about summary refs log tree commit diff
path: root/mathvec
diff options
context:
space:
mode:
authorNoah Goldstein <goldstein.w.n@gmail.com>2023-06-07 13:18:01 -0500
committerNoah Goldstein <goldstein.w.n@gmail.com>2023-06-12 11:33:39 -0500
commitaf992e7abdc9049714da76cae1e5e18bc4838fb8 (patch)
treed849302de8864185badbeba75cef1f614df45bcf /mathvec
parent5e8d1b0328a850c229146f40e18848728b104583 (diff)
downloadglibc-af992e7abdc9049714da76cae1e5e18bc4838fb8.tar.gz
glibc-af992e7abdc9049714da76cae1e5e18bc4838fb8.tar.xz
glibc-af992e7abdc9049714da76cae1e5e18bc4838fb8.zip
x86: Increase `non_temporal_threshold` to roughly `sizeof_L3 / 4`
Current `non_temporal_threshold` set to roughly '3/4 * sizeof_L3 /
ncores_per_socket'. This patch updates that value to roughly
'sizeof_L3 / 4`

The original value (specifically dividing the `ncores_per_socket`) was
done to limit the amount of other threads' data a `memcpy`/`memset`
could evict.

Dividing by 'ncores_per_socket', however leads to exceedingly low
non-temporal thresholds and leads to using non-temporal stores in
cases where REP MOVSB is multiple times faster.

Furthermore, non-temporal stores are written directly to main memory
so using it at a size much smaller than L3 can place soon to be
accessed data much further away than it otherwise could be. As well,
modern machines are able to detect streaming patterns (especially if
REP MOVSB is used) and provide LRU hints to the memory subsystem. This
in affect caps the total amount of eviction at 1/cache_associativity,
far below meaningfully thrashing the entire cache.

As best I can tell, the benchmarks that lead this small threshold
where done comparing non-temporal stores versus standard cacheable
stores. A better comparison (linked below) is to be REP MOVSB which,
on the measure systems, is nearly 2x faster than non-temporal stores
at the low-end of the previous threshold, and within 10% for over
100MB copies (well past even the current threshold). In cases with a
low number of threads competing for bandwidth, REP MOVSB is ~2x faster
up to `sizeof_L3`.

The divisor of `4` is a somewhat arbitrary value. From benchmarks it
seems Skylake and Icelake both prefer a divisor of `2`, but older CPUs
such as Broadwell prefer something closer to `8`. This patch is meant
to be followed up by another one to make the divisor cpu-specific, but
in the meantime (and for easier backporting), this patch settles on
`4` as a middle-ground.

Benchmarks comparing non-temporal stores, REP MOVSB, and cacheable
stores where done using:
https://github.com/goldsteinn/memcpy-nt-benchmarks

Sheets results (also available in pdf on the github):
https://docs.google.com/spreadsheets/d/e/2PACX-1vS183r0rW_jRX6tG_E90m9qVuFiMbRIJvi5VAE8yYOvEOIEEc3aSNuEsrFbuXw5c3nGboxMmrupZD7K/pubhtml
Reviewed-by: DJ Delorie <dj@redhat.com>
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
Diffstat (limited to 'mathvec')
0 files changed, 0 insertions, 0 deletions