cgbase.pas 30 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953
  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. Tregistermmxset = record
  301. reg0,reg1,reg2,reg3:Tregister
  302. end;
  303. { Set type definition for registers }
  304. tsuperregisterset = array[byte] of set of byte;
  305. pmmshuffle = ^tmmshuffle;
  306. { this record describes shuffle operations for mm operations; if a pointer a shuffle record
  307. passed to an mm operation is nil, it means that the whole location is moved }
  308. tmmshuffle = record
  309. { describes how many shuffles are actually described, if len=0 then
  310. moving the scalar with index 0 to the scalar with index 0 is meant,
  311. if len=-1, then a variable/unknown length is assumed }
  312. len : Shortint;
  313. { lower byte of each entry of this array describes index of the source data index while
  314. the upper byte describes the destination index }
  315. shuffles : array[1..1] of word;
  316. end;
  317. Tsuperregisterarray=array[0..$ffff] of Tsuperregister;
  318. Psuperregisterarray=^Tsuperregisterarray;
  319. Tsuperregisterworklist=object
  320. buflength,
  321. buflengthinc,
  322. length:word;
  323. buf:Psuperregisterarray;
  324. constructor init;
  325. constructor copyfrom(const x:Tsuperregisterworklist);
  326. destructor done;
  327. procedure clear;
  328. procedure add(s:tsuperregister);
  329. function addnodup(s:tsuperregister): boolean;
  330. { returns the last element and removes it from the list }
  331. function get:tsuperregister;
  332. function readidx(i:word):tsuperregister;
  333. procedure deleteidx(i:word);
  334. function delete(s:tsuperregister):boolean;
  335. end;
  336. psuperregisterworklist=^tsuperregisterworklist;
  337. const
  338. { alias for easier understanding }
  339. R_SSEREGISTER = R_MMREGISTER;
  340. { Invalid register number }
  341. RS_INVALID = high(tsuperregister);
  342. NR_INVALID = tregister($ffffffff);
  343. tcgsize2size : Array[tcgsize] of integer =
  344. (0,
  345. { integer values }
  346. 1, 2, 4, 8, 16,
  347. 1, 2, 4, 8, 16,
  348. { floating point values }
  349. 4, 8, 10, 8, 16,
  350. { multimedia values }
  351. 1, 2, 4, 8, 16, 32, 64);
  352. tfloat2tcgsize: array[tfloattype] of tcgsize =
  353. (OS_F32,OS_F64,OS_F80,OS_F80,OS_C64,OS_C64,OS_F128);
  354. tcgsize2tfloat: array[OS_F32..OS_C64] of tfloattype =
  355. (s32real,s64real,s80real,s64comp);
  356. tvarregable2tcgloc : array[tvarregable] of tcgloc = (LOC_VOID,
  357. LOC_CREGISTER,LOC_CFPUREGISTER,LOC_CMMREGISTER,LOC_CREGISTER);
  358. {$if defined(cpu64bitalu)}
  359. { operand size describing an unsigned value in a pair of int registers }
  360. OS_PAIR = OS_128;
  361. { operand size describing an signed value in a pair of int registers }
  362. OS_SPAIR = OS_S128;
  363. {$elseif defined(cpu32bitalu)}
  364. { operand size describing an unsigned value in a pair of int registers }
  365. OS_PAIR = OS_64;
  366. { operand size describing an signed value in a pair of int registers }
  367. OS_SPAIR = OS_S64;
  368. {$elseif defined(cpu16bitalu)}
  369. { operand size describing an unsigned value in a pair of int registers }
  370. OS_PAIR = OS_32;
  371. { operand size describing an signed value in a pair of int registers }
  372. OS_SPAIR = OS_S32;
  373. {$elseif defined(cpu8bitalu)}
  374. { operand size describing an unsigned value in a pair of int registers }
  375. OS_PAIR = OS_16;
  376. { operand size describing an signed value in a pair of int registers }
  377. OS_SPAIR = OS_S16;
  378. {$endif}
  379. { Table to convert tcgsize variables to the correspondending
  380. unsigned types }
  381. tcgsize2unsigned : array[tcgsize] of tcgsize = (OS_NO,
  382. OS_8, OS_16, OS_32, OS_64, OS_128,
  383. OS_8, OS_16, OS_32, OS_64, OS_128,
  384. OS_F32, OS_F64, OS_F80, OS_C64, OS_F128,
  385. OS_M8, OS_M16, OS_M32, OS_M64, OS_M128, OS_M256, OS_M512);
  386. tcgsize2signed : array[tcgsize] of tcgsize = (OS_NO,
  387. OS_S8, OS_S16, OS_S32, OS_S64, OS_S128,
  388. OS_S8, OS_S16, OS_S32, OS_S64, OS_S128,
  389. OS_F32, OS_F64, OS_F80, OS_C64, OS_F128,
  390. OS_M8, OS_M16, OS_M32, OS_M64, OS_M128, OS_M256,OS_M512);
  391. tcgloc2str : array[TCGLoc] of string[12] = (
  392. 'LOC_INVALID',
  393. 'LOC_VOID',
  394. 'LOC_CONST',
  395. 'LOC_JUMP',
  396. 'LOC_FLAGS',
  397. 'LOC_REG',
  398. 'LOC_CREG',
  399. 'LOC_FPUREG',
  400. 'LOC_CFPUREG',
  401. 'LOC_MMXREG',
  402. 'LOC_CMMXREG',
  403. 'LOC_MMREG',
  404. 'LOC_CMMREG',
  405. 'LOC_SSETREG',
  406. 'LOC_CSSETREG',
  407. 'LOC_SSETREF',
  408. 'LOC_CSSETREF',
  409. 'LOC_CREF',
  410. 'LOC_REF'
  411. );
  412. var
  413. mms_movescalar,
  414. mms_variable,
  415. mms_2,
  416. mms_4,
  417. mms_8,
  418. mms_16,
  419. mms_32 : pmmshuffle;
  420. procedure supregset_reset(var regs:tsuperregisterset;setall:boolean;
  421. maxreg:Tsuperregister);{$ifdef USEINLINE}inline;{$endif}
  422. procedure supregset_include(var regs:tsuperregisterset;s:tsuperregister);{$ifdef USEINLINE}inline;{$endif}
  423. procedure supregset_exclude(var regs:tsuperregisterset;s:tsuperregister);{$ifdef USEINLINE}inline;{$endif}
  424. function supregset_in(const regs:tsuperregisterset;s:tsuperregister):boolean;{$ifdef USEINLINE}inline;{$endif}
  425. function newreg(rt:tregistertype;sr:tsuperregister;sb:tsubregister):tregister;{$ifdef USEINLINE}inline;{$endif}
  426. function getsubreg(r:tregister):tsubregister;{$ifdef USEINLINE}inline;{$endif}
  427. function getsupreg(r:tregister):tsuperregister;{$ifdef USEINLINE}inline;{$endif}
  428. function getregtype(r:tregister):tregistertype;{$ifdef USEINLINE}inline;{$endif}
  429. procedure setsubreg(var r:tregister;sr:tsubregister);{$ifdef USEINLINE}inline;{$endif}
  430. procedure setsupreg(var r:tregister;sr:tsuperregister);{$ifdef USEINLINE}inline;{$endif}
  431. function generic_regname(r:tregister):string;
  432. {# From a constant numeric value, return the abstract code generator
  433. size.
  434. }
  435. function int_cgsize(const a: tcgint): tcgsize;{$ifdef USEINLINE}inline;{$endif}
  436. function int_float_cgsize(const a: tcgint): tcgsize;
  437. function float_array_cgsize(const a: tcgint): tcgsize;{$ifdef USEINLINE}inline;{$endif}
  438. function double_array_cgsize(const a: tcgint): tcgsize;{$ifdef USEINLINE}inline;{$endif}
  439. function tcgsize2str(cgsize: tcgsize):string;
  440. { return the inverse condition of opcmp }
  441. function inverse_opcmp(opcmp: topcmp): topcmp;{$ifdef USEINLINE}inline;{$endif}
  442. { return the opcmp needed when swapping the operands }
  443. function swap_opcmp(opcmp: topcmp): topcmp;{$ifdef USEINLINE}inline;{$endif}
  444. { return whether op is commutative }
  445. function commutativeop(op: topcg): boolean;{$ifdef USEINLINE}inline;{$endif}
  446. { returns true, if shuffle describes a real shuffle operation and not only a move }
  447. function realshuffle(shuffle : pmmshuffle) : boolean;
  448. { returns true, if the shuffle describes only a move of the scalar at index 0 }
  449. function shufflescalar(shuffle : pmmshuffle) : boolean;
  450. { removes shuffling from shuffle, this means that the destenation index of each shuffle is copied to
  451. the source }
  452. procedure removeshuffles(var shuffle : tmmshuffle);
  453. function is_float_cgsize(size: tcgsize): boolean;{$ifdef USEINLINE}inline;{$endif}
  454. implementation
  455. uses
  456. verbose,
  457. cutils;
  458. {******************************************************************************
  459. tsuperregisterworklist
  460. ******************************************************************************}
  461. constructor tsuperregisterworklist.init;
  462. begin
  463. length:=0;
  464. buflength:=0;
  465. buflengthinc:=16;
  466. buf:=nil;
  467. end;
  468. constructor Tsuperregisterworklist.copyfrom(const x:Tsuperregisterworklist);
  469. begin
  470. self:=x;
  471. if x.buf<>nil then
  472. begin
  473. getmem(buf,buflength*sizeof(Tsuperregister));
  474. move(x.buf^,buf^,length*sizeof(Tsuperregister));
  475. end;
  476. end;
  477. destructor tsuperregisterworklist.done;
  478. begin
  479. if assigned(buf) then
  480. freemem(buf);
  481. end;
  482. procedure tsuperregisterworklist.add(s:tsuperregister);
  483. begin
  484. inc(length);
  485. { Need to increase buffer length? }
  486. if length>=buflength then
  487. begin
  488. inc(buflength,buflengthinc);
  489. buflengthinc:=buflengthinc*2;
  490. if buflengthinc>256 then
  491. buflengthinc:=256;
  492. reallocmem(buf,buflength*sizeof(Tsuperregister));
  493. end;
  494. buf^[length-1]:=s;
  495. end;
  496. function tsuperregisterworklist.addnodup(s:tsuperregister): boolean;
  497. begin
  498. addnodup := false;
  499. if indexword(buf^,length,s) = -1 then
  500. begin
  501. add(s);
  502. addnodup := true;
  503. end;
  504. end;
  505. procedure tsuperregisterworklist.clear;
  506. begin
  507. length:=0;
  508. end;
  509. procedure tsuperregisterworklist.deleteidx(i:word);
  510. begin
  511. if i>=length then
  512. internalerror(200310144);
  513. buf^[i]:=buf^[length-1];
  514. dec(length);
  515. end;
  516. function tsuperregisterworklist.readidx(i:word):tsuperregister;
  517. begin
  518. if (i >= length) then
  519. internalerror(2005010601);
  520. result := buf^[i];
  521. end;
  522. function tsuperregisterworklist.get:tsuperregister;
  523. begin
  524. if length=0 then
  525. internalerror(200310142);
  526. dec(length);
  527. get:=buf^[length];
  528. end;
  529. function tsuperregisterworklist.delete(s:tsuperregister):boolean;
  530. var
  531. i:longint;
  532. begin
  533. delete:=false;
  534. { indexword in 1.0.x and 1.9.4 is broken }
  535. i:=indexword(buf^,length,s);
  536. if i<>-1 then
  537. begin
  538. deleteidx(i);
  539. delete := true;
  540. end;
  541. end;
  542. procedure supregset_reset(var regs:tsuperregisterset;setall:boolean;
  543. maxreg:Tsuperregister);{$ifdef USEINLINE}inline;{$endif}
  544. begin
  545. fillchar(regs,(maxreg+7) shr 3,-byte(setall));
  546. end;
  547. procedure supregset_include(var regs:tsuperregisterset;s:tsuperregister);{$ifdef USEINLINE}inline;{$endif}
  548. begin
  549. include(regs[s shr 8],(s and $ff));
  550. end;
  551. procedure supregset_exclude(var regs:tsuperregisterset;s:tsuperregister);{$ifdef USEINLINE}inline;{$endif}
  552. begin
  553. exclude(regs[s shr 8],(s and $ff));
  554. end;
  555. function supregset_in(const regs:tsuperregisterset;s:tsuperregister):boolean;{$ifdef USEINLINE}inline;{$endif}
  556. begin
  557. result:=(s and $ff) in regs[s shr 8];
  558. end;
  559. function newreg(rt:tregistertype;sr:tsuperregister;sb:tsubregister):tregister;{$ifdef USEINLINE}inline;{$endif}
  560. begin
  561. tregisterrec(result).regtype:=rt;
  562. tregisterrec(result).supreg:=sr;
  563. tregisterrec(result).subreg:=sb;
  564. end;
  565. function getsubreg(r:tregister):tsubregister;{$ifdef USEINLINE}inline;{$endif}
  566. begin
  567. result:=tregisterrec(r).subreg;
  568. end;
  569. function getsupreg(r:tregister):tsuperregister;{$ifdef USEINLINE}inline;{$endif}
  570. begin
  571. result:=tregisterrec(r).supreg;
  572. end;
  573. function getregtype(r:tregister):tregistertype;{$ifdef USEINLINE}inline;{$endif}
  574. begin
  575. result:=tregisterrec(r).regtype;
  576. end;
  577. procedure setsubreg(var r:tregister;sr:tsubregister);{$ifdef USEINLINE}inline;{$endif}
  578. begin
  579. tregisterrec(r).subreg:=sr;
  580. end;
  581. procedure setsupreg(var r:tregister;sr:tsuperregister);{$ifdef USEINLINE}inline;{$endif}
  582. begin
  583. tregisterrec(r).supreg:=sr;
  584. end;
  585. function generic_regname(r:tregister):string;
  586. var
  587. nr : string[12];
  588. begin
  589. str(getsupreg(r),nr);
  590. case getregtype(r) of
  591. R_INTREGISTER:
  592. result:='ireg'+nr;
  593. R_FPUREGISTER:
  594. result:='freg'+nr;
  595. R_MMREGISTER:
  596. result:='mreg'+nr;
  597. R_MMXREGISTER:
  598. result:='xreg'+nr;
  599. R_ADDRESSREGISTER:
  600. result:='areg'+nr;
  601. R_SPECIALREGISTER:
  602. result:='sreg'+nr;
  603. else
  604. begin
  605. result:='INVALID';
  606. exit;
  607. end;
  608. end;
  609. case getsubreg(r) of
  610. R_SUBNONE:
  611. ;
  612. R_SUBL:
  613. result:=result+'l';
  614. R_SUBH:
  615. result:=result+'h';
  616. R_SUBW:
  617. result:=result+'w';
  618. R_SUBD:
  619. result:=result+'d';
  620. R_SUBQ:
  621. result:=result+'q';
  622. R_SUBFS:
  623. result:=result+'fs';
  624. R_SUBFD:
  625. result:=result+'fd';
  626. R_SUBMMD:
  627. result:=result+'md';
  628. R_SUBMMS:
  629. result:=result+'ms';
  630. R_SUBMMWHOLE:
  631. result:=result+'ma';
  632. R_SUBMMX:
  633. result:=result+'mx';
  634. R_SUBMMY:
  635. result:=result+'my';
  636. R_SUBMMZ:
  637. result:=result+'mz';
  638. {$ifdef aarch64}
  639. R_SUBMM8B:
  640. result:=result+'m8b';
  641. R_SUBMM16B:
  642. result:=result+'m16b';
  643. R_SUBMM4H:
  644. result:=result+'m4h';
  645. R_SUBMM8H:
  646. result:=result+'m8h';
  647. R_SUBMM2S:
  648. result:=result+'m2s';
  649. R_SUBMM4S:
  650. result:=result+'m4s';
  651. R_SUBMM2D:
  652. result:=result+'m2d';
  653. R_SUBMMB1:
  654. result:=result+'mb1';
  655. R_SUBMMH1:
  656. result:=result+'mh1';
  657. R_SUBMMS1:
  658. result:=result+'ms1';
  659. R_SUBMMD1:
  660. result:=result+'md1';
  661. {$endif}
  662. else
  663. internalerror(200308252);
  664. end;
  665. end;
  666. function int_cgsize(const a: tcgint): tcgsize;{$ifdef USEINLINE}inline;{$endif}
  667. const
  668. size2cgsize : array[0..8] of tcgsize = (
  669. OS_NO,OS_8,OS_16,OS_NO,OS_32,OS_NO,OS_NO,OS_NO,OS_64
  670. );
  671. begin
  672. {$ifdef cpu64bitalu}
  673. if a=16 then
  674. result:=OS_128
  675. else
  676. {$endif cpu64bitalu}
  677. if a>8 then
  678. result:=OS_NO
  679. else
  680. result:=size2cgsize[a];
  681. end;
  682. function int_float_cgsize(const a: tcgint): tcgsize;
  683. begin
  684. case a of
  685. 4 :
  686. result:=OS_F32;
  687. 8 :
  688. result:=OS_F64;
  689. 10 :
  690. result:=OS_F80;
  691. 16 :
  692. result:=OS_F128;
  693. else
  694. internalerror(200603211);
  695. end;
  696. end;
  697. function float_array_cgsize(const a: tcgint): tcgsize;{$ifdef USEINLINE}inline;{$endif}
  698. begin
  699. case a of
  700. 4:
  701. result := OS_M32;
  702. 16:
  703. result := OS_M128;
  704. 32:
  705. result := OS_M256;
  706. 64:
  707. result := OS_M512;
  708. else
  709. result := int_cgsize(a);
  710. end;
  711. end;
  712. function double_array_cgsize(const a: tcgint): tcgsize;{$ifdef USEINLINE}inline;{$endif}
  713. begin
  714. case a of
  715. 8:
  716. result := OS_M64;
  717. 16:
  718. result := OS_M128;
  719. 32:
  720. result := OS_M256;
  721. 64:
  722. result := OS_M512;
  723. else
  724. result := int_cgsize(a);
  725. end;
  726. end;
  727. function tcgsize2str(cgsize: tcgsize):string;
  728. begin
  729. Str(cgsize, Result);
  730. end;
  731. function inverse_opcmp(opcmp: topcmp): topcmp;{$ifdef USEINLINE}inline;{$endif}
  732. const
  733. list: array[TOpCmp] of TOpCmp =
  734. (OC_NONE,OC_NE,OC_LTE,OC_GTE,OC_LT,OC_GT,OC_EQ,OC_A,OC_AE,
  735. OC_B,OC_BE);
  736. begin
  737. inverse_opcmp := list[opcmp];
  738. end;
  739. function swap_opcmp(opcmp: topcmp): topcmp;{$ifdef USEINLINE}inline;{$endif}
  740. const
  741. list: array[TOpCmp] of TOpCmp =
  742. (OC_NONE,OC_EQ,OC_LT,OC_GT,OC_LTE,OC_GTE,OC_NE,OC_AE,OC_A,
  743. OC_BE,OC_B);
  744. begin
  745. swap_opcmp := list[opcmp];
  746. end;
  747. function commutativeop(op: topcg): boolean;{$ifdef USEINLINE}inline;{$endif}
  748. const
  749. list: array[topcg] of boolean =
  750. (true,false,true,true,false,false,true,true,false,false,
  751. true,false,false,false,false,true,false,false);
  752. begin
  753. commutativeop := list[op];
  754. end;
  755. function realshuffle(shuffle : pmmshuffle) : boolean;
  756. var
  757. i : longint;
  758. begin
  759. realshuffle:=true;
  760. if (shuffle=nil) or (shuffle^.len<1) then
  761. realshuffle:=false
  762. else
  763. begin
  764. for i:=1 to shuffle^.len do
  765. begin
  766. if (shuffle^.shuffles[i] and $ff)<>((shuffle^.shuffles[i] and $ff00) shr 8) then
  767. exit;
  768. end;
  769. realshuffle:=false;
  770. end;
  771. end;
  772. function shufflescalar(shuffle : pmmshuffle) : boolean;
  773. begin
  774. result:=shuffle^.len=0;
  775. end;
  776. procedure removeshuffles(var shuffle : tmmshuffle);
  777. var
  778. i : longint;
  779. begin
  780. if shuffle.len=0 then
  781. exit;
  782. for i:=1 to shuffle.len do
  783. shuffle.shuffles[i]:=(shuffle.shuffles[i] and $f) or ((shuffle.shuffles[i] and $f0) shr 4);
  784. end;
  785. function is_float_cgsize(size: tcgsize): boolean;{$ifdef USEINLINE}inline;{$endif}
  786. begin
  787. result:=size in [OS_F32..OS_F128];
  788. end;
  789. procedure Initmms(var p : pmmshuffle;len : ShortInt);
  790. var
  791. i : Integer;
  792. begin
  793. Getmem(p,sizeof(tmmshuffle)+(max(len,0)-1)*2);
  794. p^.len:=len;
  795. for i:=1 to len do
  796. {$push}
  797. {$R-}
  798. p^.shuffles[i]:=i;
  799. {$pop}
  800. end;
  801. initialization
  802. Initmms(mms_movescalar,0);
  803. Initmms(mms_variable,-1);
  804. Initmms(mms_2,2);
  805. Initmms(mms_4,4);
  806. Initmms(mms_8,8);
  807. Initmms(mms_16,16);
  808. Initmms(mms_32,32);
  809. finalization
  810. Freemem(mms_movescalar);
  811. Freemem(mms_variable);
  812. Freemem(mms_2);
  813. Freemem(mms_4);
  814. Freemem(mms_8);
  815. Freemem(mms_16);
  816. Freemem(mms_32);
  817. end.