cmd/compile: run deadcode before lowered CSE

CSE can make dead values live again.
Running deadcode first avoids that;
it also makes CSE more efficient.

file    before    after     Δ       %       
api     5970616   5966520   -4096   -0.069% 
asm     4867088   4846608   -20480  -0.421% 
compile 23988320  23935072  -53248  -0.222% 
link    6084376   6080280   -4096   -0.067% 
nm      4165736   4161640   -4096   -0.098% 
objdump 4572216   4568120   -4096   -0.090% 
pprof   14452996  14457092  +4096   +0.028% 
trace   11467292  11471388  +4096   +0.036% 
total   132181100 132099180 -81920  -0.062% 

Compiler performance impact is negligible:

name        old alloc/op      new alloc/op      delta
Template         38.8MB ± 0%       38.8MB ± 0%  -0.04%  (p=0.008 n=5+5)
Unicode          28.2MB ± 0%       28.2MB ± 0%    ~     (p=1.000 n=5+5)
GoTypes           131MB ± 0%        131MB ± 0%  -0.14%  (p=0.008 n=5+5)
Compiler          606MB ± 0%        606MB ± 0%  -0.05%  (p=0.008 n=5+5)
SSA              2.14GB ± 0%       2.13GB ± 0%  -0.26%  (p=0.008 n=5+5)
Flate            24.0MB ± 0%       24.0MB ± 0%  -0.18%  (p=0.008 n=5+5)
GoParser         28.8MB ± 0%       28.8MB ± 0%  -0.15%  (p=0.008 n=5+5)
Reflect          83.8MB ± 0%       83.7MB ± 0%  -0.11%  (p=0.008 n=5+5)
Tar              36.4MB ± 0%       36.4MB ± 0%  -0.09%  (p=0.008 n=5+5)
XML              47.9MB ± 0%       47.8MB ± 0%  -0.15%  (p=0.008 n=5+5)
[Geo mean]       84.6MB            84.5MB       -0.12%

name        old allocs/op     new allocs/op     delta
Template           379k ± 0%         380k ± 0%  +0.15%  (p=0.008 n=5+5)
Unicode            340k ± 0%         340k ± 0%    ~     (p=0.738 n=5+5)
GoTypes           1.36M ± 0%        1.36M ± 0%  +0.05%  (p=0.008 n=5+5)
Compiler          5.49M ± 0%        5.49M ± 0%  +0.12%  (p=0.008 n=5+5)
SSA               17.5M ± 0%        17.5M ± 0%  -0.18%  (p=0.008 n=5+5)
Flate              235k ± 0%         235k ± 0%    ~     (p=0.079 n=5+5)
GoParser           302k ± 0%         302k ± 0%    ~     (p=0.310 n=5+5)
Reflect            976k ± 0%         977k ± 0%  +0.08%  (p=0.008 n=5+5)
Tar                352k ± 0%         352k ± 0%  +0.12%  (p=0.008 n=5+5)
XML                436k ± 0%         436k ± 0%  -0.05%  (p=0.008 n=5+5)
[Geo mean]         842k              842k       +0.03%


Change-Id: I53e8faed1859885ca5c4a5d45067a50984f3eff1
Reviewed-on: https://go-review.googlesource.com/c/go/+/175879
Run-TryBot: Josh Bleecher Snyder <[email protected]>
TryBot-Result: Gobot Gobot <[email protected]>
Reviewed-by: Keith Randall <[email protected]>

Author: josharian

src/cmd/compile/internal/ssa/compile.go

@@ -424,6 +424,7 @@ var passes = [...]pass{
 	{name: "insert resched checks", fn: insertLoopReschedChecks,
 		disabled: objabi.Preemptibleloops_enabled == 0}, // insert resched checks in loops.
 	{name: "lower", fn: lower, required: true},
+	{name: "lowered deadcode for cse", fn: deadcode}, // deadcode immediately before CSE avoids CSE making dead values live again
 	{name: "lowered cse", fn: cse},
 	{name: "elim unread autos", fn: elimUnreadAutos},
 	{name: "lowered deadcode", fn: deadcode, required: true},