It's pretty hard to get reliable CPU numbers, even with 50 runs on an
otherwise-idle physical Linux machine, but the garbage reduction
numbers are nice. To get useful time/op numbers, I modified
compilebench to report user CPU time instead of wall time:

name       old time/op     new time/op     delta
Template       547ms ± 6%      557ms ± 5%   +1.80%        (p=0.001 n=49+49)
Unicode        360ms ± 9%      365ms ± 6%     ~           (p=0.094 n=50+45)
GoTypes        1.84s ± 3%      1.82s ± 3%   -1.50%        (p=0.000 n=50+49)
Compiler       9.19s ± 2%      9.02s ± 2%   -1.87%        (p=0.000 n=45+50)

name       old alloc/op    new alloc/op    delta
Template      63.3MB ± 0%     59.1MB ± 0%   -6.72%        (p=0.000 n=50+50)
Unicode       43.1MB ± 0%     42.9MB ± 0%   -0.47%        (p=0.000 n=50+49)
GoTypes        220MB ± 0%      200MB ± 0%   -9.00%        (p=0.000 n=50+50)
Compiler      1.00GB ± 0%     0.89GB ± 0%  -10.09%        (p=0.000 n=50+49)

name       old allocs/op   new allocs/op   delta
Template        681k ± 0%       680k ± 0%   -0.16%        (p=0.000 n=50+48)
Unicode         541k ± 0%       541k ± 0%   -0.02%        (p=0.011 n=48+50)
GoTypes        2.08M ± 0%      2.08M ± 0%   -0.19%        (p=0.000 n=48+50)
Compiler       9.24M ± 0%      9.23M ± 0%   -0.11%        (p=0.000 n=50+50)

Change-Id: I1fac4ebf85a1783e3289c3ffb1ed365442837643
Reviewed-on: https://go-review.googlesource.com/20995
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Keith Randall <khr@golang.org>