Interpreter.c 43.2 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"
Simon Marlow's avatar
Simon Marlow committed
10
11
12
13
#include "rts/Bytecodes.h"

// internal headers
#include "sm/Storage.h"
Simon Marlow's avatar
Simon Marlow committed
14
#include "sm/Sanity.h"
15
#include "RtsUtils.h"
16
17
#include "Schedule.h"
#include "Updates.h"
18
#include "Prelude.h"
Simon Marlow's avatar
Simon Marlow committed
19
#include "Stable.h"
20
21
#include "Printer.h"
#include "Disassembler.h"
22
#include "Interpreter.h"
Simon Marlow's avatar
Simon Marlow committed
23
#include "ThreadPaused.h"
24
#include "Threads.h"
25

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

31
32
33
34
35
36
37
// When building the RTS in the non-dyn way on Windows, we don't
//	want declspec(__dllimport__) on the front of function prototypes
//	from libffi.
#if defined(mingw32_HOST_OS) && !defined(__PIC__)
# define LIBFFI_NOT_DLL
#endif

38
#include "ffi.h"
andy's avatar
andy committed
39

40
/* --------------------------------------------------------------------------
41
 * The bytecode interpreter
42
43
 * ------------------------------------------------------------------------*/

44
45
46
47
48
49
/* Gather stats about entry, opcode, opcode-pair frequencies.  For
   tuning the interpreter. */

/* #define INTERP_STATS */


50
/* Sp points to the lowest live word on the stack. */
51

52
#define BCO_NEXT      instrs[bciPtr++]
53
54
55
56
57
58
59
#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
60
#error Cannot cope with WORD_SIZE_IN_BITS being nether 32 nor 64
61
62
63
#endif
#define BCO_GET_LARGE_ARG ((bci & bci_FLAG_LARGE_ARGS) ? BCO_NEXT_WORD : BCO_NEXT)

64
#define BCO_PTR(n)    (W_)ptrs[n]
65
#define BCO_LIT(n)    literals[n]
66

67
68
69
70
#define LOAD_STACK_POINTERS					\
    Sp = cap->r.rCurrentTSO->sp;				\
    /* We don't change this ... */				\
    SpLim = cap->r.rCurrentTSO->stack + RESERVED_STACK_WORDS;
71

72
#define SAVE_STACK_POINTERS			\
73
    ASSERT(Sp > SpLim); \
74
    cap->r.rCurrentTSO->sp = Sp
75

76
#define RETURN_TO_SCHEDULER(todo,retcode)	\
77
78
   SAVE_STACK_POINTERS;				\
   cap->r.rCurrentTSO->what_next = (todo);	\
79
   threadPaused(cap,cap->r.rCurrentTSO);		\
80
81
   cap->r.rRet = (retcode);			\
   return cap;
82
83

#define RETURN_TO_SCHEDULER_NO_PAUSE(todo,retcode)	\
84
85
86
87
   SAVE_STACK_POINTERS;					\
   cap->r.rCurrentTSO->what_next = (todo);		\
   cap->r.rRet = (retcode);				\
   return cap;
88
89


sof's avatar
sof committed
90
STATIC_INLINE StgPtr
91
allocate_NONUPD (Capability *cap, int n_words)
92
{
93
    return allocate(cap, stg_max(sizeofW(StgHeader)+MIN_PAYLOAD_SIZE, n_words));
94
95
}

96
97
int rts_stop_next_breakpoint = 0;
int rts_stop_on_exception = 0;
98

99
#ifdef INTERP_STATS
100

101
102
103
104
105
106
107
108
109
110
111
112
113
/* 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;

114
115
int it_ofreq[27];
int it_oofreq[27][27];
116
117
int it_lastopc;

118

119
120
#define INTERP_TICK(n) (n)++

121
122
123
124
125
126
127
128
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;
129
130
131
   for (i = 0; i < 27; i++) it_ofreq[i] = 0;
   for (i = 0; i < 27; i++) 
     for (j = 0; j < 27; j++)
132
133
134
135
136
137
138
        it_oofreq[i][j] = 0;
   it_lastopc = 0;
}

void interp_shutdown ( void )
{
   int i, j, k, o_max, i_max, j_max;
139
   debugBelch("%d constrs entered -> (%d BCO, %d UPD, %d ??? )\n",
140
141
                   it_retto_BCO + it_retto_UPDATE + it_retto_other,
                   it_retto_BCO, it_retto_UPDATE, it_retto_other );
142
   debugBelch("%d total entries, %d unknown entries \n", 
143
144
145
                   it_total_entries, it_total_unknown_entries);
   for (i = 0; i < N_CLOSURE_TYPES; i++) {
     if (it_unknown_entries[i] == 0) continue;
146
     debugBelch("   type %2d: unknown entries (%4.1f%%) == %d\n",
147
148
149
150
	     i, 100.0 * ((double)it_unknown_entries[i]) / 
                        ((double)it_total_unknown_entries),
             it_unknown_entries[i]);
   }
151
   debugBelch("%d insns, %d slides, %d BCO_entries\n", 
152
                   it_insns, it_slides, it_BCO_entries);
153
   for (i = 0; i < 27; i++) 
154
      debugBelch("opcode %2d got %d\n", i, it_ofreq[i] );
155
156
157
158

   for (k = 1; k < 20; k++) {
      o_max = 0;
      i_max = j_max = 0;
159
160
      for (i = 0; i < 27; i++) {
         for (j = 0; j < 27; j++) {
161
162
163
164
165
166
167
	    if (it_oofreq[i][j] > o_max) {
               o_max = it_oofreq[i][j];
	       i_max = i; j_max = j;
	    }
	 }
      }
      
168
      debugBelch("%d:  count (%4.1f%%) %6d   is %d then %d\n",
169
170
171
172
173
174
175
                k, ((double)o_max) * 100.0 / ((double)it_insns), o_max,
                   i_max, j_max );
      it_oofreq[i_max][j_max] = 0;

   }
}

176
177
178
179
180
#else // !INTERP_STATS

#define INTERP_TICK(n) /* nothing */

#endif
181

182
183
184
185
186
187
188
189
190
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,
};

191
HsStablePtr rts_breakpoint_io_action; // points to the IO action which is executed on a breakpoint
192
193
                                // it is set in main/GHC.hs:runStmt

194
Capability *
195
interpretBCO (Capability* cap)
196
{
197
198
199
200
    // 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
201
    register StgClosure   *tagged_obj = 0, *obj;
202
    nat n, m;
203

204
205
    LOAD_STACK_POINTERS;

206
207
208
    cap->r.rHpLim = (P_)1; // HpLim is the context-switch flag; when it
                           // goes to zero we must return to the scheduler.

209
210
211
212
213
214
215
216
217
218
219
    // ------------------------------------------------------------------------
    // Case 1:
    // 
    //       We have a closure to evaluate.  Stack looks like:
    //       
    //      	|   XXXX_info   |
    //      	+---------------+
    //       Sp |      -------------------> closure
    //      	+---------------+
    //       
    if (Sp[0] == (W_)&stg_enter_info) {
220
221
       Sp++;
       goto eval;
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
    }

    // ------------------------------------------------------------------------
    // 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
239
	obj = UNTAG_CLOSURE((StgClosure *)Sp[1]);
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
	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:
256
    tagged_obj = (StgClosure*)Sp[0]; Sp++;
257
258

eval_obj:
259
    obj = UNTAG_CLOSURE(tagged_obj);
260
261
262
    INTERP_TICK(it_total_evals);

    IF_DEBUG(interpreter,
263
             debugBelch(
264
             "\n---------------------------------------------------------------\n");
265
266
267
             debugBelch("Evaluating: "); printObj(obj);
             debugBelch("Sp = %p\n", Sp);
             debugBelch("\n" );
268

269
             printStackChunk(Sp,cap->r.rCurrentTSO->stack+cap->r.rCurrentTSO->stack_size);
270
             debugBelch("\n\n");
271
            );
272

273
274
//    IF_DEBUG(sanity,checkStackChunk(Sp, cap->r.rCurrentTSO->stack+cap->r.rCurrentTSO->stack_size));
    IF_DEBUG(sanity,checkStackFrame(Sp));
275

276
    switch ( get_itbl(obj)->type ) {
277

278
279
280
281
    case IND:
    case IND_PERM:
    case IND_STATIC:
    { 
282
	tagged_obj = ((StgInd*)obj)->indirectee;
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
	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:
306
    {
307
	ASSERT(((StgBCO *)obj)->arity > 0);
308
	break;
309
    }
310
311
312
313
314
315
316
317
318
319
320
321

    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;
322
	    Sp[1] = (W_)tagged_obj;
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
	    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
342
	obj = UNTAG_CLOSURE((StgClosure*)ap->fun);
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
	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,
361
		 debugBelch("evaluating unknown closure -- yielding to sched\n"); 
362
363
364
		 printObj(obj);
	    );
	Sp -= 2;
365
	Sp[1] = (W_)tagged_obj;
366
	Sp[0] = (W_)&stg_enter_info;
367
	RETURN_TO_SCHEDULER_NO_PAUSE(ThreadRunGHC, ThreadYielding);
368
369
370
371
    }
    }

    // ------------------------------------------------------------------------
372
    // We now have an evaluated object (tagged_obj).  The next thing to
373
374
    // do is return it to the stack frame on top of the stack.
do_return:
375
    obj = UNTAG_CLOSURE(tagged_obj);
376
377
378
    ASSERT(closure_HNF(obj));

    IF_DEBUG(interpreter,
379
             debugBelch(
380
             "\n---------------------------------------------------------------\n");
381
382
383
             debugBelch("Returning: "); printObj(obj);
             debugBelch("Sp = %p\n", Sp);
             debugBelch("\n" );
384
             printStackChunk(Sp,cap->r.rCurrentTSO->stack+cap->r.rCurrentTSO->stack_size);
385
             debugBelch("\n\n");
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
    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.
436
437
438
439
440
441
442
443
        //
        // 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.
444
	INTERP_TICK(it_retto_UPDATE);
445
446
        updateThunk(cap, cap->r.rCurrentTSO, 
                    ((StgUpdateFrame *)Sp)->updatee, tagged_obj);
447
448
449
450
451
452
453
454
455
	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;
456
457
        // NB. return the untagged object; the bytecode expects it to
        // be untagged.  XXX this doesn't seem right.
458
459
460
461
462
463
464
465
466
467
	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,
468
		 debugBelch("returning to unknown frame -- yielding to sched\n"); 
469
470
471
		 printStackChunk(Sp,cap->r.rCurrentTSO->stack+cap->r.rCurrentTSO->stack_size);
	    );
	Sp -= 2;
472
	Sp[1] = (W_)tagged_obj;
473
	Sp[0] = (W_)&stg_enter_info;
474
	RETURN_TO_SCHEDULER_NO_PAUSE(ThreadRunGHC, ThreadYielding);
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
    }
    }

    // -------------------------------------------------------------------------
    // 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,
531
		     debugBelch("returning to unknown frame -- yielding to sched\n"); 
532
533
		     printStackChunk(Sp,cap->r.rCurrentTSO->stack+cap->r.rCurrentTSO->stack_size);
		);
534
	    RETURN_TO_SCHEDULER_NO_PAUSE(ThreadRunGHC, ThreadYielding);
535
536
537
538
539
540
541
542
543
544
	}
	}
    }
    // not reached.


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

do_apply:
545
    ASSERT(obj == UNTAG_CLOSURE(tagged_obj));
546
547
548
549
550
551
552
553
    // 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;
554
	    nat i, arity;
555
556
557
558

	    pap = (StgPAP *)obj;

	    // we only cope with PAPs whose function is a BCO
Simon Marlow's avatar
Simon Marlow committed
559
	    if (get_itbl(UNTAG_CLOSURE(pap->fun))->type != BCO) {
560
561
		goto defer_apply_to_sched;
	    }
562

563
564
565
566
567
568
569
570
571
572
            // 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);
            }

573
574
575
576
577
578
579
580
581
582
583
	    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++) {
584
585
		    Sp[(int)i-1] = Sp[i];
		    // ^^^^^ careful, i-1 might be negative, but i in unsigned
586
587
588
589
590
591
592
593
		}
		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
594
		obj = UNTAG_CLOSURE(pap->fun);
595
596
597
598
599
600
601
		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
602
		obj = UNTAG_CLOSURE(pap->fun);
603
604
605
606
607
		goto run_BCO_fun;
	    } 
	    else /* arity > n */ {
		// build a new PAP and return it.
		StgPAP *new_pap;
608
		new_pap = (StgPAP *)allocate(cap, PAP_sizeW(pap->n_args + m));
609
610
611
612
613
614
615
616
617
618
		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];
		}
619
		tagged_obj = (StgClosure *)new_pap;
620
621
622
623
624
625
		Sp += m;
		goto do_return;
	    }
	}	    

	case BCO: {
626
	    nat arity, i;
627
628

	    Sp++;
629
	    arity = ((StgBCO *)obj)->arity;
630
631
632
633
634
635
636
637
638
	    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++) {
639
640
		    Sp[(int)i-1] = Sp[i];
		    // ^^^^^ careful, i-1 might be negative, but i in unsigned
641
642
643
644
645
646
647
648
649
650
651
		}
		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;
652
		nat i;
653
		pap = (StgPAP *)allocate(cap, PAP_sizeW(m));
654
655
656
657
658
659
660
		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];
		}
661
		tagged_obj = (StgClosure *)pap;
662
663
664
665
666
667
668
669
670
		Sp += m;
		goto do_return;
	    }
	}

	// No point in us applying machine-code functions
	default:
	defer_apply_to_sched:
	    Sp -= 2;
671
	    Sp[1] = (W_)tagged_obj;
672
	    Sp[0] = (W_)&stg_enter_info;
673
	    RETURN_TO_SCHEDULER_NO_PAUSE(ThreadRunGHC, ThreadYielding);
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
716
717
718
    }

    // ------------------------------------------------------------------------
    // 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:

719

720
721
run_BCO_return:
    // Heap check
722
    if (doYouWantToGC(cap)) {
723
724
725
	Sp--; Sp[0] = (W_)&stg_enter_info;
	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
    goto run_BCO;
    
run_BCO_return_unboxed:
    // Heap check
733
    if (doYouWantToGC(cap)) {
734
735
	RETURN_TO_SCHEDULER(ThreadInterpret, HeapOverflow);
    }
736
737
    // Stack checks aren't necessary at return points, the stack use
    // is aggregated into the enclosing function entry point.
738

739
740
741
742
743
744
745
746
747
748
749
750
    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
751
    if (doYouWantToGC(cap)) {
752
753
754
755
756
757
	Sp -= 2; 
	Sp[1] = (W_)obj; 
	Sp[0] = (W_)&stg_apply_interp_info; // placeholder, really
	RETURN_TO_SCHEDULER(ThreadInterpret, HeapOverflow);
    }
    
758
759
    // Stack check
    if (Sp - INTERP_STACK_CHECK_THRESH < SpLim) {
760
761
762
763
764
	Sp -= 2; 
	Sp[1] = (W_)obj; 
	Sp[0] = (W_)&stg_apply_interp_info; // placeholder, really
	RETURN_TO_SCHEDULER(ThreadInterpret, StackOverflow);
    }
765

766
767
768
769
770
771
772
    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);
    {
773
	register int       bciPtr = 0; /* instruction pointer */
774
        register StgWord16 bci;
775
	register StgBCO*   bco        = (StgBCO*)obj;
776
	register StgWord16* instrs    = (StgWord16*)(bco->instrs->payload);
777
778
	register StgWord*  literals   = (StgWord*)(&bco->literals->payload[0]);
	register StgPtr*   ptrs       = (StgPtr*)(&bco->ptrs->payload[0]);
779
780
781
	int bcoSize;
    bcoSize = BCO_NEXT_WORD;
	IF_DEBUG(interpreter,debugBelch("bcoSize = %d\n", bcoSize));
782

783
784
785
#ifdef INTERP_STATS
	it_lastopc = 0; /* no opcode */
#endif
786

787
    nextInsn:
788
	ASSERT(bciPtr < bcoSize);
789
790
	IF_DEBUG(interpreter,
		 //if (do_print_stack) {
791
		 //debugBelch("\n-- BEGIN stack\n");
792
		 //printStack(Sp,cap->r.rCurrentTSO->stack+cap->r.rCurrentTSO->stack_size,iSu);
793
		 //debugBelch("-- END stack\n\n");
794
		 //}
795
		 debugBelch("Sp = %p   pc = %d      ", Sp, bciPtr);
796
797
		 disInstr(bco,bciPtr);
		 if (0) { int i;
798
		 debugBelch("\n");
799
		 for (i = 8; i >= 0; i--) {
800
		     debugBelch("%d  %p\n", i, (StgPtr)(*(Sp+i)));
801
		 }
802
		 debugBelch("\n");
803
		 }
804
805
		 //if (do_print_stack) checkStack(Sp,cap->r.rCurrentTSO->stack+cap->r.rCurrentTSO->stack_size,iSu);
	    );
806

807

808
809
810
811
812
813
814
815
816
	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

817
818
819
820
821
822
	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) {
823

824
825
826
827
828
        /* 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;
829
830
            int returning_from_break;     // are we resuming execution from a breakpoint?
                                          //  if yes, then don't break this time around
831
832
833
834
835
836
            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;

837
838
839
            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
840

841
842
            // check if we are returning from a breakpoint - this info
            // is stored in the flags field of the current TSO
843
844
            returning_from_break = cap->r.rCurrentTSO->flags & TSO_STOPPED_ON_BREAKPOINT; 

845
846
            // if we are returning from a break then skip this section
            // and continue executing
847
848
849
850
            if (!returning_from_break)
            {
               breakPoints = (StgArrWords *) BCO_PTR(arg1_brk_array);

851
               // stop the current thread if either the
852
               // "rts_stop_next_breakpoint" flag is true OR if the
853
854
               // breakpoint flag for this particular expression is
               // true
855
               if (rts_stop_next_breakpoint == rtsTrue || 
856
                   breakPoints->payload[arg2_array_index] == rtsTrue)
857
               {
858
859
                  // make sure we don't automatically stop at the
                  // next breakpoint
860
                  rts_stop_next_breakpoint = rtsFalse;
861
862
863
864
865

                  // 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
866
                  size_words = BCO_BITMAP_SIZE(obj) + 2;
867
                  new_aps = (StgAP_STACK *) allocate(cap, AP_STACK_sizeW(size_words));
868
869
870
871
872
                  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 
873
874
                  new_aps->payload[0] = (StgClosure *)&stg_apply_interp_info;
                  new_aps->payload[1] = (StgClosure *)obj;
875
876
877
878

                  // copy the contents of the top stack frame into the AP_STACK
                  for (i = 2; i < size_words; i++)
                  {
879
                     new_aps->payload[i] = (StgClosure *)Sp[i-2];
880
881
                  }

882
                  // prepare the stack so that we can call the
883
                  // rts_breakpoint_io_action and ensure that the stack is
884
885
                  // in a reasonable state for the GC and so that
                  // execution of this BCO can continue when we resume
886
                  ioAction = (StgClosure *) deRefStablePtr (rts_breakpoint_io_action);
887
888
889
                  Sp -= 8;
                  Sp[7] = (W_)obj;
                  Sp[6] = (W_)&stg_apply_interp_info;
890
891
892
893
                  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;
894
895
896
897
898
899
                  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
                  // 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
900
901
                  cap->r.rCurrentTSO->flags |= TSO_STOPPED_ON_BREAKPOINT;

902
903
904
                  // 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
905
906
907
908
909
910
911
912
913
914
                  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;
        }

915
916
917
918
	case bci_STKCHECK: {
	    // Explicit stack check at the beginning of a function
	    // *only* (stack checks in case alternatives are
	    // propagated to the enclosing function).
919
	    StgWord stk_words_reqd = BCO_GET_LARGE_ARG + 1;
920
	    if (Sp - stk_words_reqd < SpLim) {
921
922
923
		Sp -= 2; 
		Sp[1] = (W_)obj; 
		Sp[0] = (W_)&stg_apply_interp_info;
924
		RETURN_TO_SCHEDULER(ThreadInterpret, StackOverflow);
925
926
	    } else {
		goto nextInsn;
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
959
960
961
962
963
964
965
	    }
	}

	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;
966
	    Sp[-2] = (W_)&stg_ctoi_R1p_info;
967
968
969
970
971
972
973
	    Sp[-1] = BCO_PTR(o_bco);
	    Sp -= 2;
	    goto nextInsn;
	}

	case bci_PUSH_ALTS_P: {
	    int o_bco  = BCO_NEXT;
974
	    Sp[-2] = (W_)&stg_ctoi_R1unpt_info;
975
976
977
978
979
980
981
	    Sp[-1] = BCO_PTR(o_bco);
	    Sp -= 2;
	    goto nextInsn;
	}

	case bci_PUSH_ALTS_N: {
	    int o_bco  = BCO_NEXT;
982
	    Sp[-2] = (W_)&stg_ctoi_R1n_info;
983
984
985
986
987
988
989
	    Sp[-1] = BCO_PTR(o_bco);
	    Sp -= 2;
	    goto nextInsn;
	}

	case bci_PUSH_ALTS_F: {
	    int o_bco  = BCO_NEXT;
990
	    Sp[-2] = (W_)&stg_ctoi_F1_info;
991
992
993
994
995
996
997
	    Sp[-1] = BCO_PTR(o_bco);
	    Sp -= 2;
	    goto nextInsn;
	}

	case bci_PUSH_ALTS_D: {
	    int o_bco  = BCO_NEXT;
998
	    Sp[-2] = (W_)&stg_ctoi_D1_info;
999
1000
1001
1002
1003
1004
1005
	    Sp[-1] = BCO_PTR(o_bco);
	    Sp -= 2;
	    goto nextInsn;
	}

	case bci_PUSH_ALTS_L: {
	    int o_bco  = BCO_NEXT;
1006
	    Sp[-2] = (W_)&stg_ctoi_L1_info;
1007
1008
1009
1010
1011
1012
1013
	    Sp[-1] = BCO_PTR(o_bco);
	    Sp -= 2;
	    goto nextInsn;
	}

	case bci_PUSH_ALTS_V: {
	    int o_bco  = BCO_NEXT;
1014
	    Sp[-2] = (W_)&stg_ctoi_V_info;
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
1053
1054
1055
1056
1057
1058
1059
	    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++) {
1060
		Sp[i] = (W_)BCO_LIT(o_lits+i);
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
	    }
	    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
1079
	    int n_payload = BCO_NEXT;
1080
	    ap = (StgAP*)allocate(cap, AP_sizeW(n_payload));
1081
1082
1083
1084
1085
1086
1087
	    Sp[-1] = (W_)ap;
	    ap->n_args = n_payload;
	    SET_HDR(ap, &stg_AP_info, CCS_SYSTEM/*ToDo*/)
	    Sp --;
	    goto nextInsn;
	}

1088
1089
1090
	case bci_ALLOC_AP_NOUPD: {
	    StgAP* ap; 
	    int n_payload = BCO_NEXT;
1091
	    ap = (StgAP*)allocate(cap, AP_sizeW(n_payload));
1092
1093
1094
1095
1096
1097
1098
	    Sp[-1] = (W_)ap;
	    ap->n_args = n_payload;
	    SET_HDR(ap, &stg_AP_NOUPD_info, CCS_SYSTEM/*ToDo*/)
	    Sp --;
	    goto nextInsn;
	}

1099
1100
1101
	case bci_ALLOC_PAP: {
	    StgPAP* pap; 
	    int arity = BCO_NEXT;
sof's avatar
sof committed
1102
	    int n_payload = BCO_NEXT;
1103
	    pap = (StgPAP*)allocate(cap, PAP_sizeW(n_payload));
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
	    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
1115
	    int n_payload = BCO_NEXT;
1116
1117
1118
	    StgAP* ap = (StgAP*)Sp[stkoff];
	    ASSERT((int)ap->n_args == n_payload);
	    ap->fun = (StgClosure*)Sp[0];
1119
	    
1120
1121
1122
	    // The function should be a BCO, and its bitmap should
	    // cover the payload of the AP correctly.
	    ASSERT(get_itbl(ap->fun)->type == BCO
1123
1124
		   && BCO_BITMAP_SIZE(ap->fun) == ap->n_args);
	    
1125
1126
1127
1128
	    for (i = 0; i < n_payload; i++)
		ap->payload[i] = (StgClosure*)Sp[i+1];
	    Sp += n_payload+1;
	    IF_DEBUG(interpreter,
1129
		     debugBelch("\tBuilt "); 
1130
1131
1132
1133
1134
		     printObj((StgClosure*)ap);
		);
	    goto nextInsn;
	}

1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
	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;
	}

1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
	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;
1172
	    StgInfoTable* itbl = INFO_PTR_TO_STRUCT(BCO_LIT(o_itbl));
1173
1174
	    int request        = CONSTR_sizeW( itbl->layout.payload.ptrs, 
					       itbl->layout.payload.nptrs );
1175
	    StgClosure* con = (StgClosure*)allocate_NONUPD(cap,request);
1176
	    ASSERT( itbl->layout.payload.ptrs + itbl->layout.payload.nptrs > 0);
1177
	    SET_HDR(con, (StgInfoTable*)BCO_LIT(o_itbl), CCS_SYSTEM/*ToDo*/);
1178
1179
1180
1181
1182
1183
1184
	    for (i = 0; i < n_words; i++) {
		con->payload[i] = (StgClosure*)Sp[i];
	    }
	    Sp += n_words;
	    Sp --;
	    Sp[0] = (W_)con;
	    IF_DEBUG(interpreter,
1185
		     debugBelch("\tBuilt "); 
1186
1187
1188
1189
1190
1191
		     printObj((StgClosure*)con);
		);
	    goto nextInsn;
	}

	case bci_TESTLT_P: {
1192
	    unsigned int discr  = BCO_NEXT;
1193
	    int failto = BCO_GET_LARGE_ARG;
1194
	    StgClosure* con = (StgClosure*)Sp[0];
1195
	    if (GET_TAG(con) >= discr) {
1196
1197
1198
1199
1200
1201
		bciPtr = failto;
	    }
	    goto nextInsn;
	}

	case bci_TESTEQ_P: {
1202
	    unsigned int discr  = BCO_NEXT;
1203
	    int failto = BCO_GET_LARGE_ARG;
1204
	    StgClosure* con = (StgClosure*)Sp[0];
1205
	    if (GET_TAG(con) != discr) {
1206
1207
1208
1209
1210
1211
1212
1213
		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;
1214
	    int failto  = BCO_GET_LARGE_ARG;
1215
1216
1217
1218
1219
1220
1221
1222
1223
	    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;
1224
	    int failto  = BCO_GET_LARGE_ARG;
1225
1226
1227
1228
1229
1230
1231
	    I_ stackInt = (I_)Sp[1];
	    if (stackInt != (I_)BCO_LIT(discr)) {
		bciPtr = failto;
	    }
	    goto nextInsn;
	}

1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
	case bci_TESTLT_W: {
	    // There should be an Int at Sp[1], and an info table at Sp[0].
	    int discr   = BCO_NEXT;
	    int failto  = BCO_GET_LARGE_ARG;
	    W_ stackWord = (W_)Sp[1];
	    if (stackWord >= (W_)BCO_LIT(discr))
		bciPtr = failto;
	    goto nextInsn;
	}

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

1253
1254
1255
	case bci_TESTLT_D: {
	    // There should be a Double at Sp[1], and an info table at Sp[0].
	    int discr   = BCO_NEXT;
1256
	    int failto  = BCO_GET_LARGE_ARG;
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
	    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;
1269
	    int failto  = BCO_GET_LARGE_ARG;
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
	    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;
1282
	    int failto  = BCO_GET_LARGE_ARG;
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
	    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;
1295
	    int failto  = BCO_GET_LARGE_ARG;
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
	    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.
1312
	    if (cap->r.rHpLim == NULL) {
1313
1314
1315
1316
1317
1318
		Sp--; Sp[0] = (W_)&stg_enter_info;
		RETURN_TO_SCHEDULER(ThreadInterpret, ThreadYielding);
	    }
	    goto eval;

	case bci_RETURN:
1319
	    tagged_obj = (StgClosure *)Sp[0];
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
	    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: {
1356
	    void *tok;
1357
1358
1359
	    int stk_offset            = BCO_NEXT;
	    int o_itbl                = BCO_NEXT;
	    void(*marshall_fn)(void*) = (void (*)(void*))BCO_LIT(o_itbl);
1360
1361
1362
	    int ret_dyn_size = 
		RET_DYN_BITMAP_SIZE + RET_DYN_NONPTR_REGS_SIZE
		+ sizeofW(StgRetDyn);
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
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
            /* 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);
            }
1420