aoptcpu.pas 17 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490
  1. {
  2. Copyright (c) 1998-2002 by Jonas Maebe, member of the Free Pascal
  3. Development Team
  4. This unit implements the Z80 optimizer object
  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. {$i fpcdefs.inc}
  20. {$define DEBUG_AOPTCPU}
  21. Interface
  22. uses cpubase, cgbase, aasmtai, aopt,AoptObj, aoptcpub;
  23. Type
  24. TCpuAsmOptimizer = class(TAsmOptimizer)
  25. { outputs a debug message into the assembler file }
  26. procedure DebugMsg(const s: string; p: tai);
  27. { checks whether loading a new value in reg1 overwrites the entirety of reg2 }
  28. function Reg1WriteOverwritesReg2Entirely(reg1, reg2: tregister): boolean;
  29. { checks whether reading the value in reg1 depends on the value of reg2. This
  30. is very similar to SuperRegisterEquals, except it takes into account that
  31. R_SUBH and R_SUBL are independendent (e.g. reading from AL does not
  32. depend on the value in AH). }
  33. function Reg1ReadDependsOnReg2(reg1, reg2: tregister): boolean;
  34. Function GetNextInstructionUsingReg(Current: tai; Var Next: tai;reg : TRegister): Boolean;
  35. function RegLoadedWithNewValue(reg : tregister; hp : tai) : boolean; override;
  36. function InstructionLoadsFromReg(const reg : TRegister; const hp : tai) : boolean; override;
  37. { uses the same constructor as TAopObj }
  38. function PeepHoleOptPass1Cpu(var p: tai): boolean; override;
  39. procedure PeepHoleOptPass2;override;
  40. End;
  41. Implementation
  42. uses
  43. cutils,
  44. verbose,
  45. cpuinfo,
  46. aasmbase,aasmcpu,aasmdata,
  47. globals,globtype,
  48. cgutils;
  49. type
  50. TAsmOpSet = set of TAsmOp;
  51. function CanBeCond(p : tai) : boolean;
  52. begin
  53. result:=(p.typ=ait_instruction) and (taicpu(p).condition=C_None);
  54. end;
  55. function RefsEqual(const r1, r2: treference): boolean;
  56. begin
  57. refsequal :=
  58. (r1.offset = r2.offset) and
  59. (r1.base = r2.base) and
  60. (r1.index = r2.index) and (r1.scalefactor = r2.scalefactor) and
  61. (r1.symbol=r2.symbol) and (r1.refaddr = r2.refaddr) and
  62. (r1.relsymbol = r2.relsymbol);
  63. end;
  64. function MatchOperand(const oper1: TOper; const oper2: TOper): boolean; inline;
  65. begin
  66. result:=oper1.typ=oper2.typ;
  67. if result then
  68. case oper1.typ of
  69. top_const:
  70. Result:=oper1.val = oper2.val;
  71. top_reg:
  72. Result:=oper1.reg = oper2.reg;
  73. top_ref:
  74. Result:=RefsEqual(oper1.ref^, oper2.ref^);
  75. else Result:=false;
  76. end
  77. end;
  78. function MatchOperand(const oper: TOper; const reg: TRegister): boolean; inline;
  79. begin
  80. result := (oper.typ = top_reg) and (oper.reg = reg);
  81. end;
  82. function MatchInstruction(const instr: tai; const op: TAsmOp): boolean;
  83. begin
  84. result :=
  85. (instr.typ = ait_instruction) and
  86. (taicpu(instr).opcode = op);
  87. end;
  88. function MatchInstruction(const instr: tai; const ops: TAsmOpSet): boolean;
  89. begin
  90. result :=
  91. (instr.typ = ait_instruction) and
  92. (taicpu(instr).opcode in ops);
  93. end;
  94. function MatchInstruction(const instr: tai; const ops: TAsmOpSet;opcount : byte): boolean;
  95. begin
  96. result :=
  97. (instr.typ = ait_instruction) and
  98. (taicpu(instr).opcode in ops) and
  99. (taicpu(instr).ops=opcount);
  100. end;
  101. function MatchOpType(const instr : tai;ot0,ot1 : toptype) : Boolean;
  102. begin
  103. Result:=(taicpu(instr).ops=2) and
  104. (taicpu(instr).oper[0]^.typ=ot0) and
  105. (taicpu(instr).oper[1]^.typ=ot1);
  106. end;
  107. {$ifdef DEBUG_AOPTCPU}
  108. procedure TCpuAsmOptimizer.DebugMsg(const s: string;p : tai);
  109. begin
  110. asml.insertbefore(tai_comment.Create(strpnew(s)), p);
  111. end;
  112. {$else DEBUG_AOPTCPU}
  113. procedure TCpuAsmOptimizer.DebugMsg(const s: string;p : tai);inline;
  114. begin
  115. end;
  116. {$endif DEBUG_AOPTCPU}
  117. function TCpuAsmOptimizer.Reg1WriteOverwritesReg2Entirely(reg1, reg2: tregister): boolean;
  118. begin
  119. case reg1 of
  120. NR_F:
  121. result:=SuperRegistersEqual(reg2,NR_DEFAULTFLAGS);
  122. NR_AF:
  123. result:=(reg2=NR_A) or (reg2=NR_AF) or SuperRegistersEqual(reg2,NR_DEFAULTFLAGS);
  124. NR_BC:
  125. result:=(reg2=NR_B) or (reg2=NR_C) or (reg2=NR_BC);
  126. NR_DE:
  127. result:=(reg2=NR_D) or (reg2=NR_E) or (reg2=NR_DE);
  128. NR_HL:
  129. result:=(reg2=NR_H) or (reg2=NR_L) or (reg2=NR_HL);
  130. NR_F_:
  131. result:=SuperRegistersEqual(reg2,NR_F_);
  132. NR_AF_:
  133. result:=(reg2=NR_A_) or (reg2=NR_AF_) or SuperRegistersEqual(reg2,NR_F_);
  134. NR_BC_:
  135. result:=(reg2=NR_B_) or (reg2=NR_C_) or (reg2=NR_BC_);
  136. NR_DE_:
  137. result:=(reg2=NR_D_) or (reg2=NR_E_) or (reg2=NR_DE_);
  138. NR_HL_:
  139. result:=(reg2=NR_H_) or (reg2=NR_L_) or (reg2=NR_HL_);
  140. else
  141. result:=reg1=reg2;
  142. end;
  143. end;
  144. function TCpuAsmOptimizer.Reg1ReadDependsOnReg2(reg1, reg2: tregister): boolean;
  145. begin
  146. case reg1 of
  147. NR_AF:
  148. result:=(reg2=NR_A) or (reg2=NR_AF) or SuperRegistersEqual(reg2,NR_DEFAULTFLAGS);
  149. NR_A:
  150. result:=(reg2=NR_A) or (reg2=NR_AF);
  151. NR_F:
  152. result:=SuperRegistersEqual(reg2,NR_DEFAULTFLAGS);
  153. NR_BC:
  154. result:=(reg2=NR_B) or (reg2=NR_C) or (reg2=NR_BC);
  155. NR_B:
  156. result:=(reg2=NR_B) or (reg2=NR_BC);
  157. NR_C:
  158. result:=(reg2=NR_C) or (reg2=NR_BC);
  159. NR_DE:
  160. result:=(reg2=NR_D) or (reg2=NR_E) or (reg2=NR_DE);
  161. NR_D:
  162. result:=(reg2=NR_D) or (reg2=NR_DE);
  163. NR_E:
  164. result:=(reg2=NR_E) or (reg2=NR_DE);
  165. NR_HL:
  166. result:=(reg2=NR_H) or (reg2=NR_L) or (reg2=NR_HL);
  167. NR_H:
  168. result:=(reg2=NR_H) or (reg2=NR_HL);
  169. NR_L:
  170. result:=(reg2=NR_L) or (reg2=NR_HL);
  171. NR_AF_:
  172. result:=(reg2=NR_A_) or (reg2=NR_AF_) or SuperRegistersEqual(reg2,NR_F_);
  173. NR_A_:
  174. result:=(reg2=NR_A_) or (reg2=NR_AF_);
  175. NR_F_:
  176. result:=SuperRegistersEqual(reg2,NR_F_);
  177. NR_BC_:
  178. result:=(reg2=NR_B_) or (reg2=NR_C_) or (reg2=NR_BC_);
  179. NR_B_:
  180. result:=(reg2=NR_B_) or (reg2=NR_BC_);
  181. NR_C_:
  182. result:=(reg2=NR_C_) or (reg2=NR_BC_);
  183. NR_DE_:
  184. result:=(reg2=NR_D_) or (reg2=NR_E_) or (reg2=NR_DE_);
  185. NR_D_:
  186. result:=(reg2=NR_D_) or (reg2=NR_DE_);
  187. NR_E_:
  188. result:=(reg2=NR_E_) or (reg2=NR_DE_);
  189. NR_HL_:
  190. result:=(reg2=NR_H_) or (reg2=NR_L_) or (reg2=NR_HL_);
  191. NR_H_:
  192. result:=(reg2=NR_H_) or (reg2=NR_HL_);
  193. NR_L_:
  194. result:=(reg2=NR_L_) or (reg2=NR_HL_);
  195. else
  196. result:=reg1=reg2;
  197. end;
  198. end;
  199. function TCpuAsmOptimizer.GetNextInstructionUsingReg(Current: tai;
  200. var Next: tai; reg: TRegister): Boolean;
  201. begin
  202. Next:=Current;
  203. repeat
  204. Result:=GetNextInstruction(Next,Next);
  205. until not(cs_opt_level3 in current_settings.optimizerswitches) or not(Result) or (Next.typ<>ait_instruction) or (RegInInstruction(reg,Next)) or
  206. (is_calljmp(taicpu(Next).opcode));
  207. end;
  208. function TCpuAsmOptimizer.RegLoadedWithNewValue(reg: tregister; hp: tai): boolean;
  209. var
  210. p: taicpu;
  211. begin
  212. if not assigned(hp) or
  213. (hp.typ <> ait_instruction) then
  214. begin
  215. Result := false;
  216. exit;
  217. end;
  218. p := taicpu(hp);
  219. if SuperRegistersEqual(reg,NR_DEFAULTFLAGS) and (reg<>NR_AF) then
  220. begin
  221. case p.opcode of
  222. { todo: IN,INI,INIR,IND,INDR,OUT,OUTI,OTIR,OUTD,OTDR}
  223. A_PUSH,A_POP,A_EX,A_EXX,A_NOP,A_HALT,A_DI,A_EI,A_IM,A_SET,A_RES,A_JP,A_JR,A_DJNZ,A_CALL,A_RET,A_RETI,A_RETN,A_RST:
  224. result:=false;
  225. A_LD:
  226. begin
  227. if p.ops<>2 then
  228. internalerror(2020051112);
  229. { LD A,I or LD A,R ? }
  230. if (p.oper[0]^.typ=top_reg) and (p.oper[0]^.reg=NR_A) and
  231. (p.oper[1]^.typ=top_reg) and ((p.oper[1]^.reg=NR_I) or (p.oper[1]^.reg=NR_R)) then
  232. result:=(reg=NR_ADDSUBTRACTFLAG) or
  233. (reg=NR_PARITYOVERFLOWFLAG) or
  234. (reg=NR_HALFCARRYFLAG) or
  235. (reg=NR_ZEROFLAG) or
  236. (reg=NR_SIGNFLAG)
  237. else
  238. result:=false;
  239. end;
  240. A_LDI,A_LDIR,A_LDD,A_LDDR:
  241. result:=(reg=NR_ADDSUBTRACTFLAG) or
  242. (reg=NR_PARITYOVERFLOWFLAG) or
  243. (reg=NR_HALFCARRYFLAG);
  244. A_INC,A_DEC:
  245. begin
  246. if p.ops<>1 then
  247. internalerror(2020051602);
  248. if (p.oper[0]^.typ=top_reg) and ((p.oper[0]^.reg=NR_BC) or
  249. (p.oper[0]^.reg=NR_DE) or
  250. (p.oper[0]^.reg=NR_HL) or
  251. (p.oper[0]^.reg=NR_SP) or
  252. (p.oper[0]^.reg=NR_IX) or
  253. (p.oper[0]^.reg=NR_IY)) then
  254. result:=false
  255. else
  256. result:=(reg=NR_ADDSUBTRACTFLAG) or
  257. (reg=NR_PARITYOVERFLOWFLAG) or
  258. (reg=NR_HALFCARRYFLAG) or
  259. (reg=NR_ZEROFLAG) or
  260. (reg=NR_SIGNFLAG);
  261. end;
  262. A_CPI,A_CPIR,A_CPD,A_CPDR,A_RLD,A_RRD,A_BIT:
  263. result:=(reg=NR_ADDSUBTRACTFLAG) or
  264. (reg=NR_PARITYOVERFLOWFLAG) or
  265. (reg=NR_HALFCARRYFLAG) or
  266. (reg=NR_ZEROFLAG) or
  267. (reg=NR_SIGNFLAG);
  268. A_ADD:
  269. begin
  270. if p.ops<>2 then
  271. internalerror(2020051601);
  272. if (p.oper[0]^.typ=top_reg) and ((p.oper[0]^.reg=NR_HL) or (p.oper[0]^.reg=NR_IX) or (p.oper[0]^.reg=NR_IY)) then
  273. result:=(reg=NR_HALFCARRYFLAG) or
  274. (reg=NR_ADDSUBTRACTFLAG) or
  275. (reg=NR_CARRYFLAG)
  276. else
  277. result:=true;
  278. end;
  279. A_ADC,A_SUB,A_SBC,A_AND,A_OR,A_XOR,A_CP,A_NEG,A_RLC,A_RL,A_RRC,A_RR,A_SLA,A_SRA,A_SRL:
  280. result:=true;
  281. A_DAA:
  282. result:=(reg=NR_PARITYOVERFLOWFLAG) or
  283. (reg=NR_HALFCARRYFLAG) or
  284. (reg=NR_ZEROFLAG) or
  285. (reg=NR_SIGNFLAG) or
  286. (reg=NR_CARRYFLAG);
  287. A_CPL:
  288. result:=(reg=NR_HALFCARRYFLAG) or
  289. (reg=NR_ADDSUBTRACTFLAG);
  290. A_CCF,A_SCF,A_RLCA,A_RLA,A_RRCA,A_RRA:
  291. result:=(reg=NR_HALFCARRYFLAG) or
  292. (reg=NR_ADDSUBTRACTFLAG) or
  293. (reg=NR_CARRYFLAG);
  294. else
  295. internalerror(2020051111);
  296. end;
  297. end
  298. else
  299. case p.opcode of
  300. A_LD:
  301. begin
  302. if p.ops<>2 then
  303. internalerror(2020051112);
  304. result:=(p.oper[0]^.typ = top_reg) and
  305. (Reg1WriteOverwritesReg2Entirely(p.oper[0]^.reg,reg)) and
  306. ((p.oper[1]^.typ = top_const) or
  307. ((p.oper[1]^.typ = top_reg) and not(Reg1ReadDependsOnReg2(p.oper[1]^.reg,reg))) or
  308. ((p.oper[1]^.typ = top_ref) and not RegInRef(reg,p.oper[1]^.ref^)));
  309. end;
  310. { todo: all the remaining instructions }
  311. else
  312. internalerror(2020051111);
  313. end;
  314. end;
  315. function TCpuAsmOptimizer.InstructionLoadsFromReg(const reg: TRegister; const hp: tai): boolean;
  316. var
  317. p: taicpu;
  318. begin
  319. Result := false;
  320. if not (assigned(hp) and (hp.typ = ait_instruction)) then
  321. exit;
  322. p:=taicpu(hp);
  323. case p.opcode of
  324. A_LD,A_BIT,A_SET,A_RES:
  325. begin
  326. if p.ops<>2 then
  327. internalerror(2020051102);
  328. result:=((p.oper[0]^.typ=top_ref) and RegInRef(reg,p.oper[0]^.ref^)) or
  329. RegInOp(reg,p.oper[1]^);
  330. end;
  331. A_PUSH,A_INC,A_DEC,A_RLC,A_RRC,A_SLA,A_SRA,A_SRL:
  332. begin
  333. if p.ops<>1 then
  334. internalerror(2020051103);
  335. result:=RegInOp(reg,p.oper[0]^);
  336. end;
  337. A_POP:
  338. result:=(reg=NR_SP);
  339. A_EX,A_ADD,A_SUB,A_AND,A_OR,A_XOR,A_CP:
  340. begin
  341. if p.ops<>2 then
  342. internalerror(2020051104);
  343. result:=RegInOp(reg,p.oper[0]^) or
  344. RegInOp(reg,p.oper[1]^);
  345. end;
  346. A_EXX:
  347. result:=SuperRegistersEqual(reg,NR_BC) or SuperRegistersEqual(reg,NR_DE) or SuperRegistersEqual(reg,NR_HL) or
  348. SuperRegistersEqual(reg,NR_BC_) or SuperRegistersEqual(reg,NR_DE_) or SuperRegistersEqual(reg,NR_HL_);
  349. A_LDI,A_LDIR,A_LDD,A_LDDR:
  350. result:=SuperRegistersEqual(reg,NR_BC) or SuperRegistersEqual(reg,NR_DE) or SuperRegistersEqual(reg,NR_HL);
  351. A_CPI,A_CPIR,A_CPD,A_CPDR:
  352. result:=SuperRegistersEqual(reg,NR_BC) or SuperRegistersEqual(reg,NR_HL) or RegistersInterfere(reg,NR_A);
  353. A_ADC,A_SBC:
  354. begin
  355. if p.ops<>2 then
  356. internalerror(2020051105);
  357. result:=RegInOp(reg,p.oper[0]^) or
  358. RegInOp(reg,p.oper[1]^) or (reg=NR_CARRYFLAG) or (reg=NR_DEFAULTFLAGS);
  359. end;
  360. A_DAA:
  361. result:=RegistersInterfere(reg,NR_A) or (reg=NR_CARRYFLAG) or (reg=NR_HALFCARRYFLAG) or (reg=NR_ADDSUBTRACTFLAG) or (reg=NR_DEFAULTFLAGS);
  362. A_CPL,A_NEG,A_RLCA,A_RRCA:
  363. result:=RegistersInterfere(reg,NR_A);
  364. A_CCF:
  365. result:=(reg=NR_CARRYFLAG) or (reg=NR_DEFAULTFLAGS);
  366. A_SCF,A_NOP,A_HALT,A_DI,A_EI,A_IM:
  367. result:=false;
  368. A_RLA,A_RRA:
  369. result:=RegistersInterfere(reg,NR_A) or (reg=NR_CARRYFLAG) or (reg=NR_DEFAULTFLAGS);
  370. A_RL,A_RR:
  371. begin
  372. if p.ops<>1 then
  373. internalerror(2020051106);
  374. result:=RegInOp(reg,p.oper[0]^) or (reg=NR_CARRYFLAG) or (reg=NR_DEFAULTFLAGS);
  375. end;
  376. A_RLD,A_RRD:
  377. result:=RegistersInterfere(reg,NR_A) or RegistersInterfere(reg,NR_HL);
  378. A_JP,A_JR:
  379. begin
  380. if p.ops<>1 then
  381. internalerror(2020051107);
  382. if RegInOp(reg,p.oper[0]^) then
  383. result:=true
  384. else
  385. case p.condition of
  386. C_None:
  387. result:=false;
  388. C_NZ,C_Z:
  389. result:=(reg=NR_ZEROFLAG) or (reg=NR_DEFAULTFLAGS);
  390. C_NC,C_C:
  391. result:=(reg=NR_CARRYFLAG) or (reg=NR_DEFAULTFLAGS);
  392. C_PO,C_PE:
  393. result:=(reg=NR_PARITYOVERFLOWFLAG) or (reg=NR_DEFAULTFLAGS);
  394. C_P,C_M:
  395. result:=(reg=NR_SIGNFLAG) or (reg=NR_DEFAULTFLAGS);
  396. end;
  397. end;
  398. A_DJNZ:
  399. result:=RegistersInterfere(reg,NR_B);
  400. A_CALL,A_RET,A_RETI,A_RETN,A_RST:
  401. result:=true;
  402. A_IN:
  403. begin
  404. if p.ops<>2 then
  405. internalerror(2020051109);
  406. result:=(p.oper[1]^.typ=top_ref) and (p.oper[1]^.ref^.base=NR_C) and RegistersInterfere(reg,NR_BC);
  407. end;
  408. A_OUT:
  409. begin
  410. if p.ops<>2 then
  411. internalerror(2020051110);
  412. result:=RegInOp(reg,p.oper[1]^) or (p.oper[0]^.typ=top_ref) and (p.oper[0]^.ref^.base=NR_C) and RegistersInterfere(reg,NR_BC);
  413. end;
  414. A_INI,A_INIR,A_IND,A_INDR,A_OUTI,A_OTIR,A_OUTD,A_OTDR:
  415. result:=SuperRegistersEqual(reg,NR_BC) or SuperRegistersEqual(reg,NR_HL);
  416. else
  417. internalerror(2020051101);
  418. end;
  419. end;
  420. function TCpuAsmOptimizer.PeepHoleOptPass1Cpu(var p: tai): boolean;
  421. var
  422. hp1,hp2,hp3,hp4,hp5: tai;
  423. alloc, dealloc: tai_regalloc;
  424. i: integer;
  425. l: TAsmLabel;
  426. //TmpUsedRegs : TAllUsedRegs;
  427. begin
  428. result := false;
  429. //case p.typ of
  430. // ait_instruction:
  431. // begin
  432. // end;
  433. //end;
  434. end;
  435. procedure TCpuAsmOptimizer.PeepHoleOptPass2;
  436. begin
  437. end;
  438. begin
  439. casmoptimizer:=TCpuAsmOptimizer;
  440. End.