runtime: cache inner pinner on P

runtime: cache inner pinner on P

This change caches the *pinner on the P to pool it and reduce the chance
that a new allocation is made. It also makes the *pinner no longer drop
its refs array on unpin, also to avoid reallocating.

The Pinner benchmark results before and after this CL are attached at
the bottom of the commit message.

Note that these results are biased toward the current change because of
the last two benchmark changes. Reusing the pinner in the benchmark
itself achieves similar performance before this change. The benchmark
results thus basically just confirm that this change does cache the
inner pinner in a useful way. Using the previous benchmarks there's
actually a slight regression from the extra check in the cache, however
the long pole is still setPinned itself.

name                                old time/op    new time/op    delta
PinnerPinUnpinBatch-8                 42.2µs ± 2%    41.5µs ± 1%      ~     (p=0.056 n=5+5)
PinnerPinUnpinBatchDouble-8            367µs ± 1%     350µs ± 1%    -4.67%  (p=0.008 n=5+5)
PinnerPinUnpinBatchTiny-8              108µs ± 0%     102µs ± 1%    -6.22%  (p=0.008 n=5+5)
PinnerPinUnpin-8                       592ns ± 8%      40ns ± 1%   -93.29%  (p=0.008 n=5+5)
PinnerPinUnpinTiny-8                   693ns ± 9%      39ns ± 1%   -94.31%  (p=0.008 n=5+5)
PinnerPinUnpinDouble-8                 843ns ± 5%     124ns ± 3%   -85.24%  (p=0.008 n=5+5)
PinnerPinUnpinParallel-8              1.11µs ± 5%    0.00µs ± 0%   -99.55%  (p=0.008 n=5+5)
PinnerPinUnpinParallelTiny-8          1.12µs ± 8%    0.00µs ± 1%   -99.55%  (p=0.008 n=5+5)
PinnerPinUnpinParallelDouble-8        1.79µs ± 4%    0.58µs ± 6%   -67.36%  (p=0.008 n=5+5)
PinnerIsPinnedOnPinned-8              5.78ns ± 0%    5.80ns ± 1%      ~     (p=0.548 n=5+5)
PinnerIsPinnedOnUnpinned-8            4.99ns ± 1%    4.98ns ± 0%      ~     (p=0.841 n=5+5)
PinnerIsPinnedOnPinnedParallel-8      0.71ns ± 0%    0.71ns ± 0%      ~     (p=0.175 n=5+5)
PinnerIsPinnedOnUnpinnedParallel-8    0.67ns ± 1%    0.66ns ± 0%      ~     (p=0.167 n=5+5)

name                                old alloc/op   new alloc/op   delta
PinnerPinUnpinBatch-8                 20.1kB ± 0%    20.0kB ± 0%    -0.32%  (p=0.008 n=5+5)
PinnerPinUnpinBatchDouble-8           52.7kB ± 0%    52.7kB ± 0%    -0.12%  (p=0.008 n=5+5)
PinnerPinUnpinBatchTiny-8             20.1kB ± 0%    20.0kB ± 0%    -0.32%  (p=0.008 n=5+5)
PinnerPinUnpin-8                       64.0B ± 0%      0.0B       -100.00%  (p=0.008 n=5+5)
PinnerPinUnpinTiny-8                   64.0B ± 0%      0.0B       -100.00%  (p=0.008 n=5+5)
PinnerPinUnpinDouble-8                 64.0B ± 0%      0.0B       -100.00%  (p=0.008 n=5+5)
PinnerPinUnpinParallel-8               64.0B ± 0%      0.0B       -100.00%  (p=0.008 n=5+5)
PinnerPinUnpinParallelTiny-8           64.0B ± 0%      0.0B       -100.00%  (p=0.008 n=5+5)
PinnerPinUnpinParallelDouble-8         64.0B ± 0%      0.0B       -100.00%  (p=0.008 n=5+5)
PinnerIsPinnedOnPinned-8               0.00B          0.00B           ~     (all equal)
PinnerIsPinnedOnUnpinned-8             0.00B          0.00B           ~     (all equal)
PinnerIsPinnedOnPinnedParallel-8       0.00B          0.00B           ~     (all equal)
PinnerIsPinnedOnUnpinnedParallel-8     0.00B          0.00B           ~     (all equal)

name                                old allocs/op  new allocs/op  delta
PinnerPinUnpinBatch-8                   9.00 ± 0%      8.00 ± 0%   -11.11%  (p=0.008 n=5+5)
PinnerPinUnpinBatchDouble-8             11.0 ± 0%      10.0 ± 0%    -9.09%  (p=0.008 n=5+5)
PinnerPinUnpinBatchTiny-8               9.00 ± 0%      8.00 ± 0%   -11.11%  (p=0.008 n=5+5)
PinnerPinUnpin-8                        1.00 ± 0%      0.00       -100.00%  (p=0.008 n=5+5)
PinnerPinUnpinTiny-8                    1.00 ± 0%      0.00       -100.00%  (p=0.008 n=5+5)
PinnerPinUnpinDouble-8                  1.00 ± 0%      0.00       -100.00%  (p=0.008 n=5+5)
PinnerPinUnpinParallel-8                1.00 ± 0%      0.00       -100.00%  (p=0.008 n=5+5)
PinnerPinUnpinParallelTiny-8            1.00 ± 0%      0.00       -100.00%  (p=0.008 n=5+5)
PinnerPinUnpinParallelDouble-8          1.00 ± 0%      0.00       -100.00%  (p=0.008 n=5+5)
PinnerIsPinnedOnPinned-8                0.00           0.00           ~     (all equal)
PinnerIsPinnedOnUnpinned-8              0.00           0.00           ~     (all equal)
PinnerIsPinnedOnPinnedParallel-8        0.00           0.00           ~     (all equal)
PinnerIsPinnedOnUnpinnedParallel-8      0.00           0.00           ~     (all equal)

For #46787.

Change-Id: I0cdfad77b189c425868944a4faeff3d5b97417b9
Reviewed-on: https://go-review.googlesource.com/c/go/+/497615
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Ansiwen <ansiwen@gmail.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Auto-Submit: Michael Knyszek <mknyszek@google.com>