cgbase.pas 27 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887
  1. {
  2. Copyright (c) 1998-2002 by Florian Klaempfl
  3. Some basic types and constants for the code generation
  4. This program is free software; you can redistribute it and/or modify
  5. it under the terms of the GNU General Public License as published by
  6. the Free Software Foundation; either version 2 of the License, or
  7. (at your option) any later version.
  8. This program is distributed in the hope that it will be useful,
  9. but WITHOUT ANY WARRANTY; without even the implied warranty of
  10. MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  11. GNU General Public License for more details.
  12. You should have received a copy of the GNU General Public License
  13. along with this program; if not, write to the Free Software
  14. Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  15. ****************************************************************************
  16. }
  17. {# This unit exports some types which are used across the code generator }
  18. unit cgbase;
  19. {$i fpcdefs.inc}
  20. interface
  21. uses
  22. globtype,
  23. symconst;
  24. type
  25. { Location types where value can be stored }
  26. TCGLoc=(
  27. LOC_INVALID, { added for tracking problems}
  28. LOC_VOID, { no value is available }
  29. LOC_CONSTANT, { constant value }
  30. LOC_JUMP, { boolean results only, jump to false or true label }
  31. LOC_FLAGS, { boolean results only, flags are set }
  32. LOC_REGISTER, { in a processor register }
  33. LOC_CREGISTER, { Constant register which shouldn't be modified }
  34. LOC_FPUREGISTER, { FPU stack }
  35. LOC_CFPUREGISTER, { if it is a FPU register variable on the fpu stack }
  36. LOC_MMXREGISTER, { MMX register }
  37. { MMX register variable }
  38. LOC_CMMXREGISTER,
  39. { multimedia register }
  40. LOC_MMREGISTER,
  41. { Constant multimedia reg which shouldn't be modified }
  42. LOC_CMMREGISTER,
  43. { contiguous subset of bits of an integer register }
  44. LOC_SUBSETREG,
  45. LOC_CSUBSETREG,
  46. { contiguous subset of bits in memory }
  47. LOC_SUBSETREF,
  48. LOC_CSUBSETREF,
  49. { keep these last for range checking purposes }
  50. LOC_CREFERENCE, { in memory constant value reference (cannot change) }
  51. LOC_REFERENCE { in memory value }
  52. );
  53. TCGNonRefLoc=low(TCGLoc)..pred(LOC_CREFERENCE);
  54. TCGRefLoc=LOC_CREFERENCE..LOC_REFERENCE;
  55. trefaddr = (
  56. addr_no,
  57. addr_full,
  58. addr_pic,
  59. addr_pic_no_got
  60. {$IF defined(POWERPC) or defined(POWERPC64) or defined(SPARC) or defined(MIPS) or defined(SPARC64)}
  61. ,
  62. { since we have only 16bit offsets, we need to be able to specify the high
  63. and lower 16 bits of the address of a symbol of up to 64 bit }
  64. addr_low, // bits 48-63
  65. addr_high, // bits 32-47
  66. {$IF defined(POWERPC64)}
  67. addr_higher, // bits 16-31
  68. addr_highest, // bits 00-15
  69. {$ENDIF}
  70. addr_higha // bits 16-31, adjusted
  71. {$IF defined(POWERPC64)}
  72. ,
  73. addr_highera, // bits 32-47, adjusted
  74. addr_highesta // bits 48-63, adjusted
  75. {$ENDIF}
  76. {$ENDIF POWERPC or POWERPC64 or SPARC or MIPS or SPARC64}
  77. {$IFDEF MIPS}
  78. ,
  79. addr_pic_call16, // like addr_pic, but generates call16 reloc instead of got16
  80. addr_low_pic, // for large GOT model, generate got_hi16 and got_lo16 relocs
  81. addr_high_pic,
  82. addr_low_call, // counterpart of two above, generate call_hi16 and call_lo16 relocs
  83. addr_high_call
  84. {$ENDIF}
  85. {$if defined(RISCV32) or defined(RISCV64)}
  86. ,
  87. addr_hi20,
  88. addr_lo12,
  89. addr_pcrel_hi20,
  90. addr_pcrel_lo12,
  91. addr_pcrel
  92. {$endif RISCV}
  93. {$IFDEF AVR}
  94. ,addr_lo8
  95. ,addr_lo8_gs
  96. ,addr_hi8
  97. ,addr_hi8_gs
  98. {$ENDIF}
  99. {$IFDEF i8086}
  100. ,addr_dgroup // the data segment group
  101. ,addr_fardataseg // the far data segment of the current pascal module (unit or program)
  102. ,addr_seg // used for getting the segment of an object, e.g. 'mov ax, SEG symbol'
  103. {$ENDIF}
  104. {$IFDEF AARCH64}
  105. ,addr_page
  106. ,addr_pageoffset
  107. ,addr_gotpage
  108. ,addr_gotpageoffset
  109. {$ENDIF AARCH64}
  110. {$ifdef SPARC64}
  111. ,addr_gdop_hix22
  112. ,addr_gdop_lox22
  113. {$endif SPARC64}
  114. {$IFDEF ARM}
  115. ,addr_gottpoff
  116. ,addr_tpoff
  117. ,addr_tlsgd
  118. ,addr_tlsdesc
  119. ,addr_tlscall
  120. {$ENDIF}
  121. {$IFDEF i386}
  122. ,addr_ntpoff
  123. ,addr_tlsgd
  124. {$ENDIF}
  125. {$ifdef x86_64}
  126. ,addr_tpoff
  127. ,addr_tlsgd
  128. {$endif x86_64}
  129. );
  130. {# Generic opcodes, which must be supported by all processors
  131. }
  132. topcg =
  133. (
  134. OP_NONE,
  135. OP_MOVE, { replaced operation with direct load }
  136. OP_ADD, { simple addition }
  137. OP_AND, { simple logical and }
  138. OP_DIV, { simple unsigned division }
  139. OP_IDIV, { simple signed division }
  140. OP_IMUL, { simple signed multiply }
  141. OP_MUL, { simple unsigned multiply }
  142. OP_NEG, { simple negate }
  143. OP_NOT, { simple logical not }
  144. OP_OR, { simple logical or }
  145. OP_SAR, { arithmetic shift-right }
  146. OP_SHL, { logical shift left }
  147. OP_SHR, { logical shift right }
  148. OP_SUB, { simple subtraction }
  149. OP_XOR, { simple exclusive or }
  150. OP_ROL, { rotate left }
  151. OP_ROR { rotate right }
  152. );
  153. {# Generic flag values - used for jump locations }
  154. TOpCmp =
  155. (
  156. OC_NONE,
  157. OC_EQ, { equality comparison }
  158. OC_GT, { greater than (signed) }
  159. OC_LT, { less than (signed) }
  160. OC_GTE, { greater or equal than (signed) }
  161. OC_LTE, { less or equal than (signed) }
  162. OC_NE, { not equal }
  163. OC_BE, { less or equal than (unsigned) }
  164. OC_B, { less than (unsigned) }
  165. OC_AE, { greater or equal than (unsigned) }
  166. OC_A { greater than (unsigned) }
  167. );
  168. { indirect symbol flags }
  169. tindsymflag = (is_data,is_weak);
  170. tindsymflags = set of tindsymflag;
  171. { OS_NO is also used memory references with large data that can
  172. not be loaded in a register directly }
  173. TCgSize = (OS_NO,
  174. OS_8, OS_16, OS_32, OS_64, OS_128,
  175. OS_S8, OS_S16, OS_S32, OS_S64, OS_S128,
  176. { single, double, extended, comp, float128 }
  177. OS_F32, OS_F64, OS_F80, OS_C64, OS_F128,
  178. { multi-media sizes, describes only the register size but not how it is split,
  179. this information must be passed separately }
  180. OS_M8, OS_M16, OS_M32, OS_M64, OS_M128, OS_M256, OS_M512);
  181. { Register types }
  182. TRegisterType = (
  183. R_INVALIDREGISTER, { = 0 }
  184. R_INTREGISTER, { = 1 }
  185. R_FPUREGISTER, { = 2 }
  186. { used by Intel only }
  187. R_MMXREGISTER, { = 3 }
  188. R_MMREGISTER, { = 4 }
  189. R_SPECIALREGISTER, { = 5 }
  190. R_ADDRESSREGISTER, { = 6 }
  191. { used on llvm, every temp gets its own "base register" }
  192. R_TEMPREGISTER, { = 7 }
  193. { used on llvm for tracking metadata (every unique metadata has its own base register) }
  194. R_METADATAREGISTER { = 8 }
  195. );
  196. { Sub registers }
  197. TSubRegister = (
  198. R_SUBNONE, { = 0; no sub register possible }
  199. R_SUBL, { = 1; 8 bits, Like AL }
  200. R_SUBH, { = 2; 8 bits, Like AH }
  201. R_SUBW, { = 3; 16 bits, Like AX }
  202. R_SUBD, { = 4; 32 bits, Like EAX }
  203. R_SUBQ, { = 5; 64 bits, Like RAX }
  204. { For Sparc floats that use F0:F1 to store doubles }
  205. R_SUBFS, { = 6; Float that allocates 1 FPU register }
  206. R_SUBFD, { = 7; Float that allocates 2 FPU registers }
  207. R_SUBFQ, { = 8; Float that allocates 4 FPU registers }
  208. R_SUBMMS, { = 9; single scalar in multi media register }
  209. R_SUBMMD, { = 10; double scalar in multi media register }
  210. R_SUBMMWHOLE, { = 11; complete MM register, size depends on CPU }
  211. { For Intel X86 AVX-Register }
  212. R_SUBMMX, { = 12; 128 BITS }
  213. R_SUBMMY, { = 13; 256 BITS }
  214. R_SUBMMZ, { = 14; 512 BITS }
  215. { Subregisters for the flags register (x86) }
  216. R_SUBFLAGCARRY, { = 15; Carry flag }
  217. R_SUBFLAGPARITY, { = 16; Parity flag }
  218. R_SUBFLAGAUXILIARY, { = 17; Auxiliary flag }
  219. R_SUBFLAGZERO, { = 18; Zero flag }
  220. R_SUBFLAGSIGN, { = 19; Sign flag }
  221. R_SUBFLAGOVERFLOW, { = 20; Overflow flag }
  222. R_SUBFLAGINTERRUPT, { = 21; Interrupt enable flag }
  223. R_SUBFLAGDIRECTION, { = 22; Direction flag }
  224. R_SUBMM8B, { = 23; for part of v regs on aarch64 }
  225. R_SUBMM16B, { = 24; for part of v regs on aarch64 }
  226. { subregisters for the metadata register (llvm) }
  227. R_SUBMETASTRING { = 25 }
  228. );
  229. TSubRegisterSet = set of TSubRegister;
  230. TSuperRegister = type word;
  231. {
  232. The new register coding:
  233. SuperRegister (bits 0..15)
  234. Subregister (bits 16..23)
  235. Register type (bits 24..31)
  236. TRegister is defined as an enum to make it incompatible
  237. with TSuperRegister to avoid mixing them
  238. }
  239. TRegister = (
  240. TRegisterLowEnum := Low(longint),
  241. TRegisterHighEnum := High(longint)
  242. );
  243. TRegisterRec=packed record
  244. {$ifdef FPC_BIG_ENDIAN}
  245. regtype : Tregistertype;
  246. subreg : Tsubregister;
  247. supreg : Tsuperregister;
  248. {$else FPC_BIG_ENDIAN}
  249. supreg : Tsuperregister;
  250. subreg : Tsubregister;
  251. regtype : Tregistertype;
  252. {$endif FPC_BIG_ENDIAN}
  253. end;
  254. { A type to store register locations for 64 Bit values. }
  255. {$ifdef cpu64bitalu}
  256. tregister64 = tregister;
  257. tregister128 = record
  258. reglo,reghi : tregister;
  259. end;
  260. {$else cpu64bitalu}
  261. tregister64 = record
  262. reglo,reghi : tregister;
  263. end;
  264. {$endif cpu64bitalu}
  265. Tregistermmxset = record
  266. reg0,reg1,reg2,reg3:Tregister
  267. end;
  268. { Set type definition for registers }
  269. tsuperregisterset = array[byte] of set of byte;
  270. pmmshuffle = ^tmmshuffle;
  271. { this record describes shuffle operations for mm operations; if a pointer a shuffle record
  272. passed to an mm operation is nil, it means that the whole location is moved }
  273. tmmshuffle = record
  274. { describes how many shuffles are actually described, if len=0 then
  275. moving the scalar with index 0 to the scalar with index 0 is meant,
  276. if len=-1, then a variable/unknown length is assumed }
  277. len : Shortint;
  278. { lower byte of each entry of this array describes index of the source data index while
  279. the upper byte describes the destination index }
  280. shuffles : array[1..1] of word;
  281. end;
  282. Tsuperregisterarray=array[0..$ffff] of Tsuperregister;
  283. Psuperregisterarray=^Tsuperregisterarray;
  284. Tsuperregisterworklist=object
  285. buflength,
  286. buflengthinc,
  287. length:word;
  288. buf:Psuperregisterarray;
  289. constructor init;
  290. constructor copyfrom(const x:Tsuperregisterworklist);
  291. destructor done;
  292. procedure clear;
  293. procedure add(s:tsuperregister);
  294. function addnodup(s:tsuperregister): boolean;
  295. function get:tsuperregister;
  296. function readidx(i:word):tsuperregister;
  297. procedure deleteidx(i:word);
  298. function delete(s:tsuperregister):boolean;
  299. end;
  300. psuperregisterworklist=^tsuperregisterworklist;
  301. const
  302. { alias for easier understanding }
  303. R_SSEREGISTER = R_MMREGISTER;
  304. { Invalid register number }
  305. RS_INVALID = high(tsuperregister);
  306. NR_INVALID = tregister($ffffffff);
  307. tcgsize2size : Array[tcgsize] of integer =
  308. (0,
  309. { integer values }
  310. 1, 2, 4, 8, 16,
  311. 1, 2, 4, 8, 16,
  312. { floating point values }
  313. 4, 8, 10, 8, 16,
  314. { multimedia values }
  315. 1, 2, 4, 8, 16, 32, 64);
  316. tfloat2tcgsize: array[tfloattype] of tcgsize =
  317. (OS_F32,OS_F64,OS_F80,OS_F80,OS_C64,OS_C64,OS_F128);
  318. tcgsize2tfloat: array[OS_F32..OS_C64] of tfloattype =
  319. (s32real,s64real,s80real,s64comp);
  320. tvarregable2tcgloc : array[tvarregable] of tcgloc = (LOC_VOID,
  321. LOC_CREGISTER,LOC_CFPUREGISTER,LOC_CMMREGISTER,LOC_CREGISTER);
  322. {$if defined(cpu64bitalu)}
  323. { operand size describing an unsigned value in a pair of int registers }
  324. OS_PAIR = OS_128;
  325. { operand size describing an signed value in a pair of int registers }
  326. OS_SPAIR = OS_S128;
  327. {$elseif defined(cpu32bitalu)}
  328. { operand size describing an unsigned value in a pair of int registers }
  329. OS_PAIR = OS_64;
  330. { operand size describing an signed value in a pair of int registers }
  331. OS_SPAIR = OS_S64;
  332. {$elseif defined(cpu16bitalu)}
  333. { operand size describing an unsigned value in a pair of int registers }
  334. OS_PAIR = OS_32;
  335. { operand size describing an signed value in a pair of int registers }
  336. OS_SPAIR = OS_S32;
  337. {$elseif defined(cpu8bitalu)}
  338. { operand size describing an unsigned value in a pair of int registers }
  339. OS_PAIR = OS_16;
  340. { operand size describing an signed value in a pair of int registers }
  341. OS_SPAIR = OS_S16;
  342. {$endif}
  343. { Table to convert tcgsize variables to the correspondending
  344. unsigned types }
  345. tcgsize2unsigned : array[tcgsize] of tcgsize = (OS_NO,
  346. OS_8, OS_16, OS_32, OS_64, OS_128,
  347. OS_8, OS_16, OS_32, OS_64, OS_128,
  348. OS_F32, OS_F64, OS_F80, OS_C64, OS_F128,
  349. OS_M8, OS_M16, OS_M32, OS_M64, OS_M128, OS_M256, OS_M512);
  350. tcgsize2signed : array[tcgsize] of tcgsize = (OS_NO,
  351. OS_S8, OS_S16, OS_S32, OS_S64, OS_S128,
  352. OS_S8, OS_S16, OS_S32, OS_S64, OS_S128,
  353. OS_F32, OS_F64, OS_F80, OS_C64, OS_F128,
  354. OS_M8, OS_M16, OS_M32, OS_M64, OS_M128, OS_M256,OS_M512);
  355. tcgloc2str : array[TCGLoc] of string[12] = (
  356. 'LOC_INVALID',
  357. 'LOC_VOID',
  358. 'LOC_CONST',
  359. 'LOC_JUMP',
  360. 'LOC_FLAGS',
  361. 'LOC_REG',
  362. 'LOC_CREG',
  363. 'LOC_FPUREG',
  364. 'LOC_CFPUREG',
  365. 'LOC_MMXREG',
  366. 'LOC_CMMXREG',
  367. 'LOC_MMREG',
  368. 'LOC_CMMREG',
  369. 'LOC_SSETREG',
  370. 'LOC_CSSETREG',
  371. 'LOC_SSETREF',
  372. 'LOC_CSSETREF',
  373. 'LOC_CREF',
  374. 'LOC_REF'
  375. );
  376. var
  377. mms_movescalar,
  378. mms_variable,
  379. mms_2,
  380. mms_4,
  381. mms_8,
  382. mms_16,
  383. mms_32 : pmmshuffle;
  384. procedure supregset_reset(var regs:tsuperregisterset;setall:boolean;
  385. maxreg:Tsuperregister);{$ifdef USEINLINE}inline;{$endif}
  386. procedure supregset_include(var regs:tsuperregisterset;s:tsuperregister);{$ifdef USEINLINE}inline;{$endif}
  387. procedure supregset_exclude(var regs:tsuperregisterset;s:tsuperregister);{$ifdef USEINLINE}inline;{$endif}
  388. function supregset_in(const regs:tsuperregisterset;s:tsuperregister):boolean;{$ifdef USEINLINE}inline;{$endif}
  389. function newreg(rt:tregistertype;sr:tsuperregister;sb:tsubregister):tregister;{$ifdef USEINLINE}inline;{$endif}
  390. function getsubreg(r:tregister):tsubregister;{$ifdef USEINLINE}inline;{$endif}
  391. function getsupreg(r:tregister):tsuperregister;{$ifdef USEINLINE}inline;{$endif}
  392. function getregtype(r:tregister):tregistertype;{$ifdef USEINLINE}inline;{$endif}
  393. procedure setsubreg(var r:tregister;sr:tsubregister);{$ifdef USEINLINE}inline;{$endif}
  394. procedure setsupreg(var r:tregister;sr:tsuperregister);{$ifdef USEINLINE}inline;{$endif}
  395. function generic_regname(r:tregister):string;
  396. {# From a constant numeric value, return the abstract code generator
  397. size.
  398. }
  399. function int_cgsize(const a: tcgint): tcgsize;{$ifdef USEINLINE}inline;{$endif}
  400. function int_float_cgsize(const a: tcgint): tcgsize;
  401. function float_array_cgsize(const a: tcgint): tcgsize;{$ifdef USEINLINE}inline;{$endif}
  402. function double_array_cgsize(const a: tcgint): tcgsize;{$ifdef USEINLINE}inline;{$endif}
  403. function tcgsize2str(cgsize: tcgsize):string;
  404. { return the inverse condition of opcmp }
  405. function inverse_opcmp(opcmp: topcmp): topcmp;{$ifdef USEINLINE}inline;{$endif}
  406. { return the opcmp needed when swapping the operands }
  407. function swap_opcmp(opcmp: topcmp): topcmp;{$ifdef USEINLINE}inline;{$endif}
  408. { return whether op is commutative }
  409. function commutativeop(op: topcg): boolean;{$ifdef USEINLINE}inline;{$endif}
  410. { returns true, if shuffle describes a real shuffle operation and not only a move }
  411. function realshuffle(shuffle : pmmshuffle) : boolean;
  412. { returns true, if the shuffle describes only a move of the scalar at index 0 }
  413. function shufflescalar(shuffle : pmmshuffle) : boolean;
  414. { removes shuffling from shuffle, this means that the destenation index of each shuffle is copied to
  415. the source }
  416. procedure removeshuffles(var shuffle : tmmshuffle);
  417. implementation
  418. uses
  419. verbose,
  420. cutils;
  421. {******************************************************************************
  422. tsuperregisterworklist
  423. ******************************************************************************}
  424. constructor tsuperregisterworklist.init;
  425. begin
  426. length:=0;
  427. buflength:=0;
  428. buflengthinc:=16;
  429. buf:=nil;
  430. end;
  431. constructor Tsuperregisterworklist.copyfrom(const x:Tsuperregisterworklist);
  432. begin
  433. self:=x;
  434. if x.buf<>nil then
  435. begin
  436. getmem(buf,buflength*sizeof(Tsuperregister));
  437. move(x.buf^,buf^,length*sizeof(Tsuperregister));
  438. end;
  439. end;
  440. destructor tsuperregisterworklist.done;
  441. begin
  442. if assigned(buf) then
  443. freemem(buf);
  444. end;
  445. procedure tsuperregisterworklist.add(s:tsuperregister);
  446. begin
  447. inc(length);
  448. { Need to increase buffer length? }
  449. if length>=buflength then
  450. begin
  451. inc(buflength,buflengthinc);
  452. buflengthinc:=buflengthinc*2;
  453. if buflengthinc>256 then
  454. buflengthinc:=256;
  455. reallocmem(buf,buflength*sizeof(Tsuperregister));
  456. end;
  457. buf^[length-1]:=s;
  458. end;
  459. function tsuperregisterworklist.addnodup(s:tsuperregister): boolean;
  460. begin
  461. addnodup := false;
  462. if indexword(buf^,length,s) = -1 then
  463. begin
  464. add(s);
  465. addnodup := true;
  466. end;
  467. end;
  468. procedure tsuperregisterworklist.clear;
  469. begin
  470. length:=0;
  471. end;
  472. procedure tsuperregisterworklist.deleteidx(i:word);
  473. begin
  474. if i>=length then
  475. internalerror(200310144);
  476. buf^[i]:=buf^[length-1];
  477. dec(length);
  478. end;
  479. function tsuperregisterworklist.readidx(i:word):tsuperregister;
  480. begin
  481. if (i >= length) then
  482. internalerror(2005010601);
  483. result := buf^[i];
  484. end;
  485. function tsuperregisterworklist.get:tsuperregister;
  486. begin
  487. if length=0 then
  488. internalerror(200310142);
  489. get:=buf^[0];
  490. buf^[0]:=buf^[length-1];
  491. dec(length);
  492. end;
  493. function tsuperregisterworklist.delete(s:tsuperregister):boolean;
  494. var
  495. i:longint;
  496. begin
  497. delete:=false;
  498. { indexword in 1.0.x and 1.9.4 is broken }
  499. i:=indexword(buf^,length,s);
  500. if i<>-1 then
  501. begin
  502. deleteidx(i);
  503. delete := true;
  504. end;
  505. end;
  506. procedure supregset_reset(var regs:tsuperregisterset;setall:boolean;
  507. maxreg:Tsuperregister);{$ifdef USEINLINE}inline;{$endif}
  508. begin
  509. fillchar(regs,(maxreg+7) shr 3,-byte(setall));
  510. end;
  511. procedure supregset_include(var regs:tsuperregisterset;s:tsuperregister);{$ifdef USEINLINE}inline;{$endif}
  512. begin
  513. include(regs[s shr 8],(s and $ff));
  514. end;
  515. procedure supregset_exclude(var regs:tsuperregisterset;s:tsuperregister);{$ifdef USEINLINE}inline;{$endif}
  516. begin
  517. exclude(regs[s shr 8],(s and $ff));
  518. end;
  519. function supregset_in(const regs:tsuperregisterset;s:tsuperregister):boolean;{$ifdef USEINLINE}inline;{$endif}
  520. begin
  521. result:=(s and $ff) in regs[s shr 8];
  522. end;
  523. function newreg(rt:tregistertype;sr:tsuperregister;sb:tsubregister):tregister;{$ifdef USEINLINE}inline;{$endif}
  524. begin
  525. tregisterrec(result).regtype:=rt;
  526. tregisterrec(result).supreg:=sr;
  527. tregisterrec(result).subreg:=sb;
  528. end;
  529. function getsubreg(r:tregister):tsubregister;{$ifdef USEINLINE}inline;{$endif}
  530. begin
  531. result:=tregisterrec(r).subreg;
  532. end;
  533. function getsupreg(r:tregister):tsuperregister;{$ifdef USEINLINE}inline;{$endif}
  534. begin
  535. result:=tregisterrec(r).supreg;
  536. end;
  537. function getregtype(r:tregister):tregistertype;{$ifdef USEINLINE}inline;{$endif}
  538. begin
  539. result:=tregisterrec(r).regtype;
  540. end;
  541. procedure setsubreg(var r:tregister;sr:tsubregister);{$ifdef USEINLINE}inline;{$endif}
  542. begin
  543. tregisterrec(r).subreg:=sr;
  544. end;
  545. procedure setsupreg(var r:tregister;sr:tsuperregister);{$ifdef USEINLINE}inline;{$endif}
  546. begin
  547. tregisterrec(r).supreg:=sr;
  548. end;
  549. function generic_regname(r:tregister):string;
  550. var
  551. nr : string[12];
  552. begin
  553. str(getsupreg(r),nr);
  554. case getregtype(r) of
  555. R_INTREGISTER:
  556. result:='ireg'+nr;
  557. R_FPUREGISTER:
  558. result:='freg'+nr;
  559. R_MMREGISTER:
  560. result:='mreg'+nr;
  561. R_MMXREGISTER:
  562. result:='xreg'+nr;
  563. R_ADDRESSREGISTER:
  564. result:='areg'+nr;
  565. R_SPECIALREGISTER:
  566. result:='sreg'+nr;
  567. else
  568. begin
  569. result:='INVALID';
  570. exit;
  571. end;
  572. end;
  573. case getsubreg(r) of
  574. R_SUBNONE:
  575. ;
  576. R_SUBL:
  577. result:=result+'l';
  578. R_SUBH:
  579. result:=result+'h';
  580. R_SUBW:
  581. result:=result+'w';
  582. R_SUBD:
  583. result:=result+'d';
  584. R_SUBQ:
  585. result:=result+'q';
  586. R_SUBFS:
  587. result:=result+'fs';
  588. R_SUBFD:
  589. result:=result+'fd';
  590. R_SUBMMD:
  591. result:=result+'md';
  592. R_SUBMMS:
  593. result:=result+'ms';
  594. R_SUBMMWHOLE:
  595. result:=result+'ma';
  596. R_SUBMMX:
  597. result:=result+'mx';
  598. R_SUBMMY:
  599. result:=result+'my';
  600. R_SUBMMZ:
  601. result:=result+'mz';
  602. R_SUBMM8B:
  603. result:=result+'m8b';
  604. else
  605. internalerror(200308252);
  606. end;
  607. end;
  608. function int_cgsize(const a: tcgint): tcgsize;{$ifdef USEINLINE}inline;{$endif}
  609. const
  610. size2cgsize : array[0..8] of tcgsize = (
  611. OS_NO,OS_8,OS_16,OS_NO,OS_32,OS_NO,OS_NO,OS_NO,OS_64
  612. );
  613. begin
  614. {$ifdef cpu64bitalu}
  615. if a=16 then
  616. result:=OS_128
  617. else
  618. {$endif cpu64bitalu}
  619. if a>8 then
  620. result:=OS_NO
  621. else
  622. result:=size2cgsize[a];
  623. end;
  624. function int_float_cgsize(const a: tcgint): tcgsize;
  625. begin
  626. case a of
  627. 4 :
  628. result:=OS_F32;
  629. 8 :
  630. result:=OS_F64;
  631. 10 :
  632. result:=OS_F80;
  633. 16 :
  634. result:=OS_F128;
  635. else
  636. internalerror(200603211);
  637. end;
  638. end;
  639. function float_array_cgsize(const a: tcgint): tcgsize;{$ifdef USEINLINE}inline;{$endif}
  640. begin
  641. case a of
  642. 4:
  643. result := OS_M32;
  644. 16:
  645. result := OS_M128;
  646. 32:
  647. result := OS_M256;
  648. 64:
  649. result := OS_M512;
  650. else
  651. result := int_cgsize(a);
  652. end;
  653. end;
  654. function double_array_cgsize(const a: tcgint): tcgsize;{$ifdef USEINLINE}inline;{$endif}
  655. begin
  656. case a of
  657. 8:
  658. result := OS_M64;
  659. 16:
  660. result := OS_M128;
  661. 32:
  662. result := OS_M256;
  663. 64:
  664. result := OS_M512;
  665. else
  666. result := int_cgsize(a);
  667. end;
  668. end;
  669. function tcgsize2str(cgsize: tcgsize):string;
  670. begin
  671. Str(cgsize, Result);
  672. end;
  673. function inverse_opcmp(opcmp: topcmp): topcmp;{$ifdef USEINLINE}inline;{$endif}
  674. const
  675. list: array[TOpCmp] of TOpCmp =
  676. (OC_NONE,OC_NE,OC_LTE,OC_GTE,OC_LT,OC_GT,OC_EQ,OC_A,OC_AE,
  677. OC_B,OC_BE);
  678. begin
  679. inverse_opcmp := list[opcmp];
  680. end;
  681. function swap_opcmp(opcmp: topcmp): topcmp;{$ifdef USEINLINE}inline;{$endif}
  682. const
  683. list: array[TOpCmp] of TOpCmp =
  684. (OC_NONE,OC_EQ,OC_LT,OC_GT,OC_LTE,OC_GTE,OC_NE,OC_AE,OC_A,
  685. OC_BE,OC_B);
  686. begin
  687. swap_opcmp := list[opcmp];
  688. end;
  689. function commutativeop(op: topcg): boolean;{$ifdef USEINLINE}inline;{$endif}
  690. const
  691. list: array[topcg] of boolean =
  692. (true,false,true,true,false,false,true,true,false,false,
  693. true,false,false,false,false,true,false,false);
  694. begin
  695. commutativeop := list[op];
  696. end;
  697. function realshuffle(shuffle : pmmshuffle) : boolean;
  698. var
  699. i : longint;
  700. begin
  701. realshuffle:=true;
  702. if (shuffle=nil) or (shuffle^.len<1) then
  703. realshuffle:=false
  704. else
  705. begin
  706. for i:=1 to shuffle^.len do
  707. begin
  708. if (shuffle^.shuffles[i] and $ff)<>((shuffle^.shuffles[i] and $ff00) shr 8) then
  709. exit;
  710. end;
  711. realshuffle:=false;
  712. end;
  713. end;
  714. function shufflescalar(shuffle : pmmshuffle) : boolean;
  715. begin
  716. result:=shuffle^.len=0;
  717. end;
  718. procedure removeshuffles(var shuffle : tmmshuffle);
  719. var
  720. i : longint;
  721. begin
  722. if shuffle.len=0 then
  723. exit;
  724. for i:=1 to shuffle.len do
  725. shuffle.shuffles[i]:=(shuffle.shuffles[i] and $f) or ((shuffle.shuffles[i] and $f0) shr 4);
  726. end;
  727. procedure Initmms(var p : pmmshuffle;len : ShortInt);
  728. var
  729. i : Integer;
  730. begin
  731. Getmem(p,sizeof(tmmshuffle)+(max(len,0)-1)*2);
  732. p^.len:=len;
  733. for i:=1 to len do
  734. {$push}
  735. {$R-}
  736. p^.shuffles[i]:=i;
  737. {$pop}
  738. end;
  739. initialization
  740. Initmms(mms_movescalar,0);
  741. Initmms(mms_variable,-1);
  742. Initmms(mms_2,2);
  743. Initmms(mms_4,4);
  744. Initmms(mms_8,8);
  745. Initmms(mms_16,16);
  746. Initmms(mms_32,32);
  747. finalization
  748. Freemem(mms_movescalar);
  749. Freemem(mms_variable);
  750. Freemem(mms_2);
  751. Freemem(mms_4);
  752. Freemem(mms_8);
  753. Freemem(mms_16);
  754. Freemem(mms_32);
  755. end.