CL 49600043 had a non-trivial impact on some of the map benchmarks. (Perhaps other
benchmarks as well; I only happened to be looking for the regression in -test.bench=Map
in runtime.)

$ benchcmp e373697cd435.txt 884801fb67af.txt
benchmark                              old ns/op     new ns/op     delta
BenchmarkSameLengthMap                 4.42          5.24          +18.55%
BenchmarkMapStringKeysEight_1M         19.0          22.1          +16.32%
BenchmarkMapStringKeysEight_64         19.0          22.1         +16.32%
BenchmarkMapStringKeysEight_32         19.1          22.1      +15.71%
BenchmarkMegMap                        20.3        18.2          -10.34%
BenchmarkMapStringKeysEight_16         20.2          18.5          -8.42%
BenchmarkMegEmptyMap                   3.58          3.30          -7.82%
BenchmarkMapIter                       122           126           +3.28%
BenchmarkMegOneMap                     12.7          13.1          +3.15%


Dave says:

> Looks like code movement/alignment. The first has increased by .8 of ns, another
benchmark has decreased by 1.7ns. I think these numbers are low enough to be in the
noise, but it is concerning. Please consider raising an issue and cc'ing Dmitry to see
if his performance dashboard can corroborate these numbers.


This (rough) delta is very consistent locally for me; I don't believe that it is noise.
Dmitry?