emit_x86.c 36 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351
  1. /*
  2. * Basic macros to emit x86 instructions and some utils
  3. * Copyright (C) 2008,2009,2010 notaz
  4. * Copyright (C) 2019 kub
  5. *
  6. * This work is licensed under the terms of MAME license.
  7. * See COPYING file in the top-level directory.
  8. *
  9. * note:
  10. * temp registers must be eax-edx due to use of SETcc and r/w 8/16.
  11. * note about silly things like emith_eor_r_r_r:
  12. * these are here because the compiler was designed
  13. * for ARM as it's primary target.
  14. */
  15. #include <stdarg.h>
  16. enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI, // x86-64,i386 common
  17. xR8, xR9, xR10, xR11, xR12, xR13, xR14, xR15 }; // x86-64 only
  18. #define CONTEXT_REG xBP
  19. #define RET_REG xAX
  20. #define ICOND_JO 0x00
  21. #define ICOND_JNO 0x01
  22. #define ICOND_JB 0x02
  23. #define ICOND_JAE 0x03
  24. #define ICOND_JE 0x04
  25. #define ICOND_JNE 0x05
  26. #define ICOND_JBE 0x06
  27. #define ICOND_JA 0x07
  28. #define ICOND_JS 0x08
  29. #define ICOND_JNS 0x09
  30. #define ICOND_JL 0x0c
  31. #define ICOND_JGE 0x0d
  32. #define ICOND_JLE 0x0e
  33. #define ICOND_JG 0x0f
  34. #define IOP_JMP 0xeb
  35. // unified conditions (we just use rel8 jump instructions for x86)
  36. #define DCOND_EQ ICOND_JE
  37. #define DCOND_NE ICOND_JNE
  38. #define DCOND_MI ICOND_JS // MInus
  39. #define DCOND_PL ICOND_JNS // PLus or zero
  40. #define DCOND_HI ICOND_JA // higher (unsigned)
  41. #define DCOND_HS ICOND_JAE // higher || same (unsigned)
  42. #define DCOND_LO ICOND_JB // lower (unsigned)
  43. #define DCOND_LS ICOND_JBE // lower || same (unsigned)
  44. #define DCOND_GE ICOND_JGE // greater || equal (signed)
  45. #define DCOND_GT ICOND_JG // greater (signed)
  46. #define DCOND_LE ICOND_JLE // less || equal (signed)
  47. #define DCOND_LT ICOND_JL // less (signed)
  48. #define DCOND_VS ICOND_JO // oVerflow Set
  49. #define DCOND_VC ICOND_JNO // oVerflow Clear
  50. #define DCOND_CS ICOND_JB // carry set
  51. #define DCOND_CC ICOND_JAE // carry clear
  52. #define EMIT_PTR(ptr, val, type) \
  53. *(type *)(ptr) = val
  54. #define EMIT(val, type) do { \
  55. EMIT_PTR(tcache_ptr, val, type); \
  56. tcache_ptr += sizeof(type); \
  57. } while (0)
  58. #define EMIT_OP(op) do { \
  59. COUNT_OP; \
  60. if ((op) > 0xff) EMIT((op) >> 8, u8); \
  61. EMIT((u8)(op), u8); \
  62. } while (0)
  63. #define EMIT_MODRM(mod, r, rm) do { \
  64. assert((mod) < 4u); \
  65. assert((r) < 8u); \
  66. assert((rm) < 8u); \
  67. EMIT(((mod)<<6) | ((r)<<3) | (rm), u8); \
  68. } while (0)
  69. #define EMIT_SIB(scale, index, base) do { \
  70. assert((scale) < 4u); \
  71. assert((index) < 8u); \
  72. assert((base) < 8u); \
  73. EMIT(((scale)<<6) | ((index)<<3) | (base), u8); \
  74. } while (0)
  75. #define EMIT_SIB64(scale, index, base) \
  76. EMIT_SIB(scale, (index) & ~8u, (base) & ~8u)
  77. #define EMIT_REX(w,r,x,b) \
  78. EMIT(0x40 | ((w)<<3) | ((r)<<2) | ((x)<<1) | (b), u8)
  79. #define EMIT_OP_MODRM(op,mod,r,rm) do { \
  80. EMIT_OP(op); \
  81. EMIT_MODRM(mod, (r), rm); \
  82. } while (0)
  83. // 64bit friendly, rm when everything is converted
  84. #define EMIT_OP_MODRM64(op, mod, r, rm) \
  85. EMIT_OP_MODRM(op, mod, (r) & ~8u, (rm) & ~8u)
  86. #define JMP8_POS(ptr) \
  87. ptr = tcache_ptr; \
  88. tcache_ptr += 2
  89. #define JMP8_EMIT(op, ptr) \
  90. EMIT_PTR(ptr, 0x70|(op), u8); \
  91. EMIT_PTR(ptr + 1, (tcache_ptr - (ptr+2)), u8)
  92. #define JMP8_EMIT_NC(ptr) \
  93. EMIT_PTR(ptr, IOP_JMP, u8); \
  94. EMIT_PTR(ptr + 1, (tcache_ptr - (ptr+2)), u8)
  95. // _r_r
  96. #define emith_move_r_r(dst, src) do {\
  97. EMIT_REX_IF(0, dst, src); \
  98. EMIT_OP_MODRM64(0x8b, 3, dst, src); \
  99. } while (0)
  100. #define emith_move_r_r_ptr(dst, src) do { \
  101. EMIT_REX_IF(1, dst, src); \
  102. EMIT_OP_MODRM64(0x8b, 3, dst, src); \
  103. } while (0)
  104. #define emith_add_r_r(d, s) do { \
  105. EMIT_REX_IF(0, s, d); \
  106. EMIT_OP_MODRM64(0x01, 3, s, d); \
  107. } while (0)
  108. #define emith_add_r_r_ptr(d, s) do { \
  109. EMIT_REX_IF(1, s, d); \
  110. EMIT_OP_MODRM64(0x01, 3, s, d); \
  111. } while (0)
  112. #define emith_sub_r_r(d, s) do {\
  113. EMIT_REX_IF(0, s, d); \
  114. EMIT_OP_MODRM64(0x29, 3, s, d); \
  115. } while (0)
  116. #define emith_adc_r_r(d, s) do { \
  117. EMIT_REX_IF(0, s, d); \
  118. EMIT_OP_MODRM64(0x11, 3, s, d); \
  119. } while (0)
  120. #define emith_sbc_r_r(d, s) do { \
  121. EMIT_REX_IF(0, s, d); \
  122. EMIT_OP_MODRM64(0x19, 3, s, d); /* SBB */ \
  123. } while (0)
  124. #define emith_or_r_r(d, s) do { \
  125. EMIT_REX_IF(0, s, d); \
  126. EMIT_OP_MODRM64(0x09, 3, s, d); \
  127. } while (0)
  128. #define emith_and_r_r(d, s) do { \
  129. EMIT_REX_IF(0, s, d); \
  130. EMIT_OP_MODRM64(0x21, 3, s, d); \
  131. } while (0)
  132. #define emith_eor_r_r(d, s) do { \
  133. EMIT_REX_IF(0, s, d); \
  134. EMIT_OP_MODRM64(0x31, 3, s, d); /* XOR */ \
  135. } while (0)
  136. #define emith_tst_r_r(d, s) do { \
  137. EMIT_REX_IF(0, s, d); \
  138. EMIT_OP_MODRM64(0x85, 3, s, d); /* TEST */ \
  139. } while (0)
  140. #define emith_tst_r_r_ptr(d, s) do { \
  141. EMIT_REX_IF(1, s, d); \
  142. EMIT_OP_MODRM64(0x85, 3, s, d); /* TEST */ \
  143. } while (0)
  144. #define emith_cmp_r_r(d, s) do { \
  145. EMIT_REX_IF(0, s, d); \
  146. EMIT_OP_MODRM64(0x39, 3, s, d); \
  147. } while (0)
  148. // fake teq - test equivalence - get_flags(d ^ s)
  149. #define emith_teq_r_r(d, s) do { \
  150. emith_push(d); \
  151. emith_eor_r_r(d, s); \
  152. emith_pop(d); \
  153. } while (0)
  154. #define emith_mvn_r_r(d, s) do { \
  155. if (d != s) \
  156. emith_move_r_r(d, s); \
  157. EMIT_REX_IF(0, 0, d); \
  158. EMIT_OP_MODRM64(0xf7, 3, 2, d); /* NOT d */ \
  159. } while (0)
  160. #define emith_negc_r_r(d, s) do { \
  161. int tmp_ = rcache_get_tmp(); \
  162. emith_move_r_imm(tmp_, 0); \
  163. emith_sbc_r_r(tmp_, s); \
  164. emith_move_r_r(d, tmp_); \
  165. rcache_free_tmp(tmp_); \
  166. } while (0)
  167. #define emith_neg_r_r(d, s) do { \
  168. if (d != s) \
  169. emith_move_r_r(d, s); \
  170. EMIT_REX_IF(0, 0, d); \
  171. EMIT_OP_MODRM64(0xf7, 3, 3, d); /* NEG d */ \
  172. } while (0)
  173. // _r_r_r
  174. #define emith_add_r_r_r(d, s1, s2) do { \
  175. if (d == s1) { \
  176. emith_add_r_r(d, s2); \
  177. } else if (d == s2) { \
  178. emith_add_r_r(d, s1); \
  179. } else { \
  180. emith_move_r_r(d, s1); \
  181. emith_add_r_r(d, s2); \
  182. } \
  183. } while (0)
  184. #define emith_add_r_r_r_ptr(d, s1, s2) do { \
  185. if (d == s1) { \
  186. emith_add_r_r_ptr(d, s2); \
  187. } else if (d == s2) { \
  188. emith_add_r_r_ptr(d, s1); \
  189. } else { \
  190. emith_move_r_r_ptr(d, s1); \
  191. emith_add_r_r_ptr(d, s2); \
  192. } \
  193. } while (0)
  194. #define emith_sub_r_r_r(d, s1, s2) do { \
  195. if (d == s1) { \
  196. emith_sub_r_r(d, s2); \
  197. } else if (d == s2) { \
  198. emith_sub_r_r(d, s1); \
  199. } else { \
  200. emith_move_r_r(d, s1); \
  201. emith_sub_r_r(d, s2); \
  202. } \
  203. } while (0)
  204. #define emith_adc_r_r_r(d, s1, s2) do { \
  205. if (d == s1) { \
  206. emith_adc_r_r(d, s2); \
  207. } else if (d == s2) { \
  208. emith_adc_r_r(d, s1); \
  209. } else { \
  210. emith_move_r_r(d, s1); \
  211. emith_adc_r_r(d, s2); \
  212. } \
  213. } while (0)
  214. #define emith_sbc_r_r_r(d, s1, s2) do { \
  215. if (d == s1) { \
  216. emith_sbc_r_r(d, s2); \
  217. } else if (d == s2) { \
  218. emith_sbc_r_r(d, s1); \
  219. } else { \
  220. emith_move_r_r(d, s1); \
  221. emith_sbc_r_r(d, s2); \
  222. } \
  223. } while (0)
  224. #define emith_and_r_r_r(d, s1, s2) do { \
  225. if (d == s1) { \
  226. emith_and_r_r(d, s2); \
  227. } else if (d == s2) { \
  228. emith_and_r_r(d, s1); \
  229. } else { \
  230. emith_move_r_r(d, s1); \
  231. emith_and_r_r(d, s2); \
  232. } \
  233. } while (0)
  234. #define emith_or_r_r_r(d, s1, s2) do { \
  235. if (d == s1) { \
  236. emith_or_r_r(d, s2); \
  237. } else if (d == s2) { \
  238. emith_or_r_r(d, s1); \
  239. } else { \
  240. emith_move_r_r(d, s1); \
  241. emith_or_r_r(d, s2); \
  242. } \
  243. } while (0)
  244. #define emith_eor_r_r_r(d, s1, s2) do { \
  245. if (d == s1) { \
  246. emith_eor_r_r(d, s2); \
  247. } else if (d == s2) { \
  248. emith_eor_r_r(d, s1); \
  249. } else { \
  250. emith_move_r_r(d, s1); \
  251. emith_eor_r_r(d, s2); \
  252. } \
  253. } while (0)
  254. // _r_r_r_shift
  255. #define emith_add_r_r_r_lsl(d, s1, s2, lslimm) do { \
  256. int tmp_ = rcache_get_tmp(); \
  257. emith_lsl(tmp_, s2, lslimm); \
  258. emith_add_r_r_r(d, s1, tmp_); \
  259. rcache_free_tmp(tmp_); \
  260. } while (0)
  261. #define emith_add_r_r_r_lsl_ptr(d, s1, s2, lslimm) do { \
  262. int tmp_ = rcache_get_tmp(); \
  263. emith_lsl(tmp_, s2, lslimm); \
  264. emith_add_r_r_r_ptr(d, s1, tmp_); \
  265. rcache_free_tmp(tmp_); \
  266. } while (0)
  267. #define emith_add_r_r_r_lsr(d, s1, s2, lsrimm) do { \
  268. int tmp_ = rcache_get_tmp(); \
  269. emith_lsr(tmp_, s2, lsrimm); \
  270. emith_add_r_r_r(d, s1, tmp_); \
  271. rcache_free_tmp(tmp_); \
  272. } while (0)
  273. #define emith_sub_r_r_r_lsl(d, s1, s2, lslimm) do { \
  274. int tmp_ = rcache_get_tmp(); \
  275. emith_lsl(tmp_, s2, lslimm); \
  276. emith_sub_r_r_r(d, s1, tmp_); \
  277. rcache_free_tmp(tmp_); \
  278. } while (0)
  279. #define emith_or_r_r_r_lsl(d, s1, s2, lslimm) do { \
  280. int tmp_ = rcache_get_tmp(); \
  281. emith_lsl(tmp_, s2, lslimm); \
  282. emith_or_r_r_r(d, s1, tmp_); \
  283. rcache_free_tmp(tmp_); \
  284. } while (0)
  285. // _r_r_shift
  286. #define emith_or_r_r_lsl(d, s, lslimm) do { \
  287. int tmp_ = rcache_get_tmp(); \
  288. emith_lsl(tmp_, s, lslimm); \
  289. emith_or_r_r(d, tmp_); \
  290. rcache_free_tmp(tmp_); \
  291. } while (0)
  292. // d != s
  293. #define emith_eor_r_r_lsr(d, s, lsrimm) do { \
  294. emith_push(s); \
  295. emith_lsr(s, s, lsrimm); \
  296. emith_eor_r_r(d, s); \
  297. emith_pop(s); \
  298. } while (0)
  299. // _r_imm
  300. #define emith_move_r_imm(r, imm) do { \
  301. EMIT_REX_IF(0, 0, r); \
  302. EMIT_OP(0xb8 + ((r)&7)); \
  303. EMIT(imm, u32); \
  304. } while (0)
  305. #define emith_move_r_ptr_imm(r, imm) do { \
  306. if ((uintptr_t)(imm) <= UINT32_MAX) \
  307. emith_move_r_imm(r, (uintptr_t)(imm)); \
  308. else { \
  309. EMIT_REX_IF(1, 0, r); \
  310. EMIT_OP(0xb8 + ((r)&7)); \
  311. EMIT((uintptr_t)(imm), uint64_t); \
  312. } \
  313. } while (0)
  314. #define emith_move_r_imm_s8(r, imm) \
  315. emith_move_r_imm(r, (u32)(signed int)(signed char)(imm))
  316. #define emith_arith_r_imm(op, r, imm) do { \
  317. EMIT_REX_IF(0, 0, r); \
  318. EMIT_OP_MODRM64(0x81, 3, op, r); \
  319. EMIT(imm, u32); \
  320. } while (0)
  321. #define emith_add_r_imm(r, imm) \
  322. emith_arith_r_imm(0, r, imm)
  323. #define emith_or_r_imm(r, imm) \
  324. emith_arith_r_imm(1, r, imm)
  325. #define emith_adc_r_imm(r, imm) \
  326. emith_arith_r_imm(2, r, imm)
  327. #define emith_sbc_r_imm(r, imm) \
  328. emith_arith_r_imm(3, r, imm) // sbb
  329. #define emith_and_r_imm(r, imm) \
  330. emith_arith_r_imm(4, r, imm)
  331. #define emith_sub_r_imm(r, imm) \
  332. emith_arith_r_imm(5, r, imm)
  333. #define emith_eor_r_imm(r, imm) \
  334. emith_arith_r_imm(6, r, imm)
  335. #define emith_cmp_r_imm(r, imm) \
  336. emith_arith_r_imm(7, r, imm)
  337. #define emith_eor_r_imm_ptr(r, imm) do { \
  338. EMIT_REX_IF(1, 0, r); \
  339. EMIT_OP_MODRM64(0x81, 3, 6, r); \
  340. EMIT(imm, u32); \
  341. } while (0)
  342. #define emith_tst_r_imm(r, imm) do { \
  343. EMIT_REX_IF(0, 0, r); \
  344. EMIT_OP_MODRM64(0xf7, 3, 0, r); \
  345. EMIT(imm, u32); \
  346. } while (0)
  347. // fake
  348. #define emith_bic_r_imm(r, imm) \
  349. emith_arith_r_imm(4, r, ~(imm))
  350. // fake conditionals (using SJMP instead)
  351. #define emith_move_r_imm_c(cond, r, imm) \
  352. emith_move_r_imm(r, imm);
  353. #define emith_add_r_imm_c(cond, r, imm) \
  354. emith_add_r_imm(r, imm);
  355. #define emith_sub_r_imm_c(cond, r, imm) \
  356. emith_sub_r_imm(r, imm);
  357. #define emith_or_r_imm_c(cond, r, imm) \
  358. emith_or_r_imm(r, imm)
  359. #define emith_eor_r_imm_c(cond, r, imm) \
  360. emith_eor_r_imm(r, imm)
  361. #define emith_eor_r_imm_ptr_c(cond, r, imm) \
  362. emith_eor_r_imm_ptr(r, imm)
  363. #define emith_bic_r_imm_c(cond, r, imm) \
  364. emith_bic_r_imm(r, imm)
  365. #define emith_tst_r_imm_c(cond, r, imm) \
  366. emith_tst_r_imm(r, imm)
  367. #define emith_move_r_r_ptr_c(cond, d, s) \
  368. emith_move_r_r_ptr(d, s)
  369. #define emith_ror_c(cond, d, s, cnt) \
  370. emith_ror(d, s, cnt)
  371. #define emith_and_r_r_c(cond, d, s) \
  372. emith_and_r_r(d, s);
  373. #define emith_add_r_r_imm_c(cond, d, s, imm) \
  374. emith_add_r_r_imm(d, s, imm);
  375. #define emith_sub_r_r_imm_c(cond, d, s, imm) \
  376. emith_sub_r_r_imm(d, s, imm);
  377. #define emith_read8_r_r_r_c(cond, r, rs, rm) \
  378. emith_read8_r_r_r(r, rs, rm)
  379. #define emith_read8s_r_r_r_c(cond, r, rs, rm) \
  380. emith_read8s_r_r_r(r, rs, rm)
  381. #define emith_read16_r_r_r_c(cond, r, rs, rm) \
  382. emith_read16_r_r_r(r, rs, rm)
  383. #define emith_read16s_r_r_r_c(cond, r, rs, rm) \
  384. emith_read16s_r_r_r(r, rs, rm)
  385. #define emith_read_r_r_r_c(cond, r, rs, rm) \
  386. emith_read_r_r_r(r, rs, rm)
  387. #define emith_read_r_r_offs_c(cond, r, rs, offs) \
  388. emith_read_r_r_offs(r, rs, offs)
  389. #define emith_read_r_r_offs_ptr_c(cond, r, rs, offs) \
  390. emith_read_r_r_offs_ptr(r, rs, offs)
  391. #define emith_write_r_r_offs_c(cond, r, rs, offs) \
  392. emith_write_r_r_offs(r, rs, offs)
  393. #define emith_write_r_r_offs_ptr_c(cond, r, rs, offs) \
  394. emith_write_r_r_offs_ptr(r, rs, offs)
  395. #define emith_read8_r_r_offs_c(cond, r, rs, offs) \
  396. emith_read8_r_r_offs(r, rs, offs)
  397. #define emith_write8_r_r_offs_c(cond, r, rs, offs) \
  398. emith_write8_r_r_offs(r, rs, offs)
  399. #define emith_read16_r_r_offs_c(cond, r, rs, offs) \
  400. emith_read16_r_r_offs(r, rs, offs)
  401. #define emith_write16_r_r_offs_c(cond, r, rs, offs) \
  402. emith_write16_r_r_offs(r, rs, offs)
  403. #define emith_jump_reg_c(cond, r) \
  404. emith_jump_reg(r)
  405. #define emith_jump_ctx_c(cond, offs) \
  406. emith_jump_ctx(offs)
  407. #define emith_ret_c(cond) \
  408. emith_ret()
  409. // _r_r_imm - use lea
  410. #define emith_add_r_r_imm(d, s, imm) do { \
  411. EMIT_REX_IF(0, d, s); \
  412. emith_deref_modrm(0x8d, 2, d, s); \
  413. EMIT(imm, s32); \
  414. } while (0)
  415. #define emith_add_r_r_ptr_imm(d, s, imm) do { \
  416. EMIT_REX_IF(1, d, s); \
  417. emith_deref_modrm(0x8d, 2, d, s); \
  418. EMIT(imm, s32); \
  419. } while (0)
  420. #define emith_sub_r_r_imm(d, s, imm) do { \
  421. if (d != s) \
  422. emith_move_r_r(d, s); \
  423. if ((s32)(imm) != 0) \
  424. emith_sub_r_imm(d, imm); \
  425. } while (0)
  426. #define emith_and_r_r_imm(d, s, imm) do { \
  427. if (d != s) \
  428. emith_move_r_r(d, s); \
  429. if ((s32)(imm) != -1) \
  430. emith_and_r_imm(d, imm); \
  431. } while (0)
  432. #define emith_or_r_r_imm(d, s, imm) do { \
  433. if (d != s) \
  434. emith_move_r_r(d, s); \
  435. if ((s32)(imm) != 0) \
  436. emith_or_r_imm(d, imm); \
  437. } while (0)
  438. #define emith_eor_r_r_imm(d, s, imm) do { \
  439. if (d != s) \
  440. emith_move_r_r(d, s); \
  441. if ((s32)(imm) != 0) \
  442. emith_eor_r_imm(d, imm); \
  443. } while (0)
  444. // shift
  445. #define emith_shift(op, d, s, cnt) do { \
  446. if (d != s) \
  447. emith_move_r_r(d, s); \
  448. EMIT_REX_IF(0, 0, d); \
  449. EMIT_OP_MODRM64(0xc1, 3, op, d); \
  450. EMIT(cnt, u8); \
  451. } while (0)
  452. #define emith_lsl(d, s, cnt) \
  453. emith_shift(4, d, s, cnt)
  454. #define emith_lsr(d, s, cnt) \
  455. emith_shift(5, d, s, cnt)
  456. #define emith_asr(d, s, cnt) \
  457. emith_shift(7, d, s, cnt)
  458. #define emith_rol(d, s, cnt) \
  459. emith_shift(0, d, s, cnt)
  460. #define emith_ror(d, s, cnt) \
  461. emith_shift(1, d, s, cnt)
  462. #define emith_rolc(r) do { \
  463. EMIT_REX_IF(0, 0, r); \
  464. EMIT_OP_MODRM64(0xd1, 3, 2, r); \
  465. } while (0)
  466. #define emith_rorc(r) do { \
  467. EMIT_REX_IF(0, 0, r); \
  468. EMIT_OP_MODRM64(0xd1, 3, 3, r); \
  469. } while (0)
  470. // misc
  471. #define emith_push(r) do { \
  472. EMIT_REX_IF(0, 0, r); \
  473. EMIT_OP(0x50 + ((r)&7)); \
  474. } while (0)
  475. #define emith_push_imm(imm) do { \
  476. EMIT_OP(0x68); \
  477. EMIT(imm, u32); \
  478. } while (0)
  479. #define emith_pop(r) do { \
  480. EMIT_REX_IF(0, 0, r); \
  481. EMIT_OP(0x58 + ((r)&7)); \
  482. } while (0)
  483. #define emith_neg_r(r) do { \
  484. EMIT_REX_IF(0, 0, r); \
  485. EMIT_OP_MODRM64(0xf7, 3, 3, r); \
  486. } while (0)
  487. #define emith_clear_msb(d, s, count) do { \
  488. u32 t = (u32)-1; \
  489. t >>= count; \
  490. if (d != s) \
  491. emith_move_r_r(d, s); \
  492. emith_and_r_imm(d, t); \
  493. } while (0)
  494. #define emith_clear_msb_c(cond, d, s, count) do { \
  495. (void)(cond); \
  496. emith_clear_msb(d, s, count); \
  497. } while (0)
  498. #define emith_sext(d, s, bits) do { \
  499. emith_lsl(d, s, 32 - (bits)); \
  500. emith_asr(d, d, 32 - (bits)); \
  501. } while (0)
  502. #define emith_setc(r) do { \
  503. assert(is_abcdx(r)); \
  504. EMIT_REX_IF(0, 0, r); \
  505. EMIT_OP_MODRM64(0x0f92, 3, 0, r); /* SETC r */ \
  506. } while (0)
  507. // XXX: stupid mess
  508. #define emith_mul_(op, dlo, dhi, s1, s2) do { \
  509. int rmr; \
  510. if (dlo != xAX && dhi != xAX && rcache_is_hreg_used(xAX)) \
  511. emith_push(xAX); \
  512. if (dlo != xDX && dhi != xDX && rcache_is_hreg_used(xDX)) \
  513. emith_push(xDX); \
  514. if ((s1) == xAX) \
  515. rmr = s2; \
  516. else if ((s2) == xAX) \
  517. rmr = s1; \
  518. else { \
  519. emith_move_r_r(xAX, s1); \
  520. rmr = s2; \
  521. } \
  522. EMIT_REX_IF(0, 0, rmr); \
  523. EMIT_OP_MODRM64(0xf7, 3, op, rmr); /* xMUL rmr */ \
  524. if (dlo != xAX) { \
  525. EMIT_REX_IF(0, 0, dlo); \
  526. EMIT_OP(0x90 + ((dlo)&7)); /* XCHG eax, dlo */ \
  527. } \
  528. if (dhi != xDX && dhi != -1 && !(dhi == xAX && dlo == xDX)) \
  529. emith_move_r_r(dhi, (dlo == xDX ? xAX : xDX)); \
  530. if (dlo != xDX && dhi != xDX && rcache_is_hreg_used(xDX)) \
  531. emith_pop(xDX); \
  532. if (dlo != xAX && dhi != xAX && rcache_is_hreg_used(xAX)) \
  533. emith_pop(xAX); \
  534. } while (0)
  535. #define emith_mul_u64(dlo, dhi, s1, s2) \
  536. emith_mul_(4, dlo, dhi, s1, s2) /* MUL */
  537. #define emith_mul_s64(dlo, dhi, s1, s2) \
  538. emith_mul_(5, dlo, dhi, s1, s2) /* IMUL */
  539. #define emith_mul(d, s1, s2) do { \
  540. if (d == s1) { \
  541. EMIT_REX_IF(0, d, s2); \
  542. EMIT_OP_MODRM64(0x0faf, 3, d, s2); \
  543. } else if (d == s2) { \
  544. EMIT_REX_IF(0, d, s1); \
  545. EMIT_OP_MODRM64(0x0faf, 3, d, s1); \
  546. } else { \
  547. emith_move_r_r(d, s1); \
  548. EMIT_REX_IF(0, d, s2); \
  549. EMIT_OP_MODRM64(0x0faf, 3, d, s2); \
  550. } \
  551. } while (0)
  552. // (dlo,dhi) += signed(s1) * signed(s2)
  553. #define emith_mula_s64(dlo, dhi, s1, s2) do { \
  554. emith_push(dhi); \
  555. emith_push(dlo); \
  556. emith_mul_(5, dlo, dhi, s1, s2); \
  557. EMIT_REX_IF(0, dlo, xSP); \
  558. emith_deref_modrm(0x03, 0, dlo, xSP); /* add dlo, [xsp] */ \
  559. EMIT_REX_IF(0, dhi, xSP); \
  560. emith_deref_modrm(0x13, 1, dhi, xSP); /* adc dhi, [xsp+{4,8}] */ \
  561. EMIT(sizeof(void *), u8); \
  562. emith_add_r_r_ptr_imm(xSP, xSP, sizeof(void *) * 2); \
  563. } while (0)
  564. // "flag" instructions are the same
  565. #define emith_adcf_r_imm emith_adc_r_imm
  566. #define emith_subf_r_imm emith_sub_r_imm
  567. #define emith_addf_r_r emith_add_r_r
  568. #define emith_subf_r_r emith_sub_r_r
  569. #define emith_adcf_r_r emith_adc_r_r
  570. #define emith_sbcf_r_r emith_sbc_r_r
  571. #define emith_eorf_r_r emith_eor_r_r
  572. #define emith_negcf_r_r emith_negc_r_r
  573. #define emith_subf_r_r_imm emith_sub_r_r_imm
  574. #define emith_addf_r_r_r emith_add_r_r_r
  575. #define emith_subf_r_r_r emith_sub_r_r_r
  576. #define emith_adcf_r_r_r emith_adc_r_r_r
  577. #define emith_sbcf_r_r_r emith_sbc_r_r_r
  578. #define emith_eorf_r_r_r emith_eor_r_r_r
  579. #define emith_addf_r_r_r_lsr emith_add_r_r_r_lsr
  580. #define emith_lslf emith_lsl
  581. #define emith_lsrf emith_lsr
  582. #define emith_asrf emith_asr
  583. #define emith_rolf emith_rol
  584. #define emith_rorf emith_ror
  585. #define emith_rolcf emith_rolc
  586. #define emith_rorcf emith_rorc
  587. #define emith_deref_modrm(op, m, r, rs) do { \
  588. if (((rs) & 7) == 5 && m == 0) { /* xBP,xR13 not in mod 0, use mod 1 */\
  589. EMIT_OP_MODRM64(op, 1, r, rs); \
  590. EMIT(0, u8); \
  591. } else if (((rs) & 7) == 4) { /* xSP,xR12 must use SIB */ \
  592. EMIT_OP_MODRM64(op, m, r, 4); \
  593. EMIT_SIB64(0, 4, rs); \
  594. } else \
  595. EMIT_OP_MODRM64(op, m, r, rs); \
  596. } while (0)
  597. #define emith_deref_op(op, r, rs, offs) do { \
  598. /* mov r <-> [ebp+#offs] */ \
  599. if ((offs) == 0) { \
  600. emith_deref_modrm(op, 0, r, rs); \
  601. } else if (abs(offs) >= 0x80) { \
  602. emith_deref_modrm(op, 2, r, rs); \
  603. EMIT(offs, u32); \
  604. } else { \
  605. emith_deref_modrm(op, 1, r, rs); \
  606. EMIT((u8)offs, u8); \
  607. } \
  608. } while (0)
  609. #define is_abcdx(r) !((r) & ~0x3)
  610. #define emith_read_r_r_offs(r, rs, offs) do { \
  611. EMIT_REX_IF(0, r, rs); \
  612. emith_deref_op(0x8b, r, rs, offs); \
  613. } while (0)
  614. #define emith_read_r_r_offs_ptr(r, rs, offs) do { \
  615. EMIT_REX_IF(1, r, rs); \
  616. emith_deref_op(0x8b, r, rs, offs); \
  617. } while (0)
  618. #define emith_write_r_r_offs(r, rs, offs) do { \
  619. EMIT_REX_IF(0, r, rs); \
  620. emith_deref_op(0x89, r, rs, offs); \
  621. } while (0)
  622. #define emith_write_r_r_offs_ptr(r, rs, offs) do { \
  623. EMIT_REX_IF(1, r, rs); \
  624. emith_deref_op(0x89, r, rs, offs); \
  625. } while (0)
  626. #define emith_read8_r_r_offs(r, rs, offs) do { \
  627. EMIT_REX_IF(0, r, rs); \
  628. emith_deref_op(0x0fb6, r, rs, offs); \
  629. } while (0)
  630. #define emith_read8s_r_r_offs(r, rs, offs) do { \
  631. EMIT_REX_IF(0, r, rs); \
  632. emith_deref_op(0x0fbe, r, rs, offs); \
  633. } while (0)
  634. #define emith_write8_r_r_offs(r, rs, offs) do {\
  635. EMIT_REX_IF(0, r, rs); \
  636. emith_deref_op(0x88, r, rs, offs); \
  637. } while (0)
  638. #define emith_read16_r_r_offs(r, rs, offs) do { \
  639. EMIT_REX_IF(0, r, rs); \
  640. emith_deref_op(0x0fb7, r, rs, offs); \
  641. } while (0)
  642. #define emith_read16s_r_r_offs(r, rs, offs) do { \
  643. EMIT_REX_IF(0, r, rs); \
  644. emith_deref_op(0x0fbf, r, rs, offs); \
  645. } while (0)
  646. #define emith_write16_r_r_offs(r, rs, offs) do { \
  647. EMIT(0x66, u8); /* Intel SDM Vol 2a: REX must be closest to opcode */ \
  648. EMIT_REX_IF(0, r, rs); \
  649. emith_deref_op(0x89, r, rs, offs); \
  650. } while (0)
  651. #define emith_read8_r_r_r(r, rs, rm) do { \
  652. EMIT_XREX_IF(0, r, rm, rs); \
  653. EMIT_OP_MODRM64(0x0fb6, 0, r, 4); \
  654. EMIT_SIB64(0, rs, rm); /* mov r, [rm + rs * 1] */ \
  655. } while (0)
  656. #define emith_read8s_r_r_r(r, rs, rm) do { \
  657. EMIT_XREX_IF(0, r, rm, rs); \
  658. EMIT_OP_MODRM64(0x0fbe, 0, r, 4); \
  659. EMIT_SIB64(0, rs, rm); /* mov r, [rm + rs * 1] */ \
  660. } while (0)
  661. #define emith_read16_r_r_r(r, rs, rm) do { \
  662. EMIT_XREX_IF(0, r, rm, rs); \
  663. EMIT_OP_MODRM64(0x0fb7, 0, r, 4); \
  664. EMIT_SIB64(0, rs, rm); /* mov r, [rm + rs * 1] */ \
  665. } while (0)
  666. #define emith_read16s_r_r_r(r, rs, rm) do { \
  667. EMIT_XREX_IF(0, r, rm, rs); \
  668. EMIT_OP_MODRM64(0x0fbf, 0, r, 4); \
  669. EMIT_SIB64(0, rs, rm); /* mov r, [rm + rs * 1] */ \
  670. } while (0)
  671. #define emith_read_r_r_r(r, rs, rm) do { \
  672. EMIT_XREX_IF(0, r, rm, rs); \
  673. EMIT_OP_MODRM64(0x8b, 0, r, 4); \
  674. EMIT_SIB64(0, rs, rm); /* mov r, [rm + rs * 1] */ \
  675. } while (0)
  676. #define emith_read_r_r_r_ptr(r, rs, rm) do { \
  677. EMIT_XREX_IF(1, r, rm, rs); \
  678. EMIT_OP_MODRM64(0x8b, 0, r, 4); \
  679. EMIT_SIB64(0, rs, rm); /* mov r, [rm + rs * 1] */ \
  680. } while (0)
  681. #define emith_read_r_r_r_wb(r, rs, rm) do { \
  682. emith_read_r_r_r(r, rs, rm); \
  683. emith_add_r_r_ptr(rs, rm); \
  684. } while (0)
  685. #define emith_read_r_r_r_ptr_wb(r, rs, rm) do { \
  686. emith_read_r_r_r_ptr(r, rs, rm); \
  687. emith_add_r_r_ptr(rs, rm); \
  688. } while (0)
  689. #define emith_write_r_r_r(r, rs, rm) do { \
  690. EMIT_XREX_IF(0, r, rm, rs); \
  691. EMIT_OP_MODRM64(0x89, 0, r, 4); \
  692. EMIT_SIB64(0, rs, rm); /* mov [rm + rs * 1], r */ \
  693. } while (0)
  694. #define emith_write_r_r_r_ptr(r, rs, rm) do { \
  695. EMIT_XREX_IF(1, r, rm, rs); \
  696. EMIT_OP_MODRM64(0x89, 0, r, 4); \
  697. EMIT_SIB64(0, rs, rm); /* mov [rm + rs * 1], r */ \
  698. } while (0)
  699. #define emith_write_r_r_r_wb(r, rs, rm) do { \
  700. emith_write_r_r_r(r, rs, rm); \
  701. emith_add_r_r_ptr(rs, rm); \
  702. } while (0)
  703. #define emith_write_r_r_r_ptr_wb(r, rs, rm) do { \
  704. emith_write_r_r_r_ptr(r, rs, rm); \
  705. emith_add_r_r_ptr(rs, rm); \
  706. } while (0)
  707. #define emith_ctx_read(r, offs) \
  708. emith_read_r_r_offs(r, CONTEXT_REG, offs)
  709. #define emith_ctx_read_c(cond, r, offs) \
  710. emith_ctx_read(r, offs)
  711. #define emith_ctx_read_ptr(r, offs) do { \
  712. EMIT_REX_IF(1, r, CONTEXT_REG); \
  713. emith_deref_op(0x8b, r, CONTEXT_REG, offs); \
  714. } while (0)
  715. #define emith_ctx_write(r, offs) \
  716. emith_write_r_r_offs(r, CONTEXT_REG, offs)
  717. #define emith_ctx_read_multiple(r, offs, cnt, tmpr) do { \
  718. int r_ = r, offs_ = offs, cnt_ = cnt; \
  719. for (; cnt_ > 0; r_++, offs_ += 4, cnt_--) \
  720. emith_ctx_read(r_, offs_); \
  721. } while (0)
  722. #define emith_ctx_write_multiple(r, offs, cnt, tmpr) do { \
  723. int r_ = r, offs_ = offs, cnt_ = cnt; \
  724. for (; cnt_ > 0; r_++, offs_ += 4, cnt_--) \
  725. emith_ctx_write(r_, offs_); \
  726. } while (0)
  727. // assumes EBX is free
  728. #define emith_ret_to_ctx(offs) do { \
  729. emith_pop(xBX); \
  730. emith_ctx_write(xBX, offs); \
  731. } while (0)
  732. #define emith_jump(ptr) do { \
  733. u32 disp = (u8 *)(ptr) - ((u8 *)tcache_ptr + 5); \
  734. EMIT_OP(0xe9); \
  735. EMIT(disp, u32); \
  736. } while (0)
  737. #define emith_jump_patchable(target) \
  738. emith_jump(target)
  739. #define emith_jump_cond(cond, ptr) do { \
  740. u32 disp = (u8 *)(ptr) - ((u8 *)tcache_ptr + 6); \
  741. EMIT_OP(0x0f80 | (cond)); \
  742. EMIT(disp, u32); \
  743. } while (0)
  744. #define emith_jump_cond_patchable(cond, target) \
  745. emith_jump_cond(cond, target)
  746. #define emith_jump_patch(ptr, target) ({ \
  747. u32 disp_ = (u8 *)(target) - ((u8 *)(ptr) + 4); \
  748. u32 offs_ = (*(u8 *)(ptr) == 0x0f) ? 2 : 1; \
  749. EMIT_PTR((u8 *)(ptr) + offs_, disp_ - offs_, u32); \
  750. ptr; \
  751. })
  752. #define emith_jump_patch_size() 6
  753. #define emith_jump_at(ptr, target) do { \
  754. u32 disp_ = (u8 *)(target) - ((u8 *)(ptr) + 5); \
  755. EMIT_PTR(ptr, 0xe9, u8); \
  756. EMIT_PTR((u8 *)(ptr) + 1, disp_, u32); \
  757. } while (0)
  758. #define emith_call(ptr) do { \
  759. u32 disp = (u8 *)(ptr) - ((u8 *)tcache_ptr + 5); \
  760. EMIT_OP(0xe8); \
  761. EMIT(disp, u32); \
  762. } while (0)
  763. #define emith_call_cond(cond, ptr) \
  764. emith_call(ptr)
  765. #define emith_call_reg(r) \
  766. EMIT_OP_MODRM(0xff, 3, 2, r)
  767. #define emith_call_ctx(offs) do { \
  768. EMIT_OP_MODRM(0xff, 2, 2, CONTEXT_REG); \
  769. EMIT(offs, u32); \
  770. } while (0)
  771. #define emith_call_link(r, target) do { \
  772. EMIT_OP(0xe8); \
  773. EMIT(0, u32); /* call pc+0 */ \
  774. emith_pop(r); \
  775. emith_add_r_r_ptr_imm(r, r, 13); \
  776. emith_jump(target); \
  777. } while (0)
  778. #define emith_call_cleanup() \
  779. emith_add_r_r_ptr_imm(xSP, xSP, sizeof(void *)); // remove return addr
  780. #define emith_ret() \
  781. EMIT_OP(0xc3)
  782. #define emith_jump_reg(r) \
  783. EMIT_OP_MODRM(0xff, 3, 4, r)
  784. #define emith_jump_ctx(offs) do { \
  785. EMIT_OP_MODRM(0xff, 2, 4, CONTEXT_REG); \
  786. EMIT(offs, u32); \
  787. } while (0)
  788. #define emith_push_ret(r) do { \
  789. int r_ = (r >= 0 ? r : xSI); \
  790. emith_push(r_); /* always push to align */ \
  791. emith_add_r_r_ptr_imm(xSP, xSP, -8*4); /* args shadow space */ \
  792. } while (0)
  793. #define emith_pop_and_ret(r) do { \
  794. int r_ = (r >= 0 ? r : xSI); \
  795. emith_add_r_r_ptr_imm(xSP, xSP, 8*4); /* args shadow space */ \
  796. emith_pop(r_); \
  797. emith_ret(); \
  798. } while (0)
  799. #define EMITH_JMP_START(cond) { \
  800. u8 *cond_ptr; \
  801. JMP8_POS(cond_ptr)
  802. #define EMITH_JMP_END(cond) \
  803. JMP8_EMIT(cond, cond_ptr); \
  804. }
  805. #define EMITH_JMP3_START(cond) { \
  806. u8 *cond_ptr, *else_ptr; \
  807. JMP8_POS(cond_ptr)
  808. #define EMITH_JMP3_MID(cond) \
  809. JMP8_POS(else_ptr); \
  810. JMP8_EMIT(cond, cond_ptr);
  811. #define EMITH_JMP3_END() \
  812. JMP8_EMIT_NC(else_ptr); \
  813. }
  814. // "simple" jump (no more then a few insns)
  815. // ARM will use conditional instructions here
  816. #define EMITH_SJMP_START EMITH_JMP_START
  817. #define EMITH_SJMP_END EMITH_JMP_END
  818. #define EMITH_SJMP3_START EMITH_JMP3_START
  819. #define EMITH_SJMP3_MID EMITH_JMP3_MID
  820. #define EMITH_SJMP3_END EMITH_JMP3_END
  821. #define EMITH_SJMP2_START(cond) \
  822. EMITH_SJMP3_START(cond)
  823. #define EMITH_SJMP2_MID(cond) \
  824. EMITH_SJMP3_MID(cond)
  825. #define EMITH_SJMP2_END(cond) \
  826. EMITH_SJMP3_END()
  827. #define emith_pass_arg_r(arg, reg) do { \
  828. int rd = 7; \
  829. host_arg2reg(rd, arg); \
  830. emith_move_r_r_ptr(rd, reg); \
  831. } while (0)
  832. #define emith_pass_arg_imm(arg, imm) do { \
  833. int rd = 7; \
  834. host_arg2reg(rd, arg); \
  835. emith_move_r_imm(rd, imm); \
  836. } while (0)
  837. #define host_instructions_updated(base, end)
  838. #define emith_rw_offs_max() 0xffffffff
  839. #ifdef __x86_64__
  840. #define HOST_REGS 16
  841. #define PTR_SCALE 3
  842. #define NA_TMP_REG xAX // non-arg tmp from reg_temp[]
  843. #define EMIT_XREX_IF(w, r, rm, rs) do { \
  844. int xr_ = (r) > 7 ? 1 : 0; \
  845. int xb_ = (rm) > 7 ? 1 : 0; \
  846. int xx_ = (rs) > 7 ? 1 : 0; \
  847. if ((w) | xr_ | xx_ | xb_) \
  848. EMIT_REX(w, xr_, xx_, xb_); \
  849. } while (0)
  850. #define EMIT_REX_IF(w, r, rm) \
  851. EMIT_XREX_IF(w, r, rm, 0)
  852. #ifndef _WIN32
  853. #define host_arg2reg(rd, arg) \
  854. switch (arg) { \
  855. case 0: rd = xDI; break; \
  856. case 1: rd = xSI; break; \
  857. case 2: rd = xDX; break; \
  858. default: rd = xCX; break; \
  859. }
  860. #define emith_sh2_drc_entry() do { \
  861. emith_push(xBX); \
  862. emith_push(xBP); \
  863. emith_push(xR12); \
  864. emith_push(xR13); \
  865. emith_push(xR14); \
  866. emith_push(xR15); \
  867. emith_push(xSI); /* to align */ \
  868. } while (0)
  869. #define emith_sh2_drc_exit() do { \
  870. emith_pop(xSI); \
  871. emith_pop(xR15); \
  872. emith_pop(xR14); \
  873. emith_pop(xR13); \
  874. emith_pop(xR12); \
  875. emith_pop(xBP); \
  876. emith_pop(xBX); \
  877. emith_ret(); \
  878. } while (0)
  879. #else // _WIN32
  880. #define host_arg2reg(rd, arg) \
  881. switch (arg) { \
  882. case 0: rd = xCX; break; \
  883. case 1: rd = xDX; break; \
  884. case 2: rd = xR8; break; \
  885. default: rd = xR9; break; \
  886. }
  887. #define emith_sh2_drc_entry() do { \
  888. emith_push(xBX); \
  889. emith_push(xBP); \
  890. emith_push(xR12); \
  891. emith_push(xR13); \
  892. emith_push(xR14); \
  893. emith_push(xR15); \
  894. emith_push(xSI); \
  895. emith_push(xDI); \
  896. emith_add_r_r_ptr_imm(xSP, xSP, -8*5); /* align + args shadow space */ \
  897. } while (0)
  898. #define emith_sh2_drc_exit() do { \
  899. emith_add_r_r_ptr_imm(xSP, xSP, 8*5); \
  900. emith_pop(xDI); \
  901. emith_pop(xSI); \
  902. emith_pop(xR15); \
  903. emith_pop(xR14); \
  904. emith_pop(xR13); \
  905. emith_pop(xR12); \
  906. emith_pop(xBP); \
  907. emith_pop(xBX); \
  908. emith_ret(); \
  909. } while (0)
  910. #endif // _WIN32
  911. #else // !__x86_64__
  912. #define HOST_REGS 8
  913. #define PTR_SCALE 2
  914. #define NA_TMP_REG xBX // non-arg tmp from reg_temp[]
  915. #define EMIT_REX_IF(w, r, rm) do { \
  916. assert((u32)(r) < 8u); \
  917. assert((u32)(rm) < 8u); \
  918. } while (0)
  919. #define EMIT_XREX_IF(w, r, rs, rm) do { \
  920. assert((u32)(r) < 8u); \
  921. assert((u32)(rs) < 8u); \
  922. assert((u32)(rm) < 8u); \
  923. } while (0)
  924. #define host_arg2reg(rd, arg) \
  925. switch (arg) { \
  926. case 0: rd = xAX; break; \
  927. case 1: rd = xDX; break; \
  928. case 2: rd = xCX; break; \
  929. default: rd = xBX; break; \
  930. }
  931. #define emith_sh2_drc_entry() do { \
  932. emith_push(xBX); \
  933. emith_push(xBP); \
  934. emith_push(xSI); \
  935. emith_push(xDI); \
  936. } while (0)
  937. #define emith_sh2_drc_exit() do { \
  938. emith_pop(xDI); \
  939. emith_pop(xSI); \
  940. emith_pop(xBP); \
  941. emith_pop(xBX); \
  942. emith_ret(); \
  943. } while (0)
  944. #endif
  945. #define emith_save_caller_regs(mask) do { \
  946. int _c; u32 _m = mask & 0xfc7; /* AX, CX, DX, SI, DI, 8, 9, 10, 11 */ \
  947. if (__builtin_parity(_m) == 1) _m |= 0x8; /* BX for ABI align */ \
  948. for (_c = HOST_REGS; _m && _c >= 0; _m &= ~(1 << _c), _c--) \
  949. if (_m & (1 << _c)) emith_push(_c); \
  950. } while (0)
  951. #define emith_restore_caller_regs(mask) do { \
  952. int _c; u32 _m = mask & 0xfc7; \
  953. if (__builtin_parity(_m) == 1) _m |= 0x8; /* BX for ABI align */ \
  954. for (_c = 0; _m && _c < HOST_REGS; _m &= ~(1 << _c), _c++) \
  955. if (_m & (1 << _c)) emith_pop(_c); \
  956. } while (0)
  957. #define emith_sh2_rcall(a, tab, func, mask) do { \
  958. emith_lsr(mask, a, SH2_READ_SHIFT); \
  959. EMIT_XREX_IF(1, tab, tab, mask); \
  960. EMIT_OP_MODRM64(0x8d, 0, tab, 4); \
  961. EMIT_SIB64(PTR_SCALE, mask, tab); /* lea tab, [tab + mask * {4,8}] */ \
  962. EMIT_XREX_IF(1, tab, tab, mask); \
  963. EMIT_OP_MODRM64(0x8d, 0, tab, 4); \
  964. EMIT_SIB64(PTR_SCALE, mask, tab); /* lea tab, [tab + mask * {4,8}] */ \
  965. EMIT_REX_IF(1, func, tab); \
  966. emith_deref_modrm(0x8b, 0, func, tab); /* mov func, [tab] */ \
  967. EMIT_REX_IF(0, mask, tab); \
  968. emith_deref_modrm(0x8b, 1, mask, tab); \
  969. EMIT(1 << PTR_SCALE, u8); /* mov mask, [tab + {4,8}] */ \
  970. emith_add_r_r_ptr(func, func); \
  971. } while (0)
  972. #define emith_sh2_wcall(a, val, tab, func) do { \
  973. int arg2_; \
  974. host_arg2reg(arg2_, 2); \
  975. emith_lsr(func, a, SH2_WRITE_SHIFT); /* tmp = a >> WRT_SHIFT */ \
  976. EMIT_XREX_IF(1, func, tab, func); \
  977. EMIT_OP_MODRM64(0x8b, 0, func, 4); \
  978. EMIT_SIB64(PTR_SCALE, func, tab); /* mov tmp, [tab + tmp * {4,8}] */ \
  979. emith_move_r_r_ptr(arg2_, CONTEXT_REG); \
  980. emith_jump_reg(func); \
  981. } while (0)
  982. #define emith_sh2_dtbf_loop() do { \
  983. u8 *jmp0; /* negative cycles check */ \
  984. u8 *jmp1; /* unsinged overflow check */ \
  985. int cr, rn; \
  986. int tmp_ = rcache_get_tmp(); \
  987. cr = rcache_get_reg(SHR_SR, RC_GR_RMW); \
  988. rn = rcache_get_reg((op >> 8) & 0x0f, RC_GR_RMW);\
  989. emith_sub_r_imm(rn, 1); \
  990. emith_sub_r_imm(cr, (cycles+1) << 12); \
  991. cycles = 0; \
  992. emith_asr(tmp_, cr, 2+12); \
  993. JMP8_POS(jmp0); /* no negative cycles */ \
  994. emith_move_r_imm(tmp_, 0); \
  995. JMP8_EMIT(ICOND_JNS, jmp0); \
  996. emith_and_r_imm(cr, 0xffe); \
  997. emith_subf_r_r(rn, tmp_); \
  998. JMP8_POS(jmp1); /* no overflow */ \
  999. emith_neg_r(rn); /* count left */ \
  1000. emith_lsl(rn, rn, 2+12); \
  1001. emith_or_r_r(cr, rn); \
  1002. emith_or_r_imm(cr, 1); \
  1003. emith_move_r_imm(rn, 0); \
  1004. JMP8_EMIT(ICOND_JA, jmp1); \
  1005. rcache_free_tmp(tmp_); \
  1006. } while (0)
  1007. #define emith_sh2_delay_loop(cycles, reg) do { \
  1008. int sr = rcache_get_reg(SHR_SR, RC_GR_RMW, NULL); \
  1009. int t1 = rcache_get_tmp(); \
  1010. int t2 = rcache_get_tmp(); \
  1011. int t3 = rcache_get_tmp(); \
  1012. if (t3 == xAX) { t3 = t1; t1 = xAX; } /* for MUL */ \
  1013. if (t3 == xDX) { t3 = t2; t2 = xDX; } \
  1014. /* if (sr < 0) return */ \
  1015. emith_asrf(t2, sr, 12); \
  1016. EMITH_JMP_START(DCOND_LE); \
  1017. /* turns = sr.cycles / cycles */ \
  1018. emith_move_r_imm(t3, (u32)((1ULL<<32) / (cycles)) + 1); \
  1019. emith_mul_u64(t1, t2, t2, t3); /* multiply by 1/x */ \
  1020. rcache_free_tmp(t3); \
  1021. if (reg >= 0) { \
  1022. /* if (reg <= turns) turns = reg-1 */ \
  1023. t3 = rcache_get_reg(reg, RC_GR_RMW, NULL); \
  1024. emith_cmp_r_r(t3, t2); \
  1025. EMITH_SJMP_START(DCOND_HI); \
  1026. emith_sub_r_r_imm_c(DCOND_LS, t2, t3, 1); \
  1027. EMITH_SJMP_END(DCOND_HI); \
  1028. /* if (reg <= 1) turns = 0 */ \
  1029. emith_cmp_r_imm(t3, 1); \
  1030. EMITH_SJMP_START(DCOND_HI); \
  1031. emith_move_r_imm_c(DCOND_LS, t2, 0); \
  1032. EMITH_SJMP_END(DCOND_HI); \
  1033. /* reg -= turns */ \
  1034. emith_sub_r_r(t3, t2); \
  1035. } \
  1036. /* sr.cycles -= turns * cycles; */ \
  1037. emith_move_r_imm(t1, cycles); \
  1038. emith_mul_u64(t1, t2, t1, t2); \
  1039. emith_sub_r_r_r_lsl(sr, sr, t1, 12); \
  1040. EMITH_JMP_END(DCOND_LE); \
  1041. rcache_free_tmp(t1); \
  1042. rcache_free_tmp(t2); \
  1043. } while (0)
  1044. #define emith_write_sr(sr, srcr) do { \
  1045. int tmp_ = rcache_get_tmp(); \
  1046. emith_clear_msb(tmp_, srcr, 22); \
  1047. emith_bic_r_imm(sr, 0x3ff); \
  1048. emith_or_r_r(sr, tmp_); \
  1049. rcache_free_tmp(tmp_); \
  1050. } while (0)
  1051. #define emith_tpop_carry(sr, is_sub) \
  1052. emith_lsr(sr, sr, 1)
  1053. #define emith_tpush_carry(sr, is_sub) \
  1054. emith_adc_r_r(sr, sr)
  1055. /*
  1056. * if Q
  1057. * t = carry(Rn += Rm)
  1058. * else
  1059. * t = carry(Rn -= Rm)
  1060. * T ^= t
  1061. */
  1062. #define emith_sh2_div1_step(rn, rm, sr) do { \
  1063. u8 *jmp0, *jmp1; \
  1064. int tmp_ = rcache_get_tmp(); \
  1065. emith_eor_r_r(tmp_, tmp_); \
  1066. emith_tst_r_imm(sr, Q); /* if (Q ^ M) */ \
  1067. JMP8_POS(jmp0); /* je do_sub */ \
  1068. emith_add_r_r(rn, rm); \
  1069. JMP8_POS(jmp1); /* jmp done */ \
  1070. JMP8_EMIT(ICOND_JE, jmp0); /* do_sub: */ \
  1071. emith_sub_r_r(rn, rm); \
  1072. JMP8_EMIT_NC(jmp1); /* done: */ \
  1073. emith_adc_r_r(tmp_, tmp_); \
  1074. emith_eor_r_r(sr, tmp_); \
  1075. rcache_free_tmp(tmp_); \
  1076. } while (0)
  1077. /* mh:ml += rn*rm, does saturation if required by S bit. rn, rm must be TEMP */
  1078. #define emith_sh2_macl(ml, mh, rn, rm, sr) do { \
  1079. emith_tst_r_imm(sr, S); \
  1080. EMITH_SJMP_START(DCOND_EQ); \
  1081. /* MACH top 16 bits unused if saturated. sign ext for overfl detect */ \
  1082. emith_sext(mh, mh, 16); \
  1083. EMITH_SJMP_END(DCOND_EQ); \
  1084. emith_mula_s64(ml, mh, rn, rm); \
  1085. emith_tst_r_imm(sr, S); \
  1086. EMITH_SJMP_START(DCOND_EQ); \
  1087. /* overflow if top 17 bits of MACH aren't all 1 or 0 */ \
  1088. /* to check: add MACH[15] to MACH[31:16]. this is 0 if no overflow */ \
  1089. emith_asrf(rn, mh, 16); /* sum = (MACH>>16) + ((MACH>>15)&1) */ \
  1090. emith_adcf_r_imm(rn, 0); /* (MACH>>15) is in carry after shift */ \
  1091. EMITH_SJMP_START(DCOND_EQ); /* sum != 0 -> ov */ \
  1092. emith_move_r_imm_c(DCOND_NE, ml, 0x0000); /* -overflow */ \
  1093. emith_move_r_imm_c(DCOND_NE, mh, 0x8000); \
  1094. EMITH_SJMP_START(DCOND_LE); /* sum > 0 -> +ovl */ \
  1095. emith_sub_r_imm_c(DCOND_GT, ml, 1); /* 0xffffffff */ \
  1096. emith_sub_r_imm_c(DCOND_GT, mh, 1); /* 0x00007fff */ \
  1097. EMITH_SJMP_END(DCOND_LE); \
  1098. EMITH_SJMP_END(DCOND_EQ); \
  1099. EMITH_SJMP_END(DCOND_EQ); \
  1100. } while (0)
  1101. /* mh:ml += rn*rm, does saturation if required by S bit. rn, rm must be TEMP */
  1102. #define emith_sh2_macw(ml, mh, rn, rm, sr) do { \
  1103. emith_tst_r_imm(sr, S); \
  1104. EMITH_SJMP_START(DCOND_EQ); \
  1105. /* XXX: MACH should be untouched when S is set? */ \
  1106. emith_asr(mh, ml, 31); /* sign ext MACL to MACH for ovrfl check */ \
  1107. EMITH_SJMP_END(DCOND_EQ); \
  1108. emith_mula_s64(ml, mh, rn, rm); \
  1109. emith_tst_r_imm(sr, S); \
  1110. EMITH_SJMP_START(DCOND_EQ); \
  1111. /* overflow if top 33 bits of MACH:MACL aren't all 1 or 0 */ \
  1112. /* to check: add MACL[31] to MACH. this is 0 if no overflow */ \
  1113. emith_lsr(rn, ml, 31); \
  1114. emith_addf_r_r(rn, mh); /* sum = MACH + ((MACL>>31)&1) */ \
  1115. EMITH_SJMP_START(DCOND_EQ); /* sum != 0 -> overflow */ \
  1116. /* XXX: LSB signalling only in SH1, or in SH2 too? */ \
  1117. emith_move_r_imm_c(DCOND_NE, mh, 0x00000001); /* LSB of MACH */ \
  1118. emith_move_r_imm_c(DCOND_NE, ml, 0x80000000); /* negative ovrfl */ \
  1119. EMITH_SJMP_START(DCOND_LE); /* sum > 0 -> positive ovrfl */ \
  1120. emith_sub_r_imm_c(DCOND_GT, ml, 1); /* 0x7fffffff */ \
  1121. EMITH_SJMP_END(DCOND_LE); \
  1122. EMITH_SJMP_END(DCOND_EQ); \
  1123. EMITH_SJMP_END(DCOND_EQ); \
  1124. } while (0)
  1125. #define emith_pool_check() /**/
  1126. #define emith_pool_commit(j) /**/
  1127. #define emith_insn_ptr() ((u8 *)tcache_ptr)
  1128. #define emith_flush() /**/
  1129. #ifdef T
  1130. // T bit handling
  1131. #define emith_invert_cond(cond) \
  1132. ((cond) ^ 1)
  1133. static void emith_clr_t_cond(int sr)
  1134. {
  1135. emith_bic_r_imm(sr, T);
  1136. }
  1137. static void emith_set_t_cond(int sr, int cond)
  1138. {
  1139. EMITH_SJMP_START(emith_invert_cond(cond));
  1140. emith_or_r_imm_c(cond, sr, T);
  1141. EMITH_SJMP_END(emith_invert_cond(cond));
  1142. }
  1143. #define emith_get_t_cond() -1
  1144. #define emith_sync_t(sr) ((void)sr)
  1145. #define emith_invalidate_t()
  1146. static void emith_set_t(int sr, int val)
  1147. {
  1148. if (val)
  1149. emith_or_r_imm(sr, T);
  1150. else
  1151. emith_bic_r_imm(sr, T);
  1152. }
  1153. static int emith_tst_t(int sr, int tf)
  1154. {
  1155. emith_tst_r_imm(sr, T);
  1156. return tf ? DCOND_NE: DCOND_EQ;
  1157. }
  1158. #endif