[PATCH v3 3/5] malloc: Arena is not needed for tcache path in free()
H.J. Lu
hjl.tools@gmail.com
Mon Nov 25 10:34:48 GMT 2024
On Thu, Aug 29, 2024 at 2:31 PM Wangyang Guo <wangyang.guo@intel.com> wrote:
>
> Arena is not needed for _int_free_check() in non-DEBUG mode. This commit
> defers arena deference to _int_free_chunk() thus accelerate tcache path.
> When DEBUG enabled, arena can be obtained from p in do_check_inuse_chunk().
>
> Result of bench-malloc-thread benchmark
>
> Test Platform: Xeon-8380
> Ratio: New / Original time_per_iteration (Lower is Better)
>
> Threads# | Ratio
> -----------|------
> 1 thread | 0.994
> 4 threads | 0.968
>
> The data shows it can brings 3% performance gain in multi-thread scenario.
>
> ---
> Changes in v2:
> - _int_free_check() should be put outside of USE_TCACHE.
> - Link to v1: https://sourceware.org/pipermail/libc-alpha/2024-August/159360.html
> ---
> malloc/malloc.c | 10 ++++++++--
> 1 file changed, 8 insertions(+), 2 deletions(-)
>
> diff --git a/malloc/malloc.c b/malloc/malloc.c
> index 264f35e1a3..efb5292e9f 100644
> --- a/malloc/malloc.c
> +++ b/malloc/malloc.c
> @@ -2143,6 +2143,9 @@ do_check_inuse_chunk (mstate av, mchunkptr p)
> {
> mchunkptr next;
>
> + if (av == NULL)
> + av = arena_for_chunk (p);
> +
> do_check_chunk (av, p);
>
> if (chunk_is_mmapped (p))
> @@ -3447,9 +3450,10 @@ __libc_free (void *mem)
> /* Mark the chunk as belonging to the library again. */
> (void)tag_region (chunk2mem (p), memsize (p));
>
> - ar_ptr = arena_for_chunk (p);
> INTERNAL_SIZE_T size = chunksize (p);
> - _int_free_check (ar_ptr, p, size);
> + /* av is not needed for _int_free_check in non-DEBUG mode,
> + in DEBUG mode, av will fetch from p in do_check_inuse_chunk. */
> + _int_free_check (NULL, p, size);
If _int_free_check doesn't need av, can you push it further down?
>
> #if USE_TCACHE
> if (tcache_free (p, size))
> @@ -3458,6 +3462,8 @@ __libc_free (void *mem)
> return;
> }
> #endif
> +
> + ar_ptr = arena_for_chunk (p);
> _int_free_chunk (ar_ptr, p, size, 0);
> }
>
> --
> 2.43.0
>
--
H.J.
More information about the Libc-alpha
mailing list