cmd/compile: allocate less in regalloc's liveValues

Instrumenting the compiler shows that, at the end of liveValues, the
values of the workList's cap are distributed as:

  cap	 freq
  1  	 0.006
  2  	 0.002
  4  	 0.237
  8  	 0.272
  16	 0.254
  32	 0.141
  64	 0.062
  128	 0.02
  256	 0.005
  512	 0.001
  1024	 0.0

Since the initial workList slice allocation is always on the stack
(as the variable does not escape), we can aggressively pre-allocate a
big backing array at (almost) no cost. This will save several
allocations in liveValues calls that end up having a large workList,
with no performance penalties for calls that have a small workList.

name      old time/op       new time/op       delta
Template        284ms ± 3%        282ms ± 3%    ~     (p=0.201 n=20+20)
Unicode         138ms ± 7%        138ms ± 7%    ~     (p=0.718 n=20+20)
GoTypes         905ms ± 2%        895ms ± 1%  -1.10%  (p=0.003 n=19+18)
Compiler        4.26s ± 1%        4.25s ± 1%  -0.38%  (p=0.038 n=20+19)
SSA             9.85s ± 2%        9.80s ± 1%    ~     (p=0.061 n=20+19)
Flate           187ms ± 6%        186ms ± 5%    ~     (p=0.289 n=20+20)
GoParser        227ms ± 3%        225ms ± 3%    ~     (p=0.072 n=20+20)
Reflect         578ms ± 2%        575ms ± 2%    ~     (p=0.059 n=18+20)
Tar             263ms ± 2%        265ms ± 3%    ~     (p=0.224 n=19+20)
XML             323ms ± 3%        325ms ± 2%    ~     (p=0.127 n=20+20)

name      old user-time/op  new user-time/op  delta
Template        406ms ± 6%        404ms ± 4%    ~     (p=0.314 n=20+20)
Unicode         220ms ± 6%        215ms ±11%    ~     (p=0.077 n=18+20)
GoTypes         1.25s ± 3%        1.24s ± 4%    ~     (p=0.461 n=20+20)
Compiler        5.95s ± 2%        5.84s ± 5%  -1.93%  (p=0.007 n=20+20)
SSA             14.4s ± 4%        14.2s ± 4%    ~     (p=0.108 n=20+20)
Flate           257ms ± 6%        252ms ± 9%    ~     (p=0.063 n=20+20)
GoParser        317ms ± 5%        312ms ± 6%  -1.85%  (p=0.049 n=20+20)
Reflect         779ms ± 2%        774ms ± 3%    ~     (p=0.253 n=20+20)
Tar             371ms ± 4%        374ms ± 4%    ~     (p=0.327 n=20+20)
XML             440ms ± 5%        442ms ± 5%    ~     (p=0.678 n=20+20)

name      old alloc/op      new alloc/op      delta
Template       39.4MB ± 0%       39.0MB ± 0%  -0.96%  (p=0.000 n=20+20)
Unicode        29.1MB ± 0%       29.0MB ± 0%  -0.13%  (p=0.000 n=20+20)
GoTypes         117MB ± 0%        116MB ± 0%  -0.88%  (p=0.000 n=20+20)
Compiler        502MB ± 0%        498MB ± 0%  -0.77%  (p=0.000 n=19+20)
SSA            1.42GB ± 0%       1.40GB ± 0%  -0.80%  (p=0.000 n=20+20)
Flate          25.3MB ± 0%       25.0MB ± 0%  -1.10%  (p=0.000 n=20+19)
GoParser       31.3MB ± 0%       31.0MB ± 0%  -1.05%  (p=0.000 n=20+20)
Reflect        77.9MB ± 0%       77.1MB ± 0%  -1.03%  (p=0.000 n=20+20)
Tar            40.0MB ± 0%       39.7MB ± 0%  -0.80%  (p=0.000 n=20+20)
XML            45.2MB ± 0%       44.9MB ± 0%  -0.72%  (p=0.000 n=20+20)

name      old allocs/op     new allocs/op     delta
Template         392k ± 0%         386k ± 0%  -1.44%  (p=0.000 n=20+20)
Unicode          337k ± 0%         337k ± 0%  -0.22%  (p=0.000 n=20+20)
GoTypes         1.22M ± 0%        1.20M ± 0%  -1.33%  (p=0.000 n=20+20)
Compiler        4.76M ± 0%        4.71M ± 0%  -1.12%  (p=0.000 n=20+20)
SSA             11.8M ± 0%        11.7M ± 0%  -1.00%  (p=0.000 n=20+20)
Flate            241k ± 0%         238k ± 0%  -1.49%  (p=0.000 n=20+20)
GoParser         324k ± 0%         320k ± 0%  -1.17%  (p=0.000 n=20+20)
Reflect          981k ± 0%         961k ± 0%  -2.11%  (p=0.000 n=20+20)
Tar              402k ± 0%         397k ± 0%  -1.29%  (p=0.000 n=20+20)
XML              424k ± 0%         419k ± 0%  -1.10%  (p=0.000 n=19+20)

Change-Id: If46667ae98eee2d47a615cad05e18df0629d8388
Reviewed-on: https://go-review.googlesource.com/102495
Run-TryBot: Alberto Donizetti <alb.donizetti@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>

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

@@ -54,7 +54,7 @@ func liveValues(f *Func, reachable []bool) []bool {
 	}
 
 	// Find all live values
-	var q []*Value // stack-like worklist of unscanned values
+	q := make([]*Value, 0, 64) // stack-like worklist of unscanned values
 
 	// Starting set: all control values of reachable blocks are live.
 	// Calls are live (because callee can observe the memory state).