runtime: improve memmove

check memory overlap

R=rsc, r, ken, edsrzf
CC=golang-dev
https://golang.org/cl/4602047

src/pkg/runtime/386/memmove.s

@@ -32,7 +32,6 @@ TEXT runtime·memmove(SB), 7, $0
 
 /*
  * check and set for backwards
- * should we look closer for overlap?
  */
 	CMPL	SI, DI
 	JLS	back
@@ -40,6 +39,7 @@ TEXT runtime·memmove(SB), 7, $0
 /*
  * forward copy loop
  */
+forward:	
 	MOVL	BX, CX
 	SHRL	$2, CX
 	ANDL	$3, BX
@@ -50,11 +50,19 @@ TEXT runtime·memmove(SB), 7, $0
 
 	MOVL	to+0(FP),AX
 	RET
+/*
+ * check overlap
+ */
+back:
+	MOVL	SI, CX
+	ADDL	BX, CX
+	CMPL	CX, DI
+	JLS	forward
 /*
  * whole thing backwards has
  * adjusted addresses
  */
-back:
+
 	ADDL	BX, DI
 	ADDL	BX, SI
 	STD

src/pkg/runtime/amd64/memmove.s

@@ -33,7 +33,6 @@ TEXT runtime·memmove(SB), 7, $0
 
 /*
  * check and set for backwards
- * should we look closer for overlap?
  */
 	CMPQ	SI, DI
 	JLS	back
@@ -41,6 +40,7 @@ TEXT runtime·memmove(SB), 7, $0
 /*
  * forward copy loop
  */
+forward:	
 	MOVQ	BX, CX
 	SHRQ	$3, CX
 	ANDQ	$7, BX
@@ -51,11 +51,19 @@ TEXT runtime·memmove(SB), 7, $0
 
 	MOVQ	to+0(FP),AX
 	RET
+back:
+/*
+ * check overlap
+ */
+	MOVQ	SI, CX
+	ADDQ	BX, CX
+	CMPQ	CX, DI
+	JLS	forward
+	
 /*
  * whole thing backwards has
  * adjusted addresses
  */
-back:
 	ADDQ	BX, DI
 	ADDQ	BX, SI
 	STD