aboutsummaryrefslogblamecommitdiffstats
path: root/lib/hipe/x86/hipe_rtl_to_x86.erl
blob: d13f63b1d9afe053030d39ed3d76df298552e5cf (plain) (tree)
1
2
3
4
5
6
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715



                                  
                                                         
    










                                                                            





























































































































































































































                                                                             
                                


























































































































































































































































































































































































































































































                                                                            












                                                             






















































































































































                                                                         
%%% -*- erlang-indent-level: 2 -*-
%%%
%%% %CopyrightBegin%
%%% 
%%% Copyright Ericsson AB 2001-2016. All Rights Reserved.
%%% 
%%% Licensed under the Apache License, Version 2.0 (the "License");
%%% you may not use this file except in compliance with the License.
%%% You may obtain a copy of the License at
%%%
%%%     http://www.apache.org/licenses/LICENSE-2.0
%%%
%%% Unless required by applicable law or agreed to in writing, software
%%% distributed under the License is distributed on an "AS IS" BASIS,
%%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
%%% See the License for the specific language governing permissions and
%%% limitations under the License.
%%% 
%%% %CopyrightEnd%
%%%
%%%
%%% Translate 3-address RTL code to 2-address pseudo-x86 code.

-ifdef(HIPE_AMD64).
-define(HIPE_RTL_TO_X86,	hipe_rtl_to_amd64).
-define(HIPE_X86_REGISTERS,	hipe_amd64_registers).
-define(ECX,			rcx).
-define(EAX,			rax).
-else.
-define(HIPE_RTL_TO_X86,	hipe_rtl_to_x86).
-define(HIPE_X86_REGISTERS,	hipe_x86_registers).
-define(ECX,			ecx).
-define(EAX,			eax).
-endif.

-module(?HIPE_RTL_TO_X86).
-export([translate/1]).

-include("../rtl/hipe_rtl.hrl").

translate(RTL) ->	% RTL function -> x86 defun
  hipe_gensym:init(x86),
  hipe_gensym:set_var(x86, ?HIPE_X86_REGISTERS:first_virtual()),
  hipe_gensym:set_label(x86, hipe_gensym:get_label(rtl)),
  Map0 = vmap_empty(),
  {Formals, Map1} = conv_formals(hipe_rtl:rtl_params(RTL), Map0),
  OldData = hipe_rtl:rtl_data(RTL),
  {Code0, NewData} = conv_insn_list(hipe_rtl:rtl_code(RTL), Map1, OldData),
  {RegFormals,_} = split_args(Formals),
  Code =
    case RegFormals of
      [] -> Code0;
      _ -> [hipe_x86:mk_label(hipe_gensym:get_next_label(x86)) |
	    move_formals(RegFormals, Code0)]
    end,
  IsClosure = hipe_rtl:rtl_is_closure(RTL),
  IsLeaf = hipe_rtl:rtl_is_leaf(RTL),
  hipe_x86:mk_defun(hipe_rtl:rtl_fun(RTL),
		    Formals,
		    IsClosure,
		    IsLeaf,
		    Code,
		    NewData,
		    [],
		    []).

conv_insn_list([H|T], Map, Data) ->
  {NewH, NewMap, NewData1} = conv_insn(H, Map, Data),
  %% io:format("~w \n  ==>\n ~w\n- - - - - - - - -\n",[H,NewH]),
  {NewT, NewData2} = conv_insn_list(T, NewMap, NewData1),
  {NewH ++ NewT, NewData2};
conv_insn_list([], _, Data) ->
  {[], Data}.

conv_insn(I, Map, Data) ->
  case I of
    #alu{} ->
      %% dst = src1 binop src2
      BinOp = conv_binop(hipe_rtl:alu_op(I)),
      {Dst, Map0} = conv_dst(hipe_rtl:alu_dst(I), Map),
      {FixSrc1, Src1, Map1} = conv_src(hipe_rtl:alu_src1(I), Map0),
      {FixSrc2, Src2, Map2} = conv_src(hipe_rtl:alu_src2(I), Map1),
      I2 =
	case hipe_rtl:is_shift_op(hipe_rtl:alu_op(I)) of
	  true ->
	    conv_shift(Dst, Src1, BinOp, Src2);
	  false ->
	    conv_alu(Dst, Src1, BinOp, Src2, [])
	end,
      {FixSrc1++FixSrc2++I2, Map2, Data};
    #alub{} ->
      %% dst = src1 op src2; if COND goto label
      BinOp = conv_binop(hipe_rtl:alub_op(I)),
      {Dst, Map0} = conv_dst(hipe_rtl:alub_dst(I), Map),
      {FixSrc1, Src1, Map1} = conv_src(hipe_rtl:alub_src1(I), Map0),
      {FixSrc2, Src2, Map2} = conv_src(hipe_rtl:alub_src2(I), Map1),
      Cc = conv_cond(hipe_rtl:alub_cond(I)),
      I1 = [hipe_x86:mk_pseudo_jcc(Cc,
				   hipe_rtl:alub_true_label(I),
				   hipe_rtl:alub_false_label(I),
				   hipe_rtl:alub_pred(I))],
      I2 = conv_alu(Dst, Src1, BinOp, Src2, I1),
      {FixSrc1++FixSrc2++I2, Map2, Data};
    #branch{} ->
      %% <unused> = src1 - src2; if COND goto label
      {FixSrc1, Src1, Map0} = conv_src(hipe_rtl:branch_src1(I), Map),
      {FixSrc2, Src2, Map1} = conv_src(hipe_rtl:branch_src2(I), Map0),
      Cc = conv_cond(hipe_rtl:branch_cond(I)),
      I2 = conv_branch(Src1, Cc, Src2,
		       hipe_rtl:branch_true_label(I),
		       hipe_rtl:branch_false_label(I),
		       hipe_rtl:branch_pred(I)),
      {FixSrc1++FixSrc2++I2, Map1, Data};
    #call{} ->
      %%	push <arg1>
      %%	...
      %%	push <argn>
      %%	eax := call <Fun>; if exn goto <Fail> else goto Next
      %% Next:
      %%	<Dst> := eax
      %%	goto <Cont>
      {FixArgs, Args, Map0} = conv_src_list(hipe_rtl:call_arglist(I), Map),
      {Dsts, Map1} = conv_dst_list(hipe_rtl:call_dstlist(I), Map0),
      {Fun, Map2} = conv_fun(hipe_rtl:call_fun(I), Map1),
      I2 = conv_call(Dsts, Fun, Args,
		     hipe_rtl:call_continuation(I),
		     hipe_rtl:call_fail(I),
		     hipe_rtl:call_type(I)),
      %% XXX Fixme: this ++ is probably inefficient.
      {FixArgs++I2, Map2, Data};
    #comment{} ->
      I2 = [hipe_x86:mk_comment(hipe_rtl:comment_text(I))],
      {I2, Map, Data};
    #enter{} ->
      {FixArgs, Args, Map0} = conv_src_list(hipe_rtl:enter_arglist(I), Map),
      {Fun, Map1} = conv_fun(hipe_rtl:enter_fun(I), Map0),
      I2 = conv_tailcall(Fun, Args, hipe_rtl:enter_type(I)),
      {FixArgs++I2, Map1, Data};
    #goto{} ->
      I2 = [hipe_x86:mk_jmp_label(hipe_rtl:goto_label(I))],
      {I2, Map, Data};
    #label{} ->
      I2 = [hipe_x86:mk_label(hipe_rtl:label_name(I))],
      {I2, Map, Data};
    #load{} ->
      {Dst, Map0} = conv_dst(hipe_rtl:load_dst(I), Map),
      {FixSrc, Src, Map1} = conv_src(hipe_rtl:load_src(I), Map0),
      {FixOff, Off, Map2} = conv_src(hipe_rtl:load_offset(I), Map1),
      I2 = case {hipe_rtl:load_size(I), hipe_rtl:load_sign(I)} of
	     {byte, signed} ->
	       [hipe_x86:mk_movsx(hipe_x86:mk_mem(Src, Off, 'byte'), Dst)];
	     {byte, unsigned} ->
	       [hipe_x86:mk_movzx(hipe_x86:mk_mem(Src, Off, 'byte'), Dst)];
	     {int16, signed} ->
	       [hipe_x86:mk_movsx(hipe_x86:mk_mem(Src, Off, 'int16'), Dst)];
	     {int16, unsigned} ->
	       [hipe_x86:mk_movzx(hipe_x86:mk_mem(Src, Off, 'int16'), Dst)];
	     {LoadSize, LoadSign} ->
	       mk_load(LoadSize, LoadSign, Src, Off, Dst)
	   end,
      {FixSrc++FixOff++I2, Map2, Data};
    #load_address{} ->
      {Dst, Map0} = conv_dst(hipe_rtl:load_address_dst(I), Map),
      Addr = hipe_rtl:load_address_addr(I),
      Type = hipe_rtl:load_address_type(I),
      Src = hipe_x86:mk_imm_from_addr(Addr, Type),
      I2 = mk_load_address(Type, Src, Dst),
      {I2, Map0, Data};
    #load_atom{} ->
      {Dst, Map0} = conv_dst(hipe_rtl:load_atom_dst(I), Map),
      Src = hipe_x86:mk_imm_from_atom(hipe_rtl:load_atom_atom(I)),
      I2 = [hipe_x86:mk_move(Src, Dst)],
      {I2, Map0, Data};
    #move{} ->
      {Dst, Map0} = conv_dst(hipe_rtl:move_dst(I), Map),
      {FixSrc, Src, Map1} = conv_src(hipe_rtl:move_src(I), Map0),
      I2 = [hipe_x86:mk_move(Src, Dst)],
      {FixSrc++I2, Map1, Data};
    #return{} ->
      {FixArgs, Args, Map0} = conv_src_list(hipe_rtl:return_varlist(I), Map),
      %% frame will fill in npop later, hence the "mk_ret(-1)"
      I2 = move_retvals(Args, [hipe_x86:mk_ret(-1)]),
      {FixArgs++I2, Map0, Data};
    #store{} ->
      {Ptr, Map0} = conv_dst(hipe_rtl:store_base(I), Map),
      {FixSrc, Src, Map1} = conv_src(hipe_rtl:store_src(I), Map0),
      {FixOff, Off, Map2} = conv_src(hipe_rtl:store_offset(I), Map1),
      I2 = mk_store(hipe_rtl:store_size(I), Src, Ptr, Off),
      {FixSrc++FixOff++I2, Map2, Data};
    #switch{} ->	% this one also updates Data :-(
      %% from hipe_rtl2sparc, but we use a hairy addressing mode
      %% instead of doing the arithmetic manually
      Labels = hipe_rtl:switch_labels(I),
      LMap = [{label,L} || L <- Labels],
      {NewData, JTabLab} =
	case hipe_rtl:switch_sort_order(I) of
	  [] ->
	    hipe_consttab:insert_block(Data, word, LMap);
	  SortOrder ->
	    hipe_consttab:insert_sorted_block(
	      Data, word, LMap, SortOrder)
	end,
      %% no immediates allowed here
      {Index, Map1} = conv_dst(hipe_rtl:switch_src(I), Map),
      I2 = mk_jmp_switch(Index, JTabLab, Labels),
      {I2, Map1, NewData};
    #fload{} ->
      {Dst, Map0} = conv_dst(hipe_rtl:fload_dst(I), Map),
      {[], Src, Map1} = conv_src(hipe_rtl:fload_src(I), Map0),
      {[], Off, Map2} = conv_src(hipe_rtl:fload_offset(I), Map1),
      I2 = [hipe_x86:mk_fmove(hipe_x86:mk_mem(Src, Off, 'double'),Dst)],
      {I2, Map2, Data};
    #fstore{} ->
      {Dst, Map0} = conv_dst(hipe_rtl:fstore_base(I), Map),
      {[], Src, Map1} = conv_src(hipe_rtl:fstore_src(I), Map0),
      {[], Off, Map2} = conv_src(hipe_rtl:fstore_offset(I), Map1),
      I2 = [hipe_x86:mk_fmove(Src, hipe_x86:mk_mem(Dst, Off, 'double'))],
      {I2, Map2, Data};
    #fp{} ->
      {Dst, Map0} = conv_dst(hipe_rtl:fp_dst(I), Map),
      {[], Src1, Map1} = conv_src(hipe_rtl:fp_src1(I), Map0),
      {[], Src2, Map2} = conv_src(hipe_rtl:fp_src2(I), Map1),
      FpBinOp = conv_fp_binop(hipe_rtl:fp_op(I)),
      I2 = conv_fp_binary(Dst, Src1, FpBinOp, Src2),
      {I2, Map2, Data};
    #fp_unop{} ->
      {Dst, Map0} = conv_dst(hipe_rtl:fp_unop_dst(I), Map),
      {[], Src, Map1} = conv_src(hipe_rtl:fp_unop_src(I), Map0),
      FpUnOp = conv_fp_unop(hipe_rtl:fp_unop_op(I)),
      I2 = conv_fp_unary(Dst, Src, FpUnOp),
      {I2, Map1, Data};
    #fmove{} ->
      {Dst, Map0} = conv_dst(hipe_rtl:fmove_dst(I), Map),
      {[], Src, Map1} = conv_src(hipe_rtl:fmove_src(I), Map0),
      I2 = [hipe_x86:mk_fmove(Src, Dst)],
      {I2, Map1, Data};
    #fconv{} ->
      {Dst, Map0} = conv_dst(hipe_rtl:fconv_dst(I), Map),
      {[], Src, Map1} = conv_src(hipe_rtl:fconv_src(I), Map0),
      I2 = conv_fconv(Dst, Src),
      {I2, Map1, Data};
    X ->
      %% gctest??
      %% jmp, jmp_link, jsr, esr, multimove,
      %% stackneed, pop_frame, restore_frame, save_frame
      throw({?MODULE, {"unknown RTL instruction", X}})
  end.

%%% Finalise the conversion of a 3-address ALU operation, taking
%%% care to not introduce more temps and moves than necessary.

conv_alu(Dst, Src1, 'imul', Src2, Tail) ->
  mk_imul(Src1, Src2, Dst, Tail);
conv_alu(Dst, Src1, BinOp, Src2, Tail) ->
  case same_opnd(Dst, Src1) of
    true ->		% x = x op y
      [hipe_x86:mk_alu(BinOp, Src2, Dst) | Tail];	% x op= y
    false ->		% z = x op y, where z != x
      case same_opnd(Dst, Src2) of
	false ->	% z = x op y, where z != x && z != y
	  [hipe_x86:mk_move(Src1, Dst),			% z = x
	   hipe_x86:mk_alu(BinOp, Src2, Dst) | Tail];	% z op= y
	true ->		% y = x op y, where y != x
	  case binop_commutes(BinOp) of
	    true ->	% y = y op x
	      [hipe_x86:mk_alu(BinOp, Src1, Dst) | Tail]; % y op= x
	    false ->	% y = x op y, where op doesn't commute
	      Tmp = clone_dst(Dst),
	      [hipe_x86:mk_move(Src1, Tmp),		% t = x
	       hipe_x86:mk_alu(BinOp, Src2, Tmp),	% t op= y
	       hipe_x86:mk_move(Tmp, Dst) | Tail]	% y = t
	  end
      end
  end.

mk_imul(Src1, Src2, Dst, Tail) ->
  case hipe_x86:is_imm(Src1) of
    true ->
      case hipe_x86:is_imm(Src2) of
	true ->
	  mk_imul_iit(Src1, Src2, Dst, Tail);
	_ ->
	  mk_imul_itt(Src1, Src2, Dst, Tail)
      end;
    _ ->
      case hipe_x86:is_imm(Src2) of
	true ->
	  mk_imul_itt(Src2, Src1, Dst, Tail);
	_ ->
	  mk_imul_ttt(Src1, Src2, Dst, Tail)
      end
  end.

mk_imul_iit(Src1, Src2, Dst, Tail) ->
  io:format("~w: RTL mul with two immediates\n", [?MODULE]),
  Tmp2 = new_untagged_temp(),
  [hipe_x86:mk_move(Src2, Tmp2) |
   mk_imul_itt(Src1, Tmp2, Dst, Tail)].

mk_imul_itt(Src1, Src2, Dst, Tail) ->
  [hipe_x86:mk_imul(Src1, Src2, Dst) | Tail].

mk_imul_ttt(Src1, Src2, Dst, Tail) ->
  case same_opnd(Dst, Src1) of
    true ->
      [hipe_x86:mk_imul([], Src2, Dst) | Tail];
    false ->
      case same_opnd(Dst, Src2) of
	true ->
	  [hipe_x86:mk_imul([], Src1, Dst) | Tail];
	false ->
	  [hipe_x86:mk_move(Src1, Dst),
	   hipe_x86:mk_imul([], Src2, Dst) | Tail]
      end
  end.

conv_shift(Dst, Src1, BinOp, Src2) ->
  {NewSrc2,I1} =
    case hipe_x86:is_imm(Src2) of
      true ->
	{Src2, []};
      false ->
	NewSrc = hipe_x86:mk_temp(?HIPE_X86_REGISTERS:?ECX(), 'untagged'),
	{NewSrc, [hipe_x86:mk_move(Src2, NewSrc)]}
    end,
  I2 = case same_opnd(Dst, Src1) of
	 true ->	% x = x op y
	   [hipe_x86:mk_shift(BinOp, NewSrc2, Dst)];	% x op= y
	 false ->	% z = x op y, where z != x
	   case same_opnd(Dst, Src2) of
	     false ->	% z = x op y, where z != x && z != y
	       [hipe_x86:mk_move(Src1, Dst),		% z = x
		hipe_x86:mk_shift(BinOp, NewSrc2, Dst)];% z op= y
	     true ->	% y = x op y, no shift op commutes
	       Tmp = clone_dst(Dst),
	       [hipe_x86:mk_move(Src1, Tmp),		% t = x
		hipe_x86:mk_shift(BinOp, NewSrc2, Tmp),	% t op= y
		hipe_x86:mk_move(Tmp, Dst)]		% y = t
	   end
       end,
  I1 ++ I2.

%%% Finalise the conversion of a conditional branch operation, taking
%%% care to not introduce more temps and moves than necessary.

conv_branch(Src1, Cc, Src2, TrueLab, FalseLab, Pred) ->
  case hipe_x86:is_imm(Src1) of
    false ->
      mk_branch(Src1, Cc, Src2, TrueLab, FalseLab, Pred);
    true ->
      case hipe_x86:is_imm(Src2) of
	false ->
	  NewCc = commute_cc(Cc),
	  mk_branch(Src2, NewCc, Src1, TrueLab, FalseLab, Pred);
	true ->
	  %% two immediates, let the optimiser clean it up
	  Tmp = new_untagged_temp(),
	  [hipe_x86:mk_move(Src1, Tmp) |
	   mk_branch(Tmp, Cc, Src2, TrueLab, FalseLab, Pred)]
      end
  end.

mk_branch(Src1, Cc, Src2, TrueLab, FalseLab, Pred) ->
  %% PRE: not(is_imm(Src1))
  [hipe_x86:mk_cmp(Src2, Src1),
   hipe_x86:mk_pseudo_jcc(Cc, TrueLab, FalseLab, Pred)].

%%% Convert an RTL ALU or ALUB binary operator.

conv_binop(BinOp) ->
  case BinOp of
    'add'	-> 'add';
    'sub'	-> 'sub';
    'or'	-> 'or';
    'and'	-> 'and';
    'xor'	-> 'xor';
    'sll'	-> 'shl';
    'srl'	-> 'shr';
    'sra'	-> 'sar';
    'mul'	-> 'imul';
    %% andnot ???
    _		-> exit({?MODULE, {"unknown binop", BinOp}})
  end.

binop_commutes(BinOp) ->
  case BinOp of
    'add'	-> true;
    'or'	-> true;
    'and'	-> true;
    'xor'	-> true;
    _		-> false
  end.

%%% Convert an RTL conditional operator.

conv_cond(Cond) ->
  case Cond of
    eq	-> 'e';
    ne	-> 'ne';
    gt	-> 'g';
    gtu	-> 'a';
    ge	-> 'ge';
    geu	-> 'ae';
    lt	-> 'l';
    ltu	-> 'b';
    le	-> 'le';
    leu	-> 'be';
    overflow -> 'o';
    not_overflow -> 'no';
    _	-> exit({?MODULE, {"unknown rtl cond", Cond}})
  end.

commute_cc(Cc) ->	% if x Cc y, then y commute_cc(Cc) x
  case Cc of
    'e'	-> 'e';		% ==, ==
    'ne' -> 'ne';	% !=, !=
    'g'	-> 'l';		% >, <
    'a'	-> 'b';		% >u, <u
    'ge' -> 'le';	% >=, <=
    'ae' -> 'be';	% >=u, <=u
    'l'	-> 'g';		% <, >
    'b'	-> 'a';		% <u, >u
    'le' -> 'ge';	% <=, >=
    'be' -> 'ae';	% <=u, >=u
    %% overflow/not_overflow: n/a
    _	-> exit({?MODULE, {"unknown cc", Cc}})
  end.

%%% Test if Dst and Src are the same operand.

same_opnd(Dst, Src) -> Dst =:= Src.

%%% Finalise the conversion of a tailcall instruction.

conv_tailcall(Fun, Args, Linkage) ->
  Arity = length(Args),
  {RegArgs,StkArgs} = split_args(Args),
  move_actuals(RegArgs,
	       [hipe_x86:mk_pseudo_tailcall_prepare(),
		hipe_x86:mk_pseudo_tailcall(Fun, Arity, StkArgs, Linkage)]).

split_args(Args) ->
  split_args(0, ?HIPE_X86_REGISTERS:nr_args(), Args, []).
split_args(I, N, [Arg|Args], RegArgs) when I < N ->
  Reg = ?HIPE_X86_REGISTERS:arg(I),
  Temp = hipe_x86:mk_temp(Reg, 'tagged'),
  split_args(I+1, N, Args, [{Arg,Temp}|RegArgs]);
split_args(_, _, StkArgs, RegArgs) ->
  {RegArgs, StkArgs}.

move_actuals([], Rest) -> Rest;
move_actuals([{Src,Dst}|Actuals], Rest) ->
  move_actuals(Actuals, [hipe_x86:mk_move(Src, Dst) | Rest]).

move_formals([], Rest) -> Rest;
move_formals([{Dst,Src}|Formals], Rest) ->
  move_formals(Formals, [hipe_x86:mk_move(Src, Dst) | Rest]).

%%% Finalise the conversion of a call instruction.

conv_call(Dsts, Fun, Args, ContLab, ExnLab, Linkage) ->
  case hipe_x86:is_prim(Fun) of
    true ->
      conv_primop_call(Dsts, Fun, Args, ContLab, ExnLab, Linkage);
    false ->
      conv_general_call(Dsts, Fun, Args, ContLab, ExnLab, Linkage)
  end.

conv_primop_call(Dsts, Prim, Args, ContLab, ExnLab, Linkage) ->
  case hipe_x86:prim_prim(Prim) of
    'fwait' ->
      conv_fwait_call(Dsts, Args, ContLab, ExnLab, Linkage);
    _ ->
      conv_general_call(Dsts, Prim, Args, ContLab, ExnLab, Linkage)
  end.

conv_fwait_call([], [], [], [], not_remote) ->
  [hipe_x86:mk_fp_unop('fwait', [])].

conv_general_call(Dsts, Fun, Args, ContLab, ExnLab, Linkage) ->
  %% The backend does not support pseudo_calls without a
  %% continuation label, so we make sure each call has one.
  {RealContLab, Tail} =
    case do_call_results(Dsts) of
      [] ->
	%% Avoid consing up a dummy basic block if the moves list
	%% is empty, as is typical for calls to suspend/0.
	%% This should be subsumed by a general "optimise the CFG"
	%% module, and could probably be removed.
	case ContLab of
	  [] ->
	    NewContLab = hipe_gensym:get_next_label(x86),
	    {NewContLab, [hipe_x86:mk_label(NewContLab)]};
	  _ ->
	    {ContLab, []}
	end;
      Moves ->
	%% Change the call to continue at a new basic block.
	%% In this block move the result registers to the Dsts,
	%% then continue at the call's original continuation.
	%%
	%% This should be fixed to propagate "fallthrough calls"
	%% When the rest of the backend supports them.
	NewContLab = hipe_gensym:get_next_label(x86),
	case ContLab of
	  [] ->
	    %% This is just a fallthrough
	    %% No jump back after the moves.
	    {NewContLab,
	     [hipe_x86:mk_label(NewContLab) |
	      Moves]};
	  _ ->
	    %% The call has a continuation
	    %% jump to it.
	    {NewContLab,
	     [hipe_x86:mk_label(NewContLab) |
	      Moves ++
	      [hipe_x86:mk_jmp_label(ContLab)]]}
	end
    end,
  SDesc = hipe_x86:mk_sdesc(ExnLab, 0, length(Args), {}),
  CallInsn = hipe_x86:mk_pseudo_call(Fun, SDesc, RealContLab, Linkage),
  {RegArgs,StkArgs} = split_args(Args),
  do_push_args(StkArgs, move_actuals(RegArgs, [CallInsn | Tail])).

do_push_args([Arg|Args], Tail) ->
  [hipe_x86:mk_push(Arg) | do_push_args(Args, Tail)];
do_push_args([], Tail) ->
  Tail.

%%% Move return values from the return value registers.

do_call_results(DstList) ->
  do_call_results(DstList, 0, []).

do_call_results([Dst|DstList], I, Rest) ->
  Src = hipe_x86:mk_temp(?HIPE_X86_REGISTERS:ret(I), 'tagged'),
  Move = hipe_x86:mk_move(Src, Dst),
  do_call_results(DstList, I+1, [Move|Rest]);
do_call_results([], _, Insns) -> Insns.

%%% Move return values to the return value registers.

move_retvals(SrcLst, Rest) ->
  move_retvals(SrcLst, 0, Rest).

move_retvals([Src|SrcLst], I, Rest) ->
  Dst = hipe_x86:mk_temp(?HIPE_X86_REGISTERS:ret(I), 'tagged'),
  Move = hipe_x86:mk_move(Src, Dst),
  move_retvals(SrcLst, I+1, [Move|Rest]);
move_retvals([], _, Insns) -> Insns.

%%% Convert a 'fun' operand (MFA, prim, or temp)

conv_fun(Fun, Map) ->
  case hipe_rtl:is_var(Fun) of
    true ->
      conv_dst(Fun, Map);
    false ->
      case hipe_rtl:is_reg(Fun) of
	true ->
	  conv_dst(Fun, Map);
	false ->
	  case Fun of
	    Prim when is_atom(Prim) ->
	      {hipe_x86:mk_prim(Prim), Map};
	    {M,F,A} when is_atom(M), is_atom(F), is_integer(A) ->
	      {hipe_x86:mk_mfa(M,F,A), Map};
	    _ ->
	      exit({?MODULE,conv_fun,Fun})
	  end
      end
  end.

%%% Convert an RTL source operand (imm/var/reg).

conv_src(Opnd, Map) ->
  case hipe_rtl:is_imm(Opnd) of
    true ->
      conv_imm(Opnd, Map);
    false ->
      {NewOpnd,NewMap} = conv_dst(Opnd, Map),
      {[], NewOpnd, NewMap}
  end.

-ifdef(HIPE_AMD64).
conv_imm(Opnd, Map) ->
  ImmVal = hipe_rtl:imm_value(Opnd),
  case is_imm64(ImmVal) of
    true ->
      Temp = hipe_x86:mk_new_temp('untagged'),
      {[hipe_x86:mk_move64(hipe_x86:mk_imm(ImmVal), Temp)], Temp, Map};
    false ->
      {[], hipe_x86:mk_imm(ImmVal), Map}
  end.

is_imm64(Value) when is_integer(Value) ->
  (Value < -(1 bsl (32 - 1))) or (Value > (1 bsl (32 - 1)) - 1);
is_imm64({_,atom})    -> false; % Atoms are 32 bits.
is_imm64({_,c_const}) -> false; % c_consts are 32 bits.
is_imm64({_,_})       -> true . % Other relocs are 64 bits.
-else.
conv_imm(Opnd, Map) ->
  {[], hipe_x86:mk_imm(hipe_rtl:imm_value(Opnd)), Map}.
-endif.

conv_src_list([O|Os], Map) ->
  {NewInstr, V, Map1} = conv_src(O, Map),
  {Instrs, Vs, Map2} = conv_src_list(Os, Map1),
  {Instrs++NewInstr, [V|Vs], Map2};
conv_src_list([], Map) ->
  {[], [], Map}.

%%% Convert an RTL destination operand (var/reg).

conv_dst(Opnd, Map) ->
  {Name, Type} =
    case hipe_rtl:is_var(Opnd) of
      true ->
	{hipe_rtl:var_index(Opnd), 'tagged'};
      false ->
	case hipe_rtl:is_fpreg(Opnd) of
	  true ->
	    {hipe_rtl:fpreg_index(Opnd), 'double'};
	  false ->
	    {hipe_rtl:reg_index(Opnd), 'untagged'}
	end
    end,
  case ?HIPE_X86_REGISTERS:is_precoloured(Name) of
    true ->
      case ?HIPE_X86_REGISTERS:proc_offset(Name) of
	false ->
	  {hipe_x86:mk_temp(Name, Type), Map};
	Offset ->
	  Preg = ?HIPE_X86_REGISTERS:proc_pointer(),
	  Pbase = hipe_x86:mk_temp(Preg, 'untagged'),
	  Poff = hipe_x86:mk_imm(Offset),
	  {hipe_x86:mk_mem(Pbase, Poff, Type), Map}
      end;
    false ->
      case vmap_lookup(Map, Opnd) of
	{value, NewTemp} ->
	  {NewTemp, Map};
	_ ->
	  NewTemp = hipe_x86:mk_new_temp(Type),
	  {NewTemp, vmap_bind(Map, Opnd, NewTemp)}
      end
  end.

conv_dst_list([O|Os], Map) ->
  {Dst, Map1} = conv_dst(O, Map),
  {Dsts, Map2} = conv_dst_list(Os, Map1),
  {[Dst|Dsts], Map2};
conv_dst_list([], Map) ->
  {[], Map}.

conv_formals(Os, Map) ->
  conv_formals(?HIPE_X86_REGISTERS:nr_args(), Os, Map, []).

conv_formals(N, [O|Os], Map, Res) ->
  Type =
    case hipe_rtl:is_var(O) of
      true -> 'tagged';
      false ->'untagged'
    end,
  Dst =
    if N > 0 -> hipe_x86:mk_new_temp(Type);	% allocatable
       true -> hipe_x86:mk_new_nonallocatable_temp(Type)
    end,
  Map1 = vmap_bind(Map, O, Dst),
  conv_formals(N-1, Os, Map1, [Dst|Res]);
conv_formals(_, [], Map, Res) ->
  {lists:reverse(Res), Map}.

%%% typeof_src -- what's src's type?

typeof_src(Src) ->
  case hipe_x86:is_imm(Src) of
    true ->
      'untagged';
    _ ->
      typeof_dst(Src)
  end.

%%% typeof_dst -- what's dst's type?

typeof_dst(Dst) ->
  case hipe_x86:is_temp(Dst) of
    true ->
      hipe_x86:temp_type(Dst);
    _ ->
      hipe_x86:mem_type(Dst)
  end.

%%% clone_dst -- conjure up a scratch reg with same type as dst

clone_dst(Dst) ->
  hipe_x86:mk_new_temp(typeof_dst(Dst)).

%%% new_untagged_temp -- conjure up an untagged scratch reg

new_untagged_temp() ->
  hipe_x86:mk_new_temp('untagged').

%%% Map from RTL var/reg operands to x86 temps.

vmap_empty() ->
  gb_trees:empty().

vmap_lookup(Map, Key) ->
  gb_trees:lookup(Key, Map).

vmap_bind(Map, Key, Val) ->
  gb_trees:insert(Key, Val, Map).

%%% Finalise the conversion of an Integer-to-Float operation.

conv_fconv(Dst, Src) ->
  case hipe_x86:is_imm(Src) of
    false ->
      [hipe_x86:mk_fmove(Src, Dst)];
    true ->
      %% cvtsi2sd does not allow src to be an immediate
      Tmp = new_untagged_temp(),
      [hipe_x86:mk_move(Src, Tmp),
       hipe_x86:mk_fmove(Tmp, Dst)]
  end.

%%% Finalise the conversion of a 2-address FP operation.

conv_fp_unary(Dst, Src, FpUnOp) ->
  case same_opnd(Dst, Src) of
    true ->
      [hipe_x86:mk_fp_unop(FpUnOp, Dst)];
    _ ->
      [hipe_x86:mk_fmove(Src, Dst),
       hipe_x86:mk_fp_unop(FpUnOp, Dst)]
  end.

conv_fp_unop(RtlFpUnOp) ->
  case RtlFpUnOp of
    'fchs' -> 'fchs'
  end.

%%% Finalise the conversion of a 3-address FP operation.

conv_fp_binary(Dst, Src1, FpBinOp, Src2) ->
  case same_opnd(Dst, Src1) of
    true ->		% x = x op y
      [hipe_x86:mk_fp_binop(FpBinOp, Src2, Dst)];		% x op= y
    false ->		% z = x op y, where z != x
      case same_opnd(Dst, Src2) of
	false ->	% z = x op y, where z != x && z != y
	  [hipe_x86:mk_fmove(Src1, Dst),			% z = x
	   hipe_x86:mk_fp_binop(FpBinOp, Src2, Dst)];		% z op= y
	true ->		% y = x op y, where y != x
	  case fp_binop_commutes(FpBinOp) of
	    true ->	% y = y op x
	      [hipe_x86:mk_fp_binop(FpBinOp, Src1, Dst)];	% y op= x
	    false ->	% y = x op y, where op doesn't commute
	      RevFpBinOp = reverse_fp_binop(FpBinOp),
	      [hipe_x86:mk_fp_binop(RevFpBinOp, Src1, Dst)]
	  end
      end
  end.

%%% Convert an RTL FP binary operator.

conv_fp_binop(RtlFpBinOp) ->
  case RtlFpBinOp of
    'fadd' -> 'fadd';
    'fdiv' -> 'fdiv';
    'fmul' -> 'fmul';
    'fsub' -> 'fsub'
  end.

fp_binop_commutes(FpBinOp) ->
  case FpBinOp of
    'fadd'	-> true;
    'fmul'	-> true;
    _		-> false
  end.

reverse_fp_binop(FpBinOp) ->
  case FpBinOp of
    'fsub' -> 'fsubr';
    'fdiv' -> 'fdivr'
  end.

%%% Create a jmp_switch instruction.

-ifdef(HIPE_AMD64).
mk_jmp_switch(Index, JTabLab, Labels) ->
  JTabReg = hipe_x86:mk_new_temp('untagged'),
  JTabImm = hipe_x86:mk_imm_from_addr(JTabLab, constant),
  [hipe_x86:mk_move64(JTabImm, JTabReg),
   hipe_x86:mk_jmp_switch(Index, JTabReg, Labels)].
-else.
mk_jmp_switch(Index, JTabLab, Labels) ->
  %% this is equivalent to "jmp *JTabLab(,Index,4)"
  %% ("r = Index; r *= 4; r += &JTab; jmp *r" isn't as nice)
  [hipe_x86:mk_jmp_switch(Index, JTabLab, Labels)].
-endif.

%%% Finalise the translation of a load_address instruction.

-ifdef(HIPE_AMD64).
mk_load_address(Type, Src, Dst) ->
  case Type of
    c_const -> % 32 bits
      [hipe_x86:mk_move(Src, Dst)];
    _ ->
      [hipe_x86:mk_move64(Src, Dst)]
  end.
-else.
mk_load_address(_Type, Src, Dst) ->
  [hipe_x86:mk_move(Src, Dst)].
-endif.

%%% Translate 32-bit and larger loads.

-ifdef(HIPE_AMD64).
mk_load(LoadSize, LoadSign, Src, Off, Dst) ->
  case {LoadSize, LoadSign} of
    {int32, signed} ->
      [hipe_x86:mk_movsx(hipe_x86:mk_mem(Src, Off, 'int32'), Dst)];
    {int32, unsigned} ->
      %% The processor zero-extends for us. No need for 'movzx'.
      [hipe_x86:mk_move(hipe_x86:mk_mem(Src, Off, 'int32'), Dst)];
    {_, _} ->
      mk_load_word(Src, Off, Dst)
  end.
-else.
mk_load(_LoadSize, _LoadSign, Src, Off, Dst) ->
  mk_load_word(Src, Off, Dst).
-endif.

mk_load_word(Src, Off, Dst) ->
  Type = typeof_dst(Dst),
  [hipe_x86:mk_move(hipe_x86:mk_mem(Src, Off, Type), Dst)].

%%% Finalise the translation of a store instruction.

-ifdef(HIPE_AMD64).
mk_store(RtlStoreSize, Src, Ptr, Off) ->
  Type = case RtlStoreSize of
	   word ->
	     typeof_src(Src);
	   OtherType ->
	     OtherType
	 end,
  [hipe_x86:mk_move(Src, hipe_x86:mk_mem(Ptr, Off, Type))].
-else.
mk_store(RtlStoreSize, Src, Ptr, Off) ->
  case RtlStoreSize of
    word ->
      Type = typeof_src(Src),
      [hipe_x86:mk_move(Src, hipe_x86:mk_mem(Ptr, Off, Type))];
    int32 ->
      Type = typeof_src(Src),
      [hipe_x86:mk_move(Src, hipe_x86:mk_mem(Ptr, Off, Type))];
    int16 ->
      Type = 'int16',
      [hipe_x86:mk_move(Src, hipe_x86:mk_mem(Ptr, Off, Type))];
    byte ->
      Type = 'byte',
      {NewSrc, I1} = conv_small_store(Src),
      I1 ++ [hipe_x86:mk_move(NewSrc, hipe_x86:mk_mem(Ptr, Off, Type))]
  end.

conv_small_store(Src) ->
  case hipe_x86:is_imm(Src) of
    true ->
      {Src, []};
    false ->
      NewSrc = hipe_x86:mk_temp(hipe_x86_registers:eax(), 'untagged'),
      {NewSrc, [hipe_x86:mk_move(Src, NewSrc)]}
  end.
-endif.