s390x assembler pack.
authorAndy Polyakov <appro@openssl.org>
Mon, 30 Apr 2007 08:42:54 +0000 (08:42 +0000)
committerAndy Polyakov <appro@openssl.org>
Mon, 30 Apr 2007 08:42:54 +0000 (08:42 +0000)
crypto/aes/asm/aes-s390x.pl [new file with mode: 0644]
crypto/bn/asm/s390x-mont.pl [new file with mode: 0644]
crypto/sha/asm/sha1-s390x.pl [new file with mode: 0644]
crypto/sha/asm/sha512-s390x.pl [new file with mode: 0644]

diff --git a/crypto/aes/asm/aes-s390x.pl b/crypto/aes/asm/aes-s390x.pl
new file mode 100644 (file)
index 0000000..4a9713a
--- /dev/null
@@ -0,0 +1,643 @@
+#!/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 s390x.
+
+# April 2007.
+#
+# Software performance improvement over gcc-generated code is ~70% and
+# in absolute terms is ~73 cycles per byte processed with 128-bit key.
+# You're likely to exclaim "why so slow?" Keep in mind that z-CPUs are
+# *strictly* in-order execution and issued instruction [in this case
+# load value from memory is critical] has to complete before execution
+# flow proceeds. S-boxes are compressed to 2KB.
+#
+# As for hardware acceleration support. It's basically a "teaser," as
+# it can and should be improved in several ways. Most notably support
+# for CBC is not utilized, nor multiple blocks are ever processed.
+# Then software key schedule can be postponed till hardware support
+# detection... Performance improvement over assembler is reportedly
+# ~2.5x, but can reach >15x [naturally on larger chunks] if proper
+# support is implemented.
+
+$t1="%r0";
+$t2="%r1";
+$t3="%r2";     $inp="%r2";
+$out="%r3";    $mask="%r3";
+$key="%r4";
+$i1="%r5";
+$i2="%r6";
+$i3="%r7";
+$s0="%r8";
+$s1="%r9";
+$s2="%r10";
+$s3="%r11";
+$tbl="%r12";
+$rounds="%r13";
+$ra="%r14";
+$sp="%r15";
+
+sub _data_word()
+{ my $i;
+    while(defined($i=shift)) { $code.=sprintf".long\t0x%08x,0x%08x\n",$i,$i; }
+}
+
+$code=<<___;
+.text
+
+.type  AES_Te,\@object
+.align 64
+AES_Te:
+___
+&_data_word(
+       0xc66363a5, 0xf87c7c84, 0xee777799, 0xf67b7b8d,
+       0xfff2f20d, 0xd66b6bbd, 0xde6f6fb1, 0x91c5c554,
+       0x60303050, 0x02010103, 0xce6767a9, 0x562b2b7d,
+       0xe7fefe19, 0xb5d7d762, 0x4dababe6, 0xec76769a,
+       0x8fcaca45, 0x1f82829d, 0x89c9c940, 0xfa7d7d87,
+       0xeffafa15, 0xb25959eb, 0x8e4747c9, 0xfbf0f00b,
+       0x41adadec, 0xb3d4d467, 0x5fa2a2fd, 0x45afafea,
+       0x239c9cbf, 0x53a4a4f7, 0xe4727296, 0x9bc0c05b,
+       0x75b7b7c2, 0xe1fdfd1c, 0x3d9393ae, 0x4c26266a,
+       0x6c36365a, 0x7e3f3f41, 0xf5f7f702, 0x83cccc4f,
+       0x6834345c, 0x51a5a5f4, 0xd1e5e534, 0xf9f1f108,
+       0xe2717193, 0xabd8d873, 0x62313153, 0x2a15153f,
+       0x0804040c, 0x95c7c752, 0x46232365, 0x9dc3c35e,
+       0x30181828, 0x379696a1, 0x0a05050f, 0x2f9a9ab5,
+       0x0e070709, 0x24121236, 0x1b80809b, 0xdfe2e23d,
+       0xcdebeb26, 0x4e272769, 0x7fb2b2cd, 0xea75759f,
+       0x1209091b, 0x1d83839e, 0x582c2c74, 0x341a1a2e,
+       0x361b1b2d, 0xdc6e6eb2, 0xb45a5aee, 0x5ba0a0fb,
+       0xa45252f6, 0x763b3b4d, 0xb7d6d661, 0x7db3b3ce,
+       0x5229297b, 0xdde3e33e, 0x5e2f2f71, 0x13848497,
+       0xa65353f5, 0xb9d1d168, 0x00000000, 0xc1eded2c,
+       0x40202060, 0xe3fcfc1f, 0x79b1b1c8, 0xb65b5bed,
+       0xd46a6abe, 0x8dcbcb46, 0x67bebed9, 0x7239394b,
+       0x944a4ade, 0x984c4cd4, 0xb05858e8, 0x85cfcf4a,
+       0xbbd0d06b, 0xc5efef2a, 0x4faaaae5, 0xedfbfb16,
+       0x864343c5, 0x9a4d4dd7, 0x66333355, 0x11858594,
+       0x8a4545cf, 0xe9f9f910, 0x04020206, 0xfe7f7f81,
+       0xa05050f0, 0x783c3c44, 0x259f9fba, 0x4ba8a8e3,
+       0xa25151f3, 0x5da3a3fe, 0x804040c0, 0x058f8f8a,
+       0x3f9292ad, 0x219d9dbc, 0x70383848, 0xf1f5f504,
+       0x63bcbcdf, 0x77b6b6c1, 0xafdada75, 0x42212163,
+       0x20101030, 0xe5ffff1a, 0xfdf3f30e, 0xbfd2d26d,
+       0x81cdcd4c, 0x180c0c14, 0x26131335, 0xc3ecec2f,
+       0xbe5f5fe1, 0x359797a2, 0x884444cc, 0x2e171739,
+       0x93c4c457, 0x55a7a7f2, 0xfc7e7e82, 0x7a3d3d47,
+       0xc86464ac, 0xba5d5de7, 0x3219192b, 0xe6737395,
+       0xc06060a0, 0x19818198, 0x9e4f4fd1, 0xa3dcdc7f,
+       0x44222266, 0x542a2a7e, 0x3b9090ab, 0x0b888883,
+       0x8c4646ca, 0xc7eeee29, 0x6bb8b8d3, 0x2814143c,
+       0xa7dede79, 0xbc5e5ee2, 0x160b0b1d, 0xaddbdb76,
+       0xdbe0e03b, 0x64323256, 0x743a3a4e, 0x140a0a1e,
+       0x924949db, 0x0c06060a, 0x4824246c, 0xb85c5ce4,
+       0x9fc2c25d, 0xbdd3d36e, 0x43acacef, 0xc46262a6,
+       0x399191a8, 0x319595a4, 0xd3e4e437, 0xf279798b,
+       0xd5e7e732, 0x8bc8c843, 0x6e373759, 0xda6d6db7,
+       0x018d8d8c, 0xb1d5d564, 0x9c4e4ed2, 0x49a9a9e0,
+       0xd86c6cb4, 0xac5656fa, 0xf3f4f407, 0xcfeaea25,
+       0xca6565af, 0xf47a7a8e, 0x47aeaee9, 0x10080818,
+       0x6fbabad5, 0xf0787888, 0x4a25256f, 0x5c2e2e72,
+       0x381c1c24, 0x57a6a6f1, 0x73b4b4c7, 0x97c6c651,
+       0xcbe8e823, 0xa1dddd7c, 0xe874749c, 0x3e1f1f21,
+       0x964b4bdd, 0x61bdbddc, 0x0d8b8b86, 0x0f8a8a85,
+       0xe0707090, 0x7c3e3e42, 0x71b5b5c4, 0xcc6666aa,
+       0x904848d8, 0x06030305, 0xf7f6f601, 0x1c0e0e12,
+       0xc26161a3, 0x6a35355f, 0xae5757f9, 0x69b9b9d0,
+       0x17868691, 0x99c1c158, 0x3a1d1d27, 0x279e9eb9,
+       0xd9e1e138, 0xebf8f813, 0x2b9898b3, 0x22111133,
+       0xd26969bb, 0xa9d9d970, 0x078e8e89, 0x339494a7,
+       0x2d9b9bb6, 0x3c1e1e22, 0x15878792, 0xc9e9e920,
+       0x87cece49, 0xaa5555ff, 0x50282878, 0xa5dfdf7a,
+       0x038c8c8f, 0x59a1a1f8, 0x09898980, 0x1a0d0d17,
+       0x65bfbfda, 0xd7e6e631, 0x844242c6, 0xd06868b8,
+       0x824141c3, 0x299999b0, 0x5a2d2d77, 0x1e0f0f11,
+       0x7bb0b0cb, 0xa85454fc, 0x6dbbbbd6, 0x2c16163a);
+$code.=<<___;
+.size  AES_Te,.-AES_Te
+
+# void AES_encrypt(const unsigned char *in, unsigned char *out,
+#               const AES_KEY *key) {
+.globl AES_encrypt
+.type  AES_encrypt,\@function
+AES_encrypt:
+       lghi    %r0,10
+       c       %r0,240($key)
+       jne     .Lesoft
+       lghi    %r0,0           # query capability vector
+       la      %r1,16($sp)
+       .long   0xb92e0042      # km %r4,%r2
+       lg      %r0,16($sp)
+       tmhl    %r0,`0x8000>>2`
+       jz      .Lesoft
+       lghi    %r0,`0x00|0x12` # encrypt AES-128
+       la      %r1,0($key)
+       la      %r2,0($inp)
+       la      %r4,0($out)
+       lghi    %r3,16          # single block length
+       .long   0xb92e0042      # km %r4,%r2
+       br      %r14
+.Lesoft:
+       stmg    %r3,%r15,24($sp)
+
+       bras    $tbl,.Lepic
+.Lepic:        aghi    $tbl,AES_Te-.Lepic
+
+       llgf    $s0,0($inp)
+       llgf    $s1,4($inp)
+       llgf    $s2,8($inp)
+       llgf    $s3,12($inp)
+
+       llill   $mask,`0xff<<3`
+       bras    $ra,_s390x_AES_encrypt
+
+       lg      $out,24($sp)
+       st      $s0,0($out)
+       st      $s1,4($out)
+       st      $s2,8($out)
+       st      $s3,12($out)
+
+       lmg     %r6,%r15,48($sp)
+       br      %r14
+.size  AES_encrypt,.-AES_encrypt
+
+.type   _s390x_AES_encrypt,\@function
+.align 16
+_s390x_AES_encrypt:
+       x       $s0,0($key)
+       x       $s1,4($key)
+       x       $s2,8($key)
+       x       $s3,12($key)
+       l       $rounds,240($key)
+       aghi    $rounds,-1
+
+.Lenc_loop:
+       sllg    $i1,$s0,`0+3`
+       srlg    $i2,$s0,`8-3`
+       srlg    $i3,$s0,`16-3`
+       srl     $s0,`24-3`
+       nr      $s0,$mask
+       ngr     $i1,$mask
+       nr      $i2,$mask
+       nr      $i3,$mask
+       l       $s0,0($s0,$tbl) # Te0[s0>>24]
+       l       $t1,1($i1,$tbl) # Te3[s0>>0]
+       l       $t2,2($i2,$tbl) # Te2[s0>>8]
+       l       $t3,3($i3,$tbl) # Te1[s0>>16]
+
+       srlg    $i1,$s1,`16-3`  # i0
+       sllg    $i2,$s1,`0+3`
+       srlg    $i3,$s1,`8-3`
+       srl     $s1,`24-3`
+       nr      $i1,$mask
+       nr      $s1,$mask
+       ngr     $i2,$mask
+       nr      $i3,$mask
+       x       $s0,3($i1,$tbl) # Te1[s1>>16]
+       l       $s1,0($s1,$tbl) # Te0[s1>>24]
+       x       $t2,1($i2,$tbl) # Te3[s1>>0]
+       x       $t3,2($i3,$tbl) # Te2[s1>>8]
+       xr      $s1,$t1
+
+       srlg    $i1,$s2,`8-3`   # i0
+       srlg    $i2,$s2,`16-3`  # i1
+       sllg    $i3,$s2,`0+3`
+       srl     $s2,`24-3`
+       nr      $i1,$mask
+       nr      $i2,$mask
+       nr      $s2,$mask
+       ngr     $i3,$mask
+       x       $s0,2($i1,$tbl) # Te2[s2>>8]
+       x       $s1,3($i2,$tbl) # Te1[s2>>16]
+       l       $s2,0($s2,$tbl) # Te0[s2>>24]
+       x       $t3,1($i3,$tbl) # Te3[s2>>0]
+       xr      $s2,$t2
+
+       sllg    $i1,$s3,`0+3`   # i0
+       srlg    $i2,$s3,`8-3`   # i1
+       srlg    $i3,$s3,`16-3`  # i2
+       srl     $s3,`24-3`
+       ngr     $i1,$mask
+       nr      $i2,$mask
+       nr      $i3,$mask
+       nr      $s3,$mask
+       x       $s0,1($i1,$tbl) # Te3[s3>>0]
+       x       $s1,2($i2,$tbl) # Te2[s3>>8]
+       x       $s2,3($i3,$tbl) # Te1[s3>>16]
+       l       $s3,0($s3,$tbl) # Te0[s3>>24]
+       xr      $s3,$t3
+
+       la      $key,16($key)
+       x       $s0,0($key)
+       x       $s1,4($key)
+       x       $s2,8($key)
+       x       $s3,12($key)
+
+       brct    $rounds,.Lenc_loop
+
+       sllg    $i1,$s0,`0+3`
+       srlg    $i2,$s0,`8-3`
+       srlg    $i3,$s0,`16-3`
+       srl     $s0,`24-3`
+       nr      $s0,$mask
+       ngr     $i1,$mask
+       nr      $i2,$mask
+       nr      $i3,$mask
+       llgc    $s0,2($s0,$tbl) # Te4[s0>>24]
+       llgc    $t1,2($i1,$tbl) # Te4[s0>>0]
+       llgc    $t2,2($i2,$tbl) # Te4[s0>>8]
+       llgc    $t3,2($i3,$tbl) # Te4[s0>>16]
+       sll     $s0,24
+       sll     $t2,8
+       sll     $t3,16
+
+       srlg    $i1,$s1,`16-3`  # i0
+       sllg    $i2,$s1,`0+3`
+       srlg    $i3,$s1,`8-3`
+       srl     $s1,`24-3`
+       nr      $i1,$mask
+       nr      $s1,$mask
+       ngr     $i2,$mask
+       nr      $i3,$mask
+       llgc    $i1,2($i1,$tbl) # Te4[s1>>16]
+       llgc    $s1,2($s1,$tbl) # Te4[s1>>24]
+       llgc    $i2,2($i2,$tbl) # Te4[s1>>0]
+       llgc    $i3,2($i3,$tbl) # Te4[s1>>8]
+       sll     $i1,16
+       sll     $s1,24
+       sll     $i3,8
+       or      $s0,$i1
+       or      $s1,$t1
+       or      $t2,$i2
+       or      $t3,$i3
+       
+       srlg    $i1,$s2,`8-3`   # i0
+       srlg    $i2,$s2,`16-3`  # i1
+       sllg    $i3,$s2,`0+3`
+       srl     $s2,`24-3`
+       nr      $i1,$mask
+       nr      $i2,$mask
+       nr      $s2,$mask
+       ngr     $i3,$mask
+       llgc    $i1,2($i1,$tbl) # Te4[s2>>8]
+       llgc    $i2,2($i2,$tbl) # Te4[s2>>16]
+       llgc    $s2,2($s2,$tbl) # Te4[s2>>24]
+       llgc    $i3,2($i3,$tbl) # Te4[s2>>0]
+       sll     $i1,8
+       sll     $i2,16
+       sll     $s2,24
+       or      $s0,$i1
+       or      $s1,$i2
+       or      $s2,$t2
+       or      $t3,$i3
+
+       sllg    $i1,$s3,`0+3`   # i0
+       srlg    $i2,$s3,`8-3`   # i1
+       srlg    $i3,$s3,`16-3`  # i2
+       srl     $s3,`24-3`
+       ngr     $i1,$mask
+       nr      $i2,$mask
+       nr      $i3,$mask
+       nr      $s3,$mask
+       llgc    $i1,2($i1,$tbl) # Te4[s3>>0]
+       llgc    $i2,2($i2,$tbl) # Te4[s3>>8]
+       llgc    $i3,2($i3,$tbl) # Te4[s3>>16]
+       llgc    $s3,2($s3,$tbl) # Te4[s3>>24]
+       sll     $i2,8
+       sll     $i3,16
+       sll     $s3,24
+       or      $s0,$i1
+       or      $s1,$i2
+       or      $s2,$i3
+       or      $s3,$t3
+
+       x       $s0,16($key)
+       x       $s1,20($key)
+       x       $s2,24($key)
+       x       $s3,28($key)
+
+       br      $ra     
+.size  _s390x_AES_encrypt,.-_s390x_AES_encrypt
+___
+
+$code.=<<___;
+.type  AES_Td,\@object
+.align 64
+AES_Td:
+___
+&_data_word(
+       0x51f4a750, 0x7e416553, 0x1a17a4c3, 0x3a275e96,
+       0x3bab6bcb, 0x1f9d45f1, 0xacfa58ab, 0x4be30393,
+       0x2030fa55, 0xad766df6, 0x88cc7691, 0xf5024c25,
+       0x4fe5d7fc, 0xc52acbd7, 0x26354480, 0xb562a38f,
+       0xdeb15a49, 0x25ba1b67, 0x45ea0e98, 0x5dfec0e1,
+       0xc32f7502, 0x814cf012, 0x8d4697a3, 0x6bd3f9c6,
+       0x038f5fe7, 0x15929c95, 0xbf6d7aeb, 0x955259da,
+       0xd4be832d, 0x587421d3, 0x49e06929, 0x8ec9c844,
+       0x75c2896a, 0xf48e7978, 0x99583e6b, 0x27b971dd,
+       0xbee14fb6, 0xf088ad17, 0xc920ac66, 0x7dce3ab4,
+       0x63df4a18, 0xe51a3182, 0x97513360, 0x62537f45,
+       0xb16477e0, 0xbb6bae84, 0xfe81a01c, 0xf9082b94,
+       0x70486858, 0x8f45fd19, 0x94de6c87, 0x527bf8b7,
+       0xab73d323, 0x724b02e2, 0xe31f8f57, 0x6655ab2a,
+       0xb2eb2807, 0x2fb5c203, 0x86c57b9a, 0xd33708a5,
+       0x302887f2, 0x23bfa5b2, 0x02036aba, 0xed16825c,
+       0x8acf1c2b, 0xa779b492, 0xf307f2f0, 0x4e69e2a1,
+       0x65daf4cd, 0x0605bed5, 0xd134621f, 0xc4a6fe8a,
+       0x342e539d, 0xa2f355a0, 0x058ae132, 0xa4f6eb75,
+       0x0b83ec39, 0x4060efaa, 0x5e719f06, 0xbd6e1051,
+       0x3e218af9, 0x96dd063d, 0xdd3e05ae, 0x4de6bd46,
+       0x91548db5, 0x71c45d05, 0x0406d46f, 0x605015ff,
+       0x1998fb24, 0xd6bde997, 0x894043cc, 0x67d99e77,
+       0xb0e842bd, 0x07898b88, 0xe7195b38, 0x79c8eedb,
+       0xa17c0a47, 0x7c420fe9, 0xf8841ec9, 0x00000000,
+       0x09808683, 0x322bed48, 0x1e1170ac, 0x6c5a724e,
+       0xfd0efffb, 0x0f853856, 0x3daed51e, 0x362d3927,
+       0x0a0fd964, 0x685ca621, 0x9b5b54d1, 0x24362e3a,
+       0x0c0a67b1, 0x9357e70f, 0xb4ee96d2, 0x1b9b919e,
+       0x80c0c54f, 0x61dc20a2, 0x5a774b69, 0x1c121a16,
+       0xe293ba0a, 0xc0a02ae5, 0x3c22e043, 0x121b171d,
+       0x0e090d0b, 0xf28bc7ad, 0x2db6a8b9, 0x141ea9c8,
+       0x57f11985, 0xaf75074c, 0xee99ddbb, 0xa37f60fd,
+       0xf701269f, 0x5c72f5bc, 0x44663bc5, 0x5bfb7e34,
+       0x8b432976, 0xcb23c6dc, 0xb6edfc68, 0xb8e4f163,
+       0xd731dcca, 0x42638510, 0x13972240, 0x84c61120,
+       0x854a247d, 0xd2bb3df8, 0xaef93211, 0xc729a16d,
+       0x1d9e2f4b, 0xdcb230f3, 0x0d8652ec, 0x77c1e3d0,
+       0x2bb3166c, 0xa970b999, 0x119448fa, 0x47e96422,
+       0xa8fc8cc4, 0xa0f03f1a, 0x567d2cd8, 0x223390ef,
+       0x87494ec7, 0xd938d1c1, 0x8ccaa2fe, 0x98d40b36,
+       0xa6f581cf, 0xa57ade28, 0xdab78e26, 0x3fadbfa4,
+       0x2c3a9de4, 0x5078920d, 0x6a5fcc9b, 0x547e4662,
+       0xf68d13c2, 0x90d8b8e8, 0x2e39f75e, 0x82c3aff5,
+       0x9f5d80be, 0x69d0937c, 0x6fd52da9, 0xcf2512b3,
+       0xc8ac993b, 0x10187da7, 0xe89c636e, 0xdb3bbb7b,
+       0xcd267809, 0x6e5918f4, 0xec9ab701, 0x834f9aa8,
+       0xe6956e65, 0xaaffe67e, 0x21bccf08, 0xef15e8e6,
+       0xbae79bd9, 0x4a6f36ce, 0xea9f09d4, 0x29b07cd6,
+       0x31a4b2af, 0x2a3f2331, 0xc6a59430, 0x35a266c0,
+       0x744ebc37, 0xfc82caa6, 0xe090d0b0, 0x33a7d815,
+       0xf104984a, 0x41ecdaf7, 0x7fcd500e, 0x1791f62f,
+       0x764dd68d, 0x43efb04d, 0xccaa4d54, 0xe49604df,
+       0x9ed1b5e3, 0x4c6a881b, 0xc12c1fb8, 0x4665517f,
+       0x9d5eea04, 0x018c355d, 0xfa877473, 0xfb0b412e,
+       0xb3671d5a, 0x92dbd252, 0xe9105633, 0x6dd64713,
+       0x9ad7618c, 0x37a10c7a, 0x59f8148e, 0xeb133c89,
+       0xcea927ee, 0xb761c935, 0xe11ce5ed, 0x7a47b13c,
+       0x9cd2df59, 0x55f2733f, 0x1814ce79, 0x73c737bf,
+       0x53f7cdea, 0x5ffdaa5b, 0xdf3d6f14, 0x7844db86,
+       0xcaaff381, 0xb968c43e, 0x3824342c, 0xc2a3405f,
+       0x161dc372, 0xbce2250c, 0x283c498b, 0xff0d9541,
+       0x39a80171, 0x080cb3de, 0xd8b4e49c, 0x6456c190,
+       0x7bcb8461, 0xd532b670, 0x486c5c74, 0xd0b85742);
+$code.=<<___;
+.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
+.size  AES_Td,.-AES_Td
+
+# void AES_decrypt(const unsigned char *in, unsigned char *out,
+#               const AES_KEY *key) {
+.globl AES_decrypt
+.type  AES_decrypt,\@function
+AES_decrypt:
+       lghi    %r0,10
+       c       %r0,240($key)
+       jne     .Ldsoft
+       lghi    %r0,0           # query capability vector
+       la      %r1,16($sp)
+       .long   0xb92e0042      # km %r4,%r2
+       lg      %r0,16($sp)
+       tmhl    %r0,`0x8000>>2`
+       jz      .Ldsoft
+       lghi    %r0,`0x80|0x12` # decrypt AES-128
+       la      %r1,160($key)
+       la      %r2,0($inp)
+       la      %r4,0($out)
+       lghi    %r3,16          # single block length
+       .long   0xb92e0042      # km %r4,%r2
+       br      %r14
+.Ldsoft:
+       stmg    %r3,%r15,24($sp)
+
+       bras    $tbl,.Ldpic
+.Ldpic:        aghi    $tbl,AES_Td-.Ldpic
+
+       llgf    $s0,0($inp)
+       llgf    $s1,4($inp)
+       llgf    $s2,8($inp)
+       llgf    $s3,12($inp)
+
+       llill   $mask,`0xff<<3`
+       bras    $ra,_s390x_AES_decrypt
+
+       lg      $out,24($sp)
+       st      $s0,0($out)
+       st      $s1,4($out)
+       st      $s2,8($out)
+       st      $s3,12($out)
+
+       lmg     %r6,%r15,48($sp)
+       br      %r14
+.size  AES_decrypt,.-AES_decrypt
+
+.type   _s390x_AES_decrypt,\@function
+.align 16
+_s390x_AES_decrypt:
+       x       $s0,0($key)
+       x       $s1,4($key)
+       x       $s2,8($key)
+       x       $s3,12($key)
+       l       $rounds,240($key)
+       aghi    $rounds,-1
+
+.Ldec_loop:
+       srlg    $i1,$s0,`16-3`
+       srlg    $i2,$s0,`8-3`
+       sllg    $i3,$s0,`0+3`
+       srl     $s0,`24-3`
+       nr      $s0,$mask
+       nr      $i1,$mask
+       nr      $i2,$mask
+       ngr     $i3,$mask
+       l       $s0,0($s0,$tbl) # Td0[s0>>24]
+       l       $t1,3($i1,$tbl) # Td1[s0>>16]
+       l       $t2,2($i2,$tbl) # Td2[s0>>8]
+       l       $t3,1($i3,$tbl) # Td3[s0>>0]
+
+       sllg    $i1,$s1,`0+3`   # i0
+       srlg    $i2,$s1,`16-3`
+       srlg    $i3,$s1,`8-3`
+       srl     $s1,`24-3`
+       ngr     $i1,$mask
+       nr      $s1,$mask
+       nr      $i2,$mask
+       nr      $i3,$mask
+       x       $s0,1($i1,$tbl) # Td3[s1>>0]
+       l       $s1,0($s1,$tbl) # Td0[s1>>24]
+       x       $t2,3($i2,$tbl) # Td1[s1>>16]
+       x       $t3,2($i3,$tbl) # Td2[s1>>8]
+       xr      $s1,$t1
+
+       srlg    $i1,$s2,`8-3`   # i0
+       sllg    $i2,$s2,`0+3`   # i1
+       srlg    $i3,$s2,`16-3`
+       srl     $s2,`24-3`
+       nr      $i1,$mask
+       ngr     $i2,$mask
+       nr      $s2,$mask
+       nr      $i3,$mask
+       x       $s0,2($i1,$tbl) # Td2[s2>>8]
+       x       $s1,1($i2,$tbl) # Td3[s2>>0]
+       l       $s2,0($s2,$tbl) # Td0[s2>>24]
+       x       $t3,3($i3,$tbl) # Td1[s2>>16]
+       xr      $s2,$t2
+
+       srlg    $i1,$s3,`16-3`  # i0
+       srlg    $i2,$s3,`8-3`   # i1
+       sllg    $i3,$s3,`0+3`   # i2
+       srl     $s3,`24-3`
+       nr      $i1,$mask
+       nr      $i2,$mask
+       ngr     $i3,$mask
+       nr      $s3,$mask
+       x       $s0,3($i1,$tbl) # Td1[s3>>16]
+       x       $s1,2($i2,$tbl) # Td2[s3>>8]
+       x       $s2,1($i3,$tbl) # Td3[s3>>0]
+       l       $s3,0($s3,$tbl) # Td0[s3>>24]
+       xr      $s3,$t3
+
+       la      $key,16($key)
+       x       $s0,0($key)
+       x       $s1,4($key)
+       x       $s2,8($key)
+       x       $s3,12($key)
+
+       brct    $rounds,.Ldec_loop
+
+       l       $t1,`2048+0`($tbl)      # prefetch Td4
+       l       $t2,`2048+32`($tbl)
+       l       $t3,`2048+64`($tbl)
+       l       $i1,`2048+96`($tbl)
+       l       $i2,`2048+128`($tbl)
+       l       $i3,`2048+160`($tbl)
+       l       $t1,`2048+192`($tbl)
+       l       $t2,`2048+224`($tbl)
+       llill   $mask,0xff
+
+       srlg    $i3,$s0,24      # i0
+       srlg    $i1,$s0,16
+       srlg    $i2,$s0,8
+       nr      $s0,$mask       # i3
+       nr      $i1,$mask
+       nr      $i2,$mask
+       llgc    $i3,2048($i3,$tbl)      # Td4[s0>>24]
+       llgc    $t1,2048($i1,$tbl)      # Td4[s0>>16]
+       llgc    $t2,2048($i2,$tbl)      # Td4[s0>>8]
+       llgc    $t3,2048($s0,$tbl)      # Td4[s0>>0]
+       sllg    $s0,$i3,24
+       sll     $t1,16
+       sll     $t2,8
+
+       srlg    $i1,$s1,24
+       srlg    $i2,$s1,16
+       srlg    $i3,$s1,8
+       nr      $s1,$mask       # i0
+       nr      $i2,$mask
+       nr      $i3,$mask
+       llgc    $s1,2048($s1,$tbl)      # Td4[s1>>0]
+       llgc    $i1,2048($i1,$tbl)      # Td4[s1>>24]
+       llgc    $i2,2048($i2,$tbl)      # Td4[s1>>16]
+       llgc    $i3,2048($i3,$tbl)      # Td4[s1>>8]
+       sll     $i1,24
+       sll     $i2,16
+       sll     $i3,8
+       or      $s0,$s1
+       or      $t1,$i1
+       or      $t2,$i2
+       or      $t3,$i3
+
+       srlg    $i1,$s2,8       # i0
+       srlg    $i2,$s2,24
+       srlg    $i3,$s2,16
+       nr      $s2,$mask       # i1
+       nr      $i1,$mask
+       nr      $i3,$mask
+       llgc    $i1,2048($i1,$tbl)      # Td4[s2>>8]
+       llgc    $s1,2048($s2,$tbl)      # Td4[s2>>0]
+       llgc    $i2,2048($i2,$tbl)      # Td4[s2>>24]
+       llgc    $i3,2048($i3,$tbl)      # Td4[s2>>16]
+       sll     $i1,8
+       sll     $i2,24
+       sll     $i3,16
+       or      $s0,$i1
+       or      $s1,$t1
+       or      $t2,$i2
+       or      $t3,$i3
+
+       srlg    $i1,$s3,16      # i0
+       srlg    $i2,$s3,8       # i1
+       srlg    $i3,$s3,24
+       nr      $s3,$mask       # i2
+       nr      $i1,$mask
+       nr      $i2,$mask
+       llgc    $i1,2048($i1,$tbl)      # Td4[s3>>16]
+       llgc    $i2,2048($i2,$tbl)      # Td4[s3>>8]
+       llgc    $s2,2048($s3,$tbl)      # Td4[s3>>0]
+       llgc    $s3,2048($i3,$tbl)      # Td4[s3>>24]
+       sll     $i1,16
+       sll     $i2,8
+       sll     $s3,24
+       or      $s0,$i1
+       or      $s1,$i2
+       or      $s2,$t2
+       or      $s3,$t3
+
+       x       $s0,16($key)
+       x       $s1,20($key)
+       x       $s2,24($key)
+       x       $s3,28($key)
+
+       br      $ra     
+.size  _s390x_AES_decrypt,.-_s390x_AES_decrypt
+.string        "AES for s390x, CRYPTOGAMS by <appro\@openssl.org>"
+___
+
+$code =~ s/\`([^\`]*)\`/eval $1/gem;
+print $code;
diff --git a/crypto/bn/asm/s390x-mont.pl b/crypto/bn/asm/s390x-mont.pl
new file mode 100644 (file)
index 0000000..5d1b9fd
--- /dev/null
@@ -0,0 +1,223 @@
+#!/usr/bin/env perl
+
+# ====================================================================
+# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+
+# April 2007.
+#
+# Performance improvement over vanilla C code varies from 85% to 45%
+# depending on key length and benchmark. Unfortunately in this context
+# these are not very impressive results [for code that utilizes "wide"
+# 64x64=128-bit multiplication, which is not commonly available to C
+# programmers], at least hand-coded bn_asm.c replacement is known to
+# provide 30-40% better results for longest keys. Well, on a second
+# thought it's not very surprising, because z-CPUs are single-issue
+# and _strictly_ in-order execution, while bn_mul_mont is more or less
+# dependent on CPU ability to pipe-line instructions and have several
+# of them "in-flight" at the same time. I mean while other methods,
+# for example Karatsuba, aim to minimize amount of multiplications at
+# the cost of other operations increase, bn_mul_mont aim to neatly
+# "overlap" multiplications and the other operations [and on most
+# platforms even minimize the amount of the other operations, in
+# particular references to memory]. But it's possible to improve this
+# module performance by implementing dedicated squaring code-path and
+# possibly by unrolling loops...
+
+$mn0="%r0";
+$num="%r1";
+
+# int bn_mul_mont(
+$rp="%r2";             # BN_ULONG *rp,
+$ap="%r3";             # const BN_ULONG *ap,
+$bp="%r4";             # const BN_ULONG *bp,
+$np="%r5";             # const BN_ULONG *np,
+$n0="%r6";             # const BN_ULONG *n0,
+#$num="160(%r15)"      # int num);
+
+$bi="%r2";     # zaps rp
+$j="%r7";
+
+$ahi="%r8";
+$alo="%r9";
+$nhi="%r10";
+$nlo="%r11";
+$AHI="%r12";
+$NHI="%r13";
+$fp="%r14";
+$sp="%r15";
+
+$code.=<<___;
+.text
+.globl bn_mul_mont
+.type  bn_mul_mont,\@function
+bn_mul_mont:
+       lgf     $num,164($sp)   # pull $num
+       sla     $num,3          # $num to enumerate bytes
+       la      $rp,0($num,$rp) # pointers to point at the vectors' ends
+       la      $ap,0($num,$ap)
+       la      $bp,0($num,$bp)
+       la      $np,0($num,$np)
+
+       stmg    %r2,%r15,16($sp)
+
+       cghi    $num,16         #
+       lghi    %r2,0           #
+       blr     %r14            # if($num<16) return 0;
+
+       lcgr    $num,$num       # -$num
+       lgr     %r0,$sp
+       lgr     $fp,$sp
+       aghi    $fp,-160-8      # leave room for carry bit
+       la      $sp,0($num,$fp) # alloca
+       stg     %r0,0($sp)
+       aghi    $fp,160-8       # $fp to point at tp[$num-1]
+
+       la      $bp,0($num,$bp) # restore $bp
+       lg      $n0,0($n0)      # pull n0
+
+       lg      $bi,0($bp)
+       lg      $alo,0($num,$ap)
+       mlgr    $ahi,$bi        # ap[0]*bp[0]
+       lgr     $AHI,$ahi
+
+       lgr     $mn0,$alo       # "tp[0]"*n0
+       msgr    $mn0,$n0
+
+       lg      $nlo,0($num,$np)#
+       mlgr    $nhi,$mn0       # np[0]*m1
+       algr    $nlo,$alo       # +="tp[0]"
+       lghi    $NHI,0
+       alcgr   $NHI,$nhi
+
+       lgr     $j,$num
+       aghi    $j,8            # j=1
+.L1st:
+       lg      $alo,0($j,$ap)
+       mlgr    $ahi,$bi        # ap[j]*bp[0]
+       algr    $alo,$AHI
+       lghi    $AHI,0
+       alcgr   $AHI,$ahi
+
+       lg      $nlo,0($j,$np)
+       mlgr    $nhi,$mn0       # np[j]*m1
+       algr    $nlo,$NHI
+       lghi    $NHI,0
+       alcgr   $nhi,$NHI       # +="tp[j]"
+       algr    $nlo,$alo
+       alcgr   $NHI,$nhi
+
+       stg     $nlo,0($j,$fp)  # tp[j-1]=
+       aghi    $j,8            # j++
+       jnz     .L1st
+
+       algr    $NHI,$AHI
+       lghi    $AHI,0
+       alcgr   $AHI,$AHI       # upmost overflow bit
+       stg     $NHI,0($fp)
+       stg     $AHI,8($fp)
+       la      $bp,8($bp)      # bp++
+
+.Louter:
+       lg      $bi,0($bp)      # bp[i]
+       lg      $alo,0($num,$ap)
+       mlgr    $ahi,$bi        # ap[0]*bp[i]
+       alg     $alo,8($num,$fp)# +=tp[0]
+       lghi    $AHI,0
+       alcgr   $AHI,$ahi
+
+       lgr     $mn0,$alo
+       msgr    $mn0,$n0                # tp[0]*n0
+
+       lg      $nlo,0($num,$np)# np[0]
+       mlgr    $nhi,$mn0       # np[0]*m1
+       algr    $nlo,$alo       # +="tp[0]"
+       lghi    $NHI,0
+       alcgr   $NHI,$nhi
+
+       lgr     $j,$num
+       aghi    $j,8            # j=1
+.Linner:
+       lg      $alo,0($j,$ap)
+       mlgr    $ahi,$bi        # ap[j]*bp[i]
+       algr    $alo,$AHI
+       lghi    $AHI,0
+       alcgr   $ahi,$AHI
+       alg     $alo,8($j,$fp)  # +=tp[j]
+       alcgr   $AHI,$ahi
+
+       lg      $nlo,0($j,$np)
+       mlgr    $nhi,$mn0       # np[j]*m1
+       algr    $nlo,$NHI
+       lghi    $NHI,0
+       alcgr   $nhi,$NHI
+       algr    $nlo,$alo       # +="tp[j]"
+       alcgr   $NHI,$nhi
+
+       stg     $nlo,0($j,$fp)  # tp[j-1]=
+       aghi    $j,8            # j++
+       jnz     .Linner
+
+       algr    $NHI,$AHI
+       lghi    $AHI,0
+       alcgr   $AHI,$AHI
+       alg     $NHI,8($fp)     # accumulate previous upmost overflow bit
+       lghi    $ahi,0
+       alcgr   $AHI,$ahi       # new upmost overflow bit
+       stg     $NHI,0($fp)
+       stg     $AHI,8($fp)
+
+       la      $bp,8($bp)      # bp++
+       clg     $bp,16+32($fp)  # compare to &bp[num]
+       jne     .Louter
+___
+
+undef $bi;
+$count=$ap; undef $ap;
+
+$code.=<<___;
+       lg      $rp,16+16($fp)  # reincarnate rp
+       lgr     $j,$num
+       ltgr    $AHI,$AHI
+       jnz     .Lsub           # upmost overflow bit is not zero
+       #slg    $NHI,-8($np)    # tp[num-1]-np[num-1]
+       lghi    $count,-8               # buggy assembler
+       slg     $NHI,0($count,$np)      # buggy assembler
+       jnle    .Lsub           # branch if not borrow 
+
+.Lcopy:        lg      $alo,8($j,$fp)
+       stg     $j,8($j,$fp)
+       stg     $alo,0($j,$rp)
+       aghi    $j,8
+       jnz     .Lcopy
+.Lexit:
+       lmg     %r6,%r15,16+48($fp)
+       lghi    %r2,1           # signal "processed"
+       br      %r14
+
+.Lsub: lcgr    $count,$num
+       sra     $count,3        # incidentally clears "borrow"
+.Lsubloop:
+       lg      $alo,8($j,$fp)
+       slbg    $alo,0($j,$np)
+       stg     $alo,0($j,$rp)
+       la      $j,8($j)
+       brct    $count,.Lsubloop
+       lghi    $ahi,0
+       slbgr   $AHI,$ahi
+       lgr     $j,$num
+       jle     .Lcopy          # branch if borrow
+
+.Lzap: stg     $j,8($j,$fp)
+       aghi    $j,8
+       jnz     .Lzap
+       j       .Lexit
+.size  bn_mul_mont,.-bn_mul_mont
+.string        "Montgomery Multiplication for s390x, CRYPTOGAMS by <appro\@openssl.org>"
+___
+
+print $code;
+close STDOUT;
diff --git a/crypto/sha/asm/sha1-s390x.pl b/crypto/sha/asm/sha1-s390x.pl
new file mode 100644 (file)
index 0000000..9f4436d
--- /dev/null
@@ -0,0 +1,221 @@
+#!/usr/bin/env perl
+
+# ====================================================================
+# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+
+# SHA1 block procedure for s390x.
+
+# April 2007.
+#
+# Performance is >30% better than gcc 3.3 generated code. But the real
+# twist is that SHA1 hardware support is detected and utilized. In
+# which case performance can reach further >8x for larger chunks.
+
+$kimdfunc=1;   # magic function code for kimd instruction
+
+$output=shift;
+open STDOUT,">$output";
+
+$t0="%r0";
+$t1="%r1";
+$ctx="%r2";
+$inp="%r3";
+$len="%r4";
+
+$A="%r5";
+$B="%r6";
+$C="%r7";
+$D="%r8";
+$E="%r9";      @V=($A,$B,$C,$D,$E);
+$K_00_19="%r10";
+$K_20_39="%r11";
+$K_40_59="%r12";
+$K_60_79="%r13";
+$Xi="%r14";
+$sp="%r15";
+
+$frame=160+16*4;
+
+sub BODY_00_15 {
+my ($i,$a,$b,$c,$d,$e)=@_;
+my $xi=($i&1)?$Xi:$t1;
+
+$code.=<<___ if ($i<16 && !($i&1));
+       lg      $Xi,`$i*4`($inp)
+___
+$code.=<<___;
+       alr     $e,$K_00_19     ### $i
+       rll     $t0,$a,5
+       alr     $e,$t0
+       lr      $t0,$d
+       xr      $t0,$c
+       nr      $t0,$b
+       xr      $t0,$d
+       alr     $e,$t0
+       rll     $b,$b,30
+___
+$code.=<<___ if ($i<16 && !($i&1));
+       srlg    $xi,$Xi,32
+       stg     $Xi,`160+$i*4`($sp)
+___
+$code.=<<___;
+       alr     $e,$xi
+___
+}
+
+sub Xupdate {
+my $i=shift;
+
+return if ($i&1);      # Xupdate is vectorized and executed every 2nd cycle
+$code.=<<___;
+       lg      $Xi,`160+4*($i%16)`($sp)        ### Xupdate($i)
+       xg      $Xi,`160+4*(($i+2)%16)`($sp)
+       xg      $Xi,`160+4*(($i+8)%16)`($sp)
+___
+if ((($i+13)%16)==15) {
+$code.=<<___;
+       llgf    $t0,`160+4*15`($sp)
+       x       $Xi,`160+0`($sp)
+       sllg    $t0,$t0,32
+       xgr     $Xi,$t0
+___
+} else {
+$code.=<<___;
+       xg      $Xi,`160+4*(($i+13)%16)`($sp)
+___
+}
+$code.=<<___;
+       rll     $Xi,$Xi,1
+       rllg    $t1,$Xi,32
+       rll     $t1,$t1,1
+       rllg    $Xi,$t1,32
+       stg     $Xi,`160+4*($i%16)`($sp)
+___
+}
+
+sub BODY_16_19 {
+       &Xupdate(@_[0]);
+       &BODY_00_15(@_);
+}
+
+sub BODY_20_39 {
+my ($i,$a,$b,$c,$d,$e)=@_;
+my $xi=($i&1)?$Xi:$t1;
+my $K_XX_XX=($i<40)?$K_20_39:$K_60_79;
+
+       &Xupdate($i);
+$code.=<<___;
+       alr     $e,$K_XX_XX     ### $i
+       rll     $t0,$a,5
+       alr     $e,$t0
+       lr      $t0,$b
+       xr      $t0,$c
+       xr      $t0,$d
+       alr     $e,$t0
+       rll     $b,$b,30
+       alr     $e,$xi
+___
+}
+
+sub BODY_40_59 {
+my ($i,$a,$b,$c,$d,$e)=@_;
+my $xi=($i&1)?$Xi:$t1;
+
+       &Xupdate($i);
+$code.=<<___;
+       alr     $e,$K_40_59     ### $i
+       rll     $t0,$a,5
+       alr     $e,$t0
+       lr      $t0,$b
+       or      $t0,$c
+       nr      $t0,$d
+       alr     $e,$xi
+       lr      $t1,$b
+       nr      $t1,$c
+       or      $t0,$t1
+       alr     $e,$t0
+       rll     $b,$b,30
+___
+}
+
+$code.=<<___;
+.text
+.globl sha1_block_data_order
+.type  sha1_block_data_order,\@function
+sha1_block_data_order:
+___
+$code.=<<___ if ($kimdfunc);
+       lghi    %r0,0
+       la      %r1,16($sp)
+       .long   0xb93e0002      # kimd %r0,%r2
+       lg      %r0,16($sp)
+       tmhh    %r0,`0x8000>>$kimdfunc`
+       jz      .Lsoftware
+       lghi    %r0,$kimdfunc
+       lgr     %r1,$ctx
+       lgr     %r2,$inp
+       sllg    %r3,$len,6
+       .long   0xb93e0002      # kimd %r0,%r2
+       br      %r14
+.Lsoftware:
+___
+$code.=<<___;
+       stmg    %r6,%r15,48($sp)
+       lgr     %r0,$sp
+       aghi    $sp,-$frame
+       stg     %r0,0($sp)
+
+       sllg    $len,$len,6
+       la      $len,0($inp,$len)
+
+       llgf    $A,0($ctx)
+       llgf    $B,4($ctx)
+       llgf    $C,8($ctx)
+       llgf    $D,12($ctx)
+       llgf    $E,16($ctx)
+
+       llilh   $K_00_19,0x5a82
+       oill    $K_00_19,0x7999
+       llilh   $K_20_39,0x6ed9
+       oill    $K_20_39,0xeba1
+       llilh   $K_40_59,0x8f1b
+       oill    $K_40_59,0xbcdc
+       llilh   $K_60_79,0xca62
+       oill    $K_60_79,0xc1d6
+.Lloop:
+___
+for ($i=0;$i<16;$i++)  { &BODY_00_15($i,@V); unshift(@V,pop(@V)); }
+for (;$i<20;$i++)      { &BODY_16_19($i,@V); unshift(@V,pop(@V)); }
+for (;$i<40;$i++)      { &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
+for (;$i<60;$i++)      { &BODY_40_59($i,@V); unshift(@V,pop(@V)); }
+for (;$i<80;$i++)      { &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
+$code.=<<___;
+
+       al      $A,0($ctx)
+       al      $B,4($ctx)
+       al      $C,8($ctx)
+       al      $D,12($ctx)
+       al      $E,16($ctx)
+       st      $A,0($ctx)
+       st      $B,4($ctx)
+       st      $C,8($ctx)
+       st      $D,12($ctx)
+       st      $E,16($ctx)
+       la      $inp,64($inp)
+       clgr    $inp,$len
+       jne     .Lloop
+
+       lmg     %r6,%r15,`$frame+48`($sp)
+       br      %r14
+.size  sha1_block_data_order,.-sha1_block_data_order
+.string        "SHA1 block transform for s390x, CRYPTOGAMS by <appro\@openssl.org>"
+___
+
+$code =~ s/\`([^\`]*)\`/eval $1/gem;
+
+print $code;
+close STDOUT;
diff --git a/crypto/sha/asm/sha512-s390x.pl b/crypto/sha/asm/sha512-s390x.pl
new file mode 100644 (file)
index 0000000..67a17d3
--- /dev/null
@@ -0,0 +1,288 @@
+#!/usr/bin/env perl
+
+# ====================================================================
+# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+
+# SHA256/512 block procedures for s390x.
+
+# April 2007.
+#
+# sha256_block_data_order is reportedly >3 times faster than gcc 3.3
+# generated code (must to be a bug in compiler, as improvement is
+# "pathologically" high, in particular in comparison to other SHA
+# modules). But the real twist is that it detects if hardware support
+# for SHA256 is available and in such case utilizes it. Then the
+# performance can reach >12x of assembler one for larger chunks.
+#
+# sha512_block_data_order is ~70% faster than gcc 3.3 generated code.
+
+$t0="%r0";
+$t1="%r1";
+$ctx="%r2";
+$inp="%r3";
+$len="%r4";    # used as index in inner loop
+
+$A="%r5";
+$B="%r6";
+$C="%r7";
+$D="%r8";
+$E="%r9";
+$F="%r10";
+$G="%r11";
+$H="%r12";     @V=($A,$B,$C,$D,$E,$F,$G,$H);
+$tbl="%r13";
+$T1="%r14";
+$sp="%r15";
+
+$output=shift;
+open STDOUT,">$output";
+
+if ($output =~ /512/) {
+       $label="512";
+       $SZ=8;
+       $LD="lg";       # load from memory
+       $ST="stg";      # store to memory
+       $ADD="alg";     # add with memory operand
+       $ROT="rllg";    # rotate left
+       $SHR="srlg";    # logical right shift [see even at the end]
+       @Sigma0=(25,30,36);
+       @Sigma1=(23,46,50);
+       @sigma0=(56,63, 7);
+       @sigma1=( 3,45, 6);
+       $rounds=80;
+       $kimdfunc=0;    # 0 means unknown/unsupported/unimplemented
+} else {
+       $label="256";
+       $SZ=4;
+       $LD="llgf";     # load from memory
+       $ST="st";       # store to memory
+       $ADD="al";      # add with memory operand
+       $ROT="rll";     # rotate left
+       $SHR="srl";     # logical right shift
+       @Sigma0=(10,19,30);
+       @Sigma1=( 7,21,26);
+       @sigma0=(14,25, 3);
+       @sigma1=(13,15,10);
+       $rounds=64;
+       $kimdfunc=2;    # magic function code for kimd instruction
+}
+$Func="sha${label}_block_data_order";
+$Table="K${label}";
+$frame=160+16*$SZ;
+
+sub BODY_00_15 {
+my ($i,$a,$b,$c,$d,$e,$f,$g,$h) = @_;
+
+$code.=<<___ if ($i<16);
+       $LD     $T1,`$i*$SZ`($inp)      ### $i
+___
+$code.=<<___;
+       $ROT    $t0,$e,$Sigma1[0]
+       $ROT    $t1,$e,$Sigma1[1]
+       xgr     $t0,$t1
+       $ROT    $t1,$t1,`$Sigma1[2]-$Sigma1[1]`
+       xgr     $t0,$t1                 # Sigma1(e)
+       $ST     $T1,`160+$SZ*($i%16)`($sp)
+       algr    $T1,$t0                 # T1+=Sigma1(e)
+       algr    $T1,$h                  # T1+=h
+       $ADD    $T1,`$i*$SZ`($len,$tbl) # T1+=K[i]
+       lgr     $t0,$f
+       xgr     $t0,$g
+       ngr     $t0,$e
+       xgr     $t0,$g                  # Ch(e,f,g)
+       algr    $T1,$t0                 # T1+=Ch(e,f,g)
+       $ROT    $h,$a,$Sigma0[0]
+       $ROT    $t0,$a,$Sigma0[1]
+       xgr     $h,$t0
+       $ROT    $t0,$t0,`$Sigma0[2]-$Sigma0[1]`
+       xgr     $h,$t0                  # h=Sigma0(a)
+       lgr     $t0,$a
+       ogr     $t0,$b
+       ngr     $t0,$c
+       lgr     $t1,$a
+       ngr     $t1,$b
+       ogr     $t0,$t1                 # Maj(a,b,c)
+       algr    $h,$t0                  # h+=Maj(a,b,c)
+       algr    $d,$T1                  # d+=T1
+       algr    $h,$T1                  # h+=T1
+___
+}
+
+sub BODY_16_XX {
+my ($i,$a,$b,$c,$d,$e,$f,$g,$h) = @_;
+
+$code.=<<___;
+       $LD     $T1,`160+$SZ*(($i+1)%16)`($sp)  ### $i
+       $LD     $t1,`160+$SZ*(($i+14)%16)`($sp)
+       $ROT    $t0,$T1,$sigma0[0]
+       $SHR    $T1,$sigma0[2]
+       xgr     $T1,$t0
+       $ROT    $t0,$t0,`$sigma0[1]-$sigma0[0]`
+       xgr     $T1,$t0                         # sigma0(X[i+1])
+       $ROT    $t0,$t1,$sigma1[0]
+       $ADD    $T1,`160+$SZ*($i%16)`($sp)      # +=X[i]
+       $SHR    $t1,$sigma1[2]
+       xgr     $t1,$t0
+       $ADD    $T1,`160+$SZ*(($i+9)%16)`($sp)  # +=X[i+9]
+       $ROT    $t0,$t0,`$sigma1[1]-$sigma1[0]`
+       xgr     $t1,$t0                         # sigma1(X[i+14])
+       algr    $T1,$t1                         # +=sigma1(X[i+14])
+___
+       &BODY_00_15(@_);
+}
+
+$code.=<<___;
+.text
+.align 64
+.type  $Table,\@object
+$Table:
+___
+$code.=<<___ if ($SZ==4);
+       .long   0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5
+       .long   0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5
+       .long   0xd807aa98,0x12835b01,0x243185be,0x550c7dc3
+       .long   0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174
+       .long   0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc
+       .long   0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da
+       .long   0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7
+       .long   0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967
+       .long   0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13
+       .long   0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85
+       .long   0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3
+       .long   0xd192e819,0xd6990624,0xf40e3585,0x106aa070
+       .long   0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5
+       .long   0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3
+       .long   0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208
+       .long   0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2
+___
+$code.=<<___ if ($SZ==8);
+       .quad   0x428a2f98d728ae22,0x7137449123ef65cd
+       .quad   0xb5c0fbcfec4d3b2f,0xe9b5dba58189dbbc
+       .quad   0x3956c25bf348b538,0x59f111f1b605d019
+       .quad   0x923f82a4af194f9b,0xab1c5ed5da6d8118
+       .quad   0xd807aa98a3030242,0x12835b0145706fbe
+       .quad   0x243185be4ee4b28c,0x550c7dc3d5ffb4e2
+       .quad   0x72be5d74f27b896f,0x80deb1fe3b1696b1
+       .quad   0x9bdc06a725c71235,0xc19bf174cf692694
+       .quad   0xe49b69c19ef14ad2,0xefbe4786384f25e3
+       .quad   0x0fc19dc68b8cd5b5,0x240ca1cc77ac9c65
+       .quad   0x2de92c6f592b0275,0x4a7484aa6ea6e483
+       .quad   0x5cb0a9dcbd41fbd4,0x76f988da831153b5
+       .quad   0x983e5152ee66dfab,0xa831c66d2db43210
+       .quad   0xb00327c898fb213f,0xbf597fc7beef0ee4
+       .quad   0xc6e00bf33da88fc2,0xd5a79147930aa725
+       .quad   0x06ca6351e003826f,0x142929670a0e6e70
+       .quad   0x27b70a8546d22ffc,0x2e1b21385c26c926
+       .quad   0x4d2c6dfc5ac42aed,0x53380d139d95b3df
+       .quad   0x650a73548baf63de,0x766a0abb3c77b2a8
+       .quad   0x81c2c92e47edaee6,0x92722c851482353b
+       .quad   0xa2bfe8a14cf10364,0xa81a664bbc423001
+       .quad   0xc24b8b70d0f89791,0xc76c51a30654be30
+       .quad   0xd192e819d6ef5218,0xd69906245565a910
+       .quad   0xf40e35855771202a,0x106aa07032bbd1b8
+       .quad   0x19a4c116b8d2d0c8,0x1e376c085141ab53
+       .quad   0x2748774cdf8eeb99,0x34b0bcb5e19b48a8
+       .quad   0x391c0cb3c5c95a63,0x4ed8aa4ae3418acb
+       .quad   0x5b9cca4f7763e373,0x682e6ff3d6b2b8a3
+       .quad   0x748f82ee5defb2fc,0x78a5636f43172f60
+       .quad   0x84c87814a1f0ab72,0x8cc702081a6439ec
+       .quad   0x90befffa23631e28,0xa4506cebde82bde9
+       .quad   0xbef9a3f7b2c67915,0xc67178f2e372532b
+       .quad   0xca273eceea26619c,0xd186b8c721c0c207
+       .quad   0xeada7dd6cde0eb1e,0xf57d4f7fee6ed178
+       .quad   0x06f067aa72176fba,0x0a637dc5a2c898a6
+       .quad   0x113f9804bef90dae,0x1b710b35131c471b
+       .quad   0x28db77f523047d84,0x32caab7b40c72493
+       .quad   0x3c9ebe0a15c9bebc,0x431d67c49c100d4c
+       .quad   0x4cc5d4becb3e42b6,0x597f299cfc657e2a
+       .quad   0x5fcb6fab3ad6faec,0x6c44198c4a475817
+___
+$code.=<<___;
+.size  $Table,.-$Table
+.globl $Func
+.type  $Func,\@function
+$Func:
+___
+$code.=<<___ if ($kimdfunc);
+       lghi    %r0,0
+       la      %r1,16($sp)
+       .long   0xb93e0002      # kimd %r0,%r2
+       lg      %r0,16($sp)
+       tmhh    %r0,`0x8000>>$kimdfunc`
+       jz      .Lsoftware
+       lghi    %r0,$kimdfunc
+       lgr     %r1,$ctx
+       lgr     %r2,$inp
+       sllg    %r3,$len,`log(16*$SZ)/log(2)`
+       .long   0xb93e0002      # kimd %r0,%r2
+       br      %r14
+.Lsoftware:
+___
+$code.=<<___;
+       sllg    $len,$len,`log(16*$SZ)/log(2)`
+       la      $len,0($inp,$len)
+       stmg    $len,%r15,32($sp)
+       lgr     %r0,$sp
+       aghi    $sp,-$frame
+       stg     %r0,0($sp)
+
+       bras    $tbl,.Lpic
+.Lpic: aghi    $tbl,$Table-.Lpic
+
+       $LD     $A,`0*$SZ`($ctx)
+       $LD     $B,`1*$SZ`($ctx)
+       $LD     $C,`2*$SZ`($ctx)
+       $LD     $D,`3*$SZ`($ctx)
+       $LD     $E,`4*$SZ`($ctx)
+       $LD     $F,`5*$SZ`($ctx)
+       $LD     $G,`6*$SZ`($ctx)
+       $LD     $H,`7*$SZ`($ctx)
+
+.Lloop:
+       lghi    $len,0
+___
+for ($i=0;$i<16;$i++)  { &BODY_00_15($i,@V); unshift(@V,pop(@V)); }
+$code.=".Lrounds_16_xx:\n";
+for (;$i<32;$i++)      { &BODY_16_XX($i,@V); unshift(@V,pop(@V)); }
+$code.=<<___;
+       aghi    $len,`16*$SZ`
+       lghi    $t0,`($rounds-16)*$SZ`
+       clgr    $len,$t0
+       jne     .Lrounds_16_xx
+
+       $ADD    $A,`0*$SZ`($ctx)
+       $ADD    $B,`1*$SZ`($ctx)
+       $ADD    $C,`2*$SZ`($ctx)
+       $ADD    $D,`3*$SZ`($ctx)
+       $ADD    $E,`4*$SZ`($ctx)
+       $ADD    $F,`5*$SZ`($ctx)
+       $ADD    $G,`6*$SZ`($ctx)
+       $ADD    $H,`7*$SZ`($ctx)
+       $ST     $A,`0*$SZ`($ctx)
+       $ST     $B,`1*$SZ`($ctx)
+       $ST     $C,`2*$SZ`($ctx)
+       $ST     $D,`3*$SZ`($ctx)
+       $ST     $E,`4*$SZ`($ctx)
+       $ST     $F,`5*$SZ`($ctx)
+       $ST     $G,`6*$SZ`($ctx)
+       $ST     $H,`7*$SZ`($ctx)
+       la      $inp,`16*$SZ`($inp)
+       clg     $inp,`$frame+32`($sp)
+       jne     .Lloop
+
+       lmg     %r6,%r15,`$frame+48`($sp)       
+       br      %r14
+.size  $Func,.-$Func
+.string        "SHA${label} block transform for s390x, CRYPTOGAMS by <appro\@openssl.org>"
+___
+
+$code =~ s/\`([^\`]*)\`/eval $1/gem;
+# unlike 32-bit shift 64-bit one takes three arguments
+$code =~ s/(srlg\s+)(%r[0-9]+),/$1$2,$2,/gm;
+
+print $code;
+close STDOUT;