bytes: add assembly version of Equal for ARM

BenchmarkEqual32                       662          159  -75.98%
BenchmarkEqual4K                     76545        13719  -82.08%
BenchmarkEqual4M                  90136700     23588870  -73.83%
BenchmarkEqual64M               2147483647   1419616000  -42.63%

BenchmarkEqual32                     48.32       201.15    4.16x
BenchmarkEqual4K                     53.51       298.56    5.58x
BenchmarkEqual4M                     46.53       177.81    3.82x
BenchmarkEqual64M                    27.12        47.27    1.74x

R=golang-dev, qyzhai, minux.ma, rsc, iant, nigeltao
CC=golang-dev
https://golang.org/cl/6118049

diff --git a/src/pkg/bytes/asm_arm.s b/src/pkg/bytes/asm_arm.s
index bc8e00695ce6dd6b58b4adeec211fe0d4edec413..e0170b3aad15c9c93306f51fe7a378827799c47d 100644 (file)
--- a/src/pkg/bytes/asm_arm.s
+++ b/src/pkg/bytes/asm_arm.s
@@ -27,4 +27,30 @@ _notfound:
        RET
 
 TEXT ·Equal(SB),7,$0
-       B       ·equalPortable(SB)
+       MOVW    alen+4(FP), R1
+       MOVW    blen+16(FP), R3
+       
+       CMP     R1, R3          // unequal lengths are not equal
+       B.NE    _notequal
+
+       MOVW    aptr+0(FP), R0
+       MOVW    bptr+12(FP), R2
+       ADD     R0, R1          // end
+
+_next:
+       CMP     R0, R1
+       B.EQ    _equal          // reached the end
+       MOVBU.P 1(R0), R4
+       MOVBU.P 1(R2), R5
+       CMP     R4, R5
+       B.EQ    _next
+
+_notequal:
+       MOVW    $0, R0
+       MOVW    R0, equal+24(FP)
+       RET
+
+_equal:
+       MOVW    $1, R0
+       MOVW    R0, equal+24(FP)
+       RET