2
0

s390x-gf2m.pl 5.0 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221
  1. #!/usr/bin/env perl
  2. #
  3. # ====================================================================
  4. # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
  5. # project. The module is, however, dual licensed under OpenSSL and
  6. # CRYPTOGAMS licenses depending on where you obtain it. For further
  7. # details see http://www.openssl.org/~appro/cryptogams/.
  8. # ====================================================================
  9. #
  10. # May 2011
  11. #
  12. # The module implements bn_GF2m_mul_2x2 polynomial multiplication used
  13. # in bn_gf2m.c. It's kind of low-hanging mechanical port from C for
  14. # the time being... gcc 4.3 appeared to generate poor code, therefore
  15. # the effort. And indeed, the module delivers 55%-90%(*) improvement
  16. # on haviest ECDSA verify and ECDH benchmarks for 163- and 571-bit
  17. # key lengths on z990, 30%-55%(*) - on z10, and 70%-110%(*) - on z196.
  18. # This is for 64-bit build. In 32-bit "highgprs" case improvement is
  19. # even higher, for example on z990 it was measured 80%-150%. ECDSA
  20. # sign is modest 9%-12% faster. Keep in mind that these coefficients
  21. # are not ones for bn_GF2m_mul_2x2 itself, as not all CPU time is
  22. # burnt in it...
  23. #
  24. # (*) gcc 4.1 was observed to deliver better results than gcc 4.3,
  25. # so that improvement coefficients can vary from one specific
  26. # setup to another.
  27. $flavour = shift;
  28. if ($flavour =~ /3[12]/) {
  29. $SIZE_T=4;
  30. $g="";
  31. } else {
  32. $SIZE_T=8;
  33. $g="g";
  34. }
  35. while (($output=shift) && ($output!~/^\w[\w\-]*\.\w+$/)) {}
  36. open STDOUT,">$output";
  37. $stdframe=16*$SIZE_T+4*8;
  38. $rp="%r2";
  39. $a1="%r3";
  40. $a0="%r4";
  41. $b1="%r5";
  42. $b0="%r6";
  43. $ra="%r14";
  44. $sp="%r15";
  45. @T=("%r0","%r1");
  46. @i=("%r12","%r13");
  47. ($a1,$a2,$a4,$a8,$a12,$a48)=map("%r$_",(6..11));
  48. ($lo,$hi,$b)=map("%r$_",(3..5)); $a=$lo; $mask=$a8;
  49. $code.=<<___;
  50. .text
  51. .type _mul_1x1,\@function
  52. .align 16
  53. _mul_1x1:
  54. lgr $a1,$a
  55. sllg $a2,$a,1
  56. sllg $a4,$a,2
  57. sllg $a8,$a,3
  58. srag $lo,$a1,63 # broadcast 63rd bit
  59. nihh $a1,0x1fff
  60. srag @i[0],$a2,63 # broadcast 62nd bit
  61. nihh $a2,0x3fff
  62. srag @i[1],$a4,63 # broadcast 61st bit
  63. nihh $a4,0x7fff
  64. ngr $lo,$b
  65. ngr @i[0],$b
  66. ngr @i[1],$b
  67. lghi @T[0],0
  68. lgr $a12,$a1
  69. stg @T[0],`$stdframe+0*8`($sp) # tab[0]=0
  70. xgr $a12,$a2
  71. stg $a1,`$stdframe+1*8`($sp) # tab[1]=a1
  72. lgr $a48,$a4
  73. stg $a2,`$stdframe+2*8`($sp) # tab[2]=a2
  74. xgr $a48,$a8
  75. stg $a12,`$stdframe+3*8`($sp) # tab[3]=a1^a2
  76. xgr $a1,$a4
  77. stg $a4,`$stdframe+4*8`($sp) # tab[4]=a4
  78. xgr $a2,$a4
  79. stg $a1,`$stdframe+5*8`($sp) # tab[5]=a1^a4
  80. xgr $a12,$a4
  81. stg $a2,`$stdframe+6*8`($sp) # tab[6]=a2^a4
  82. xgr $a1,$a48
  83. stg $a12,`$stdframe+7*8`($sp) # tab[7]=a1^a2^a4
  84. xgr $a2,$a48
  85. stg $a8,`$stdframe+8*8`($sp) # tab[8]=a8
  86. xgr $a12,$a48
  87. stg $a1,`$stdframe+9*8`($sp) # tab[9]=a1^a8
  88. xgr $a1,$a4
  89. stg $a2,`$stdframe+10*8`($sp) # tab[10]=a2^a8
  90. xgr $a2,$a4
  91. stg $a12,`$stdframe+11*8`($sp) # tab[11]=a1^a2^a8
  92. xgr $a12,$a4
  93. stg $a48,`$stdframe+12*8`($sp) # tab[12]=a4^a8
  94. srlg $hi,$lo,1
  95. stg $a1,`$stdframe+13*8`($sp) # tab[13]=a1^a4^a8
  96. sllg $lo,$lo,63
  97. stg $a2,`$stdframe+14*8`($sp) # tab[14]=a2^a4^a8
  98. srlg @T[0],@i[0],2
  99. stg $a12,`$stdframe+15*8`($sp) # tab[15]=a1^a2^a4^a8
  100. lghi $mask,`0xf<<3`
  101. sllg $a1,@i[0],62
  102. sllg @i[0],$b,3
  103. srlg @T[1],@i[1],3
  104. ngr @i[0],$mask
  105. sllg $a2,@i[1],61
  106. srlg @i[1],$b,4-3
  107. xgr $hi,@T[0]
  108. ngr @i[1],$mask
  109. xgr $lo,$a1
  110. xgr $hi,@T[1]
  111. xgr $lo,$a2
  112. xg $lo,$stdframe(@i[0],$sp)
  113. srlg @i[0],$b,8-3
  114. ngr @i[0],$mask
  115. ___
  116. for($n=1;$n<14;$n++) {
  117. $code.=<<___;
  118. lg @T[1],$stdframe(@i[1],$sp)
  119. srlg @i[1],$b,`($n+2)*4`-3
  120. sllg @T[0],@T[1],`$n*4`
  121. ngr @i[1],$mask
  122. srlg @T[1],@T[1],`64-$n*4`
  123. xgr $lo,@T[0]
  124. xgr $hi,@T[1]
  125. ___
  126. push(@i,shift(@i)); push(@T,shift(@T));
  127. }
  128. $code.=<<___;
  129. lg @T[1],$stdframe(@i[1],$sp)
  130. sllg @T[0],@T[1],`$n*4`
  131. srlg @T[1],@T[1],`64-$n*4`
  132. xgr $lo,@T[0]
  133. xgr $hi,@T[1]
  134. lg @T[0],$stdframe(@i[0],$sp)
  135. sllg @T[1],@T[0],`($n+1)*4`
  136. srlg @T[0],@T[0],`64-($n+1)*4`
  137. xgr $lo,@T[1]
  138. xgr $hi,@T[0]
  139. br $ra
  140. .size _mul_1x1,.-_mul_1x1
  141. .globl bn_GF2m_mul_2x2
  142. .type bn_GF2m_mul_2x2,\@function
  143. .align 16
  144. bn_GF2m_mul_2x2:
  145. stm${g} %r3,%r15,3*$SIZE_T($sp)
  146. lghi %r1,-$stdframe-128
  147. la %r0,0($sp)
  148. la $sp,0(%r1,$sp) # alloca
  149. st${g} %r0,0($sp) # back chain
  150. ___
  151. if ($SIZE_T==8) {
  152. my @r=map("%r$_",(6..9));
  153. $code.=<<___;
  154. bras $ra,_mul_1x1 # a1·b1
  155. stmg $lo,$hi,16($rp)
  156. lg $a,`$stdframe+128+4*$SIZE_T`($sp)
  157. lg $b,`$stdframe+128+6*$SIZE_T`($sp)
  158. bras $ra,_mul_1x1 # a0·b0
  159. stmg $lo,$hi,0($rp)
  160. lg $a,`$stdframe+128+3*$SIZE_T`($sp)
  161. lg $b,`$stdframe+128+5*$SIZE_T`($sp)
  162. xg $a,`$stdframe+128+4*$SIZE_T`($sp)
  163. xg $b,`$stdframe+128+6*$SIZE_T`($sp)
  164. bras $ra,_mul_1x1 # (a0+a1)·(b0+b1)
  165. lmg @r[0],@r[3],0($rp)
  166. xgr $lo,$hi
  167. xgr $hi,@r[1]
  168. xgr $lo,@r[0]
  169. xgr $hi,@r[2]
  170. xgr $lo,@r[3]
  171. xgr $hi,@r[3]
  172. xgr $lo,$hi
  173. stg $hi,16($rp)
  174. stg $lo,8($rp)
  175. ___
  176. } else {
  177. $code.=<<___;
  178. sllg %r3,%r3,32
  179. sllg %r5,%r5,32
  180. or %r3,%r4
  181. or %r5,%r6
  182. bras $ra,_mul_1x1
  183. rllg $lo,$lo,32
  184. rllg $hi,$hi,32
  185. stmg $lo,$hi,0($rp)
  186. ___
  187. }
  188. $code.=<<___;
  189. lm${g} %r6,%r15,`$stdframe+128+6*$SIZE_T`($sp)
  190. br $ra
  191. .size bn_GF2m_mul_2x2,.-bn_GF2m_mul_2x2
  192. .string "GF(2^m) Multiplication for s390x, CRYPTOGAMS by <appro\@openssl.org>"
  193. ___
  194. $code =~ s/\`([^\`]*)\`/eval($1)/gem;
  195. print $code;
  196. close STDOUT;