cmd/compile: reduce allocations in regAllocState.regalloc

name      old time/op       new time/op       delta
Template        281ms ± 2%        282ms ± 3%    ~     (p=0.428 n=19+20)
Unicode         138ms ± 6%        138ms ± 7%    ~     (p=0.813 n=19+20)
GoTypes         901ms ± 2%        895ms ± 2%    ~     (p=0.050 n=19+20)
Compiler        4.25s ± 1%        4.23s ± 1%  -0.31%  (p=0.031 n=19+18)
SSA             9.77s ± 1%        9.78s ± 1%    ~     (p=0.512 n=20+20)
Flate           187ms ± 3%        187ms ± 4%    ~     (p=0.687 n=20+19)
GoParser        224ms ± 4%        222ms ± 3%    ~     (p=0.301 n=20+20)
Reflect         576ms ± 2%        576ms ± 2%    ~     (p=0.620 n=20+20)
Tar             262ms ± 3%        263ms ± 3%    ~     (p=0.599 n=19+18)
XML             322ms ± 4%        322ms ± 2%    ~     (p=0.512 n=20+20)

name      old user-time/op  new user-time/op  delta
Template        403ms ± 3%        399ms ± 5%    ~     (p=0.149 n=17+20)
Unicode         217ms ±12%        217ms ± 9%    ~     (p=0.883 n=20+20)
GoTypes         1.24s ± 3%        1.24s ± 3%    ~     (p=0.718 n=20+20)
Compiler        5.90s ± 3%        5.84s ± 5%    ~     (p=0.217 n=18+20)
SSA             14.0s ± 6%        14.1s ± 5%    ~     (p=0.235 n=19+20)
Flate           253ms ± 6%        254ms ± 5%    ~     (p=0.749 n=20+19)
GoParser        309ms ± 7%        307ms ± 5%    ~     (p=0.398 n=20+20)
Reflect         772ms ± 3%        771ms ± 3%    ~     (p=0.901 n=20+19)
Tar             368ms ± 5%        369ms ± 8%    ~     (p=0.429 n=20+20)
XML             435ms ± 5%        434ms ± 5%    ~     (p=0.841 n=20+20)

name      old alloc/op      new alloc/op      delta
Template       39.0MB ± 0%       38.9MB ± 0%  -0.21%  (p=0.000 n=20+19)
Unicode        29.0MB ± 0%       29.0MB ± 0%  -0.03%  (p=0.000 n=20+20)
GoTypes         116MB ± 0%        115MB ± 0%  -0.33%  (p=0.000 n=20+20)
Compiler        498MB ± 0%        496MB ± 0%  -0.37%  (p=0.000 n=19+20)
SSA            1.41GB ± 0%       1.40GB ± 0%  -0.24%  (p=0.000 n=20+20)
Flate          25.0MB ± 0%       25.0MB ± 0%  -0.22%  (p=0.000 n=20+19)
GoParser       31.0MB ± 0%       30.9MB ± 0%  -0.23%  (p=0.000 n=20+17)
Reflect        77.1MB ± 0%       77.0MB ± 0%  -0.12%  (p=0.000 n=20+20)
Tar            39.7MB ± 0%       39.6MB ± 0%  -0.17%  (p=0.000 n=20+20)
XML            44.9MB ± 0%       44.8MB ± 0%  -0.29%  (p=0.000 n=20+20)

name      old allocs/op     new allocs/op     delta
Template         386k ± 0%         385k ± 0%  -0.28%  (p=0.000 n=20+20)
Unicode          337k ± 0%         336k ± 0%  -0.07%  (p=0.000 n=20+20)
GoTypes         1.20M ± 0%        1.20M ± 0%  -0.41%  (p=0.000 n=20+20)
Compiler        4.71M ± 0%        4.68M ± 0%  -0.52%  (p=0.000 n=20+20)
SSA             11.7M ± 0%        11.6M ± 0%  -0.31%  (p=0.000 n=20+19)
Flate            238k ± 0%         237k ± 0%  -0.28%  (p=0.000 n=18+20)
GoParser         320k ± 0%         319k ± 0%  -0.34%  (p=0.000 n=20+19)
Reflect          961k ± 0%         959k ± 0%  -0.12%  (p=0.000 n=20+20)
Tar              397k ± 0%         396k ± 0%  -0.23%  (p=0.000 n=20+20)
XML              419k ± 0%         417k ± 0%  -0.39%  (p=0.000 n=20+19)

Change-Id: Ic7ec3614808d9892c1cab3991b996b7a3b8eff21
Reviewed-on: https://go-review.googlesource.com/102676
Run-TryBot: Alberto Donizetti <alb.donizetti@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>

diff --git a/src/cmd/compile/internal/ssa/regalloc.go b/src/cmd/compile/internal/ssa/regalloc.go
index 291ca91c6129fa11d02cf19039906290f8329ba4..abfdfc1b24591bcf47bd7d138de03ca39cad91cf 100644 (file)
--- a/src/cmd/compile/internal/ssa/regalloc.go
+++ b/src/cmd/compile/internal/ssa/regalloc.go
@@ -963,7 +963,10 @@ func (s *regAllocState) regalloc(f *Func) {
                        }
 
                        // Save the starting state for use by merge edges.
-                       var regList []startReg
+                       // We append to a stack allocated variable that we'll
+                       // later copy into s.startRegs in one fell swoop, to save
+                       // on allocations.
+                       regList := make([]startReg, 0, 32)
                        for r := register(0); r < s.numRegs; r++ {
                                v := s.regs[r].v
                                if v == nil {
@@ -976,7 +979,8 @@ func (s *regAllocState) regalloc(f *Func) {
                                }
                                regList = append(regList, startReg{r, v, s.regs[r].c, s.values[v.ID].uses.pos})
                        }
-                       s.startRegs[b.ID] = regList
+                       s.startRegs[b.ID] = make([]startReg, len(regList))
+                       copy(s.startRegs[b.ID], regList)
 
                        if s.f.pass.debug > regDebug {
                                fmt.Printf("after phis\n")