Interpreter.c 42.3 KB
Newer Older
1
/* -----------------------------------------------------------------------------
2
 * Bytecode interpreter
3
 *
4
 * Copyright (c) The GHC Team, 1994-2002.
5
6
 * ---------------------------------------------------------------------------*/

7
#include "PosixSource.h"
8
9
#include "Rts.h"
#include "RtsAPI.h"
10
#include "RtsUtils.h"
11
12
13
14
#include "Closures.h"
#include "TSO.h"
#include "Schedule.h"
#include "RtsFlags.h"
15
#include "LdvProfile.h"
16
#include "Updates.h"
17
#include "Sanity.h"
18
#include "Liveness.h"
19
#include "Prelude.h"
20

21
#include "Bytecodes.h"
22
23
#include "Printer.h"
#include "Disassembler.h"
24
#include "Interpreter.h"
25

26
27
28
29
30
#include <string.h>     /* for memcpy */
#ifdef HAVE_ERRNO_H
#include <errno.h>
#endif

31
#include "ffi.h"
andy's avatar
andy committed
32

33
/* --------------------------------------------------------------------------
34
 * The bytecode interpreter
35
36
 * ------------------------------------------------------------------------*/

37
38
39
40
41
42
/* Gather stats about entry, opcode, opcode-pair frequencies.  For
   tuning the interpreter. */

/* #define INTERP_STATS */


43
/* Sp points to the lowest live word on the stack. */
44

45
#define BCO_NEXT      instrs[bciPtr++]
46
47
48
49
50
51
52
#define BCO_NEXT_32   (bciPtr += 2, (((StgWord) instrs[bciPtr-2]) << 16) + ((StgWord) instrs[bciPtr-1]))
#define BCO_NEXT_64   (bciPtr += 4, (((StgWord) instrs[bciPtr-4]) << 48) + (((StgWord) instrs[bciPtr-3]) << 32) + (((StgWord) instrs[bciPtr-2]) << 16) + ((StgWord) instrs[bciPtr-1]))
#if WORD_SIZE_IN_BITS == 32
#define BCO_NEXT_WORD BCO_NEXT_32
#elif WORD_SIZE_IN_BITS == 64
#define BCO_NEXT_WORD BCO_NEXT_64
#else
Ian Lynagh's avatar
Ian Lynagh committed
53
#error Cannot cope with WORD_SIZE_IN_BITS being nether 32 nor 64
54
55
56
#endif
#define BCO_GET_LARGE_ARG ((bci & bci_FLAG_LARGE_ARGS) ? BCO_NEXT_WORD : BCO_NEXT)

57
#define BCO_PTR(n)    (W_)ptrs[n]
58
#define BCO_LIT(n)    literals[n]
59

60
61
62
63
#define LOAD_STACK_POINTERS					\
    Sp = cap->r.rCurrentTSO->sp;				\
    /* We don't change this ... */				\
    SpLim = cap->r.rCurrentTSO->stack + RESERVED_STACK_WORDS;
64

65
#define SAVE_STACK_POINTERS			\
66
    ASSERT(Sp > SpLim); \
67
    cap->r.rCurrentTSO->sp = Sp
68

69
#define RETURN_TO_SCHEDULER(todo,retcode)	\
70
71
   SAVE_STACK_POINTERS;				\
   cap->r.rCurrentTSO->what_next = (todo);	\
72
   threadPaused(cap,cap->r.rCurrentTSO);		\
73
74
   cap->r.rRet = (retcode);			\
   return cap;
75
76

#define RETURN_TO_SCHEDULER_NO_PAUSE(todo,retcode)	\
77
78
79
80
   SAVE_STACK_POINTERS;					\
   cap->r.rCurrentTSO->what_next = (todo);		\
   cap->r.rRet = (retcode);				\
   return cap;
81
82


sof's avatar
sof committed
83
STATIC_INLINE StgPtr
84
allocate_NONUPD (int n_words)
85
{
86
    return allocate(stg_max(sizeofW(StgHeader)+MIN_PAYLOAD_SIZE, n_words));
87
88
}

89
90
int rts_stop_next_breakpoint = 0;
int rts_stop_on_exception = 0;
91

92
#ifdef INTERP_STATS
93

94
95
96
97
98
99
100
101
102
103
104
105
106
/* Hacky stats, for tuning the interpreter ... */
int it_unknown_entries[N_CLOSURE_TYPES];
int it_total_unknown_entries;
int it_total_entries;

int it_retto_BCO;
int it_retto_UPDATE;
int it_retto_other;

int it_slides;
int it_insns;
int it_BCO_entries;

107
108
int it_ofreq[27];
int it_oofreq[27][27];
109
110
int it_lastopc;

111

112
113
#define INTERP_TICK(n) (n)++

114
115
116
117
118
119
120
121
void interp_startup ( void )
{
   int i, j;
   it_retto_BCO = it_retto_UPDATE = it_retto_other = 0;
   it_total_entries = it_total_unknown_entries = 0;
   for (i = 0; i < N_CLOSURE_TYPES; i++)
      it_unknown_entries[i] = 0;
   it_slides = it_insns = it_BCO_entries = 0;
122
123
124
   for (i = 0; i < 27; i++) it_ofreq[i] = 0;
   for (i = 0; i < 27; i++) 
     for (j = 0; j < 27; j++)
125
126
127
128
129
130
131
        it_oofreq[i][j] = 0;
   it_lastopc = 0;
}

void interp_shutdown ( void )
{
   int i, j, k, o_max, i_max, j_max;
132
   debugBelch("%d constrs entered -> (%d BCO, %d UPD, %d ??? )\n",
133
134
                   it_retto_BCO + it_retto_UPDATE + it_retto_other,
                   it_retto_BCO, it_retto_UPDATE, it_retto_other );
135
   debugBelch("%d total entries, %d unknown entries \n", 
136
137
138
                   it_total_entries, it_total_unknown_entries);
   for (i = 0; i < N_CLOSURE_TYPES; i++) {
     if (it_unknown_entries[i] == 0) continue;
139
     debugBelch("   type %2d: unknown entries (%4.1f%%) == %d\n",
140
141
142
143
	     i, 100.0 * ((double)it_unknown_entries[i]) / 
                        ((double)it_total_unknown_entries),
             it_unknown_entries[i]);
   }
144
   debugBelch("%d insns, %d slides, %d BCO_entries\n", 
145
                   it_insns, it_slides, it_BCO_entries);
146
   for (i = 0; i < 27; i++) 
147
      debugBelch("opcode %2d got %d\n", i, it_ofreq[i] );
148
149
150
151

   for (k = 1; k < 20; k++) {
      o_max = 0;
      i_max = j_max = 0;
152
153
      for (i = 0; i < 27; i++) {
         for (j = 0; j < 27; j++) {
154
155
156
157
158
159
160
	    if (it_oofreq[i][j] > o_max) {
               o_max = it_oofreq[i][j];
	       i_max = i; j_max = j;
	    }
	 }
      }
      
161
      debugBelch("%d:  count (%4.1f%%) %6d   is %d then %d\n",
162
163
164
165
166
167
168
                k, ((double)o_max) * 100.0 / ((double)it_insns), o_max,
                   i_max, j_max );
      it_oofreq[i_max][j_max] = 0;

   }
}

169
170
171
172
173
#else // !INTERP_STATS

#define INTERP_TICK(n) /* nothing */

#endif
174

175
176
177
178
179
180
181
182
183
static StgWord app_ptrs_itbl[] = {
    (W_)&stg_ap_p_info,
    (W_)&stg_ap_pp_info,
    (W_)&stg_ap_ppp_info,
    (W_)&stg_ap_pppp_info,
    (W_)&stg_ap_ppppp_info,
    (W_)&stg_ap_pppppp_info,
};

184
HsStablePtr rts_breakpoint_io_action; // points to the IO action which is executed on a breakpoint
185
186
                                // it is set in main/GHC.hs:runStmt

187
Capability *
188
interpretBCO (Capability* cap)
189
{
190
191
192
193
    // Use of register here is primarily to make it clear to compilers
    // that these entities are non-aliasable.
    register StgPtr       Sp;    // local state -- stack pointer
    register StgPtr       SpLim; // local state -- stack lim pointer
194
    register StgClosure   *tagged_obj = 0, *obj;
195
    nat n, m;
196

197
198
    LOAD_STACK_POINTERS;

199
200
201
202
203
204
205
206
207
208
209
    // ------------------------------------------------------------------------
    // Case 1:
    // 
    //       We have a closure to evaluate.  Stack looks like:
    //       
    //      	|   XXXX_info   |
    //      	+---------------+
    //       Sp |      -------------------> closure
    //      	+---------------+
    //       
    if (Sp[0] == (W_)&stg_enter_info) {
210
211
       Sp++;
       goto eval;
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
    }

    // ------------------------------------------------------------------------
    // Case 2:
    // 
    //       We have a BCO application to perform.  Stack looks like:
    //
    //      	|     ....      |
    //      	+---------------+
    //      	|     arg1      |
    //      	+---------------+
    //      	|     BCO       |
    //      	+---------------+
    //       Sp |   RET_BCO     |
    //      	+---------------+
    //       
    else if (Sp[0] == (W_)&stg_apply_interp_info) {
Simon Marlow's avatar
Simon Marlow committed
229
	obj = UNTAG_CLOSURE((StgClosure *)Sp[1]);
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
	Sp += 2;
	goto run_BCO_fun;
    }

    // ------------------------------------------------------------------------
    // Case 3:
    //
    //       We have an unboxed value to return.  See comment before
    //       do_return_unboxed, below.
    //
    else {
	goto do_return_unboxed;
    }

    // Evaluate the object on top of the stack.
eval:
246
    tagged_obj = (StgClosure*)Sp[0]; Sp++;
247
248

eval_obj:
249
    obj = UNTAG_CLOSURE(tagged_obj);
250
251
252
    INTERP_TICK(it_total_evals);

    IF_DEBUG(interpreter,
253
             debugBelch(
254
             "\n---------------------------------------------------------------\n");
255
256
257
             debugBelch("Evaluating: "); printObj(obj);
             debugBelch("Sp = %p\n", Sp);
             debugBelch("\n" );
258

259
             printStackChunk(Sp,cap->r.rCurrentTSO->stack+cap->r.rCurrentTSO->stack_size);
260
             debugBelch("\n\n");
261
            );
262

263
    IF_DEBUG(sanity,checkStackChunk(Sp, cap->r.rCurrentTSO->stack+cap->r.rCurrentTSO->stack_size));
264

265
    switch ( get_itbl(obj)->type ) {
266

267
268
269
270
271
272
    case IND:
    case IND_OLDGEN:
    case IND_PERM:
    case IND_OLDGEN_PERM:
    case IND_STATIC:
    { 
273
	tagged_obj = ((StgInd*)obj)->indirectee;
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
	goto eval_obj;
    }
    
    case CONSTR:
    case CONSTR_1_0:
    case CONSTR_0_1:
    case CONSTR_2_0:
    case CONSTR_1_1:
    case CONSTR_0_2:
    case CONSTR_STATIC:
    case CONSTR_NOCAF_STATIC:
    case FUN:
    case FUN_1_0:
    case FUN_0_1:
    case FUN_2_0:
    case FUN_1_1:
    case FUN_0_2:
    case FUN_STATIC:
    case PAP:
	// already in WHNF
	break;
	
    case BCO:
297
    {
298
	ASSERT(((StgBCO *)obj)->arity > 0);
299
	break;
300
    }
301
302
303
304
305
306
307
308
309
310
311
312

    case AP:	/* Copied from stg_AP_entry. */
    {
	nat i, words;
	StgAP *ap;
	
	ap = (StgAP*)obj;
	words = ap->n_args;
	
	// Stack check
	if (Sp - (words+sizeofW(StgUpdateFrame)) < SpLim) {
	    Sp -= 2;
313
	    Sp[1] = (W_)tagged_obj;
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
	    Sp[0] = (W_)&stg_enter_info;
	    RETURN_TO_SCHEDULER(ThreadInterpret, StackOverflow);
	}
	
	/* Ok; we're safe.  Party on.  Push an update frame. */
	Sp -= sizeofW(StgUpdateFrame);
	{
	    StgUpdateFrame *__frame;
	    __frame = (StgUpdateFrame *)Sp;
	    SET_INFO(__frame, (StgInfoTable *)&stg_upd_frame_info);
	    __frame->updatee = (StgClosure *)(ap);
	}
	
	/* Reload the stack */
	Sp -= words;
	for (i=0; i < words; i++) {
	    Sp[i] = (W_)ap->payload[i];
	}

Simon Marlow's avatar
Simon Marlow committed
333
	obj = UNTAG_CLOSURE((StgClosure*)ap->fun);
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
	ASSERT(get_itbl(obj)->type == BCO);
	goto run_BCO_fun;
    }

    default:
#ifdef INTERP_STATS
    { 
	int j;
	
	j = get_itbl(obj)->type;
	ASSERT(j >= 0 && j < N_CLOSURE_TYPES);
	it_unknown_entries[j]++;
	it_total_unknown_entries++;
    }
#endif
    {
	// Can't handle this object; yield to scheduler
	IF_DEBUG(interpreter,
352
		 debugBelch("evaluating unknown closure -- yielding to sched\n"); 
353
354
355
		 printObj(obj);
	    );
	Sp -= 2;
356
	Sp[1] = (W_)tagged_obj;
357
	Sp[0] = (W_)&stg_enter_info;
358
	RETURN_TO_SCHEDULER_NO_PAUSE(ThreadRunGHC, ThreadYielding);
359
360
361
362
    }
    }

    // ------------------------------------------------------------------------
363
    // We now have an evaluated object (tagged_obj).  The next thing to
364
365
    // do is return it to the stack frame on top of the stack.
do_return:
366
    obj = UNTAG_CLOSURE(tagged_obj);
367
368
369
    ASSERT(closure_HNF(obj));

    IF_DEBUG(interpreter,
370
             debugBelch(
371
             "\n---------------------------------------------------------------\n");
372
373
374
             debugBelch("Returning: "); printObj(obj);
             debugBelch("Sp = %p\n", Sp);
             debugBelch("\n" );
375
             printStackChunk(Sp,cap->r.rCurrentTSO->stack+cap->r.rCurrentTSO->stack_size);
376
             debugBelch("\n\n");
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
    IF_DEBUG(sanity,checkStackChunk(Sp, cap->r.rCurrentTSO->stack+cap->r.rCurrentTSO->stack_size));

    switch (get_itbl((StgClosure *)Sp)->type) {

    case RET_SMALL: {
	const StgInfoTable *info;

	// NOTE: not using get_itbl().
	info = ((StgClosure *)Sp)->header.info;
	if (info == (StgInfoTable *)&stg_ap_v_info) {
	    n = 1; m = 0; goto do_apply;
	}
	if (info == (StgInfoTable *)&stg_ap_f_info) {
	    n = 1; m = 1; goto do_apply;
	}
	if (info == (StgInfoTable *)&stg_ap_d_info) {
	    n = 1; m = sizeofW(StgDouble); goto do_apply;
	}
	if (info == (StgInfoTable *)&stg_ap_l_info) {
	    n = 1; m = sizeofW(StgInt64); goto do_apply;
	}
	if (info == (StgInfoTable *)&stg_ap_n_info) {
	    n = 1; m = 1; goto do_apply;
	}
	if (info == (StgInfoTable *)&stg_ap_p_info) {
	    n = 1; m = 1; goto do_apply;
	}
	if (info == (StgInfoTable *)&stg_ap_pp_info) {
	    n = 2; m = 2; goto do_apply;
	}
	if (info == (StgInfoTable *)&stg_ap_ppp_info) {
	    n = 3; m = 3; goto do_apply;
	}
	if (info == (StgInfoTable *)&stg_ap_pppp_info) {
	    n = 4; m = 4; goto do_apply;
	}
	if (info == (StgInfoTable *)&stg_ap_ppppp_info) {
	    n = 5; m = 5; goto do_apply;
	}
	if (info == (StgInfoTable *)&stg_ap_pppppp_info) {
	    n = 6; m = 6; goto do_apply;
	}
	goto do_return_unrecognised;
    }

    case UPDATE_FRAME:
	// Returning to an update frame: do the update, pop the update
	// frame, and continue with the next stack frame.
427
428
429
430
431
432
433
434
        //
        // NB. we must update with the *tagged* pointer.  Some tags
        // are not optional, and if we omit the tag bits when updating
        // then bad things can happen (albeit very rarely).  See #1925.
        // What happened was an indirection was created with an
        // untagged pointer, and this untagged pointer was propagated
        // to a PAP by the GC, violating the invariant that PAPs
        // always contain a tagged pointer to the function.
435
	INTERP_TICK(it_retto_UPDATE);
436
	UPD_IND(((StgUpdateFrame *)Sp)->updatee, tagged_obj); 
437
438
439
440
441
442
443
444
445
	Sp += sizeofW(StgUpdateFrame);
	goto do_return;

    case RET_BCO:
	// Returning to an interpreted continuation: put the object on
	// the stack, and start executing the BCO.
	INTERP_TICK(it_retto_BCO);
	Sp--;
	Sp[0] = (W_)obj;
446
447
        // NB. return the untagged object; the bytecode expects it to
        // be untagged.  XXX this doesn't seem right.
448
449
450
451
452
453
454
455
456
457
	obj = (StgClosure*)Sp[2];
	ASSERT(get_itbl(obj)->type == BCO);
	goto run_BCO_return;

    default:
    do_return_unrecognised:
    {
	// Can't handle this return address; yield to scheduler
	INTERP_TICK(it_retto_other);
	IF_DEBUG(interpreter,
458
		 debugBelch("returning to unknown frame -- yielding to sched\n"); 
459
460
461
		 printStackChunk(Sp,cap->r.rCurrentTSO->stack+cap->r.rCurrentTSO->stack_size);
	    );
	Sp -= 2;
462
	Sp[1] = (W_)tagged_obj;
463
	Sp[0] = (W_)&stg_enter_info;
464
	RETURN_TO_SCHEDULER_NO_PAUSE(ThreadRunGHC, ThreadYielding);
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
    }
    }

    // -------------------------------------------------------------------------
    // Returning an unboxed value.  The stack looks like this:
    //
    // 	  |     ....      |
    // 	  +---------------+
    // 	  |     fv2       |
    // 	  +---------------+
    // 	  |     fv1       |
    // 	  +---------------+
    // 	  |     BCO       |
    // 	  +---------------+
    // 	  | stg_ctoi_ret_ |
    // 	  +---------------+
    // 	  |    retval     |
    // 	  +---------------+
    // 	  |   XXXX_info   |
    // 	  +---------------+
    //
    // where XXXX_info is one of the stg_gc_unbx_r1_info family.
    //
    // We're only interested in the case when the real return address
    // is a BCO; otherwise we'll return to the scheduler.

do_return_unboxed:
    { 
	int offset;
	
	ASSERT( Sp[0] == (W_)&stg_gc_unbx_r1_info
		|| Sp[0] == (W_)&stg_gc_unpt_r1_info
		|| Sp[0] == (W_)&stg_gc_f1_info
		|| Sp[0] == (W_)&stg_gc_d1_info
		|| Sp[0] == (W_)&stg_gc_l1_info
		|| Sp[0] == (W_)&stg_gc_void_info // VoidRep
	    );

	// get the offset of the stg_ctoi_ret_XXX itbl
	offset = stack_frame_sizeW((StgClosure *)Sp);

	switch (get_itbl((StgClosure *)Sp+offset)->type) {

	case RET_BCO:
	    // Returning to an interpreted continuation: put the object on
	    // the stack, and start executing the BCO.
	    INTERP_TICK(it_retto_BCO);
	    obj = (StgClosure*)Sp[offset+1];
	    ASSERT(get_itbl(obj)->type == BCO);
	    goto run_BCO_return_unboxed;

	default:
	{
	    // Can't handle this return address; yield to scheduler
	    INTERP_TICK(it_retto_other);
	    IF_DEBUG(interpreter,
521
		     debugBelch("returning to unknown frame -- yielding to sched\n"); 
522
523
		     printStackChunk(Sp,cap->r.rCurrentTSO->stack+cap->r.rCurrentTSO->stack_size);
		);
524
	    RETURN_TO_SCHEDULER_NO_PAUSE(ThreadRunGHC, ThreadYielding);
525
526
527
528
529
530
531
532
533
534
	}
	}
    }
    // not reached.


    // -------------------------------------------------------------------------
    // Application...

do_apply:
535
    ASSERT(obj == UNTAG_CLOSURE(tagged_obj));
536
537
538
539
540
541
542
543
    // we have a function to apply (obj), and n arguments taking up m
    // words on the stack.  The info table (stg_ap_pp_info or whatever)
    // is on top of the arguments on the stack.
    {
	switch (get_itbl(obj)->type) {

	case PAP: {
	    StgPAP *pap;
544
	    nat i, arity;
545
546
547
548

	    pap = (StgPAP *)obj;

	    // we only cope with PAPs whose function is a BCO
Simon Marlow's avatar
Simon Marlow committed
549
	    if (get_itbl(UNTAG_CLOSURE(pap->fun))->type != BCO) {
550
551
		goto defer_apply_to_sched;
	    }
552

553
554
555
556
557
558
559
560
561
562
            // Stack check: we're about to unpack the PAP onto the
            // stack.  The (+1) is for the (arity < n) case, where we
            // also need space for an extra info pointer.
            if (Sp - (pap->n_args + 1) < SpLim) {
                Sp -= 2;
                Sp[1] = (W_)tagged_obj;
                Sp[0] = (W_)&stg_enter_info;
                RETURN_TO_SCHEDULER(ThreadInterpret, StackOverflow);
            }

563
564
565
566
567
568
569
570
571
572
573
	    Sp++;
	    arity = pap->arity;
	    ASSERT(arity > 0);
	    if (arity < n) {
		// n must be greater than 1, and the only kinds of
		// application we support with more than one argument
		// are all pointers...
		//
		// Shuffle the args for this function down, and put
		// the appropriate info table in the gap.
		for (i = 0; i < arity; i++) {
574
575
		    Sp[(int)i-1] = Sp[i];
		    // ^^^^^ careful, i-1 might be negative, but i in unsigned
576
577
578
579
580
581
582
583
		}
		Sp[arity-1] = app_ptrs_itbl[n-arity-1];
		Sp--;
		// unpack the PAP's arguments onto the stack
		Sp -= pap->n_args;
		for (i = 0; i < pap->n_args; i++) {
		    Sp[i] = (W_)pap->payload[i];
		}
Simon Marlow's avatar
Simon Marlow committed
584
		obj = UNTAG_CLOSURE(pap->fun);
585
586
587
588
589
590
591
		goto run_BCO_fun;
	    } 
	    else if (arity == n) {
		Sp -= pap->n_args;
		for (i = 0; i < pap->n_args; i++) {
		    Sp[i] = (W_)pap->payload[i];
		}
Simon Marlow's avatar
Simon Marlow committed
592
		obj = UNTAG_CLOSURE(pap->fun);
593
594
595
596
597
		goto run_BCO_fun;
	    } 
	    else /* arity > n */ {
		// build a new PAP and return it.
		StgPAP *new_pap;
598
		new_pap = (StgPAP *)allocate(PAP_sizeW(pap->n_args + m));
599
600
601
602
603
604
605
606
607
608
		SET_HDR(new_pap,&stg_PAP_info,CCCS);
		new_pap->arity = pap->arity - n;
		new_pap->n_args = pap->n_args + m;
		new_pap->fun = pap->fun;
		for (i = 0; i < pap->n_args; i++) {
		    new_pap->payload[i] = pap->payload[i];
		}
		for (i = 0; i < m; i++) {
		    new_pap->payload[pap->n_args + i] = (StgClosure *)Sp[i];
		}
609
		tagged_obj = (StgClosure *)new_pap;
610
611
612
613
614
615
		Sp += m;
		goto do_return;
	    }
	}	    

	case BCO: {
616
	    nat arity, i;
617
618

	    Sp++;
619
	    arity = ((StgBCO *)obj)->arity;
620
621
622
623
624
625
626
627
628
	    ASSERT(arity > 0);
	    if (arity < n) {
		// n must be greater than 1, and the only kinds of
		// application we support with more than one argument
		// are all pointers...
		//
		// Shuffle the args for this function down, and put
		// the appropriate info table in the gap.
		for (i = 0; i < arity; i++) {
629
630
		    Sp[(int)i-1] = Sp[i];
		    // ^^^^^ careful, i-1 might be negative, but i in unsigned
631
632
633
634
635
636
637
638
639
640
641
		}
		Sp[arity-1] = app_ptrs_itbl[n-arity-1];
		Sp--;
		goto run_BCO_fun;
	    } 
	    else if (arity == n) {
		goto run_BCO_fun;
	    }
	    else /* arity > n */ {
		// build a PAP and return it.
		StgPAP *pap;
642
643
		nat i;
		pap = (StgPAP *)allocate(PAP_sizeW(m));
644
645
646
647
648
649
650
		SET_HDR(pap, &stg_PAP_info,CCCS);
		pap->arity = arity - n;
		pap->fun = obj;
		pap->n_args = m;
		for (i = 0; i < m; i++) {
		    pap->payload[i] = (StgClosure *)Sp[i];
		}
651
		tagged_obj = (StgClosure *)pap;
652
653
654
655
656
657
658
659
660
		Sp += m;
		goto do_return;
	    }
	}

	// No point in us applying machine-code functions
	default:
	defer_apply_to_sched:
	    Sp -= 2;
661
	    Sp[1] = (W_)tagged_obj;
662
	    Sp[0] = (W_)&stg_enter_info;
663
	    RETURN_TO_SCHEDULER_NO_PAUSE(ThreadRunGHC, ThreadYielding);
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
    }

    // ------------------------------------------------------------------------
    // Ok, we now have a bco (obj), and its arguments are all on the
    // stack.  We can start executing the byte codes.
    //
    // The stack is in one of two states.  First, if this BCO is a
    // function:
    //
    // 	  |     ....      |
    // 	  +---------------+
    // 	  |     arg2      |
    // 	  +---------------+
    // 	  |     arg1      |
    // 	  +---------------+
    //
    // Second, if this BCO is a continuation:
    //
    // 	  |     ....      |
    // 	  +---------------+
    // 	  |     fv2       |
    // 	  +---------------+
    // 	  |     fv1       |
    // 	  +---------------+
    // 	  |     BCO       |
    // 	  +---------------+
    // 	  | stg_ctoi_ret_ |
    // 	  +---------------+
    // 	  |    retval     |
    // 	  +---------------+
    // 
    // where retval is the value being returned to this continuation.
    // In the event of a stack check, heap check, or context switch,
    // we need to leave the stack in a sane state so the garbage
    // collector can find all the pointers.
    //
    //  (1) BCO is a function:  the BCO's bitmap describes the
    //      pointerhood of the arguments.
    //
    //  (2) BCO is a continuation: BCO's bitmap describes the
    //      pointerhood of the free variables.
    //
    // Sadly we have three different kinds of stack/heap/cswitch check
    // to do:

709

710
711
712
713
714
715
run_BCO_return:
    // Heap check
    if (doYouWantToGC()) {
	Sp--; Sp[0] = (W_)&stg_enter_info;
	RETURN_TO_SCHEDULER(ThreadInterpret, HeapOverflow);
    }
716
717
    // Stack checks aren't necessary at return points, the stack use
    // is aggregated into the enclosing function entry point.
718

719
720
721
722
723
724
725
    goto run_BCO;
    
run_BCO_return_unboxed:
    // Heap check
    if (doYouWantToGC()) {
	RETURN_TO_SCHEDULER(ThreadInterpret, HeapOverflow);
    }
726
727
    // Stack checks aren't necessary at return points, the stack use
    // is aggregated into the enclosing function entry point.
728

729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
    goto run_BCO;
    
run_BCO_fun:
    IF_DEBUG(sanity,
	     Sp -= 2; 
	     Sp[1] = (W_)obj; 
	     Sp[0] = (W_)&stg_apply_interp_info;
	     checkStackChunk(Sp,SpLim);
	     Sp += 2;
	);

    // Heap check
    if (doYouWantToGC()) {
	Sp -= 2; 
	Sp[1] = (W_)obj; 
	Sp[0] = (W_)&stg_apply_interp_info; // placeholder, really
	RETURN_TO_SCHEDULER(ThreadInterpret, HeapOverflow);
    }
    
748
749
    // Stack check
    if (Sp - INTERP_STACK_CHECK_THRESH < SpLim) {
750
751
752
753
754
	Sp -= 2; 
	Sp[1] = (W_)obj; 
	Sp[0] = (W_)&stg_apply_interp_info; // placeholder, really
	RETURN_TO_SCHEDULER(ThreadInterpret, StackOverflow);
    }
755

756
757
758
759
760
761
762
763
    goto run_BCO;
    
    // Now, actually interpret the BCO... (no returning to the
    // scheduler again until the stack is in an orderly state).
run_BCO:
    INTERP_TICK(it_BCO_entries);
    {
	register int       bciPtr     = 1; /* instruction pointer */
764
        register StgWord16 bci;
765
	register StgBCO*   bco        = (StgBCO*)obj;
766
	register StgWord16* instrs    = (StgWord16*)(bco->instrs->payload);
767
768
	register StgWord*  literals   = (StgWord*)(&bco->literals->payload[0]);
	register StgPtr*   ptrs       = (StgPtr*)(&bco->ptrs->payload[0]);
769

770
771
772
#ifdef INTERP_STATS
	it_lastopc = 0; /* no opcode */
#endif
773

774
775
776
777
    nextInsn:
	ASSERT(bciPtr <= instrs[0]);
	IF_DEBUG(interpreter,
		 //if (do_print_stack) {
778
		 //debugBelch("\n-- BEGIN stack\n");
779
		 //printStack(Sp,cap->r.rCurrentTSO->stack+cap->r.rCurrentTSO->stack_size,iSu);
780
		 //debugBelch("-- END stack\n\n");
781
		 //}
782
		 debugBelch("Sp = %p   pc = %d      ", Sp, bciPtr);
783
784
		 disInstr(bco,bciPtr);
		 if (0) { int i;
785
		 debugBelch("\n");
786
		 for (i = 8; i >= 0; i--) {
787
		     debugBelch("%d  %p\n", i, (StgPtr)(*(Sp+i)));
788
		 }
789
		 debugBelch("\n");
790
		 }
791
792
		 //if (do_print_stack) checkStack(Sp,cap->r.rCurrentTSO->stack+cap->r.rCurrentTSO->stack_size,iSu);
	    );
793

794

795
796
797
798
799
800
801
802
803
	INTERP_TICK(it_insns);

#ifdef INTERP_STATS
	ASSERT( (int)instrs[bciPtr] >= 0 && (int)instrs[bciPtr] < 27 );
	it_ofreq[ (int)instrs[bciPtr] ] ++;
	it_oofreq[ it_lastopc ][ (int)instrs[bciPtr] ] ++;
	it_lastopc = (int)instrs[bciPtr];
#endif

804
805
806
807
808
809
	bci = BCO_NEXT;
    /* We use the high 8 bits for flags, only the highest of which is
     * currently allocated */
    ASSERT((bci & 0xFF00) == (bci & 0x8000));

    switch (bci & 0xFF) {
810

811
812
813
814
815
        /* check for a breakpoint on the beginning of a let binding */
        case bci_BRK_FUN: 
        {
            int arg1_brk_array, arg2_array_index, arg3_freeVars;
            StgArrWords *breakPoints;
816
817
            int returning_from_break;     // are we resuming execution from a breakpoint?
                                          //  if yes, then don't break this time around
818
819
820
821
822
823
            StgClosure *ioAction;         // the io action to run at a breakpoint

            StgAP_STACK *new_aps;         // a closure to save the top stack frame on the heap
            int i;
            int size_words;

824
825
826
            arg1_brk_array      = BCO_NEXT;  // 1st arg of break instruction
            arg2_array_index    = BCO_NEXT;  // 2nd arg of break instruction
            arg3_freeVars       = BCO_NEXT;  // 3rd arg of break instruction
827

828
829
            // check if we are returning from a breakpoint - this info
            // is stored in the flags field of the current TSO
830
831
            returning_from_break = cap->r.rCurrentTSO->flags & TSO_STOPPED_ON_BREAKPOINT; 

832
833
            // if we are returning from a break then skip this section
            // and continue executing
834
835
836
837
            if (!returning_from_break)
            {
               breakPoints = (StgArrWords *) BCO_PTR(arg1_brk_array);

838
               // stop the current thread if either the
839
               // "rts_stop_next_breakpoint" flag is true OR if the
840
841
               // breakpoint flag for this particular expression is
               // true
842
               if (rts_stop_next_breakpoint == rtsTrue || 
843
                   breakPoints->payload[arg2_array_index] == rtsTrue)
844
               {
845
846
                  // make sure we don't automatically stop at the
                  // next breakpoint
847
                  rts_stop_next_breakpoint = rtsFalse;
848
849
850
851
852

                  // allocate memory for a new AP_STACK, enough to
                  // store the top stack frame plus an
                  // stg_apply_interp_info pointer and a pointer to
                  // the BCO
853
854
855
856
857
858
859
                  size_words = BCO_BITMAP_SIZE(obj) + 2;
                  new_aps = (StgAP_STACK *) allocate (AP_STACK_sizeW(size_words));
                  SET_HDR(new_aps,&stg_AP_STACK_info,CCS_SYSTEM); 
                  new_aps->size = size_words;
                  new_aps->fun = &stg_dummy_ret_closure; 

                  // fill in the payload of the AP_STACK 
860
861
                  new_aps->payload[0] = (StgClosure *)&stg_apply_interp_info;
                  new_aps->payload[1] = (StgClosure *)obj;
862
863
864
865

                  // copy the contents of the top stack frame into the AP_STACK
                  for (i = 2; i < size_words; i++)
                  {
866
                     new_aps->payload[i] = (StgClosure *)Sp[i-2];
867
868
                  }

869
                  // prepare the stack so that we can call the
870
                  // rts_breakpoint_io_action and ensure that the stack is
871
872
                  // in a reasonable state for the GC and so that
                  // execution of this BCO can continue when we resume
873
                  ioAction = (StgClosure *) deRefStablePtr (rts_breakpoint_io_action);
874
875
876
877
                  Sp -= 9;
                  Sp[8] = (W_)obj;   
                  Sp[7] = (W_)&stg_apply_interp_info;
                  Sp[6] = (W_)&stg_noforceIO_info;     // see [unreg] below
878
879
880
881
                  Sp[5] = (W_)new_aps;                 // the AP_STACK
                  Sp[4] = (W_)BCO_PTR(arg3_freeVars);  // the info about local vars of the breakpoint
                  Sp[3] = (W_)False_closure;            // True <=> a breakpoint
                  Sp[2] = (W_)&stg_ap_pppv_info;
882
883
                  Sp[1] = (W_)ioAction;                // apply the IO action to its two arguments above
                  Sp[0] = (W_)&stg_enter_info;         // get ready to run the IO action
884
885
886
887
                  // Note [unreg]: in unregisterised mode, the return
                  // convention for IO is different.  The
                  // stg_noForceIO_info stack frame is necessary to
                  // account for this difference.
888

889
890
891
892
                  // set the flag in the TSO to say that we are now
                  // stopping at a breakpoint so that when we resume
                  // we don't stop on the same breakpoint that we
                  // already stopped at just now
893
894
                  cap->r.rCurrentTSO->flags |= TSO_STOPPED_ON_BREAKPOINT;

895
896
897
                  // stop this thread and return to the scheduler -
                  // eventually we will come back and the IO action on
                  // the top of the stack will be executed
898
899
900
901
902
903
904
905
906
907
                  RETURN_TO_SCHEDULER_NO_PAUSE(ThreadRunGHC, ThreadYielding);
               }
            }
            // record that this thread is not stopped at a breakpoint anymore
            cap->r.rCurrentTSO->flags &= ~TSO_STOPPED_ON_BREAKPOINT;

            // continue normal execution of the byte code instructions
	    goto nextInsn;
        }

908
909
910
911
	case bci_STKCHECK: {
	    // Explicit stack check at the beginning of a function
	    // *only* (stack checks in case alternatives are
	    // propagated to the enclosing function).
912
	    StgWord stk_words_reqd = BCO_GET_LARGE_ARG + 1;
913
	    if (Sp - stk_words_reqd < SpLim) {
914
915
916
		Sp -= 2; 
		Sp[1] = (W_)obj; 
		Sp[0] = (W_)&stg_apply_interp_info;
917
		RETURN_TO_SCHEDULER(ThreadInterpret, StackOverflow);
918
919
	    } else {
		goto nextInsn;
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
	    }
	}

	case bci_PUSH_L: {
	    int o1 = BCO_NEXT;
	    Sp[-1] = Sp[o1];
	    Sp--;
	    goto nextInsn;
	}

	case bci_PUSH_LL: {
	    int o1 = BCO_NEXT;
	    int o2 = BCO_NEXT;
	    Sp[-1] = Sp[o1];
	    Sp[-2] = Sp[o2];
	    Sp -= 2;
	    goto nextInsn;
	}

	case bci_PUSH_LLL: {
	    int o1 = BCO_NEXT;
	    int o2 = BCO_NEXT;
	    int o3 = BCO_NEXT;
	    Sp[-1] = Sp[o1];
	    Sp[-2] = Sp[o2];
	    Sp[-3] = Sp[o3];
	    Sp -= 3;
	    goto nextInsn;
	}

	case bci_PUSH_G: {
	    int o1 = BCO_NEXT;
	    Sp[-1] = BCO_PTR(o1);
	    Sp -= 1;
	    goto nextInsn;
	}

	case bci_PUSH_ALTS: {
	    int o_bco  = BCO_NEXT;
959
	    Sp[-2] = (W_)&stg_ctoi_R1p_info;
960
961
962
963
964
965
966
	    Sp[-1] = BCO_PTR(o_bco);
	    Sp -= 2;
	    goto nextInsn;
	}

	case bci_PUSH_ALTS_P: {
	    int o_bco  = BCO_NEXT;
967
	    Sp[-2] = (W_)&stg_ctoi_R1unpt_info;
968
969
970
971
972
973
974
	    Sp[-1] = BCO_PTR(o_bco);
	    Sp -= 2;
	    goto nextInsn;
	}

	case bci_PUSH_ALTS_N: {
	    int o_bco  = BCO_NEXT;
975
	    Sp[-2] = (W_)&stg_ctoi_R1n_info;
976
977
978
979
980
981
982
	    Sp[-1] = BCO_PTR(o_bco);
	    Sp -= 2;
	    goto nextInsn;
	}

	case bci_PUSH_ALTS_F: {
	    int o_bco  = BCO_NEXT;
983
	    Sp[-2] = (W_)&stg_ctoi_F1_info;
984
985
986
987
988
989
990
	    Sp[-1] = BCO_PTR(o_bco);
	    Sp -= 2;
	    goto nextInsn;
	}

	case bci_PUSH_ALTS_D: {
	    int o_bco  = BCO_NEXT;
991
	    Sp[-2] = (W_)&stg_ctoi_D1_info;
992
993
994
995
996
997
998
	    Sp[-1] = BCO_PTR(o_bco);
	    Sp -= 2;
	    goto nextInsn;
	}

	case bci_PUSH_ALTS_L: {
	    int o_bco  = BCO_NEXT;
999
	    Sp[-2] = (W_)&stg_ctoi_L1_info;
1000
1001
1002
1003
1004
1005
1006
	    Sp[-1] = BCO_PTR(o_bco);
	    Sp -= 2;
	    goto nextInsn;
	}

	case bci_PUSH_ALTS_V: {
	    int o_bco  = BCO_NEXT;
1007
	    Sp[-2] = (W_)&stg_ctoi_V_info;
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
	    Sp[-1] = BCO_PTR(o_bco);
	    Sp -= 2;
	    goto nextInsn;
	}

	case bci_PUSH_APPLY_N:
	    Sp--; Sp[0] = (W_)&stg_ap_n_info;
	    goto nextInsn;
	case bci_PUSH_APPLY_V:
	    Sp--; Sp[0] = (W_)&stg_ap_v_info;
	    goto nextInsn;
	case bci_PUSH_APPLY_F:
	    Sp--; Sp[0] = (W_)&stg_ap_f_info;
	    goto nextInsn;
	case bci_PUSH_APPLY_D:
	    Sp--; Sp[0] = (W_)&stg_ap_d_info;
	    goto nextInsn;
	case bci_PUSH_APPLY_L:
	    Sp--; Sp[0] = (W_)&stg_ap_l_info;
	    goto nextInsn;
	case bci_PUSH_APPLY_P:
	    Sp--; Sp[0] = (W_)&stg_ap_p_info;
	    goto nextInsn;
	case bci_PUSH_APPLY_PP:
	    Sp--; Sp[0] = (W_)&stg_ap_pp_info;
	    goto nextInsn;
	case bci_PUSH_APPLY_PPP:
	    Sp--; Sp[0] = (W_)&stg_ap_ppp_info;
	    goto nextInsn;
	case bci_PUSH_APPLY_PPPP:
	    Sp--; Sp[0] = (W_)&stg_ap_pppp_info;
	    goto nextInsn;
	case bci_PUSH_APPLY_PPPPP:
	    Sp--; Sp[0] = (W_)&stg_ap_ppppp_info;
	    goto nextInsn;
	case bci_PUSH_APPLY_PPPPPP:
	    Sp--; Sp[0] = (W_)&stg_ap_pppppp_info;
	    goto nextInsn;
	    
	case bci_PUSH_UBX: {
	    int i;
	    int o_lits = BCO_NEXT;
	    int n_words = BCO_NEXT;
	    Sp -= n_words;
	    for (i = 0; i < n_words; i++) {
1053
		Sp[i] = (W_)BCO_LIT(o_lits+i);
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
	    }
	    goto nextInsn;
	}

	case bci_SLIDE: {
	    int n  = BCO_NEXT;
	    int by = BCO_NEXT;
	    /* a_1, .. a_n, b_1, .. b_by, s => a_1, .. a_n, s */
	    while(--n >= 0) {
		Sp[n+by] = Sp[n];
	    }
	    Sp += by;
	    INTERP_TICK(it_slides);
	    goto nextInsn;
	}

	case bci_ALLOC_AP: {
	    StgAP* ap; 
sof's avatar
sof committed
1072
	    int n_payload = BCO_NEXT;
1073
	    ap = (StgAP*)allocate(AP_sizeW(n_payload));
1074
1075
1076
1077
1078
1079
1080
	    Sp[-1] = (W_)ap;
	    ap->n_args = n_payload;
	    SET_HDR(ap, &stg_AP_info, CCS_SYSTEM/*ToDo*/)
	    Sp --;
	    goto nextInsn;
	}

1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
	case bci_ALLOC_AP_NOUPD: {
	    StgAP* ap; 
	    int n_payload = BCO_NEXT;
	    ap = (StgAP*)allocate(AP_sizeW(n_payload));
	    Sp[-1] = (W_)ap;
	    ap->n_args = n_payload;
	    SET_HDR(ap, &stg_AP_NOUPD_info, CCS_SYSTEM/*ToDo*/)
	    Sp --;
	    goto nextInsn;
	}

1092
1093
1094
	case bci_ALLOC_PAP: {
	    StgPAP* pap; 
	    int arity = BCO_NEXT;
sof's avatar
sof committed
1095
	    int n_payload = BCO_NEXT;
1096
	    pap = (StgPAP*)allocate(PAP_sizeW(n_payload));
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
	    Sp[-1] = (W_)pap;
	    pap->n_args = n_payload;
	    pap->arity = arity;
	    SET_HDR(pap, &stg_PAP_info, CCS_SYSTEM/*ToDo*/)
	    Sp --;
	    goto nextInsn;
	}

	case bci_MKAP: {
	    int i;
	    int stkoff = BCO_NEXT;
sof's avatar
sof committed
1108
	    int n_payload = BCO_NEXT;
1109
1110
1111
	    StgAP* ap = (StgAP*)Sp[stkoff];
	    ASSERT((int)ap->n_args == n_payload);
	    ap->fun = (StgClosure*)Sp[0];
1112
	    
1113
1114
1115
	    // The function should be a BCO, and its bitmap should
	    // cover the payload of the AP correctly.
	    ASSERT(get_itbl(ap->fun)->type == BCO
1116
1117
		   && BCO_BITMAP_SIZE(ap->fun) == ap->n_args);
	    
1118
1119
1120
1121
	    for (i = 0; i < n_payload; i++)
		ap->payload[i] = (StgClosure*)Sp[i+1];
	    Sp += n_payload+1;
	    IF_DEBUG(interpreter,
1122
		     debugBelch("\tBuilt "); 
1123
1124
1125
1126
1127
		     printObj((StgClosure*)ap);
		);
	    goto nextInsn;
	}

1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
	case bci_MKPAP: {
	    int i;
	    int stkoff = BCO_NEXT;
	    int n_payload = BCO_NEXT;
	    StgPAP* pap = (StgPAP*)Sp[stkoff];
	    ASSERT((int)pap->n_args == n_payload);
	    pap->fun = (StgClosure*)Sp[0];
	    
	    // The function should be a BCO
	    ASSERT(get_itbl(pap->fun)->type == BCO);
	    
	    for (i = 0; i < n_payload; i++)
		pap->payload[i] = (StgClosure*)Sp[i+1];
	    Sp += n_payload+1;
	    IF_DEBUG(interpreter,
		     debugBelch("\tBuilt "); 
		     printObj((StgClosure*)pap);
		);
	    goto nextInsn;
	}

1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
	case bci_UNPACK: {
	    /* Unpack N ptr words from t.o.s constructor */
	    int i;
	    int n_words = BCO_NEXT;
	    StgClosure* con = (StgClosure*)Sp[0];
	    Sp -= n_words;
	    for (i = 0; i < n_words; i++) {
		Sp[i] = (W_)con->payload[i];
	    }
	    goto nextInsn;
	}

	case bci_PACK: {
	    int i;
	    int o_itbl         = BCO_NEXT;
	    int n_words        = BCO_NEXT;
1165
	    StgInfoTable* itbl = INFO_PTR_TO_STRUCT(BCO_LIT(o_itbl));
1166
1167
1168
1169
	    int request        = CONSTR_sizeW( itbl->layout.payload.ptrs, 
					       itbl->layout.payload.nptrs );
	    StgClosure* con = (StgClosure*)allocate_NONUPD(request);
	    ASSERT( itbl->layout.payload.ptrs + itbl->layout.payload.nptrs > 0);
1170
	    SET_HDR(con, (StgInfoTable*)BCO_LIT(o_itbl), CCS_SYSTEM/*ToDo*/);
1171
1172
1173
1174
1175
1176
1177
	    for (i = 0; i < n_words; i++) {
		con->payload[i] = (StgClosure*)Sp[i];
	    }
	    Sp += n_words;
	    Sp --;
	    Sp[0] = (W_)con;
	    IF_DEBUG(interpreter,
1178
		     debugBelch("\tBuilt "); 
1179
1180
1181
1182
1183
1184
		     printObj((StgClosure*)con);
		);
	    goto nextInsn;
	}

	case bci_TESTLT_P: {
1185
	    unsigned int discr  = BCO_NEXT;
1186
1187
	    int failto = BCO_NEXT;
	    StgClosure* con = (StgClosure*)Sp[0];
1188
	    if (GET_TAG(con) >= discr) {
1189
1190
1191
1192
1193
1194
		bciPtr = failto;
	    }
	    goto nextInsn;
	}

	case bci_TESTEQ_P: {
1195
	    unsigned int discr  = BCO_NEXT;
1196
1197
	    int failto = BCO_NEXT;
	    StgClosure* con = (StgClosure*)Sp[0];
1198
	    if (GET_TAG(con) != discr) {
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
		bciPtr = failto;
	    }
	    goto nextInsn;
	}

	case bci_TESTLT_I: {
	    // There should be an Int at Sp[1], and an info table at Sp[0].
	    int discr   = BCO_NEXT;
	    int failto  = BCO_NEXT;
	    I_ stackInt = (I_)Sp[1];
	    if (stackInt >= (I_)BCO_LIT(discr))
		bciPtr = failto;
	    goto nextInsn;
	}

	case bci_TESTEQ_I: {
	    // There should be an Int at Sp[1], and an info table at Sp[0].
	    int discr   = BCO_NEXT;
	    int failto  = BCO_NEXT;
	    I_ stackInt = (I_)Sp[1];
	    if (stackInt != (I_)BCO_LIT(discr)) {
		bciPtr = failto;
	    }
	    goto nextInsn;
	}

	case bci_TESTLT_D: {
	    // There should be a Double at Sp[1], and an info table at Sp[0].
	    int discr   = BCO_NEXT;
	    int failto  = BCO_NEXT;
	    StgDouble stackDbl, discrDbl;
	    stackDbl = PK_DBL( & Sp[1] );
	    discrDbl = PK_DBL( & BCO_LIT(discr) );
	    if (stackDbl >= discrDbl) {
		bciPtr = failto;
	    }
	    goto nextInsn;
	}

	case bci_TESTEQ_D: {
	    // There should be a Double at Sp[1], and an info table at Sp[0].
	    int discr   = BCO_NEXT;
	    int failto  = BCO_NEXT;
	    StgDouble stackDbl, discrDbl;
	    stackDbl = PK_DBL( & Sp[1] );
	    discrDbl = PK_DBL( & BCO_LIT(discr) );
	    if (stackDbl != discrDbl) {
		bciPtr = failto;
	    }
	    goto nextInsn;
	}

	case bci_TESTLT_F: {
	    // There should be a Float at Sp[1], and an info table at Sp[0].
	    int discr   = BCO_NEXT;
	    int failto  = BCO_NEXT;
	    StgFloat stackFlt, discrFlt;
	    stackFlt = PK_FLT( & Sp[1] );
	    discrFlt = PK_FLT( & BCO_LIT(discr) );
	    if (stackFlt >= discrFlt) {
		bciPtr = failto;
	    }
	    goto nextInsn;
	}

	case bci_TESTEQ_F: {
	    // There should be a Float at Sp[1], and an info table at Sp[0].
	    int discr   = BCO_NEXT;
	    int failto  = BCO_NEXT;
	    StgFloat stackFlt, discrFlt;
	    stackFlt = PK_FLT( & Sp[1] );
	    discrFlt = PK_FLT( & BCO_LIT(discr) );
	    if (stackFlt != discrFlt) {
		bciPtr = failto;
	    }
	    goto nextInsn;
	}

	// Control-flow ish things
	case bci_ENTER:
	    // Context-switch check.  We put it here to ensure that
	    // the interpreter has done at least *some* work before
	    // context switching: sometimes the scheduler can invoke
	    // the interpreter with context_switch == 1, particularly
	    // if the -C0 flag has been given on the cmd line.
1284
	    if (cap->context_switch) {
1285
1286
1287
1288
1289
1290
		Sp--; Sp[0] = (W_)&stg_enter_info;
		RETURN_TO_SCHEDULER(ThreadInterpret, ThreadYielding);
	    }
	    goto eval;

	case bci_RETURN:
1291
	    tagged_obj = (StgClosure *)Sp[0];
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
	    Sp++;
	    goto do_return;

	case bci_RETURN_P:
	    Sp--;
	    Sp[0] = (W_)&stg_gc_unpt_r1_info;
	    goto do_return_unboxed;
	case bci_RETURN_N:
	    Sp--;
	    Sp[0] = (W_)&stg_gc_unbx_r1_info;
	    goto do_return_unboxed;
	case bci_RETURN_F:
	    Sp--;
	    Sp[0] = (W_)&stg_gc_f1_info;
	    goto do_return_unboxed;
	case bci_RETURN_D:
	    Sp--;
	    Sp[0] = (W_)&stg_gc_d1_info;
	    goto do_return_unboxed;
	case bci_RETURN_L:
	    Sp--;
	    Sp[0] = (W_)&stg_gc_l1_info;
	    goto do_return_unboxed;
	case bci_RETURN_V:
	    Sp--;
	    Sp[0] = (W_)&stg_gc_void_info;
	    goto do_return_unboxed;

	case bci_SWIZZLE: {
	    int stkoff = BCO_NEXT;
	    signed short n = (signed short)(BCO_NEXT);
	    Sp[stkoff] += (W_)n;
	    goto nextInsn;
	}

	case bci_CCALL: {
1328
	    void *tok;
1329
1330
1331
	    int stk_offset            = BCO_NEXT;
	    int o_itbl                = BCO_NEXT;
	    void(*marshall_fn)(void*) = (void (*)(void*))BCO_LIT(o_itbl);
1332
1333
1334
	    int ret_dyn_size = 
		RET_DYN_BITMAP_SIZE + RET_DYN_NONPTR_REGS_SIZE
		+ sizeofW(StgRetDyn);
1335

1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
            /* the stack looks like this:
               
               |             |  <- Sp + stk_offset
               +-------------+  
               |             |
               |    args     |
               |             |  <- Sp + ret_size + 1
               +-------------+
               |    C fun    |  <- Sp + ret_size
               +-------------+
               |     ret     |  <- Sp
               +-------------+

               ret is a placeholder for the return address, and may be
               up to 2 words.

               We need to copy the args out of the TSO, because when
               we call suspendThread() we no longer own the TSO stack,
               and it may move at any time - indeed suspendThread()
               itself may do stack squeezing and move our args.
               So we make a copy of the argument block.
            */

#define ROUND_UP_WDS(p)  ((((StgWord)(p)) + sizeof(W_)-1)/sizeof(W_))

            ffi_cif *cif = (ffi_cif *)marshall_fn;
            nat nargs = cif->nargs;
            nat ret_size;
            nat i;
            StgPtr p;
            W_ ret[2];                  // max needed
	    W_ *arguments[stk_offset];  // max needed
            void *argptrs[nargs];
            void (*fn)(void);

            if (cif->rtype->type == FFI_TYPE_VOID) {
                // necessary because cif->rtype->size == 1 for void,
                // but the bytecode generator has not pushed a
                // placeholder in this case.
                ret_size = 0;
            } else {
                ret_size = ROUND_UP_WDS(cif->rtype->size);
            }

	    memcpy(arguments, Sp+ret_size+1, 
                   sizeof(W_) * (stk_offset-1-ret_size));
            
            // libffi expects the args as an array of pointers to
            // values, so we have to construct this array before making
            // the call.
            p = (StgPtr)arguments;
            for (i = 0; i < nargs; i++) {
                argptrs[i] = (void *)p;
                // get the size from the cif
                p += ROUND_UP_WDS(cif->arg_types[i]->size);
            }
1392

1393
1394
            // this is the function we're going to call
            fn = (void(*)(void))Sp[ret_size];
1395

1396
1397
1398
	    // Restore the Haskell thread's current value of errno
	    errno = cap->r.rCurrentTSO->saved_errno;

1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
	    // There are a bunch of non-ptr words on the stack (the
	    // ccall args, the ccall fun address and space for the
	    // result), which we need to cover with an info table
	    // since we might GC during this call.
	    //
	    // We know how many (non-ptr) words there are before the
	    // next valid stack frame: it is the stk_offset arg to the
	    // CCALL instruction.   So we build a RET_DYN stack frame
	    // on the stack frame to describe this chunk of stack.
	    //
1409
	    Sp -= ret_dyn_size;
1410
	    ((StgRetDyn *)Sp)->liveness = R1_PTR | N_NONPTRS(stk_offset);
1411
1412
	    ((StgRetDyn *)Sp)->info = (StgInfoTable *)&stg_gc_gen_info;

1413
1414
1415
1416
1417
            // save obj (pointer to the current BCO), since this
            // might move during the call.  We use the R1 slot in the
            // RET_DYN frame for this, hence R1_PTR above.
            ((StgRetDyn *)Sp)->payload[0] = (StgClosure *)obj;

1418
	    SAVE_STACK_POINTERS;
1419
	    tok = suspendThread(&cap->r);
1420

1421
	    // We already made a copy of the arguments above.
1422
            ffi_call(cif, fn, ret, argptrs);
1423

1424
	    // And restart the thread again, popping the RET_DYN frame.
1425
	    cap = (Capability *)((void *)((unsigned char*)resumeThread(tok) - sizeof(StgFunTable)));
1426
	    LOAD_STACK_POINTERS;
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436

            // Re-load the pointer to the BCO from the RET_DYN frame,
            // it might have moved during the call.  Also reload the
            // pointers to the components of the BCO.
            obj        = ((StgRetDyn *)Sp)->payload[0];
            bco        = (StgBCO*)obj;
            instrs     = (StgWord16*)(bco->instrs->payload);
            literals   = (StgWord*)(&bco->literals->payload[0]);
            ptrs       = (StgPtr*)(&bco->ptrs->payload[0]);

1437
	    Sp += ret_dyn_size;
1438
	    
1439
1440
1441
	    // Save the Haskell thread's current value of errno
	    cap->r.rCurrentTSO->saved_errno = errno;
		
1442
1443
1444
	    // Copy the return value back to the TSO stack.  It is at
            // most 2 words large, and resides at arguments[0].
            memcpy(Sp, ret, sizeof(W_) * stg_min(stk_offset,ret_size));
1445

1446
1447
1448
1449
1450
1451
1452
1453
1454
	    goto nextInsn;
	}

	case bci_JMP: {
	    /* BCO_NEXT modifies bciPtr, so be conservative. */
	    int nextpc = BCO_NEXT;
	    bciPtr     = nextpc;
	    goto nextInsn;
	}
<