cmd/compile: memory clearing optimization for arm64

Use "STP (ZR, ZR), O(R)" instead of "MOVD ZR, O(R)" to implement memory clearing.
Also improve assembler supports to STP/LDP.
Results (A57@2GHzx8):

benchmark                   old ns/op     new ns/op     delta
BenchmarkClearFat8-8        1.00          1.00          +0.00%
BenchmarkClearFat12-8       1.01          1.01          +0.00%
BenchmarkClearFat16-8       1.01          1.01          +0.00%
BenchmarkClearFat24-8       1.52          1.52          +0.00%
BenchmarkClearFat32-8       3.00          2.02          -32.67%
BenchmarkClearFat40-8       3.50          2.52          -28.00%
BenchmarkClearFat48-8       3.50          3.03          -13.43%
BenchmarkClearFat56-8       4.00          3.50          -12.50%
BenchmarkClearFat64-8       4.25          4.00          -5.88%
BenchmarkClearFat128-8      8.01          8.01          +0.00%
BenchmarkClearFat256-8      16.1          16.0          -0.62%
BenchmarkClearFat512-8      32.1          32.0          -0.31%
BenchmarkClearFat1024-8     64.1          64.1          +0.00%

Change-Id: Ie5f5eac271ff685884775005825f206167a5c146
Reviewed-on: https://go-review.googlesource.com/55610
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>

1 files changed, 8 insertions, 5 deletions

diff --git a/src/cmd/compile/internal/ssa/gen/ARM64Ops.go b/src/cmd/compile/internal/ssa/gen/ARM64Ops.go
index 3b3d494c54..1cac97f3ae 100644
--- a/src/cmd/compile/internal/ssa/gen/ARM64Ops.go
+++ b/src/cmd/compile/internal/ssa/gen/ARM64Ops.go
@@ -144,6 +144,7 @@ func init() {
 		gpload    = regInfo{inputs: []regMask{gpspsbg}, outputs: []regMask{gp}}
 		gpstore   = regInfo{inputs: []regMask{gpspsbg, gpg}}
 		gpstore0  = regInfo{inputs: []regMask{gpspsbg}}
+		gpstore2  = regInfo{inputs: []regMask{gpspsbg, gpg, gpg}}
 		gpxchg    = regInfo{inputs: []regMask{gpspsbg, gpg}, outputs: []regMask{gp}}
 		gpcas     = regInfo{inputs: []regMask{gpspsbg, gpg, gpg}, outputs: []regMask{gp}}
 		fp01      = regInfo{inputs: nil, outputs: []regMask{fp}}
@@ -275,13 +276,15 @@ func init() {
 		{name: "MOVHstore", argLength: 3, reg: gpstore, aux: "SymOff", asm: "MOVH", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"},   // store 2 bytes of arg1 to arg0 + auxInt + aux.  arg2=mem.
 		{name: "MOVWstore", argLength: 3, reg: gpstore, aux: "SymOff", asm: "MOVW", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"},   // store 4 bytes of arg1 to arg0 + auxInt + aux.  arg2=mem.
 		{name: "MOVDstore", argLength: 3, reg: gpstore, aux: "SymOff", asm: "MOVD", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"},   // store 8 bytes of arg1 to arg0 + auxInt + aux.  arg2=mem.
+		{name: "STP", argLength: 4, reg: gpstore2, aux: "SymOff", asm: "STP", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"},         // store 16 bytes of arg1 and arg2 to arg0 + auxInt + aux.  arg3=mem.
 		{name: "FMOVSstore", argLength: 3, reg: fpstore, aux: "SymOff", asm: "FMOVS", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"}, // store 4 bytes of arg1 to arg0 + auxInt + aux.  arg2=mem.
 		{name: "FMOVDstore", argLength: 3, reg: fpstore, aux: "SymOff", asm: "FMOVD", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"}, // store 8 bytes of arg1 to arg0 + auxInt + aux.  arg2=mem.
 
 		{name: "MOVBstorezero", argLength: 2, reg: gpstore0, aux: "SymOff", asm: "MOVB", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"}, // store 1 byte of zero to arg0 + auxInt + aux.  arg1=mem.
 		{name: "MOVHstorezero", argLength: 2, reg: gpstore0, aux: "SymOff", asm: "MOVH", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"}, // store 2 bytes of zero to arg0 + auxInt + aux.  arg1=mem.
 		{name: "MOVWstorezero", argLength: 2, reg: gpstore0, aux: "SymOff", asm: "MOVW", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"}, // store 4 bytes of zero to arg0 + auxInt + aux.  arg1=mem.
-		{name: "MOVDstorezero", argLength: 2, reg: gpstore0, aux: "SymOff", asm: "MOVD", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"}, // store 8 bytes of zero to arg0 + auxInt + aux.  ar12=mem.
+		{name: "MOVDstorezero", argLength: 2, reg: gpstore0, aux: "SymOff", asm: "MOVD", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"}, // store 8 bytes of zero to arg0 + auxInt + aux.  arg1=mem.
+		{name: "MOVQstorezero", argLength: 2, reg: gpstore0, aux: "SymOff", asm: "STP", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"},  // store 16 bytes of zero to arg0 + auxInt + aux.  arg1=mem.
 
 		// conversions
 		{name: "MOVBreg", argLength: 1, reg: gp11, asm: "MOVB"},   // move from arg0, sign-extended from byte
@@ -347,7 +350,7 @@ func init() {
 			aux:       "Int64",
 			argLength: 2,
 			reg: regInfo{
-				inputs:   []regMask{gp},
+				inputs:   []regMask{buildReg("R16")},
 				clobbers: buildReg("R16 R30"),
 			},
 			faultOnNilArg0: true,
@@ -355,14 +358,14 @@ func init() {
 
 		// large zeroing
 		// arg0 = address of memory to zero (in R16 aka arm64.REGRT1, changed as side effect)
-		// arg1 = address of the last element to zero
+		// arg1 = address of the last 16-byte unit to zero
 		// arg2 = mem
 		// returns mem
-		//	MOVD.P	ZR, 8(R16)
+		//	STP.P	(ZR,ZR), 16(R16)
 		//	CMP	Rarg1, R16
 		//	BLE	-2(PC)
 		// Note: the-end-of-the-memory may be not a valid pointer. it's a problem if it is spilled.
-		// the-end-of-the-memory - 8 is with the area to zero, ok to spill.
+		// the-end-of-the-memory - 16 is with the area to zero, ok to spill.
 		{
 			name:      "LoweredZero",
 			argLength: 3,