sparcv9-gf2m.pl 4.4 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190
  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. # October 2012
  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... Except that it has two code paths: one suitable
  15. # for all SPARCv9 processors and one for VIS3-capable ones. Former
  16. # delivers ~25-45% more, more for longer keys, heaviest DH and DSA
  17. # verify operations on venerable UltraSPARC II. On T4 VIS3 code is
  18. # ~100-230% faster than gcc-generated code and ~35-90% faster than
  19. # the pure SPARCv9 code path.
  20. $locals=16*8;
  21. $tab="%l0";
  22. @T=("%g2","%g3");
  23. @i=("%g4","%g5");
  24. ($a1,$a2,$a4,$a8,$a12,$a48)=map("%o$_",(0..5));
  25. ($lo,$hi,$b)=("%g1",$a8,"%o7"); $a=$lo;
  26. $code.=<<___;
  27. #include <sparc_arch.h>
  28. #ifdef __arch64__
  29. .register %g2,#scratch
  30. .register %g3,#scratch
  31. #endif
  32. #ifdef __PIC__
  33. SPARC_PIC_THUNK(%g1)
  34. #endif
  35. .globl bn_GF2m_mul_2x2
  36. .align 16
  37. bn_GF2m_mul_2x2:
  38. SPARC_LOAD_ADDRESS_LEAF(OPENSSL_sparcv9cap_P,%g1,%g5)
  39. ld [%g1+0],%g1 ! OPENSSL_sparcv9cap_P[0]
  40. andcc %g1, SPARCV9_VIS3, %g0
  41. bz,pn %icc,.Lsoftware
  42. nop
  43. sllx %o1, 32, %o1
  44. sllx %o3, 32, %o3
  45. or %o2, %o1, %o1
  46. or %o4, %o3, %o3
  47. .word 0x95b262ab ! xmulx %o1, %o3, %o2
  48. .word 0x99b262cb ! xmulxhi %o1, %o3, %o4
  49. srlx %o2, 32, %o1 ! 13 cycles later
  50. st %o2, [%o0+0]
  51. st %o1, [%o0+4]
  52. srlx %o4, 32, %o3
  53. st %o4, [%o0+8]
  54. retl
  55. st %o3, [%o0+12]
  56. .align 16
  57. .Lsoftware:
  58. save %sp,-STACK_FRAME-$locals,%sp
  59. sllx %i1,32,$a
  60. mov -1,$a12
  61. sllx %i3,32,$b
  62. or %i2,$a,$a
  63. srlx $a12,1,$a48 ! 0x7fff...
  64. or %i4,$b,$b
  65. srlx $a12,2,$a12 ! 0x3fff...
  66. add %sp,STACK_BIAS+STACK_FRAME,$tab
  67. sllx $a,2,$a4
  68. mov $a,$a1
  69. sllx $a,1,$a2
  70. srax $a4,63,@i[1] ! broadcast 61st bit
  71. and $a48,$a4,$a4 ! (a<<2)&0x7fff...
  72. srlx $a48,2,$a48
  73. srax $a2,63,@i[0] ! broadcast 62nd bit
  74. and $a12,$a2,$a2 ! (a<<1)&0x3fff...
  75. srax $a1,63,$lo ! broadcast 63rd bit
  76. and $a48,$a1,$a1 ! (a<<0)&0x1fff...
  77. sllx $a1,3,$a8
  78. and $b,$lo,$lo
  79. and $b,@i[0],@i[0]
  80. and $b,@i[1],@i[1]
  81. stx %g0,[$tab+0*8] ! tab[0]=0
  82. xor $a1,$a2,$a12
  83. stx $a1,[$tab+1*8] ! tab[1]=a1
  84. stx $a2,[$tab+2*8] ! tab[2]=a2
  85. xor $a4,$a8,$a48
  86. stx $a12,[$tab+3*8] ! tab[3]=a1^a2
  87. xor $a4,$a1,$a1
  88. stx $a4,[$tab+4*8] ! tab[4]=a4
  89. xor $a4,$a2,$a2
  90. stx $a1,[$tab+5*8] ! tab[5]=a1^a4
  91. xor $a4,$a12,$a12
  92. stx $a2,[$tab+6*8] ! tab[6]=a2^a4
  93. xor $a48,$a1,$a1
  94. stx $a12,[$tab+7*8] ! tab[7]=a1^a2^a4
  95. xor $a48,$a2,$a2
  96. stx $a8,[$tab+8*8] ! tab[8]=a8
  97. xor $a48,$a12,$a12
  98. stx $a1,[$tab+9*8] ! tab[9]=a1^a8
  99. xor $a4,$a1,$a1
  100. stx $a2,[$tab+10*8] ! tab[10]=a2^a8
  101. xor $a4,$a2,$a2
  102. stx $a12,[$tab+11*8] ! tab[11]=a1^a2^a8
  103. xor $a4,$a12,$a12
  104. stx $a48,[$tab+12*8] ! tab[12]=a4^a8
  105. srlx $lo,1,$hi
  106. stx $a1,[$tab+13*8] ! tab[13]=a1^a4^a8
  107. sllx $lo,63,$lo
  108. stx $a2,[$tab+14*8] ! tab[14]=a2^a4^a8
  109. srlx @i[0],2,@T[0]
  110. stx $a12,[$tab+15*8] ! tab[15]=a1^a2^a4^a8
  111. sllx @i[0],62,$a1
  112. sllx $b,3,@i[0]
  113. srlx @i[1],3,@T[1]
  114. and @i[0],`0xf<<3`,@i[0]
  115. sllx @i[1],61,$a2
  116. ldx [$tab+@i[0]],@i[0]
  117. srlx $b,4-3,@i[1]
  118. xor @T[0],$hi,$hi
  119. and @i[1],`0xf<<3`,@i[1]
  120. xor $a1,$lo,$lo
  121. ldx [$tab+@i[1]],@i[1]
  122. xor @T[1],$hi,$hi
  123. xor @i[0],$lo,$lo
  124. srlx $b,8-3,@i[0]
  125. xor $a2,$lo,$lo
  126. and @i[0],`0xf<<3`,@i[0]
  127. ___
  128. for($n=1;$n<14;$n++) {
  129. $code.=<<___;
  130. sllx @i[1],`$n*4`,@T[0]
  131. ldx [$tab+@i[0]],@i[0]
  132. srlx @i[1],`64-$n*4`,@T[1]
  133. xor @T[0],$lo,$lo
  134. srlx $b,`($n+2)*4`-3,@i[1]
  135. xor @T[1],$hi,$hi
  136. and @i[1],`0xf<<3`,@i[1]
  137. ___
  138. push(@i,shift(@i)); push(@T,shift(@T));
  139. }
  140. $code.=<<___;
  141. sllx @i[1],`$n*4`,@T[0]
  142. ldx [$tab+@i[0]],@i[0]
  143. srlx @i[1],`64-$n*4`,@T[1]
  144. xor @T[0],$lo,$lo
  145. sllx @i[0],`($n+1)*4`,@T[0]
  146. xor @T[1],$hi,$hi
  147. srlx @i[0],`64-($n+1)*4`,@T[1]
  148. xor @T[0],$lo,$lo
  149. xor @T[1],$hi,$hi
  150. srlx $lo,32,%i1
  151. st $lo,[%i0+0]
  152. st %i1,[%i0+4]
  153. srlx $hi,32,%i2
  154. st $hi,[%i0+8]
  155. st %i2,[%i0+12]
  156. ret
  157. restore
  158. .type bn_GF2m_mul_2x2,#function
  159. .size bn_GF2m_mul_2x2,.-bn_GF2m_mul_2x2
  160. .asciz "GF(2^m) Multiplication for SPARCv9, CRYPTOGAMS by <appro\@openssl.org>"
  161. .align 4
  162. ___
  163. $code =~ s/\`([^\`]*)\`/eval($1)/gem;
  164. print $code;
  165. close STDOUT;