Use table lookup for run_quantize_{floor,ceil}().

Reduce run quantization overhead by generating lookup tables during
bootstrapping, and using the tables for all subsequent run quantization.
This commit is contained in:
Jason Evans 2016-02-22 16:20:56 -08:00
parent 08551eee58
commit 0da8ce1e96
4 changed files with 90 additions and 32 deletions

View File

@ -494,7 +494,7 @@ extern size_t map_bias; /* Number of arena chunk header pages. */
extern size_t map_misc_offset; extern size_t map_misc_offset;
extern size_t arena_maxrun; /* Max run size for arenas. */ extern size_t arena_maxrun; /* Max run size for arenas. */
extern size_t large_maxclass; /* Max large size class. */ extern size_t large_maxclass; /* Max large size class. */
extern size_t small_maxrun; /* Max run size for small size classes. */ extern size_t run_quantize_max; /* Max run_quantize_*() input. */
extern unsigned nlclasses; /* Number of large size classes. */ extern unsigned nlclasses; /* Number of large size classes. */
extern unsigned nhclasses; /* Number of huge size classes. */ extern unsigned nhclasses; /* Number of huge size classes. */

View File

@ -447,6 +447,7 @@ rtree_val_read
rtree_val_write rtree_val_write
run_quantize_ceil run_quantize_ceil
run_quantize_floor run_quantize_floor
run_quantize_max
s2u s2u
s2u_compute s2u_compute
s2u_lookup s2u_lookup

View File

@ -21,8 +21,11 @@ size_t map_bias;
size_t map_misc_offset; size_t map_misc_offset;
size_t arena_maxrun; /* Max run size for arenas. */ size_t arena_maxrun; /* Max run size for arenas. */
size_t large_maxclass; /* Max large size class. */ size_t large_maxclass; /* Max large size class. */
size_t small_maxrun; /* Max run size for small size classes. */ size_t run_quantize_max; /* Max run_quantize_*() input. */
static size_t small_maxrun; /* Max run size for small size classes. */
static bool *small_run_tab; /* Valid small run page multiples. */ static bool *small_run_tab; /* Valid small run page multiples. */
static size_t *run_quantize_floor_tab; /* run_quantize_floor() memoization. */
static size_t *run_quantize_ceil_tab; /* run_quantize_ceil() memoization. */
unsigned nlclasses; /* Number of large size classes. */ unsigned nlclasses; /* Number of large size classes. */
unsigned nhclasses; /* Number of huge size classes. */ unsigned nhclasses; /* Number of huge size classes. */
@ -100,12 +103,8 @@ arena_run_comp(const arena_chunk_map_misc_t *a, const arena_chunk_map_misc_t *b)
rb_gen(static UNUSED, arena_run_tree_, arena_run_tree_t, arena_chunk_map_misc_t, rb_gen(static UNUSED, arena_run_tree_, arena_run_tree_t, arena_chunk_map_misc_t,
rb_link, arena_run_comp) rb_link, arena_run_comp)
#ifdef JEMALLOC_JET
#undef run_quantize_floor
#define run_quantize_floor JEMALLOC_N(run_quantize_floor_impl)
#endif
static size_t static size_t
run_quantize_floor(size_t size) run_quantize_floor_compute(size_t size)
{ {
size_t qsize; size_t qsize;
@ -123,18 +122,13 @@ run_quantize_floor(size_t size)
*/ */
qsize = index2size(size2index(size - large_pad + 1) - 1) + large_pad; qsize = index2size(size2index(size - large_pad + 1) - 1) + large_pad;
if (qsize <= SMALL_MAXCLASS + large_pad) if (qsize <= SMALL_MAXCLASS + large_pad)
return (run_quantize_floor(size - large_pad)); return (run_quantize_floor_compute(size - large_pad));
assert(qsize <= size); assert(qsize <= size);
return (qsize); return (qsize);
} }
#ifdef JEMALLOC_JET
#undef run_quantize_floor
#define run_quantize_floor JEMALLOC_N(run_quantize_floor)
run_quantize_t *run_quantize_floor = JEMALLOC_N(run_quantize_floor_impl);
#endif
static size_t static size_t
run_quantize_ceil_hard(size_t size) run_quantize_ceil_compute_hard(size_t size)
{ {
size_t large_run_size_next; size_t large_run_size_next;
@ -167,14 +161,10 @@ run_quantize_ceil_hard(size_t size)
} }
} }
#ifdef JEMALLOC_JET
#undef run_quantize_ceil
#define run_quantize_ceil JEMALLOC_N(run_quantize_ceil_impl)
#endif
static size_t static size_t
run_quantize_ceil(size_t size) run_quantize_ceil_compute(size_t size)
{ {
size_t qsize = run_quantize_floor(size); size_t qsize = run_quantize_floor_compute(size);
if (qsize < size) { if (qsize < size) {
/* /*
@ -185,10 +175,51 @@ run_quantize_ceil(size_t size)
* search would potentially find sufficiently aligned available * search would potentially find sufficiently aligned available
* memory somewhere lower. * memory somewhere lower.
*/ */
qsize = run_quantize_ceil_hard(qsize); qsize = run_quantize_ceil_compute_hard(qsize);
} }
return (qsize); return (qsize);
} }
#ifdef JEMALLOC_JET
#undef run_quantize_floor
#define run_quantize_floor JEMALLOC_N(run_quantize_floor_impl)
#endif
static size_t
run_quantize_floor(size_t size)
{
size_t ret;
assert(size > 0);
assert(size <= run_quantize_max);
assert((size & PAGE_MASK) == 0);
ret = run_quantize_floor_tab[(size >> LG_PAGE) - 1];
assert(ret == run_quantize_floor_compute(size));
return (ret);
}
#ifdef JEMALLOC_JET
#undef run_quantize_floor
#define run_quantize_floor JEMALLOC_N(run_quantize_floor)
run_quantize_t *run_quantize_floor = JEMALLOC_N(run_quantize_floor_impl);
#endif
#ifdef JEMALLOC_JET
#undef run_quantize_ceil
#define run_quantize_ceil JEMALLOC_N(run_quantize_ceil_impl)
#endif
static size_t
run_quantize_ceil(size_t size)
{
size_t ret;
assert(size > 0);
assert(size <= run_quantize_max);
assert((size & PAGE_MASK) == 0);
ret = run_quantize_ceil_tab[(size >> LG_PAGE) - 1];
assert(ret == run_quantize_ceil_compute(size));
return (ret);
}
#ifdef JEMALLOC_JET #ifdef JEMALLOC_JET
#undef run_quantize_ceil #undef run_quantize_ceil
#define run_quantize_ceil JEMALLOC_N(run_quantize_ceil) #define run_quantize_ceil JEMALLOC_N(run_quantize_ceil)
@ -3522,6 +3553,35 @@ small_run_size_init(void)
return (false); return (false);
} }
static bool
run_quantize_init(void)
{
unsigned i;
run_quantize_max = chunksize + large_pad;
run_quantize_floor_tab = (size_t *)base_alloc(sizeof(size_t) *
(run_quantize_max >> LG_PAGE));
if (run_quantize_floor_tab == NULL)
return (true);
run_quantize_ceil_tab = (size_t *)base_alloc(sizeof(size_t) *
(run_quantize_max >> LG_PAGE));
if (run_quantize_ceil_tab == NULL)
return (true);
for (i = 1; i <= run_quantize_max >> LG_PAGE; i++) {
size_t run_size = i << LG_PAGE;
run_quantize_floor_tab[i-1] =
run_quantize_floor_compute(run_size);
run_quantize_ceil_tab[i-1] =
run_quantize_ceil_compute(run_size);
}
return (false);
}
bool bool
arena_boot(void) arena_boot(void)
{ {
@ -3570,7 +3630,12 @@ arena_boot(void)
nhclasses = NSIZES - nlclasses - NBINS; nhclasses = NSIZES - nlclasses - NBINS;
bin_info_init(); bin_info_init();
return (small_run_size_init()); if (small_run_size_init())
return (true);
if (run_quantize_init())
return (true);
return (false);
} }
void void

View File

@ -93,19 +93,14 @@ TEST_END
TEST_BEGIN(test_monotonic) TEST_BEGIN(test_monotonic)
{ {
bool cache_oblivious;
unsigned nbins, nlruns, i; unsigned nbins, nlruns, i;
size_t sz, max_run_size, floor_prev, ceil_prev; size_t sz, floor_prev, ceil_prev;
/* /*
* Iterate over all run sizes and verify that * Iterate over all run sizes and verify that
* run_quantize_{floor,ceil}() are monotonic. * run_quantize_{floor,ceil}() are monotonic.
*/ */
sz = sizeof(bool);
assert_d_eq(mallctl("config.cache_oblivious", &cache_oblivious, &sz,
NULL, 0), 0, "Unexpected mallctl failure");
sz = sizeof(unsigned); sz = sizeof(unsigned);
assert_d_eq(mallctl("arenas.nbins", &nbins, &sz, NULL, 0), 0, assert_d_eq(mallctl("arenas.nbins", &nbins, &sz, NULL, 0), 0,
"Unexpected mallctl failure"); "Unexpected mallctl failure");
@ -114,12 +109,9 @@ TEST_BEGIN(test_monotonic)
assert_d_eq(mallctl("arenas.nlruns", &nlruns, &sz, NULL, 0), 0, assert_d_eq(mallctl("arenas.nlruns", &nlruns, &sz, NULL, 0), 0,
"Unexpected mallctl failure"); "Unexpected mallctl failure");
max_run_size = (large_maxclass > small_maxrun) ? large_maxclass :
small_maxrun;
floor_prev = 0; floor_prev = 0;
ceil_prev = 0; ceil_prev = 0;
for (i = 1; i < max_run_size >> LG_PAGE; i++) { for (i = 1; i < run_quantize_max >> LG_PAGE; i++) {
size_t run_size, floor, ceil; size_t run_size, floor, ceil;
run_size = i << LG_PAGE; run_size = i << LG_PAGE;