ap),${fai}        ; ap[j,j+1]
+       flddx           $idx($np),${fni}        ; np[j,j+1]
+        addl           $ab1,$nm1,$nm1
+        extrd,u        $nm1,31,32,$hi1
+
+       xmpyu           ${fai}L,${fbi},${fab0}  ; ap[j]*bp[0]
+       xmpyu           ${fni}L,${fm0}R,${fnm0} ; np[j]*m
+       ldd             -16($xfer),$ab0
+       fstds           ${fab0},-16($xfer)
+        addl           $hi0,$ab0,$ab0
+        extrd,u        $ab0,31,32,$hi0
+       ldd             -8($xfer),$nm0
+       fstds           ${fnm0},-8($xfer)
+        extrd,u        $ab0,63,32,$ab0
+        addl           $hi1,$nm0,$nm0
+       stw             $nm1,-4($tp)            ; tp[j-1]
+        addl           $ab0,$nm0,$nm0
+        stw,ma         $nm0,8($tp)             ; tp[j-1]
+       addib,<>        8,$idx,L\$1st           ; j++++
+        extrd,u        $nm0,31,32,$hi1
+
+       xmpyu           ${fai}R,${fbi},${fab1}  ; ap[j]*bp[0]
+       xmpyu           ${fni}R,${fm0}R,${fnm1} ; np[j]*m
+       ldd             0($xfer),$ab1
+       fstds           ${fab1},0($xfer)
+        addl           $hi0,$ab1,$ab1
+        extrd,u        $ab1,31,32,$hi0
+       ldd             8($xfer),$nm1
+       fstds           ${fnm1},8($xfer)
+        extrd,u        $ab1,63,32,$ab1
+        addl           $hi1,$nm1,$nm1
+       ldd             -16($xfer),$ab0
+        addl           $ab1,$nm1,$nm1
+       ldd             -8($xfer),$nm0
+        extrd,u        $nm1,31,32,$hi1
+
+        addl           $hi0,$ab0,$ab0
+        extrd,u        $ab0,31,32,$hi0
+       stw             $nm1,-4($tp)            ; tp[j-1]
+        extrd,u        $ab0,63,32,$ab0
+        addl           $hi1,$nm0,$nm0
+       ldd             0($xfer),$ab1
+        addl           $ab0,$nm0,$nm0
+       ldd,mb          8($xfer),$nm1
+        extrd,u        $nm0,31,32,$hi1
+       stw,ma          $nm0,8($tp)             ; tp[j-1]
+
+       ldo             -1($num),$num           ; i--
+       subi            0,$arrsz,$idx           ; j=0
+___
+$code.=<<___ if ($BN_SZ==4);
+       fldws,ma        4($bp),${fbi}           ; bp[1]
+___
+$code.=<<___ if ($BN_SZ==8);
+       fldws           0($bp),${fbi}           ; bp[1] in flipped word order
+___
+$code.=<<___;
+        flddx          $idx($ap),${fai}        ; ap[0,1]
+        flddx          $idx($np),${fni}        ; np[0,1]
+        fldws          8($xfer),${fti}R        ; tp[0]
+       addl            $hi0,$ab1,$ab1
+        extrd,u        $ab1,31,32,$hi0
+        extrd,u        $ab1,63,32,$ab1
+        ldo            8($idx),$idx            ; j++++
+        xmpyu          ${fai}L,${fbi},${fab0}  ; ap[0]*bp[1]
+        xmpyu          ${fai}R,${fbi},${fab1}  ; ap[1]*bp[1]
+       addl            $hi1,$nm1,$nm1
+       addl            $ab1,$nm1,$nm1
+       extrd,u         $nm1,31,32,$hi1
+        fstws,mb       ${fab0}L,-8($xfer)      ; save high part
+       stw             $nm1,-4($tp)            ; tp[j-1]
+
+        fcpy,sgl       %fr0,${fti}L            ; zero high part
+        fcpy,sgl       %fr0,${fab0}L
+       addl            $hi1,$hi0,$hi0
+       extrd,u         $hi0,31,32,$hi1
+        fcnvxf,dbl,dbl ${fti},${fti}           ; 32-bit unsigned int -> double
+        fcnvxf,dbl,dbl ${fab0},${fab0}
+       stw             $hi0,0($tp)
+       stw             $hi1,4($tp)
+
+       fadd,dbl        ${fti},${fab0},${fab0}  ; add tp[0]
+       fcnvfx,dbl,dbl  ${fab0},${fab0}         ; double -> 33-bit unsigned int
+       xmpyu           ${fn0},${fab0}R,${fm0}
+       ldo             `$LOCALS+32+4`($fp),$tp
+L\$outer
+       xmpyu           ${fni}L,${fm0}R,${fnm0} ; np[0]*m
+       xmpyu           ${fni}R,${fm0}R,${fnm1} ; np[1]*m
+       fstds           ${fab0},-16($xfer)      ; 33-bit value
+       fstds           ${fnm0},-8($xfer)
+        flddx          $idx($ap),${fai}        ; ap[2]
+        flddx          $idx($np),${fni}        ; np[2]
+        ldo            8($idx),$idx            ; j++++
+       ldd             -16($xfer),$ab0         ; 33-bit value
+       ldd             -8($xfer),$nm0
+       ldw             0($xfer),$hi0           ; high part
+
+       xmpyu           ${fai}L,${fbi},${fab0}  ; ap[j]*bp[i]
+       xmpyu           ${fni}L,${fm0}R,${fnm0} ; np[j]*m
+        extrd,u        $ab0,31,32,$ti0         ; carry bit
+        extrd,u        $ab0,63,32,$ab0
+       fstds           ${fab1},0($xfer)
+        addl           $ti0,$hi0,$hi0          ; account carry bit
+       fstds           ${fnm1},8($xfer)
+        addl           $ab0,$nm0,$nm0          ; low part is discarded
+       ldw             0($tp),$ti1             ; tp[1]
+        extrd,u        $nm0,31,32,$hi1
+       fstds           ${fab0},-16($xfer)
+       fstds           ${fnm0},-8($xfer)
+\f
+L\$inner
+       xmpyu           ${fai}R,${fbi},${fab1}  ; ap[j+1]*bp[i]
+       xmpyu           ${fni}R,${fm0}R,${fnm1} ; np[j+1]*m
+       ldd             0($xfer),$ab1
+       fstds           ${fab1},0($xfer)
+        addl           $hi0,$ti1,$ti1
+        addl           $ti1,$ab1,$ab1
+       ldd             8($xfer),$nm1
+       fstds           ${fnm1},8($xfer)
+        extrd,u        $ab1,31,32,$hi0
+        extrd,u        $ab1,63,32,$ab1
+       flddx           $idx($ap),${fai}        ; ap[j,j+1]
+       flddx           $idx($np),${fni}        ; np[j,j+1]
+        addl           $hi1,$nm1,$nm1
+        addl           $ab1,$nm1,$nm1
+       ldw             4($tp),$ti0             ; tp[j]
+       stw             $nm1,-4($tp)            ; tp[j-1]
+
+       xmpyu           ${fai}L,${fbi},${fab0}  ; ap[j]*bp[i]
+       xmpyu           ${fni}L,${fm0}R,${fnm0} ; np[j]*m
+       ldd             -16($xfer),$ab0
+       fstds           ${fab0},-16($xfer)
+        addl           $hi0,$ti0,$ti0
+        addl           $ti0,$ab0,$ab0
+       ldd             -8($xfer),$nm0
+       fstds           ${fnm0},-8($xfer)
+        extrd,u        $ab0,31,32,$hi0
+        extrd,u        $nm1,31,32,$hi1
+       ldw             8($tp),$ti1             ; tp[j]
+        extrd,u        $ab0,63,32,$ab0
+        addl           $hi1,$nm0,$nm0
+        addl           $ab0,$nm0,$nm0
+        stw,ma         $nm0,8($tp)             ; tp[j-1]
+       addib,<>        8,$idx,L\$inner         ; j++++
+        extrd,u        $nm0,31,32,$hi1
+
+       xmpyu           ${fai}R,${fbi},${fab1}  ; ap[j]*bp[i]
+       xmpyu           ${fni}R,${fm0}R,${fnm1} ; np[j]*m
+       ldd             0($xfer),$ab1
+       fstds           ${fab1},0($xfer)
+        addl           $hi0,$ti1,$ti1
+        addl           $ti1,$ab1,$ab1
+       ldd             8($xfer),$nm1
+       fstds           ${fnm1},8($xfer)
+        extrd,u        $ab1,31,32,$hi0
+        extrd,u        $ab1,63,32,$ab1
+       ldw             4($tp),$ti0             ; tp[j]
+        addl           $hi1,$nm1,$nm1
+        addl           $ab1,$nm1,$nm1
+       ldd             -16($xfer),$ab0
+       ldd             -8($xfer),$nm0
+        extrd,u        $nm1,31,32,$hi1
+
+       addl            $hi0,$ab0,$ab0
+        addl           $ti0,$ab0,$ab0
+        stw            $nm1,-4($tp)            ; tp[j-1]
+        extrd,u        $ab0,31,32,$hi0
+       ldw             8($tp),$ti1             ; tp[j]
+        extrd,u        $ab0,63,32,$ab0
+        addl           $hi1,$nm0,$nm0
+       ldd             0($xfer),$ab1
+        addl           $ab0,$nm0,$nm0
+       ldd,mb          8($xfer),$nm1
+        extrd,u        $nm0,31,32,$hi1
+        stw,ma         $nm0,8($tp)             ; tp[j-1]
+
+       addib,=         -1,$num,L\$outerdone    ; i--
+       subi            0,$arrsz,$idx           ; j=0
+___
+$code.=<<___ if ($BN_SZ==4);
+       fldws,ma        4($bp),${fbi}           ; bp[i]
+___
+$code.=<<___ if ($BN_SZ==8);
+       ldi             12,$ti0                 ; bp[i] in flipped word order
+       addl,ev         %r0,$num,$num
+       ldi             -4,$ti0
+       addl            $ti0,$bp,$bp
+       fldws           0($bp),${fbi}
+___
+$code.=<<___;
+        flddx          $idx($ap),${fai}        ; ap[0]
+       addl            $hi0,$ab1,$ab1
+        flddx          $idx($np),${fni}        ; np[0]
+        fldws          8($xfer),${fti}R        ; tp[0]
+       addl            $ti1,$ab1,$ab1
+       extrd,u         $ab1,31,32,$hi0
+       extrd,u         $ab1,63,32,$ab1
+
+        ldo            8($idx),$idx            ; j++++
+        xmpyu          ${fai}L,${fbi},${fab0}  ; ap[0]*bp[i]
+        xmpyu          ${fai}R,${fbi},${fab1}  ; ap[1]*bp[i]
+       ldw             4($tp),$ti0             ; tp[j]
+
+       addl            $hi1,$nm1,$nm1
+        fstws,mb       ${fab0}L,-8($xfer)      ; save high part
+       addl            $ab1,$nm1,$nm1
+       extrd,u         $nm1,31,32,$hi1
+        fcpy,sgl       %fr0,${fti}L            ; zero high part
+        fcpy,sgl       %fr0,${fab0}L
+       stw             $nm1,-4($tp)            ; tp[j-1]
+
+        fcnvxf,dbl,dbl ${fti},${fti}           ; 32-bit unsigned int -> double
+        fcnvxf,dbl,dbl ${fab0},${fab0}
+       addl            $hi1,$hi0,$hi0
+        fadd,dbl       ${fti},${fab0},${fab0}  ; add tp[0]
+       addl            $ti0,$hi0,$hi0
+       extrd,u         $hi0,31,32,$hi1
+        fcnvfx,dbl,dbl ${fab0},${fab0}         ; double -> 33-bit unsigned int
+       stw             $hi0,0($tp)
+       stw             $hi1,4($tp)
+        xmpyu          ${fn0},${fab0}R,${fm0}
+
+       b               L\$outer
+       ldo             `$LOCALS+32+4`($fp),$tp
+\f
+L\$outerdone
+       addl            $hi0,$ab1,$ab1
+       addl            $ti1,$ab1,$ab1
+       extrd,u         $ab1,31,32,$hi0
+       extrd,u         $ab1,63,32,$ab1
+
+       ldw             4($tp),$ti0             ; tp[j]
+
+       addl            $hi1,$nm1,$nm1
+       addl            $ab1,$nm1,$nm1
+       extrd,u         $nm1,31,32,$hi1
+       stw             $nm1,-4($tp)            ; tp[j-1]
+
+       addl            $hi1,$hi0,$hi0
+       addl            $ti0,$hi0,$hi0
+       extrd,u         $hi0,31,32,$hi1
+       stw             $hi0,0($tp)
+       stw             $hi1,4($tp)
+
+       ldo             `$LOCALS+32`($fp),$tp
+       sub             %r0,%r0,%r0             ; clear borrow
+___
+$code.=<<___ if ($BN_SZ==4);
+       ldws,ma         4($tp),$ti0
+       extru,=         $rp,31,3,%r0            ; is rp 64-bit aligned?
+       b               L\$sub_pa11
+       addl            $tp,$arrsz,$tp
+L\$sub
+       ldwx            $idx($np),$hi0
+       subb            $ti0,$hi0,$hi1
+       ldwx            $idx($tp),$ti0
+       addib,<>        4,$idx,L\$sub
+       stws,ma         $hi1,4($rp)
+
+       subb            $ti0,%r0,$hi1
+       ldo             -4($tp),$tp
+___
+$code.=<<___ if ($BN_SZ==8);
+       ldd,ma          8($tp),$ti0
+L\$sub
+       ldd             $idx($np),$hi0
+       shrpd           $ti0,$ti0,32,$ti0       ; flip word order
+       std             $ti0,-8($tp)            ; save flipped value
+       sub,db          $ti0,$hi0,$hi1
+       ldd,ma          8($tp),$ti0
+       addib,<>        8,$idx,L\$sub
+       std,ma          $hi1,8($rp)
+
+       extrd,u         $ti0,31,32,$ti0         ; carry in flipped word order
+       sub,db          $ti0,%r0,$hi1
+       ldo             -8($tp),$tp
+___
+$code.=<<___;
+       and             $tp,$hi1,$ap
+       andcm           $rp,$hi1,$bp
+       or              $ap,$bp,$np
+
+       sub             $rp,$arrsz,$rp          ; rewind rp
+       subi            0,$arrsz,$idx
+       ldo             `$LOCALS+32`($fp),$tp
+L\$copy
+       ldd             $idx($np),$hi0
+       std,ma          %r0,8($tp)
+       addib,<>        8,$idx,.-8              ; L\$copy
+       std,ma          $hi0,8($rp)     
+___
+
+if ($BN_SZ==4) {                               # PA-RISC 1.1 code-path
+$ablo=$ab0;
+$abhi=$ab1;
+$nmlo0=$nm0;
+$nmhi0=$nm1;
+$nmlo1="%r9";
+$nmhi1="%r8";
+
+$code.=<<___;
+       b               L\$done
+       nop
+
+       .ALIGN          8
+L\$parisc11
+       xmpyu           ${fai}L,${fbi},${fab0}  ; ap[j]*bp[0]
+       xmpyu           ${fni}L,${fm0}R,${fnm0} ; np[j]*m
+       ldw             -12($xfer),$ablo
+       ldw             -16($xfer),$hi0
+       ldw             -4($xfer),$nmlo0
+       ldw             -8($xfer),$nmhi0
+       fstds           ${fab0},-16($xfer)
+       fstds           ${fnm0},-8($xfer)
+
+        ldo            8($idx),$idx            ; j++++
+        add            $ablo,$nmlo0,$nmlo0     ; discarded
+        addc           %r0,$nmhi0,$hi1
+       ldw             4($xfer),$ablo
+       ldw             0($xfer),$abhi
+       nop
+\f
+L\$1st_pa11
+       xmpyu           ${fai}R,${fbi},${fab1}  ; ap[j+1]*bp[0]
+       flddx           $idx($ap),${fai}        ; ap[j,j+1]
+       xmpyu           ${fni}R,${fm0}R,${fnm1} ; np[j+1]*m
+       flddx           $idx($np),${fni}        ; np[j,j+1]
+        add            $hi0,$ablo,$ablo
+       ldw             12($xfer),$nmlo1
+        addc           %r0,$abhi,$hi0
+       ldw             8($xfer),$nmhi1
+        add            $ablo,$nmlo1,$nmlo1
+       fstds           ${fab1},0($xfer)
+        addc           %r0,$nmhi1,$nmhi1
+       fstds           ${fnm1},8($xfer)
+        add            $hi1,$nmlo1,$nmlo1
+       ldw             -12($xfer),$ablo
+        addc           %r0,$nmhi1,$hi1
+       ldw             -16($xfer),$abhi
+
+       xmpyu           ${fai}L,${fbi},${fab0}  ; ap[j]*bp[0]
+       ldw             -4($xfer),$nmlo0
+       xmpyu           ${fni}L,${fm0}R,${fnm0} ; np[j]*m
+       ldw             -8($xfer),$nmhi0
+        add            $hi0,$ablo,$ablo
+       stw             $nmlo1,-4($tp)          ; tp[j-1]
+        addc           %r0,$abhi,$hi0
+       fstds           ${fab0},-16($xfer)
+        add            $ablo,$nmlo0,$nmlo0
+       fstds           ${fnm0},-8($xfer)
+        addc           %r0,$nmhi0,$nmhi0
+       ldw             0($xfer),$abhi
+        add            $hi1,$nmlo0,$nmlo0
+       ldw             4($xfer),$ablo
+        stws,ma        $nmlo0,8($tp)           ; tp[j-1]
+       addib,<>        8,$idx,L\$1st_pa11      ; j++++
+        addc           %r0,$nmhi0,$hi1
+
+        ldw            8($xfer),$nmhi1
+        ldw            12($xfer),$nmlo1
+       xmpyu           ${fai}R,${fbi},${fab1}  ; ap[j]*bp[0]
+       xmpyu           ${fni}R,${fm0}R,${fnm1} ; np[j]*m
+        add            $hi0,$ablo,$ablo
+       fstds           ${fab1},0($xfer)
+        addc           %r0,$abhi,$hi0
+       fstds           ${fnm1},8($xfer)
+        add            $ablo,$nmlo1,$nmlo1
+       ldw             -16($xfer),$abhi
+        addc           %r0,$nmhi1,$nmhi1
+       ldw             -12($xfer),$ablo
+        add            $hi1,$nmlo1,$nmlo1
+       ldw             -8($xfer),$nmhi0
+        addc           %r0,$nmhi1,$hi1
+       ldw             -4($xfer),$nmlo0
+
+        add            $hi0,$ablo,$ablo
+       stw             $nmlo1,-4($tp)          ; tp[j-1]
+        addc           %r0,$abhi,$hi0
+       ldw             0($xfer),$abhi
+        add            $ablo,$nmlo0,$nmlo0
+       ldw             4($xfer),$ablo
+        addc           %r0,$nmhi0,$nmhi0
+       ldws,mb         8($xfer),$nmhi1
+        add            $hi1,$nmlo0,$nmlo0
+       ldw             4($xfer),$nmlo1
+        addc           %r0,$nmhi0,$hi1
+       stws,ma         $nmlo0,8($tp)           ; tp[j-1]
+
+       ldo             -1($num),$num           ; i--
+       subi            0,$arrsz,$idx           ; j=0
+
+        fldws,ma       4($bp),${fbi}           ; bp[1]
+        flddx          $idx($ap),${fai}        ; ap[0,1]
+        flddx          $idx($np),${fni}        ; np[0,1]
+        fldws          8($xfer),${fti}R        ; tp[0]
+       add             $hi0,$ablo,$ablo
+       addc            %r0,$abhi,$hi0
+        ldo            8($idx),$idx            ; j++++
+        xmpyu          ${fai}L,${fbi},${fab0}  ; ap[0]*bp[1]
+        xmpyu          ${fai}R,${fbi},${fab1}  ; ap[1]*bp[1]
+       add             $hi1,$nmlo1,$nmlo1
+       addc            %r0,$nmhi1,$nmhi1
+       add             $ablo,$nmlo1,$nmlo1
+       addc            %r0,$nmhi1,$hi1
+        fstws,mb       ${fab0}L,-8($xfer)      ; save high part
+       stw             $nmlo1,-4($tp)          ; tp[j-1]
+
+        fcpy,sgl       %fr0,${fti}L            ; zero high part
+        fcpy,sgl       %fr0,${fab0}L
+       add             $hi1,$hi0,$hi0
+       addc            %r0,%r0,$hi1
+        fcnvxf,dbl,dbl ${fti},${fti}           ; 32-bit unsigned int -> double
+        fcnvxf,dbl,dbl ${fab0},${fab0}
+       stw             $hi0,0($tp)
+       stw             $hi1,4($tp)
+
+       fadd,dbl        ${fti},${fab0},${fab0}  ; add tp[0]
+       fcnvfx,dbl,dbl  ${fab0},${fab0}         ; double -> 33-bit unsigned int
+       xmpyu           ${fn0},${fab0}R,${fm0}
+       ldo             `$LOCALS+32+4`($fp),$tp
+L\$outer_pa11
+       xmpyu           ${fni}L,${fm0}R,${fnm0} ; np[0]*m
+       xmpyu           ${fni}R,${fm0}R,${fnm1} ; np[1]*m
+       fstds           ${fab0},-16($xfer)      ; 33-bit value
+       fstds           ${fnm0},-8($xfer)
+        flddx          $idx($ap),${fai}        ; ap[2,3]
+        flddx          $idx($np),${fni}        ; np[2,3]
+       ldw             -16($xfer),$abhi        ; carry bit actually
+        ldo            8($idx),$idx            ; j++++
+       ldw             -12($xfer),$ablo
+       ldw             -8($xfer),$nmhi0
+       ldw             -4($xfer),$nmlo0
+       ldw             0($xfer),$hi0           ; high part
+
+       xmpyu           ${fai}L,${fbi},${fab0}  ; ap[j]*bp[i]
+       xmpyu           ${fni}L,${fm0}R,${fnm0} ; np[j]*m
+       fstds           ${fab1},0($xfer)
+        addl           $abhi,$hi0,$hi0         ; account carry bit
+       fstds           ${fnm1},8($xfer)
+        add            $ablo,$nmlo0,$nmlo0     ; discarded
+       ldw             0($tp),$ti1             ; tp[1]
+        addc           %r0,$nmhi0,$hi1
+       fstds           ${fab0},-16($xfer)
+       fstds           ${fnm0},-8($xfer)
+       ldw             4($xfer),$ablo
+       ldw             0($xfer),$abhi
+\f
+L\$inner_pa11
+       xmpyu           ${fai}R,${fbi},${fab1}  ; ap[j+1]*bp[i]
+       flddx           $idx($ap),${fai}        ; ap[j,j+1]
+       xmpyu           ${fni}R,${fm0}R,${fnm1} ; np[j+1]*m
+       flddx           $idx($np),${fni}        ; np[j,j+1]
+        add            $hi0,$ablo,$ablo
+       ldw             4($tp),$ti0             ; tp[j]
+        addc           %r0,$abhi,$abhi
+       ldw             12($xfer),$nmlo1
+        add            $ti1,$ablo,$ablo
+       ldw             8($xfer),$nmhi1
+        addc           %r0,$abhi,$hi0
+       fstds           ${fab1},0($xfer)
+        add            $ablo,$nmlo1,$nmlo1
+       fstds           ${fnm1},8($xfer)
+        addc           %r0,$nmhi1,$nmhi1
+       ldw             -12($xfer),$ablo
+        add            $hi1,$nmlo1,$nmlo1
+       ldw             -16($xfer),$abhi
+        addc           %r0,$nmhi1,$hi1
+
+       xmpyu           ${fai}L,${fbi},${fab0}  ; ap[j]*bp[i]
+       ldw             8($tp),$ti1             ; tp[j]
+       xmpyu           ${fni}L,${fm0}R,${fnm0} ; np[j]*m
+       ldw             -4($xfer),$nmlo0
+        add            $hi0,$ablo,$ablo
+       ldw             -8($xfer),$nmhi0
+        addc           %r0,$abhi,$abhi
+       stw             $nmlo1,-4($tp)          ; tp[j-1]
+        add            $ti0,$ablo,$ablo
+       fstds           ${fab0},-16($xfer)
+        addc           %r0,$abhi,$hi0
+       fstds           ${fnm0},-8($xfer)
+        add            $ablo,$nmlo0,$nmlo0
+       ldw             4($xfer),$ablo
+        addc           %r0,$nmhi0,$nmhi0
+       ldw             0($xfer),$abhi
+        add            $hi1,$nmlo0,$nmlo0
+        stws,ma        $nmlo0,8($tp)           ; tp[j-1]
+       addib,<>        8,$idx,L\$inner_pa11    ; j++++
+        addc           %r0,$nmhi0,$hi1
+
+       xmpyu           ${fai}R,${fbi},${fab1}  ; ap[j]*bp[i]
+       ldw             12($xfer),$nmlo1
+       xmpyu           ${fni}R,${fm0}R,${fnm1} ; np[j]*m
+       ldw             8($xfer),$nmhi1
+        add            $hi0,$ablo,$ablo
+       ldw             4($tp),$ti0             ; tp[j]
+        addc           %r0,$abhi,$abhi
+       fstds           ${fab1},0($xfer)
+        add            $ti1,$ablo,$ablo
+       fstds           ${fnm1},8($xfer)
+        addc           %r0,$abhi,$hi0
+       ldw             -16($xfer),$abhi
+        add            $ablo,$nmlo1,$nmlo1
+       ldw             -12($xfer),$ablo
+        addc           %r0,$nmhi1,$nmhi1
+       ldw             -8($xfer),$nmhi0
+        add            $hi1,$nmlo1,$nmlo1
+       ldw             -4($xfer),$nmlo0
+        addc           %r0,$nmhi1,$hi1
+
+       add             $hi0,$ablo,$ablo
+        stw            $nmlo1,-4($tp)          ; tp[j-1]
+       addc            %r0,$abhi,$abhi
+        add            $ti0,$ablo,$ablo
+       ldw             8($tp),$ti1             ; tp[j]
+        addc           %r0,$abhi,$hi0
+       ldw             0($xfer),$abhi
+        add            $ablo,$nmlo0,$nmlo0
+       ldw             4($xfer),$ablo
+        addc           %r0,$nmhi0,$nmhi0
+       ldws,mb         8($xfer),$nmhi1
+        add            $hi1,$nmlo0,$nmlo0
+       ldw             4($xfer),$nmlo1
+        addc           %r0,$nmhi0,$hi1
+        stws,ma        $nmlo0,8($tp)           ; tp[j-1]
+
+       addib,=         -1,$num,L\$outerdone_pa11; i--
+       subi            0,$arrsz,$idx           ; j=0
+
+        fldws,ma       4($bp),${fbi}           ; bp[i]
+        flddx          $idx($ap),${fai}        ; ap[0]
+       add             $hi0,$ablo,$ablo
+       addc            %r0,$abhi,$abhi
+        flddx          $idx($np),${fni}        ; np[0]
+        fldws          8($xfer),${fti}R        ; tp[0]
+       add             $ti1,$ablo,$ablo
+       addc            %r0,$abhi,$hi0
+
+        ldo            8($idx),$idx            ; j++++
+        xmpyu          ${fai}L,${fbi},${fab0}  ; ap[0]*bp[i]
+        xmpyu          ${fai}R,${fbi},${fab1}  ; ap[1]*bp[i]
+       ldw             4($tp),$ti0             ; tp[j]
+
+       add             $hi1,$nmlo1,$nmlo1
+       addc            %r0,$nmhi1,$nmhi1
+        fstws,mb       ${fab0}L,-8($xfer)      ; save high part
+       add             $ablo,$nmlo1,$nmlo1
+       addc            %r0,$nmhi1,$hi1
+        fcpy,sgl       %fr0,${fti}L            ; zero high part
+        fcpy,sgl       %fr0,${fab0}L
+       stw             $nmlo1,-4($tp)          ; tp[j-1]
+
+        fcnvxf,dbl,dbl ${fti},${fti}           ; 32-bit unsigned int -> double
+        fcnvxf,dbl,dbl ${fab0},${fab0}
+       add             $hi1,$hi0,$hi0
+       addc            %r0,%r0,$hi1
+        fadd,dbl       ${fti},${fab0},${fab0}  ; add tp[0]
+       add             $ti0,$hi0,$hi0
+       addc            %r0,$hi1,$hi1
+        fcnvfx,dbl,dbl ${fab0},${fab0}         ; double -> 33-bit unsigned int
+       stw             $hi0,0($tp)
+       stw             $hi1,4($tp)
+        xmpyu          ${fn0},${fab0}R,${fm0}
+
+       b               L\$outer_pa11
+       ldo             `$LOCALS+32+4`($fp),$tp
+\f
+L\$outerdone_pa11
+       add             $hi0,$ablo,$ablo
+       addc            %r0,$abhi,$abhi
+       add             $ti1,$ablo,$ablo
+       addc            %r0,$abhi,$hi0
+
+       ldw             4($tp),$ti0             ; tp[j]
+
+       add             $hi1,$nmlo1,$nmlo1
+       addc            %r0,$nmhi1,$nmhi1
+       add             $ablo,$nmlo1,$nmlo1
+       addc            %r0,$nmhi1,$hi1
+       stw             $nmlo1,-4($tp)          ; tp[j-1]
+
+       add             $hi1,$hi0,$hi0
+       addc            %r0,%r0,$hi1
+       add             $ti0,$hi0,$hi0
+       addc            %r0,$hi1,$hi1
+       stw             $hi0,0($tp)
+       stw             $hi1,4($tp)
+
+       ldo             `$LOCALS+32+4`($fp),$tp
+       sub             %r0,%r0,%r0             ; clear borrow
+       ldw             -4($tp),$ti0
+       addl            $tp,$arrsz,$tp
+L\$sub_pa11
+       ldwx            $idx($np),$hi0
+       subb            $ti0,$hi0,$hi1
+       ldwx            $idx($tp),$ti0
+       addib,<>        4,$idx,L\$sub_pa11
+       stws,ma         $hi1,4($rp)
+
+       subb            $ti0,%r0,$hi1
+       ldo             -4($tp),$tp
+       and             $tp,$hi1,$ap
+       andcm           $rp,$hi1,$bp
+       or              $ap,$bp,$np
+
+       sub             $rp,$arrsz,$rp          ; rewind rp
+       subi            0,$arrsz,$idx
+       ldo             `$LOCALS+32`($fp),$tp
+L\$copy_pa11
+       ldwx            $idx($np),$hi0
+       stws,ma         %r0,4($tp)
+       addib,<>        4,$idx,L\$copy_pa11
+       stws,ma         $hi0,4($rp)     
+
+       nop                                     ; alignment
+L\$done
+___
+}
+\f
+$code.=<<___;
+       ldi             1,%r28                  ; signal "handled"
+       ldo             $FRAME($fp),%sp         ; destroy tp[num+1]
+
+       $POP    `-$FRAME-$SAVED_RP`(%sp),%r2    ; standard epilogue
+       $POP    `-$FRAME+1*$SIZE_T`(%sp),%r4
+       $POP    `-$FRAME+2*$SIZE_T`(%sp),%r5
+       $POP    `-$FRAME+3*$SIZE_T`(%sp),%r6
+       $POP    `-$FRAME+4*$SIZE_T`(%sp),%r7
+       $POP    `-$FRAME+5*$SIZE_T`(%sp),%r8
+       $POP    `-$FRAME+6*$SIZE_T`(%sp),%r9
+       $POP    `-$FRAME+7*$SIZE_T`(%sp),%r10
+L\$abort
+       bv      (%r2)
+       .EXIT
+       $POPMB  -$FRAME(%sp),%r3
+       .PROCEND
+       .STRINGZ "Montgomery Multiplication for PA-RISC, CRYPTOGAMS by <appro\@openssl.org>"
+___
+\f
+# Explicitly encode PA-RISC 2.0 instructions used in this module, so
+# that it can be compiled with .LEVEL 1.0. It should be noted that I
+# wouldn't have to do this, if GNU assembler understood .ALLOW 2.0
+# directive...
+
+my $ldd = sub {
+  my ($mod,$args) = @_;
+  my $orig = "ldd$mod\t$args";
+
+    if ($args =~ /%r([0-9]+)\(%r([0-9]+)\),%r([0-9]+)/)                # format 4
+    {  my $opcode=(0x03<<26)|($2<<21)|($1<<16)|(3<<6)|$3;
+       sprintf "\t.WORD\t0x%08x\t; %s",$opcode,$orig;
+    }
+    elsif ($args =~ /(\-?[0-9]+)\(%r([0-9]+)\),%r([0-9]+)/)    # format 5
+    {  my $opcode=(0x03<<26)|($2<<21)|(1<<12)|(3<<6)|$3;
+       $opcode|=(($1&0xF)<<17)|(($1&0x10)<<12);                # encode offset
+       $opcode|=(1<<5)  if ($mod =~ /^,m/);
+       $opcode|=(1<<13) if ($mod =~ /^,mb/);
+       sprintf "\t.WORD\t0x%08x\t; %s",$opcode,$orig;
+    }
+    else { "\t".$orig; }
+};
+
+my $std = sub {
+  my ($mod,$args) = @_;
+  my $orig = "std$mod\t$args";
+
+    if ($args =~ /%r([0-9]+),(\-?[0-9]+)\(%r([0-9]+)\)/)       # format 6
+    {  my $opcode=(0x03<<26)|($3<<21)|($1<<16)|(1<<12)|(0xB<<6);
+       $opcode|=(($2&0xF)<<1)|(($2&0x10)>>4);                  # encode offset
+       $opcode|=(1<<5)  if ($mod =~ /^,m/);
+       $opcode|=(1<<13) if ($mod =~ /^,mb/);
+       sprintf "\t.WORD\t0x%08x\t; %s",$opcode,$orig;
+    }
+    else { "\t".$orig; }
+};
+
+my $extrd = sub {
+  my ($mod,$args) = @_;
+  my $orig = "extrd$mod\t$args";
+
+    # I only have ",u" completer, it's implicitly encoded...
+    if ($args =~ /%r([0-9]+),([0-9]+),([0-9]+),%r([0-9]+)/)    # format 15
+    {  my $opcode=(0x36<<26)|($1<<21)|($4<<16);
+       my $len=32-$3;
+       $opcode |= (($2&0x20)<<6)|(($2&0x1f)<<5);               # encode pos
+       $opcode |= (($len&0x20)<<7)|($len&0x1f);                # encode len
+       sprintf "\t.WORD\t0x%08x\t; %s",$opcode,$orig;
+    }
+    elsif ($args =~ /%r([0-9]+),%sar,([0-9]+),%r([0-9]+)/)     # format 12
+    {  my $opcode=(0x34<<26)|($1<<21)|($3<<16)|(2<<11)|(1<<9);
+       my $len=32-$2;
+       $opcode |= (($len&0x20)<<3)|($len&0x1f);                # encode len
+       $opcode |= (1<<13) if ($mod =~ /,\**=/);
+       sprintf "\t.WORD\t0x%08x\t; %s",$opcode,$orig;
+    }
+    else { "\t".$orig; }
+};
+
+my $shrpd = sub {
+  my ($mod,$args) = @_;
+  my $orig = "shrpd$mod\t$args";
+
+    if ($args =~ /%r([0-9]+),%r([0-9]+),([0-9]+),%r([0-9]+)/)  # format 14
+    {  my $opcode=(0x34<<26)|($2<<21)|($1<<16)|(1<<10)|$4;
+       my $cpos=63-$3;
+       $opcode |= (($cpos&0x20)<<6)|(($cpos&0x1f)<<5);         # encode sa
+       sprintf "\t.WORD\t0x%08x\t; %s",$opcode,$orig;
+    }
+    else { "\t".$orig; }
+};
+
+my $sub = sub {
+  my ($mod,$args) = @_;
+  my $orig = "sub$mod\t$args";
+
+    if ($mod eq ",db" && $args =~ /%r([0-9]+),%r([0-9]+),%r([0-9]+)/) {
+       my $opcode=(0x02<<26)|($2<<21)|($1<<16)|$3;
+       $opcode|=(1<<10);       # e1
+       $opcode|=(1<<8);        # e2
+       $opcode|=(1<<5);        # d
+       sprintf "\t.WORD\t0x%08x\t; %s",$opcode,$orig
+    }
+    else { "\t".$orig; }
+};
+
+sub assemble {
+  my ($mnemonic,$mod,$args)=@_;
+  my $opcode = eval("\$$mnemonic");
+
+    ref($opcode) eq 'CODE' ? &$opcode($mod,$args) : "\t$mnemonic$mod\t$args";
+}
+
+foreach (split("\n",$code)) {
+       s/\`([^\`]*)\`/eval $1/ge;
+       # flip word order in 64-bit mode...
+       s/(xmpyu\s+)($fai|$fni)([LR])/$1.$2.($3 eq "L"?"R":"L")/e if ($BN_SZ==8);
+       # assemble 2.0 instructions in 32-bit mode...
+       s/^\s+([a-z]+)([\S]*)\s+([\S]*)/&assemble($1,$2,$3)/e if ($BN_SZ==4);
+
+       print $_,"\n";
+}
+close STDOUT;
diff --git a/crypto/pariscid.pl b/crypto/pariscid.pl
new file mode 100644 (file)
index 0000000..477ec9b
--- /dev/null
@@ -0,0 +1,224 @@
+#!/usr/bin/env perl
+
+$flavour = shift;
+$output = shift;
+open STDOUT,">$output";
+
+if ($flavour =~ /64/) {
+       $LEVEL          ="2.0W";
+       $SIZE_T         =8;
+       $ST             ="std";
+} else {
+       $LEVEL          ="1.1";
+       $SIZE_T         =4;
+       $ST             ="stw";
+}
+
+$rp="%r2";
+$sp="%r30";
+$rv="%r28";
+
+$code=<<___;
+       .LEVEL  $LEVEL
+       .SPACE  \$TEXT\$
+       .SUBSPA \$CODE\$,QUAD=0,ALIGN=8,ACCESS=0x2C,CODE_ONLY
+
+       .EXPORT OPENSSL_cpuid_setup,ENTRY
+       .ALIGN  8
+OPENSSL_cpuid_setup
+       .PROC
+       .CALLINFO       NO_CALLS
+       .ENTRY
+       bv      ($rp)
+       .EXIT
+       nop
+       .PROCEND
+
+       .EXPORT OPENSSL_rdtsc,ENTRY
+       .ALIGN  8
+OPENSSL_rdtsc
+       .PROC
+       .CALLINFO       NO_CALLS
+       .ENTRY
+       mfctl   %cr16,$rv
+       bv      ($rp)
+       .EXIT
+       nop
+       .PROCEND
+
+       .EXPORT OPENSSL_wipe_cpu,ENTRY
+       .ALIGN  8
+OPENSSL_wipe_cpu
+       .PROC
+       .CALLINFO       NO_CALLS
+       .ENTRY
+       xor             %r0,%r0,%r1
+       fcpy,dbl        %fr0,%fr4
+       xor             %r0,%r0,%r19
+       fcpy,dbl        %fr0,%fr5
+       xor             %r0,%r0,%r20
+       fcpy,dbl        %fr0,%fr6
+       xor             %r0,%r0,%r21
+       fcpy,dbl        %fr0,%fr7
+       xor             %r0,%r0,%r22
+       fcpy,dbl        %fr0,%fr8
+       xor             %r0,%r0,%r23
+       fcpy,dbl        %fr0,%fr9
+       xor             %r0,%r0,%r24
+       fcpy,dbl        %fr0,%fr10
+       xor             %r0,%r0,%r25
+       fcpy,dbl        %fr0,%fr11
+       xor             %r0,%r0,%r26
+       fcpy,dbl        %fr0,%fr22
+       xor             %r0,%r0,%r29
+       fcpy,dbl        %fr0,%fr23
+       xor             %r0,%r0,%r31
+       fcpy,dbl        %fr0,%fr24
+       fcpy,dbl        %fr0,%fr25
+       fcpy,dbl        %fr0,%fr26
+       fcpy,dbl        %fr0,%fr27
+       fcpy,dbl        %fr0,%fr28
+       fcpy,dbl        %fr0,%fr29
+       fcpy,dbl        %fr0,%fr30
+       fcpy,dbl        %fr0,%fr31
+       bv              ($rp)
+       .EXIT
+       ldo             0($sp),$rv
+       .PROCEND
+___
+{
+my $inp="%r26";
+my $len="%r25";
+
+$code.=<<___;
+       .EXPORT OPENSSL_cleanse,ENTRY,ARGW0=GR,ARGW1=GR
+       .ALIGN  8
+OPENSSL_cleanse
+       .PROC
+       .CALLINFO       NO_CALLS
+       .ENTRY
+       cmpib,*=        0,$len,Ldone
+       nop
+       cmpib,*>>=      15,$len,Little
+       ldi             $SIZE_T-1,%r1
+
+Lalign
+       and,*<>         $inp,%r1,%r28
+       b,n             Laligned
+       stb             %r0,0($inp)
+       ldo             -1($len),$len
+       b               Lalign
+       ldo             1($inp),$inp
+
+Laligned
+       andcm           $len,%r1,%r28
+Lot
+       $ST             %r0,0($inp)
+       addib,*<>       -$SIZE_T,%r28,Lot
+       ldo             $SIZE_T($inp),$inp
+
+       and,*<>         $len,%r1,$len
+       b,n             Ldone
+Little
+       stb             %r0,0($inp)
+       addib,*<>       -1,$len,Little
+       ldo             1($inp),$inp
+Ldone
+       bv              ($rp)
+       .EXIT
+       nop
+       .PROCEND
+___
+}
+{
+my ($out,$cnt,$max)=("%r26","%r25","%r24");
+my ($tick,$lasttick)=("%r23","%r22");
+my ($diff,$lastdiff)=("%r21","%r20");
+
+$code.=<<___;
+       .EXPORT OPENSSL_instrument_bus,ENTRY,ARGW0=GR,ARGW1=GR
+       .ALIGN  8
+OPENSSL_instrument_bus
+       .PROC
+       .CALLINFO       NO_CALLS
+       .ENTRY
+       copy            $cnt,$rv
+       mfctl           %cr16,$tick
+       copy            $tick,$lasttick
+       ldi             0,$diff
+
+       fdc             0($out)
+       ldw             0($out),$tick
+       add             $diff,$tick,$tick
+       stw             $tick,0($out)
+Loop
+       mfctl           %cr16,$tick
+       sub             $tick,$lasttick,$diff
+       copy            $tick,$lasttick
+
+       fdc             0($out)
+       ldw             0($out),$tick
+       add             $diff,$tick,$tick
+       stw             $tick,0($out)
+
+       addib,<>        -1,$cnt,Loop
+       addi            4,$out,$out
+
+       bv              ($rp)
+       .EXIT
+       sub             $rv,$cnt,$rv
+       .PROCEND
+
+       .EXPORT OPENSSL_instrument_bus2,ENTRY,ARGW0=GR,ARGW1=GR
+       .ALIGN  8
+OPENSSL_instrument_bus2
+       .PROC
+       .CALLINFO       NO_CALLS
+       .ENTRY
+       copy            $cnt,$rv
+       sub             %r0,$cnt,$cnt
+
+       mfctl           %cr16,$tick
+       copy            $tick,$lasttick
+       ldi             0,$diff
+
+       fdc             0($out)
+       ldw             0($out),$tick
+       add             $diff,$tick,$tick
+       stw             $tick,0($out)
+
+       mfctl           %cr16,$tick
+       sub             $tick,$lasttick,$diff
+       copy            $tick,$lasttick
+Loop2
+       copy            $diff,$lastdiff
+       fdc             0($out)
+       ldw             0($out),$tick
+       add             $diff,$tick,$tick
+       stw             $tick,0($out)
+
+       addib,=         -1,$max,Ldone2
+       nop
+
+       mfctl           %cr16,$tick
+       sub             $tick,$lasttick,$diff
+       copy            $tick,$lasttick
+       cmpclr,<>       $lastdiff,$diff,$tick
+       ldi             1,$tick
+
+       ldi             1,%r1
+       xor             %r1,$tick,$tick
+       addb,<>         $tick,$cnt,Loop2
+       shladd,l        $tick,2,$out,$out
+Ldone2
+       bv              ($rp)
+       .EXIT
+       add             $rv,$cnt,$rv
+       .PROCEND
+___
+}
+$code =~ s/cmpib,\*/comib,/gm if ($SIZE_T==4);
+$code =~ s/,\*/,/gm if ($SIZE_T==4);
+print $code;
+close STDOUT;
+
diff --git a/crypto/rc4/asm/rc4-parisc.pl b/crypto/rc4/asm/rc4-parisc.pl
new file mode 100644 (file)
index 0000000..9e68196
--- /dev/null
@@ -0,0 +1,313 @@
+#!/usr/bin/env perl
+
+# ====================================================================
+# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+
+# RC4 for PA-RISC.
+
+# June 2009.
+#
+# Performance is 33% better than gcc 3.2 generated code on PA-7100LC.
+# For reference, [4x] unrolled loop is >40% faster than folded one.
+# It's possible to unroll loop 8 times on PA-RISC 2.0, but improvement
+# is believed to be not sufficient to justify the effort...
+#
+# Special thanks to polarhome.com for providing HP-UX account.
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+
+$flavour = shift;
+$output = shift;
+open STDOUT,">$output";
+
+if ($flavour =~ /64/) {
+       $LEVEL          ="2.0W";
+       $SIZE_T         =8;
+       $FRAME_MARKER   =80;
+       $SAVED_RP       =16;
+       $PUSH           ="std";
+       $PUSHMA         ="std,ma";
+       $POP            ="ldd";
+       $POPMB          ="ldd,mb";
+} else {
+       $LEVEL          ="1.0";
+       $SIZE_T         =4;
+       $FRAME_MARKER   =48;
+       $SAVED_RP       =20;
+       $PUSH           ="stw";
+       $PUSHMA         ="stwm";
+       $POP            ="ldw";
+       $POPMB          ="ldwm";
+}
+
+$FRAME=4*$SIZE_T+$FRAME_MARKER;        # 4 saved regs + frame marker
+                               #                [+ argument transfer]
+$SZ=1;                         # defaults to RC4_CHAR
+if (open CONF,"<${dir}../../opensslconf.h") {
+    while(<CONF>) {
+       if (m/#\s*define\s+RC4_INT\s+(.*)/) {
+           $SZ = ($1=~/char$/) ? 1 : 4;
+           last;
+       }
+    }
+    close CONF;
+}
+
+if ($SZ==1) {  # RC4_CHAR
+    $LD="ldb";
+    $LDX="ldbx";
+    $MKX="addl";
+    $ST="stb";
+} else {       # RC4_INT (~5% faster than RC4_CHAR on PA-7100LC)
+    $LD="ldw";
+    $LDX="ldwx,s";
+    $MKX="sh2addl";
+    $ST="stw";
+}
+
+$key="%r26";
+$len="%r25";
+$inp="%r24";
+$out="%r23";
+
+@XX=("%r19","%r20");
+@TX=("%r21","%r22");
+$YY="%r28";
+$TY="%r29";
+
+$acc="%r1";
+$ix="%r2";
+$iy="%r3";
+$dat0="%r4";
+$dat1="%r5";
+$rem="%r6";
+$mask="%r31";
+
+sub unrolledloopbody {
+for ($i=0;$i<4;$i++) {
+$code.=<<___;
+       ldo     1($XX[0]),$XX[1]
+       `sprintf("$LDX  %$TY(%$key),%$dat1") if ($i>0)` 
+       and     $mask,$XX[1],$XX[1]
+       $LDX    $YY($key),$TY
+       $MKX    $YY,$key,$ix
+       $LDX    $XX[1]($key),$TX[1]
+       $MKX    $XX[0],$key,$iy
+       $ST     $TX[0],0($ix)
+       comclr,<> $XX[1],$YY,%r0        ; conditional
+       copy    $TX[0],$TX[1]           ; move
+       `sprintf("%sdep %$dat1,%d,8,%$acc",$i==1?"z":"",8*($i-1)+7) if ($i>0)`
+       $ST     $TY,0($iy)
+       addl    $TX[0],$TY,$TY
+       addl    $TX[1],$YY,$YY
+       and     $mask,$TY,$TY
+       and     $mask,$YY,$YY
+___
+push(@TX,shift(@TX)); push(@XX,shift(@XX));    # "rotate" registers
+} }
+
+sub foldedloop {
+my ($label,$count)=@_;
+$code.=<<___;
+$label
+       $MKX    $YY,$key,$iy
+       $LDX    $YY($key),$TY
+       $MKX    $XX[0],$key,$ix
+       $ST     $TX[0],0($iy)
+       ldo     1($XX[0]),$XX[0]
+       $ST     $TY,0($ix)
+       addl    $TX[0],$TY,$TY
+       ldbx    $inp($out),$dat1
+       and     $mask,$TY,$TY
+       and     $mask,$XX[0],$XX[0]
+       $LDX    $TY($key),$acc
+       $LDX    $XX[0]($key),$TX[0]
+       ldo     1($out),$out
+       xor     $dat1,$acc,$acc
+       addl    $TX[0],$YY,$YY
+       stb     $acc,-1($out)
+       addib,<> -1,$count,$label       ; $count is always small
+       and     $mask,$YY,$YY
+___
+}
+
+$code=<<___;
+       .LEVEL  $LEVEL
+       .SPACE  \$TEXT\$
+       .SUBSPA \$CODE\$,QUAD=0,ALIGN=8,ACCESS=0x2C,CODE_ONLY
+
+       .EXPORT RC4,ENTRY,ARGW0=GR,ARGW1=GR,ARGW2=GR,ARGW3=GR
+RC4
+       .PROC
+       .CALLINFO       FRAME=`$FRAME-4*$SIZE_T`,NO_CALLS,SAVE_RP,ENTRY_GR=6
+       .ENTRY
+       $PUSH   %r2,-$SAVED_RP(%sp)     ; standard prologue
+       $PUSHMA %r3,$FRAME(%sp)
+       $PUSH   %r4,`-$FRAME+1*$SIZE_T`(%sp)
+       $PUSH   %r5,`-$FRAME+2*$SIZE_T`(%sp)
+       $PUSH   %r6,`-$FRAME+3*$SIZE_T`(%sp)
+
+       cmpib,*= 0,$len,L\$abort
+       sub     $inp,$out,$inp          ; distance between $inp and $out
+
+       $LD     `0*$SZ`($key),$XX[0]
+       $LD     `1*$SZ`($key),$YY
+       ldo     `2*$SZ`($key),$key
+
+       ldi     0xff,$mask
+       ldi     3,$dat0         
+
+       ldo     1($XX[0]),$XX[0]        ; warm up loop
+       and     $mask,$XX[0],$XX[0]
+       $LDX    $XX[0]($key),$TX[0]
+       addl    $TX[0],$YY,$YY
+       cmpib,*>>= 6,$len,L\$oop1       ; is $len large enough to bother?
+       and     $mask,$YY,$YY
+
+       and,<>  $out,$dat0,$rem         ; is $out aligned?
+       b       L\$alignedout
+       subi    4,$rem,$rem
+       sub     $len,$rem,$len
+___
+&foldedloop("L\$alignout",$rem);       # process till $out is aligned
+
+$code.=<<___;
+L\$alignedout                          ; $len is at least 4 here
+       and,<>  $inp,$dat0,$acc         ; is $inp aligned?
+       b       L\$oop4
+       sub     $inp,$acc,$rem          ; align $inp
+
+       sh3addl $acc,%r0,$acc
+       subi    32,$acc,$acc
+       mtctl   $acc,%cr11              ; load %sar with vshd align factor
+       ldwx    $rem($out),$dat0
+       ldo     4($rem),$rem
+L\$oop4misalignedinp
+___
+&unrolledloopbody();
+$code.=<<___;
+       $LDX    $TY($key),$ix
+       ldwx    $rem($out),$dat1
+       ldo     -4($len),$len
+       or      $ix,$acc,$acc           ; last piece, no need to dep
+       vshd    $dat0,$dat1,$iy         ; align data
+       copy    $dat1,$dat0
+       xor     $iy,$acc,$acc
+       stw     $acc,0($out)
+       cmpib,*<< 3,$len,L\$oop4misalignedinp
+       ldo     4($out),$out
+       cmpib,*= 0,$len,L\$done
+       nop
+       b       L\$oop1
+       nop
+
+       .ALIGN  8
+L\$oop4
+___
+&unrolledloopbody();
+$code.=<<___;
+       $LDX    $TY($key),$ix
+       ldwx    $inp($out),$dat0
+       ldo     -4($len),$len
+       or      $ix,$acc,$acc           ; last piece, no need to dep
+       xor     $dat0,$acc,$acc
+       stw     $acc,0($out)
+       cmpib,*<< 3,$len,L\$oop4
+       ldo     4($out),$out
+       cmpib,*= 0,$len,L\$done
+       nop
+___
+&foldedloop("L\$oop1",$len);
+$code.=<<___;
+L\$done
+       $POP    `-$FRAME-$SAVED_RP`(%sp),%r2
+       ldo     -1($XX[0]),$XX[0]       ; chill out loop
+       sub     $YY,$TX[0],$YY
+       and     $mask,$XX[0],$XX[0]
+       and     $mask,$YY,$YY
+       $ST     $XX[0],`-2*$SZ`($key)
+       $ST     $YY,`-1*$SZ`($key)
+       $POP    `-$FRAME+1*$SIZE_T`(%sp),%r4
+       $POP    `-$FRAME+2*$SIZE_T`(%sp),%r5
+       $POP    `-$FRAME+3*$SIZE_T`(%sp),%r6
+L\$abort
+       bv      (%r2)
+       .EXIT
+       $POPMB  -$FRAME(%sp),%r3
+       .PROCEND
+___
+
+$code.=<<___;
+
+       .EXPORT RC4_set_key,ENTRY,ARGW0=GR,ARGW1=GR,ARGW2=GR
+       .ALIGN  8
+RC4_set_key
+       .PROC
+       .CALLINFO       NO_CALLS
+       .ENTRY
+       $ST     %r0,`0*$SZ`($key)
+       $ST     %r0,`1*$SZ`($key)
+       ldo     `2*$SZ`($key),$key
+       copy    %r0,@XX[0]
+L\$1st
+       $ST     @XX[0],0($key)
+       ldo     1(@XX[0]),@XX[0]
+       bb,>=   @XX[0],`31-8`,L\$1st    ; @XX[0]<256
+       ldo     $SZ($key),$key
+
+       ldo     `-256*$SZ`($key),$key   ; rewind $key
+       addl    $len,$inp,$inp          ; $inp to point at the end
+       sub     %r0,$len,%r23           ; inverse index
+       copy    %r0,@XX[0]
+       copy    %r0,@XX[1]
+       ldi     0xff,$mask
+
+L\$2nd
+       $LDX    @XX[0]($key),@TX[0]
+       ldbx    %r23($inp),@TX[1]
+       addi,nuv 1,%r23,%r23            ; increment and conditional
+       sub     %r0,$len,%r23           ; inverse index
+       addl    @TX[0],@XX[1],@XX[1]
+       addl    @TX[1],@XX[1],@XX[1]
+       and     $mask,@XX[1],@XX[1]
+       $MKX    @XX[0],$key,$TY
+       $LDX    @XX[1]($key),@TX[1]
+       $MKX    @XX[1],$key,$YY
+       ldo     1(@XX[0]),@XX[0]
+       $ST     @TX[0],0($YY)
+       bb,>=   @XX[0],`31-8`,L\$2nd    ; @XX[0]<256
+       $ST     @TX[1],0($TY)
+
+       bv,n    (%r2)
+       .EXIT
+       nop
+       .PROCEND
+
+       .EXPORT RC4_options,ENTRY
+       .ALIGN  8
+RC4_options
+       .PROC
+       .CALLINFO       NO_CALLS
+       .ENTRY
+       blr     %r0,%r28
+       ldi     3,%r1
+L\$pic
+       andcm   %r28,%r1,%r28
+       bv      (%r2)
+       .EXIT
+       ldo     L\$opts-L\$pic(%r28),%r28
+       .PROCEND
+       .ALIGN  8
+L\$opts
+       .STRINGZ "rc4(4x,`$SZ==1?"char":"int"`)"
+       .STRINGZ "RC4 for PA-RISC, CRYPTOGAMS by <appro\@openssl.org>"
+___
+$code =~ s/\`([^\`]*)\`/eval $1/gem;
+$code =~ s/cmpib,\*/comib,/gm if ($SIZE_T==4);
+
+print $code;
+close STDOUT;
diff --git a/crypto/sha/asm/sha1-parisc.pl b/crypto/sha/asm/sha1-parisc.pl
new file mode 100644 (file)
index 0000000..6d7bf49
--- /dev/null
@@ -0,0 +1,259 @@
+#!/usr/bin/env perl
+
+# ====================================================================
+# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+
+# SHA1 block procedure for PA-RISC.
+
+# June 2009.
+#
+# On PA-7100LC performance is >30% better than gcc 3.2 generated code
+# for aligned input and >50% better for unaligned. Compared to vendor
+# compiler on PA-8600 it's almost 60% faster in 64-bit build and just
+# few percent faster in 32-bit one (this for aligned input, data for
+# unaligned input is not available).
+#
+# Special thanks to polarhome.com for providing HP-UX account.
+
+$flavour = shift;
+$output = shift;
+open STDOUT,">$output";
+
+if ($flavour =~ /64/) {
+       $LEVEL          ="2.0W";
+       $SIZE_T         =8;
+       $FRAME_MARKER   =80;
+       $SAVED_RP       =16;
+       $PUSH           ="std";
+       $PUSHMA         ="std,ma";
+       $POP            ="ldd";
+       $POPMB          ="ldd,mb";
+} else {
+       $LEVEL          ="1.0";
+       $SIZE_T         =4;
+       $FRAME_MARKER   =48;
+       $SAVED_RP       =20;
+       $PUSH           ="stw";
+       $PUSHMA         ="stwm";
+       $POP            ="ldw";
+       $POPMB          ="ldwm";
+}
+
+$FRAME=14*$SIZE_T+$FRAME_MARKER;# 14 saved regs + frame marker
+                               #                 [+ argument transfer]
+$ctx="%r26";           # arg0
+$inp="%r25";           # arg1
+$num="%r24";           # arg2
+
+$t0="%r28";
+$t1="%r29";
+$K="%r31";
+
+@X=("%r1", "%r2", "%r3", "%r4", "%r5", "%r6", "%r7", "%r8",
+    "%r9", "%r10","%r11","%r12","%r13","%r14","%r15","%r16",$t0);
+
+@V=($A,$B,$C,$D,$E)=("%r19","%r20","%r21","%r22","%r23");
+
+sub BODY_00_19 {
+my ($i,$a,$b,$c,$d,$e)=@_;
+my $j=$i+1;
+$code.=<<___ if ($i<15);
+       addl    $K,$e,$e        ; $i
+       shd     $a,$a,27,$t1
+       addl    @X[$i],$e,$e
+       and     $c,$b,$t0
+       addl    $t1,$e,$e
+       andcm   $d,$b,$t1
+       shd     $b,$b,2,$b
+       or      $t1,$t0,$t0
+       addl    $t0,$e,$e
+___
+$code.=<<___ if ($i>=15);      # with forward Xupdate
+       addl    $K,$e,$e        ; $i
+       shd     $a,$a,27,$t1
+       xor     @X[($j+2)%16],@X[$j%16],@X[$j%16]
+       addl    @X[$i%16],$e,$e
+       and     $c,$b,$t0
+       xor     @X[($j+8)%16],@X[$j%16],@X[$j%16]
+       addl    $t1,$e,$e
+       andcm   $d,$b,$t1
+       shd     $b,$b,2,$b
+       or      $t1,$t0,$t0
+       xor     @X[($j+13)%16],@X[$j%16],@X[$j%16]
+       add     $t0,$e,$e
+       shd     @X[$j%16],@X[$j%16],31,@X[$j%16]
+___
+}
+
+sub BODY_20_39 {
+my ($i,$a,$b,$c,$d,$e)=@_;
+my $j=$i+1;
+$code.=<<___ if ($i<79);
+       xor     @X[($j+2)%16],@X[$j%16],@X[$j%16]       ; $i
+       addl    $K,$e,$e
+       shd     $a,$a,27,$t1
+       xor     @X[($j+8)%16],@X[$j%16],@X[$j%16]
+       addl    @X[$i%16],$e,$e
+       xor     $b,$c,$t0
+       xor     @X[($j+13)%16],@X[$j%16],@X[$j%16]
+       addl    $t1,$e,$e
+       shd     $b,$b,2,$b
+       xor     $d,$t0,$t0
+       shd     @X[$j%16],@X[$j%16],31,@X[$j%16]
+       addl    $t0,$e,$e
+___
+$code.=<<___ if ($i==79);      # with context load
+       ldw     0($ctx),@X[0]   ; $i
+       addl    $K,$e,$e
+       shd     $a,$a,27,$t1
+       ldw     4($ctx),@X[1]
+       addl    @X[$i%16],$e,$e
+       xor     $b,$c,$t0
+       ldw     8($ctx),@X[2]
+       addl    $t1,$e,$e
+       shd     $b,$b,2,$b
+       xor     $d,$t0,$t0
+       ldw     12($ctx),@X[3]
+       addl    $t0,$e,$e
+       ldw     16($ctx),@X[4]
+___
+}
+
+sub BODY_40_59 {
+my ($i,$a,$b,$c,$d,$e)=@_;
+my $j=$i+1;
+$code.=<<___;
+       shd     $a,$a,27,$t1    ; $i
+       addl    $K,$e,$e
+       xor     @X[($j+2)%16],@X[$j%16],@X[$j%16]
+       xor     $d,$c,$t0
+       addl    @X[$i%16],$e,$e
+       xor     @X[($j+8)%16],@X[$j%16],@X[$j%16]
+       and     $b,$t0,$t0
+       addl    $t1,$e,$e
+       shd     $b,$b,2,$b
+       xor     @X[($j+13)%16],@X[$j%16],@X[$j%16]
+       addl    $t0,$e,$e
+       and     $d,$c,$t1
+       shd     @X[$j%16],@X[$j%16],31,@X[$j%16]
+       addl    $t1,$e,$e
+___
+}
+
+$code=<<___;
+       .LEVEL  $LEVEL
+       .SPACE  \$TEXT\$
+       .SUBSPA \$CODE\$,QUAD=0,ALIGN=8,ACCESS=0x2C,CODE_ONLY
+
+       .EXPORT sha1_block_data_order,ENTRY,ARGW0=GR,ARGW1=GR,ARGW2=GR
+sha1_block_data_order
+       .PROC
+       .CALLINFO       FRAME=`$FRAME-14*$SIZE_T`,NO_CALLS,SAVE_RP,ENTRY_GR=16
+       .ENTRY
+       $PUSH   %r2,-$SAVED_RP(%sp)     ; standard prologue
+       $PUSHMA %r3,$FRAME(%sp)
+       $PUSH   %r4,`-$FRAME+1*$SIZE_T`(%sp)
+       $PUSH   %r5,`-$FRAME+2*$SIZE_T`(%sp)
+       $PUSH   %r6,`-$FRAME+3*$SIZE_T`(%sp)
+       $PUSH   %r7,`-$FRAME+4*$SIZE_T`(%sp)
+       $PUSH   %r8,`-$FRAME+5*$SIZE_T`(%sp)
+       $PUSH   %r9,`-$FRAME+6*$SIZE_T`(%sp)
+       $PUSH   %r10,`-$FRAME+7*$SIZE_T`(%sp)
+       $PUSH   %r11,`-$FRAME+8*$SIZE_T`(%sp)
+       $PUSH   %r12,`-$FRAME+9*$SIZE_T`(%sp)
+       $PUSH   %r13,`-$FRAME+10*$SIZE_T`(%sp)
+       $PUSH   %r14,`-$FRAME+11*$SIZE_T`(%sp)
+       $PUSH   %r15,`-$FRAME+12*$SIZE_T`(%sp)
+       $PUSH   %r16,`-$FRAME+13*$SIZE_T`(%sp)
+
+       ldw     0($ctx),$A
+       ldw     4($ctx),$B
+       ldw     8($ctx),$C
+       ldw     12($ctx),$D
+       ldw     16($ctx),$E
+
+       extru   $inp,31,2,$t0           ; t0=inp&3;
+       sh3addl $t0,%r0,$t0             ; t0*=8;
+       subi    32,$t0,$t0              ; t0=32-t0;
+       mtctl   $t0,%cr11               ; %sar=t0;
+
+L\$oop
+       ldi     3,$t0
+       andcm   $inp,$t0,$t0            ; 64-bit neutral
+___
+       for ($i=0;$i<15;$i++) {         # load input block
+       $code.="\tldw   `4*$i`($t0),@X[$i]\n";          }
+$code.=<<___;
+       cmpb,*= $inp,$t0,L\$aligned
+       ldw     60($t0),@X[15]
+       ldw     64($t0),@X[16]
+___
+       for ($i=0;$i<16;$i++) {         # align input
+       $code.="\tvshd  @X[$i],@X[$i+1],@X[$i]\n";      }
+$code.=<<___;
+L\$aligned
+       ldil    L'0x5a827000,$K         ; K_00_19
+       ldo     0x999($K),$K
+___
+for ($i=0;$i<20;$i++)   { &BODY_00_19($i,@V); unshift(@V,pop(@V)); }
+$code.=<<___;
+       ldil    L'0x6ed9e000,$K         ; K_20_39
+       ldo     0xba1($K),$K
+___
+
+for (;$i<40;$i++)       { &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
+$code.=<<___;
+       ldil    L'0x8f1bb000,$K         ; K_40_59
+       ldo     0xcdc($K),$K
+___
+
+for (;$i<60;$i++)       { &BODY_40_59($i,@V); unshift(@V,pop(@V)); }
+$code.=<<___;
+       ldil    L'0xca62c000,$K         ; K_60_79
+       ldo     0x1d6($K),$K
+___
+for (;$i<80;$i++)       { &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
+
+$code.=<<___;
+       addl    @X[0],$A,$A
+       addl    @X[1],$B,$B
+       addl    @X[2],$C,$C
+       addl    @X[3],$D,$D
+       addl    @X[4],$E,$E
+       stw     $A,0($ctx)
+       stw     $B,4($ctx)
+       stw     $C,8($ctx)
+       stw     $D,12($ctx)
+       stw     $E,16($ctx)
+       addib,*<> -1,$num,L\$oop
+       ldo     64($inp),$inp
+
+       $POP    `-$FRAME-$SAVED_RP`(%sp),%r2    ; standard epilogue
+       $POP    `-$FRAME+1*$SIZE_T`(%sp),%r4
+       $POP    `-$FRAME+2*$SIZE_T`(%sp),%r5
+       $POP    `-$FRAME+3*$SIZE_T`(%sp),%r6
+       $POP    `-$FRAME+4*$SIZE_T`(%sp),%r7
+       $POP    `-$FRAME+5*$SIZE_T`(%sp),%r8
+       $POP    `-$FRAME+6*$SIZE_T`(%sp),%r9
+       $POP    `-$FRAME+7*$SIZE_T`(%sp),%r10
+       $POP    `-$FRAME+8*$SIZE_T`(%sp),%r11
+       $POP    `-$FRAME+9*$SIZE_T`(%sp),%r12
+       $POP    `-$FRAME+10*$SIZE_T`(%sp),%r13
+       $POP    `-$FRAME+11*$SIZE_T`(%sp),%r14
+       $POP    `-$FRAME+12*$SIZE_T`(%sp),%r15
+       $POP    `-$FRAME+13*$SIZE_T`(%sp),%r16
+       bv      (%r2)
+       .EXIT
+       $POPMB  -$FRAME(%sp),%r3
+       .PROCEND
+       .STRINGZ "SHA1 block transform for PA-RISC, CRYPTOGAMS by <appro\@openssl.org>"
+___
+
+$code =~ s/\`([^\`]*)\`/eval $1/gem;
+$code =~ s/,\*/,/gm if ($SIZE_T==4);
+print $code;
+close STDOUT;
diff --git a/crypto/sha/asm/sha512-parisc.pl b/crypto/sha/asm/sha512-parisc.pl
new file mode 100755 (executable)
index 0000000..e24ee58
--- /dev/null
@@ -0,0 +1,791 @@
+#!/usr/bin/env perl
+
+# ====================================================================
+# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+
+# SHA256/512 block procedure for PA-RISC.
+
+# June 2009.
+#
+# SHA256 performance is >75% better than gcc 3.2 generated code on
+# PA-7100LC. Compared to code generated by vendor compiler this
+# implementation is almost 70% faster in 64-bit build, but delivers
+# virtually same performance in 32-bit build on PA-8600.
+#
+# SHA512 performance is >2.9x better than gcc 3.2 generated code on
+# PA-7100LC, PA-RISC 1.1 processor. Then implementation detects if the
+# code is executed on PA-RISC 2.0 processor and switches to 64-bit
+# code path delivering adequate peformance even in "blended" 32-bit
+# build. Though 64-bit code is not any faster than code generated by
+# vendor compiler on PA-8600...
+#
+# Special thanks to polarhome.com for providing HP-UX account.
+
+$flavour = shift;
+$output = shift;
+open STDOUT,">$output";
+
+if ($flavour =~ /64/) {
+       $LEVEL          ="2.0W";
+       $SIZE_T         =8;
+       $FRAME_MARKER   =80;
+       $SAVED_RP       =16;
+       $PUSH           ="std";
+       $PUSHMA         ="std,ma";
+       $POP            ="ldd";
+       $POPMB          ="ldd,mb";
+} else {
+       $LEVEL          ="1.0";
+       $SIZE_T         =4;
+       $FRAME_MARKER   =48;
+       $SAVED_RP       =20;
+       $PUSH           ="stw";
+       $PUSHMA         ="stwm";
+       $POP            ="ldw";
+       $POPMB          ="ldwm";
+}
+
+if ($output =~ /512/) {
+       $func="sha512_block_data_order";
+       $SZ=8;
+       @Sigma0=(28,34,39);
+       @Sigma1=(14,18,41);
+       @sigma0=(1,  8, 7);
+       @sigma1=(19,61, 6);
+       $rounds=80;
+       $LAST10BITS=0x017;
+       $LD="ldd";
+       $LDM="ldd,ma";
+       $ST="std";
+} else {
+       $func="sha256_block_data_order";
+       $SZ=4;
+       @Sigma0=( 2,13,22);
+       @Sigma1=( 6,11,25);
+       @sigma0=( 7,18, 3);
+       @sigma1=(17,19,10);
+       $rounds=64;
+       $LAST10BITS=0x0f2;
+       $LD="ldw";
+       $LDM="ldwm";
+       $ST="stw";
+}
+
+$FRAME=16*$SIZE_T+$FRAME_MARKER;# 16 saved regs + frame marker
+                               #                 [+ argument transfer]
+$XOFF=16*$SZ+32;               # local variables
+$FRAME+=$XOFF;
+$XOFF+=$FRAME_MARKER;          # distance between %sp and local variables
+
+$ctx="%r26";   # zapped by $a0
+$inp="%r25";   # zapped by $a1
+$num="%r24";   # zapped by $t0
+
+$a0 ="%r26";
+$a1 ="%r25";
+$t0 ="%r24";
+$t1 ="%r29";
+$Tbl="%r31";
+
+@V=($A,$B,$C,$D,$E,$F,$G,$H)=("%r17","%r18","%r19","%r20","%r21","%r22","%r23","%r28");
+
+@X=("%r1", "%r2", "%r3", "%r4", "%r5", "%r6", "%r7", "%r8",
+    "%r9", "%r10","%r11","%r12","%r13","%r14","%r15","%r16",$inp);
+
+sub ROUND_00_15 {
+my ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_;
+$code.=<<___;
+       _ror    $e,$Sigma1[0],$a0
+       and     $f,$e,$t0
+       _ror    $e,$Sigma1[1],$a1
+       addl    $t1,$h,$h
+       andcm   $g,$e,$t1
+       xor     $a1,$a0,$a0
+       _ror    $a1,`$Sigma1[2]-$Sigma1[1]`,$a1
+       or      $t0,$t1,$t1             ; Ch(e,f,g)
+       addl    @X[$i%16],$h,$h
+       xor     $a0,$a1,$a1             ; Sigma1(e)
+       addl    $t1,$h,$h
+       _ror    $a,$Sigma0[0],$a0
+       addl    $a1,$h,$h
+
+       _ror    $a,$Sigma0[1],$a1
+       and     $a,$b,$t0
+       and     $a,$c,$t1
+       xor     $a1,$a0,$a0
+       _ror    $a1,`$Sigma0[2]-$Sigma0[1]`,$a1
+       xor     $t1,$t0,$t0
+       and     $b,$c,$t1
+       xor     $a0,$a1,$a1             ; Sigma0(a)
+       addl    $h,$d,$d
+       xor     $t1,$t0,$t0             ; Maj(a,b,c)
+       `"$LDM  $SZ($Tbl),$t1" if ($i<15)`
+       addl    $a1,$h,$h
+       addl    $t0,$h,$h
+
+___
+}
+
+sub ROUND_16_xx {
+my ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_;
+$i-=16;
+$code.=<<___;
+       _ror    @X[($i+1)%16],$sigma0[0],$a0
+       _ror    @X[($i+1)%16],$sigma0[1],$a1
+       addl    @X[($i+9)%16],@X[$i],@X[$i]
+       _ror    @X[($i+14)%16],$sigma1[0],$t0
+       _ror    @X[($i+14)%16],$sigma1[1],$t1
+       xor     $a1,$a0,$a0
+       _shr    @X[($i+1)%16],$sigma0[2],$a1
+       xor     $t1,$t0,$t0
+       _shr    @X[($i+14)%16],$sigma1[2],$t1
+       xor     $a1,$a0,$a0             ; sigma0(X[(i+1)&0x0f])
+       xor     $t1,$t0,$t0             ; sigma1(X[(i+14)&0x0f])
+       $LDM    $SZ($Tbl),$t1
+       addl    $a0,@X[$i],@X[$i]
+       addl    $t0,@X[$i],@X[$i]
+___
+$code.=<<___ if ($i==15);
+       extru   $t1,31,10,$a1
+       comiclr,<> $LAST10BITS,$a1,%r0
+       ldo     1($Tbl),$Tbl            ; signal end of $Tbl
+___
+&ROUND_00_15($i+16,$a,$b,$c,$d,$e,$f,$g,$h);
+}
+
+$code=<<___;
+       .LEVEL  $LEVEL
+       .SPACE  \$TEXT\$
+       .SUBSPA \$CODE\$,QUAD=0,ALIGN=8,ACCESS=0x2C,CODE_ONLY
+
+       .ALIGN  64
+L\$table
+___
+$code.=<<___ if ($SZ==8);
+       .WORD   0x428a2f98,0xd728ae22,0x71374491,0x23ef65cd
+       .WORD   0xb5c0fbcf,0xec4d3b2f,0xe9b5dba5,0x8189dbbc
+       .WORD   0x3956c25b,0xf348b538,0x59f111f1,0xb605d019
+       .WORD   0x923f82a4,0xaf194f9b,0xab1c5ed5,0xda6d8118
+       .WORD   0xd807aa98,0xa3030242,0x12835b01,0x45706fbe
+       .WORD   0x243185be,0x4ee4b28c,0x550c7dc3,0xd5ffb4e2
+       .WORD   0x72be5d74,0xf27b896f,0x80deb1fe,0x3b1696b1
+       .WORD   0x9bdc06a7,0x25c71235,0xc19bf174,0xcf692694
+       .WORD   0xe49b69c1,0x9ef14ad2,0xefbe4786,0x384f25e3
+       .WORD   0x0fc19dc6,0x8b8cd5b5,0x240ca1cc,0x77ac9c65
+       .WORD   0x2de92c6f,0x592b0275,0x4a7484aa,0x6ea6e483
+       .WORD   0x5cb0a9dc,0xbd41fbd4,0x76f988da,0x831153b5
+       .WORD   0x983e5152,0xee66dfab,0xa831c66d,0x2db43210
+       .WORD   0xb00327c8,0x98fb213f,0xbf597fc7,0xbeef0ee4
+       .WORD   0xc6e00bf3,0x3da88fc2,0xd5a79147,0x930aa725
+       .WORD   0x06ca6351,0xe003826f,0x14292967,0x0a0e6e70
+       .WORD   0x27b70a85,0x46d22ffc,0x2e1b2138,0x5c26c926
+       .WORD   0x4d2c6dfc,0x5ac42aed,0x53380d13,0x9d95b3df
+       .WORD   0x650a7354,0x8baf63de,0x766a0abb,0x3c77b2a8
+       .WORD   0x81c2c92e,0x47edaee6,0x92722c85,0x1482353b
+       .WORD   0xa2bfe8a1,0x4cf10364,0xa81a664b,0xbc423001
+       .WORD   0xc24b8b70,0xd0f89791,0xc76c51a3,0x0654be30
+       .WORD   0xd192e819,0xd6ef5218,0xd6990624,0x5565a910
+       .WORD   0xf40e3585,0x5771202a,0x106aa070,0x32bbd1b8
+       .WORD   0x19a4c116,0xb8d2d0c8,0x1e376c08,0x5141ab53
+       .WORD   0x2748774c,0xdf8eeb99,0x34b0bcb5,0xe19b48a8
+       .WORD   0x391c0cb3,0xc5c95a63,0x4ed8aa4a,0xe3418acb
+       .WORD   0x5b9cca4f,0x7763e373,0x682e6ff3,0xd6b2b8a3
+       .WORD   0x748f82ee,0x5defb2fc,0x78a5636f,0x43172f60
+       .WORD   0x84c87814,0xa1f0ab72,0x8cc70208,0x1a6439ec
+       .WORD   0x90befffa,0x23631e28,0xa4506ceb,0xde82bde9
+       .WORD   0xbef9a3f7,0xb2c67915,0xc67178f2,0xe372532b
+       .WORD   0xca273ece,0xea26619c,0xd186b8c7,0x21c0c207
+       .WORD   0xeada7dd6,0xcde0eb1e,0xf57d4f7f,0xee6ed178
+       .WORD   0x06f067aa,0x72176fba,0x0a637dc5,0xa2c898a6
+       .WORD   0x113f9804,0xbef90dae,0x1b710b35,0x131c471b
+       .WORD   0x28db77f5,0x23047d84,0x32caab7b,0x40c72493
+       .WORD   0x3c9ebe0a,0x15c9bebc,0x431d67c4,0x9c100d4c
+       .WORD   0x4cc5d4be,0xcb3e42b6,0x597f299c,0xfc657e2a
+       .WORD   0x5fcb6fab,0x3ad6faec,0x6c44198c,0x4a475817
+___
+$code.=<<___ if ($SZ==4);
+       .WORD   0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5
+       .WORD   0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5
+       .WORD   0xd807aa98,0x12835b01,0x243185be,0x550c7dc3
+       .WORD   0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174
+       .WORD   0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc
+       .WORD   0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da
+       .WORD   0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7
+       .WORD   0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967
+       .WORD   0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13
+       .WORD   0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85
+       .WORD   0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3
+       .WORD   0xd192e819,0xd6990624,0xf40e3585,0x106aa070
+       .WORD   0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5
+       .WORD   0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3
+       .WORD   0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208
+       .WORD   0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2
+___
+$code.=<<___;
+
+       .EXPORT $func,ENTRY,ARGW0=GR,ARGW1=GR,ARGW2=GR
+       .ALIGN  64
+$func
+       .PROC
+       .CALLINFO       FRAME=`$FRAME-16*$SIZE_T`,NO_CALLS,SAVE_RP,ENTRY_GR=18
+       .ENTRY
+       $PUSH   %r2,-$SAVED_RP(%sp)     ; standard prologue
+       $PUSHMA %r3,$FRAME(%sp)
+       $PUSH   %r4,`-$FRAME+1*$SIZE_T`(%sp)
+       $PUSH   %r5,`-$FRAME+2*$SIZE_T`(%sp)
+       $PUSH   %r6,`-$FRAME+3*$SIZE_T`(%sp)
+       $PUSH   %r7,`-$FRAME+4*$SIZE_T`(%sp)
+       $PUSH   %r8,`-$FRAME+5*$SIZE_T`(%sp)
+       $PUSH   %r9,`-$FRAME+6*$SIZE_T`(%sp)
+       $PUSH   %r10,`-$FRAME+7*$SIZE_T`(%sp)
+       $PUSH   %r11,`-$FRAME+8*$SIZE_T`(%sp)
+       $PUSH   %r12,`-$FRAME+9*$SIZE_T`(%sp)
+       $PUSH   %r13,`-$FRAME+10*$SIZE_T`(%sp)
+       $PUSH   %r14,`-$FRAME+11*$SIZE_T`(%sp)
+       $PUSH   %r15,`-$FRAME+12*$SIZE_T`(%sp)
+       $PUSH   %r16,`-$FRAME+13*$SIZE_T`(%sp)
+       $PUSH   %r17,`-$FRAME+14*$SIZE_T`(%sp)
+       $PUSH   %r18,`-$FRAME+15*$SIZE_T`(%sp)
+
+       _shl    $num,`log(16*$SZ)/log(2)`,$num
+       addl    $inp,$num,$num          ; $num to point at the end of $inp
+
+       $PUSH   $num,`-$FRAME_MARKER-4*$SIZE_T`(%sp)    ; save arguments
+       $PUSH   $inp,`-$FRAME_MARKER-3*$SIZE_T`(%sp)
+       $PUSH   $ctx,`-$FRAME_MARKER-2*$SIZE_T`(%sp)
+
+       blr     %r0,$Tbl
+       ldi     3,$t1
+L\$pic
+       andcm   $Tbl,$t1,$Tbl           ; wipe privilege level
+       ldo     L\$table-L\$pic($Tbl),$Tbl
+___
+$code.=<<___ if ($SZ==8 && $SIZE_T==4);
+       ldi     31,$t1
+       mtctl   $t1,%cr11
+       extrd,u,*= $t1,%sar,1,$t1       ; executes on PA-RISC 1.0
+       b       L\$parisc1
+       nop
+___
+$code.=<<___;
+       $LD     `0*$SZ`($ctx),$A        ; load context
+       $LD     `1*$SZ`($ctx),$B
+       $LD     `2*$SZ`($ctx),$C
+       $LD     `3*$SZ`($ctx),$D
+       $LD     `4*$SZ`($ctx),$E
+       $LD     `5*$SZ`($ctx),$F
+       $LD     `6*$SZ`($ctx),$G
+       $LD     `7*$SZ`($ctx),$H
+
+       extru   $inp,31,`log($SZ)/log(2)`,$t0
+       sh3addl $t0,%r0,$t0
+       subi    `8*$SZ`,$t0,$t0
+       mtctl   $t0,%cr11               ; load %sar with align factor
+
+L\$oop
+       ldi     `$SZ-1`,$t0
+       $LDM    $SZ($Tbl),$t1
+       andcm   $inp,$t0,$t0            ; align $inp
+___
+       for ($i=0;$i<15;$i++) {         # load input block
+       $code.="\t$LD   `$SZ*$i`($t0),@X[$i]\n";                }
+$code.=<<___;
+       cmpb,*= $inp,$t0,L\$aligned
+       $LD     `$SZ*15`($t0),@X[15]
+       $LD     `$SZ*16`($t0),@X[16]
+___
+       for ($i=0;$i<16;$i++) {         # align data
+       $code.="\t_align        @X[$i],@X[$i+1],@X[$i]\n";      }
+$code.=<<___;
+L\$aligned
+       nop     ; otherwise /usr/ccs/bin/as is confused by below .WORD
+___
+
+for($i=0;$i<16;$i++)   { &ROUND_00_15($i,@V); unshift(@V,pop(@V)); }
+$code.=<<___;
+L\$rounds
+       nop     ; otherwise /usr/ccs/bin/as is confused by below .WORD
+___
+for(;$i<32;$i++)       { &ROUND_16_xx($i,@V); unshift(@V,pop(@V)); }
+$code.=<<___;
+       bb,>=   $Tbl,31,L\$rounds       ; end of $Tbl signalled?
+       nop
+
+       $POP    `-$FRAME_MARKER-2*$SIZE_T`(%sp),$ctx    ; restore arguments
+       $POP    `-$FRAME_MARKER-3*$SIZE_T`(%sp),$inp
+       $POP    `-$FRAME_MARKER-4*$SIZE_T`(%sp),$num
+       ldo     `-$rounds*$SZ-1`($Tbl),$Tbl             ; rewind $Tbl
+
+       $LD     `0*$SZ`($ctx),@X[0]     ; load context
+       $LD     `1*$SZ`($ctx),@X[1]
+       $LD     `2*$SZ`($ctx),@X[2]
+       $LD     `3*$SZ`($ctx),@X[3]
+       $LD     `4*$SZ`($ctx),@X[4]
+       $LD     `5*$SZ`($ctx),@X[5]
+       addl    @X[0],$A,$A
+       $LD     `6*$SZ`($ctx),@X[6]
+       addl    @X[1],$B,$B
+       $LD     `7*$SZ`($ctx),@X[7]
+       ldo     `16*$SZ`($inp),$inp     ; advance $inp
+
+       $ST     $A,`0*$SZ`($ctx)        ; save context
+       addl    @X[2],$C,$C
+       $ST     $B,`1*$SZ`($ctx)
+       addl    @X[3],$D,$D
+       $ST     $C,`2*$SZ`($ctx)
+       addl    @X[4],$E,$E
+       $ST     $D,`3*$SZ`($ctx)
+       addl    @X[5],$F,$F
+       $ST     $E,`4*$SZ`($ctx)
+       addl    @X[6],$G,$G
+       $ST     $F,`5*$SZ`($ctx)
+       addl    @X[7],$H,$H
+       $ST     $G,`6*$SZ`($ctx)
+       $ST     $H,`7*$SZ`($ctx)
+
+       cmpb,*<>,n $inp,$num,L\$oop
+       $PUSH   $inp,`-$FRAME_MARKER-3*$SIZE_T`(%sp)    ; save $inp
+___
+if ($SZ==8 && $SIZE_T==4)      # SHA512 for 32-bit PA-RISC 1.0
+{{
+$code.=<<___;
+       b       L\$done
+       nop
+
+       .ALIGN  64
+L\$parisc1
+___
+
+@V=(  $Ahi,  $Alo,  $Bhi,  $Blo,  $Chi,  $Clo,  $Dhi,  $Dlo,
+      $Ehi,  $Elo,  $Fhi,  $Flo,  $Ghi,  $Glo,  $Hhi,  $Hlo) = 
+   ( "%r1", "%r2", "%r3", "%r4", "%r5", "%r6", "%r7", "%r8",
+     "%r9","%r10","%r11","%r12","%r13","%r14","%r15","%r16");
+$a0 ="%r17";
+$a1 ="%r18";
+$a2 ="%r19";
+$a3 ="%r20";
+$t0 ="%r21";
+$t1 ="%r22";
+$t2 ="%r28";
+$t3 ="%r29";
+$Tbl="%r31";
+
+@X=("%r23","%r24","%r25","%r26");      # zaps $num,$inp,$ctx
+
+sub ROUND_00_15_pa1 {
+my ($i,$ahi,$alo,$bhi,$blo,$chi,$clo,$dhi,$dlo,
+       $ehi,$elo,$fhi,$flo,$ghi,$glo,$hhi,$hlo,$flag)=@_;
+my ($Xhi,$Xlo,$Xnhi,$Xnlo) = @X;
+
+$code.=<<___ if (!$flag);
+       ldw     `-$XOFF+8*(($i+1)%16)`(%sp),$Xnhi
+       ldw     `-$XOFF+8*(($i+1)%16)+4`(%sp),$Xnlo     ; load X[i+1]
+___
+$code.=<<___;
+       shd     $ehi,$elo,$Sigma1[0],$t0
+        add    $Xlo,$hlo,$hlo
+       shd     $elo,$ehi,$Sigma1[0],$t1
+        addc   $Xhi,$hhi,$hhi          ; h += X[i]
+       shd     $ehi,$elo,$Sigma1[1],$t2
+        ldwm   8($Tbl),$Xhi
+       shd     $elo,$ehi,$Sigma1[1],$t3
+        ldw    -4($Tbl),$Xlo           ; load K[i]
+       xor     $t2,$t0,$t0
+       xor     $t3,$t1,$t1
+        and    $flo,$elo,$a0
+        and    $fhi,$ehi,$a1
+       shd     $ehi,$elo,$Sigma1[2],$t2
+        andcm  $glo,$elo,$a2
+       shd     $elo,$ehi,$Sigma1[2],$t3
+        andcm  $ghi,$ehi,$a3
+       xor     $t2,$t0,$t0
+       xor     $t3,$t1,$t1             ; Sigma1(e)
+       add     $Xlo,$hlo,$hlo
+        xor    $a2,$a0,$a0
+       addc    $Xhi,$hhi,$hhi          ; h += K[i]
+        xor    $a3,$a1,$a1             ; Ch(e,f,g)
+
+        add    $t0,$hlo,$hlo
+       shd     $ahi,$alo,$Sigma0[0],$t0
+        addc   $t1,$hhi,$hhi           ; h += Sigma1(e)
+       shd     $alo,$ahi,$Sigma0[0],$t1        
+        add    $a0,$hlo,$hlo
+       shd     $ahi,$alo,$Sigma0[1],$t2
+        addc   $a1,$hhi,$hhi           ; h += Ch(e,f,g)
+       shd     $alo,$ahi,$Sigma0[1],$t3
+
+       xor     $t2,$t0,$t0
+       xor     $t3,$t1,$t1
+       shd     $ahi,$alo,$Sigma0[2],$t2
+       and     $alo,$blo,$a0
+       shd     $alo,$ahi,$Sigma0[2],$t3
+       and     $ahi,$bhi,$a1
+       xor     $t2,$t0,$t0
+       xor     $t3,$t1,$t1             ; Sigma0(a)
+
+       and     $alo,$clo,$a2
+       and     $ahi,$chi,$a3
+       xor     $a2,$a0,$a0
+        add    $hlo,$dlo,$dlo
+       xor     $a3,$a1,$a1
+        addc   $hhi,$dhi,$dhi          ; d += h
+       and     $blo,$clo,$a2
+        add    $t0,$hlo,$hlo
+       and     $bhi,$chi,$a3
+        addc   $t1,$hhi,$hhi           ; h += Sigma0(a)
+       xor     $a2,$a0,$a0
+        add    $a0,$hlo,$hlo
+       xor     $a3,$a1,$a1             ; Maj(a,b,c)
+        addc   $a1,$hhi,$hhi           ; h += Maj(a,b,c)
+
+___
+$code.=<<___ if ($i==15 && $flag);
+       extru   $Xlo,31,10,$Xlo
+       comiclr,= $LAST10BITS,$Xlo,%r0
+       b       L\$rounds_pa1
+       nop
+___
+push(@X,shift(@X)); push(@X,shift(@X));
+}
+
+sub ROUND_16_xx_pa1 {
+my ($Xhi,$Xlo,$Xnhi,$Xnlo) = @X;
+my ($i)=shift;
+$i-=16;
+$code.=<<___;
+       ldw     `-$XOFF+8*(($i+1)%16)`(%sp),$Xnhi
+       ldw     `-$XOFF+8*(($i+1)%16)+4`(%sp),$Xnlo     ; load X[i+1]
+       ldw     `-$XOFF+8*(($i+9)%16)`(%sp),$a1
+       ldw     `-$XOFF+8*(($i+9)%16)+4`(%sp),$a0       ; load X[i+9]
+       ldw     `-$XOFF+8*(($i+14)%16)`(%sp),$a3
+       ldw     `-$XOFF+8*(($i+14)%16)+4`(%sp),$a2      ; load X[i+14]
+       shd     $Xnhi,$Xnlo,$sigma0[0],$t0
+       shd     $Xnlo,$Xnhi,$sigma0[0],$t1
+        add    $a0,$Xlo,$Xlo
+       shd     $Xnhi,$Xnlo,$sigma0[1],$t2
+        addc   $a1,$Xhi,$Xhi
+       shd     $Xnlo,$Xnhi,$sigma0[1],$t3
+       xor     $t2,$t0,$t0
+       shd     $Xnhi,$Xnlo,$sigma0[2],$t2
+       xor     $t3,$t1,$t1
+       extru   $Xnhi,`31-$sigma0[2]`,`32-$sigma0[2]`,$t3
+       xor     $t2,$t0,$t0
+        shd    $a3,$a2,$sigma1[0],$a0
+       xor     $t3,$t1,$t1             ; sigma0(X[i+1)&0x0f])
+        shd    $a2,$a3,$sigma1[0],$a1
+       add     $t0,$Xlo,$Xlo
+        shd    $a3,$a2,$sigma1[1],$t2
+       addc    $t1,$Xhi,$Xhi
+        shd    $a2,$a3,$sigma1[1],$t3
+       xor     $t2,$a0,$a0
+       shd     $a3,$a2,$sigma1[2],$t2
+       xor     $t3,$a1,$a1
+       extru   $a3,`31-$sigma1[2]`,`32-$sigma1[2]`,$t3
+       xor     $t2,$a0,$a0
+       xor     $t3,$a1,$a1             ; sigma0(X[i+14)&0x0f])
+       add     $a0,$Xlo,$Xlo
+       addc    $a1,$Xhi,$Xhi
+
+       stw     $Xhi,`-$XOFF+8*($i%16)`(%sp)
+       stw     $Xlo,`-$XOFF+8*($i%16)+4`(%sp)
+___
+&ROUND_00_15_pa1($i,@_,1);
+}
+$code.=<<___;
+       ldw     `0*4`($ctx),$Ahi                ; load context
+       ldw     `1*4`($ctx),$Alo
+       ldw     `2*4`($ctx),$Bhi
+       ldw     `3*4`($ctx),$Blo
+       ldw     `4*4`($ctx),$Chi
+       ldw     `5*4`($ctx),$Clo
+       ldw     `6*4`($ctx),$Dhi
+       ldw     `7*4`($ctx),$Dlo
+       ldw     `8*4`($ctx),$Ehi
+       ldw     `9*4`($ctx),$Elo
+       ldw     `10*4`($ctx),$Fhi
+       ldw     `11*4`($ctx),$Flo
+       ldw     `12*4`($ctx),$Ghi
+       ldw     `13*4`($ctx),$Glo
+       ldw     `14*4`($ctx),$Hhi
+       ldw     `15*4`($ctx),$Hlo
+
+       extru   $inp,31,2,$t0
+       sh3addl $t0,%r0,$t0
+       subi    32,$t0,$t0
+       mtctl   $t0,%cr11               ; load %sar with align factor
+
+L\$oop_pa1
+       extru   $inp,31,2,$a3
+       comib,= 0,$a3,L\$aligned_pa1
+       sub     $inp,$a3,$inp
+
+       ldw     `0*4`($inp),$X[0]
+       ldw     `1*4`($inp),$X[1]
+       ldw     `2*4`($inp),$t2
+       ldw     `3*4`($inp),$t3
+       ldw     `4*4`($inp),$a0
+       ldw     `5*4`($inp),$a1
+       ldw     `6*4`($inp),$a2
+       ldw     `7*4`($inp),$a3
+       vshd    $X[0],$X[1],$X[0]
+       vshd    $X[1],$t2,$X[1]
+       stw     $X[0],`-$XOFF+0*4`(%sp)
+       ldw     `8*4`($inp),$t0
+       vshd    $t2,$t3,$t2
+       stw     $X[1],`-$XOFF+1*4`(%sp)
+       ldw     `9*4`($inp),$t1
+       vshd    $t3,$a0,$t3
+___
+{
+my @t=($t2,$t3,$a0,$a1,$a2,$a3,$t0,$t1);
+for ($i=2;$i<=(128/4-8);$i++) {
+$code.=<<___;
+       stw     $t[0],`-$XOFF+$i*4`(%sp)
+       ldw     `(8+$i)*4`($inp),$t[0]
+       vshd    $t[1],$t[2],$t[1]
+___
+push(@t,shift(@t));
+}
+for (;$i<(128/4-1);$i++) {
+$code.=<<___;
+       stw     $t[0],`-$XOFF+$i*4`(%sp)
+       vshd    $t[1],$t[2],$t[1]
+___
+push(@t,shift(@t));
+}
+$code.=<<___;
+       b       L\$collected_pa1
+       stw     $t[0],`-$XOFF+$i*4`(%sp)
+
+___
+}
+$code.=<<___;
+L\$aligned_pa1
+       ldw     `0*4`($inp),$X[0]
+       ldw     `1*4`($inp),$X[1]
+       ldw     `2*4`($inp),$t2
+       ldw     `3*4`($inp),$t3
+       ldw     `4*4`($inp),$a0
+       ldw     `5*4`($inp),$a1
+       ldw     `6*4`($inp),$a2
+       ldw     `7*4`($inp),$a3
+       stw     $X[0],`-$XOFF+0*4`(%sp)
+       ldw     `8*4`($inp),$t0
+       stw     $X[1],`-$XOFF+1*4`(%sp)
+       ldw     `9*4`($inp),$t1
+___
+{
+my @t=($t2,$t3,$a0,$a1,$a2,$a3,$t0,$t1);
+for ($i=2;$i<(128/4-8);$i++) {
+$code.=<<___;
+       stw     $t[0],`-$XOFF+$i*4`(%sp)
+       ldw     `(8+$i)*4`($inp),$t[0]
+___
+push(@t,shift(@t));
+}
+for (;$i<128/4;$i++) {
+$code.=<<___;
+       stw     $t[0],`-$XOFF+$i*4`(%sp)
+___
+push(@t,shift(@t));
+}
+$code.="L\$collected_pa1\n";
+}
+
+for($i=0;$i<16;$i++)   { &ROUND_00_15_pa1($i,@V); unshift(@V,pop(@V)); unshift(@V,pop(@V)); }
+$code.="L\$rounds_pa1\n";
+for(;$i<32;$i++)       { &ROUND_16_xx_pa1($i,@V); unshift(@V,pop(@V)); unshift(@V,pop(@V)); }
+
+$code.=<<___;
+       $POP    `-$FRAME_MARKER-2*$SIZE_T`(%sp),$ctx    ; restore arguments
+       $POP    `-$FRAME_MARKER-3*$SIZE_T`(%sp),$inp
+       $POP    `-$FRAME_MARKER-4*$SIZE_T`(%sp),$num
+       ldo     `-$rounds*$SZ`($Tbl),$Tbl               ; rewind $Tbl
+
+       ldw     `0*4`($ctx),$t1         ; update context
+       ldw     `1*4`($ctx),$t0
+       ldw     `2*4`($ctx),$t3
+       ldw     `3*4`($ctx),$t2
+       ldw     `4*4`($ctx),$a1
+       ldw     `5*4`($ctx),$a0
+       ldw     `6*4`($ctx),$a3
+       add     $t0,$Alo,$Alo
+       ldw     `7*4`($ctx),$a2
+       addc    $t1,$Ahi,$Ahi
+       ldw     `8*4`($ctx),$t1
+       add     $t2,$Blo,$Blo
+       ldw     `9*4`($ctx),$t0
+       addc    $t3,$Bhi,$Bhi
+       ldw     `10*4`($ctx),$t3
+       add     $a0,$Clo,$Clo
+       ldw     `11*4`($ctx),$t2
+       addc    $a1,$Chi,$Chi
+       ldw     `12*4`($ctx),$a1
+       add     $a2,$Dlo,$Dlo
+       ldw     `13*4`($ctx),$a0
+       addc    $a3,$Dhi,$Dhi
+       ldw     `14*4`($ctx),$a3
+       add     $t0,$Elo,$Elo
+       ldw     `15*4`($ctx),$a2
+       addc    $t1,$Ehi,$Ehi
+       stw     $Ahi,`0*4`($ctx)
+       add     $t2,$Flo,$Flo
+       stw     $Alo,`1*4`($ctx)
+       addc    $t3,$Fhi,$Fhi
+       stw     $Bhi,`2*4`($ctx)
+       add     $a0,$Glo,$Glo
+       stw     $Blo,`3*4`($ctx)
+       addc    $a1,$Ghi,$Ghi
+       stw     $Chi,`4*4`($ctx)
+       add     $a2,$Hlo,$Hlo
+       stw     $Clo,`5*4`($ctx)
+       addc    $a3,$Hhi,$Hhi
+       stw     $Dhi,`6*4`($ctx)
+       ldo     `16*$SZ`($inp),$inp     ; advance $inp
+       stw     $Dlo,`7*4`($ctx)
+       stw     $Ehi,`8*4`($ctx)
+       stw     $Elo,`9*4`($ctx)
+       stw     $Fhi,`10*4`($ctx)
+       stw     $Flo,`11*4`($ctx)
+       stw     $Ghi,`12*4`($ctx)
+       stw     $Glo,`13*4`($ctx)
+       stw     $Hhi,`14*4`($ctx)
+       comb,=  $inp,$num,L\$done
+       stw     $Hlo,`15*4`($ctx)
+       b       L\$oop_pa1
+       $PUSH   $inp,`-$FRAME_MARKER-3*$SIZE_T`(%sp)    ; save $inp
+L\$done
+___
+}}
+$code.=<<___;
+       $POP    `-$FRAME-$SAVED_RP`(%sp),%r2            ; standard epilogue
+       $POP    `-$FRAME+1*$SIZE_T`(%sp),%r4
+       $POP    `-$FRAME+2*$SIZE_T`(%sp),%r5
+       $POP    `-$FRAME+3*$SIZE_T`(%sp),%r6
+       $POP    `-$FRAME+4*$SIZE_T`(%sp),%r7
+       $POP    `-$FRAME+5*$SIZE_T`(%sp),%r8
+       $POP    `-$FRAME+6*$SIZE_T`(%sp),%r9
+       $POP    `-$FRAME+7*$SIZE_T`(%sp),%r10
+       $POP    `-$FRAME+8*$SIZE_T`(%sp),%r11
+       $POP    `-$FRAME+9*$SIZE_T`(%sp),%r12
+       $POP    `-$FRAME+10*$SIZE_T`(%sp),%r13
+       $POP    `-$FRAME+11*$SIZE_T`(%sp),%r14
+       $POP    `-$FRAME+12*$SIZE_T`(%sp),%r15
+       $POP    `-$FRAME+13*$SIZE_T`(%sp),%r16
+       $POP    `-$FRAME+14*$SIZE_T`(%sp),%r17
+       $POP    `-$FRAME+15*$SIZE_T`(%sp),%r18
+       bv      (%r2)
+       .EXIT
+       $POPMB  -$FRAME(%sp),%r3
+       .PROCEND
+       .STRINGZ "SHA`64*$SZ` block transform for PA-RISC, CRYPTOGAMS by <appro\@openssl.org>"
+___
+
+# Explicitly encode PA-RISC 2.0 instructions used in this module, so
+# that it can be compiled with .LEVEL 1.0. It should be noted that I
+# wouldn't have to do this, if GNU assembler understood .ALLOW 2.0
+# directive...
+
+my $ldd = sub {
+  my ($mod,$args) = @_;
+  my $orig = "ldd$mod\t$args";
+
+    if ($args =~ /(\-?[0-9]+)\(%r([0-9]+)\),%r([0-9]+)/) # format 3 suffices
+    {  my $opcode=(0x14<<26)|($2<<21)|($3<<16)|(($1&0x1FF8)<<1)|(($1>>13)&1);
+       $opcode|=(1<<3) if ($mod =~ /^,m/);
+       $opcode|=(1<<2) if ($mod =~ /^,mb/);
+       sprintf "\t.WORD\t0x%08x\t; %s",$opcode,$orig;
+    }
+    else { "\t".$orig; }
+};
+
+my $std = sub {
+  my ($mod,$args) = @_;
+  my $orig = "std$mod\t$args";
+
+    if ($args =~ /%r([0-9]+),(\-?[0-9]+)\(%r([0-9]+)\)/) # format 3 suffices
+    {  my $opcode=(0x1c<<26)|($3<<21)|($1<<16)|(($2&0x1FF8)<<1)|(($2>>13)&1);
+       sprintf "\t.WORD\t0x%08x\t; %s",$opcode,$orig;
+    }
+    else { "\t".$orig; }
+};
+
+my $extrd = sub {
+  my ($mod,$args) = @_;
+  my $orig = "extrd$mod\t$args";
+
+    # I only have ",u" completer, it's implicitly encoded...
+    if ($args =~ /%r([0-9]+),([0-9]+),([0-9]+),%r([0-9]+)/)    # format 15
+    {  my $opcode=(0x36<<26)|($1<<21)|($4<<16);
+       my $len=32-$3;
+       $opcode |= (($2&0x20)<<6)|(($2&0x1f)<<5);               # encode pos
+       $opcode |= (($len&0x20)<<7)|($len&0x1f);                # encode len
+       sprintf "\t.WORD\t0x%08x\t; %s",$opcode,$orig;
+    }
+    elsif ($args =~ /%r([0-9]+),%sar,([0-9]+),%r([0-9]+)/)     # format 12
+    {  my $opcode=(0x34<<26)|($1<<21)|($3<<16)|(2<<11)|(1<<9);
+       my $len=32-$2;
+       $opcode |= (($len&0x20)<<3)|($len&0x1f);                # encode len
+       $opcode |= (1<<13) if ($mod =~ /,\**=/);
+       sprintf "\t.WORD\t0x%08x\t; %s",$opcode,$orig;
+    }
+    else { "\t".$orig; }
+};
+
+my $shrpd = sub {
+  my ($mod,$args) = @_;
+  my $orig = "shrpd$mod\t$args";
+
+    if ($args =~ /%r([0-9]+),%r([0-9]+),([0-9]+),%r([0-9]+)/)  # format 14
+    {  my $opcode=(0x34<<26)|($2<<21)|($1<<16)|(1<<10)|$4;
+       my $cpos=63-$3;
+       $opcode |= (($cpos&0x20)<<6)|(($cpos&0x1f)<<5);         # encode sa
+       sprintf "\t.WORD\t0x%08x\t; %s",$opcode,$orig;
+    }
+    elsif ($args =~ /%r([0-9]+),%r([0-9]+),%sar,%r([0-9]+)/)   # format 11
+    {  sprintf "\t.WORD\t0x%08x\t; %s",
+               (0x34<<26)|($2<<21)|($1<<16)|(1<<9)|$3,$orig;
+    }
+    else { "\t".$orig; }
+};
+
+sub assemble {
+  my ($mnemonic,$mod,$args)=@_;
+  my $opcode = eval("\$$mnemonic");
+
+    ref($opcode) eq 'CODE' ? &$opcode($mod,$args) : "\t$mnemonic$mod\t$args";
+}
+
+foreach (split("\n",$code)) {
+       s/\`([^\`]*)\`/eval $1/ge;
+
+       s/shd\s+(%r[0-9]+),(%r[0-9]+),([0-9]+)/
+               $3>31 ? sprintf("shd\t%$2,%$1,%d",$3-32)        # rotation for >=32
+               :       sprintf("shd\t%$1,%$2,%d",$3)/e                 or
+       # translate made up instructons: _ror, _shr, _align, _shl
+       s/_ror(\s+)(%r[0-9]+),/
+               ($SZ==4 ? "shd" : "shrpd")."$1$2,$2,"/e                 or
+
+       s/_shr(\s+%r[0-9]+),([0-9]+),/
+               $SZ==4 ? sprintf("extru%s,%d,%d,",$1,31-$2,32-$2)
+               :        sprintf("extrd,u%s,%d,%d,",$1,63-$2,64-$2)/e   or
+
+       s/_align(\s+%r[0-9]+,%r[0-9]+),/
+               ($SZ==4 ? "vshd$1," : "shrpd$1,%sar,")/e                or
+
+       s/_shl(\s+%r[0-9]+),([0-9]+),/
+               $SIZE_T==4 ? sprintf("zdep%s,%d,%d,",$1,31-$2,32-$2)
+               :            sprintf("depd,z%s,%d,%d,",$1,63-$2,64-$2)/e;
+
+       s/^\s+([a-z]+)([\S]*)\s+([\S]*)/&assemble($1,$2,$3)/e if ($SIZE_T==4);
+
+       s/cmpb,\*/comb,/ if ($SIZE_T==4);
+
+       print $_,"\n";
+}
+
+close STDOUT;