RetainerProfile.c 66 KB
Newer Older
1
/* -----------------------------------------------------------------------------
2
 * $Id: RetainerProfile.c,v 1.10 2003/05/16 14:39:29 simonmar Exp $
3 4 5 6 7 8 9 10 11 12
 *
 * (c) The GHC Team, 2001
 * Author: Sungwoo Park
 *
 * Retainer profiling.
 *
 * ---------------------------------------------------------------------------*/

#ifdef PROFILING

13 14 15 16 17 18 19
// Turn off inlining when debugging - it obfuscates things
#ifdef DEBUG
#define INLINE
#else
#define INLINE inline
#endif

20 21
#include <stdio.h>

22 23 24 25 26 27 28 29 30 31 32
#include "Rts.h"
#include "RtsUtils.h"
#include "RetainerProfile.h"
#include "RetainerSet.h"
#include "Schedule.h"
#include "Printer.h"
#include "Storage.h"
#include "StoragePriv.h"
#include "RtsFlags.h"
#include "Weak.h"
#include "Sanity.h"
33
#include "StablePriv.h"
34 35 36 37
#include "Profiling.h"
#include "Stats.h"
#include "BlockAlloc.h"
#include "ProfHeap.h"
38
#include "Apply.h"
39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

/*
  Note: what to change in order to plug-in a new retainer profiling scheme?
    (1) type retainer in ../includes/StgRetainerProf.h
    (2) retainer function R(), i.e., getRetainerFrom()
    (3) the two hashing functions, hashKeySingleton() and hashKeyAddElement(),
        in RetainerSet.h, if needed.
    (4) printRetainer() and printRetainerSetShort() in RetainerSet.c.
 */

/* -----------------------------------------------------------------------------
 * Declarations...
 * -------------------------------------------------------------------------- */

static nat retainerGeneration;	// generation

static nat numObjectVisited;	// total number of objects visited
static nat timesAnyObjectVisited; // number of times any objects are visited

/*
  The rs field in the profile header of any object points to its retainer
  set in an indirect way: if flip is 0, it points to the retainer set;
  if flip is 1, it points to the next byte after the retainer set (even
  for NULL pointers). Therefore, with flip 1, (rs ^ 1) is the actual
  pointer. See retainerSetOf().
 */

66
StgWord flip = 0;     // flip bit
67 68 69 70 71
                      // must be 0 if DEBUG_RETAINER is on (for static closures)

#define setRetainerSetToNull(c)   \
  (c)->header.prof.hp.rs = (RetainerSet *)((StgWord)NULL | flip)

72
static void retainStack(StgClosure *, retainer, StgPtr, StgPtr);
73
static void retainClosure(StgClosure *, StgClosure *, retainer);
74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113
#ifdef DEBUG_RETAINER
static void belongToHeap(StgPtr p);
#endif

#ifdef DEBUG_RETAINER
/*
  cStackSize records how many times retainStack() has been invoked recursively,
  that is, the number of activation records for retainStack() on the C stack.
  maxCStackSize records its max value.
  Invariants:
    cStackSize <= maxCStackSize
 */
static nat cStackSize, maxCStackSize;

static nat sumOfNewCost;	// sum of the cost of each object, computed
				// when the object is first visited
static nat sumOfNewCostExtra;   // for those objects not visited during
                                // retainer profiling, e.g., MUT_VAR
static nat costArray[N_CLOSURE_TYPES];

nat sumOfCostLinear;		// sum of the costs of all object, computed
				// when linearly traversing the heap after
				// retainer profiling
nat costArrayLinear[N_CLOSURE_TYPES];
#endif

/* -----------------------------------------------------------------------------
 * Retainer stack - header
 *   Note:
 *     Although the retainer stack implementation could be separated *
 *     from the retainer profiling engine, there does not seem to be
 *     any advantage in doing that; retainer stack is an integral part
 *     of retainer profiling engine and cannot be use elsewhere at
 *     all.
 * -------------------------------------------------------------------------- */

typedef enum {
    posTypeStep,
    posTypePtrs,
    posTypeSRT,
114
    posTypeLargeSRT,
115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136
} nextPosType;

typedef union {
    // fixed layout or layout specified by a field in the closure
    StgWord step;

    // layout.payload
    struct {
    // See StgClosureInfo in InfoTables.h
#if SIZEOF_VOID_P == 8
	StgWord32 pos;
	StgWord32 ptrs;
#else
	StgWord16 pos;
	StgWord16 ptrs;
#endif
	StgPtr payload;
    } ptrs;

    // SRT
    struct {
	StgClosure **srt;
137
	StgWord    srt_bitmap;
138
    } srt;
139 140 141 142 143 144 145

    // Large SRT
    struct {
	StgLargeSRT *srt;
	StgWord offset;
    } large_srt;
	
146 147 148 149 150 151 152 153 154
} nextPos;

typedef struct {
    nextPosType type;
    nextPos next;
} stackPos;

typedef struct {
    StgClosure *c;
155
    retainer c_child_r;
156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175
    stackPos info;
} stackElement;

/*
  Invariants:
    firstStack points to the first block group.
    currentStack points to the block group currently being used.
    currentStack->free == stackLimit.
    stackTop points to the topmost byte in the stack of currentStack.
    Unless the whole stack is empty, stackTop must point to the topmost
    object (or byte) in the whole stack. Thus, it is only when the whole stack
    is empty that stackTop == stackLimit (not during the execution of push()
    and pop()).
    stackBottom == currentStack->start.
    stackLimit == currentStack->start + BLOCK_SIZE_W * currentStack->blocks.
  Note:
    When a current stack becomes empty, stackTop is set to point to
    the topmost element on the previous block group so as to satisfy
    the invariants described above.
 */
sof's avatar
sof committed
176
static bdescr *firstStack = NULL;
177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211
static bdescr *currentStack;
static stackElement *stackBottom, *stackTop, *stackLimit;

/*
  currentStackBoundary is used to mark the current stack chunk.
  If stackTop == currentStackBoundary, it means that the current stack chunk
  is empty. It is the responsibility of the user to keep currentStackBoundary
  valid all the time if it is to be employed.
 */
static stackElement *currentStackBoundary;

/*
  stackSize records the current size of the stack.
  maxStackSize records its high water mark.
  Invariants:
    stackSize <= maxStackSize
  Note:
    stackSize is just an estimate measure of the depth of the graph. The reason
    is that some heap objects have only a single child and may not result
    in a new element being pushed onto the stack. Therefore, at the end of
    retainer profiling, maxStackSize + maxCStackSize is some value no greater
    than the actual depth of the graph.
 */
#ifdef DEBUG_RETAINER
static int stackSize, maxStackSize;
#endif

// number of blocks allocated for one stack
#define BLOCKS_IN_STACK 1

/* -----------------------------------------------------------------------------
 * Add a new block group to the stack.
 * Invariants:
 *  currentStack->link == s.
 * -------------------------------------------------------------------------- */
212
static INLINE void
213 214 215 216 217 218 219 220 221 222 223 224 225 226
newStackBlock( bdescr *bd )
{
    currentStack = bd;
    stackTop     = (stackElement *)(bd->start + BLOCK_SIZE_W * bd->blocks);
    stackBottom  = (stackElement *)bd->start;
    stackLimit   = (stackElement *)stackTop;
    bd->free     = (StgPtr)stackLimit;
}

/* -----------------------------------------------------------------------------
 * Return to the previous block group.
 * Invariants:
 *   s->link == currentStack.
 * -------------------------------------------------------------------------- */
227
static INLINE void
228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268
returnToOldStack( bdescr *bd )
{
    currentStack = bd;
    stackTop = (stackElement *)bd->free;
    stackBottom = (stackElement *)bd->start;
    stackLimit = (stackElement *)(bd->start + BLOCK_SIZE_W * bd->blocks);
    bd->free = (StgPtr)stackLimit;
}

/* -----------------------------------------------------------------------------
 *  Initializes the traverse stack.
 * -------------------------------------------------------------------------- */
static void
initializeTraverseStack( void )
{
    if (firstStack != NULL) {
	freeChain(firstStack);
    }

    firstStack = allocGroup(BLOCKS_IN_STACK);
    firstStack->link = NULL;
    firstStack->u.back = NULL;

    newStackBlock(firstStack);
}

/* -----------------------------------------------------------------------------
 * Frees all the block groups in the traverse stack.
 * Invariants:
 *   firstStack != NULL
 * -------------------------------------------------------------------------- */
static void
closeTraverseStack( void )
{
    freeChain(firstStack);
    firstStack = NULL;
}

/* -----------------------------------------------------------------------------
 * Returns rtsTrue if the whole stack is empty.
 * -------------------------------------------------------------------------- */
269
static INLINE rtsBool
270 271 272 273 274
isEmptyRetainerStack( void )
{
    return (firstStack == currentStack) && stackTop == stackLimit;
}

sof's avatar
sof committed
275 276 277
/* -----------------------------------------------------------------------------
 * Returns size of stack
 * -------------------------------------------------------------------------- */
278
#ifdef DEBUG
sof's avatar
sof committed
279
lnat
280
retainerStackBlocks( void )
sof's avatar
sof committed
281 282 283 284 285 286 287 288 289
{
    bdescr* bd;
    lnat res = 0;

    for (bd = firstStack; bd != NULL; bd = bd->link) 
      res += bd->blocks;

    return res;
}
290
#endif
sof's avatar
sof committed
291

292 293 294 295
/* -----------------------------------------------------------------------------
 * Returns rtsTrue if stackTop is at the stack boundary of the current stack,
 * i.e., if the current stack chunk is empty.
 * -------------------------------------------------------------------------- */
296
static INLINE rtsBool
297 298 299 300 301 302 303 304 305 306
isOnBoundary( void )
{
    return stackTop == currentStackBoundary;
}

/* -----------------------------------------------------------------------------
 * Initializes *info from ptrs and payload.
 * Invariants:
 *   payload[] begins with ptrs pointers followed by non-pointers.
 * -------------------------------------------------------------------------- */
307
static INLINE void
308 309 310 311 312 313 314 315 316 317 318
init_ptrs( stackPos *info, nat ptrs, StgPtr payload )
{
    info->type              = posTypePtrs;
    info->next.ptrs.pos     = 0;
    info->next.ptrs.ptrs    = ptrs;
    info->next.ptrs.payload = payload;
}

/* -----------------------------------------------------------------------------
 * Find the next object from *info.
 * -------------------------------------------------------------------------- */
319
static INLINE StgClosure *
320 321 322 323 324 325 326 327 328 329 330 331
find_ptrs( stackPos *info )
{
    if (info->next.ptrs.pos < info->next.ptrs.ptrs) {
	return (StgClosure *)info->next.ptrs.payload[info->next.ptrs.pos++];
    } else {
	return NULL;
    }
}

/* -----------------------------------------------------------------------------
 *  Initializes *info from SRT information stored in *infoTable.
 * -------------------------------------------------------------------------- */
332
static INLINE void
333
init_srt_fun( stackPos *info, StgFunInfoTable *infoTable )
334
{
335 336 337 338 339 340 341 342 343
    if (infoTable->i.srt_bitmap == (StgHalfWord)(-1)) {
	info->type = posTypeLargeSRT;
	info->next.large_srt.srt = (StgLargeSRT *)infoTable->srt;
	info->next.large_srt.offset = 0;
    } else {
	info->type = posTypeSRT;
	info->next.srt.srt = (StgClosure **)(infoTable->srt);
	info->next.srt.srt_bitmap = infoTable->i.srt_bitmap;
    }
344 345
}

346
static INLINE void
347 348
init_srt_thunk( stackPos *info, StgThunkInfoTable *infoTable )
{
349 350 351 352 353 354 355 356 357
    if (infoTable->i.srt_bitmap == (StgHalfWord)(-1)) {
	info->type = posTypeLargeSRT;
	info->next.large_srt.srt = (StgLargeSRT *)infoTable->srt;
	info->next.large_srt.offset = 0;
    } else {
	info->type = posTypeSRT;
	info->next.srt.srt = (StgClosure **)(infoTable->srt);
	info->next.srt.srt_bitmap = infoTable->i.srt_bitmap;
    }
358 359 360 361 362
}

/* -----------------------------------------------------------------------------
 * Find the next object from *info.
 * -------------------------------------------------------------------------- */
363
static INLINE StgClosure *
364 365 366
find_srt( stackPos *info )
{
    StgClosure *c;
367
    StgWord bitmap;
368

369 370 371 372 373
    if (info->type == posTypeSRT) {
	// Small SRT bitmap
	bitmap = info->next.srt.srt_bitmap;
	while (bitmap != 0) {
	    if ((bitmap & 1) != 0) {
374
#ifdef ENABLE_WIN32_DLL_SUPPORT
375 376 377 378 379
		
		if ((unsigned long)(*(info->next.srt.srt)) & 0x1)
		    c = (* (StgClosure **)((unsigned long)*(info->next.srt.srt)) & ~0x1);
		else
		    c = *(info->next.srt.srt);
380
#else
381
		c = *(info->next.srt.srt);
382
#endif
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
		bitmap = bitmap >> 1;
		info->next.srt.srt++;
		info->next.srt.srt_bitmap = bitmap;
		return c;
	    }
	    bitmap = bitmap >> 1;
	    info->next.srt.srt++;
	}
	// bitmap is now zero...
	return NULL;
    }
    else {
	// Large SRT bitmap
	nat i = info->next.large_srt.offset;
	StgWord bitmap;

	// Follow the pattern from GC.c:scavenge_large_srt_bitmap().
	bitmap = info->next.large_srt.srt->l.bitmap[i / BITS_IN(W_)];
	bitmap = bitmap >> (i % BITS_IN(StgWord));
	while (i < info->next.large_srt.srt->l.size) {
	    if ((bitmap & 1) != 0) {
		c = ((StgClosure **)info->next.large_srt.srt->srt)[i];
		i++;
		info->next.large_srt.offset = i;
		return c;
	    }
	    i++;
	    if (i % BITS_IN(W_) == 0) {
		bitmap = info->next.large_srt.srt->l.bitmap[i / BITS_IN(W_)];
	    } else {
		bitmap = bitmap >> 1;
	    }
	}
	// reached the end of this bitmap.
	info->next.large_srt.offset = i;
418 419 420 421 422 423 424 425 426 427 428 429 430 431 432
	return NULL;
    }
}

/* -----------------------------------------------------------------------------
 *  push() pushes a stackElement representing the next child of *c
 *  onto the traverse stack. If *c has no child, *first_child is set
 *  to NULL and nothing is pushed onto the stack. If *c has only one
 *  child, *c_chlid is set to that child and nothing is pushed onto
 *  the stack.  If *c has more than two children, *first_child is set
 *  to the first child and a stackElement representing the second
 *  child is pushed onto the stack.

 *  Invariants:
 *     *c_child_r is the most recent retainer of *c's children.
433
 *     *c is not any of TSO, AP, PAP, AP_STACK, which means that
434 435 436
 *        there cannot be any stack objects.
 *  Note: SRTs are considered to  be children as well.
 * -------------------------------------------------------------------------- */
437
static INLINE void
438
push( StgClosure *c, retainer c_child_r, StgClosure **first_child )
439 440 441 442 443 444 445 446 447
{
    stackElement se;
    bdescr *nbd;      // Next Block Descriptor

#ifdef DEBUG_RETAINER
    // fprintf(stderr, "push(): stackTop = 0x%x, currentStackBoundary = 0x%x\n", stackTop, currentStackBoundary);
#endif

    ASSERT(get_itbl(c)->type != TSO);
448
    ASSERT(get_itbl(c)->type != AP_STACK);
449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547

    //
    // fill in se
    //

    se.c = c;
    se.c_child_r = c_child_r;

    // fill in se.info
    switch (get_itbl(c)->type) {
	// no child, no SRT
    case CONSTR_0_1:
    case CONSTR_0_2:
    case CAF_BLACKHOLE:
    case BLACKHOLE:
    case SE_BLACKHOLE:
    case SE_CAF_BLACKHOLE:
    case ARR_WORDS:
	*first_child = NULL;
	return;

	// one child (fixed), no SRT
    case MUT_VAR:
    case MUT_CONS:
	*first_child = ((StgMutVar *)c)->var;
	return;
    case BLACKHOLE_BQ:
	// blocking_queue must be TSO and the head of a linked list of TSOs.
	// Shoule it be a child? Seems to be yes.
	*first_child = (StgClosure *)((StgBlockingQueue *)c)->blocking_queue;
	return;
    case THUNK_SELECTOR:
	*first_child = ((StgSelector *)c)->selectee;
	return;
    case IND_PERM:
    case IND_OLDGEN_PERM:
    case IND_OLDGEN:
	*first_child = ((StgIndOldGen *)c)->indirectee;
	return;
    case CONSTR_1_0:
    case CONSTR_1_1:
	*first_child = c->payload[0];
	return;

	// For CONSTR_2_0 and MVAR, we use se.info.step to record the position
	// of the next child. We do not write a separate initialization code.
	// Also we do not have to initialize info.type;

	// two children (fixed), no SRT
	// need to push a stackElement, but nothing to store in se.info
    case CONSTR_2_0:
	*first_child = c->payload[0];         // return the first pointer
	// se.info.type = posTypeStep;
	// se.info.next.step = 2;            // 2 = second
	break;

	// three children (fixed), no SRT
	// need to push a stackElement
    case MVAR:
	// head must be TSO and the head of a linked list of TSOs.
	// Shoule it be a child? Seems to be yes.
	*first_child = (StgClosure *)((StgMVar *)c)->head;
	// se.info.type = posTypeStep;
	se.info.next.step = 2;            // 2 = second
	break;

	// three children (fixed), no SRT
    case WEAK:
	*first_child = ((StgWeak *)c)->key;
	// se.info.type = posTypeStep;
	se.info.next.step = 2;
	break;

	// layout.payload.ptrs, no SRT
    case CONSTR:
    case FOREIGN:
    case STABLE_NAME:
    case BCO:
    case CONSTR_STATIC:
	init_ptrs(&se.info, get_itbl(c)->layout.payload.ptrs,
		  (StgPtr)c->payload);
	*first_child = find_ptrs(&se.info);
	if (*first_child == NULL)
	    return;   // no child
	break;

	// StgMutArrPtr.ptrs, no SRT
    case MUT_ARR_PTRS:
    case MUT_ARR_PTRS_FROZEN:
	init_ptrs(&se.info, ((StgMutArrPtrs *)c)->ptrs,
		  (StgPtr)(((StgMutArrPtrs *)c)->payload));
	*first_child = find_ptrs(&se.info);
	if (*first_child == NULL)
	    return;
	break;

    // layout.payload.ptrs, SRT
    case FUN:           // *c is a heap object.
    case FUN_2_0:
548 549 550 551 552 553 554
	init_ptrs(&se.info, get_itbl(c)->layout.payload.ptrs, (StgPtr)c->payload);
	*first_child = find_ptrs(&se.info);
	if (*first_child == NULL)
	    // no child from ptrs, so check SRT
	    goto fun_srt_only;
	break;

555 556 557 558 559 560
    case THUNK:
    case THUNK_2_0:
	init_ptrs(&se.info, get_itbl(c)->layout.payload.ptrs, (StgPtr)c->payload);
	*first_child = find_ptrs(&se.info);
	if (*first_child == NULL)
	    // no child from ptrs, so check SRT
561
	    goto thunk_srt_only;
562 563 564 565 566
	break;

	// 1 fixed child, SRT
    case FUN_1_0:
    case FUN_1_1:
567 568 569 570 571
	*first_child = c->payload[0];
	ASSERT(*first_child != NULL);
	init_srt_fun(&se.info, get_fun_itbl(c));
	break;

572 573 574 575
    case THUNK_1_0:
    case THUNK_1_1:
	*first_child = c->payload[0];
	ASSERT(*first_child != NULL);
576
	init_srt_thunk(&se.info, get_thunk_itbl(c));
577 578 579
	break;

    case FUN_STATIC:      // *c is a heap object.
580
	ASSERT(get_itbl(c)->srt_bitmap != 0);
581 582
    case FUN_0_1:
    case FUN_0_2:
583 584 585 586 587 588 589 590 591
    fun_srt_only:
        init_srt_fun(&se.info, get_fun_itbl(c));
	*first_child = find_srt(&se.info);
	if (*first_child == NULL)
	    return;     // no child
	break;

    // SRT only
    case THUNK_STATIC:
592
	ASSERT(get_itbl(c)->srt_bitmap != 0);
593 594
    case THUNK_0_1:
    case THUNK_0_2:
595 596
    thunk_srt_only:
        init_srt_thunk(&se.info, get_thunk_itbl(c));
597 598 599 600 601 602 603
	*first_child = find_srt(&se.info);
	if (*first_child == NULL)
	    return;     // no child
	break;

	// cannot appear
    case PAP:
604 605
    case AP:
    case AP_STACK:
606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679
    case TSO:
    case IND_STATIC:
    case CONSTR_INTLIKE:
    case CONSTR_CHARLIKE:
    case CONSTR_NOCAF_STATIC:
	// stack objects
    case UPDATE_FRAME:
    case CATCH_FRAME:
    case STOP_FRAME:
    case RET_DYN:
    case RET_BCO:
    case RET_SMALL:
    case RET_VEC_SMALL:
    case RET_BIG:
    case RET_VEC_BIG:
	// invalid objects
    case IND:
    case BLOCKED_FETCH:
    case FETCH_ME:
    case FETCH_ME_BQ:
    case RBH:
    case REMOTE_REF:
    case EVACUATED:
    case INVALID_OBJECT:
    default:
	barf("Invalid object *c in push()");
	return;
    }

    if (stackTop - 1 < stackBottom) {
#ifdef DEBUG_RETAINER
	// fprintf(stderr, "push() to the next stack.\n");
#endif
	// currentStack->free is updated when the active stack is switched
	// to the next stack.
	currentStack->free = (StgPtr)stackTop;

	if (currentStack->link == NULL) {
	    nbd = allocGroup(BLOCKS_IN_STACK);
	    nbd->link = NULL;
	    nbd->u.back = currentStack;
	    currentStack->link = nbd;
	} else
	    nbd = currentStack->link;

	newStackBlock(nbd);
    }

    // adjust stackTop (acutal push)
    stackTop--;
    // If the size of stackElement was huge, we would better replace the
    // following statement by either a memcpy() call or a switch statement
    // on the type of the element. Currently, the size of stackElement is
    // small enough (5 words) that this direct assignment seems to be enough.
    *stackTop = se;

#ifdef DEBUG_RETAINER
    stackSize++;
    if (stackSize > maxStackSize) maxStackSize = stackSize;
    // ASSERT(stackSize >= 0);
    // fprintf(stderr, "stackSize = %d\n", stackSize);
#endif
}

/* -----------------------------------------------------------------------------
 *  popOff() and popOffReal(): Pop a stackElement off the traverse stack.
 *  Invariants:
 *    stackTop cannot be equal to stackLimit unless the whole stack is
 *    empty, in which case popOff() is not allowed.
 *  Note:
 *    You can think of popOffReal() as a part of popOff() which is
 *    executed at the end of popOff() in necessary. Since popOff() is
 *    likely to be executed quite often while popOffReal() is not, we
 *    separate popOffReal() from popOff(), which is declared as an
680
 *    INLINE function (for the sake of execution speed).  popOffReal()
681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729
 *    is called only within popOff() and nowhere else.
 * -------------------------------------------------------------------------- */
static void
popOffReal(void)
{
    bdescr *pbd;    // Previous Block Descriptor

#ifdef DEBUG_RETAINER
    // fprintf(stderr, "pop() to the previous stack.\n");
#endif

    ASSERT(stackTop + 1 == stackLimit);
    ASSERT(stackBottom == (stackElement *)currentStack->start);

    if (firstStack == currentStack) {
	// The stack is completely empty.
	stackTop++;
	ASSERT(stackTop == stackLimit);
#ifdef DEBUG_RETAINER
	stackSize--;
	if (stackSize > maxStackSize) maxStackSize = stackSize;
	/*
	  ASSERT(stackSize >= 0);
	  fprintf(stderr, "stackSize = %d\n", stackSize);
	*/
#endif
	return;
    }

    // currentStack->free is updated when the active stack is switched back
    // to the previous stack.
    currentStack->free = (StgPtr)stackLimit;

    // find the previous block descriptor
    pbd = currentStack->u.back;
    ASSERT(pbd != NULL);

    returnToOldStack(pbd);

#ifdef DEBUG_RETAINER
    stackSize--;
    if (stackSize > maxStackSize) maxStackSize = stackSize;
    /*
      ASSERT(stackSize >= 0);
      fprintf(stderr, "stackSize = %d\n", stackSize);
    */
#endif
}

730
static INLINE void
731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771
popOff(void) {
#ifdef DEBUG_RETAINER
    // fprintf(stderr, "\tpopOff(): stackTop = 0x%x, currentStackBoundary = 0x%x\n", stackTop, currentStackBoundary);
#endif

    ASSERT(stackTop != stackLimit);
    ASSERT(!isEmptyRetainerStack());

    // <= (instead of <) is wrong!
    if (stackTop + 1 < stackLimit) {
	stackTop++;
#ifdef DEBUG_RETAINER
	stackSize--;
	if (stackSize > maxStackSize) maxStackSize = stackSize;
	/*
	  ASSERT(stackSize >= 0);
	  fprintf(stderr, "stackSize = %d\n", stackSize);
	*/
#endif
	return;
    }

    popOffReal();
}

/* -----------------------------------------------------------------------------
 *  Finds the next object to be considered for retainer profiling and store
 *  its pointer to *c.
 *  Test if the topmost stack element indicates that more objects are left,
 *  and if so, retrieve the first object and store its pointer to *c. Also,
 *  set *cp and *r appropriately, both of which are stored in the stack element.
 *  The topmost stack element then is overwritten so as for it to now denote
 *  the next object.
 *  If the topmost stack element indicates no more objects are left, pop
 *  off the stack element until either an object can be retrieved or
 *  the current stack chunk becomes empty, indicated by rtsTrue returned by
 *  isOnBoundary(), in which case *c is set to NULL.
 *  Note:
 *    It is okay to call this function even when the current stack chunk
 *    is empty.
 * -------------------------------------------------------------------------- */
772
static INLINE void
773
pop( StgClosure **c, StgClosure **cp, retainer *r )
774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847
{
    stackElement *se;

#ifdef DEBUG_RETAINER
    // fprintf(stderr, "pop(): stackTop = 0x%x, currentStackBoundary = 0x%x\n", stackTop, currentStackBoundary);
#endif

    do {
	if (isOnBoundary()) {     // if the current stack chunk is depleted
	    *c = NULL;
	    return;
	}

	se = stackTop;

	switch (get_itbl(se->c)->type) {
	    // two children (fixed), no SRT
	    // nothing in se.info
	case CONSTR_2_0:
	    *c = se->c->payload[1];
	    *cp = se->c;
	    *r = se->c_child_r;
	    popOff();
	    return;

	    // three children (fixed), no SRT
	    // need to push a stackElement
	case MVAR:
	    if (se->info.next.step == 2) {
		*c = (StgClosure *)((StgMVar *)se->c)->tail;
		se->info.next.step++;             // move to the next step
		// no popOff
	    } else {
		*c = ((StgMVar *)se->c)->value;
		popOff();
	    }
	    *cp = se->c;
	    *r = se->c_child_r;
	    return;

	    // three children (fixed), no SRT
	case WEAK:
	    if (se->info.next.step == 2) {
		*c = ((StgWeak *)se->c)->value;
		se->info.next.step++;
		// no popOff
	    } else {
		*c = ((StgWeak *)se->c)->finalizer;
		popOff();
	    }
	    *cp = se->c;
	    *r = se->c_child_r;
	    return;

	case CONSTR:
	case FOREIGN:
	case STABLE_NAME:
	case BCO:
	case CONSTR_STATIC:
	    // StgMutArrPtr.ptrs, no SRT
	case MUT_ARR_PTRS:
	case MUT_ARR_PTRS_FROZEN:
	    *c = find_ptrs(&se->info);
	    if (*c == NULL) {
		popOff();
		break;
	    }
	    *cp = se->c;
	    *r = se->c_child_r;
	    return;

	    // layout.payload.ptrs, SRT
	case FUN:         // always a heap object
	case FUN_2_0:
848 849 850 851 852 853 854 855 856 857 858
	    if (se->info.type == posTypePtrs) {
		*c = find_ptrs(&se->info);
		if (*c != NULL) {
		    *cp = se->c;
		    *r = se->c_child_r;
		    return;
		}
		init_srt_fun(&se->info, get_fun_itbl(se->c));
	    }
	    goto do_srt;

859 860 861 862 863 864 865 866 867
	case THUNK:
	case THUNK_2_0:
	    if (se->info.type == posTypePtrs) {
		*c = find_ptrs(&se->info);
		if (*c != NULL) {
		    *cp = se->c;
		    *r = se->c_child_r;
		    return;
		}
868
		init_srt_thunk(&se->info, get_thunk_itbl(se->c));
869
	    }
870
	    goto do_srt;
871 872

	    // SRT
873
	do_srt:
874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911
	case THUNK_STATIC:
	case FUN_STATIC:
	case FUN_0_1:
	case FUN_0_2:
	case THUNK_0_1:
	case THUNK_0_2:
	case FUN_1_0:
	case FUN_1_1:
	case THUNK_1_0:
	case THUNK_1_1:
	    *c = find_srt(&se->info);
	    if (*c != NULL) {
		*cp = se->c;
		*r = se->c_child_r;
		return;
	    }
	    popOff();
	    break;

	    // no child (fixed), no SRT
	case CONSTR_0_1:
	case CONSTR_0_2:
	case CAF_BLACKHOLE:
	case BLACKHOLE:
	case SE_BLACKHOLE:
	case SE_CAF_BLACKHOLE:
	case ARR_WORDS:
	    // one child (fixed), no SRT
	case MUT_VAR:
	case MUT_CONS:
	case BLACKHOLE_BQ:
	case THUNK_SELECTOR:
	case IND_PERM:
	case IND_OLDGEN_PERM:
	case IND_OLDGEN:
	case CONSTR_1_1:
	    // cannot appear
	case PAP:
912 913
	case AP:
	case AP_STACK:
914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980
	case TSO:
	case IND_STATIC:
	case CONSTR_INTLIKE:
	case CONSTR_CHARLIKE:
	case CONSTR_NOCAF_STATIC:
	    // stack objects
	case RET_DYN:
	case UPDATE_FRAME:
	case CATCH_FRAME:
	case STOP_FRAME:
	case RET_BCO:
	case RET_SMALL:
	case RET_VEC_SMALL:
	case RET_BIG:
	case RET_VEC_BIG:
	    // invalid objects
	case IND:
	case BLOCKED_FETCH:
	case FETCH_ME:
	case FETCH_ME_BQ:
	case RBH:
	case REMOTE_REF:
	case EVACUATED:
	case INVALID_OBJECT:
	default:
	    barf("Invalid object *c in pop()");
	    return;
	}
    } while (rtsTrue);
}

/* -----------------------------------------------------------------------------
 * RETAINER PROFILING ENGINE
 * -------------------------------------------------------------------------- */

void
initRetainerProfiling( void )
{
    initializeAllRetainerSet();
    retainerGeneration = 0;
}

/* -----------------------------------------------------------------------------
 *  This function must be called before f-closing prof_file.
 * -------------------------------------------------------------------------- */
void
endRetainerProfiling( void )
{
#ifdef SECOND_APPROACH
    outputAllRetainerSet(prof_file);
#endif
}

/* -----------------------------------------------------------------------------
 *  Returns the actual pointer to the retainer set of the closure *c.
 *  It may adjust RSET(c) subject to flip.
 *  Side effects:
 *    RSET(c) is initialized to NULL if its current value does not
 *    conform to flip.
 *  Note:
 *    Even though this function has side effects, they CAN be ignored because
 *    subsequent calls to retainerSetOf() always result in the same return value
 *    and retainerSetOf() is the only way to retrieve retainerSet of a given
 *    closure.
 *    We have to perform an XOR (^) operation each time a closure is examined.
 *    The reason is that we do not know when a closure is visited last.
 * -------------------------------------------------------------------------- */
981
static INLINE void
982 983 984 985 986 987 988 989 990 991
maybeInitRetainerSet( StgClosure *c )
{
    if (!isRetainerSetFieldValid(c)) {
	setRetainerSetToNull(c);
    }
}

/* -----------------------------------------------------------------------------
 * Returns rtsTrue if *c is a retainer.
 * -------------------------------------------------------------------------- */
992
static INLINE rtsBool
993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016
isRetainer( StgClosure *c )
{
    switch (get_itbl(c)->type) {
	//
	//  True case
	//
	// TSOs MUST be retainers: they constitute the set of roots.
    case TSO:

	// mutable objects
    case MVAR:
    case MUT_VAR:
    case MUT_CONS:
    case MUT_ARR_PTRS:
    case MUT_ARR_PTRS_FROZEN:

	// thunks are retainers.
    case THUNK:
    case THUNK_1_0:
    case THUNK_0_1:
    case THUNK_2_0:
    case THUNK_1_1:
    case THUNK_0_2:
    case THUNK_SELECTOR:
1017 1018
    case AP:
    case AP_STACK:
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 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112

	// Static thunks, or CAFS, are obviously retainers.
    case THUNK_STATIC:

	// WEAK objects are roots; there is separate code in which traversing
	// begins from WEAK objects.
    case WEAK:
	return rtsTrue;

	//
	// False case
	//

	// constructors
    case CONSTR:
    case CONSTR_1_0:
    case CONSTR_0_1:
    case CONSTR_2_0:
    case CONSTR_1_1:
    case CONSTR_0_2:
	// functions
    case FUN:
    case FUN_1_0:
    case FUN_0_1:
    case FUN_2_0:
    case FUN_1_1:
    case FUN_0_2:
	// partial applications
    case PAP:
	// blackholes
    case CAF_BLACKHOLE:
    case BLACKHOLE:
    case SE_BLACKHOLE:
    case SE_CAF_BLACKHOLE:
    case BLACKHOLE_BQ:
	// indirection
    case IND_PERM:
    case IND_OLDGEN_PERM:
    case IND_OLDGEN:
	// static objects
    case CONSTR_STATIC:
    case FUN_STATIC:
	// misc
    case FOREIGN:
    case STABLE_NAME:
    case BCO:
    case ARR_WORDS:
	return rtsFalse;

	//
	// Error case
	//
	// IND_STATIC cannot be *c, *cp, *r in the retainer profiling loop.
    case IND_STATIC:
	// CONSTR_INTLIKE, CONSTR_CHARLIKE, and CONSTR_NOCAF_STATIC
	// cannot be *c, *cp, *r in the retainer profiling loop.
    case CONSTR_INTLIKE:
    case CONSTR_CHARLIKE:
    case CONSTR_NOCAF_STATIC:
	// Stack objects are invalid because they are never treated as
	// legal objects during retainer profiling.
    case UPDATE_FRAME:
    case CATCH_FRAME:
    case STOP_FRAME:
    case RET_DYN:
    case RET_BCO:
    case RET_SMALL:
    case RET_VEC_SMALL:
    case RET_BIG:
    case RET_VEC_BIG:
	// other cases
    case IND:
    case BLOCKED_FETCH:
    case FETCH_ME:
    case FETCH_ME_BQ:
    case RBH:
    case REMOTE_REF:
    case EVACUATED:
    case INVALID_OBJECT:
    default:
	barf("Invalid object in isRetainer(): %d", get_itbl(c)->type);
	return rtsFalse;
    }
}

/* -----------------------------------------------------------------------------
 *  Returns the retainer function value for the closure *c, i.e., R(*c).
 *  This function does NOT return the retainer(s) of *c.
 *  Invariants:
 *    *c must be a retainer.
 *  Note:
 *    Depending on the definition of this function, the maintenance of retainer
 *    sets can be made easier. If most retainer sets are likely to be created
 *    again across garbage collections, refreshAllRetainerSet() in
1113
 *    RetainerSet.c can simply do nothing.
1114 1115 1116 1117
 *    If this is not the case, we can free all the retainer sets and
 *    re-initialize the hash table.
 *    See refreshAllRetainerSet() in RetainerSet.c.
 * -------------------------------------------------------------------------- */
1118
static INLINE retainer
1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141
getRetainerFrom( StgClosure *c )
{
    ASSERT(isRetainer(c));

#if defined(RETAINER_SCHEME_INFO)
    // Retainer scheme 1: retainer = info table
    return get_itbl(c);
#elif defined(RETAINER_SCHEME_CCS)
    // Retainer scheme 2: retainer = cost centre stack
    return c->header.prof.ccs;
#elif defined(RETAINER_SCHEME_CC)
    // Retainer scheme 3: retainer = cost centre
    return c->header.prof.ccs->cc;
#endif
}

/* -----------------------------------------------------------------------------
 *  Associates the retainer set *s with the closure *c, that is, *s becomes
 *  the retainer set of *c.
 *  Invariants:
 *    c != NULL
 *    s != NULL
 * -------------------------------------------------------------------------- */
1142
static INLINE void
1143
associate( StgClosure *c, RetainerSet *s )
1144 1145 1146 1147 1148 1149
{
    // StgWord has the same size as pointers, so the following type
    // casting is okay.
    RSET(c) = (RetainerSet *)((StgWord)s | flip);
}

1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177
/* -----------------------------------------------------------------------------
   Call retainClosure for each of the closures covered by a large bitmap.
   -------------------------------------------------------------------------- */

static void
retain_large_bitmap (StgPtr p, StgLargeBitmap *large_bitmap, nat size,
		     StgClosure *c, retainer c_child_r)
{
    nat i, b;
    StgWord bitmap;
    
    b = 0;
    bitmap = large_bitmap->bitmap[b];
    for (i = 0; i < size; ) {
	if ((bitmap & 1) == 0) {
	    retainClosure((StgClosure *)*p, c, c_child_r);
	}
	i++;
	p++;
	if (i % BITS_IN(W_) == 0) {
	    b++;
	    bitmap = large_bitmap->bitmap[b];
	} else {
	    bitmap = bitmap >> 1;
	}
    }
}

1178
static INLINE StgPtr
1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192
retain_small_bitmap (StgPtr p, nat size, StgWord bitmap,
		     StgClosure *c, retainer c_child_r)
{
    while (size > 0) {
	if ((bitmap & 1) == 0) {
	    retainClosure((StgClosure *)*p, c, c_child_r);
	}
	p++;
	bitmap = bitmap >> 1;
	size--;
    }
    return p;
}

1193 1194 1195 1196 1197 1198 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
/* -----------------------------------------------------------------------------
 * Call retainClosure for each of the closures in an SRT.
 * ------------------------------------------------------------------------- */

static void
retain_large_srt_bitmap (StgLargeSRT *srt, StgClosure *c, retainer c_child_r)
{
    nat i, b, size;
    StgWord bitmap;
    StgClosure **p;
    
    b = 0;
    p = (StgClosure **)srt->srt;
    size   = srt->l.size;
    bitmap = srt->l.bitmap[b];
    for (i = 0; i < size; ) {
	if ((bitmap & 1) != 0) {
	    retainClosure((StgClosure *)*p, c, c_child_r);
	}
	i++;
	p++;
	if (i % BITS_IN(W_) == 0) {
	    b++;
	    bitmap = srt->l.bitmap[b];
	} else {
	    bitmap = bitmap >> 1;
	}
    }
}

static INLINE void
retainSRT (StgClosure **srt, nat srt_bitmap, StgClosure *c, retainer c_child_r)
{
  nat bitmap;
  StgClosure **p;

  bitmap = srt_bitmap;
  p = srt;

  if (bitmap == (StgHalfWord)(-1)) {  
      retain_large_srt_bitmap( (StgLargeSRT *)srt, c, c_child_r );
      return;
  }

  while (bitmap != 0) {
      if ((bitmap & 1) != 0) {
#ifdef ENABLE_WIN32_DLL_SUPPORT
	  if ( (unsigned long)(*srt) & 0x1 ) {
	      retainClosure(*stgCast(StgClosure**,(stgCast(unsigned long, *srt) & ~0x1)), 
			    c, c_child_r);
	  } else {
	      retainClosure(*srt,c,c_child_r);
	  }
#else
	  retainClosure(*srt,c,c_child_r);
#endif
      }
      p++;
      bitmap = bitmap >> 1;
  }
}

1255 1256 1257 1258 1259 1260
/* -----------------------------------------------------------------------------
 *  Process all the objects in the stack chunk from stackStart to stackEnd
 *  with *c and *c_child_r being their parent and their most recent retainer,
 *  respectively. Treat stackOptionalFun as another child of *c if it is
 *  not NULL.
 *  Invariants:
1261
 *    *c is one of the following: TSO, AP_STACK.
1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273
 *    If *c is TSO, c == c_child_r.
 *    stackStart < stackEnd.
 *    RSET(c) and RSET(c_child_r) are valid, i.e., their
 *    interpretation conforms to the current value of flip (even when they
 *    are interpreted to be NULL).
 *    If *c is TSO, its state is not any of ThreadRelocated, ThreadComplete,
 *    or ThreadKilled, which means that its stack is ready to process.
 *  Note:
 *    This code was almost plagiarzied from GC.c! For each pointer,
 *    retainClosure() is invoked instead of evacuate().
 * -------------------------------------------------------------------------- */
static void
1274
retainStack( StgClosure *c, retainer c_child_r,
1275
	     StgPtr stackStart, StgPtr stackEnd )
1276 1277
{
    stackElement *oldStackBoundary;
1278 1279
    StgPtr p;
    StgRetInfoTable *info;
1280
    StgWord32 bitmap;
1281
    nat size;
1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300

#ifdef DEBUG_RETAINER
    cStackSize++;
    if (cStackSize > maxCStackSize) maxCStackSize = cStackSize;
#endif

    /*
      Each invocation of retainStack() creates a new virtual
      stack. Since all such stacks share a single common stack, we
      record the current currentStackBoundary, which will be restored
      at the exit.
    */
    oldStackBoundary = currentStackBoundary;
    currentStackBoundary = stackTop;

#ifdef DEBUG_RETAINER
    // fprintf(stderr, "retainStack() called: oldStackBoundary = 0x%x, currentStackBoundary = 0x%x\n", oldStackBoundary, currentStackBoundary);
#endif

1301 1302 1303 1304 1305
    ASSERT(get_itbl(c)->type != TSO || 
	   (((StgTSO *)c)->what_next != ThreadRelocated &&
	    ((StgTSO *)c)->what_next != ThreadComplete &&
	    ((StgTSO *)c)->what_next != ThreadKilled));
    
1306 1307
    p = stackStart;
    while (p < stackEnd) {
1308
	info = get_ret_itbl((StgClosure *)p);
1309

1310
	switch(info->i.type) {
1311 1312 1313 1314 1315 1316 1317 1318 1319 1320

	case UPDATE_FRAME:
	    retainClosure(((StgUpdateFrame *)p)->updatee, c, c_child_r);
	    p += sizeofW(StgUpdateFrame);
	    continue;

	case STOP_FRAME:
	case CATCH_FRAME:
	case RET_SMALL:
	case RET_VEC_SMALL:
1321 1322
	    bitmap = BITMAP_BITS(info->i.layout.bitmap);
	    size   = BITMAP_SIZE(info->i.layout.bitmap);
1323
	    p++;
1324 1325
	    p = retain_small_bitmap(p, size, bitmap, c, c_child_r);

1326
	follow_srt:
1327
	    retainSRT((StgClosure **)info->srt, info->i.srt_bitmap, c, c_child_r);
1328
	    continue;
1329

1330 1331 1332 1333 1334 1335 1336 1337 1338 1339
	case RET_BCO: {
	    StgBCO *bco;
	    
	    p++;
	    retainClosure((StgClosure *)*p, c, c_child_r);
	    bco = (StgBCO *)*p;
	    p++;
	    size = BCO_BITMAP_SIZE(bco);
	    retain_large_bitmap(p, BCO_BITMAP(bco), size, c, c_child_r);
	    p += size;
1340
	    continue;
1341
	}
1342

1343
	    // large bitmap (> 32 entries, or > 64 on a 64-bit machine) 
1344 1345
	case RET_BIG:
	case RET_VEC_BIG:
1346
	    size = info->i.layout.large_bitmap->size;
1347
	    p++;
1348 1349 1350 1351 1352
	    retain_large_bitmap(p, info->i.layout.large_bitmap,
				size, c, c_child_r);
	    p += size;
	    // and don't forget to follow the SRT 
	    goto follow_srt;
1353

1354 1355 1356 1357 1358 1359 1360 1361
	    // Dynamic bitmap: the mask is stored on the stack 
	case RET_DYN: {
	    StgWord dyn;
	    dyn = ((StgRetDyn *)p)->liveness;

	    // traverse the bitmap first
	    bitmap = GET_LIVENESS(dyn);
	    p      = (P_)&((StgRetDyn *)p)->payload[0];
1362
	    size   = RET_DYN_BITMAP_SIZE;
1363 1364 1365
	    p = retain_small_bitmap(p, size, bitmap, c, c_child_r);
	    
	    // skip over the non-ptr words
1366
	    p += GET_NONPTRS(dyn) + RET_DYN_NONPTR_REGS_SIZE;
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
	    
	    // follow the ptr words
	    for (size = GET_PTRS(dyn); size > 0; size--) {
		retainClosure((StgClosure *)*p, c, c_child_r);
		p++;
	    }
	    continue;
	}

	case RET_FUN: {
	    StgRetFun *ret_fun = (StgRetFun *)p;
	    StgFunInfoTable *fun_info;
	    
	    retainClosure(ret_fun->fun, c, c_child_r);
	    fun_info = get_fun_itbl(ret_fun->fun);
	    
	    p = (P_)&ret_fun->payload;
	    switch (fun_info->fun_type) {
	    case ARG_GEN:
		bitmap = BITMAP_BITS(fun_info->bitmap);
		size = BITMAP_SIZE(fun_info->bitmap);
		p = retain_small_bitmap(p, size, bitmap, c, c_child_r);
		break;
	    case ARG_GEN_BIG:
		size = ((StgLargeBitmap *)fun_info->bitmap)->size;
		retain_large_bitmap(p, (StgLargeBitmap *)fun_info->bitmap, 
				    size, c, c_child_r);
		p += size;
		break;
	    default:
		bitmap = BITMAP_BITS(stg_arg_bitmaps[fun_info->fun_type]);
		size = BITMAP_SIZE(stg_arg_bitmaps[fun_info->fun_type]);
		p = retain_small_bitmap(p, size, bitmap, c, c_child_r);
		break;
1401
	    }
1402
	    goto follow_srt;
1403
	}
1404

1405 1406
	default:
	    barf("Invalid object found in retainStack(): %d",
1407
		 (int)(info->i.type));
1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421
	}
    }

    // restore currentStackBoundary
    currentStackBoundary = oldStackBoundary;
#ifdef DEBUG_RETAINER
    // fprintf(stderr, "retainStack() finished: currentStackBoundary = 0x%x\n", currentStackBoundary);
#endif

#ifdef DEBUG_RETAINER
    cStackSize--;
#endif
}

1422 1423 1424 1425
/* ----------------------------------------------------------------------------
 * Call retainClosure for each of the children of a PAP/AP
 * ------------------------------------------------------------------------- */

1426
static INLINE StgPtr
1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464
retain_PAP (StgPAP *pap, retainer c_child_r)
{
    StgPtr p;
    StgWord bitmap, size;
    StgFunInfoTable *fun_info;

    retainClosure(pap->fun, (StgClosure *)pap, c_child_r);
    fun_info = get_fun_itbl(pap->fun);
    ASSERT(fun_info->i.type != PAP);

    p = (StgPtr)pap->payload;
    size = pap->n_args;

    switch (fun_info->fun_type) {
    case ARG_GEN:
	bitmap = BITMAP_BITS(fun_info->bitmap);
	p = retain_small_bitmap(p, pap->n_args, bitmap, 
				(StgClosure *)pap, c_child_r);
	break;
    case ARG_GEN_BIG:
	retain_large_bitmap(p, (StgLargeBitmap *)fun_info->bitmap,
			    size, (StgClosure *)pap, c_child_r);
	p += size;
	break;
    case ARG_BCO:
	retain_large_bitmap((StgPtr)pap->payload, BCO_BITMAP(pap->fun),
			    size, (StgClosure *)pap, c_child_r);
	p += size;
	break;
    default:
	bitmap = BITMAP_BITS(stg_arg_bitmaps[fun_info->fun_type]);
	p = retain_small_bitmap(p, pap->n_args, bitmap, 
				(StgClosure *)pap, c_child_r);
	break;
    }
    return p;
}

1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478
/* -----------------------------------------------------------------------------
 *  Compute the retainer set of *c0 and all its desecents by traversing.
 *  *cp0 is the parent of *c0, and *r0 is the most recent retainer of *c0.
 *  Invariants:
 *    c0 = cp0 = r0 holds only for root objects.
 *    RSET(cp0) and RSET(r0) are valid, i.e., their
 *    interpretation conforms to the current value of flip (even when they
 *    are interpreted to be NULL).
 *    However, RSET(c0) may be corrupt, i.e., it may not conform to
 *    the current value of flip. If it does not, during the execution
 *    of this function, RSET(c0) must be initialized as well as all
 *    its descendants.
 *  Note:
 *    stackTop must be the same at the beginning and the exit of this function.
1479
 *    *c0 can be TSO (as well as AP_STACK).
1480 1481
 * -------------------------------------------------------------------------- */
static void
1482
retainClosure( StgClosure *c0, StgClosure *cp0, retainer r0 )
1483 1484 1485 1486 1487 1488
{
    // c = Current closure
    // cp = Current closure's Parent
    // r = current closures' most recent Retainer
    // c_child_r = current closure's children's most recent retainer
    // first_child = first child of c
1489
    StgClosure *c, *cp, *first_child;
1490
    RetainerSet *s, *retainerSetOfc;
1491
    retainer r, c_child_r;
1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590
    StgWord typeOfc;

#ifdef DEBUG_RETAINER
    // StgPtr oldStackTop;
#endif

#ifdef DEBUG_RETAINER
    // oldStackTop = stackTop;
    // fprintf(stderr, "retainClosure() called: c0 = 0x%x, cp0 = 0x%x, r0 = 0x%x\n", c0, cp0, r0);
#endif

    // (c, cp, r) = (c0, cp0, r0)
    c = c0;
    cp = cp0;
    r = r0;
    goto inner_loop;

loop:
    //fprintf(stderr, "loop");
    // pop to (c, cp, r);
    pop(&c, &cp, &r);

    if (c == NULL) {
#ifdef DEBUG_RETAINER
	// fprintf(stderr, "retainClosure() ends: oldStackTop = 0x%x, stackTop = 0x%x\n", oldStackTop, stackTop);
#endif
	return;
    }

    //fprintf(stderr, "inner_loop");

inner_loop:
    // c  = current closure under consideration,
    // cp = current closure's parent,
    // r  = current closure's most recent retainer
    //
    // Loop invariants (on the meaning of c, cp, r, and their retainer sets):
    //   RSET(cp) and RSET(r) are valid.
    //   RSET(c) is valid only if c has been visited before.
    //
    // Loop invariants (on the relation between c, cp, and r)
    //   if cp is not a retainer, r belongs to RSET(cp).
    //   if cp is a retainer, r == cp.

    typeOfc = get_itbl(c)->type;

#ifdef DEBUG_RETAINER
    switch (typeOfc) {
    case IND_STATIC:
    case CONSTR_INTLIKE:
    case CONSTR_CHARLIKE:
    case CONSTR_NOCAF_STATIC:
    case CONSTR_STATIC:
    case THUNK_STATIC:
    case FUN_STATIC:
	break;
    default:
	if (retainerSetOf(c) == NULL) {   // first visit?
	    costArray[typeOfc] += cost(c);
	    sumOfNewCost += cost(c);
	}
	break;
    }
#endif

    // special cases
    switch (typeOfc) {
    case TSO:
	if (((StgTSO *)c)->what_next == ThreadComplete ||
	    ((StgTSO *)c)->what_next == ThreadKilled) {
#ifdef DEBUG_RETAINER
	    fprintf(stderr, "ThreadComplete or ThreadKilled encountered in retainClosure()\n");
#endif
	    goto loop;
	}
	if (((StgTSO *)c)->what_next == ThreadRelocated) {
#ifdef DEBUG_RETAINER
	    fprintf(stderr, "ThreadRelocated encountered in retainClosure()\n");
#endif
	    c = (StgClosure *)((StgTSO *)c)->link;
	    goto inner_loop;
	}
	break;

    case IND_STATIC:
	// We just skip IND_STATIC, so its retainer set is never computed.
	c = ((StgIndStatic *)c)->indirectee;
	goto inner_loop;
    case CONSTR_INTLIKE:
    case CONSTR_CHARLIKE:
	// static objects with no pointers out, so goto loop.
    case CONSTR_NOCAF_STATIC:
	// It is not just enough not to compute the retainer set for *c; it is
	// mandatory because CONSTR_NOCAF_STATIC are not reachable from
	// scavenged_static_objects, the list from which is assumed to traverse
	// all static objects after major garbage collections.
	goto loop;
    case THUNK_STATIC:
    case FUN_STATIC:
1591
	if (get_itbl(c)->srt_bitmap == 0) {
1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645
	    // No need to compute the retainer set; no dynamic objects
	    // are reachable from *c.
	    //
	    // Static objects: if we traverse all the live closures,
	    // including static closures, during each heap census then
	    // we will observe that some static closures appear and
	    // disappear.  eg. a closure may contain a pointer to a
	    // static function 'f' which is not otherwise reachable
	    // (it doesn't indirectly point to any CAFs, so it doesn't
	    // appear in any SRTs), so we would find 'f' during
	    // traversal.  However on the next sweep there may be no
	    // closures pointing to 'f'.
	    //
	    // We must therefore ignore static closures whose SRT is
	    // empty, because these are exactly the closures that may
	    // "appear".  A closure with a non-empty SRT, and which is
	    // still required, will always be reachable.
	    //
	    // But what about CONSTR_STATIC?  Surely these may be able
	    // to appear, and they don't have SRTs, so we can't
	    // check.  So for now, we're calling
	    // resetStaticObjectForRetainerProfiling() from the
	    // garbage collector to reset the retainer sets in all the
	    // reachable static objects.
	    goto loop;
	}
    default:
	break;
    }

    // The above objects are ignored in computing the average number of times
    // an object is visited.
    timesAnyObjectVisited++;

    // If this is the first visit to c, initialize its retainer set.
    maybeInitRetainerSet(c);
    retainerSetOfc = retainerSetOf(c);

    // Now compute s:
    //    isRetainer(cp) == rtsTrue => s == NULL
    //    isRetainer(cp) == rtsFalse => s == cp.retainer
    if (isRetainer(cp))
	s = NULL;
    else
	s = retainerSetOf(cp);

    // (c, cp, r, s) is available.

    // (c, cp, r, s, R_r) is available, so compute the retainer set for *c.
    if (retainerSetOfc == NULL) {
	// This is the first visit to *c.
	numObjectVisited++;

	if (s == NULL)
1646
	    associate(c, singleton(r));
1647 1648
	else
	    // s is actually the retainer set of *c!
1649
	    associate(c, s);
1650 1651

	// compute c_child_r
1652
	c_child_r = isRetainer(c) ? getRetainerFrom(c) : r;
1653 1654
    } else {
	// This is not the first visit to *c.
1655
	if (isMember(r, retainerSetOfc))
1656 1657 1658
	    goto loop;          // no need to process child

	if (s == NULL)
1659
	    associate(c, addElement(r, retainerSetOfc));
1660 1661 1662 1663 1664 1665
	else {
	    // s is not NULL and cp is not a retainer. This means that
	    // each time *cp is visited, so is *c. Thus, if s has
	    // exactly one more element in its retainer set than c, s
	    // is also the new retainer set for *c.
	    if (s->num == retainerSetOfc->num + 1) {
1666
		associate(c, s);
1667 1668 1669
	    }
	    // Otherwise, just add R_r to the current retainer set of *c.
	    else {
1670
		associate(c, addElement(r, retainerSetOfc));
1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685
	    }
	}

	if (isRetainer(c))
	    goto loop;          // no need to process child

	// compute c_child_r
	c_child_r = r;
    }

    // now, RSET() of all of *c, *cp, and *r is valid.
    // (c, c_child_r) are available.

    // process child

1686 1687 1688 1689 1690
    // Special case closures: we process these all in one go rather
    // than attempting to save the current position, because doing so
    // would be hard.
    switch (typeOfc) {
    case TSO:
1691 1692 1693 1694
	retainStack(c, c_child_r,
		    ((StgTSO *)c)->sp,
		    ((StgTSO *)c)->stack + ((StgTSO *)c)->stack_size);
	goto loop;
1695 1696 1697 1698

    case PAP:
    case AP:
	retain_PAP((StgPAP *)c, c_child_r);
1699
	goto loop;
1700 1701 1702

    case AP_STACK:
	retainClosure(((StgAP_STACK *)c)->fun, c, c_child_r);
1703
	retainStack(c, c_child_r,
1704 1705 1706
		    (StgPtr)((StgAP_STACK *)c)->payload,
		    (StgPtr)((StgAP_STACK *)c)->payload +
		             ((StgAP_STACK *)c)->size);
1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736
	goto loop;
    }

    push(c, c_child_r, &first_child);

    // If first_child is null, c has no child.
    // If first_child is not null, the top stack element points to the next
    // object. push() may or may not push a stackElement on the stack.
    if (first_child == NULL)
	goto loop;

    // (c, cp, r) = (first_child, c, c_child_r)
    r = c_child_r;
    cp = c;
    c = first_child;
    goto inner_loop;
}

/* -----------------------------------------------------------------------------
 *  Compute the retainer set for every object reachable from *tl.
 * -------------------------------------------------------------------------- */
static void
retainRoot( StgClosure **tl )
{
    // We no longer assume that only TSOs and WEAKs are roots; any closure can
    // be a root.

    ASSERT(isEmptyRetainerStack());
    currentStackBoundary = stackTop;

1737 1738 1739 1740 1741
    if (isRetainer(*tl)) {
	retainClosure(*tl, *tl, getRetainerFrom(*tl));
    } else {
	retainClosure(*tl, *tl, CCS_SYSTEM);
    }
1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772

    // NOT TRUE: ASSERT(isMember(getRetainerFrom(*tl), retainerSetOf(*tl)));
    // *tl might be a TSO which is ThreadComplete, in which
    // case we ignore it for the purposes of retainer profiling.
}

/* -----------------------------------------------------------------------------
 *  Compute the retainer set for each of the objects in the heap.
 * -------------------------------------------------------------------------- */
static void
computeRetainerSet( void )
{
    StgWeak *weak;
    RetainerSet *rtl;
    nat g;
    StgMutClosure *ml;
#ifdef DEBUG_RETAINER
    RetainerSet tmpRetainerSet;
#endif

    GetRoots(retainRoot);	// for scheduler roots

    // This function is called after a major GC, when key, value, and finalizer
    // all are guaranteed to be valid, or reachable.
    //
    // The following code assumes that WEAK objects are considered to be roots
    // for retainer profilng.
    for (weak = weak_ptr_list; weak != NULL; weak = weak->link)
	// retainRoot((StgClosure *)weak);
	retainRoot((StgClosure **)&weak);

1773 1774 1775
    // Consider roots from the stable ptr table.
    markStablePtrTable(retainRoot);

1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061
    // The following code resets the rs field of each unvisited mutable
    // object (computing sumOfNewCostExtra and updating costArray[] when
    // debugging retainer profiler).
    for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
	ASSERT(g != 0 ||
	       (generations[g].mut_list == END_MUT_LIST &&
		generations[g].mut_once_list == END_MUT_LIST));

	// Todo:
	// I think traversing through mut_list is unnecessary.
	// Think about removing this part.
	for (ml = generations[g].mut_list; ml != END_MUT_LIST;
	     ml = ml->mut_link) {

	    maybeInitRetainerSet((StgClosure *)ml);
	    rtl = retainerSetOf((StgClosure *)ml);

#ifdef DEBUG_RETAINER
	    if (rtl == NULL) {
		// first visit to *ml
		// This is a violation of the interface rule!
		RSET(ml) = (RetainerSet *)((StgWord)(&tmpRetainerSet) | flip);

		switch (get_itbl((StgClosure *)ml)->type) {
		case IND_STATIC:
		    // no cost involved
		    break;
		case CONSTR_INTLIKE:
		case CONSTR_CHARLIKE:
		case CONSTR_NOCAF_STATIC:
		case CONSTR_STATIC:
		case THUNK_STATIC:
		case FUN_STATIC:
		    barf("Invalid object in computeRetainerSet(): %d", get_itbl((StgClosure*)ml)->type);
		    break;
		default:
		    // dynamic objects
		    costArray[get_itbl((StgClosure *)ml)->type] += cost((StgClosure *)ml);
		    sumOfNewCostExtra += cost((StgClosure *)ml);
		    break;
		}
	    }
#endif
	}

	// Traversing through mut_once_list is, in contrast, necessary
	// because we can find MUT_VAR objects which have not been
	// visited during retainer profiling.
	for (ml = generations[g].mut_once_list; ml != END_MUT_LIST;
	     ml = ml->mut_link) {

	    maybeInitRetainerSet((StgClosure *)ml);
	    rtl = retainerSetOf((StgClosure *)ml);
#ifdef DEBUG_RETAINER
	    if (rtl == NULL) {
		// first visit to *ml
		// This is a violation of the interface rule!
		RSET(ml) = (RetainerSet *)((StgWord)(&tmpRetainerSet) | flip);

		switch (get_itbl((StgClosure *)ml)->type) {
		case IND_STATIC:
		    // no cost involved
		    break;
		case CONSTR_INTLIKE:
		case CONSTR_CHARLIKE:
		case CONSTR_NOCAF_STATIC:
		case CONSTR_STATIC:
		case THUNK_STATIC:
		case FUN_STATIC:
		    barf("Invalid object in computeRetainerSet(): %d", get_itbl((StgClosure*)ml)->type);
		    break;
		default:
		    // dynamic objects
		    costArray[get_itbl((StgClosure *)ml)->type] += cost((StgClosure *)ml);
		    sumOfNewCostExtra += cost((StgClosure *)ml);
		    break;
		}
	    }
#endif
	}
    }
}

/* -----------------------------------------------------------------------------
 *  Traverse all static objects for which we compute retainer sets,
 *  and reset their rs fields to NULL, which is accomplished by
 *  invoking maybeInitRetainerSet(). This function must be called
 *  before zeroing all objects reachable from scavenged_static_objects
 *  in the case of major gabage collections. See GarbageCollect() in
 *  GC.c.
 *  Note:
 *    The mut_once_list of the oldest generation must also be traversed?
 *    Why? Because if the evacuation of an object pointed to by a static
 *    indirection object fails, it is put back to the mut_once_list of
 *    the oldest generation.
 *    However, this is not necessary because any static indirection objects
 *    are just traversed through to reach dynamic objects. In other words,
 *    they are not taken into consideration in computing retainer sets.
 * -------------------------------------------------------------------------- */
void
resetStaticObjectForRetainerProfiling( void )
{
#ifdef DEBUG_RETAINER
    nat count;
#endif
    StgClosure *p;

#ifdef DEBUG_RETAINER
    count = 0;
#endif
    p = scavenged_static_objects;
    while (p != END_OF_STATIC_LIST) {
#ifdef DEBUG_RETAINER
	count++;
#endif
	switch (get_itbl(p)->type) {
	case IND_STATIC:
	    // Since we do not compute the retainer set of any
	    // IND_STATIC object, we don't have to reset its retainer
	    // field.
	    p = IND_STATIC_LINK(p);
	    break;
	case THUNK_STATIC:
	    maybeInitRetainerSet(p);
	    p = THUNK_STATIC_LINK(p);
	    break;
	case FUN_STATIC:
	    maybeInitRetainerSet(p);
	    p = FUN_STATIC_LINK(p);
	    break;
	case CONSTR_STATIC:
	    maybeInitRetainerSet(p);
	    p = STATIC_LINK(get_itbl(p), p);
	    break;
	default:
	    barf("resetStaticObjectForRetainerProfiling: %p (%s)",
		 p, get_itbl(p)->type);
	    break;
	}
    }
#ifdef DEBUG_RETAINER
    // fprintf(stderr, "count in scavenged_static_objects = %d\n", count);
#endif
}

/* -----------------------------------------------------------------------------
 * Perform retainer profiling.
 * N is the oldest generation being profilied, where the generations are
 * numbered starting at 0.
 * Invariants:
 * Note:
 *   This function should be called only immediately after major garbage
 *   collection.
 * ------------------------------------------------------------------------- */
void
retainerProfile(void)
{
#ifdef DEBUG_RETAINER
  nat i;
  nat totalHeapSize;        // total raw heap size (computed by linear scanning)
#endif

#ifdef DEBUG_RETAINER
  fprintf(stderr, " < retainerProfile() invoked : %d>\n", retainerGeneration);
#endif

  stat_startRP();

  // We haven't flipped the bit yet.
#ifdef DEBUG_RETAINER
  fprintf(stderr, "Before traversing:\n");
  sumOfCostLinear = 0;
  for (i = 0;i < N_CLOSURE_TYPES; i++)
    costArrayLinear[i] = 0;
  totalHeapSize = checkHeapSanityForRetainerProfiling();

  fprintf(stderr, "\tsumOfCostLinear = %d, totalHeapSize = %d\n", sumOfCostLinear, totalHeapSize);
  /*
  fprintf(stderr, "costArrayLinear[] = ");
  for (i = 0;i < N_CLOSURE_TYPES; i++)
    fprintf(stderr, "[%u:%u] ", i, costArrayLinear[i]);
  fprintf(stderr, "\n");
  */

  ASSERT(sumOfCostLinear == totalHeapSize);

/*
#define pcostArrayLinear(index) \
  if (costArrayLinear[index] > 0) \
    fprintf(stderr, "costArrayLinear[" #index "] = %u\n", costArrayLinear[index])
  pcostArrayLinear(THUNK_STATIC);
  pcostArrayLinear(FUN_STATIC);
  pcostArrayLinear(CONSTR_STATIC);
  pcostArrayLinear(CONSTR_NOCAF_STATIC);
  pcostArrayLinear(CONSTR_INTLIKE);
  pcostArrayLinear(CONSTR_CHARLIKE);
*/
#endif

  // Now we flips flip.
  flip = flip ^ 1;

#ifdef DEBUG_RETAINER
  stackSize = 0;
  maxStackSize = 0;
  cStackSize = 0;
  maxCStackSize = 0;
#endif
  numObjectVisited = 0;
  timesAnyObjectVisited = 0;

#ifdef DEBUG_RETAINER
  fprintf(stderr, "During traversing:\n");
  sumOfNewCost = 0;
  sumOfNewCostExtra = 0;
  for (i = 0;i < N_CLOSURE_TYPES; i++)
    costArray[i] = 0;
#endif

  /*
    We initialize the traverse stack each time the retainer profiling is
    performed (because the traverse stack size varies on each retainer profiling
    and this operation is not costly anyhow). However, we just refresh the
    retainer sets.
   */
  initializeTraverseStack();
#ifdef DEBUG_RETAINER
  initializeAllRetainerSet();
#else
  refreshAllRetainerSet();
#endif
  computeRetainerSet();

#ifdef DEBUG_RETAINER
  fprintf(stderr, "After traversing:\n");
  sumOfCostLinear = 0;
  for (i = 0;i < N_CLOSURE_TYPES; i++)
    costArrayLinear[i] = 0;
  totalHeapSize = checkHeapSanityForRetainerProfiling();

  fprintf(stderr, "\tsumOfCostLinear = %d, totalHeapSize = %d\n", sumOfCostLinear, totalHeapSize);
  ASSERT(sumOfCostLinear == totalHeapSize);

  // now, compare the two results
  /*
    Note:
      costArray[] must be exactly the same as costArrayLinear[].
      Known exceptions:
        1) Dead weak pointers, whose type is CONSTR. These objects are not
           reachable from any roots.
  */
  fprintf(stderr, "Comparison:\n");
  fprintf(stderr, "\tcostArrayLinear[] (must be empty) = ");
  for (i = 0;i < N_CLOSURE_TYPES; i++)
    if (costArray[i] != costArrayLinear[i])
      // nothing should be printed except MUT_VAR after major GCs
      fprintf(stderr, "[%u:%u] ", i, costArrayLinear[i]);
  fprintf(stderr, "\n");

  fprintf(stderr, "\tsumOfNewCost = %u\n", sumOfNewCost);
  fprintf(stderr, "\tsumOfNewCostExtra = %u\n", sumOfNewCostExtra);
  fprintf(stderr, "\tcostArray[] (must be empty) = ");
  for (i = 0;i < N_CLOSURE_TYPES; i++)
    if (costArray[i] != costArrayLinear[i])
      // nothing should be printed except MUT_VAR after major GCs
      fprintf(stderr, "[%u:%u] ", i, costArray[i]);
  fprintf(stderr, "\n");

  // only for major garbage collection
  ASSERT(sumOfNewCost + sumOfNewCostExtra == sumOfCostLinear);
#endif

  // post-processing
  closeTraverseStack();
#ifdef DEBUG_RETAINER
  closeAllRetainerSet();
#else
  // Note that there is no post-processing for the retainer sets.
#endif
  retainerGeneration++;

  stat_endRP(
    retainerGeneration - 1,   // retainerGeneration has just been incremented!
#ifdef DEBUG_RETAINER
    maxCStackSize, maxStackSize,
#endif
2062
    (double)timesAnyObjectVisited / numObjectVisited);
2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120
}

/* -----------------------------------------------------------------------------
 * DEBUGGING CODE
 * -------------------------------------------------------------------------- */

#ifdef DEBUG_RETAINER

#define LOOKS_LIKE_PTR(r) ((LOOKS_LIKE_STATIC_CLOSURE(r) || \
        ((HEAP_ALLOCED(r) && Bdescr((P_)r)->free != (void *)-1))) && \
        ((StgWord)(*(StgPtr)r)!=0xaaaaaaaa))

static nat
sanityCheckHeapClosure( StgClosure *c )
{
    StgInfoTable *info;

    ASSERT(LOOKS_LIKE_GHC_INFO(c->header.info));
    ASSERT(!closure_STATIC(c));
    ASSERT(LOOKS_LIKE_PTR(c));

    if ((((StgWord)RSET(c) & 1) ^ flip) != 0) {
	if (get_itbl(c)->type == CONSTR &&
	    !strcmp(get_itbl(c)->prof.closure_type, "DEAD_WEAK") &&
	    !strcmp(get_itbl(c)->prof.closure_desc, "DEAD_WEAK")) {
	    fprintf(stderr, "\tUnvisited dead weak pointer object found: c = %p\n", c);
	    costArray[get_itbl(c)->type] += cost(c);
	    sumOfNewCost += cost(c);
	} else
	    fprintf(stderr,
		    "Unvisited object: flip = %d, c = %p(%d, %s, %s), rs = %p\n",
		    flip, c, get_itbl(c)->type,
		    get_itbl(c)->prof.closure_type, get_itbl(c)->prof.closure_desc,
		    RSET(c));
    } else {
	// fprintf(stderr, "sanityCheckHeapClosure) S: flip = %d, c = %p(%d), rs = %p\n", flip, c, get_itbl(c)->type, RSET(c));
    }

    info = get_itbl(c);
    switch (info->type) {
    case TSO:
	return tso_sizeW((StgTSO *)c);

    case THUNK:
    case THUNK_1_0:
    case THUNK_0_1:
    case THUNK_2_0:
    case THUNK_1_1:
    case THUNK_0_2:
	return stg_max(sizeW_fromITBL(info), sizeofW(StgHeader) + MIN_UPD_SIZE);

    case MVAR:
	return sizeofW(StgMVar);

    case MUT_ARR_PTRS:
    case MUT_ARR_PTRS_FROZEN:
	return mut_arr_ptrs_sizeW((StgMutArrPtrs *)c);

2121
    case AP:
2122 2123 2124
    case PAP:
	return pap_sizeW((StgPAP *)c);

2125 2126 2127
    case AP:
	return ap_stack_sizeW((StgAP_STACK *)c);

2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385
    case ARR_WORDS:
	return arr_words_sizeW((StgArrWords *)c);

    case CONSTR:
    case CONSTR_1_0:
    case CONSTR_0_1:
    case CONSTR_2_0:
    case CONSTR_1_1:
    case CONSTR_0_2:
    case FUN:
    case FUN_1_0:
    case FUN_0_1:
    case FUN_2_0:
    case FUN_1_1:
    case FUN_0_2:
    case WEAK:
    case MUT_VAR:
    case MUT_CONS:
    case CAF_BLACKHOLE:
    case BLACKHOLE:
    case SE_BLACKHOLE:
    case SE_CAF_BLACKHOLE:
    case BLACKHOLE_BQ:
    case IND_PERM:
    case IND_OLDGEN:
    case IND_OLDGEN_PERM:
    case FOREIGN:
    case BCO:
    case STABLE_NAME:
	return sizeW_fromITBL(info);

    case THUNK_SELECTOR:
	return sizeofW(StgHeader) + MIN_UPD_SIZE;

	/*
	  Error case
	*/
    case IND_STATIC:
    case CONSTR_STATIC:
    case FUN_STATIC:
    case THUNK_STATIC:
    case CONSTR_INTLIKE:
    case CONSTR_CHARLIKE:
    case CONSTR_NOCAF_STATIC:
    case UPDATE_FRAME:
    case CATCH_FRAME:
    case STOP_FRAME:
    case RET_DYN:
    case RET_BCO:
    case RET_SMALL:
    case RET_VEC_SMALL:
    case RET_BIG:
    case RET_VEC_BIG:
    case IND:
    case BLOCKED_FETCH:
    case FETCH_ME:
    case FETCH_ME_BQ:
    case RBH:
    case REMOTE_REF:
    case EVACUATED:
    case INVALID_OBJECT:
    default:
	barf("Invalid object in sanityCheckHeapClosure(): %d",
	     get_itbl(c)->type);
	return 0;
    }
}

static nat
heapCheck( bdescr *bd )
{
    StgPtr p;
    static nat costSum, size;

    costSum = 0;
    while (bd != NULL) {
	p = bd->start;
	while (p < bd->free) {
	    size = sanityCheckHeapClosure((StgClosure *)p);
	    sumOfCostLinear += size;
	    costArrayLinear[get_itbl((StgClosure *)p)->type] += size;
	    p += size;
	    // no need for slop check; I think slops are not used currently.
	}
	ASSERT(p == bd->free);
	costSum += bd->free - bd->start;
	bd = bd->link;
    }

    return costSum;
}

static nat
smallObjectPoolCheck(void)
{
    bdescr *bd;
    StgPtr p;
    static nat costSum, size;

    bd = small_alloc_list;
    costSum = 0;

    // first block
    if (bd == NULL)
	return costSum;

    p = bd->start;
    while (p < alloc_Hp) {
	size = sanityCheckHeapClosure((StgClosure *)p);
	sumOfCostLinear += size;
	costArrayLinear[get_itbl((StgClosure *)p)->type] += size;
	p += size;
    }
    ASSERT(p == alloc_Hp);
    costSum += alloc_Hp - bd->start;

    bd = bd->link;
    while (bd != NULL) {
	p = bd->start;
	while (p < bd->free) {
	    size = sanityCheckHeapClosure((StgClosure *)p);
	    sumOfCostLinear += size;
	    costArrayLinear[get_itbl((StgClosure *)p)->type] += size;
	    p += size;
	}
	ASSERT(p == bd->free);
	costSum += bd->free - bd->start;
	bd = bd->link;
    }

    return costSum;
}

static nat
chainCheck(bdescr *bd)
{
    nat costSum, size;

    costSum = 0;
    while (bd != NULL) {
	// bd->free - bd->start is not an accurate measurement of the
	// object size.  Actually it is always zero, so we compute its
	// size explicitly.
	size = sanityCheckHeapClosure((StgClosure *)bd->start);
	sumOfCostLinear += size;
	costArrayLinear[get_itbl((StgClosure *)bd->start)->type] += size;
	costSum += size;
	bd = bd->link;
    }

    return costSum;
}

static nat
checkHeapSanityForRetainerProfiling( void )
{
    nat costSum, g, s;

    costSum = 0;
    fprintf(stderr, "START: sumOfCostLinear = %d, costSum = %d\n", sumOfCostLinear, costSum);
    if (RtsFlags.GcFlags.generations == 1) {
	costSum += heapCheck(g0s0->to_blocks);
	fprintf(stderr, "heapCheck: sumOfCostLinear = %d, costSum = %d\n", sumOfCostLinear, costSum);
	costSum += chainCheck(g0s0->large_objects);
	fprintf(stderr, "chainCheck: sumOfCostLinear = %d, costSum = %d\n", sumOfCostLinear, costSum);
    } else {
	for (g = 0; g < RtsFlags.GcFlags.generations; g++)
	for (s = 0; s < generations[g].n_steps; s++) {
	    /*
	      After all live objects have been scavenged, the garbage
	      collector may create some objects in
	      scheduleFinalizers(). These objects are created throught
	      allocate(), so the small object pool or the large object
	      pool of the g0s0 may not be empty.
	    */
	    if (g == 0 && s == 0) {
		costSum += smallObjectPoolCheck();
		fprintf(stderr, "smallObjectPoolCheck(): sumOfCostLinear = %d, costSum = %d\n", sumOfCostLinear, costSum);
		costSum += chainCheck(generations[g].steps[s].large_objects);
		fprintf(stderr, "chainCheck(): sumOfCostLinear = %d, costSum = %d\n", sumOfCostLinear, costSum);
	    } else {
		costSum += heapCheck(generations[g].steps[s].blocks);
		fprintf(stderr, "heapCheck(): sumOfCostLinear = %d, costSum = %d\n", sumOfCostLinear, costSum);
		costSum += chainCheck(generations[g].steps[s].large_objects);
		fprintf(stderr, "chainCheck(): sumOfCostLinear = %d, costSum = %d\n", sumOfCostLinear, costSum);
	    }
	}
    }

    return costSum;
}

void
findPointer(StgPtr p)
{
    StgPtr q, r, e;
    bdescr *bd;
    nat g, s;

    for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
	for (s = 0; s < generations[g].n_steps; s++) {
	    // if (g == 0 && s == 0) continue;
	    bd = generations[g].steps[s].blocks;
	    for (; bd; bd = bd->link) {
		for (q = bd->start; q < bd->free; q++) {
		    if (*q == (StgWord)p) {
			r =