cpupara.pas 22 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603
  1. {
  2. Copyright (c) 2002 by Florian Klaempfl
  3. Generates the argument location information for x86-64 target
  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 bymethodpointer
  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. unit cpupara;
  18. {$i fpcdefs.inc}
  19. interface
  20. uses
  21. globtype,
  22. cpubase,cgbase,
  23. symconst,symtype,symsym,symdef,
  24. aasmtai,aasmdata,
  25. parabase,paramgr;
  26. type
  27. tx86_64paramanager = class(tparamanager)
  28. private
  29. procedure create_funcretloc_info(p : tabstractprocdef; side: tcallercallee);
  30. procedure create_paraloc_info_intern(p : tabstractprocdef; side: tcallercallee;paras:tparalist;
  31. var intparareg,mmparareg,parasize:longint;varargsparas: boolean);
  32. public
  33. function param_use_paraloc(const cgpara:tcgpara):boolean;override;
  34. function push_addr_param(varspez:tvarspez;def : tdef;calloption : tproccalloption) : boolean;override;
  35. function ret_in_param(def : tdef;calloption : tproccalloption) : boolean;override;
  36. procedure getintparaloc(calloption : tproccalloption; nr : longint;var cgpara:TCGPara);override;
  37. function get_volatile_registers_int(calloption : tproccalloption):tcpuregisterset;override;
  38. function get_volatile_registers_mm(calloption : tproccalloption):tcpuregisterset;override;
  39. function get_volatile_registers_fpu(calloption : tproccalloption):tcpuregisterset;override;
  40. function create_paraloc_info(p : tabstractprocdef; side: tcallercallee):longint;override;
  41. function create_varargs_paraloc_info(p : tabstractprocdef; varargspara:tvarargsparalist):longint;override;
  42. end;
  43. implementation
  44. uses
  45. cutils,verbose,
  46. systems,
  47. defutil,
  48. cgutils;
  49. const
  50. paraintsupregs : array[0..5] of tsuperregister = (RS_RDI,RS_RSI,RS_RDX,RS_RCX,RS_R8,RS_R9);
  51. parammsupregs : array[0..7] of tsuperregister = (RS_XMM0,RS_XMM1,RS_XMM2,RS_XMM3,RS_XMM4,RS_XMM5,RS_XMM6,RS_XMM7);
  52. paraintsupregs_winx64 : array[0..3] of tsuperregister = (RS_RCX,RS_RDX,RS_R8,RS_R9);
  53. parammsupregs_winx64 : array[0..3] of tsuperregister = (RS_XMM0,RS_XMM1,RS_XMM2,RS_XMM3);
  54. procedure getvalueparaloc(p : tdef;var loc1,loc2:tcgloc);
  55. begin
  56. loc1:=LOC_INVALID;
  57. loc2:=LOC_INVALID;
  58. case p.typ of
  59. orddef:
  60. begin
  61. loc1:=LOC_REGISTER;
  62. {$warning TODO 128bit also needs lochigh}
  63. end;
  64. floatdef:
  65. begin
  66. case tfloatdef(p).floattype of
  67. s80real:
  68. loc1:=LOC_REFERENCE;
  69. s32real,
  70. s64real :
  71. loc1:=LOC_MMREGISTER;
  72. s64currency,
  73. s64comp :
  74. loc1:=LOC_REGISTER;
  75. s128real:
  76. begin
  77. loc1:=LOC_MMREGISTER;
  78. loc2:=LOC_MMREGISTER;
  79. {$warning TODO float 128bit needs SSEUP lochigh}
  80. end;
  81. end;
  82. end;
  83. recorddef:
  84. begin
  85. { win64 abi }
  86. if ((target_info.system=system_x86_64_win64) and (p.size<=8)) or
  87. { linux abi }
  88. ((target_info.system<>system_x86_64_win64) and (p.size<=16)) then
  89. loc1:=LOC_REGISTER
  90. else
  91. loc1:=LOC_REFERENCE;
  92. end;
  93. objectdef:
  94. begin
  95. if is_object(p) then
  96. begin
  97. { win64 abi }
  98. if ((target_info.system=system_x86_64_win64) and (p.size<=8)) or
  99. { linux abi }
  100. ((target_info.system<>system_x86_64_win64) and (p.size<=16)) then
  101. loc1:=LOC_REGISTER
  102. else
  103. loc1:=LOC_REFERENCE;
  104. end
  105. else
  106. loc1:=LOC_REGISTER;
  107. end;
  108. arraydef:
  109. begin
  110. if not(is_special_array(p)) and
  111. (
  112. { win64 abi }
  113. ((target_info.system=system_x86_64_win64) and (p.size<=8)) or
  114. { linux abi }
  115. ((target_info.system<>system_x86_64_win64) and (p.size<=16))
  116. ) then
  117. loc1:=LOC_REGISTER
  118. else
  119. loc1:=LOC_REFERENCE;
  120. end;
  121. variantdef:
  122. { linux abi }
  123. if target_info.system<>system_x86_64_win64 then
  124. loc1:=LOC_REGISTER
  125. else
  126. loc1:=LOC_REFERENCE;
  127. stringdef:
  128. if is_shortstring(p) or is_longstring(p) then
  129. begin
  130. { handle long and shortstrings like arrays }
  131. { win64 abi }
  132. if ((target_info.system=system_x86_64_win64) and (p.size<=8)) or
  133. { linux abi }
  134. ((target_info.system<>system_x86_64_win64) and (p.size<=16)) then
  135. loc1:=LOC_REGISTER
  136. else
  137. loc1:=LOC_REFERENCE;
  138. end
  139. else
  140. loc1:=LOC_REGISTER;
  141. setdef:
  142. if is_smallset(p) then
  143. loc1:=LOC_REGISTER
  144. else
  145. loc1:=LOC_REFERENCE;
  146. procvardef:
  147. begin
  148. { This is a record < 16 bytes }
  149. if (po_methodpointer in tprocvardef(p).procoptions) then
  150. begin
  151. loc1:=LOC_REGISTER;
  152. loc2:=LOC_REGISTER;
  153. end
  154. else
  155. loc1:=LOC_REGISTER;
  156. end;
  157. else
  158. begin
  159. { default for pointers,enums,etc }
  160. loc1:=LOC_REGISTER;
  161. end;
  162. end;
  163. end;
  164. function tx86_64paramanager.ret_in_param(def : tdef;calloption : tproccalloption) : boolean;
  165. begin
  166. if target_info.system=system_x86_64_win64 then
  167. result:=(calloption=pocall_safecall) or
  168. (def.size>8) or not(def.size in [1,2,4,8])
  169. else
  170. result:=inherited ret_in_param(def,calloption);
  171. end;
  172. function tx86_64paramanager.param_use_paraloc(const cgpara:tcgpara):boolean;
  173. var
  174. paraloc : pcgparalocation;
  175. begin
  176. if not assigned(cgpara.location) then
  177. internalerror(200410102);
  178. result:=true;
  179. { All locations are LOC_REFERENCE }
  180. paraloc:=cgpara.location;
  181. while assigned(paraloc) do
  182. begin
  183. if (paraloc^.loc<>LOC_REFERENCE) then
  184. begin
  185. result:=false;
  186. exit;
  187. end;
  188. paraloc:=paraloc^.next;
  189. end;
  190. end;
  191. { true if a parameter is too large to copy and only the address is pushed }
  192. function tx86_64paramanager.push_addr_param(varspez:tvarspez;def : tdef;calloption : tproccalloption) : boolean;
  193. begin
  194. result:=false;
  195. { var,out always require address }
  196. if varspez in [vs_var,vs_out] then
  197. begin
  198. result:=true;
  199. exit;
  200. end;
  201. { Only vs_const, vs_value here }
  202. case def.typ of
  203. variantdef,
  204. formaldef :
  205. result:=true;
  206. recorddef :
  207. result:=(varspez=vs_const) or ((target_info.system=system_x86_64_win64) and (def.size>8));
  208. arraydef :
  209. begin
  210. result:=not(
  211. { cdecl array of const need to be ignored and therefor be puhsed
  212. as value parameter with length 0 }
  213. (calloption in [pocall_cdecl,pocall_cppdecl]) and
  214. (is_array_of_const(def) or
  215. is_dynamic_array(def))
  216. );
  217. end;
  218. objectdef :
  219. result:=is_object(def);
  220. stringdef :
  221. result:=(tstringdef(def).stringtype in [st_shortstring,st_longstring]);
  222. procvardef :
  223. result:=(po_methodpointer in tprocvardef(def).procoptions);
  224. setdef :
  225. result:=(tsetdef(def).settype<>smallset);
  226. end;
  227. end;
  228. function tx86_64paramanager.get_volatile_registers_int(calloption : tproccalloption):tcpuregisterset;
  229. begin
  230. if target_info.system=system_x86_64_win64 then
  231. result:=[RS_RAX,RS_RCX,RS_RDX,RS_R8,RS_R9,RS_R10,RS_R11]
  232. else
  233. result:=[RS_RAX,RS_RCX,RS_RDX,RS_RSI,RS_RDI,RS_R8,RS_R9,RS_R10,RS_R11];
  234. end;
  235. function tx86_64paramanager.get_volatile_registers_mm(calloption : tproccalloption):tcpuregisterset;
  236. begin
  237. if target_info.system=system_x86_64_win64 then
  238. result:=[RS_XMM0..RS_XMM5]
  239. else
  240. result:=[RS_XMM0..RS_XMM15];
  241. end;
  242. function tx86_64paramanager.get_volatile_registers_fpu(calloption : tproccalloption):tcpuregisterset;
  243. begin
  244. result:=[RS_ST0..RS_ST7];
  245. end;
  246. procedure tx86_64paramanager.getintparaloc(calloption : tproccalloption; nr : longint;var cgpara:TCGPara);
  247. var
  248. paraloc : pcgparalocation;
  249. begin
  250. cgpara.reset;
  251. cgpara.size:=OS_INT;
  252. cgpara.intsize:=tcgsize2size[OS_INT];
  253. cgpara.alignment:=get_para_align(calloption);
  254. paraloc:=cgpara.add_location;
  255. with paraloc^ do
  256. begin
  257. size:=OS_INT;
  258. if target_info.system=system_x86_64_win64 then
  259. begin
  260. if nr<1 then
  261. internalerror(200304303)
  262. else if nr<=high(paraintsupregs_winx64)+1 then
  263. begin
  264. loc:=LOC_REGISTER;
  265. register:=newreg(R_INTREGISTER,paraintsupregs_winx64[nr-1],R_SUBWHOLE);
  266. end
  267. else
  268. begin
  269. loc:=LOC_REFERENCE;
  270. reference.index:=NR_STACK_POINTER_REG;
  271. reference.offset:=(nr-6)*sizeof(aint);
  272. end;
  273. end
  274. else
  275. begin
  276. if nr<1 then
  277. internalerror(200304303)
  278. else if nr<=high(paraintsupregs)+1 then
  279. begin
  280. loc:=LOC_REGISTER;
  281. register:=newreg(R_INTREGISTER,paraintsupregs[nr-1],R_SUBWHOLE);
  282. end
  283. else
  284. begin
  285. loc:=LOC_REFERENCE;
  286. reference.index:=NR_STACK_POINTER_REG;
  287. reference.offset:=(nr-6)*sizeof(aint);
  288. end;
  289. end;
  290. end;
  291. end;
  292. procedure tx86_64paramanager.create_funcretloc_info(p : tabstractprocdef; side: tcallercallee);
  293. var
  294. retcgsize : tcgsize;
  295. begin
  296. { Constructors return self instead of a boolean }
  297. if (p.proctypeoption=potype_constructor) then
  298. retcgsize:=OS_ADDR
  299. else
  300. retcgsize:=def_cgsize(p.returndef);
  301. location_reset(p.funcretloc[side],LOC_INVALID,OS_NO);
  302. { void has no location }
  303. if is_void(p.returndef) then
  304. begin
  305. location_reset(p.funcretloc[side],LOC_VOID,OS_NO);
  306. exit;
  307. end;
  308. { Return is passed as var parameter }
  309. if ret_in_param(p.returndef,p.proccalloption) then
  310. begin
  311. p.funcretloc[side].loc:=LOC_REFERENCE;
  312. p.funcretloc[side].size:=retcgsize;
  313. exit;
  314. end;
  315. { Return in FPU register? }
  316. if p.returndef.typ=floatdef then
  317. begin
  318. case tfloatdef(p.returndef).floattype of
  319. s32real,s64real:
  320. begin
  321. p.funcretloc[side].loc:=LOC_MMREGISTER;
  322. p.funcretloc[side].register:=NR_MM_RESULT_REG;
  323. p.funcretloc[side].size:=retcgsize;
  324. end;
  325. s64currency,
  326. s64comp,
  327. s80real:
  328. begin
  329. p.funcretloc[side].loc:=LOC_FPUREGISTER;
  330. p.funcretloc[side].register:=NR_FPU_RESULT_REG;
  331. p.funcretloc[side].size:=retcgsize;
  332. end;
  333. else
  334. internalerror(200405034);
  335. end;
  336. end
  337. else
  338. { Return in register }
  339. begin
  340. p.funcretloc[side].loc:=LOC_REGISTER;
  341. p.funcretloc[side].size:=retcgsize;
  342. if side=callerside then
  343. p.funcretloc[side].register:=newreg(R_INTREGISTER,RS_FUNCTION_RESULT_REG,cgsize2subreg(retcgsize))
  344. else
  345. p.funcretloc[side].register:=newreg(R_INTREGISTER,RS_FUNCTION_RETURN_REG,cgsize2subreg(retcgsize));
  346. end;
  347. end;
  348. procedure tx86_64paramanager.create_paraloc_info_intern(p : tabstractprocdef; side: tcallercallee;paras:tparalist;
  349. var intparareg,mmparareg,parasize:longint;varargsparas: boolean);
  350. var
  351. hp : tparavarsym;
  352. paraloc : pcgparalocation;
  353. subreg : tsubregister;
  354. pushaddr : boolean;
  355. paracgsize : tcgsize;
  356. loc : array[1..2] of tcgloc;
  357. paralen,
  358. locidx,
  359. l,i,
  360. varalign,
  361. paraalign : longint;
  362. begin
  363. paraalign:=get_para_align(p.proccalloption);
  364. { Register parameters are assigned from left to right }
  365. for i:=0 to paras.count-1 do
  366. begin
  367. hp:=tparavarsym(paras[i]);
  368. pushaddr:=push_addr_param(hp.varspez,hp.vardef,p.proccalloption);
  369. if pushaddr then
  370. begin
  371. loc[1]:=LOC_REGISTER;
  372. loc[2]:=LOC_INVALID;
  373. paracgsize:=OS_ADDR;
  374. paralen:=sizeof(aint);
  375. end
  376. else
  377. begin
  378. getvalueparaloc(hp.vardef,loc[1],loc[2]);
  379. paralen:=push_size(hp.varspez,hp.vardef,p.proccalloption);
  380. paracgsize:=def_cgsize(hp.vardef);
  381. end;
  382. { cheat for now, we should copy the value to an mm reg as well (FK) }
  383. if varargsparas and
  384. (target_info.system = system_x86_64_win64) and
  385. (hp.vardef.typ = floatdef) then
  386. begin
  387. loc[1] := LOC_REGISTER;
  388. loc[2] := LOC_INVALID;
  389. if paracgsize = OS_F64 then
  390. paracgsize := OS_64
  391. else
  392. paracgsize := OS_32;
  393. end;
  394. hp.paraloc[side].reset;
  395. hp.paraloc[side].size:=paracgsize;
  396. hp.paraloc[side].intsize:=paralen;
  397. hp.paraloc[side].Alignment:=paraalign;
  398. if paralen>0 then
  399. begin
  400. locidx:=1;
  401. while (paralen>0) do
  402. begin
  403. if locidx>2 then
  404. internalerror(200501283);
  405. { Enough registers free? }
  406. case loc[locidx] of
  407. LOC_REGISTER :
  408. begin
  409. { winx64 uses different registers }
  410. if ((target_info.system=system_x86_64_win64) and
  411. (intparareg>high(paraintsupregs_winx64))) or
  412. ((target_info.system<>system_x86_64_win64) and
  413. (intparareg>high(paraintsupregs))) then
  414. loc[locidx]:=LOC_REFERENCE;
  415. end;
  416. LOC_MMREGISTER :
  417. begin
  418. { winx64 uses different registers }
  419. if ((target_info.system=system_x86_64_win64) and
  420. (mmparareg>high(parammsupregs_winx64))) or
  421. ((target_info.system<>system_x86_64_win64) and
  422. (mmparareg>high(parammsupregs))) then
  423. loc[locidx]:=LOC_REFERENCE;
  424. end;
  425. end;
  426. { Allocate }
  427. case loc[locidx] of
  428. LOC_REGISTER :
  429. begin
  430. paraloc:=hp.paraloc[side].add_location;
  431. paraloc^.loc:=LOC_REGISTER;
  432. if (paracgsize=OS_NO) or (loc[2]<>LOC_INVALID) then
  433. begin
  434. paraloc^.size:=OS_INT;
  435. subreg:=R_SUBWHOLE;
  436. end
  437. else
  438. begin
  439. paraloc^.size:=paracgsize;
  440. { s64comp is pushed in an int register }
  441. if paraloc^.size=OS_C64 then
  442. paraloc^.size:=OS_64;
  443. subreg:=cgsize2subreg(paraloc^.size);
  444. end;
  445. { winx64 uses different registers }
  446. if target_info.system=system_x86_64_win64 then
  447. paraloc^.register:=newreg(R_INTREGISTER,paraintsupregs_winx64[intparareg],subreg)
  448. else
  449. paraloc^.register:=newreg(R_INTREGISTER,paraintsupregs[intparareg],subreg);
  450. { matching mm register must be skipped }
  451. if target_info.system=system_x86_64_win64 then
  452. inc(mmparareg);
  453. inc(intparareg);
  454. dec(paralen,tcgsize2size[paraloc^.size]);
  455. end;
  456. LOC_MMREGISTER :
  457. begin
  458. paraloc:=hp.paraloc[side].add_location;
  459. paraloc^.loc:=LOC_MMREGISTER;
  460. { winx64 uses different registers }
  461. if target_info.system=system_x86_64_win64 then
  462. paraloc^.register:=newreg(R_MMREGISTER,parammsupregs_winx64[mmparareg],R_SUBNONE)
  463. else
  464. paraloc^.register:=newreg(R_MMREGISTER,parammsupregs[mmparareg],R_SUBNONE);
  465. if paracgsize=OS_F128 then
  466. paraloc^.size:=OS_F64
  467. else
  468. paraloc^.size:=paracgsize;
  469. { matching int register must be skipped }
  470. if target_info.system=system_x86_64_win64 then
  471. inc(intparareg);
  472. inc(mmparareg);
  473. dec(paralen,tcgsize2size[paraloc^.size]);
  474. end;
  475. LOC_REFERENCE :
  476. begin
  477. paraloc:=hp.paraloc[side].add_location;
  478. paraloc^.loc:=LOC_REFERENCE;
  479. if paracgsize in [OS_F32,OS_F64,OS_F80,OS_F128] then
  480. paraloc^.size:=int_float_cgsize(paralen)
  481. else
  482. paraloc^.size:=int_cgsize(paralen);
  483. if side=callerside then
  484. paraloc^.reference.index:=NR_STACK_POINTER_REG
  485. else
  486. paraloc^.reference.index:=NR_FRAME_POINTER_REG;
  487. varalign:=used_align(size_2_align(paralen),paraalign,paraalign);
  488. paraloc^.reference.offset:=parasize;
  489. parasize:=align(parasize+paralen,varalign);
  490. paralen:=0;
  491. end;
  492. end;
  493. if (locidx<2) and
  494. (loc[locidx+1]<>LOC_INVALID) then
  495. inc(locidx);
  496. end;
  497. end
  498. else
  499. begin
  500. paraloc:=hp.paraloc[side].add_location;
  501. paraloc^.loc:=LOC_VOID;
  502. end;
  503. end;
  504. { Register parameters are assigned from left-to-right, but the
  505. offsets on the stack are right-to-left. There is no need
  506. to reverse the offset, only adapt the calleeside with the
  507. start offset of the first param on the stack }
  508. if side=calleeside then
  509. begin
  510. for i:=0 to paras.count-1 do
  511. begin
  512. hp:=tparavarsym(paras[i]);
  513. with hp.paraloc[side].location^ do
  514. if (loc=LOC_REFERENCE) then
  515. inc(reference.offset,target_info.first_parm_offset);
  516. end;
  517. end;
  518. end;
  519. function tx86_64paramanager.create_varargs_paraloc_info(p : tabstractprocdef; varargspara:tvarargsparalist):longint;
  520. var
  521. intparareg,mmparareg,
  522. parasize : longint;
  523. begin
  524. intparareg:=0;
  525. mmparareg:=0;
  526. if target_info.system=system_x86_64_win64 then
  527. parasize:=4*8
  528. else
  529. parasize:=0;
  530. { calculate the registers for the normal parameters }
  531. create_paraloc_info_intern(p,callerside,p.paras,intparareg,mmparareg,parasize,false);
  532. { append the varargs }
  533. create_paraloc_info_intern(p,callerside,varargspara,intparareg,mmparareg,parasize,true);
  534. { store used no. of SSE registers, that needs to be passed in %AL }
  535. varargspara.mmregsused:=mmparareg;
  536. result:=parasize;
  537. end;
  538. function tx86_64paramanager.create_paraloc_info(p : tabstractprocdef; side: tcallercallee):longint;
  539. var
  540. intparareg,mmparareg,
  541. parasize : longint;
  542. begin
  543. intparareg:=0;
  544. mmparareg:=0;
  545. if target_info.system=system_x86_64_win64 then
  546. parasize:=4*8
  547. else
  548. parasize:=0;
  549. create_paraloc_info_intern(p,side,p.paras,intparareg,mmparareg,parasize,false);
  550. { Create Function result paraloc }
  551. create_funcretloc_info(p,side);
  552. { We need to return the size allocated on the stack }
  553. result:=parasize;
  554. end;
  555. begin
  556. paramanager:=tx86_64paramanager.create;
  557. end.