6a, 6l: add PREFETCH instructions

R=ken2
CC=golang-dev
https://golang.org/cl/5989073

src/cmd/6a/lex.c

@@ -1001,6 +1001,10 @@ struct
 	"XORPS",	LTYPE3,	AXORPS,
 	"CRC32B",	LTYPE4, ACRC32B,
 	"CRC32Q",	LTYPE4, ACRC32Q,
+	"PREFETCHT0",		LTYPE2,	APREFETCHT0,
+	"PREFETCHT1",		LTYPE2,	APREFETCHT1,
+	"PREFETCHT2",		LTYPE2,	APREFETCHT2,
+	"PREFETCHNTA",		LTYPE2,	APREFETCHNTA,
 
 	0
 };

src/cmd/6l/6.out.h

@@ -736,6 +736,11 @@ enum	as
 	ACRC32B,
 	ACRC32Q,
 	AIMUL3Q,
+	
+	APREFETCHT0,
+	APREFETCHT1,
+	APREFETCHT2,
+	APREFETCHNTA,
 
 	ALAST
 };

src/cmd/6l/optab.c

@@ -539,6 +539,11 @@ uchar	ycrc32l[] =
 {
 	Yml,	Yrl,	Zlitm_r,	0,
 };
+uchar	yprefetch[] =
+{
+	Ym,	Ynone,	Zm_o,	2,
+	0,
+};
 
 /*
  * You are doasm, holding in your hand a Prog* with p->as set to, say, ACRC32,
@@ -1270,8 +1275,13 @@ Optab optab[] =
 	{ AXADDQ,	yrl_ml,	Pw, 0x0f,0xc1 },
 	{ AXADDW,	yrl_ml,	Pe, 0x0f,0xc1 },
 
-	{ ACRC32B,       ycrc32l,Px, 0xf2,0x0f,0x38,0xf0,0},
-	{ ACRC32Q,       ycrc32l,Pw, 0xf2,0x0f,0x38,0xf1,0},
+	{ ACRC32B,       ycrc32l,Px, 0xf2,0x0f,0x38,0xf0,0 },
+	{ ACRC32Q,       ycrc32l,Pw, 0xf2,0x0f,0x38,0xf1,0 },
+	
+	{ APREFETCHT0,	yprefetch,	Pm,	0x18,(01) },
+	{ APREFETCHT1,	yprefetch,	Pm,	0x18,(02) },
+	{ APREFETCHT2,	yprefetch,	Pm,	0x18,(03) },
+	{ APREFETCHNTA,	yprefetch,	Pm,	0x18,(00) },
 
 	{ AEND },
 	0

src/pkg/runtime/asm_amd64.s

@@ -443,8 +443,7 @@ TEXT runtime·atomicstore64(SB), 7, $0
 
 TEXT runtime·prefetch(SB), 7, $0
 	MOVQ    8(SP), AX
-	// PREFETCHNTA (AX)
-	BYTE $0x0f; BYTE $0x18; BYTE $0x00
+	PREFETCHNTA	(AX)
 	RET
 
 // void jmpdefer(fn, sp);