cmd/compile/internal/ssa: combine consecutive BigEndian stores on arm64

This optimization mirrors that which is already implemented for AMD64.  The
optimization specifically targets the binary.BigEndian.PutUint* functions.

encoding-binary results on Amberwing:
name                   old time/op    new time/op    delta
ReadSlice1000Int32s      9.83µs ± 2%    9.78µs ± 1%     ~     (p=0.362 n=9+10)
ReadStruct               5.24µs ± 3%    5.19µs ± 2%     ~     (p=0.285 n=10+10)
ReadInts                 8.35µs ± 8%    8.44µs ± 3%     ~     (p=0.323 n=10+10)
WriteInts                3.38µs ± 3%    3.44µs ±15%     ~     (p=0.921 n=9+10)
WriteSlice1000Int32s     11.4µs ± 6%    10.2µs ± 4%   -9.94%  (p=0.000 n=10+10)
PutUint16                 510ns ±12%     500ns ± 0%     ~     (p=0.586 n=10+7)
PutUint32                 530ns ±15%     490ns ±12%     ~     (p=0.086 n=10+10)
PutUint64                 550ns ± 0%     470ns ± 6%  -14.52%  (p=0.000 n=7+10)
LittleEndianPutUint16     500ns ± 0%     475ns ±16%     ~     (p=0.120 n=7+10)
LittleEndianPutUint32     450ns ± 0%     517ns ±16%  +14.81%  (p=0.004 n=8+9)
LittleEndianPutUint64     550ns ± 0%     485ns ±13%  -11.82%  (p=0.000 n=8+10)
PutUvarint32              685ns ±12%     622ns ± 4%   -9.17%  (p=0.005 n=10+9)
PutUvarint64              735ns ± 9%     711ns ± 9%     ~     (p=0.272 n=10+9)
[Geo mean]               1.47µs         1.42µs        -3.87%

name                   old speed      new speed      delta
ReadSlice1000Int32s     407MB/s ± 2%   409MB/s ± 1%     ~     (p=0.362 n=9+10)
ReadStruct             14.3MB/s ± 3%  14.4MB/s ± 2%     ~     (p=0.250 n=10+10)
ReadInts               3.59MB/s ± 7%  3.56MB/s ± 4%     ~     (p=0.340 n=10+10)
WriteInts              8.87MB/s ± 3%  8.74MB/s ±13%     ~     (p=0.890 n=9+10)
WriteSlice1000Int32s    352MB/s ± 6%   391MB/s ± 4%  +11.03%  (p=0.000 n=10+10)
PutUint16              3.95MB/s ±13%  4.00MB/s ± 0%     ~     (p=0.312 n=10+7)
PutUint32              7.62MB/s ±17%  8.21MB/s ±11%     ~     (p=0.086 n=10+10)
PutUint64              14.6MB/s ± 0%  17.1MB/s ± 6%  +17.28%  (p=0.000 n=7+10)
LittleEndianPutUint16  4.00MB/s ± 0%  4.23MB/s ±18%     ~     (p=0.176 n=7+10)
LittleEndianPutUint32  8.89MB/s ± 0%  7.64MB/s ±20%  -14.05%  (p=0.001 n=8+10)
LittleEndianPutUint64  14.6MB/s ± 0%  16.6MB/s ±12%  +13.86%  (p=0.000 n=8+10)
PutUvarint32           5.86MB/s ±14%  6.44MB/s ± 5%   +9.84%  (p=0.006 n=10+9)
PutUvarint64           10.9MB/s ± 8%  11.3MB/s ± 9%     ~     (p=0.373 n=10+9)
[Geo mean]             14.2MB/s       14.8MB/s        +3.93%

go1 results on Amberwing:
RegexpMatchEasy0_32       254ns ± 0%     254ns ± 0%    ~     (all equal)
RegexpMatchEasy0_1K       547ns ± 0%     547ns ± 0%    ~     (all equal)
RegexpMatchEasy1_32       252ns ± 0%     253ns ± 1%    ~     (p=0.294 n=8+10)
RegexpMatchEasy1_1K       782ns ± 0%     783ns ± 1%    ~     (p=0.529 n=8+9)
RegexpMatchMedium_32      316ns ± 0%     316ns ± 0%    ~     (all equal)
RegexpMatchMedium_1K     51.5µs ± 0%    51.5µs ± 0%    ~     (p=0.645 n=10+9)
RegexpMatchHard_32       2.75µs ± 0%    2.75µs ± 0%    ~     (all equal)
RegexpMatchHard_1K       78.7µs ± 0%    78.7µs ± 0%    ~     (p=0.754 n=10+10)
FmtFprintfEmpty          57.0ns ± 0%    57.0ns ± 0%    ~     (all equal)
FmtFprintfString          111ns ± 0%     111ns ± 0%    ~     (all equal)
FmtFprintfInt             114ns ± 0%     114ns ± 1%    ~     (p=0.065 n=9+10)
FmtFprintfIntInt          182ns ± 0%     178ns ± 0%  -2.20%  (p=0.000 n=10+10)
FmtFprintfPrefixedInt     225ns ± 0%     227ns ± 0%  +0.89%  (p=0.000 n=10+10)
FmtFprintfFloat           307ns ± 0%     307ns ± 0%    ~     (p=1.000 n=9+9)
FmtManyArgs               697ns ± 0%     701ns ± 2%    ~     (p=0.108 n=9+10)
Gzip                      436ms ± 0%     437ms ± 0%  +0.23%  (p=0.000 n=10+8)
HTTPClientServer         88.8µs ± 2%    89.6µs ± 1%  +0.98%  (p=0.019 n=10+10)
JSONEncode               20.1ms ± 1%    20.2ms ± 1%  +0.48%  (p=0.007 n=10+10)
JSONDecode               94.7ms ± 1%    94.1ms ± 0%  -0.62%  (p=0.000 n=10+9)
GobDecode                12.6ms ± 2%    12.6ms ± 1%    ~     (p=0.360 n=10+8)
GobEncode                12.0ms ± 1%    11.9ms ± 1%  -1.34%  (p=0.000 n=10+10)
Mandelbrot200            5.05ms ± 0%    5.05ms ± 0%  +0.12%  (p=0.000 n=10+10)
TimeParse                 448ns ± 0%     448ns ± 0%    ~     (p=0.529 n=8+9)
TimeFormat                501ns ± 1%     501ns ± 1%    ~     (p=1.000 n=10+9)
Template                 90.6ms ± 0%    89.1ms ± 0%  -1.67%  (p=0.000 n=9+9)
GoParse                  6.01ms ± 0%    5.96ms ± 0%  -0.83%  (p=0.000 n=10+9)
BinaryTree17              11.7s ± 0%     11.7s ± 0%    ~     (p=0.481 n=10+10)
Revcomp                   675ms ± 0%     675ms ± 0%    ~     (p=0.436 n=9+9)
Fannkuch11                3.26s ± 0%     3.27s ± 1%  +0.57%  (p=0.000 n=10+10)
[Geo mean]               67.4µs         67.3µs       -0.10%

name                   old speed      new speed      delta
RegexpMatchEasy0_32     126MB/s ± 0%   126MB/s ± 0%    ~     (p=0.353 n=10+7)
RegexpMatchEasy0_1K    1.87GB/s ± 0%  1.87GB/s ± 0%    ~     (p=0.275 n=8+10)
RegexpMatchEasy1_32     127MB/s ± 0%   126MB/s ± 1%    ~     (p=0.110 n=8+10)
RegexpMatchEasy1_1K    1.31GB/s ± 0%  1.31GB/s ± 1%    ~     (p=0.079 n=8+10)
RegexpMatchMedium_32   3.16MB/s ± 0%  3.16MB/s ± 0%    ~     (all equal)
RegexpMatchMedium_1K   19.9MB/s ± 0%  19.9MB/s ± 0%    ~     (p=0.889 n=10+9)
RegexpMatchHard_32     11.7MB/s ± 0%  11.7MB/s ± 0%    ~     (all equal)
RegexpMatchHard_1K     13.0MB/s ± 0%  13.0MB/s ± 0%    ~     (p=1.000 n=10+10)
Gzip                   44.5MB/s ± 0%  44.4MB/s ± 0%  -0.22%  (p=0.000 n=10+8)
JSONEncode             96.6MB/s ± 1%  96.1MB/s ± 1%  -0.48%  (p=0.007 n=10+10)
JSONDecode             20.5MB/s ± 1%  20.6MB/s ± 0%  +0.63%  (p=0.000 n=10+9)
GobDecode              61.0MB/s ± 2%  61.1MB/s ± 1%    ~     (p=0.372 n=10+8)
GobEncode              63.8MB/s ± 1%  64.7MB/s ± 1%  +1.36%  (p=0.000 n=10+10)
Template               21.4MB/s ± 0%  21.8MB/s ± 0%  +1.69%  (p=0.000 n=9+9)
GoParse                9.63MB/s ± 0%  9.71MB/s ± 0%  +0.84%  (p=0.000 n=9+8)
Revcomp                 377MB/s ± 0%   376MB/s ± 0%    ~     (p=0.399 n=9+9)
[Geo mean]             56.2MB/s       56.3MB/s       +0.20%

Change-Id: Ic915373f5ef512f9fbc45745860e5db7f6de6286
Reviewed-on: https://go-review.googlesource.com/97755
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>

src/cmd/compile/internal/gc/asm_test.go

@@ -3318,6 +3318,33 @@ var linuxARM64Tests = []*asmTest{
 		pos: []string{"MOVD"},
 		neg: []string{"MOVB", "MOVH", "MOVW"},
 	},
+	{
+		fn: `
+		func $(b []byte, v uint16) {
+			binary.BigEndian.PutUint16(b, v)
+		}
+		`,
+		pos: []string{"MOVH"},
+		neg: []string{"MOVB"},
+	},
+	{
+		fn: `
+		func $(b []byte, v uint32) {
+			binary.BigEndian.PutUint32(b, v)
+		}
+		`,
+		pos: []string{"MOVW"},
+		neg: []string{"MOVB", "MOVH"},
+	},
+	{
+		fn: `
+		func $(b []byte, v uint64) {
+			binary.BigEndian.PutUint64(b, v)
+		}
+		`,
+		pos: []string{"MOVD"},
+		neg: []string{"MOVB", "MOVH", "MOVW"},
+	},
 }
 
 var linuxMIPSTests = []*asmTest{

src/cmd/compile/internal/ssa/gen/ARM64.rules

@@ -1545,6 +1545,63 @@
 	&& isSamePtr(ptr0, ptr1)
 	&& clobber(x)
 	-> (MOVDstore [i-4] {s} ptr0 w0 mem)
+(MOVBstore [i] {s} ptr w
+	x0:(MOVBstore [i-1] {s} ptr (SRLconst [8] w)
+	x1:(MOVBstore [i-2] {s} ptr (SRLconst [16] w)
+	x2:(MOVBstore [i-3] {s} ptr (SRLconst [24] w)
+	x3:(MOVBstore [i-4] {s} ptr (SRLconst [32] w)
+	x4:(MOVBstore [i-5] {s} ptr (SRLconst [40] w)
+	x5:(MOVBstore [i-6] {s} ptr (SRLconst [48] w)
+	x6:(MOVBstore [i-7] {s} ptr (SRLconst [56] w) mem))))))))
+	&& x0.Uses == 1
+	&& x1.Uses == 1
+	&& x2.Uses == 1
+	&& x3.Uses == 1
+	&& x4.Uses == 1
+	&& x5.Uses == 1
+	&& x6.Uses == 1
+	&& clobber(x0)
+	&& clobber(x1)
+	&& clobber(x2)
+	&& clobber(x3)
+	&& clobber(x4)
+	&& clobber(x5)
+	&& clobber(x6)
+	-> (MOVDstore [i-7] {s} ptr (REV <w.Type> w) mem)
+(MOVBstore [i] {s} ptr w
+	x0:(MOVBstore [i-1] {s} ptr (SRLconst [8] (MOVWUreg w))
+	x1:(MOVBstore [i-2] {s} ptr (SRLconst [16] (MOVWUreg w))
+	x2:(MOVBstore [i-3] {s} ptr (SRLconst [24] (MOVWUreg w)) mem))))
+	&& x0.Uses == 1
+	&& x1.Uses == 1
+	&& x2.Uses == 1
+	&& clobber(x0)
+	&& clobber(x1)
+	&& clobber(x2)
+	-> (MOVWstore [i-3] {s} ptr (REVW <w.Type> w) mem)
+(MOVBstore [i] {s} ptr w
+	x0:(MOVBstore [i-1] {s} ptr (SRLconst [8] w)
+	x1:(MOVBstore [i-2] {s} ptr (SRLconst [16] w)
+	x2:(MOVBstore [i-3] {s} ptr (SRLconst [24] w) mem))))
+	&& x0.Uses == 1
+	&& x1.Uses == 1
+	&& x2.Uses == 1
+	&& clobber(x0)
+	&& clobber(x1)
+	&& clobber(x2)
+	-> (MOVWstore [i-3] {s} ptr (REVW <w.Type> w) mem)
+(MOVBstore [i] {s} ptr w x:(MOVBstore [i-1] {s} ptr (SRLconst [8] w) mem))
+	&& x.Uses == 1
+	&& clobber(x)
+	-> (MOVHstore [i-1] {s} ptr (REV16W <w.Type> w) mem)
+(MOVBstore [i] {s} ptr w x:(MOVBstore [i-1] {s} ptr (SRLconst [8] (MOVHUreg w)) mem))
+	&& x.Uses == 1
+	&& clobber(x)
+	-> (MOVHstore [i-1] {s} ptr (REV16W <w.Type> w) mem)
+(MOVBstore [i] {s} ptr w x:(MOVBstore [i-1] {s} ptr (SRLconst [8] (MOVWUreg w)) mem))
+	&& x.Uses == 1
+	&& clobber(x)
+	-> (MOVHstore [i-1] {s} ptr (REV16W <w.Type> w) mem)
 
 // FP simplification
 (FNEGS (FMULS x y)) -> (FNMULS x y)