void
push_scanned_block (bdescr *bd, gen_workspace *ws)
{
ASSERT(bd != NULL);
ASSERT(bd->link == NULL);
ASSERT(bd->gen == ws->gen);
ASSERT(bd->u.scan == bd->free);
if (bd->start + bd->blocks * BLOCK_SIZE_W - bd->free > WORK_UNIT_WORDS)
{
// a partially full block: put it on the part_list list.
bd->link = ws->part_list;
ws->part_list = bd;
ws->n_part_blocks += bd->blocks;
IF_DEBUG(sanity,
ASSERT(countBlocks(ws->part_list) == ws->n_part_blocks));
}
else
{
// put the scan block on the ws->scavd_list.
bd->link = ws->scavd_list;
ws->scavd_list = bd;
ws->n_scavd_blocks += bd->blocks;
IF_DEBUG(sanity,
ASSERT(countBlocks(ws->scavd_list) == ws->n_scavd_blocks));
}
}
void
push_scanned_block (bdescr *bd, gen_workspace *ws)
{
ASSERT(bd != NULL);
ASSERT(bd->link == NULL);
ASSERT(bd->gen == ws->gen);
ASSERT(bd->u.scan == bd->free);
if (bd->blocks == 1 &&
bd->start + BLOCK_SIZE_W - bd->free > WORK_UNIT_WORDS)
{
// A partially full block: put it on the part_list list.
// Only for single objects - see Note [big objects]
bd->link = ws->part_list;
ws->part_list = bd;
ws->n_part_blocks += bd->blocks;
ws->n_part_words += bd->free - bd->start;
IF_DEBUG(sanity,
ASSERT(countBlocks(ws->part_list) == ws->n_part_blocks));
}
else
{
// put the scan block on the ws->scavd_list.
bd->link = ws->scavd_list;
ws->scavd_list = bd;
ws->n_scavd_blocks += bd->blocks;
ws->n_scavd_words += bd->free - bd->start;
IF_DEBUG(sanity,
ASSERT(countBlocks(ws->scavd_list) == ws->n_scavd_blocks));
}
}
static void checkGeneration (generation *gen,
rtsBool after_major_gc USED_IF_THREADS)
{
nat n;
gen_workspace *ws;
ASSERT(countBlocks(gen->blocks) == gen->n_blocks);
ASSERT(countBlocks(gen->large_objects) == gen->n_large_blocks);
#if defined(THREADED_RTS)
// heap sanity checking doesn't work with SMP, because we can't
// zero the slop (see Updates.h). However, we can sanity-check
// the heap after a major gc, because there is no slop.
if (!after_major_gc) return;
#endif
checkHeapChain(gen->blocks);
for (n = 0; n < n_capabilities; n++) {
ws = &gc_threads[n]->gens[gen->no];
checkHeapChain(ws->todo_bd);
checkHeapChain(ws->part_list);
checkHeapChain(ws->scavd_list);
}
checkLargeObjects(gen->large_objects);
}
static W_
genBlocks (generation *gen)
{
ASSERT(countBlocks(gen->blocks) == gen->n_blocks);
ASSERT(countBlocks(gen->large_objects) == gen->n_large_blocks);
return gen->n_blocks + gen->n_old_blocks +
countAllocdBlocks(gen->large_objects);
}
static W_
genBlocks (generation *gen)
{
ASSERT(countBlocks(gen->blocks) == gen->n_blocks);
ASSERT(countBlocks(gen->large_objects) == gen->n_large_blocks);
ASSERT(countCompactBlocks(gen->compact_objects) == gen->n_compact_blocks);
ASSERT(countCompactBlocks(gen->compact_blocks_in_import) == gen->n_compact_blocks_in_import);
return gen->n_blocks + gen->n_old_blocks +
countAllocdBlocks(gen->large_objects) +
countAllocdCompactBlocks(gen->compact_objects) +
countAllocdCompactBlocks(gen->compact_blocks_in_import);
}
void bigWigAverageOverBed(char *inBw, char *inBed, char *outTab)
/* bigWigAverageOverBed - Compute average score of big wig over each bed, which may have introns. */
{
struct bed *bedList;
int fieldCount;
bedLoadAllReturnFieldCount(inBed, &bedList, &fieldCount);
checkUniqueNames(bedList);
struct bbiFile *bbi = bigWigFileOpen(inBw);
FILE *f = mustOpen(outTab, "w");
FILE *bedF = NULL;
if (bedOut != NULL)
bedF = mustOpen(bedOut, "w");
/* Count up number of blocks in file. It takes about 1/100th of of second to
* look up a single block in a bigWig. On the other hand to stream through
* the whole file setting a array of doubles takes about 30 seconds, so we change
* strategy at 3,000 blocks.
* I (Jim) usually avoid having two paths through the code like this, and am tempted
* to always go the ~30 second chromosome-at-a-time way. On the other hand the block-way
* was developed first, and it was useful to have both ways to test against each other.
* (This found a bug where the chromosome way wasn't handling beds in chromosomes not
* covered by the bigWig for instance). Since this code is not likely to change too
* much, keeping both implementations in seems reasonable. */
int blockCount = countBlocks(bedList, fieldCount);
verbose(2, "Got %d blocks, if >= 3000 will use chromosome-at-a-time method\n", blockCount);
if (blockCount < 3000)
averageFetchingEachBlock(bbi, bedList, fieldCount, f, bedF);
else
averageFetchingEachChrom(bbi, &bedList, fieldCount, f, bedF);
carefulClose(&bedF);
carefulClose(&f);
}
static int dir_mdu(direntry_t *entry, MainParam_t *mp)
{
Arg_t *parentArg = (Arg_t *) (mp->arg);
Arg_t arg;
int ret;
arg = *parentArg;
arg.mp.arg = (void *) &arg;
arg.parent = parentArg;
arg.inDir = 1;
/* account for the space occupied by the directory itself */
if(!isRootDir(entry->Dir)) {
arg.blocks = countBlocks(entry->Dir,
getStart(entry->Dir, &entry->dir));
} else {
arg.blocks = 0;
}
/* recursion */
ret = mp->loop(mp->File, &arg.mp, "*");
if(!arg.summary || !parentArg->inDir) {
printf("%-7d ", arg.blocks);
fprintPwd(stdout, entry,0);
fputc('\n', stdout);
}
arg.parent->blocks += arg.blocks;
return ret;
}
W_ gcThreadLiveBlocks (nat i, nat g)
{
W_ blocks;
blocks = countBlocks(gc_threads[i]->gens[g].todo_bd);
blocks += gc_threads[i]->gens[g].n_part_blocks;
blocks += gc_threads[i]->gens[g].n_scavd_blocks;
return blocks;
}
static int file_mdu(direntry_t *entry, MainParam_t *mp)
{
unsigned int blocks;
Arg_t * arg = (Arg_t *) (mp->arg);
blocks = countBlocks(entry->Dir,getStart(entry->Dir, &entry->dir));
if(arg->all || !arg->inDir) {
printf("%-7d ", blocks);
fprintPwd(stdout, entry,0);
fputc('\n', stdout);
}
arg->blocks += blocks;
return GOT_ONE;
}
BCRS* CreateBCRS(MATRIX* my_m) {
int i, jj, k, index, block_number = 0;
int nrows = my_m->nrows;
int ncols = my_m->ncols;
int nnz = my_m->nnz;
double** mal = my_m->mel;
double current;
BCRS* cc = (BCRS*)malloc(sizeof(BCRS));
int nblocks = countBlocks(my_m);
int* colInd = malloc(nblocks * sizeof(int));
int* rowPtr = malloc((nrows+1) * sizeof(int));
int* nnzPtr = malloc((nblocks+1) * sizeof(int));
double* value = malloc(nnz * sizeof(double));
for(i=0; i<nrows+1; ++i) {
rowPtr[i] = 0;
}
index = 0;
for(k=0; k<nrows; k++){
for(jj=0; jj<ncols; ++jj){
if (mal[k][jj] != 0) {
if ((jj == 0) || (jj != 0 &&(mal[k][jj-1] == 0))) {
colInd[block_number] = jj;
nnzPtr[block_number] = index;
block_number++;
}
value[index] = mal[k][jj];
index++;
}
}
rowPtr[k+1] = block_number;
}
nnzPtr[nblocks] = nnz;
cc->colInd = colInd;
cc->rowPtr = rowPtr;
cc->value = value;
cc->nnzPtr = nnzPtr;
cc->nrows = nrows;
cc->ncols = ncols;
cc->nnz = nnz;
cc->nblocks = nblocks;
return cc;
}
//
// Resize each of the nurseries to the specified size.
//
static void
resizeNurseriesEach (W_ blocks)
{
uint32_t i, node;
bdescr *bd;
W_ nursery_blocks;
nursery *nursery;
for (i = 0; i < n_nurseries; i++) {
nursery = &nurseries[i];
nursery_blocks = nursery->n_blocks;
if (nursery_blocks == blocks) continue;
node = capNoToNumaNode(i);
if (nursery_blocks < blocks) {
debugTrace(DEBUG_gc, "increasing size of nursery to %d blocks",
blocks);
nursery->blocks = allocNursery(node, nursery->blocks,
blocks-nursery_blocks);
}
else
{
bdescr *next_bd;
debugTrace(DEBUG_gc, "decreasing size of nursery to %d blocks",
blocks);
bd = nursery->blocks;
while (nursery_blocks > blocks) {
next_bd = bd->link;
next_bd->u.back = NULL;
nursery_blocks -= bd->blocks; // might be a large block
freeGroup(bd);
bd = next_bd;
}
nursery->blocks = bd;
// might have gone just under, by freeing a large block, so make
// up the difference.
if (nursery_blocks < blocks) {
nursery->blocks = allocNursery(node, nursery->blocks,
blocks-nursery_blocks);
}
}
nursery->n_blocks = blocks;
ASSERT(countBlocks(nursery->blocks) == nursery->n_blocks);
}
}
static void
resizeNursery (nursery *nursery, W_ blocks)
{
bdescr *bd;
W_ nursery_blocks;
nursery_blocks = nursery->n_blocks;
if (nursery_blocks == blocks) return;
if (nursery_blocks < blocks) {
debugTrace(DEBUG_gc, "increasing size of nursery to %d blocks",
blocks);
nursery->blocks = allocNursery(nursery->blocks, blocks-nursery_blocks);
}
else {
bdescr *next_bd;
debugTrace(DEBUG_gc, "decreasing size of nursery to %d blocks",
blocks);
bd = nursery->blocks;
while (nursery_blocks > blocks) {
next_bd = bd->link;
next_bd->u.back = NULL;
nursery_blocks -= bd->blocks; // might be a large block
freeGroup(bd);
bd = next_bd;
}
nursery->blocks = bd;
// might have gone just under, by freeing a large block, so make
// up the difference.
if (nursery_blocks < blocks) {
nursery->blocks = allocNursery(nursery->blocks, blocks-nursery_blocks);
}
}
nursery->n_blocks = blocks;
ASSERT(countBlocks(nursery->blocks) == nursery->n_blocks);
}
void
sweep(generation *gen)
{
bdescr *bd, *prev, *next;
uint32_t i;
W_ freed, resid, fragd, blocks, live;
ASSERT(countBlocks(gen->old_blocks) == gen->n_old_blocks);
live = 0; // estimate of live data in this gen
freed = 0;
fragd = 0;
blocks = 0;
prev = NULL;
for (bd = gen->old_blocks; bd != NULL; bd = next)
{
next = bd->link;
if (!(bd->flags & BF_MARKED)) {
prev = bd;
continue;
}
blocks++;
resid = 0;
for (i = 0; i < BLOCK_SIZE_W / BITS_IN(W_); i++)
{
if (bd->u.bitmap[i] != 0) resid++;
}
live += resid * BITS_IN(W_);
if (resid == 0)
{
freed++;
gen->n_old_blocks--;
if (prev == NULL) {
gen->old_blocks = next;
} else {
prev->link = next;
}
freeGroup(bd);
}
else
{
prev = bd;
if (resid < (BLOCK_SIZE_W * 3) / (BITS_IN(W_) * 4)) {
fragd++;
bd->flags |= BF_FRAGMENTED;
}
bd->flags |= BF_SWEPT;
}
}
gen->live_estimate = live;
debugTrace(DEBUG_gc, "sweeping: %d blocks, %d were copied, %d freed (%d%%), %d are fragmented, live estimate: %ld%%",
gen->n_old_blocks + freed,
gen->n_old_blocks - blocks + freed,
freed,
blocks == 0 ? 0 : (freed * 100) / blocks,
fragd,
(unsigned long)((blocks - freed) == 0 ? 0 : ((live / BLOCK_SIZE_W) * 100) / (blocks - freed)));
ASSERT(countBlocks(gen->old_blocks) == gen->n_old_blocks);
}
void
memInventory (rtsBool show)
{
nat g, i;
W_ gen_blocks[RtsFlags.GcFlags.generations];
W_ nursery_blocks, retainer_blocks,
arena_blocks, exec_blocks;
W_ live_blocks = 0, free_blocks = 0;
rtsBool leak;
// count the blocks we current have
for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
gen_blocks[g] = 0;
for (i = 0; i < n_capabilities; i++) {
gen_blocks[g] += countBlocks(capabilities[i].mut_lists[g]);
gen_blocks[g] += countBlocks(gc_threads[i]->gens[g].part_list);
gen_blocks[g] += countBlocks(gc_threads[i]->gens[g].scavd_list);
gen_blocks[g] += countBlocks(gc_threads[i]->gens[g].todo_bd);
}
gen_blocks[g] += genBlocks(&generations[g]);
}
nursery_blocks = 0;
for (i = 0; i < n_capabilities; i++) {
ASSERT(countBlocks(nurseries[i].blocks) == nurseries[i].n_blocks);
nursery_blocks += nurseries[i].n_blocks;
if (capabilities[i].pinned_object_block != NULL) {
nursery_blocks += capabilities[i].pinned_object_block->blocks;
}
nursery_blocks += countBlocks(capabilities[i].pinned_object_blocks);
}
retainer_blocks = 0;
#ifdef PROFILING
if (RtsFlags.ProfFlags.doHeapProfile == HEAP_BY_RETAINER) {
retainer_blocks = retainerStackBlocks();
}
#endif
// count the blocks allocated by the arena allocator
arena_blocks = arenaBlocks();
// count the blocks containing executable memory
exec_blocks = countAllocdBlocks(exec_block);
/* count the blocks on the free list */
free_blocks = countFreeList();
live_blocks = 0;
for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
live_blocks += gen_blocks[g];
}
live_blocks += nursery_blocks +
+ retainer_blocks + arena_blocks + exec_blocks;
#define MB(n) (((double)(n) * BLOCK_SIZE_W) / ((1024*1024)/sizeof(W_)))
leak = live_blocks + free_blocks != mblocks_allocated * BLOCKS_PER_MBLOCK;
if (show || leak)
{
if (leak) {
debugBelch("Memory leak detected:\n");
} else {
debugBelch("Memory inventory:\n");
}
for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
debugBelch(" gen %d blocks : %5" FMT_Word " blocks (%6.1lf MB)\n", g,
gen_blocks[g], MB(gen_blocks[g]));
}
debugBelch(" nursery : %5" FMT_Word " blocks (%6.1lf MB)\n",
nursery_blocks, MB(nursery_blocks));
debugBelch(" retainer : %5" FMT_Word " blocks (%6.1lf MB)\n",
retainer_blocks, MB(retainer_blocks));
debugBelch(" arena blocks : %5" FMT_Word " blocks (%6.1lf MB)\n",
arena_blocks, MB(arena_blocks));
debugBelch(" exec : %5" FMT_Word " blocks (%6.1lf MB)\n",
exec_blocks, MB(exec_blocks));
debugBelch(" free : %5" FMT_Word " blocks (%6.1lf MB)\n",
free_blocks, MB(free_blocks));
debugBelch(" total : %5" FMT_Word " blocks (%6.1lf MB)\n",
live_blocks + free_blocks, MB(live_blocks+free_blocks));
if (leak) {
debugBelch("\n in system : %5" FMT_Word " blocks (%" FMT_Word " MB)\n",
mblocks_allocated * BLOCKS_PER_MBLOCK, mblocks_allocated);
}
}
if (leak) {
debugBelch("\n");
findMemoryLeak();
}
ASSERT(n_alloc_blocks == live_blocks);
ASSERT(!leak);
}