dynasm/dasm_mips.lua

   1 ------------------------------------------------------------------------------
   2 -- DynASM MIPS module.
   3 --
   4 -- Copyright (C) 2005-2012 Mike Pall. All rights reserved.
   5 -- See dynasm.lua for full copyright notice.
   6 ------------------------------------------------------------------------------
   7
   8 -- Module information:
   9 local _info = {
  10   arch =        "mips",
  11   description = "DynASM MIPS module",
  12   version =     "1.3.0",
  13   vernum =       10300,
  14   release =     "2012-01-23",
  15   author =      "Mike Pall",
  16   license =     "MIT",
  17 }
  18
  19 -- Exported glue functions for the arch-specific module.
  20 local _M = { _info = _info }
  21
  22 -- Cache library functions.
  23 local type, tonumber, pairs, ipairs = type, tonumber, pairs, ipairs
  24 local assert, setmetatable = assert, setmetatable
  25 local _s = string
  26 local sub, format, byte, char = _s.sub, _s.format, _s.byte, _s.char
  27 local match, gmatch = _s.match, _s.gmatch
  28 local concat, sort = table.concat, table.sort
  29
  30 -- Inherited tables and callbacks.
  31 local g_opt, g_arch
  32 local wline, werror, wfatal, wwarn
  33
  34 -- Action name list.
  35 -- CHECK: Keep this in sync with the C code!
  36 local action_names = {
  37   "STOP", "SECTION", "ESC", "REL_EXT",
  38   "ALIGN", "REL_LG", "LABEL_LG",
  39   "REL_PC", "LABEL_PC", "IMM",
  40 }
  41
  42 -- Maximum number of section buffer positions for dasm_put().
  43 -- CHECK: Keep this in sync with the C code!
  44 local maxsecpos = 25 -- Keep this low, to avoid excessively long C lines.
  45
  46 -- Action name -> action number.
  47 local map_action = {}
  48 for n,name in ipairs(action_names) do
  49   map_action[name] = n-1
  50 end
  51
  52 -- Action list buffer.
  53 local actlist = {}
  54
  55 -- Argument list for next dasm_put(). Start with offset 0 into action list.
  56 local actargs = { 0 }
  57
  58 -- Current number of section buffer positions for dasm_put().
  59 local secpos = 1
  60
  61 ------------------------------------------------------------------------------
  62
  63 -- Return 8 digit hex number.
  64 local function tohex(x)
  65   return sub(format("%08x", x), -8) -- Avoid 64 bit portability problem in Lua.
  66 end
  67
  68 -- Dump action names and numbers.
  69 local function dumpactions(out)
  70   out:write("DynASM encoding engine action codes:\n")
  71   for n,name in ipairs(action_names) do
  72     local num = map_action[name]
  73     out:write(format("  %-10s %02X  %d\n", name, num, num))
  74   end
  75   out:write("\n")
  76 end
  77
  78 -- Write action list buffer as a huge static C array.
  79 local function writeactions(out, name)
  80   local nn = #actlist
  81   if nn == 0 then nn = 1; actlist[0] = map_action.STOP end
  82   out:write("static const unsigned int ", name, "[", nn, "] = {\n")
  83   for i = 1,nn-1 do
  84     assert(out:write("0x", tohex(actlist[i]), ",\n"))
  85   end
  86   assert(out:write("0x", tohex(actlist[nn]), "\n};\n\n"))
  87 end
  88
  89 ------------------------------------------------------------------------------
  90
  91 -- Add word to action list.
  92 local function wputxw(n)
  93   assert(n >= 0 and n <= 0xffffffff and n % 1 == 0, "word out of range")
  94   actlist[#actlist+1] = n
  95 end
  96
  97 -- Add action to list with optional arg. Advance buffer pos, too.
  98 local function waction(action, val, a, num)
  99   local w = assert(map_action[action], "bad action name `"..action.."'")
 100   wputxw(0xff000000 + w * 0x10000 + (val or 0))
 101   if a then actargs[#actargs+1] = a end
 102   if a or num then secpos = secpos + (num or 1) end
 103 end
 104
 105 -- Flush action list (intervening C code or buffer pos overflow).
 106 local function wflush(term)
 107   if #actlist == actargs[1] then return end -- Nothing to flush.
 108   if not term then waction("STOP") end -- Terminate action list.
 109   wline(format("dasm_put(Dst, %s);", concat(actargs, ", ")), true)
 110   actargs = { #actlist } -- Actionlist offset is 1st arg to next dasm_put().
 111   secpos = 1 -- The actionlist offset occupies a buffer position, too.
 112 end
 113
 114 -- Put escaped word.
 115 local function wputw(n)
 116   if n >= 0xff000000 then waction("ESC") end
 117   wputxw(n)
 118 end
 119
 120 -- Reserve position for word.
 121 local function wpos()
 122   local pos = #actlist+1
 123   actlist[pos] = ""
 124   return pos
 125 end
 126
 127 -- Store word to reserved position.
 128 local function wputpos(pos, n)
 129   assert(n >= 0 and n <= 0xffffffff and n % 1 == 0, "word out of range")
 130   actlist[pos] = n
 131 end
 132
 133 ------------------------------------------------------------------------------
 134
 135 -- Global label name -> global label number. With auto assignment on 1st use.
 136 local next_global = 20
 137 local map_global = setmetatable({}, { __index = function(t, name)
 138   if not match(name, "^[%a_][%w_]*$") then werror("bad global label") end
 139   local n = next_global
 140   if n > 2047 then werror("too many global labels") end
 141   next_global = n + 1
 142   t[name] = n
 143   return n
 144 end})
 145
 146 -- Dump global labels.
 147 local function dumpglobals(out, lvl)
 148   local t = {}
 149   for name, n in pairs(map_global) do t[n] = name end
 150   out:write("Global labels:\n")
 151   for i=20,next_global-1 do
 152     out:write(format("  %s\n", t[i]))
 153   end
 154   out:write("\n")
 155 end
 156
 157 -- Write global label enum.
 158 local function writeglobals(out, prefix)
 159   local t = {}
 160   for name, n in pairs(map_global) do t[n] = name end
 161   out:write("enum {\n")
 162   for i=20,next_global-1 do
 163     out:write("  ", prefix, t[i], ",\n")
 164   end
 165   out:write("  ", prefix, "_MAX\n};\n")
 166 end
 167
 168 -- Write global label names.
 169 local function writeglobalnames(out, name)
 170   local t = {}
 171   for name, n in pairs(map_global) do t[n] = name end
 172   out:write("static const char *const ", name, "[] = {\n")
 173   for i=20,next_global-1 do
 174     out:write("  \"", t[i], "\",\n")
 175   end
 176   out:write("  (const char *)0\n};\n")
 177 end
 178
 179 ------------------------------------------------------------------------------
 180
 181 -- Extern label name -> extern label number. With auto assignment on 1st use.
 182 local next_extern = 0
 183 local map_extern_ = {}
 184 local map_extern = setmetatable({}, { __index = function(t, name)
 185   -- No restrictions on the name for now.
 186   local n = next_extern
 187   if n > 2047 then werror("too many extern labels") end
 188   next_extern = n + 1
 189   t[name] = n
 190   map_extern_[n] = name
 191   return n
 192 end})
 193
 194 -- Dump extern labels.
 195 local function dumpexterns(out, lvl)
 196   out:write("Extern labels:\n")
 197   for i=0,next_extern-1 do
 198     out:write(format("  %s\n", map_extern_[i]))
 199   end
 200   out:write("\n")
 201 end
 202
 203 -- Write extern label names.
 204 local function writeexternnames(out, name)
 205   out:write("static const char *const ", name, "[] = {\n")
 206   for i=0,next_extern-1 do
 207     out:write("  \"", map_extern_[i], "\",\n")
 208   end
 209   out:write("  (const char *)0\n};\n")
 210 end
 211
 212 ------------------------------------------------------------------------------
 213
 214 -- Arch-specific maps.
 215 local map_archdef = { sp="r29", ra="r31" } -- Ext. register name -> int. name.
 216
 217 local map_type = {}             -- Type name -> { ctype, reg }
 218 local ctypenum = 0              -- Type number (for Dt... macros).
 219
 220 -- Reverse defines for registers.
 221 function _M.revdef(s)
 222   if s == "r29" then return "sp"
 223   elseif s == "r31" then return "ra" end
 224   return s
 225 end
 226
 227 ------------------------------------------------------------------------------
 228
 229 -- Template strings for MIPS instructions.
 230 local map_op = {
 231   -- First-level opcodes.
 232   j_1 =         "08000000J",
 233   jal_1 =       "0c000000J",
 234   b_1 =         "10000000B",
 235   beqz_2 =      "10000000SB",
 236   beq_3 =       "10000000STB",
 237   bnez_2 =      "14000000SB",
 238   bne_3 =       "14000000STB",
 239   blez_2 =      "18000000SB",
 240   bgtz_2 =      "1c000000SB",
 241   addi_3 =      "20000000TSI",
 242   li_2 =        "24000000TI",
 243   addiu_3 =     "24000000TSI",
 244   slti_3 =      "28000000TSI",
 245   sltiu_3 =     "2c000000TSI",
 246   andi_3 =      "30000000TSU",
 247   lu_2 =        "34000000TU",
 248   ori_3 =       "34000000TSU",
 249   xori_3 =      "38000000TSU",
 250   lui_2 =       "3c000000TU",
 251   beqzl_2 =     "50000000SB",
 252   beql_3 =      "50000000STB",
 253   bnezl_2 =     "54000000SB",
 254   bnel_3 =      "54000000STB",
 255   blezl_2 =     "58000000SB",
 256   bgtzl_2 =     "5c000000SB",
 257   lb_2 =        "80000000TO",
 258   lh_2 =        "84000000TO",
 259   lwl_2 =       "88000000TO",
 260   lw_2 =        "8c000000TO",
 261   lbu_2 =       "90000000TO",
 262   lhu_2 =       "94000000TO",
 263   lwr_2 =       "98000000TO",
 264   sb_2 =        "a0000000TO",
 265   sh_2 =        "a4000000TO",
 266   swl_2 =       "a8000000TO",
 267   sw_2 =        "ac000000TO",
 268   swr_2 =       "b8000000TO",
 269   cache_2 =     "bc000000NO",
 270   ll_2 =        "c0000000TO",
 271   lwc1_2 =      "c4000000HO",
 272   pref_2 =      "cc000000NO",
 273   ldc1_2 =      "d4000000HO",
 274   sc_2 =        "e0000000TO",
 275   swc1_2 =      "e4000000HO",
 276   sdc1_2 =      "f4000000HO",
 277
 278   -- Opcode SPECIAL.
 279   nop_0 =       "00000000",
 280   sll_3 =       "00000000DTA",
 281   movf_2 =      "00000001DS",
 282   movf_3 =      "00000001DSC",
 283   movt_2 =      "00010001DS",
 284   movt_3 =      "00010001DSC",
 285   srl_3 =       "00000002DTA",
 286   rotr_3 =      "00200002DTA",
 287   sra_3 =       "00000003DTA",
 288   sllv_3 =      "00000004DTS",
 289   srlv_3 =      "00000006DTS",
 290   rotrv_3 =     "00000046DTS",
 291   srav_3 =      "00000007DTS",
 292   jr_1 =        "00000008S",
 293   jalr_1 =      "0000f809S",
 294   jalr_2 =      "00000009DS",
 295   movz_3 =      "0000000aDST",
 296   movn_3 =      "0000000bDST",
 297   syscall_0 =   "0000000c",
 298   syscall_1 =   "0000000cY",
 299   break_0 =     "0000000d",
 300   break_1 =     "0000000dY",
 301   sync_0 =      "0000000f",
 302   mfhi_1 =      "00000010D",
 303   mthi_1 =      "00000011S",
 304   mflo_1 =      "00000012D",
 305   mtlo_1 =      "00000013S",
 306   mult_2 =      "00000018ST",
 307   multu_2 =     "00000019ST",
 308   div_2 =       "0000001aST",
 309   divu_2 =      "0000001bST",
 310   add_3 =       "00000020DST",
 311   move_2 =      "00000021DS",
 312   addu_3 =      "00000021DST",
 313   sub_3 =       "00000022DST",
 314   negu_2 =      "00000023DT",
 315   subu_3 =      "00000023DST",
 316   and_3 =       "00000024DST",
 317   or_3 =        "00000025DST",
 318   xor_3 =       "00000026DST",
 319   not_2 =       "00000027DS",
 320   nor_3 =       "00000027DST",
 321   slt_3 =       "0000002aDST",
 322   sltu_3 =      "0000002bDST",
 323   tge_2 =       "00000030ST",
 324   tge_3 =       "00000030STZ",
 325   tgeu_2 =      "00000031ST",
 326   tgeu_3 =      "00000031STZ",
 327   tlt_2 =       "00000032ST",
 328   tlt_3 =       "00000032STZ",
 329   tltu_2 =      "00000033ST",
 330   tltu_3 =      "00000033STZ",
 331   teq_2 =       "00000034ST",
 332   teq_3 =       "00000034STZ",
 333   tne_2 =       "00000036ST",
 334   tne_3 =       "00000036STZ",
 335
 336   -- Opcode REGIMM.
 337   bltz_2 =      "04000000SB",
 338   bgez_2 =      "04010000SB",
 339   bltzl_2 =     "04020000SB",
 340   bgezl_2 =     "04030000SB",
 341   tgei_2 =      "04080000SI",
 342   tgeiu_2 =     "04090000SI",
 343   tlti_2 =      "040a0000SI",
 344   tltiu_2 =     "040b0000SI",
 345   teqi_2 =      "040c0000SI",
 346   tnei_2 =      "040e0000SI",
 347   bltzal_2 =    "04100000SB",
 348   bal_1 =       "04110000B",
 349   bgezal_2 =    "04110000SB",
 350   bltzall_2 =   "04120000SB",
 351   bgezall_2 =   "04130000SB",
 352   synci_1 =     "041f0000O",
 353
 354   -- Opcode SPECIAL2.
 355   madd_2 =      "70000000ST",
 356   maddu_2 =     "70000001ST",
 357   mul_3 =       "70000002DST",
 358   msub_2 =      "70000004ST",
 359   msubu_2 =     "70000005ST",
 360   clz_2 =       "70000020DS=",
 361   clo_2 =       "70000021DS=",
 362   sdbbp_0 =     "7000003f",
 363   sdbbp_1 =     "7000003fY",
 364
 365   -- Opcode SPECIAL3.
 366   ext_4 =       "7c000000TSAM", -- Note: last arg is msbd = size-1
 367   ins_4 =       "7c000004TSAM", -- Note: last arg is msb = pos+size-1
 368   wsbh_2 =      "7c0000a0DT",
 369   seb_2 =       "7c000420DT",
 370   seh_2 =       "7c000620DT",
 371   rdhwr_2 =     "7c00003bTD",
 372
 373   -- Opcode COP0.
 374   mfc0_2 =      "40000000TD",
 375   mfc0_3 =      "40000000TDW",
 376   mtc0_2 =      "40800000TD",
 377   mtc0_3 =      "40800000TDW",
 378   rdpgpr_2 =    "41400000DT",
 379   di_0 =        "41606000",
 380   di_1 =        "41606000T",
 381   ei_0 =        "41606020",
 382   ei_1 =        "41606020T",
 383   wrpgpr_2 =    "41c00000DT",
 384   tlbr_0 =      "42000001",
 385   tlbwi_0 =     "42000002",
 386   tlbwr_0 =     "42000006",
 387   tlbp_0 =      "42000008",
 388   eret_0 =      "42000018",
 389   deret_0 =     "4200001f",
 390   wait_0 =      "42000020",
 391
 392   -- Opcode COP1.
 393   mfc1_2 =      "44000000TG",
 394   cfc1_2 =      "44400000TG",
 395   mfhc1_2 =     "44600000TG",
 396   mtc1_2 =      "44800000TG",
 397   ctc1_2 =      "44c00000TG",
 398   mthc1_2 =     "44e00000TG",
 399
 400   bc1f_1 =      "45000000B",
 401   bc1f_2 =      "45000000CB",
 402   bc1t_1 =      "45010000B",
 403   bc1t_2 =      "45010000CB",
 404   bc1fl_1 =     "45020000B",
 405   bc1fl_2 =     "45020000CB",
 406   bc1tl_1 =     "45030000B",
 407   bc1tl_2 =     "45030000CB",
 408
 409   ["add.s_3"] =         "46000000FGH",
 410   ["sub.s_3"] =         "46000001FGH",
 411   ["mul.s_3"] =         "46000002FGH",
 412   ["div.s_3"] =         "46000003FGH",
 413   ["sqrt.s_2"] =        "46000004FG",
 414   ["abs.s_2"] =         "46000005FG",
 415   ["mov.s_2"] =         "46000006FG",
 416   ["neg.s_2"] =         "46000007FG",
 417   ["round.l.s_2"] =     "46000008FG",
 418   ["trunc.l.s_2"] =     "46000009FG",
 419   ["ceil.l.s_2"] =      "4600000aFG",
 420   ["floor.l.s_2"] =     "4600000bFG",
 421   ["round.w.s_2"] =     "4600000cFG",
 422   ["trunc.w.s_2"] =     "4600000dFG",
 423   ["ceil.w.s_2"] =      "4600000eFG",
 424   ["floor.w.s_2"] =     "4600000fFG",
 425   ["movf.s_2"] =        "46000011FG",
 426   ["movf.s_3"] =        "46000011FGC",
 427   ["movt.s_2"] =        "46010011FG",
 428   ["movt.s_3"] =        "46010011FGC",
 429   ["movz.s_3"] =        "46000012FGT",
 430   ["movn.s_3"] =        "46000013FGT",
 431   ["recip.s_2"] =       "46000015FG",
 432   ["rsqrt.s_2"] =       "46000016FG",
 433   ["cvt.d.s_2"] =       "46000021FG",
 434   ["cvt.w.s_2"] =       "46000024FG",
 435   ["cvt.l.s_2"] =       "46000025FG",
 436   ["cvt.ps.s_3"] =      "46000026FGH",
 437   ["c.f.s_2"] =         "46000030GH",
 438   ["c.f.s_3"] =         "46000030VGH",
 439   ["c.un.s_2"] =        "46000031GH",
 440   ["c.un.s_3"] =        "46000031VGH",
 441   ["c.eq.s_2"] =        "46000032GH",
 442   ["c.eq.s_3"] =        "46000032VGH",
 443   ["c.ueq.s_2"] =       "46000033GH",
 444   ["c.ueq.s_3"] =       "46000033VGH",
 445   ["c.olt.s_2"] =       "46000034GH",
 446   ["c.olt.s_3"] =       "46000034VGH",
 447   ["c.ult.s_2"] =       "46000035GH",
 448   ["c.ult.s_3"] =       "46000035VGH",
 449   ["c.ole.s_2"] =       "46000036GH",
 450   ["c.ole.s_3"] =       "46000036VGH",
 451   ["c.ule.s_2"] =       "46000037GH",
 452   ["c.ule.s_3"] =       "46000037VGH",
 453   ["c.sf.s_2"] =        "46000038GH",
 454   ["c.sf.s_3"] =        "46000038VGH",
 455   ["c.ngle.s_2"] =      "46000039GH",
 456   ["c.ngle.s_3"] =      "46000039VGH",
 457   ["c.seq.s_2"] =       "4600003aGH",
 458   ["c.seq.s_3"] =       "4600003aVGH",
 459   ["c.ngl.s_2"] =       "4600003bGH",
 460   ["c.ngl.s_3"] =       "4600003bVGH",
 461   ["c.lt.s_2"] =        "4600003cGH",
 462   ["c.lt.s_3"] =        "4600003cVGH",
 463   ["c.nge.s_2"] =       "4600003dGH",
 464   ["c.nge.s_3"] =       "4600003dVGH",
 465   ["c.le.s_2"] =        "4600003eGH",
 466   ["c.le.s_3"] =        "4600003eVGH",
 467   ["c.ngt.s_2"] =       "4600003fGH",
 468   ["c.ngt.s_3"] =       "4600003fVGH",
 469
 470   ["add.d_3"] =         "46200000FGH",
 471   ["sub.d_3"] =         "46200001FGH",
 472   ["mul.d_3"] =         "46200002FGH",
 473   ["div.d_3"] =         "46200003FGH",
 474   ["sqrt.d_2"] =        "46200004FG",
 475   ["abs.d_2"] =         "46200005FG",
 476   ["mov.d_2"] =         "46200006FG",
 477   ["neg.d_2"] =         "46200007FG",
 478   ["round.l.d_2"] =     "46200008FG",
 479   ["trunc.l.d_2"] =     "46200009FG",
 480   ["ceil.l.d_2"] =      "4620000aFG",
 481   ["floor.l.d_2"] =     "4620000bFG",
 482   ["round.w.d_2"] =     "4620000cFG",
 483   ["trunc.w.d_2"] =     "4620000dFG",
 484   ["ceil.w.d_2"] =      "4620000eFG",
 485   ["floor.w.d_2"] =     "4620000fFG",
 486   ["movf.d_2"] =        "46200011FG",
 487   ["movf.d_3"] =        "46200011FGC",
 488   ["movt.d_2"] =        "46210011FG",
 489   ["movt.d_3"] =        "46210011FGC",
 490   ["movz.d_3"] =        "46200012FGT",
 491   ["movn.d_3"] =        "46200013FGT",
 492   ["recip.d_2"] =       "46200015FG",
 493   ["rsqrt.d_2"] =       "46200016FG",
 494   ["cvt.s.d_2"] =       "46200020FG",
 495   ["cvt.w.d_2"] =       "46200024FG",
 496   ["cvt.l.d_2"] =       "46200025FG",
 497   ["c.f.d_2"] =         "46200030GH",
 498   ["c.f.d_3"] =         "46200030VGH",
 499   ["c.un.d_2"] =        "46200031GH",
 500   ["c.un.d_3"] =        "46200031VGH",
 501   ["c.eq.d_2"] =        "46200032GH",
 502   ["c.eq.d_3"] =        "46200032VGH",
 503   ["c.ueq.d_2"] =       "46200033GH",
 504   ["c.ueq.d_3"] =       "46200033VGH",
 505   ["c.olt.d_2"] =       "46200034GH",
 506   ["c.olt.d_3"] =       "46200034VGH",
 507   ["c.ult.d_2"] =       "46200035GH",
 508   ["c.ult.d_3"] =       "46200035VGH",
 509   ["c.ole.d_2"] =       "46200036GH",
 510   ["c.ole.d_3"] =       "46200036VGH",
 511   ["c.ule.d_2"] =       "46200037GH",
 512   ["c.ule.d_3"] =       "46200037VGH",
 513   ["c.sf.d_2"] =        "46200038GH",
 514   ["c.sf.d_3"] =        "46200038VGH",
 515   ["c.ngle.d_2"] =      "46200039GH",
 516   ["c.ngle.d_3"] =      "46200039VGH",
 517   ["c.seq.d_2"] =       "4620003aGH",
 518   ["c.seq.d_3"] =       "4620003aVGH",
 519   ["c.ngl.d_2"] =       "4620003bGH",
 520   ["c.ngl.d_3"] =       "4620003bVGH",
 521   ["c.lt.d_2"] =        "4620003cGH",
 522   ["c.lt.d_3"] =        "4620003cVGH",
 523   ["c.nge.d_2"] =       "4620003dGH",
 524   ["c.nge.d_3"] =       "4620003dVGH",
 525   ["c.le.d_2"] =        "4620003eGH",
 526   ["c.le.d_3"] =        "4620003eVGH",
 527   ["c.ngt.d_2"] =       "4620003fGH",
 528   ["c.ngt.d_3"] =       "4620003fVGH",
 529
 530   ["add.ps_3"] =        "46c00000FGH",
 531   ["sub.ps_3"] =        "46c00001FGH",
 532   ["mul.ps_3"] =        "46c00002FGH",
 533   ["abs.ps_2"] =        "46c00005FG",
 534   ["mov.ps_2"] =        "46c00006FG",
 535   ["neg.ps_2"] =        "46c00007FG",
 536   ["movf.ps_2"] =       "46c00011FG",
 537   ["movf.ps_3"] =       "46c00011FGC",
 538   ["movt.ps_2"] =       "46c10011FG",
 539   ["movt.ps_3"] =       "46c10011FGC",
 540   ["movz.ps_3"] =       "46c00012FGT",
 541   ["movn.ps_3"] =       "46c00013FGT",
 542   ["cvt.s.pu_2"] =      "46c00020FG",
 543   ["cvt.s.pl_2"] =      "46c00028FG",
 544   ["pll.ps_3"] =        "46c0002cFGH",
 545   ["plu.ps_3"] =        "46c0002dFGH",
 546   ["pul.ps_3"] =        "46c0002eFGH",
 547   ["puu.ps_3"] =        "46c0002fFGH",
 548   ["c.f.ps_2"] =        "46c00030GH",
 549   ["c.f.ps_3"] =        "46c00030VGH",
 550   ["c.un.ps_2"] =       "46c00031GH",
 551   ["c.un.ps_3"] =       "46c00031VGH",
 552   ["c.eq.ps_2"] =       "46c00032GH",
 553   ["c.eq.ps_3"] =       "46c00032VGH",
 554   ["c.ueq.ps_2"] =      "46c00033GH",
 555   ["c.ueq.ps_3"] =      "46c00033VGH",
 556   ["c.olt.ps_2"] =      "46c00034GH",
 557   ["c.olt.ps_3"] =      "46c00034VGH",
 558   ["c.ult.ps_2"] =      "46c00035GH",
 559   ["c.ult.ps_3"] =      "46c00035VGH",
 560   ["c.ole.ps_2"] =      "46c00036GH",
 561   ["c.ole.ps_3"] =      "46c00036VGH",
 562   ["c.ule.ps_2"] =      "46c00037GH",
 563   ["c.ule.ps_3"] =      "46c00037VGH",
 564   ["c.sf.ps_2"] =       "46c00038GH",
 565   ["c.sf.ps_3"] =       "46c00038VGH",
 566   ["c.ngle.ps_2"] =     "46c00039GH",
 567   ["c.ngle.ps_3"] =     "46c00039VGH",
 568   ["c.seq.ps_2"] =      "46c0003aGH",
 569   ["c.seq.ps_3"] =      "46c0003aVGH",
 570   ["c.ngl.ps_2"] =      "46c0003bGH",
 571   ["c.ngl.ps_3"] =      "46c0003bVGH",
 572   ["c.lt.ps_2"] =       "46c0003cGH",
 573   ["c.lt.ps_3"] =       "46c0003cVGH",
 574   ["c.nge.ps_2"] =      "46c0003dGH",
 575   ["c.nge.ps_3"] =      "46c0003dVGH",
 576   ["c.le.ps_2"] =       "46c0003eGH",
 577   ["c.le.ps_3"] =       "46c0003eVGH",
 578   ["c.ngt.ps_2"] =      "46c0003fGH",
 579   ["c.ngt.ps_3"] =      "46c0003fVGH",
 580
 581   ["cvt.s.w_2"] =       "46800020FG",
 582   ["cvt.d.w_2"] =       "46800021FG",
 583
 584   ["cvt.s.l_2"] =       "46a00020FG",
 585   ["cvt.d.l_2"] =       "46a00021FG",
 586
 587   -- Opcode COP1X.
 588   lwxc1_2 =             "4c000000FX",
 589   ldxc1_2 =             "4c000001FX",
 590   luxc1_2 =             "4c000005FX",
 591   swxc1_2 =             "4c000008FX",
 592   sdxc1_2 =             "4c000009FX",
 593   suxc1_2 =             "4c00000dFX",
 594   prefx_2 =             "4c00000fMX",
 595   ["alnv.ps_4"] =       "4c00001eFGHS",
 596   ["madd.s_4"] =        "4c000020FRGH",
 597   ["madd.d_4"] =        "4c000021FRGH",
 598   ["madd.ps_4"] =       "4c000026FRGH",
 599   ["msub.s_4"] =        "4c000028FRGH",
 600   ["msub.d_4"] =        "4c000029FRGH",
 601   ["msub.ps_4"] =       "4c00002eFRGH",
 602   ["nmadd.s_4"] =       "4c000030FRGH",
 603   ["nmadd.d_4"] =       "4c000031FRGH",
 604   ["nmadd.ps_4"] =      "4c000036FRGH",
 605   ["nmsub.s_4"] =       "4c000038FRGH",
 606   ["nmsub.d_4"] =       "4c000039FRGH",
 607   ["nmsub.ps_4"] =      "4c00003eFRGH",
 608 }
 609
 610 ------------------------------------------------------------------------------
 611
 612 local function parse_gpr(expr)
 613   local tname, ovreg = match(expr, "^([%w_]+):(r[1-3]?[0-9])$")
 614   local tp = map_type[tname or expr]
 615   if tp then
 616     local reg = ovreg or tp.reg
 617     if not reg then
 618       werror("type `"..(tname or expr).."' needs a register override")
 619     end
 620     expr = reg
 621   end
 622   local r = match(expr, "^r([1-3]?[0-9])$")
 623   if r then
 624     r = tonumber(r)
 625     if r <= 31 then return r, tp end
 626   end
 627   werror("bad register name `"..expr.."'")
 628 end
 629
 630 local function parse_fpr(expr)
 631   local r = match(expr, "^f([1-3]?[0-9])$")
 632   if r then
 633     r = tonumber(r)
 634     if r <= 31 then return r end
 635   end
 636   werror("bad register name `"..expr.."'")
 637 end
 638
 639 local function parse_imm(imm, bits, shift, scale, signed)
 640   local n = tonumber(imm)
 641   if n then
 642     if n % 2^scale == 0 then
 643       n = n / 2^scale
 644       if signed then
 645         if n >= 0 then
 646           if n < 2^(bits-1) then return n*2^shift end
 647         else
 648           if n >= -(2^(bits-1))-1 then return (n+2^bits)*2^shift end
 649         end
 650       else
 651         if n >= 0 and n <= 2^bits-1 then return n*2^shift end
 652       end
 653     end
 654     werror("out of range immediate `"..imm.."'")
 655   elseif match(imm, "^[rf]([1-3]?[0-9])$") or
 656          match(imm, "^([%w_]+):([rf][1-3]?[0-9])$") then
 657     werror("expected immediate operand, got register")
 658   else
 659     waction("IMM", (signed and 32768 or 0)+scale*1024+bits*32+shift, imm)
 660     return 0
 661   end
 662 end
 663
 664 local function parse_disp(disp)
 665   local imm, reg = match(disp, "^(.*)%(([%w_:]+)%)$")
 666   if imm then
 667     local r = parse_gpr(reg)*2^21
 668     local extname = match(imm, "^extern%s+(%S+)$")
 669     if extname then
 670       waction("REL_EXT", map_extern[extname], nil, 1)
 671       return r
 672     else
 673       return r + parse_imm(imm, 16, 0, 0, true)
 674     end
 675   end
 676   local reg, tailr = match(disp, "^([%w_:]+)%s*(.*)$")
 677   if reg and tailr ~= "" then
 678     local r, tp = parse_gpr(reg)
 679     if tp then
 680       waction("IMM", 32768+16*32, format(tp.ctypefmt, tailr))
 681       return r*2^21
 682     end
 683   end
 684   werror("bad displacement `"..disp.."'")
 685 end
 686
 687 local function parse_index(idx)
 688   local rt, rs = match(idx, "^(.*)%(([%w_:]+)%)$")
 689   if rt then
 690     rt = parse_gpr(rt)
 691     rs = parse_gpr(rs)
 692     return rt*2^16 + rs*2^21
 693   end
 694   werror("bad index `"..idx.."'")
 695 end
 696
 697 local function parse_label(label, def)
 698   local prefix = sub(label, 1, 2)
 699   -- =>label (pc label reference)
 700   if prefix == "=>" then
 701     return "PC", 0, sub(label, 3)
 702   end
 703   -- ->name (global label reference)
 704   if prefix == "->" then
 705     return "LG", map_global[sub(label, 3)]
 706   end
 707   if def then
 708     -- [1-9] (local label definition)
 709     if match(label, "^[1-9]$") then
 710       return "LG", 10+tonumber(label)
 711     end
 712   else
 713     -- [<>][1-9] (local label reference)
 714     local dir, lnum = match(label, "^([<>])([1-9])$")
 715     if dir then -- Fwd: 1-9, Bkwd: 11-19.
 716       return "LG", lnum + (dir == ">" and 0 or 10)
 717     end
 718     -- extern label (extern label reference)
 719     local extname = match(label, "^extern%s+(%S+)$")
 720     if extname then
 721       return "EXT", map_extern[extname]
 722     end
 723   end
 724   werror("bad label `"..label.."'")
 725 end
 726
 727 ------------------------------------------------------------------------------
 728
 729 -- Handle opcodes defined with template strings.
 730 map_op[".template__"] = function(params, template, nparams)
 731   if not params then return sub(template, 9) end
 732   local op = tonumber(sub(template, 1, 8), 16)
 733   local n = 1
 734
 735   -- Limit number of section buffer positions used by a single dasm_put().
 736   -- A single opcode needs a maximum of 2 positions (ins/ext).
 737   if secpos+2 > maxsecpos then wflush() end
 738   local pos = wpos()
 739
 740   -- Process each character.
 741   for p in gmatch(sub(template, 9), ".") do
 742     if p == "D" then
 743       op = op + parse_gpr(params[n]) * 2^11; n = n + 1
 744     elseif p == "T" then
 745       op = op + parse_gpr(params[n]) * 2^16; n = n + 1
 746     elseif p == "S" then
 747       op = op + parse_gpr(params[n]) * 2^21; n = n + 1
 748     elseif p == "F" then
 749       op = op + parse_fpr(params[n]) * 2^6; n = n + 1
 750     elseif p == "G" then
 751       op = op + parse_fpr(params[n]) * 2^11; n = n + 1
 752     elseif p == "H" then
 753       op = op + parse_fpr(params[n]) * 2^16; n = n + 1
 754     elseif p == "R" then
 755       op = op + parse_fpr(params[n]) * 2^21; n = n + 1
 756     elseif p == "I" then
 757       op = op + parse_imm(params[n], 16, 0, 0, true); n = n + 1
 758     elseif p == "U" then
 759       op = op + parse_imm(params[n], 16, 0, 0, false); n = n + 1
 760     elseif p == "O" then
 761       op = op + parse_disp(params[n]); n = n + 1
 762     elseif p == "X" then
 763       op = op + parse_index(params[n]); n = n + 1
 764     elseif p == "B" or p == "J" then
 765       local mode, n, s = parse_label(params[n], false)
 766       if p == "B" then n = n + 2048 end
 767       waction("REL_"..mode, n, s, 1)
 768       n = n + 1
 769     elseif p == "A" then
 770       op = op + parse_imm(params[n], 5, 6, 0, false); n = n + 1
 771     elseif p == "M" then
 772       op = op + parse_imm(params[n], 5, 11, 0, false); n = n + 1
 773     elseif p == "N" then
 774       op = op + parse_imm(params[n], 5, 16, 0, false); n = n + 1
 775     elseif p == "C" then
 776       op = op + parse_imm(params[n], 3, 18, 0, false); n = n + 1
 777     elseif p == "V" then
 778       op = op + parse_imm(params[n], 3, 8, 0, false); n = n + 1
 779     elseif p == "W" then
 780       op = op + parse_imm(params[n], 3, 0, 0, false); n = n + 1
 781     elseif p == "Y" then
 782       op = op + parse_imm(params[n], 20, 6, 0, false); n = n + 1
 783     elseif p == "Z" then
 784       op = op + parse_imm(params[n], 10, 6, 0, false); n = n + 1
 785     elseif p == "=" then
 786       local d = ((op - op % 2^11) / 2^11) % 32
 787       op = op + d * 2^16 -- Copy D to T for clz, clo.
 788     else
 789       assert(false)
 790     end
 791   end
 792   wputpos(pos, op)
 793 end
 794
 795 ------------------------------------------------------------------------------
 796
 797 -- Pseudo-opcode to mark the position where the action list is to be emitted.
 798 map_op[".actionlist_1"] = function(params)
 799   if not params then return "cvar" end
 800   local name = params[1] -- No syntax check. You get to keep the pieces.
 801   wline(function(out) writeactions(out, name) end)
 802 end
 803
 804 -- Pseudo-opcode to mark the position where the global enum is to be emitted.
 805 map_op[".globals_1"] = function(params)
 806   if not params then return "prefix" end
 807   local prefix = params[1] -- No syntax check. You get to keep the pieces.
 808   wline(function(out) writeglobals(out, prefix) end)
 809 end
 810
 811 -- Pseudo-opcode to mark the position where the global names are to be emitted.
 812 map_op[".globalnames_1"] = function(params)
 813   if not params then return "cvar" end
 814   local name = params[1] -- No syntax check. You get to keep the pieces.
 815   wline(function(out) writeglobalnames(out, name) end)
 816 end
 817
 818 -- Pseudo-opcode to mark the position where the extern names are to be emitted.
 819 map_op[".externnames_1"] = function(params)
 820   if not params then return "cvar" end
 821   local name = params[1] -- No syntax check. You get to keep the pieces.
 822   wline(function(out) writeexternnames(out, name) end)
 823 end
 824
 825 ------------------------------------------------------------------------------
 826
 827 -- Label pseudo-opcode (converted from trailing colon form).
 828 map_op[".label_1"] = function(params)
 829   if not params then return "[1-9] | ->global | =>pcexpr" end
 830   if secpos+1 > maxsecpos then wflush() end
 831   local mode, n, s = parse_label(params[1], true)
 832   if mode == "EXT" then werror("bad label definition") end
 833   waction("LABEL_"..mode, n, s, 1)
 834 end
 835
 836 ------------------------------------------------------------------------------
 837
 838 -- Pseudo-opcodes for data storage.
 839 map_op[".long_*"] = function(params)
 840   if not params then return "imm..." end
 841   for _,p in ipairs(params) do
 842     local n = tonumber(p)
 843     if not n then werror("bad immediate `"..p.."'") end
 844     if n < 0 then n = n + 2^32 end
 845     wputw(n)
 846     if secpos+2 > maxsecpos then wflush() end
 847   end
 848 end
 849
 850 -- Alignment pseudo-opcode.
 851 map_op[".align_1"] = function(params)
 852   if not params then return "numpow2" end
 853   if secpos+1 > maxsecpos then wflush() end
 854   local align = tonumber(params[1])
 855   if align then
 856     local x = align
 857     -- Must be a power of 2 in the range (2 ... 256).
 858     for i=1,8 do
 859       x = x / 2
 860       if x == 1 then
 861         waction("ALIGN", align-1, nil, 1) -- Action byte is 2**n-1.
 862         return
 863       end
 864     end
 865   end
 866   werror("bad alignment")
 867 end
 868
 869 ------------------------------------------------------------------------------
 870
 871 -- Pseudo-opcode for (primitive) type definitions (map to C types).
 872 map_op[".type_3"] = function(params, nparams)
 873   if not params then
 874     return nparams == 2 and "name, ctype" or "name, ctype, reg"
 875   end
 876   local name, ctype, reg = params[1], params[2], params[3]
 877   if not match(name, "^[%a_][%w_]*$") then
 878     werror("bad type name `"..name.."'")
 879   end
 880   local tp = map_type[name]
 881   if tp then
 882     werror("duplicate type `"..name.."'")
 883   end
 884   -- Add #type to defines. A bit unclean to put it in map_archdef.
 885   map_archdef["#"..name] = "sizeof("..ctype..")"
 886   -- Add new type and emit shortcut define.
 887   local num = ctypenum + 1
 888   map_type[name] = {
 889     ctype = ctype,
 890     ctypefmt = format("Dt%X(%%s)", num),
 891     reg = reg,
 892   }
 893   wline(format("#define Dt%X(_V) (int)(ptrdiff_t)&(((%s *)0)_V)", num, ctype))
 894   ctypenum = num
 895 end
 896 map_op[".type_2"] = map_op[".type_3"]
 897
 898 -- Dump type definitions.
 899 local function dumptypes(out, lvl)
 900   local t = {}
 901   for name in pairs(map_type) do t[#t+1] = name end
 902   sort(t)
 903   out:write("Type definitions:\n")
 904   for _,name in ipairs(t) do
 905     local tp = map_type[name]
 906     local reg = tp.reg or ""
 907     out:write(format("  %-20s %-20s %s\n", name, tp.ctype, reg))
 908   end
 909   out:write("\n")
 910 end
 911
 912 ------------------------------------------------------------------------------
 913
 914 -- Set the current section.
 915 function _M.section(num)
 916   waction("SECTION", num)
 917   wflush(true) -- SECTION is a terminal action.
 918 end
 919
 920 ------------------------------------------------------------------------------
 921
 922 -- Dump architecture description.
 923 function _M.dumparch(out)
 924   out:write(format("DynASM %s version %s, released %s\n\n",
 925     _info.arch, _info.version, _info.release))
 926   dumpactions(out)
 927 end
 928
 929 -- Dump all user defined elements.
 930 function _M.dumpdef(out, lvl)
 931   dumptypes(out, lvl)
 932   dumpglobals(out, lvl)
 933   dumpexterns(out, lvl)
 934 end
 935
 936 ------------------------------------------------------------------------------
 937
 938 -- Pass callbacks from/to the DynASM core.
 939 function _M.passcb(wl, we, wf, ww)
 940   wline, werror, wfatal, wwarn = wl, we, wf, ww
 941   return wflush
 942 end
 943
 944 -- Setup the arch-specific module.
 945 function _M.setup(arch, opt)
 946   g_arch, g_opt = arch, opt
 947 end
 948
 949 -- Merge the core maps and the arch-specific maps.
 950 function _M.mergemaps(map_coreop, map_def)
 951   setmetatable(map_op, { __index = map_coreop })
 952   setmetatable(map_def, { __index = map_archdef })
 953   return map_op, map_def
 954 end
 955
 956 return _M
 957
 958 ------------------------------------------------------------------------------
 959