project-base/server-skynet-source-3rd-jemalloc
2
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
eeabdd2466
server-skynet-source-3rd-je.../include/jemalloc/internal/rtree_inlines.h

510 lines
17 KiB
C
Raw Normal View History

Break up headers into constituent parts This is part of a broader change to make header files better represent the dependencies between one another (see https://github.com/jemalloc/jemalloc/issues/533). It breaks up component headers into smaller parts that can be made to have a simpler dependency graph. For the autogenerated headers (smoothstep.h and size_classes.h), no splitting was necessary, so I didn't add support to emit multiple headers.
2017-01-11 10:06:31 +08:00
#ifndef JEMALLOC_INTERNAL_RTREE_INLINES_H
#define JEMALLOC_INTERNAL_RTREE_INLINES_H
Port to Mac OS X. Add Mac OS X support, based in large part on the OS X support in Mozilla's version of jemalloc.
2010-09-06 01:35:13 +08:00
#ifndef JEMALLOC_ENABLE_INLINE
Determine rtree levels at compile time. Rather than dynamically building a table to aid per level computations, define a constant table at compile time. Omit both high and low insignificant bits. Use one to three tree levels, depending on the number of significant bits.
2017-02-07 05:17:12 +08:00
uintptr_t rtree_leafkey(uintptr_t key);
uintptr_t rtree_subkey(uintptr_t key, unsigned level);
Implement compact rtree leaf element representation. If a single virtual adddress pointer has enough unused bits to pack {szind_t, extent_t *, bool, bool}, use a single pointer-sized field in each rtree leaf element, rather than using three separate fields. This has little impact on access speed (fewer loads/stores, but more bit twiddling), except that denser representation increases TLB effectiveness.
2017-03-21 07:38:21 +08:00
# ifdef RTREE_LEAF_COMPACT
uintptr_t rtree_leaf_elm_bits_read(tsdn_t *tsdn, rtree_t *rtree,
rtree_leaf_elm_t *elm, bool acquired, bool dependent);
extent_t *rtree_leaf_elm_bits_extent_get(uintptr_t bits);
szind_t rtree_leaf_elm_bits_szind_get(uintptr_t bits);
bool rtree_leaf_elm_bits_slab_get(uintptr_t bits);
bool rtree_leaf_elm_bits_locked_get(uintptr_t bits);
# endif
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
extent_t *rtree_leaf_elm_extent_read(tsdn_t *tsdn, rtree_t *rtree,
rtree_leaf_elm_t *elm, bool acquired, bool dependent);
szind_t rtree_leaf_elm_szind_read(tsdn_t *tsdn, rtree_t *rtree,
rtree_leaf_elm_t *elm, bool acquired, bool dependent);
bool rtree_leaf_elm_slab_read(tsdn_t *tsdn, rtree_t *rtree,
rtree_leaf_elm_t *elm, bool acquired, bool dependent);
void rtree_leaf_elm_extent_write(tsdn_t *tsdn, rtree_t *rtree,
rtree_leaf_elm_t *elm, bool acquired, extent_t *extent);
void rtree_leaf_elm_szind_write(tsdn_t *tsdn, rtree_t *rtree,
rtree_leaf_elm_t *elm, bool acquired, szind_t szind);
void rtree_leaf_elm_slab_write(tsdn_t *tsdn, rtree_t *rtree,
rtree_leaf_elm_t *elm, bool acquired, bool slab);
void rtree_leaf_elm_write(tsdn_t *tsdn, rtree_t *rtree, rtree_leaf_elm_t *elm,
bool acquired, extent_t *extent, szind_t szind, bool slab);
Fix a race in rtree_szind_slab_update() for RTREE_LEAF_COMPACT.
2017-03-27 19:08:51 +08:00
void rtree_leaf_elm_szind_slab_update(tsdn_t *tsdn, rtree_t *rtree,
rtree_leaf_elm_t *elm, szind_t szind, bool slab);
Split rtree_elm_t into rtree_{node,leaf}_elm_t. This allows leaf elements to differ in size from internal node elements. In principle it would be more correct to use a different type for each level of the tree, but due to implementation details related to atomic operations, we use casts anyway, thus counteracting the value of additional type correctness. Furthermore, such a scheme would require function code generation (via cpp macros), as well as either unwieldy type names for leaves or type aliases, e.g. typedef struct rtree_elm_d2_s rtree_leaf_elm_t; This alternate strategy would be more correct, and with less code duplication, but probably not worth the complexity.
2017-03-17 00:46:42 +08:00
rtree_leaf_elm_t *rtree_leaf_elm_lookup(tsdn_t *tsdn, rtree_t *rtree,
Add rtree lookup path caching. rtree-based extent lookups remain more expensive than chunk-based run lookups, but with this optimization the fast path slowdown is ~3 CPU cycles per metadata lookup (on Intel Core i7-4980HQ), versus ~11 cycles prior. The path caching speedup tends to degrade gracefully unless allocated memory is spread far apart (as is the case when using a mixture of sbrk() and mmap()).
2016-06-03 09:43:10 +08:00
rtree_ctx_t *rtree_ctx, uintptr_t key, bool dependent, bool init_missing);
Make non-essential inline rtree functions static functions.
2017-02-04 11:44:33 +08:00
bool rtree_write(tsdn_t *tsdn, rtree_t *rtree, rtree_ctx_t *rtree_ctx,
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
uintptr_t key, extent_t *extent, szind_t szind, bool slab);
rtree_leaf_elm_t *rtree_read(tsdn_t *tsdn, rtree_t *rtree,
rtree_ctx_t *rtree_ctx, uintptr_t key, bool dependent);
extent_t *rtree_extent_read(tsdn_t *tsdn, rtree_t *rtree,
rtree_ctx_t *rtree_ctx, uintptr_t key, bool dependent);
szind_t rtree_szind_read(tsdn_t *tsdn, rtree_t *rtree,
rtree_ctx_t *rtree_ctx, uintptr_t key, bool dependent);
bool rtree_extent_szind_read(tsdn_t *tsdn, rtree_t *rtree,
rtree_ctx_t *rtree_ctx, uintptr_t key, bool dependent, extent_t **r_extent,
szind_t *r_szind);
bool rtree_szind_slab_read(tsdn_t *tsdn, rtree_t *rtree, rtree_ctx_t *rtree_ctx,
uintptr_t key, bool dependent, szind_t *r_szind, bool *r_slab);
Split rtree_elm_t into rtree_{node,leaf}_elm_t. This allows leaf elements to differ in size from internal node elements. In principle it would be more correct to use a different type for each level of the tree, but due to implementation details related to atomic operations, we use casts anyway, thus counteracting the value of additional type correctness. Furthermore, such a scheme would require function code generation (via cpp macros), as well as either unwieldy type names for leaves or type aliases, e.g. typedef struct rtree_elm_d2_s rtree_leaf_elm_t; This alternate strategy would be more correct, and with less code duplication, but probably not worth the complexity.
2017-03-17 00:46:42 +08:00
rtree_leaf_elm_t *rtree_leaf_elm_acquire(tsdn_t *tsdn, rtree_t *rtree,
Add rtree lookup path caching. rtree-based extent lookups remain more expensive than chunk-based run lookups, but with this optimization the fast path slowdown is ~3 CPU cycles per metadata lookup (on Intel Core i7-4980HQ), versus ~11 cycles prior. The path caching speedup tends to degrade gracefully unless allocated memory is spread far apart (as is the case when using a mixture of sbrk() and mmap()).
2016-06-03 09:43:10 +08:00
rtree_ctx_t *rtree_ctx, uintptr_t key, bool dependent, bool init_missing);
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
void rtree_leaf_elm_release(tsdn_t *tsdn, rtree_t *rtree,
Split rtree_elm_t into rtree_{node,leaf}_elm_t. This allows leaf elements to differ in size from internal node elements. In principle it would be more correct to use a different type for each level of the tree, but due to implementation details related to atomic operations, we use casts anyway, thus counteracting the value of additional type correctness. Furthermore, such a scheme would require function code generation (via cpp macros), as well as either unwieldy type names for leaves or type aliases, e.g. typedef struct rtree_elm_d2_s rtree_leaf_elm_t; This alternate strategy would be more correct, and with less code duplication, but probably not worth the complexity.
2017-03-17 00:46:42 +08:00
rtree_leaf_elm_t *elm);
Fix a race in rtree_szind_slab_update() for RTREE_LEAF_COMPACT.
2017-03-27 19:08:51 +08:00
void rtree_szind_slab_update(tsdn_t *tsdn, rtree_t *rtree,
rtree_ctx_t *rtree_ctx, uintptr_t key, szind_t szind, bool slab);
Make non-essential inline rtree functions static functions.
2017-02-04 11:44:33 +08:00
void rtree_clear(tsdn_t *tsdn, rtree_t *rtree, rtree_ctx_t *rtree_ctx,
Add rtree lookup path caching. rtree-based extent lookups remain more expensive than chunk-based run lookups, but with this optimization the fast path slowdown is ~3 CPU cycles per metadata lookup (on Intel Core i7-4980HQ), versus ~11 cycles prior. The path caching speedup tends to degrade gracefully unless allocated memory is spread far apart (as is the case when using a mixture of sbrk() and mmap()).
2016-06-03 09:43:10 +08:00
uintptr_t key);
Port to Mac OS X. Add Mac OS X support, based in large part on the OS X support in Mozilla's version of jemalloc.
2010-09-06 01:35:13 +08:00
#endif
Add the "stats.cactive" mallctl. Add the "stats.cactive" mallctl, which can be used to efficiently and repeatedly query approximately how much active memory the application is utilizing.
2011-03-19 08:56:14 +08:00
#if (defined(JEMALLOC_ENABLE_INLINE) || defined(JEMALLOC_RTREE_C_))
Replace rtree path cache with LRU cache. Rework rtree_ctx_t to encapsulate an rtree leaf LRU lookup cache rather than a single-path element lookup cache. The replacement is logically much simpler, as well as slightly faster in the fast path case and less prone to degraded performance during non-trivial sequences of lookups.
2017-02-04 12:21:56 +08:00
JEMALLOC_ALWAYS_INLINE uintptr_t
Determine rtree levels at compile time. Rather than dynamically building a table to aid per level computations, define a constant table at compile time. Omit both high and low insignificant bits. Use one to three tree levels, depending on the number of significant bits.
2017-02-07 05:17:12 +08:00
rtree_leafkey(uintptr_t key) {
Replace rtree path cache with LRU cache. Rework rtree_ctx_t to encapsulate an rtree leaf LRU lookup cache rather than a single-path element lookup cache. The replacement is logically much simpler, as well as slightly faster in the fast path case and less prone to degraded performance during non-trivial sequences of lookups.
2017-02-04 12:21:56 +08:00
unsigned ptrbits = ZU(1) << (LG_SIZEOF_PTR+3);
Determine rtree levels at compile time. Rather than dynamically building a table to aid per level computations, define a constant table at compile time. Omit both high and low insignificant bits. Use one to three tree levels, depending on the number of significant bits.
2017-02-07 05:17:12 +08:00
unsigned cumbits = (rtree_levels[RTREE_HEIGHT-1].cumbits -
rtree_levels[RTREE_HEIGHT-1].bits);
Replace rtree path cache with LRU cache. Rework rtree_ctx_t to encapsulate an rtree leaf LRU lookup cache rather than a single-path element lookup cache. The replacement is logically much simpler, as well as slightly faster in the fast path case and less prone to degraded performance during non-trivial sequences of lookups.
2017-02-04 12:21:56 +08:00
unsigned maskbits = ptrbits - cumbits;
uintptr_t mask = ~((ZU(1) << maskbits) - 1);
return (key & mask);
Port to Mac OS X. Add Mac OS X support, based in large part on the OS X support in Mozilla's version of jemalloc.
2010-09-06 01:35:13 +08:00
}
Always inline performance-critical rtree operations.
2016-03-24 07:14:41 +08:00
JEMALLOC_ALWAYS_INLINE uintptr_t
Determine rtree levels at compile time. Rather than dynamically building a table to aid per level computations, define a constant table at compile time. Omit both high and low insignificant bits. Use one to three tree levels, depending on the number of significant bits.
2017-02-07 05:17:12 +08:00
rtree_subkey(uintptr_t key, unsigned level) {
Replace rtree path cache with LRU cache. Rework rtree_ctx_t to encapsulate an rtree leaf LRU lookup cache rather than a single-path element lookup cache. The replacement is logically much simpler, as well as slightly faster in the fast path case and less prone to degraded performance during non-trivial sequences of lookups.
2017-02-04 12:21:56 +08:00
unsigned ptrbits = ZU(1) << (LG_SIZEOF_PTR+3);
Determine rtree levels at compile time. Rather than dynamically building a table to aid per level computations, define a constant table at compile time. Omit both high and low insignificant bits. Use one to three tree levels, depending on the number of significant bits.
2017-02-07 05:17:12 +08:00
unsigned cumbits = rtree_levels[level].cumbits;
Replace rtree path cache with LRU cache. Rework rtree_ctx_t to encapsulate an rtree leaf LRU lookup cache rather than a single-path element lookup cache. The replacement is logically much simpler, as well as slightly faster in the fast path case and less prone to degraded performance during non-trivial sequences of lookups.
2017-02-04 12:21:56 +08:00
unsigned shiftbits = ptrbits - cumbits;
Determine rtree levels at compile time. Rather than dynamically building a table to aid per level computations, define a constant table at compile time. Omit both high and low insignificant bits. Use one to three tree levels, depending on the number of significant bits.
2017-02-07 05:17:12 +08:00
unsigned maskbits = rtree_levels[level].bits;
Fix rtree_subkey() regression. Fix rtree_subkey() to use uintptr_t rather than unsigned for key bitmasking. This regression was introduced by 4a346f55939af4f200121cc4454089592d952f18 (Replace rtree path cache with LRU cache.).
2017-02-10 04:31:11 +08:00
uintptr_t mask = (ZU(1) << maskbits) - 1;
Replace rtree path cache with LRU cache. Rework rtree_ctx_t to encapsulate an rtree leaf LRU lookup cache rather than a single-path element lookup cache. The replacement is logically much simpler, as well as slightly faster in the fast path case and less prone to degraded performance during non-trivial sequences of lookups.
2017-02-04 12:21:56 +08:00
return ((key >> shiftbits) & mask);
Refactor rtree to be lock-free. Recent huge allocation refactoring associates huge allocations with arenas, but it remains necessary to quickly look up huge allocation metadata during reallocation/deallocation. A global radix tree remains a good solution to this problem, but locking would have become the primary bottleneck after (upcoming) migration of chunk management from global to per arena data structures. This lock-free implementation uses double-checked reads to traverse the tree, so that in the steady state, each read or write requires only a single atomic operation. This implementation also assures that no more than two tree levels actually exist, through a combination of careful virtual memory allocation which makes large sparse nodes cheap, and skipping the root node on x64 (possible because the top 16 bits are all 0 in practice).
2015-01-31 14:54:08 +08:00
}
Port to Mac OS X. Add Mac OS X support, based in large part on the OS X support in Mozilla's version of jemalloc.
2010-09-06 01:35:13 +08:00
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
/*
* Atomic getters.
*
* dependent: Reading a value on behalf of a pointer to a valid allocation
* is guaranteed to be a clean read even without synchronization,
* because the rtree update became visible in memory before the
* pointer came into existence.
* !dependent: An arbitrary read, e.g. on behalf of ivsalloc(), may not be
* dependent on a previous rtree write, which means a stale read
* could result if synchronization were omitted here.
*/
Implement compact rtree leaf element representation. If a single virtual adddress pointer has enough unused bits to pack {szind_t, extent_t *, bool, bool}, use a single pointer-sized field in each rtree leaf element, rather than using three separate fields. This has little impact on access speed (fewer loads/stores, but more bit twiddling), except that denser representation increases TLB effectiveness.
2017-03-21 07:38:21 +08:00
# ifdef RTREE_LEAF_COMPACT
JEMALLOC_ALWAYS_INLINE uintptr_t
rtree_leaf_elm_bits_read(tsdn_t *tsdn, rtree_t *rtree, rtree_leaf_elm_t *elm,
bool acquired, bool dependent) {
if (config_debug && acquired) {
assert(dependent);
rtree_leaf_elm_witness_access(tsdn, rtree, elm);
}
return (uintptr_t)atomic_load_p(&elm->le_bits, dependent
? ATOMIC_RELAXED : ATOMIC_ACQUIRE);
}
JEMALLOC_ALWAYS_INLINE extent_t *
rtree_leaf_elm_bits_extent_get(uintptr_t bits) {
/* Restore sign-extended high bits, mask slab and lock bits. */
return (extent_t *)((uintptr_t)((intptr_t)(bits << RTREE_NHIB) >>
RTREE_NHIB) & ~((uintptr_t)0x3));
}
JEMALLOC_ALWAYS_INLINE szind_t
rtree_leaf_elm_bits_szind_get(uintptr_t bits) {
return (szind_t)(bits >> LG_VADDR);
}
JEMALLOC_ALWAYS_INLINE bool
rtree_leaf_elm_bits_slab_get(uintptr_t bits) {
return (bool)((bits >> 1) & (uintptr_t)0x1);
}
JEMALLOC_ALWAYS_INLINE bool
rtree_leaf_elm_bits_locked_get(uintptr_t bits) {
return (bool)(bits & (uintptr_t)0x1);
}
# endif
Rename extent_node_t to extent_t.
2016-03-24 12:09:28 +08:00
JEMALLOC_ALWAYS_INLINE extent_t *
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
rtree_leaf_elm_extent_read(tsdn_t *tsdn, rtree_t *rtree, rtree_leaf_elm_t *elm,
bool acquired, bool dependent) {
if (config_debug && acquired) {
assert(dependent);
rtree_leaf_elm_witness_access(tsdn, rtree, elm);
Avoid atomic operations for dependent rtree reads.
2015-05-16 08:02:30 +08:00
}
Add element acquire/release capabilities to rtree. This makes it possible to acquire short-term "ownership" of rtree elements so that it is possible to read an extent pointer *and* read the extent's contents with a guarantee that the element will not be modified until the ownership is released. This is intended as a mechanism for resolving rtree read/write races rather than as a way to lock extents.
2016-03-28 18:06:35 +08:00
Implement compact rtree leaf element representation. If a single virtual adddress pointer has enough unused bits to pack {szind_t, extent_t *, bool, bool}, use a single pointer-sized field in each rtree leaf element, rather than using three separate fields. This has little impact on access speed (fewer loads/stores, but more bit twiddling), except that denser representation increases TLB effectiveness.
2017-03-21 07:38:21 +08:00
#ifdef RTREE_LEAF_COMPACT
uintptr_t bits = rtree_leaf_elm_bits_read(tsdn, rtree, elm, acquired,
dependent);
assert(!acquired || rtree_leaf_elm_bits_locked_get(bits));
return rtree_leaf_elm_bits_extent_get(bits);
#else
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
extent_t *extent = (extent_t *)atomic_load_p(&elm->le_extent, dependent
? ATOMIC_RELAXED : ATOMIC_ACQUIRE);
assert(!acquired || ((uintptr_t)extent & (uintptr_t)0x1) ==
(uintptr_t)0x1);
/* Mask lock bit. */
Add element acquire/release capabilities to rtree. This makes it possible to acquire short-term "ownership" of rtree elements so that it is possible to read an extent pointer *and* read the extent's contents with a guarantee that the element will not be modified until the ownership is released. This is intended as a mechanism for resolving rtree read/write races rather than as a way to lock extents.
2016-03-28 18:06:35 +08:00
extent = (extent_t *)((uintptr_t)extent & ~((uintptr_t)0x1));
Remove extraneous parens around return arguments. This resolves #540.
2017-01-20 10:15:45 +08:00
return extent;
Implement compact rtree leaf element representation. If a single virtual adddress pointer has enough unused bits to pack {szind_t, extent_t *, bool, bool}, use a single pointer-sized field in each rtree leaf element, rather than using three separate fields. This has little impact on access speed (fewer loads/stores, but more bit twiddling), except that denser representation increases TLB effectiveness.
2017-03-21 07:38:21 +08:00
#endif
Refactor rtree to be lock-free. Recent huge allocation refactoring associates huge allocations with arenas, but it remains necessary to quickly look up huge allocation metadata during reallocation/deallocation. A global radix tree remains a good solution to this problem, but locking would have become the primary bottleneck after (upcoming) migration of chunk management from global to per arena data structures. This lock-free implementation uses double-checked reads to traverse the tree, so that in the steady state, each read or write requires only a single atomic operation. This implementation also assures that no more than two tree levels actually exist, through a combination of careful virtual memory allocation which makes large sparse nodes cheap, and skipping the root node on x64 (possible because the top 16 bits are all 0 in practice).
2015-01-31 14:54:08 +08:00
}
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
JEMALLOC_ALWAYS_INLINE szind_t
rtree_leaf_elm_szind_read(tsdn_t *tsdn, rtree_t *rtree, rtree_leaf_elm_t *elm,
bool acquired, bool dependent) {
if (config_debug && acquired) {
assert(dependent);
rtree_leaf_elm_witness_access(tsdn, rtree, elm);
}
Implement compact rtree leaf element representation. If a single virtual adddress pointer has enough unused bits to pack {szind_t, extent_t *, bool, bool}, use a single pointer-sized field in each rtree leaf element, rather than using three separate fields. This has little impact on access speed (fewer loads/stores, but more bit twiddling), except that denser representation increases TLB effectiveness.
2017-03-21 07:38:21 +08:00
#ifdef RTREE_LEAF_COMPACT
uintptr_t bits = rtree_leaf_elm_bits_read(tsdn, rtree, elm, acquired,
dependent);
assert(!acquired || rtree_leaf_elm_bits_locked_get(bits));
return rtree_leaf_elm_bits_szind_get(bits);
#else
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
return (szind_t)atomic_load_u(&elm->le_szind, dependent ? ATOMIC_RELAXED
: ATOMIC_ACQUIRE);
Implement compact rtree leaf element representation. If a single virtual adddress pointer has enough unused bits to pack {szind_t, extent_t *, bool, bool}, use a single pointer-sized field in each rtree leaf element, rather than using three separate fields. This has little impact on access speed (fewer loads/stores, but more bit twiddling), except that denser representation increases TLB effectiveness.
2017-03-21 07:38:21 +08:00
#endif
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
}
JEMALLOC_ALWAYS_INLINE bool
rtree_leaf_elm_slab_read(tsdn_t *tsdn, rtree_t *rtree, rtree_leaf_elm_t *elm,
bool acquired, bool dependent) {
if (config_debug && acquired) {
assert(dependent);
rtree_leaf_elm_witness_access(tsdn, rtree, elm);
}
Implement compact rtree leaf element representation. If a single virtual adddress pointer has enough unused bits to pack {szind_t, extent_t *, bool, bool}, use a single pointer-sized field in each rtree leaf element, rather than using three separate fields. This has little impact on access speed (fewer loads/stores, but more bit twiddling), except that denser representation increases TLB effectiveness.
2017-03-21 07:38:21 +08:00
#ifdef RTREE_LEAF_COMPACT
uintptr_t bits = rtree_leaf_elm_bits_read(tsdn, rtree, elm, acquired,
dependent);
assert(!acquired || rtree_leaf_elm_bits_locked_get(bits));
return rtree_leaf_elm_bits_slab_get(bits);
#else
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
return atomic_load_b(&elm->le_slab, dependent ? ATOMIC_RELAXED :
ATOMIC_ACQUIRE);
Implement compact rtree leaf element representation. If a single virtual adddress pointer has enough unused bits to pack {szind_t, extent_t *, bool, bool}, use a single pointer-sized field in each rtree leaf element, rather than using three separate fields. This has little impact on access speed (fewer loads/stores, but more bit twiddling), except that denser representation increases TLB effectiveness.
2017-03-21 07:38:21 +08:00
#endif
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
}
JEMALLOC_INLINE void
rtree_leaf_elm_extent_write(tsdn_t *tsdn, rtree_t *rtree, rtree_leaf_elm_t *elm,
bool acquired, extent_t *extent) {
if (config_debug && acquired) {
rtree_leaf_elm_witness_access(tsdn, rtree, elm);
}
assert(((uintptr_t)extent & (uintptr_t)0x1) == (uintptr_t)0x0);
Implement compact rtree leaf element representation. If a single virtual adddress pointer has enough unused bits to pack {szind_t, extent_t *, bool, bool}, use a single pointer-sized field in each rtree leaf element, rather than using three separate fields. This has little impact on access speed (fewer loads/stores, but more bit twiddling), except that denser representation increases TLB effectiveness.
2017-03-21 07:38:21 +08:00
#ifdef RTREE_LEAF_COMPACT
uintptr_t old_bits = rtree_leaf_elm_bits_read(tsdn, rtree, elm,
acquired, acquired);
uintptr_t bits = ((uintptr_t)rtree_leaf_elm_bits_szind_get(old_bits) <<
LG_VADDR) | ((uintptr_t)extent & (((uintptr_t)0x1 << LG_VADDR) - 1))
| ((uintptr_t)rtree_leaf_elm_bits_slab_get(old_bits) << 1) |
(uintptr_t)acquired;
atomic_store_p(&elm->le_bits, (void *)bits, ATOMIC_RELEASE);
#else
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
if (acquired) {
/* Overlay lock bit. */
extent = (extent_t *)((uintptr_t)extent | (uintptr_t)0x1);
}
atomic_store_p(&elm->le_extent, extent, ATOMIC_RELEASE);
Implement compact rtree leaf element representation. If a single virtual adddress pointer has enough unused bits to pack {szind_t, extent_t *, bool, bool}, use a single pointer-sized field in each rtree leaf element, rather than using three separate fields. This has little impact on access speed (fewer loads/stores, but more bit twiddling), except that denser representation increases TLB effectiveness.
2017-03-21 07:38:21 +08:00
#endif
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
}
Refactor rtree to be lock-free. Recent huge allocation refactoring associates huge allocations with arenas, but it remains necessary to quickly look up huge allocation metadata during reallocation/deallocation. A global radix tree remains a good solution to this problem, but locking would have become the primary bottleneck after (upcoming) migration of chunk management from global to per arena data structures. This lock-free implementation uses double-checked reads to traverse the tree, so that in the steady state, each read or write requires only a single atomic operation. This implementation also assures that no more than two tree levels actually exist, through a combination of careful virtual memory allocation which makes large sparse nodes cheap, and skipping the root node on x64 (possible because the top 16 bits are all 0 in practice).
2015-01-31 14:54:08 +08:00
JEMALLOC_INLINE void
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
rtree_leaf_elm_szind_write(tsdn_t *tsdn, rtree_t *rtree, rtree_leaf_elm_t *elm,
bool acquired, szind_t szind) {
if (config_debug && acquired) {
rtree_leaf_elm_witness_access(tsdn, rtree, elm);
}
assert(szind <= NSIZES);
Implement compact rtree leaf element representation. If a single virtual adddress pointer has enough unused bits to pack {szind_t, extent_t *, bool, bool}, use a single pointer-sized field in each rtree leaf element, rather than using three separate fields. This has little impact on access speed (fewer loads/stores, but more bit twiddling), except that denser representation increases TLB effectiveness.
2017-03-21 07:38:21 +08:00
#ifdef RTREE_LEAF_COMPACT
uintptr_t old_bits = rtree_leaf_elm_bits_read(tsdn, rtree, elm,
acquired, acquired);
uintptr_t bits = ((uintptr_t)szind << LG_VADDR) |
((uintptr_t)rtree_leaf_elm_bits_extent_get(old_bits) &
(((uintptr_t)0x1 << LG_VADDR) - 1)) |
((uintptr_t)rtree_leaf_elm_bits_slab_get(old_bits) << 1) |
(uintptr_t)acquired;
atomic_store_p(&elm->le_bits, (void *)bits, ATOMIC_RELEASE);
#else
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
atomic_store_u(&elm->le_szind, szind, ATOMIC_RELEASE);
Implement compact rtree leaf element representation. If a single virtual adddress pointer has enough unused bits to pack {szind_t, extent_t *, bool, bool}, use a single pointer-sized field in each rtree leaf element, rather than using three separate fields. This has little impact on access speed (fewer loads/stores, but more bit twiddling), except that denser representation increases TLB effectiveness.
2017-03-21 07:38:21 +08:00
#endif
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
}
JEMALLOC_INLINE void
rtree_leaf_elm_slab_write(tsdn_t *tsdn, rtree_t *rtree, rtree_leaf_elm_t *elm,
bool acquired, bool slab) {
if (config_debug && acquired) {
rtree_leaf_elm_witness_access(tsdn, rtree, elm);
}
Implement compact rtree leaf element representation. If a single virtual adddress pointer has enough unused bits to pack {szind_t, extent_t *, bool, bool}, use a single pointer-sized field in each rtree leaf element, rather than using three separate fields. This has little impact on access speed (fewer loads/stores, but more bit twiddling), except that denser representation increases TLB effectiveness.
2017-03-21 07:38:21 +08:00
#ifdef RTREE_LEAF_COMPACT
uintptr_t old_bits = rtree_leaf_elm_bits_read(tsdn, rtree, elm,
acquired, acquired);
uintptr_t bits = ((uintptr_t)rtree_leaf_elm_bits_szind_get(old_bits) <<
LG_VADDR) | ((uintptr_t)rtree_leaf_elm_bits_extent_get(old_bits) &
(((uintptr_t)0x1 << LG_VADDR) - 1)) | ((uintptr_t)slab << 1) |
(uintptr_t)acquired;
atomic_store_p(&elm->le_bits, (void *)bits, ATOMIC_RELEASE);
#else
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
atomic_store_b(&elm->le_slab, slab, ATOMIC_RELEASE);
Implement compact rtree leaf element representation. If a single virtual adddress pointer has enough unused bits to pack {szind_t, extent_t *, bool, bool}, use a single pointer-sized field in each rtree leaf element, rather than using three separate fields. This has little impact on access speed (fewer loads/stores, but more bit twiddling), except that denser representation increases TLB effectiveness.
2017-03-21 07:38:21 +08:00
#endif
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
}
JEMALLOC_INLINE void
rtree_leaf_elm_write(tsdn_t *tsdn, rtree_t *rtree, rtree_leaf_elm_t *elm,
bool acquired, extent_t *extent, szind_t szind, bool slab) {
Implement compact rtree leaf element representation. If a single virtual adddress pointer has enough unused bits to pack {szind_t, extent_t *, bool, bool}, use a single pointer-sized field in each rtree leaf element, rather than using three separate fields. This has little impact on access speed (fewer loads/stores, but more bit twiddling), except that denser representation increases TLB effectiveness.
2017-03-21 07:38:21 +08:00
if (config_debug && acquired) {
rtree_leaf_elm_witness_access(tsdn, rtree, elm);
}
assert(!slab || szind < NBINS);
Fix a race in rtree_szind_slab_update() for RTREE_LEAF_COMPACT.
2017-03-27 19:08:51 +08:00
#ifdef RTREE_LEAF_COMPACT
Implement compact rtree leaf element representation. If a single virtual adddress pointer has enough unused bits to pack {szind_t, extent_t *, bool, bool}, use a single pointer-sized field in each rtree leaf element, rather than using three separate fields. This has little impact on access speed (fewer loads/stores, but more bit twiddling), except that denser representation increases TLB effectiveness.
2017-03-21 07:38:21 +08:00
uintptr_t bits = ((uintptr_t)szind << LG_VADDR) |
((uintptr_t)extent & (((uintptr_t)0x1 << LG_VADDR) - 1)) |
((uintptr_t)slab << 1) |
(uintptr_t)acquired;
atomic_store_p(&elm->le_bits, (void *)bits, ATOMIC_RELEASE);
#else
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
rtree_leaf_elm_slab_write(tsdn, rtree, elm, acquired, slab);
rtree_leaf_elm_szind_write(tsdn, rtree, elm, acquired, szind);
/*
* Write extent last, since the element is atomically considered valid
* as soon as the extent field is non-NULL.
*/
rtree_leaf_elm_extent_write(tsdn, rtree, elm, acquired, extent);
Implement compact rtree leaf element representation. If a single virtual adddress pointer has enough unused bits to pack {szind_t, extent_t *, bool, bool}, use a single pointer-sized field in each rtree leaf element, rather than using three separate fields. This has little impact on access speed (fewer loads/stores, but more bit twiddling), except that denser representation increases TLB effectiveness.
2017-03-21 07:38:21 +08:00
#endif
Refactor rtree to be lock-free. Recent huge allocation refactoring associates huge allocations with arenas, but it remains necessary to quickly look up huge allocation metadata during reallocation/deallocation. A global radix tree remains a good solution to this problem, but locking would have become the primary bottleneck after (upcoming) migration of chunk management from global to per arena data structures. This lock-free implementation uses double-checked reads to traverse the tree, so that in the steady state, each read or write requires only a single atomic operation. This implementation also assures that no more than two tree levels actually exist, through a combination of careful virtual memory allocation which makes large sparse nodes cheap, and skipping the root node on x64 (possible because the top 16 bits are all 0 in practice).
2015-01-31 14:54:08 +08:00
}
Fix a race in rtree_szind_slab_update() for RTREE_LEAF_COMPACT.
2017-03-27 19:08:51 +08:00
JEMALLOC_INLINE void
rtree_leaf_elm_szind_slab_update(tsdn_t *tsdn, rtree_t *rtree,
rtree_leaf_elm_t *elm, szind_t szind, bool slab) {
assert(!slab || szind < NBINS);
/*
* The caller implicitly assures that it is the only writer to the szind
* and slab fields, and that the extent field cannot currently change.
*/
#ifdef RTREE_LEAF_COMPACT
/*
* Another thread may concurrently acquire the elm, which means that
* even though the szind and slab fields will not be concurrently
* modified by another thread, the fact that the lock is embedded in the
* same word requires that a CAS operation be used here.
*/
uintptr_t old_bits = rtree_leaf_elm_bits_read(tsdn, rtree, elm, false,
true) & ~((uintptr_t)0x1); /* Mask lock bit. */
uintptr_t bits = ((uintptr_t)szind << LG_VADDR) |
((uintptr_t)rtree_leaf_elm_bits_extent_get(old_bits) &
(((uintptr_t)0x1 << LG_VADDR) - 1)) |
((uintptr_t)slab << 1);
spin_t spinner = SPIN_INITIALIZER;
while (true) {
if (likely(atomic_compare_exchange_strong_p(&elm->le_bits,
(void **)&old_bits, (void *)bits, ATOMIC_ACQUIRE,
ATOMIC_RELAXED))) {
break;
}
spin_adaptive(&spinner);
}
#else
/* No need to lock. */
rtree_leaf_elm_slab_write(tsdn, rtree, elm, false, slab);
rtree_leaf_elm_szind_write(tsdn, rtree, elm, false, szind);
#endif
}
Split rtree_elm_t into rtree_{node,leaf}_elm_t. This allows leaf elements to differ in size from internal node elements. In principle it would be more correct to use a different type for each level of the tree, but due to implementation details related to atomic operations, we use casts anyway, thus counteracting the value of additional type correctness. Furthermore, such a scheme would require function code generation (via cpp macros), as well as either unwieldy type names for leaves or type aliases, e.g. typedef struct rtree_elm_d2_s rtree_leaf_elm_t; This alternate strategy would be more correct, and with less code duplication, but probably not worth the complexity.
2017-03-17 00:46:42 +08:00
JEMALLOC_ALWAYS_INLINE rtree_leaf_elm_t *
rtree_leaf_elm_lookup(tsdn_t *tsdn, rtree_t *rtree, rtree_ctx_t *rtree_ctx,
Update brace style. Add braces around single-line blocks, and remove line breaks before function-opening braces. This resolves #537.
2017-01-16 08:56:30 +08:00
uintptr_t key, bool dependent, bool init_missing) {
Remove rtree support for 0 (NULL) keys. NULL can never actually be inserted in practice, and removing support allows a branch to be removed from the fast path.
2017-02-05 18:50:59 +08:00
assert(key != 0);
Add element acquire/release capabilities to rtree. This makes it possible to acquire short-term "ownership" of rtree elements so that it is possible to read an extent pointer *and* read the extent's contents with a guarantee that the element will not be modified until the ownership is released. This is intended as a mechanism for resolving rtree read/write races rather than as a way to lock extents.
2016-03-28 18:06:35 +08:00
assert(!dependent || !init_missing);
Refactor rtree to be lock-free. Recent huge allocation refactoring associates huge allocations with arenas, but it remains necessary to quickly look up huge allocation metadata during reallocation/deallocation. A global radix tree remains a good solution to this problem, but locking would have become the primary bottleneck after (upcoming) migration of chunk management from global to per arena data structures. This lock-free implementation uses double-checked reads to traverse the tree, so that in the steady state, each read or write requires only a single atomic operation. This implementation also assures that no more than two tree levels actually exist, through a combination of careful virtual memory allocation which makes large sparse nodes cheap, and skipping the root node on x64 (possible because the top 16 bits are all 0 in practice).
2015-01-31 14:54:08 +08:00
Remove rtree support for 0 (NULL) keys. NULL can never actually be inserted in practice, and removing support allows a branch to be removed from the fast path.
2017-02-05 18:50:59 +08:00
uintptr_t leafkey = rtree_leafkey(key);
Replace rtree path cache with LRU cache. Rework rtree_ctx_t to encapsulate an rtree leaf LRU lookup cache rather than a single-path element lookup cache. The replacement is logically much simpler, as well as slightly faster in the fast path case and less prone to degraded performance during non-trivial sequences of lookups.
2017-02-04 12:21:56 +08:00
#define RTREE_CACHE_CHECK(i) do { \
Remove rtree support for 0 (NULL) keys. NULL can never actually be inserted in practice, and removing support allows a branch to be removed from the fast path.
2017-02-05 18:50:59 +08:00
if (likely(rtree_ctx->cache[i].leafkey == leafkey)) { \
Split rtree_elm_t into rtree_{node,leaf}_elm_t. This allows leaf elements to differ in size from internal node elements. In principle it would be more correct to use a different type for each level of the tree, but due to implementation details related to atomic operations, we use casts anyway, thus counteracting the value of additional type correctness. Furthermore, such a scheme would require function code generation (via cpp macros), as well as either unwieldy type names for leaves or type aliases, e.g. typedef struct rtree_elm_d2_s rtree_leaf_elm_t; This alternate strategy would be more correct, and with less code duplication, but probably not worth the complexity.
2017-03-17 00:46:42 +08:00
rtree_leaf_elm_t *leaf = rtree_ctx->cache[i].leaf; \
Remove the leafkey NULL check in leaf_elm_lookup.
2017-03-29 08:14:43 +08:00
assert(leaf != NULL); \
if (i > 0) { \
Simplify rtree cache replacement policy. To avoid memmove on free() fast path, simplify the cache replacement policy to only bubble up the cache hit element by 1.
2017-03-28 02:48:39 +08:00
/* Bubble up by one. */ \
Remove the leafkey NULL check in leaf_elm_lookup.
2017-03-29 08:14:43 +08:00
rtree_ctx->cache[i] = rtree_ctx->cache[i - 1]; \
rtree_ctx->cache[i - 1].leafkey = leafkey; \
rtree_ctx->cache[i - 1].leaf = leaf; \
Replace rtree path cache with LRU cache. Rework rtree_ctx_t to encapsulate an rtree leaf LRU lookup cache rather than a single-path element lookup cache. The replacement is logically much simpler, as well as slightly faster in the fast path case and less prone to degraded performance during non-trivial sequences of lookups.
2017-02-04 12:21:56 +08:00
} \
Remove the leafkey NULL check in leaf_elm_lookup.
2017-03-29 08:14:43 +08:00
uintptr_t subkey = rtree_subkey(key, RTREE_HEIGHT-1); \
return &leaf[subkey]; \
Remove rtree support for 0 (NULL) keys. NULL can never actually be inserted in practice, and removing support allows a branch to be removed from the fast path.
2017-02-05 18:50:59 +08:00
} \
Replace rtree path cache with LRU cache. Rework rtree_ctx_t to encapsulate an rtree leaf LRU lookup cache rather than a single-path element lookup cache. The replacement is logically much simpler, as well as slightly faster in the fast path case and less prone to degraded performance during non-trivial sequences of lookups.
2017-02-04 12:21:56 +08:00
} while (0)
Simplify rtree cache replacement policy. To avoid memmove on free() fast path, simplify the cache replacement policy to only bubble up the cache hit element by 1.
2017-03-28 02:48:39 +08:00
/* Check the first cache entry. */
Remove rtree support for 0 (NULL) keys. NULL can never actually be inserted in practice, and removing support allows a branch to be removed from the fast path.
2017-02-05 18:50:59 +08:00
RTREE_CACHE_CHECK(0);
/*
* Search the remaining cache elements, and on success move the matching
Simplify rtree cache replacement policy. To avoid memmove on free() fast path, simplify the cache replacement policy to only bubble up the cache hit element by 1.
2017-03-28 02:48:39 +08:00
* element up by one slot.
Remove rtree support for 0 (NULL) keys. NULL can never actually be inserted in practice, and removing support allows a branch to be removed from the fast path.
2017-02-05 18:50:59 +08:00
*/
Simplify rtree cache replacement policy. To avoid memmove on free() fast path, simplify the cache replacement policy to only bubble up the cache hit element by 1.
2017-03-28 02:48:39 +08:00
for (unsigned i = 1; i < RTREE_CTX_NCACHE; i++) {
Remove rtree support for 0 (NULL) keys. NULL can never actually be inserted in practice, and removing support allows a branch to be removed from the fast path.
2017-02-05 18:50:59 +08:00
RTREE_CACHE_CHECK(i);
}
#undef RTREE_CACHE_CHECK
Refactor rtree to be lock-free. Recent huge allocation refactoring associates huge allocations with arenas, but it remains necessary to quickly look up huge allocation metadata during reallocation/deallocation. A global radix tree remains a good solution to this problem, but locking would have become the primary bottleneck after (upcoming) migration of chunk management from global to per arena data structures. This lock-free implementation uses double-checked reads to traverse the tree, so that in the steady state, each read or write requires only a single atomic operation. This implementation also assures that no more than two tree levels actually exist, through a combination of careful virtual memory allocation which makes large sparse nodes cheap, and skipping the root node on x64 (possible because the top 16 bits are all 0 in practice).
2015-01-31 14:54:08 +08:00
Split rtree_elm_t into rtree_{node,leaf}_elm_t. This allows leaf elements to differ in size from internal node elements. In principle it would be more correct to use a different type for each level of the tree, but due to implementation details related to atomic operations, we use casts anyway, thus counteracting the value of additional type correctness. Furthermore, such a scheme would require function code generation (via cpp macros), as well as either unwieldy type names for leaves or type aliases, e.g. typedef struct rtree_elm_d2_s rtree_leaf_elm_t; This alternate strategy would be more correct, and with less code duplication, but probably not worth the complexity.
2017-03-17 00:46:42 +08:00
return rtree_leaf_elm_lookup_hard(tsdn, rtree, rtree_ctx, key,
dependent, init_missing);
Refactor rtree to be lock-free. Recent huge allocation refactoring associates huge allocations with arenas, but it remains necessary to quickly look up huge allocation metadata during reallocation/deallocation. A global radix tree remains a good solution to this problem, but locking would have become the primary bottleneck after (upcoming) migration of chunk management from global to per arena data structures. This lock-free implementation uses double-checked reads to traverse the tree, so that in the steady state, each read or write requires only a single atomic operation. This implementation also assures that no more than two tree levels actually exist, through a combination of careful virtual memory allocation which makes large sparse nodes cheap, and skipping the root node on x64 (possible because the top 16 bits are all 0 in practice).
2015-01-31 14:54:08 +08:00
}
Port to Mac OS X. Add Mac OS X support, based in large part on the OS X support in Mozilla's version of jemalloc.
2010-09-06 01:35:13 +08:00
Refactor rtree to be lock-free. Recent huge allocation refactoring associates huge allocations with arenas, but it remains necessary to quickly look up huge allocation metadata during reallocation/deallocation. A global radix tree remains a good solution to this problem, but locking would have become the primary bottleneck after (upcoming) migration of chunk management from global to per arena data structures. This lock-free implementation uses double-checked reads to traverse the tree, so that in the steady state, each read or write requires only a single atomic operation. This implementation also assures that no more than two tree levels actually exist, through a combination of careful virtual memory allocation which makes large sparse nodes cheap, and skipping the root node on x64 (possible because the top 16 bits are all 0 in practice).
2015-01-31 14:54:08 +08:00
JEMALLOC_INLINE bool
Add rtree lookup path caching. rtree-based extent lookups remain more expensive than chunk-based run lookups, but with this optimization the fast path slowdown is ~3 CPU cycles per metadata lookup (on Intel Core i7-4980HQ), versus ~11 cycles prior. The path caching speedup tends to degrade gracefully unless allocated memory is spread far apart (as is the case when using a mixture of sbrk() and mmap()).
2016-06-03 09:43:10 +08:00
rtree_write(tsdn_t *tsdn, rtree_t *rtree, rtree_ctx_t *rtree_ctx, uintptr_t key,
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
extent_t *extent, szind_t szind, bool slab) {
/* Use rtree_clear() to set the extent to NULL. */
assert(extent != NULL);
Port to Mac OS X. Add Mac OS X support, based in large part on the OS X support in Mozilla's version of jemalloc.
2010-09-06 01:35:13 +08:00
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
rtree_leaf_elm_t *elm = rtree_leaf_elm_lookup(tsdn, rtree, rtree_ctx,
key, false, true);
Update brace style. Add braces around single-line blocks, and remove line breaks before function-opening braces. This resolves #537.
2017-01-16 08:56:30 +08:00
if (elm == NULL) {
Remove extraneous parens around return arguments. This resolves #540.
2017-01-20 10:15:45 +08:00
return true;
Update brace style. Add braces around single-line blocks, and remove line breaks before function-opening braces. This resolves #537.
2017-01-16 08:56:30 +08:00
}
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
assert(rtree_leaf_elm_extent_read(tsdn, rtree, elm, false, false) ==
NULL);
rtree_leaf_elm_write(tsdn, rtree, elm, false, extent, szind, slab);
Add element acquire/release capabilities to rtree. This makes it possible to acquire short-term "ownership" of rtree elements so that it is possible to read an extent pointer *and* read the extent's contents with a guarantee that the element will not be modified until the ownership is released. This is intended as a mechanism for resolving rtree read/write races rather than as a way to lock extents.
2016-03-28 18:06:35 +08:00
Remove extraneous parens around return arguments. This resolves #540.
2017-01-20 10:15:45 +08:00
return false;
Add element acquire/release capabilities to rtree. This makes it possible to acquire short-term "ownership" of rtree elements so that it is possible to read an extent pointer *and* read the extent's contents with a guarantee that the element will not be modified until the ownership is released. This is intended as a mechanism for resolving rtree read/write races rather than as a way to lock extents.
2016-03-28 18:06:35 +08:00
}
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
JEMALLOC_ALWAYS_INLINE rtree_leaf_elm_t *
Add rtree lookup path caching. rtree-based extent lookups remain more expensive than chunk-based run lookups, but with this optimization the fast path slowdown is ~3 CPU cycles per metadata lookup (on Intel Core i7-4980HQ), versus ~11 cycles prior. The path caching speedup tends to degrade gracefully unless allocated memory is spread far apart (as is the case when using a mixture of sbrk() and mmap()).
2016-06-03 09:43:10 +08:00
rtree_read(tsdn_t *tsdn, rtree_t *rtree, rtree_ctx_t *rtree_ctx, uintptr_t key,
Update brace style. Add braces around single-line blocks, and remove line breaks before function-opening braces. This resolves #537.
2017-01-16 08:56:30 +08:00
bool dependent) {
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
rtree_leaf_elm_t *elm = rtree_leaf_elm_lookup(tsdn, rtree, rtree_ctx,
key, dependent, false);
if (!dependent && elm == NULL) {
return NULL;
}
assert(elm != NULL);
return elm;
}
Add element acquire/release capabilities to rtree. This makes it possible to acquire short-term "ownership" of rtree elements so that it is possible to read an extent pointer *and* read the extent's contents with a guarantee that the element will not be modified until the ownership is released. This is intended as a mechanism for resolving rtree read/write races rather than as a way to lock extents.
2016-03-28 18:06:35 +08:00
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
JEMALLOC_ALWAYS_INLINE extent_t *
rtree_extent_read(tsdn_t *tsdn, rtree_t *rtree, rtree_ctx_t *rtree_ctx,
uintptr_t key, bool dependent) {
rtree_leaf_elm_t *elm = rtree_read(tsdn, rtree, rtree_ctx, key,
dependent);
Optimize a branch out of rtree_read() if !dependent.
2017-02-04 12:17:47 +08:00
if (!dependent && elm == NULL) {
Remove extraneous parens around return arguments. This resolves #540.
2017-01-20 10:15:45 +08:00
return NULL;
Update brace style. Add braces around single-line blocks, and remove line breaks before function-opening braces. This resolves #537.
2017-01-16 08:56:30 +08:00
}
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
return rtree_leaf_elm_extent_read(tsdn, rtree, elm, false, dependent);
}
JEMALLOC_ALWAYS_INLINE szind_t
rtree_szind_read(tsdn_t *tsdn, rtree_t *rtree, rtree_ctx_t *rtree_ctx,
uintptr_t key, bool dependent) {
rtree_leaf_elm_t *elm = rtree_read(tsdn, rtree, rtree_ctx, key,
dependent);
if (!dependent && elm == NULL) {
return NSIZES;
}
return rtree_leaf_elm_szind_read(tsdn, rtree, elm, false, dependent);
}
/*
* rtree_slab_read() is intentionally omitted because slab is always read in
* conjunction with szind, which makes rtree_szind_slab_read() a better choice.
*/
JEMALLOC_ALWAYS_INLINE bool
rtree_extent_szind_read(tsdn_t *tsdn, rtree_t *rtree, rtree_ctx_t *rtree_ctx,
uintptr_t key, bool dependent, extent_t **r_extent, szind_t *r_szind) {
rtree_leaf_elm_t *elm = rtree_read(tsdn, rtree, rtree_ctx, key,
dependent);
if (!dependent && elm == NULL) {
return true;
}
*r_extent = rtree_leaf_elm_extent_read(tsdn, rtree, elm, false,
dependent);
*r_szind = rtree_leaf_elm_szind_read(tsdn, rtree, elm, false,
dependent);
return false;
}
Add element acquire/release capabilities to rtree. This makes it possible to acquire short-term "ownership" of rtree elements so that it is possible to read an extent pointer *and* read the extent's contents with a guarantee that the element will not be modified until the ownership is released. This is intended as a mechanism for resolving rtree read/write races rather than as a way to lock extents.
2016-03-28 18:06:35 +08:00
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
JEMALLOC_ALWAYS_INLINE bool
rtree_szind_slab_read(tsdn_t *tsdn, rtree_t *rtree, rtree_ctx_t *rtree_ctx,
uintptr_t key, bool dependent, szind_t *r_szind, bool *r_slab) {
rtree_leaf_elm_t *elm = rtree_read(tsdn, rtree, rtree_ctx, key,
dependent);
if (!dependent && elm == NULL) {
return true;
}
*r_szind = rtree_leaf_elm_szind_read(tsdn, rtree, elm, false,
dependent);
*r_slab = rtree_leaf_elm_slab_read(tsdn, rtree, elm, false, dependent);
return false;
}
Split rtree_elm_t into rtree_{node,leaf}_elm_t. This allows leaf elements to differ in size from internal node elements. In principle it would be more correct to use a different type for each level of the tree, but due to implementation details related to atomic operations, we use casts anyway, thus counteracting the value of additional type correctness. Furthermore, such a scheme would require function code generation (via cpp macros), as well as either unwieldy type names for leaves or type aliases, e.g. typedef struct rtree_elm_d2_s rtree_leaf_elm_t; This alternate strategy would be more correct, and with less code duplication, but probably not worth the complexity.
2017-03-17 00:46:42 +08:00
JEMALLOC_INLINE rtree_leaf_elm_t *
rtree_leaf_elm_acquire(tsdn_t *tsdn, rtree_t *rtree, rtree_ctx_t *rtree_ctx,
Update brace style. Add braces around single-line blocks, and remove line breaks before function-opening braces. This resolves #537.
2017-01-16 08:56:30 +08:00
uintptr_t key, bool dependent, bool init_missing) {
Split rtree_elm_t into rtree_{node,leaf}_elm_t. This allows leaf elements to differ in size from internal node elements. In principle it would be more correct to use a different type for each level of the tree, but due to implementation details related to atomic operations, we use casts anyway, thus counteracting the value of additional type correctness. Furthermore, such a scheme would require function code generation (via cpp macros), as well as either unwieldy type names for leaves or type aliases, e.g. typedef struct rtree_elm_d2_s rtree_leaf_elm_t; This alternate strategy would be more correct, and with less code duplication, but probably not worth the complexity.
2017-03-17 00:46:42 +08:00
rtree_leaf_elm_t *elm = rtree_leaf_elm_lookup(tsdn, rtree, rtree_ctx,
key, dependent, init_missing);
Update brace style. Add braces around single-line blocks, and remove line breaks before function-opening braces. This resolves #537.
2017-01-16 08:56:30 +08:00
if (!dependent && elm == NULL) {
Remove extraneous parens around return arguments. This resolves #540.
2017-01-20 10:15:45 +08:00
return NULL;
Update brace style. Add braces around single-line blocks, and remove line breaks before function-opening braces. This resolves #537.
2017-01-16 08:56:30 +08:00
}
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
assert(elm != NULL);
Replace rtree path cache with LRU cache. Rework rtree_ctx_t to encapsulate an rtree leaf LRU lookup cache rather than a single-path element lookup cache. The replacement is logically much simpler, as well as slightly faster in the fast path case and less prone to degraded performance during non-trivial sequences of lookups.
2017-02-04 12:21:56 +08:00
Spin adaptively in rtree_elm_acquire().
2017-02-06 15:58:02 +08:00
spin_t spinner = SPIN_INITIALIZER;
while (true) {
Replace rtree path cache with LRU cache. Rework rtree_ctx_t to encapsulate an rtree leaf LRU lookup cache rather than a single-path element lookup cache. The replacement is logically much simpler, as well as slightly faster in the fast path case and less prone to degraded performance during non-trivial sequences of lookups.
2017-02-04 12:21:56 +08:00
/* The least significant bit serves as a lock. */
Implement compact rtree leaf element representation. If a single virtual adddress pointer has enough unused bits to pack {szind_t, extent_t *, bool, bool}, use a single pointer-sized field in each rtree leaf element, rather than using three separate fields. This has little impact on access speed (fewer loads/stores, but more bit twiddling), except that denser representation increases TLB effectiveness.
2017-03-21 07:38:21 +08:00
#ifdef RTREE_LEAF_COMPACT
# define RTREE_FIELD_WITH_LOCK le_bits
#else
# define RTREE_FIELD_WITH_LOCK le_extent
#endif
void *bits = atomic_load_p(&elm->RTREE_FIELD_WITH_LOCK,
Convert rtree code to use C11 atomics In the process, I changed the implementation of rtree_elm_acquire so that it won't even try to CAS if its initial read (getting the extent + lock bit) indicates that the CAS is doomed to fail. This can significantly improve performance under contention.
2017-03-10 06:49:32 +08:00
ATOMIC_RELAXED);
Implement compact rtree leaf element representation. If a single virtual adddress pointer has enough unused bits to pack {szind_t, extent_t *, bool, bool}, use a single pointer-sized field in each rtree leaf element, rather than using three separate fields. This has little impact on access speed (fewer loads/stores, but more bit twiddling), except that denser representation increases TLB effectiveness.
2017-03-21 07:38:21 +08:00
if (likely(((uintptr_t)bits & (uintptr_t)0x1) == 0)) {
void *locked = (void *)((uintptr_t)bits |
(uintptr_t)0x1);
Convert rtree code to use C11 atomics In the process, I changed the implementation of rtree_elm_acquire so that it won't even try to CAS if its initial read (getting the extent + lock bit) indicates that the CAS is doomed to fail. This can significantly improve performance under contention.
2017-03-10 06:49:32 +08:00
if (likely(atomic_compare_exchange_strong_p(
Implement compact rtree leaf element representation. If a single virtual adddress pointer has enough unused bits to pack {szind_t, extent_t *, bool, bool}, use a single pointer-sized field in each rtree leaf element, rather than using three separate fields. This has little impact on access speed (fewer loads/stores, but more bit twiddling), except that denser representation increases TLB effectiveness.
2017-03-21 07:38:21 +08:00
&elm->RTREE_FIELD_WITH_LOCK, &bits, locked,
Convert rtree code to use C11 atomics In the process, I changed the implementation of rtree_elm_acquire so that it won't even try to CAS if its initial read (getting the extent + lock bit) indicates that the CAS is doomed to fail. This can significantly improve performance under contention.
2017-03-10 06:49:32 +08:00
ATOMIC_ACQUIRE, ATOMIC_RELAXED))) {
break;
}
Spin adaptively in rtree_elm_acquire().
2017-02-06 15:58:02 +08:00
}
spin_adaptive(&spinner);
Implement compact rtree leaf element representation. If a single virtual adddress pointer has enough unused bits to pack {szind_t, extent_t *, bool, bool}, use a single pointer-sized field in each rtree leaf element, rather than using three separate fields. This has little impact on access speed (fewer loads/stores, but more bit twiddling), except that denser representation increases TLB effectiveness.
2017-03-21 07:38:21 +08:00
#undef RTREE_FIELD_WITH_LOCK
Spin adaptively in rtree_elm_acquire().
2017-02-06 15:58:02 +08:00
}
Add element acquire/release capabilities to rtree. This makes it possible to acquire short-term "ownership" of rtree elements so that it is possible to read an extent pointer *and* read the extent's contents with a guarantee that the element will not be modified until the ownership is released. This is intended as a mechanism for resolving rtree read/write races rather than as a way to lock extents.
2016-03-28 18:06:35 +08:00
Update brace style. Add braces around single-line blocks, and remove line breaks before function-opening braces. This resolves #537.
2017-01-16 08:56:30 +08:00
if (config_debug) {
Split rtree_elm_t into rtree_{node,leaf}_elm_t. This allows leaf elements to differ in size from internal node elements. In principle it would be more correct to use a different type for each level of the tree, but due to implementation details related to atomic operations, we use casts anyway, thus counteracting the value of additional type correctness. Furthermore, such a scheme would require function code generation (via cpp macros), as well as either unwieldy type names for leaves or type aliases, e.g. typedef struct rtree_elm_d2_s rtree_leaf_elm_t; This alternate strategy would be more correct, and with less code duplication, but probably not worth the complexity.
2017-03-17 00:46:42 +08:00
rtree_leaf_elm_witness_acquire(tsdn, rtree, key, elm);
Update brace style. Add braces around single-line blocks, and remove line breaks before function-opening braces. This resolves #537.
2017-01-16 08:56:30 +08:00
}
Add rtree element witnesses.
2016-04-18 03:55:10 +08:00
Remove extraneous parens around return arguments. This resolves #540.
2017-01-20 10:15:45 +08:00
return elm;
Add element acquire/release capabilities to rtree. This makes it possible to acquire short-term "ownership" of rtree elements so that it is possible to read an extent pointer *and* read the extent's contents with a guarantee that the element will not be modified until the ownership is released. This is intended as a mechanism for resolving rtree read/write races rather than as a way to lock extents.
2016-03-28 18:06:35 +08:00
}
JEMALLOC_INLINE void
Incorporate szind/slab into rtree leaves. Expand and restructure the rtree API such that all common operations can be achieved with minimal work, regardless of whether the rtree leaf fields are independent versus packed into a single atomic pointer.
2017-03-17 08:57:52 +08:00
rtree_leaf_elm_release(tsdn_t *tsdn, rtree_t *rtree, rtree_leaf_elm_t *elm) {
extent_t *extent = rtree_leaf_elm_extent_read(tsdn, rtree, elm, true,
true);
rtree_leaf_elm_extent_write(tsdn, rtree, elm, false, extent);
Update brace style. Add braces around single-line blocks, and remove line breaks before function-opening braces. This resolves #537.
2017-01-16 08:56:30 +08:00
if (config_debug) {
Split rtree_elm_t into rtree_{node,leaf}_elm_t. This allows leaf elements to differ in size from internal node elements. In principle it would be more correct to use a different type for each level of the tree, but due to implementation details related to atomic operations, we use casts anyway, thus counteracting the value of additional type correctness. Furthermore, such a scheme would require function code generation (via cpp macros), as well as either unwieldy type names for leaves or type aliases, e.g. typedef struct rtree_elm_d2_s rtree_leaf_elm_t; This alternate strategy would be more correct, and with less code duplication, but probably not worth the complexity.
2017-03-17 00:46:42 +08:00
rtree_leaf_elm_witness_release(tsdn, rtree, elm);
Update brace style. Add braces around single-line blocks, and remove line breaks before function-opening braces. This resolves #537.
2017-01-16 08:56:30 +08:00
}
Add element acquire/release capabilities to rtree. This makes it possible to acquire short-term "ownership" of rtree elements so that it is possible to read an extent pointer *and* read the extent's contents with a guarantee that the element will not be modified until the ownership is released. This is intended as a mechanism for resolving rtree read/write races rather than as a way to lock extents.
2016-03-28 18:06:35 +08:00
}
Fix a race in rtree_szind_slab_update() for RTREE_LEAF_COMPACT.
2017-03-27 19:08:51 +08:00
JEMALLOC_INLINE void
rtree_szind_slab_update(tsdn_t *tsdn, rtree_t *rtree, rtree_ctx_t *rtree_ctx,
uintptr_t key, szind_t szind, bool slab) {
assert(!slab || szind < NBINS);
rtree_leaf_elm_t *elm = rtree_read(tsdn, rtree, rtree_ctx, key, true);
rtree_leaf_elm_szind_slab_update(tsdn, rtree, elm, szind, slab);
}
Add element acquire/release capabilities to rtree. This makes it possible to acquire short-term "ownership" of rtree elements so that it is possible to read an extent pointer *and* read the extent's contents with a guarantee that the element will not be modified until the ownership is released. This is intended as a mechanism for resolving rtree read/write races rather than as a way to lock extents.
2016-03-28 18:06:35 +08:00
JEMALLOC_INLINE void
Update brace style. Add braces around single-line blocks, and remove line breaks before function-opening braces. This resolves #537.
2017-01-16 08:56:30 +08:00
rtree_clear(tsdn_t *tsdn, rtree_t *rtree, rtree_ctx_t *rtree_ctx,
uintptr_t key) {
Simplify rtree_clear() to avoid locking.
2017-03-27 19:10:11 +08:00
rtree_leaf_elm_t *elm = rtree_read(tsdn, rtree, rtree_ctx, key, true);
assert(rtree_leaf_elm_extent_read(tsdn, rtree, elm, false, false) !=
NULL);
rtree_leaf_elm_write(tsdn, rtree, elm, false, NULL, NSIZES, false);
Port to Mac OS X. Add Mac OS X support, based in large part on the OS X support in Mozilla's version of jemalloc.
2010-09-06 01:35:13 +08:00
}
#endif
Break up headers into constituent parts This is part of a broader change to make header files better represent the dependencies between one another (see https://github.com/jemalloc/jemalloc/issues/533). It breaks up component headers into smaller parts that can be made to have a simpler dependency graph. For the autogenerated headers (smoothstep.h and size_classes.h), no splitting was necessary, so I didn't add support to emit multiple headers.
2017-01-11 10:06:31 +08:00
#endif /* JEMALLOC_INTERNAL_RTREE_INLINES_H */
Reference in New Issue Copy Permalink