cgbase.pas 30 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949
  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. addr_got_pcrel_hi,
  93. addr_plt
  94. {$endif RISCV}
  95. {$IFDEF AVR}
  96. ,addr_lo8
  97. ,addr_lo8_gs
  98. ,addr_hi8
  99. ,addr_hi8_gs
  100. {$ENDIF}
  101. {$IFDEF Z80}
  102. ,addr_lo8
  103. ,addr_hi8
  104. {$ENDIF}
  105. {$IFDEF i8086}
  106. ,addr_dgroup // the data segment group
  107. ,addr_fardataseg // the far data segment of the current pascal module (unit or program)
  108. ,addr_seg // used for getting the segment of an object, e.g. 'mov ax, SEG symbol'
  109. {$ENDIF}
  110. {$IFDEF AARCH64}
  111. ,addr_page
  112. ,addr_pageoffset
  113. ,addr_gotpage
  114. ,addr_gotpageoffset
  115. {$ENDIF AARCH64}
  116. {$ifdef SPARC64}
  117. ,addr_gdop_hix22
  118. ,addr_gdop_lox22
  119. {$endif SPARC64}
  120. {$IFDEF ARM}
  121. ,addr_gottpoff
  122. ,addr_tpoff
  123. ,addr_tlsgd
  124. ,addr_tlsdesc
  125. ,addr_tlscall
  126. {$ENDIF}
  127. {$IFDEF i386}
  128. ,addr_ntpoff
  129. ,addr_tlsgd
  130. {$ENDIF}
  131. {$ifdef x86_64}
  132. ,addr_tpoff
  133. ,addr_tlsgd
  134. {$endif x86_64}
  135. );
  136. {# Generic opcodes, which must be supported by all processors
  137. }
  138. topcg =
  139. (
  140. OP_NONE,
  141. OP_MOVE, { replaced operation with direct load }
  142. OP_ADD, { simple addition }
  143. OP_AND, { simple logical and }
  144. OP_DIV, { simple unsigned division }
  145. OP_IDIV, { simple signed division }
  146. OP_IMUL, { simple signed multiply }
  147. OP_MUL, { simple unsigned multiply }
  148. OP_NEG, { simple negate }
  149. OP_NOT, { simple logical not }
  150. OP_OR, { simple logical or }
  151. OP_SAR, { arithmetic shift-right }
  152. OP_SHL, { logical shift left }
  153. OP_SHR, { logical shift right }
  154. OP_SUB, { simple subtraction }
  155. OP_XOR, { simple exclusive or }
  156. OP_ROL, { rotate left }
  157. OP_ROR { rotate right }
  158. );
  159. {# Generic flag values - used for jump locations }
  160. TOpCmp =
  161. (
  162. OC_NONE,
  163. OC_EQ, { equality comparison }
  164. OC_GT, { greater than (signed) }
  165. OC_LT, { less than (signed) }
  166. OC_GTE, { greater or equal than (signed) }
  167. OC_LTE, { less or equal than (signed) }
  168. OC_NE, { not equal }
  169. OC_BE, { less or equal than (unsigned) }
  170. OC_B, { less than (unsigned) }
  171. OC_AE, { greater or equal than (unsigned) }
  172. OC_A { greater than (unsigned) }
  173. );
  174. { indirect symbol flags }
  175. tindsymflag = (is_data,is_weak);
  176. tindsymflags = set of tindsymflag;
  177. { OS_NO is also used memory references with large data that can
  178. not be loaded in a register directly }
  179. TCgSize = (OS_NO,
  180. OS_8, OS_16, OS_32, OS_64, OS_128,
  181. OS_S8, OS_S16, OS_S32, OS_S64, OS_S128,
  182. { single, double, extended, comp, float128 }
  183. OS_F32, OS_F64, OS_F80, OS_C64, OS_F128,
  184. { multi-media sizes, describes only the register size but not how it is split,
  185. this information must be passed separately }
  186. OS_M8, OS_M16, OS_M32, OS_M64, OS_M128, OS_M256, OS_M512);
  187. { Register types }
  188. TRegisterType = (
  189. R_INVALIDREGISTER, { = 0 }
  190. R_INTREGISTER, { = 1 }
  191. R_FPUREGISTER, { = 2 }
  192. { used by Intel only }
  193. R_MMXREGISTER, { = 3 }
  194. R_MMREGISTER, { = 4 }
  195. R_SPECIALREGISTER, { = 5 }
  196. R_ADDRESSREGISTER, { = 6 }
  197. { used on llvm, every temp gets its own "base register" }
  198. R_TEMPREGISTER, { = 7 }
  199. { used on llvm for tracking metadata (every unique metadata has its own base register) }
  200. R_METADATAREGISTER,{ = 8 }
  201. { optional MAC16 (16 bit multiply-accumulate) registers on Xtensa }
  202. R_MAC16REGISTER { = 9 }
  203. { do not add more than 16 elements (ifdef by cpu type if needed)
  204. so we can store this in one nibble and pack TRegister
  205. if the supreg width should be extended }
  206. );
  207. { Sub registers }
  208. TSubRegister = (
  209. R_SUBNONE, { = 0; no sub register possible }
  210. R_SUBL, { = 1; 8 bits, Like AL }
  211. R_SUBH, { = 2; 8 bits, Like AH }
  212. R_SUBW, { = 3; 16 bits, Like AX }
  213. R_SUBD, { = 4; 32 bits, Like EAX }
  214. R_SUBQ, { = 5; 64 bits, Like RAX }
  215. { For Sparc floats that use F0:F1 to store doubles }
  216. R_SUBFS, { = 6; Float that allocates 1 FPU register }
  217. R_SUBFD, { = 7; Float that allocates 2 FPU registers }
  218. R_SUBFQ, { = 8; Float that allocates 4 FPU registers }
  219. R_SUBMMS, { = 9; single scalar in multi media register }
  220. R_SUBMMD, { = 10; double scalar in multi media register }
  221. R_SUBMMWHOLE, { = 11; complete MM register, size depends on CPU }
  222. { For Intel X86 AVX-Register }
  223. R_SUBMMX, { = 12; 128 BITS }
  224. R_SUBMMY, { = 13; 256 BITS }
  225. R_SUBMMZ, { = 14; 512 BITS }
  226. {$ifdef Z80}
  227. { Subregisters for the flags register (Z80) }
  228. R_SUBFLAGCARRY, { = 15; Carry flag }
  229. R_SUBFLAGADDSUBTRACT, { = 16; Add/Subtract flag }
  230. R_SUBFLAGPARITYOVERFLOW, { = 17; Parity/Overflow flag }
  231. R_SUBFLAGUNUSEDBIT3, { = 18; Unused flag (bit 3) }
  232. R_SUBFLAGHALFCARRY, { = 19; Half Carry flag }
  233. R_SUBFLAGUNUSEDBIT5, { = 20; Unused flag (bit 5) }
  234. R_SUBFLAGZERO, { = 21; Zero flag }
  235. R_SUBFLAGSIGN, { = 22; Sign flag }
  236. {$else Z80}
  237. { Subregisters for the flags register (x86) }
  238. R_SUBFLAGCARRY, { = 15; Carry flag }
  239. R_SUBFLAGPARITY, { = 16; Parity flag }
  240. R_SUBFLAGAUXILIARY, { = 17; Auxiliary flag }
  241. R_SUBFLAGZERO, { = 18; Zero flag }
  242. R_SUBFLAGSIGN, { = 19; Sign flag }
  243. R_SUBFLAGOVERFLOW, { = 20; Overflow flag }
  244. R_SUBFLAGINTERRUPT, { = 21; Interrupt enable flag }
  245. R_SUBFLAGDIRECTION, { = 22; Direction flag }
  246. {$endif Z80}
  247. { subregisters for the metadata register (llvm) }
  248. R_SUBMETASTRING { = 23 }
  249. {$ifdef aarch64}
  250. , R_SUBMM8B { = 24; for arrangement of v regs on aarch64 }
  251. , R_SUBMM16B { = 25; for arrangement of v regs on aarch64 }
  252. , R_SUBMM4H { = 26; for arrangement of v regs on aarch64 }
  253. , R_SUBMM8H { = 27; for arrangement of v regs on aarch64 }
  254. , R_SUBMM2S { = 28; for arrangement of v regs on aarch64 }
  255. , R_SUBMM4S { = 29; for arrangement of v regs on aarch64 }
  256. , R_SUBMM1D { = 30; for arrangement of v regs on aarch64 }
  257. , R_SUBMM2D { = 31; for arrangement of v regs on aarch64 }
  258. , R_SUBMMB1 { = 32; for arrangement of v regs on aarch64; for use with ldN/stN }
  259. , R_SUBMMH1 { = 33; for arrangement of v regs on aarch64; for use with ldN/stN }
  260. , R_SUBMMS1 { = 34; for arrangement of v regs on aarch64; for use with ldN/stN }
  261. , R_SUBMMD1 { = 35; for arrangement of v regs on aarch64; for use with ldN/stN }
  262. {$endif aarch64}
  263. );
  264. TSubRegisterSet = set of TSubRegister;
  265. TSuperRegister = type word;
  266. {
  267. The new register coding:
  268. SuperRegister (bits 0..15)
  269. Subregister (bits 16..23)
  270. Register type (bits 24..31)
  271. TRegister is defined as an enum to make it incompatible
  272. with TSuperRegister to avoid mixing them
  273. }
  274. TRegister = (
  275. TRegisterLowEnum := Low(longint),
  276. TRegisterHighEnum := High(longint)
  277. );
  278. TRegisterRec=packed record
  279. {$ifdef FPC_BIG_ENDIAN}
  280. regtype : Tregistertype;
  281. subreg : Tsubregister;
  282. supreg : Tsuperregister;
  283. {$else FPC_BIG_ENDIAN}
  284. supreg : Tsuperregister;
  285. subreg : Tsubregister;
  286. regtype : Tregistertype;
  287. {$endif FPC_BIG_ENDIAN}
  288. end;
  289. { A type to store register locations for 64 Bit values. }
  290. {$ifdef cpu64bitalu}
  291. tregister64 = tregister;
  292. tregister128 = record
  293. reglo,reghi : tregister;
  294. end;
  295. {$else cpu64bitalu}
  296. tregister64 = record
  297. reglo,reghi : tregister;
  298. end;
  299. {$endif cpu64bitalu}
  300. { Set type definition for registers }
  301. tsuperregisterset = array[byte] of set of byte;
  302. pmmshuffle = ^tmmshuffle;
  303. { this record describes shuffle operations for mm operations; if a pointer a shuffle record
  304. passed to an mm operation is nil, it means that the whole location is moved }
  305. tmmshuffle = record
  306. { describes how many shuffles are actually described, if len=0 then
  307. moving the scalar with index 0 to the scalar with index 0 is meant,
  308. if len=-1, then a variable/unknown length is assumed }
  309. len : Shortint;
  310. { lower byte of each entry of this array describes index of the source data index while
  311. the upper byte describes the destination index }
  312. shuffles : array[1..1] of word;
  313. end;
  314. Tsuperregisterarray=array[0..$ffff] of Tsuperregister;
  315. Psuperregisterarray=^Tsuperregisterarray;
  316. Tsuperregisterworklist=object
  317. buflength,
  318. buflengthinc,
  319. length:word;
  320. buf:Psuperregisterarray;
  321. constructor init;
  322. constructor copyfrom(const x:Tsuperregisterworklist);
  323. destructor done;
  324. procedure clear;
  325. procedure add(s:tsuperregister);
  326. function addnodup(s:tsuperregister): boolean;
  327. { returns the last element and removes it from the list }
  328. function get:tsuperregister;
  329. function readidx(i:word):tsuperregister;
  330. procedure deleteidx(i:word);
  331. function delete(s:tsuperregister):boolean;
  332. end;
  333. psuperregisterworklist=^tsuperregisterworklist;
  334. const
  335. { alias for easier understanding }
  336. R_SSEREGISTER = R_MMREGISTER;
  337. { Invalid register number }
  338. RS_INVALID = high(tsuperregister);
  339. NR_INVALID = tregister($ffffffff);
  340. tcgsize2size : Array[tcgsize] of integer =
  341. (0,
  342. { integer values }
  343. 1, 2, 4, 8, 16,
  344. 1, 2, 4, 8, 16,
  345. { floating point values }
  346. 4, 8, 10, 8, 16,
  347. { multimedia values }
  348. 1, 2, 4, 8, 16, 32, 64);
  349. tfloat2tcgsize: array[tfloattype] of tcgsize =
  350. (OS_F32,OS_F64,OS_F80,OS_F80,OS_C64,OS_C64,OS_F128);
  351. tcgsize2tfloat: array[OS_F32..OS_C64] of tfloattype =
  352. (s32real,s64real,s80real,s64comp);
  353. tvarregable2tcgloc : array[tvarregable] of tcgloc = (LOC_VOID,
  354. LOC_CREGISTER,LOC_CFPUREGISTER,LOC_CMMREGISTER,LOC_CREGISTER);
  355. {$if defined(cpu64bitalu)}
  356. { operand size describing an unsigned value in a pair of int registers }
  357. OS_PAIR = OS_128;
  358. { operand size describing an signed value in a pair of int registers }
  359. OS_SPAIR = OS_S128;
  360. {$elseif defined(cpu32bitalu)}
  361. { operand size describing an unsigned value in a pair of int registers }
  362. OS_PAIR = OS_64;
  363. { operand size describing an signed value in a pair of int registers }
  364. OS_SPAIR = OS_S64;
  365. {$elseif defined(cpu16bitalu)}
  366. { operand size describing an unsigned value in a pair of int registers }
  367. OS_PAIR = OS_32;
  368. { operand size describing an signed value in a pair of int registers }
  369. OS_SPAIR = OS_S32;
  370. {$elseif defined(cpu8bitalu)}
  371. { operand size describing an unsigned value in a pair of int registers }
  372. OS_PAIR = OS_16;
  373. { operand size describing an signed value in a pair of int registers }
  374. OS_SPAIR = OS_S16;
  375. {$endif}
  376. { Table to convert tcgsize variables to the correspondending
  377. unsigned types }
  378. tcgsize2unsigned : array[tcgsize] of tcgsize = (OS_NO,
  379. OS_8, OS_16, OS_32, OS_64, OS_128,
  380. OS_8, OS_16, OS_32, OS_64, OS_128,
  381. OS_F32, OS_F64, OS_F80, OS_C64, OS_F128,
  382. OS_M8, OS_M16, OS_M32, OS_M64, OS_M128, OS_M256, OS_M512);
  383. tcgsize2signed : array[tcgsize] of tcgsize = (OS_NO,
  384. OS_S8, OS_S16, OS_S32, OS_S64, OS_S128,
  385. OS_S8, OS_S16, OS_S32, OS_S64, OS_S128,
  386. OS_F32, OS_F64, OS_F80, OS_C64, OS_F128,
  387. OS_M8, OS_M16, OS_M32, OS_M64, OS_M128, OS_M256,OS_M512);
  388. tcgloc2str : array[TCGLoc] of string[12] = (
  389. 'LOC_INVALID',
  390. 'LOC_VOID',
  391. 'LOC_CONST',
  392. 'LOC_JUMP',
  393. 'LOC_FLAGS',
  394. 'LOC_REG',
  395. 'LOC_CREG',
  396. 'LOC_FPUREG',
  397. 'LOC_CFPUREG',
  398. 'LOC_MMXREG',
  399. 'LOC_CMMXREG',
  400. 'LOC_MMREG',
  401. 'LOC_CMMREG',
  402. 'LOC_SSETREG',
  403. 'LOC_CSSETREG',
  404. 'LOC_SSETREF',
  405. 'LOC_CSSETREF',
  406. 'LOC_CREF',
  407. 'LOC_REF'
  408. );
  409. var
  410. mms_movescalar,
  411. mms_variable,
  412. mms_2,
  413. mms_4,
  414. mms_8,
  415. mms_16,
  416. mms_32 : pmmshuffle;
  417. procedure supregset_reset(var regs:tsuperregisterset;setall:boolean;
  418. maxreg:Tsuperregister);{$ifdef USEINLINE}inline;{$endif}
  419. procedure supregset_include(var regs:tsuperregisterset;s:tsuperregister);{$ifdef USEINLINE}inline;{$endif}
  420. procedure supregset_exclude(var regs:tsuperregisterset;s:tsuperregister);{$ifdef USEINLINE}inline;{$endif}
  421. function supregset_in(const regs:tsuperregisterset;s:tsuperregister):boolean;{$ifdef USEINLINE}inline;{$endif}
  422. function newreg(rt:tregistertype;sr:tsuperregister;sb:tsubregister):tregister;{$ifdef USEINLINE}inline;{$endif}
  423. function getsubreg(r:tregister):tsubregister;{$ifdef USEINLINE}inline;{$endif}
  424. function getsupreg(r:tregister):tsuperregister;{$ifdef USEINLINE}inline;{$endif}
  425. function getregtype(r:tregister):tregistertype;{$ifdef USEINLINE}inline;{$endif}
  426. procedure setsubreg(var r:tregister;sr:tsubregister);{$ifdef USEINLINE}inline;{$endif}
  427. procedure setsupreg(var r:tregister;sr:tsuperregister);{$ifdef USEINLINE}inline;{$endif}
  428. function generic_regname(r:tregister):string;
  429. {# From a constant numeric value, return the abstract code generator
  430. size.
  431. }
  432. function int_cgsize(const a: tcgint): tcgsize;{$ifdef USEINLINE}inline;{$endif}
  433. function int_float_cgsize(const a: tcgint): tcgsize;
  434. function float_array_cgsize(const a: tcgint): tcgsize;{$ifdef USEINLINE}inline;{$endif}
  435. function double_array_cgsize(const a: tcgint): tcgsize;{$ifdef USEINLINE}inline;{$endif}
  436. function tcgsize2str(cgsize: tcgsize):string;
  437. { return the inverse condition of opcmp }
  438. function inverse_opcmp(opcmp: topcmp): topcmp;{$ifdef USEINLINE}inline;{$endif}
  439. { return the opcmp needed when swapping the operands }
  440. function swap_opcmp(opcmp: topcmp): topcmp;{$ifdef USEINLINE}inline;{$endif}
  441. { return whether op is commutative }
  442. function commutativeop(op: topcg): boolean;{$ifdef USEINLINE}inline;{$endif}
  443. { returns true, if shuffle describes a real shuffle operation and not only a move }
  444. function realshuffle(shuffle : pmmshuffle) : boolean;
  445. { returns true, if the shuffle describes only a move of the scalar at index 0 }
  446. function shufflescalar(shuffle : pmmshuffle) : boolean;
  447. { removes shuffling from shuffle, this means that the destenation index of each shuffle is copied to
  448. the source }
  449. procedure removeshuffles(var shuffle : tmmshuffle);
  450. function is_float_cgsize(size: tcgsize): boolean;{$ifdef USEINLINE}inline;{$endif}
  451. implementation
  452. uses
  453. verbose,
  454. cutils;
  455. {******************************************************************************
  456. tsuperregisterworklist
  457. ******************************************************************************}
  458. constructor tsuperregisterworklist.init;
  459. begin
  460. length:=0;
  461. buflength:=0;
  462. buflengthinc:=16;
  463. buf:=nil;
  464. end;
  465. constructor Tsuperregisterworklist.copyfrom(const x:Tsuperregisterworklist);
  466. begin
  467. self:=x;
  468. if x.buf<>nil then
  469. begin
  470. getmem(buf,buflength*sizeof(Tsuperregister));
  471. move(x.buf^,buf^,length*sizeof(Tsuperregister));
  472. end;
  473. end;
  474. destructor tsuperregisterworklist.done;
  475. begin
  476. if assigned(buf) then
  477. freemem(buf);
  478. end;
  479. procedure tsuperregisterworklist.add(s:tsuperregister);
  480. begin
  481. inc(length);
  482. { Need to increase buffer length? }
  483. if length>=buflength then
  484. begin
  485. inc(buflength,buflengthinc);
  486. buflengthinc:=buflengthinc*2;
  487. if buflengthinc>256 then
  488. buflengthinc:=256;
  489. reallocmem(buf,buflength*sizeof(Tsuperregister));
  490. end;
  491. buf^[length-1]:=s;
  492. end;
  493. function tsuperregisterworklist.addnodup(s:tsuperregister): boolean;
  494. begin
  495. addnodup := false;
  496. if indexword(buf^,length,s) = -1 then
  497. begin
  498. add(s);
  499. addnodup := true;
  500. end;
  501. end;
  502. procedure tsuperregisterworklist.clear;
  503. begin
  504. length:=0;
  505. end;
  506. procedure tsuperregisterworklist.deleteidx(i:word);
  507. begin
  508. if i>=length then
  509. internalerror(200310144);
  510. buf^[i]:=buf^[length-1];
  511. dec(length);
  512. end;
  513. function tsuperregisterworklist.readidx(i:word):tsuperregister;
  514. begin
  515. if (i >= length) then
  516. internalerror(2005010601);
  517. result := buf^[i];
  518. end;
  519. function tsuperregisterworklist.get:tsuperregister;
  520. begin
  521. if length=0 then
  522. internalerror(200310142);
  523. dec(length);
  524. get:=buf^[length];
  525. end;
  526. function tsuperregisterworklist.delete(s:tsuperregister):boolean;
  527. var
  528. i:longint;
  529. begin
  530. delete:=false;
  531. { indexword in 1.0.x and 1.9.4 is broken }
  532. i:=indexword(buf^,length,s);
  533. if i<>-1 then
  534. begin
  535. deleteidx(i);
  536. delete := true;
  537. end;
  538. end;
  539. procedure supregset_reset(var regs:tsuperregisterset;setall:boolean;
  540. maxreg:Tsuperregister);{$ifdef USEINLINE}inline;{$endif}
  541. begin
  542. fillchar(regs,(maxreg+7) shr 3,-byte(setall));
  543. end;
  544. procedure supregset_include(var regs:tsuperregisterset;s:tsuperregister);{$ifdef USEINLINE}inline;{$endif}
  545. begin
  546. include(regs[s shr 8],(s and $ff));
  547. end;
  548. procedure supregset_exclude(var regs:tsuperregisterset;s:tsuperregister);{$ifdef USEINLINE}inline;{$endif}
  549. begin
  550. exclude(regs[s shr 8],(s and $ff));
  551. end;
  552. function supregset_in(const regs:tsuperregisterset;s:tsuperregister):boolean;{$ifdef USEINLINE}inline;{$endif}
  553. begin
  554. result:=(s and $ff) in regs[s shr 8];
  555. end;
  556. function newreg(rt:tregistertype;sr:tsuperregister;sb:tsubregister):tregister;{$ifdef USEINLINE}inline;{$endif}
  557. begin
  558. tregisterrec(result).regtype:=rt;
  559. tregisterrec(result).supreg:=sr;
  560. tregisterrec(result).subreg:=sb;
  561. end;
  562. function getsubreg(r:tregister):tsubregister;{$ifdef USEINLINE}inline;{$endif}
  563. begin
  564. result:=tregisterrec(r).subreg;
  565. end;
  566. function getsupreg(r:tregister):tsuperregister;{$ifdef USEINLINE}inline;{$endif}
  567. begin
  568. result:=tregisterrec(r).supreg;
  569. end;
  570. function getregtype(r:tregister):tregistertype;{$ifdef USEINLINE}inline;{$endif}
  571. begin
  572. result:=tregisterrec(r).regtype;
  573. end;
  574. procedure setsubreg(var r:tregister;sr:tsubregister);{$ifdef USEINLINE}inline;{$endif}
  575. begin
  576. tregisterrec(r).subreg:=sr;
  577. end;
  578. procedure setsupreg(var r:tregister;sr:tsuperregister);{$ifdef USEINLINE}inline;{$endif}
  579. begin
  580. tregisterrec(r).supreg:=sr;
  581. end;
  582. function generic_regname(r:tregister):string;
  583. var
  584. nr : string[12];
  585. begin
  586. str(getsupreg(r),nr);
  587. case getregtype(r) of
  588. R_INTREGISTER:
  589. result:='ireg'+nr;
  590. R_FPUREGISTER:
  591. result:='freg'+nr;
  592. R_MMREGISTER:
  593. result:='mreg'+nr;
  594. R_MMXREGISTER:
  595. result:='xreg'+nr;
  596. R_ADDRESSREGISTER:
  597. result:='areg'+nr;
  598. R_SPECIALREGISTER:
  599. result:='sreg'+nr;
  600. else
  601. begin
  602. result:='INVALID';
  603. exit;
  604. end;
  605. end;
  606. case getsubreg(r) of
  607. R_SUBNONE:
  608. ;
  609. R_SUBL:
  610. result:=result+'l';
  611. R_SUBH:
  612. result:=result+'h';
  613. R_SUBW:
  614. result:=result+'w';
  615. R_SUBD:
  616. result:=result+'d';
  617. R_SUBQ:
  618. result:=result+'q';
  619. R_SUBFS:
  620. result:=result+'fs';
  621. R_SUBFD:
  622. result:=result+'fd';
  623. R_SUBMMD:
  624. result:=result+'md';
  625. R_SUBMMS:
  626. result:=result+'ms';
  627. R_SUBMMWHOLE:
  628. result:=result+'ma';
  629. R_SUBMMX:
  630. result:=result+'mx';
  631. R_SUBMMY:
  632. result:=result+'my';
  633. R_SUBMMZ:
  634. result:=result+'mz';
  635. {$ifdef aarch64}
  636. R_SUBMM8B:
  637. result:=result+'m8b';
  638. R_SUBMM16B:
  639. result:=result+'m16b';
  640. R_SUBMM4H:
  641. result:=result+'m4h';
  642. R_SUBMM8H:
  643. result:=result+'m8h';
  644. R_SUBMM2S:
  645. result:=result+'m2s';
  646. R_SUBMM4S:
  647. result:=result+'m4s';
  648. R_SUBMM2D:
  649. result:=result+'m2d';
  650. R_SUBMMB1:
  651. result:=result+'mb1';
  652. R_SUBMMH1:
  653. result:=result+'mh1';
  654. R_SUBMMS1:
  655. result:=result+'ms1';
  656. R_SUBMMD1:
  657. result:=result+'md1';
  658. {$endif}
  659. else
  660. internalerror(200308252);
  661. end;
  662. end;
  663. function int_cgsize(const a: tcgint): tcgsize;{$ifdef USEINLINE}inline;{$endif}
  664. const
  665. size2cgsize : array[0..8] of tcgsize = (
  666. OS_NO,OS_8,OS_16,OS_NO,OS_32,OS_NO,OS_NO,OS_NO,OS_64
  667. );
  668. begin
  669. {$ifdef cpu64bitalu}
  670. if a=16 then
  671. result:=OS_128
  672. else
  673. {$endif cpu64bitalu}
  674. if a>8 then
  675. result:=OS_NO
  676. else
  677. result:=size2cgsize[a];
  678. end;
  679. function int_float_cgsize(const a: tcgint): tcgsize;
  680. begin
  681. case a of
  682. 4 :
  683. result:=OS_F32;
  684. 8 :
  685. result:=OS_F64;
  686. 10 :
  687. result:=OS_F80;
  688. 16 :
  689. result:=OS_F128;
  690. else
  691. internalerror(200603211);
  692. end;
  693. end;
  694. function float_array_cgsize(const a: tcgint): tcgsize;{$ifdef USEINLINE}inline;{$endif}
  695. begin
  696. case a of
  697. 4:
  698. result := OS_M32;
  699. 16:
  700. result := OS_M128;
  701. 32:
  702. result := OS_M256;
  703. 64:
  704. result := OS_M512;
  705. else
  706. result := int_cgsize(a);
  707. end;
  708. end;
  709. function double_array_cgsize(const a: tcgint): tcgsize;{$ifdef USEINLINE}inline;{$endif}
  710. begin
  711. case a of
  712. 8:
  713. result := OS_M64;
  714. 16:
  715. result := OS_M128;
  716. 32:
  717. result := OS_M256;
  718. 64:
  719. result := OS_M512;
  720. else
  721. result := int_cgsize(a);
  722. end;
  723. end;
  724. function tcgsize2str(cgsize: tcgsize):string;
  725. begin
  726. Str(cgsize, Result);
  727. end;
  728. function inverse_opcmp(opcmp: topcmp): topcmp;{$ifdef USEINLINE}inline;{$endif}
  729. const
  730. list: array[TOpCmp] of TOpCmp =
  731. (OC_NONE,OC_NE,OC_LTE,OC_GTE,OC_LT,OC_GT,OC_EQ,OC_A,OC_AE,
  732. OC_B,OC_BE);
  733. begin
  734. inverse_opcmp := list[opcmp];
  735. end;
  736. function swap_opcmp(opcmp: topcmp): topcmp;{$ifdef USEINLINE}inline;{$endif}
  737. const
  738. list: array[TOpCmp] of TOpCmp =
  739. (OC_NONE,OC_EQ,OC_LT,OC_GT,OC_LTE,OC_GTE,OC_NE,OC_AE,OC_A,
  740. OC_BE,OC_B);
  741. begin
  742. swap_opcmp := list[opcmp];
  743. end;
  744. function commutativeop(op: topcg): boolean;{$ifdef USEINLINE}inline;{$endif}
  745. const
  746. list: array[topcg] of boolean =
  747. (true,false,true,true,false,false,true,true,false,false,
  748. true,false,false,false,false,true,false,false);
  749. begin
  750. commutativeop := list[op];
  751. end;
  752. function realshuffle(shuffle : pmmshuffle) : boolean;
  753. var
  754. i : longint;
  755. begin
  756. realshuffle:=true;
  757. if (shuffle=nil) or (shuffle^.len<1) then
  758. realshuffle:=false
  759. else
  760. begin
  761. for i:=1 to shuffle^.len do
  762. begin
  763. if (shuffle^.shuffles[i] and $ff)<>((shuffle^.shuffles[i] and $ff00) shr 8) then
  764. exit;
  765. end;
  766. realshuffle:=false;
  767. end;
  768. end;
  769. function shufflescalar(shuffle : pmmshuffle) : boolean;
  770. begin
  771. result:=shuffle^.len=0;
  772. end;
  773. procedure removeshuffles(var shuffle : tmmshuffle);
  774. var
  775. i : longint;
  776. begin
  777. if shuffle.len=0 then
  778. exit;
  779. for i:=1 to shuffle.len do
  780. shuffle.shuffles[i]:=(shuffle.shuffles[i] and $f) or ((shuffle.shuffles[i] and $f0) shr 4);
  781. end;
  782. function is_float_cgsize(size: tcgsize): boolean;{$ifdef USEINLINE}inline;{$endif}
  783. begin
  784. result:=size in [OS_F32..OS_F128];
  785. end;
  786. procedure Initmms(var p : pmmshuffle;len : ShortInt);
  787. var
  788. i : Integer;
  789. begin
  790. Getmem(p,sizeof(tmmshuffle)+(max(len,0)-1)*2);
  791. p^.len:=len;
  792. for i:=1 to len do
  793. {$push}
  794. {$R-}
  795. p^.shuffles[i]:=i;
  796. {$pop}
  797. end;
  798. initialization
  799. Initmms(mms_movescalar,0);
  800. Initmms(mms_variable,-1);
  801. Initmms(mms_2,2);
  802. Initmms(mms_4,4);
  803. Initmms(mms_8,8);
  804. Initmms(mms_16,16);
  805. Initmms(mms_32,32);
  806. finalization
  807. Freemem(mms_movescalar);
  808. Freemem(mms_variable);
  809. Freemem(mms_2);
  810. Freemem(mms_4);
  811. Freemem(mms_8);
  812. Freemem(mms_16);
  813. Freemem(mms_32);
  814. end.