cmd/compile: optimize ARM code with CMN/TST/TEQ

CMN/TST/TEQ were supported since ARMv4, which can be used to
simplify comparisons.

This patch implements the optimization and here are the benchmark
results.

1. A special test case got 18.21% improvement.
name                     old time/op    new time/op    delta
TSTTEQ-4                    806µs ± 1%     659µs ± 0%  -18.21%  (p=0.000 n=20+18)
(https://github.com/benshi001/ugo1/blob/master/tstteq_test.go)

2. There is no regression in the compilecmp benchmark.
name        old time/op       new time/op       delta
Template          2.31s ± 1%        2.30s ± 1%    ~     (p=0.661 n=10+9)
Unicode           1.32s ± 3%        1.32s ± 5%    ~     (p=0.280 n=10+10)
GoTypes           7.69s ± 1%        7.65s ± 0%  -0.52%  (p=0.027 n=10+8)
Compiler          36.5s ± 1%        36.4s ± 1%    ~     (p=0.546 n=9+9)
SSA               85.1s ± 2%        84.9s ± 1%    ~     (p=0.529 n=10+10)
Flate             1.43s ± 2%        1.43s ± 2%    ~     (p=0.661 n=10+9)
GoParser          1.81s ± 2%        1.81s ± 1%    ~     (p=0.796 n=10+10)
Reflect           5.10s ± 2%        5.09s ± 1%    ~     (p=0.853 n=10+10)
Tar               2.47s ± 1%        2.48s ± 1%    ~     (p=0.123 n=10+10)
XML               2.59s ± 1%        2.58s ± 1%    ~     (p=0.853 n=10+10)
[Geo mean]        4.78s             4.77s       -0.17%

name        old user-time/op  new user-time/op  delta
Template          2.72s ± 3%        2.73s ± 2%    ~     (p=0.928 n=10+10)
Unicode           1.58s ± 4%        1.60s ± 1%    ~     (p=0.087 n=10+9)
GoTypes           9.41s ± 2%        9.36s ± 1%    ~     (p=0.060 n=10+10)
Compiler          44.4s ± 2%        44.2s ± 2%    ~     (p=0.289 n=10+10)
SSA                110s ± 2%         110s ± 1%    ~     (p=0.739 n=10+10)
Flate             1.67s ± 2%        1.63s ± 3%    ~     (p=0.063 n=10+10)
GoParser          2.12s ± 1%        2.12s ± 2%    ~     (p=0.840 n=10+10)
Reflect           5.94s ± 1%        5.98s ± 1%    ~     (p=0.063 n=9+10)
Tar               3.01s ± 2%        3.02s ± 2%    ~     (p=0.584 n=10+10)
XML               3.04s ± 3%        3.02s ± 2%    ~     (p=0.696 n=10+10)
[Geo mean]        5.73s             5.72s       -0.20%

name        old text-bytes    new text-bytes    delta
HelloSize         579kB ± 0%        579kB ± 0%    ~     (all equal)

name        old data-bytes    new data-bytes    delta
HelloSize        5.46kB ± 0%       5.46kB ± 0%    ~     (all equal)

name        old bss-bytes     new bss-bytes     delta
HelloSize        72.8kB ± 0%       72.8kB ± 0%    ~     (all equal)

name        old exe-bytes     new exe-bytes     delta
HelloSize        1.03MB ± 0%       1.03MB ± 0%    ~     (all equal)

3. There is little change in the go1 benchmark (excluding the noise).
name                     old time/op    new time/op     delta
BinaryTree17-4              40.3s ± 1%      40.6s ± 1%  +0.80%  (p=0.000 n=30+30)
Fannkuch11-4                24.2s ± 1%      24.1s ± 0%    ~     (p=0.093 n=30+30)
FmtFprintfEmpty-4           834ns ± 0%      826ns ± 0%  -0.93%  (p=0.000 n=29+24)
FmtFprintfString-4         1.39µs ± 1%     1.36µs ± 0%  -2.02%  (p=0.000 n=30+30)
FmtFprintfInt-4            1.43µs ± 1%     1.44µs ± 1%    ~     (p=0.155 n=30+29)
FmtFprintfIntInt-4         2.09µs ± 0%     2.11µs ± 0%  +1.16%  (p=0.000 n=28+30)
FmtFprintfPrefixedInt-4    2.33µs ± 1%     2.36µs ± 0%  +1.25%  (p=0.000 n=30+30)
FmtFprintfFloat-4          4.27µs ± 1%     4.32µs ± 1%  +1.27%  (p=0.000 n=30+30)
FmtManyArgs-4              8.18µs ± 0%     8.14µs ± 0%  -0.46%  (p=0.000 n=25+27)
GobDecode-4                 101ms ± 1%      101ms ± 1%    ~     (p=0.182 n=29+29)
GobEncode-4                89.6ms ± 1%     87.8ms ± 2%  -2.02%  (p=0.000 n=30+29)
Gzip-4                      4.07s ± 1%      4.08s ± 1%    ~     (p=0.173 n=30+27)
Gunzip-4                    602ms ± 1%      600ms ± 1%  -0.29%  (p=0.000 n=29+28)
HTTPClientServer-4          679µs ± 4%      683µs ± 3%    ~     (p=0.197 n=30+30)
JSONEncode-4                241ms ± 1%      239ms ± 1%  -0.84%  (p=0.000 n=30+30)
JSONDecode-4                903ms ± 1%      882ms ± 1%  -2.33%  (p=0.000 n=30+30)
Mandelbrot200-4            41.8ms ± 0%     41.8ms ± 0%    ~     (p=0.719 n=30+30)
GoParse-4                  45.5ms ± 1%     45.8ms ± 1%  +0.52%  (p=0.000 n=30+30)
RegexpMatchEasy0_32-4      1.27µs ± 1%     1.27µs ± 0%  -0.60%  (p=0.000 n=30+30)
RegexpMatchEasy0_1K-4      7.77µs ± 6%     7.69µs ± 4%  -0.96%  (p=0.040 n=30+30)
RegexpMatchEasy1_32-4      1.29µs ± 1%     1.28µs ± 1%  -0.54%  (p=0.000 n=30+30)
RegexpMatchEasy1_1K-4      10.3µs ± 6%     10.2µs ± 3%    ~     (p=0.453 n=30+27)
RegexpMatchMedium_32-4     1.98µs ± 1%     2.00µs ± 1%  +0.85%  (p=0.000 n=30+29)
RegexpMatchMedium_1K-4      503µs ± 0%      503µs ± 1%    ~     (p=0.752 n=30+30)
RegexpMatchHard_32-4       27.1µs ± 1%     26.5µs ± 0%  -1.96%  (p=0.000 n=30+24)
RegexpMatchHard_1K-4        809µs ± 1%      799µs ± 1%  -1.29%  (p=0.000 n=29+30)
Revcomp-4                  67.3ms ± 2%     67.2ms ± 1%    ~     (p=0.265 n=29+29)
Template-4                  1.08s ± 1%      1.07s ± 0%  -1.39%  (p=0.000 n=30+22)
TimeParse-4                6.93µs ± 1%     6.96µs ± 1%  +0.40%  (p=0.005 n=30+30)
TimeFormat-4               13.3µs ± 0%     13.3µs ± 1%    ~     (p=0.734 n=30+30)
[Geo mean]                  709µs           707µs       -0.32%

name                     old speed      new speed       delta
GobDecode-4              7.59MB/s ± 1%   7.57MB/s ± 1%    ~     (p=0.145 n=29+29)
GobEncode-4              8.56MB/s ± 1%   8.74MB/s ± 1%  +2.07%  (p=0.000 n=30+29)
Gzip-4                   4.76MB/s ± 1%   4.75MB/s ± 1%  -0.25%  (p=0.037 n=30+30)
Gunzip-4                 32.2MB/s ± 1%   32.3MB/s ± 1%  +0.29%  (p=0.000 n=29+28)
JSONEncode-4             8.04MB/s ± 1%   8.11MB/s ± 1%  +0.85%  (p=0.000 n=30+30)
JSONDecode-4             2.15MB/s ± 1%   2.20MB/s ± 1%  +2.29%  (p=0.000 n=30+30)
GoParse-4                1.27MB/s ± 1%   1.26MB/s ± 1%  -0.73%  (p=0.000 n=30+30)
RegexpMatchEasy0_32-4    25.1MB/s ± 1%   25.3MB/s ± 0%  +0.61%  (p=0.000 n=30+30)
RegexpMatchEasy0_1K-4     131MB/s ± 6%    133MB/s ± 4%  +1.35%  (p=0.009 n=28+30)
RegexpMatchEasy1_32-4    24.9MB/s ± 1%   25.0MB/s ± 1%  +0.54%  (p=0.000 n=30+30)
RegexpMatchEasy1_1K-4    99.2MB/s ± 6%  100.2MB/s ± 3%    ~     (p=0.448 n=30+27)
RegexpMatchMedium_32-4    503kB/s ± 1%    500kB/s ± 0%  -0.66%  (p=0.002 n=30+24)
RegexpMatchMedium_1K-4   2.04MB/s ± 0%   2.04MB/s ± 1%    ~     (p=0.358 n=30+30)
RegexpMatchHard_32-4     1.18MB/s ± 1%   1.20MB/s ± 1%  +1.75%  (p=0.000 n=30+30)
RegexpMatchHard_1K-4     1.26MB/s ± 1%   1.28MB/s ± 1%  +1.42%  (p=0.000 n=30+30)
Revcomp-4                37.8MB/s ± 2%   37.8MB/s ± 1%    ~     (p=0.266 n=29+29)
Template-4               1.80MB/s ± 1%   1.82MB/s ± 1%  +1.46%  (p=0.000 n=30+30)
[Geo mean]               6.91MB/s        6.96MB/s       +0.70%

fixes #21583

Change-Id: I24065a80588ccae7de3ad732a3cfb0026cf7e214
Reviewed-on: https://go-review.googlesource.com/67490
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

1 files changed, 6 insertions, 6 deletions

diff --git a/src/cmd/compile/internal/arm/ssa.go b/src/cmd/compile/internal/arm/ssa.go
index 4188775329..aee9ff3014 100644
--- a/src/cmd/compile/internal/arm/ssa.go
+++ b/src/cmd/compile/internal/arm/ssa.go
@@ -473,17 +473,17 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		p := s.Prog(v.Op.Asm())
 		p.From.Type = obj.TYPE_REG
 		p.From.Reg = v.Args[0].Reg()
-	case ssa.OpARMCMPshiftLL:
+	case ssa.OpARMCMPshiftLL, ssa.OpARMCMNshiftLL, ssa.OpARMTSTshiftLL, ssa.OpARMTEQshiftLL:
 		genshift(s, v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), 0, arm.SHIFT_LL, v.AuxInt)
-	case ssa.OpARMCMPshiftRL:
+	case ssa.OpARMCMPshiftRL, ssa.OpARMCMNshiftRL, ssa.OpARMTSTshiftRL, ssa.OpARMTEQshiftRL:
 		genshift(s, v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), 0, arm.SHIFT_LR, v.AuxInt)
-	case ssa.OpARMCMPshiftRA:
+	case ssa.OpARMCMPshiftRA, ssa.OpARMCMNshiftRA, ssa.OpARMTSTshiftRA, ssa.OpARMTEQshiftRA:
 		genshift(s, v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), 0, arm.SHIFT_AR, v.AuxInt)
-	case ssa.OpARMCMPshiftLLreg:
+	case ssa.OpARMCMPshiftLLreg, ssa.OpARMCMNshiftLLreg, ssa.OpARMTSTshiftLLreg, ssa.OpARMTEQshiftLLreg:
 		genregshift(s, v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Args[2].Reg(), 0, arm.SHIFT_LL)
-	case ssa.OpARMCMPshiftRLreg:
+	case ssa.OpARMCMPshiftRLreg, ssa.OpARMCMNshiftRLreg, ssa.OpARMTSTshiftRLreg, ssa.OpARMTEQshiftRLreg:
 		genregshift(s, v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Args[2].Reg(), 0, arm.SHIFT_LR)
-	case ssa.OpARMCMPshiftRAreg:
+	case ssa.OpARMCMPshiftRAreg, ssa.OpARMCMNshiftRAreg, ssa.OpARMTSTshiftRAreg, ssa.OpARMTEQshiftRAreg:
 		genregshift(s, v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Args[2].Reg(), 0, arm.SHIFT_AR)
 	case ssa.OpARMMOVWaddr:
 		p := s.Prog(arm.AMOVW)