[PATCH v4 0/7] malloc: tcache improvements

[Cupertino Miranda]

Hi everyone,

This new version is the result of the recent interactions with Wilco
Dijkstra relating performance degradation detected in previous version.
I will reply to this cover letter with an attached spreadsheet and chart
image.

Presented results were collected with mimalloc-bench, executed with the
following code variations:
 - master as the baseline,
 - v3 (patches 1 to 3),
 - v3 (all 5 patches),
 - inline (with __always_inline in tcache_free),
 - v4 (recent changes to improve performance: added fast_csize2tidx,
   trim spurious conditions and adding __glibc_(un)likely to
   conditions).

All results were taken without any tunables set, i.e. the tcache should
behave as originally intended in glibc malloc.

Looking forward to your comments.

Best regards,
Cupertino

Changes from v3:
 - force inline of _int_free. This might make sense as a generic patch.
 - performance improvements
Changes from v2:
 - Redesign implementation.
 - Dropped strategy to remove old chunks from tcache to a better
   split and consolidate strategy (future patch, still work in
   progress).
 - Fixed regressions in v2. 
Changes from v1:
 - Typo in the assigment to mp_.tcache_bins in do_set_cache_max
   function. Was setting it to MAX_TCACHE_SIZE instead of
   TCACHE_MAX_BINS.
 - Fixed testsuite regressions.

Cupertino Miranda (7):
  malloc: missing initialization of tcache in _mid_memalign
  malloc: mangle tcache entries pointers
  malloc: replace _int_free by _int_free_chunk in sysmalloc.
  malloc: add tcache support for large chunk caching
  malloc: add testing for large tcache support.
  malloc: force inline of _int_free
  malloc: performance improvements

 Rules           |  19 ++++++
 malloc/Makefile |  15 +++++
 malloc/arena.c  |   2 +-
 malloc/malloc.c | 156 ++++++++++++++++++++++++++++++++----------------
 4 files changed, 139 insertions(+), 53 deletions(-)

-- 
2.39.5