[dev.garbage] runtime: add prefetcht0, prefetcht1, prefetcht2, prefetcht3, prefetchnta for GC

We don't know what we need yet, so add them all.
Add them even on x86 architectures (as no-ops) so that
the GC can refer to them unconditionally.

Eventually we'll know what we want and probably
have just one 'prefetch' with an appropriate meaning
on each architecture.

LGTM=rlh
R=rlh
CC=golang-codereviews
https://golang.org/cl/179160043

src/runtime/asm_386.s

@@ -2285,3 +2285,23 @@ TEXT runtime·getg(SB),NOSPLIT,$0-4
 	MOVL	AX, ret+0(FP)
 	RET
 
+TEXT runtime·prefetcht0(SB),NOSPLIT,$0-4
+	MOVL	addr+0(FP), AX
+	PREFETCHT0	(AX)
+	RET
+
+TEXT runtime·prefetcht1(SB),NOSPLIT,$0-4
+	MOVL	addr+0(FP), AX
+	PREFETCHT1	(AX)
+	RET
+
+
+TEXT runtime·prefetcht2(SB),NOSPLIT,$0-4
+	MOVL	addr+0(FP), AX
+	PREFETCHT2	(AX)
+	RET
+
+TEXT runtime·prefetchnta(SB),NOSPLIT,$0-4
+	MOVL	addr+0(FP), AX
+	PREFETCHNTA	(AX)
+	RET

src/runtime/asm_amd64.s

@@ -2228,3 +2228,23 @@ TEXT runtime·getg(SB),NOSPLIT,$0-8
 	MOVQ	g(CX), AX
 	MOVQ	AX, ret+0(FP)
 	RET
+
+TEXT runtime·prefetcht0(SB),NOSPLIT,$0-8
+	MOVQ	addr+0(FP), AX
+	PREFETCHT0	(AX)
+	RET
+
+TEXT runtime·prefetcht1(SB),NOSPLIT,$0-8
+	MOVQ	addr+0(FP), AX
+	PREFETCHT1	(AX)
+	RET
+
+TEXT runtime·prefetcht2(SB),NOSPLIT,$0-8
+	MOVQ	addr+0(FP), AX
+	PREFETCHT2	(AX)
+	RET
+
+TEXT runtime·prefetchnta(SB),NOSPLIT,$0-8
+	MOVQ	addr+0(FP), AX
+	PREFETCHNTA	(AX)
+	RET

src/runtime/asm_amd64p32.s

@@ -1079,3 +1079,24 @@ TEXT runtime·getg(SB),NOSPLIT,$0-4
 	MOVL	g(CX), AX
 	MOVL	AX, ret+0(FP)
 	RET
+
+TEXT runtime·prefetcht0(SB),NOSPLIT,$0-4
+	MOVL	addr+0(FP), AX
+	PREFETCHT0	(AX)
+	RET
+
+TEXT runtime·prefetcht1(SB),NOSPLIT,$0-4
+	MOVL	addr+0(FP), AX
+	PREFETCHT1	(AX)
+	RET
+
+
+TEXT runtime·prefetcht2(SB),NOSPLIT,$0-4
+	MOVL	addr+0(FP), AX
+	PREFETCHT2	(AX)
+	RET
+
+TEXT runtime·prefetchnta(SB),NOSPLIT,$0-4
+	MOVL	addr+0(FP), AX
+	PREFETCHNTA	(AX)
+	RET

src/runtime/asm_arm.s

@@ -1320,3 +1320,15 @@ TEXT runtime·goexit(SB),NOSPLIT,$-4-0
 TEXT runtime·getg(SB),NOSPLIT,$-4-4
 	MOVW	g, ret+0(FP)
 	RET
+
+TEXT runtime·prefetcht0(SB),NOSPLIT,$0-4
+	RET
+
+TEXT runtime·prefetcht1(SB),NOSPLIT,$0-4
+	RET
+
+TEXT runtime·prefetcht2(SB),NOSPLIT,$0-4
+	RET
+
+TEXT runtime·prefetchnta(SB),NOSPLIT,$0-4
+	RET

src/runtime/asm_power64x.s

@@ -986,3 +986,15 @@ TEXT _cgo_topofstack(SB),NOSPLIT,$0
 TEXT runtime·goexit(SB),NOSPLIT,$-8-0
 	MOVD	R0, R0	// NOP
 	BL	runtime·goexit1(SB)	// does not return
+
+TEXT runtime·prefetcht0(SB),NOSPLIT,$0-8
+	RETURN
+
+TEXT runtime·prefetcht1(SB),NOSPLIT,$0-8
+	RETURN
+
+TEXT runtime·prefetcht2(SB),NOSPLIT,$0-8
+	RETURN
+
+TEXT runtime·prefetchnta(SB),NOSPLIT,$0-8
+	RETURN

src/runtime/runtime1.go

@@ -97,7 +97,10 @@ func testAtomic64() {
 
 	z64 = 42
 	x64 = 0
-	// TODO: PREFETCH((unsafe.Pointer)(&z64))
+	prefetcht0(uintptr(unsafe.Pointer(&z64)))
+	prefetcht1(uintptr(unsafe.Pointer(&z64)))
+	prefetcht2(uintptr(unsafe.Pointer(&z64)))
+	prefetchnta(uintptr(unsafe.Pointer(&z64)))
 	if cas64(&z64, x64, 1) {
 		gothrow("cas64 failed")
 	}

src/runtime/stubs.go

@@ -231,3 +231,8 @@ func call536870912(fn, arg unsafe.Pointer, n, retoffset uint32)
 func call1073741824(fn, arg unsafe.Pointer, n, retoffset uint32)
 
 func systemstack_switch()
+
+func prefetcht0(addr uintptr)
+func prefetcht1(addr uintptr)
+func prefetcht2(addr uintptr)
+func prefetchnta(addr uintptr)