cmd/compile: match more patterns for shortcircuit

cmd/compile: match more patterns for shortcircuit

This CL tries to generalize the pattern matching of certain
shortcircuit-able CFGs a bit more:
For a shortcircuit-able CFG:
p   q
 \ /
  b
 / \
t   u
Where the constant branch is t, and b has multiple phi values
other than the control phi.

For the non-constant branch target u, we try to match the
"diamond" shape CFG:
p   q
 \ /
  b
 / \
t   u
 \ /
  m

or

p   q
 \ /
  b
  |\
  | u
  |/
  m

Instead of matching u as a single block, we know try to
generalize it as a subgraph that satisfy condition:
it's a DAG that has a single entry point u, and has a
path to m.

compilebench stats:
                         │   old.txt   │              new.txt               │
                         │   sec/op    │   sec/op     vs base               │
Template                   109.4m ± 3%   109.8m ± 3%       ~ (p=0.796 n=10)
Unicode                    94.23m ± 1%   93.85m ± 1%       ~ (p=0.631 n=10)
GoTypes                    538.2m ± 1%   538.8m ± 1%       ~ (p=0.912 n=10)
Compiler                   90.21m ± 1%   90.02m ± 1%       ~ (p=0.436 n=10)
SSA                         3.318 ± 1%    3.323 ± 1%       ~ (p=1.000 n=10)
Flate                      69.38m ± 1%   69.57m ± 2%       ~ (p=0.529 n=10)
GoParser                   128.5m ± 1%   127.4m ± 1%       ~ (p=0.075 n=10)
Reflect                    267.4m ± 1%   267.2m ± 2%       ~ (p=0.739 n=10)
Tar                        127.7m ± 2%   126.4m ± 1%       ~ (p=0.353 n=10)
XML                        149.5m ± 1%   149.6m ± 2%       ~ (p=0.684 n=10)
LinkCompiler               390.0m ± 1%   388.4m ± 2%       ~ (p=0.353 n=10)
ExternalLinkCompiler        1.296 ± 0%    1.296 ± 1%       ~ (p=0.971 n=10)
LinkWithoutDebugCompiler   226.3m ± 1%   225.5m ± 1%       ~ (p=0.393 n=10)
StdCmd                      13.26 ± 0%    13.25 ± 1%       ~ (p=0.529 n=10)
geomean                    319.3m        318.8m       -0.17%

                         │   old.txt   │               new.txt               │
                         │ user-sec/op │ user-sec/op   vs base               │
Template                   293.1m ± 3%   291.4m ± 11%       ~ (p=0.436 n=10)
Unicode                    91.09m ± 5%   87.61m ±  7%       ~ (p=0.165 n=10)
GoTypes                     1.932 ± 3%    1.926 ±  3%       ~ (p=0.739 n=10)
Compiler                   125.8m ± 3%   121.5m ± 10%       ~ (p=0.481 n=10)
SSA                         18.93 ± 3%    18.89 ±  1%       ~ (p=0.684 n=10)
Flate                      158.5m ± 5%   160.0m ±  7%       ~ (p=0.971 n=10)
GoParser                   316.0m ± 9%   327.4m ±  7%       ~ (p=0.052 n=10)
Reflect                    845.6m ± 6%   861.6m ±  3%       ~ (p=0.579 n=10)
Tar                        358.1m ± 5%   348.5m ±  4%       ~ (p=0.089 n=10)
XML                        382.4m ± 4%   392.2m ±  3%       ~ (p=0.143 n=10)
LinkCompiler               609.1m ± 4%   627.9m ±  3%       ~ (p=0.123 n=10)
ExternalLinkCompiler        1.336 ± 2%    1.343 ±  4%       ~ (p=0.565 n=10)
LinkWithoutDebugCompiler   248.7m ± 3%   248.0m ±  1%       ~ (p=0.853 n=10)
geomean                    506.4m        506.8m        +0.08%

          │   old.txt    │               new.txt               │
          │  text-bytes  │  text-bytes   vs base               │
HelloSize   965.8Ki ± 0%   965.0Ki ± 0%  -0.08% (p=0.000 n=10)
CmdGoSize   12.30Mi ± 0%   12.29Mi ± 0%  -0.08% (p=0.000 n=10)
geomean     3.406Mi        3.403Mi       -0.08%

          │   old.txt    │                new.txt                │
          │  data-bytes  │  data-bytes   vs base                 │
HelloSize   15.08Ki ± 0%   15.08Ki ± 0%       ~ (p=1.000 n=10) ¹
CmdGoSize   408.5Ki ± 0%   408.5Ki ± 0%       ~ (p=1.000 n=10) ¹
geomean     78.49Ki        78.49Ki       +0.00%
¹ all samples are equal

          │   old.txt    │                new.txt                │
          │  bss-bytes   │  bss-bytes    vs base                 │
HelloSize   142.0Ki ± 0%   142.0Ki ± 0%       ~ (p=1.000 n=10) ¹
CmdGoSize   206.4Ki ± 0%   206.4Ki ± 0%       ~ (p=1.000 n=10) ¹
geomean     171.2Ki        171.2Ki       +0.00%
¹ all samples are equal

          │   old.txt    │               new.txt               │
          │  exe-bytes   │  exe-bytes    vs base               │
HelloSize   1.466Mi ± 0%   1.462Mi ± 0%  -0.27% (p=0.000 n=10)
CmdGoSize   18.19Mi ± 0%   18.17Mi ± 0%  -0.10% (p=0.000 n=10)
geomean     5.164Mi        5.154Mi       -0.18%

Fixes #72132

Change-Id: I3d1cb10b6a158c5750adc23c79709d63dbd771f3
Reviewed-on: https://go-review.googlesource.com/c/go/+/656255
Auto-Submit: Junyang Shao <shaojunyang@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Keith Randall <khr@google.com>