aoptcpu.pas 7.8 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279
  1. {
  2. Copyright (c) 1998-2004 by Jonas Maebe
  3. This unit calls the optimization procedures to optimize the assembler
  4. code for sparc
  5. This program is free software; you can redistribute it and/or modify
  6. it under the terms of the GNU General Public License as published by
  7. the Free Software Foundation; either version 2 of the License, or
  8. (at your option) any later version.
  9. This program is distributed in the hope that it will be useful,
  10. but WITHOUT ANY WARRANTY; without even the implied warranty of
  11. MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  12. GNU General Public License for more details.
  13. You should have received a copy of the GNU General Public License
  14. along with this program; if not, write to the Free Software
  15. Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  16. ****************************************************************************
  17. }
  18. unit aoptcpu;
  19. {$i fpcdefs.inc}
  20. interface
  21. uses cpubase, aasmtai, aopt, aoptx86;
  22. type
  23. TCpuAsmOptimizer = class(TX86AsmOptimizer)
  24. function PrePeepHoleOptsCpu(var p: tai): boolean; override;
  25. function PeepHoleOptPass1Cpu(var p: tai): boolean; override;
  26. function PeepHoleOptPass2Cpu(var p: tai): boolean; override;
  27. function PostPeepHoleOptsCpu(var p : tai) : boolean; override;
  28. end;
  29. implementation
  30. uses
  31. globals,
  32. aasmcpu;
  33. function TCpuAsmOptimizer.PrePeepHoleOptsCpu(var p : tai) : boolean;
  34. begin
  35. result := false;
  36. case p.typ of
  37. ait_instruction:
  38. begin
  39. case taicpu(p).opcode of
  40. A_IMUL:
  41. result:=PrePeepholeOptIMUL(p);
  42. A_SAR,A_SHR:
  43. result:=PrePeepholeOptSxx(p);
  44. A_AND:
  45. Result:=PrePeepholeOptAND(p);
  46. else
  47. ;
  48. end;
  49. end;
  50. else
  51. ;
  52. end;
  53. end;
  54. function TCpuAsmOptimizer.PeepHoleOptPass1Cpu(var p: tai): boolean;
  55. begin
  56. result:=False;
  57. case p.typ of
  58. ait_instruction:
  59. begin
  60. case taicpu(p).opcode of
  61. A_ADD:
  62. Result:=OptPass1ADD(p);
  63. A_AND:
  64. Result:=OptPass1AND(p);
  65. A_IMUL:
  66. Result:=OptPass1Imul(p);
  67. A_MOV:
  68. Result:=OptPass1MOV(p);
  69. A_MOVSX,
  70. A_MOVSXD,
  71. A_MOVZX:
  72. Result:=OptPass1Movx(p);
  73. A_MOVDQA,
  74. A_MOVAPD,
  75. A_MOVAPS,
  76. A_MOVUPD,
  77. A_MOVUPS,
  78. A_VMOVAPS,
  79. A_VMOVAPD,
  80. A_VMOVUPS,
  81. A_VMOVUPD:
  82. result:=OptPass1_V_MOVAP(p);
  83. A_VMINSS,
  84. A_VMINSD,
  85. A_VMAXSS,
  86. A_VMAXSD,
  87. A_VSQRTSD,
  88. A_VSQRTSS,
  89. A_VDIVSD,
  90. A_VDIVSS,
  91. A_VSUBSD,
  92. A_VSUBSS,
  93. A_VMULSD,
  94. A_VMULSS,
  95. A_VADDSD,
  96. A_VADDSS,
  97. A_VANDPD,
  98. A_VANDPS,
  99. A_VORPD,
  100. A_VORPS:
  101. result:=OptPass1VOP(p);
  102. A_MULSD,
  103. A_MULSS,
  104. A_ADDSD,
  105. A_ADDSS:
  106. result:=OptPass1OP(p);
  107. A_VMOVSD,
  108. A_VMOVSS,
  109. A_MOVSD,
  110. A_MOVSS:
  111. result:=OptPass1MOVXX(p);
  112. A_LEA:
  113. result:=OptPass1LEA(p);
  114. A_SUB:
  115. result:=OptPass1Sub(p);
  116. A_SHL,A_SAL:
  117. result:=OptPass1SHLSAL(p);
  118. A_SHR:
  119. result:=OptPass1SHR(p);
  120. A_FSTP,A_FISTP:
  121. result:=OptPass1FSTP(p);
  122. A_FLD:
  123. result:=OptPass1FLD(p);
  124. A_CMP:
  125. result:=OptPass1Cmp(p);
  126. A_VPXORD,
  127. A_VPXORQ,
  128. A_VXORPS,
  129. A_VXORPD,
  130. A_VPXOR:
  131. Result:=OptPass1VPXor(p);
  132. A_VMOVDQA,
  133. A_VMOVDQU:
  134. Result:=OptPass1VMOVDQ(p);
  135. A_XORPS,
  136. A_XORPD,
  137. A_PXOR:
  138. Result:=OptPass1PXor(p);
  139. A_TEST:
  140. Result:=OptPass1Test(p);
  141. A_Jcc:
  142. Result:=OptPass1Jcc(p);
  143. A_SHRX,
  144. A_SHLX:
  145. Result:=OptPass1SHXX(p);
  146. A_VCVTSS2SD,
  147. A_CVTSS2SD:
  148. Result:=OptPass1_V_Cvtss2sd(p);
  149. else
  150. ;
  151. end;
  152. end;
  153. else
  154. ;
  155. end;
  156. { If this flag is set, force another run of pass 1 even if p wasn't
  157. changed }
  158. if aoc_ForceNewIteration in OptsToCheck then
  159. begin
  160. Exclude(OptsToCheck, aoc_ForceNewIteration);
  161. if not Result then
  162. begin
  163. if (p.typ in SkipInstr) then
  164. UpdateUsedRegs(p);
  165. p := tai(p.Next);
  166. Result := True;
  167. end;
  168. end;
  169. end;
  170. function TCpuAsmOptimizer.PeepHoleOptPass2Cpu(var p : tai) : boolean;
  171. begin
  172. Result := False;
  173. case p.typ of
  174. ait_instruction:
  175. begin
  176. case taicpu(p).opcode of
  177. A_MOV:
  178. Result:=OptPass2MOV(p);
  179. A_MOVZX:
  180. Result:=OptPass2Movx(p);
  181. A_IMUL:
  182. Result:=OptPass2Imul(p);
  183. A_JMP:
  184. Result:=OptPass2Jmp(p);
  185. A_Jcc:
  186. Result:=OptPass2Jcc(p);
  187. A_Lea:
  188. Result:=OptPass2Lea(p);
  189. A_SUB:
  190. Result:=OptPass2SUB(p);
  191. A_ADD:
  192. Result:=OptPass2ADD(p);
  193. A_SETcc:
  194. result:=OptPass2SETcc(p);
  195. else
  196. ;
  197. end;
  198. end;
  199. else
  200. ;
  201. end;
  202. end;
  203. function TCpuAsmOptimizer.PostPeepHoleOptsCpu(var p: tai): boolean;
  204. begin
  205. result := false;
  206. case p.typ of
  207. ait_instruction:
  208. begin
  209. case taicpu(p).opcode of
  210. A_MOV:
  211. Result:=PostPeepholeOptMov(p);
  212. A_AND:
  213. Result:=PostPeepholeOptAnd(p);
  214. A_MOVSX,
  215. A_MOVSXD:
  216. Result:=PostPeepholeOptMOVSX(p);
  217. A_MOVZX:
  218. Result:=PostPeepholeOptMovzx(p);
  219. A_CMP:
  220. Result:=PostPeepholeOptCmp(p);
  221. A_OR,
  222. A_TEST:
  223. Result:=PostPeepholeOptTestOr(p);
  224. A_XOR:
  225. Result:=PostPeepholeOptXor(p);
  226. A_CALL:
  227. Result:=PostPeepholeOptCall(p);
  228. A_LEA:
  229. Result:=PostPeepholeOptLea(p);
  230. A_PUSH:
  231. Result:=PostPeepholeOptPush(p);
  232. A_SHR:
  233. Result:=PostPeepholeOptShr(p);
  234. A_ADD,
  235. A_SUB:
  236. Result:=PostPeepholeOptADDSUB(p);
  237. A_VPXOR:
  238. Result:=PostPeepholeOptVPXOR(p);
  239. else
  240. ;
  241. end;
  242. { Optimise any reference-type operands (if Result is True, the
  243. instruction will be checked on the next iteration) }
  244. if not Result then
  245. OptimizeRefs(taicpu(p));
  246. end;
  247. else
  248. ;
  249. end;
  250. end;
  251. begin
  252. casmoptimizer := TCpuAsmOptimizer;
  253. end.