Aggressively optimise fast path for allocation

This change introduces a per small sizeclass free list.  That can be
used to access the free objects for that sizeclass with minimal
calculations being required.

It changes to a partial bump ptr.  We bump allocate a whole OS
page worth of objects at a go, so we don't switch as frequently
between bump and free list allocation.

The code for the fast paths has been restructured to minimise the
work required on the common case, and also it is all inlined for the
common case.

Allocating a zero sized object is moved off the fast path.  Ask for 1
byte if you want to be fast.
This commit is contained in:
Matthew Parkinson
2019-06-28 11:50:17 +01:00
parent b0c1531221
commit 7a198cbda5
7 changed files with 299 additions and 191 deletions

View File

@@ -23,10 +23,8 @@ namespace snmalloc
ModArray<NUM_SIZECLASSES, size_t> size;
ModArray<NUM_SIZECLASSES, size_t> cache_friendly_mask;
ModArray<NUM_SIZECLASSES, size_t> inverse_cache_friendly_mask;
ModArray<NUM_SMALL_CLASSES, uint16_t> bump_ptr_start;
ModArray<NUM_SMALL_CLASSES, uint16_t> short_bump_ptr_start;
ModArray<NUM_SMALL_CLASSES, uint16_t> initial_link_ptr;
ModArray<NUM_SMALL_CLASSES, uint16_t> short_initial_link_ptr;
ModArray<NUM_SMALL_CLASSES, uint16_t> initial_offset_ptr;
ModArray<NUM_SMALL_CLASSES, uint16_t> short_initial_offset_ptr;
ModArray<NUM_MEDIUM_CLASSES, uint16_t> medium_slab_slots;
@@ -35,10 +33,8 @@ namespace snmalloc
size(),
cache_friendly_mask(),
inverse_cache_friendly_mask(),
bump_ptr_start(),
short_bump_ptr_start(),
initial_link_ptr(),
short_initial_link_ptr(),
initial_offset_ptr(),
short_initial_offset_ptr(),
medium_slab_slots()
{
size_t curr = 1;
@@ -71,19 +67,9 @@ namespace snmalloc
size_t correction = SLAB_SIZE % size[i];
// First element in the block is the link
initial_link_ptr[i] = static_cast<uint16_t>(correction);
short_initial_link_ptr[i] =
initial_offset_ptr[i] = static_cast<uint16_t>(correction);
short_initial_offset_ptr[i] =
static_cast<uint16_t>(header_size + short_correction);
// Move to object after link.
auto short_after_link = short_initial_link_ptr[i] + size[i];
size_t after_link = initial_link_ptr[i] + size[i];
// Bump ptr has bottom bit set.
// In case we only have one object on this slab check for wrap around.
short_bump_ptr_start[i] =
static_cast<uint16_t>((short_after_link + 1) % SLAB_SIZE);
bump_ptr_start[i] = static_cast<uint16_t>((after_link + 1) % SLAB_SIZE);
}
for (sizeclass_t i = NUM_SMALL_CLASSES; i < NUM_SIZECLASSES; i++)
@@ -97,21 +83,12 @@ namespace snmalloc
static constexpr SizeClassTable sizeclass_metadata = SizeClassTable();
static inline constexpr uint16_t
get_initial_bumpptr(sizeclass_t sc, bool is_short)
get_initial_offset(sizeclass_t sc, bool is_short)
{
if (is_short)
return sizeclass_metadata.short_bump_ptr_start[sc];
return sizeclass_metadata.short_initial_offset_ptr[sc];
return sizeclass_metadata.bump_ptr_start[sc];
}
static inline constexpr uint16_t
get_initial_link(sizeclass_t sc, bool is_short)
{
if (is_short)
return sizeclass_metadata.short_initial_link_ptr[sc];
return sizeclass_metadata.initial_link_ptr[sc];
return sizeclass_metadata.initial_offset_ptr[sc];
}
constexpr static inline size_t sizeclass_to_size(sizeclass_t sizeclass)