cmd/compile: use desired info when allocating registers for live values

When allocting registers for live values, use desired register if available,
this is helpful for some cases, such as (*entry).delete, which can save a
few of copies.
Besides, this patch allows more debugging information to be printed out.

Test results of compilecmp on Linux/amd64:
name                      old time/op                 new time/op                 delta
Template                    326729362.060000ns +- 3%    329227238.775510ns +- 4%  +0.76%  (p=0.038 n=50+49)
Unicode                     157671860.391304ns +- 6%    156917927.320000ns +- 6%    ~     (p=0.291 n=46+50)
GoTypes                    1065591138.304348ns +- 2%   1063695977.434783ns +- 1%    ~     (p=0.208 n=46+46)
Compiler                   5053424790.760001ns +- 2%   5052729636.551020ns +- 3%    ~     (p=0.908 n=50+49)
SSA                       12392067635.866669ns +- 2%  12319786960.460005ns +- 2%  -0.58%  (p=0.008 n=45+50)
Flate                       212609767.340000ns +- 5%    213011228.085106ns +- 5%    ~     (p=0.685 n=50+47)
GoParser                    266870495.100000ns +- 4%    266962314.280000ns +- 3%    ~     (p=0.975 n=50+50)
Reflect                     660164306.551021ns +- 2%    658284470.729167ns +- 2%    ~     (p=0.069 n=49+48)
Tar                         292805895.720000ns +- 4%    292103626.954545ns +- 2%    ~     (p=0.321 n=50+44)
XML                         386294811.700000ns +- 4%    386665088.820000ns +- 4%    ~     (p=0.786 n=50+50)
LinkCompiler                548495788.659575ns +- 5%    549359489.102041ns +- 4%    ~     (p=0.855 n=47+49)
ExternalLinkCompiler       1810414270.280000ns +- 2%   1806872224.673470ns +- 2%    ~     (p=0.313 n=50+49)
LinkWithoutDebugCompiler    340888843.795918ns +- 5%    340341541.100000ns +- 6%    ~     (p=0.735 n=49+50)
[Geo mean]                   664550174.613777ns          664090221.153575ns       -0.07%

name                      old user-time/op            new user-time/op            delta
Template                    565202800.000000ns +-16%    595351040.000000ns +-16%  +5.33%  (p=0.001 n=50+50)
Unicode                     378444740.000000ns +-14%    373825183.673469ns +-17%    ~     (p=0.458 n=50+49)
GoTypes                    2052073341.463415ns +-12%   2059679864.864865ns +- 7%    ~     (p=0.381 n=41+37)
Compiler                   9913371980.000000ns +-20%   9848836720.000002ns +-19%    ~     (p=0.781 n=50+50)
SSA                       25013846224.489799ns +-17%  24571896183.673466ns +-17%    ~     (p=0.132 n=49+49)
Flate                       314422702.127660ns +-17%    314831666.666667ns +-11%    ~     (p=0.427 n=47+45)
GoParser                    419496060.000000ns +- 9%    417403460.000000ns +-11%    ~     (p=0.512 n=50+50)
Reflect                    1233632469.387755ns +-17%   1193061073.170732ns +-13%  -3.29%  (p=0.030 n=49+41)
Tar                         509855937.500000ns +-10%    508700740.000000ns +-14%    ~     (p=0.890 n=48+50)
XML                         703511425.531915ns +-12%    694007591.836735ns +-11%    ~     (p=0.164 n=47+49)
LinkCompiler                993137687.500000ns +- 6%    991914714.285714ns +- 8%    ~     (p=0.860 n=48+49)
ExternalLinkCompiler       2193851840.000001ns +- 3%   2186672183.673470ns +- 5%    ~     (p=0.320 n=50+49)
LinkWithoutDebugCompiler    420800875.000000ns +-10%    422062640.000000ns +- 9%    ~     (p=0.840 n=48+50)
[Geo mean]                  1145156131.480097ns         1142033233.550961ns       -0.27%

name                      old alloc/op                new alloc/op                delta
Template                                36.3MB +- 0%                36.3MB +- 0%    ~     (p=0.886 n=50+49)
Unicode                                 30.1MB +- 0%                30.1MB +- 0%    ~     (p=0.792 n=50+50)
GoTypes                                  118MB +- 0%                 118MB +- 0%    ~     (p=1.000 n=47+48)
Compiler                                 562MB +- 0%                 562MB +- 0%    ~     (p=0.205 n=50+49)
SSA                                     1.42GB +- 0%                1.42GB +- 0%  -0.12%  (p=0.000 n=50+50)
Flate                                   22.8MB +- 0%                22.8MB +- 0%    ~     (p=0.384 n=50+47)
GoParser                                28.0MB +- 0%                28.0MB +- 0%  -0.02%  (p=0.013 n=50+50)
Reflect                                 78.0MB +- 0%                78.0MB +- 0%    ~     (p=0.384 n=46+48)
Tar                                     34.1MB +- 0%                34.1MB +- 0%    ~     (p=0.072 n=50+50)
XML                                     43.1MB +- 0%                43.1MB +- 0%  -0.04%  (p=0.000 n=49+50)
LinkCompiler                            98.5MB +- 0%                98.5MB +- 0%  +0.01%  (p=0.012 n=50+43)
ExternalLinkCompiler                    89.6MB +- 0%                89.6MB +- 0%    ~     (p=0.762 n=50+50)
LinkWithoutDebugCompiler                56.9MB +- 0%                56.9MB +- 0%    ~     (p=0.268 n=49+48)
[Geo mean]                               77.7MB                      77.7MB       -0.01%

name                      old allocs/op               new allocs/op               delta
Template                                  367k +- 0%                  367k +- 0%  -0.01%  (p=0.002 n=50+49)
Unicode                                   345k +- 0%                  345k +- 0%    ~     (p=0.981 n=50+50)
GoTypes                                  1.28M +- 0%                 1.28M +- 0%  -0.00%  (p=0.002 n=49+50)
Compiler                                 5.39M +- 0%                 5.39M +- 0%  -0.00%  (p=0.000 n=50+50)
SSA                                      13.9M +- 0%                 13.9M +- 0%  +0.01%  (p=0.000 n=50+50)
Flate                                     230k +- 0%                  230k +- 0%    ~     (p=0.815 n=50+50)
GoParser                                  292k +- 0%                  292k +- 0%  -0.01%  (p=0.000 n=50+50)
Reflect                                   977k +- 0%                  977k +- 0%  -0.00%  (p=0.035 n=50+50)
Tar                                       343k +- 0%                  343k +- 0%  -0.01%  (p=0.008 n=48+50)
XML                                       418k +- 0%                  418k +- 0%  -0.01%  (p=0.000 n=50+50)
LinkCompiler                              516k +- 0%                  516k +- 0%  +0.01%  (p=0.002 n=50+48)
ExternalLinkCompiler                      570k +- 0%                  570k +- 0%    ~     (p=0.430 n=46+50)
LinkWithoutDebugCompiler                  169k +- 0%                  169k +- 0%    ~     (p=0.706 n=49+49)
[Geo mean]                                 672k                        672k       -0.00%

name                      old maxRSS/op               new maxRSS/op               delta
Template                                 34.3M +- 5%                 34.7M +- 4%  +1.24%  (p=0.004 n=50+50)
Unicode                                  36.2M +- 5%                 36.1M +- 8%    ~     (p=0.785 n=50+50)
GoTypes                                  75.7M +- 7%                 76.1M +- 6%    ~     (p=0.544 n=50+50)
Compiler                                  304M +- 7%                  304M +- 7%    ~     (p=0.744 n=50+50)
SSA                                       721M +- 6%                  723M +- 7%    ~     (p=0.724 n=49+50)
Flate                                    26.1M +- 3%                 26.1M +- 5%    ~     (p=0.649 n=48+49)
GoParser                                 29.3M +- 5%                 29.3M +- 4%    ~     (p=0.809 n=50+50)
Reflect                                  56.0M +- 6%                 56.3M +- 5%    ~     (p=0.350 n=50+50)
Tar                                      34.1M +- 3%                 33.9M +- 5%    ~     (p=0.121 n=49+50)
XML                                      39.6M +- 5%                 39.9M +- 4%    ~     (p=0.109 n=50+50)
LinkCompiler                              168M +- 1%                  168M +- 1%    ~     (p=0.578 n=49+48)
ExternalLinkCompiler                      179M +- 1%                  179M +- 2%    ~     (p=0.522 n=46+46)
LinkWithoutDebugCompiler                  137M +- 3%                  137M +- 3%    ~     (p=0.463 n=41+50)
[Geo mean]                                79.3M                       79.5M       +0.20%

name                      old text-bytes              new text-bytes              delta
HelloSize                                812kB +- 0%                 811kB +- 0%  -0.05%  (p=0.000 n=50+50)

name                      old data-bytes              new data-bytes              delta
HelloSize                               13.3kB +- 0%                13.3kB +- 0%    ~     (all equal)

name                      old bss-bytes               new bss-bytes               delta
HelloSize                                206kB +- 0%                 206kB +- 0%    ~     (all equal)

name                      old exe-bytes               new exe-bytes               delta
HelloSize                               1.21MB +- 0%                1.21MB +- 0%  +0.02%  (p=0.000 n=50+50)

file      before    after     Δ       %
addr2line 4052949   4052453   -496    -0.012%
api       4948171   4947163   -1008   -0.020%
asm       4888889   4888049   -840    -0.017%
buildid   2617545   2617673   +128    +0.005%
cgo       4521681   4516801   -4880   -0.108%
compile   19139091  19137683  -1408   -0.007%
cover     4843191   4840359   -2832   -0.058%
dist      3473677   3474717   +1040   +0.030%
doc       3821592   3821552   -40     -0.001%
fix       3220587   3220059   -528    -0.016%
link      6587368   6582696   -4672   -0.071%
nm        3999858   3999186   -672    -0.017%
objdump   4409161   4408217   -944    -0.021%
pack      2394038   2393846   -192    -0.008%
pprof     13601271  13602487  +1216   +0.009%
test2json 2645148   2644604   -544    -0.021%
trace     10357878  10356862  -1016   -0.010%
vet       6779482   6778706   -776    -0.011%
total     106301577 106283113 -18464  -0.017%

Change-Id: I63ac6e224e1a4756ddc1bfc4aabbaeb92d7d4273
Reviewed-on: https://go-review.googlesource.com/c/go/+/263599
Run-TryBot: eric fang <eric.fang@arm.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: eric fang <eric.fang@arm.com>
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
This commit is contained in:
eric fang 2020-10-19 03:57:15 +00:00 committed by Keith Randall
parent 1c0d1f4e97
commit b2a8317b31

View File

@ -1582,6 +1582,19 @@ func (s *regAllocState) regalloc(f *Func) {
}
v := s.orig[vid]
m := s.compatRegs(v.Type) &^ s.used
// Used desired register if available.
outerloop:
for _, e := range desired.entries {
if e.ID != v.ID {
continue
}
for _, r := range e.regs {
if r != noRegister && m>>r&1 != 0 {
m = regMask(1) << r
break outerloop
}
}
}
if m&^desired.avoid != 0 {
m &^= desired.avoid
}
@ -1643,7 +1656,9 @@ func (s *regAllocState) regalloc(f *Func) {
// we'll rematerialize during the merge.
continue
}
//fmt.Printf("live-at-end spill for %s at %s\n", s.orig[e.ID], b)
if s.f.pass.debug > regDebug {
fmt.Printf("live-at-end spill for %s at %s\n", s.orig[e.ID], b)
}
spill := s.makeSpill(s.orig[e.ID], b)
s.spillLive[b.ID] = append(s.spillLive[b.ID], spill.ID)
}
@ -2514,7 +2529,7 @@ func (s *regAllocState) computeLive() {
for _, b := range f.Blocks {
fmt.Printf(" %s:", b)
for _, x := range s.live[b.ID] {
fmt.Printf(" v%d", x.ID)
fmt.Printf(" v%d(%d)", x.ID, x.dist)
for _, e := range s.desired[b.ID].entries {
if e.ID != x.ID {
continue