Run initcgo for all amd64 targets, not just GNU/Linux.

This is required to make cgo export work on Darwin.  Note that
this corrects the stack alignment when calling initcgo to that
required by gcc on amd64.

R=rsc
CC=golang-dev
https://golang.org/cl/907041

src/pkg/runtime/amd64/asm.s

@@ -9,10 +9,16 @@ TEXT	_rt0_amd64(SB),7,$-8
 	MOVQ	0(DI), AX		// argc
 	LEAQ	8(DI), BX		// argv
 	SUBQ	$(4*8+7), SP		// 2args 2auto
-	ANDQ	$~7, SP
+	ANDQ	$~15, SP
 	MOVQ	AX, 16(SP)
 	MOVQ	BX, 24(SP)
 
+	// if there is an initcgo, call it.
+	MOVQ	initcgo(SB), AX
+	TESTQ	AX, AX
+	JZ	2(PC)
+	CALL	AX
+
 	// set the per-goroutine and per-mach registers
 	LEAQ	m0(SB), m
 	LEAQ	g0(SB), g
@@ -372,4 +378,5 @@ TEXT getcallersp(SB),7,$0
 	MOVQ	sp+0(FP), AX
 	RET
 
+GLOBL initcgo(SB), $8
 GLOBL libcgo_set_scheduler(SB), $8

src/pkg/runtime/linux/amd64/rt0.s

@@ -5,13 +5,6 @@
 // Darwin and Linux use the same linkage to main
 
 TEXT	_rt0_amd64_linux(SB),7,$-8
-	MOVQ	initcgo(SB), AX
-	TESTQ	AX, AX
-	JZ	2(PC)
-	CALL	AX
-
 	MOVQ	$_rt0_amd64(SB), AX
 	MOVQ	SP, DI
 	JMP	AX
-
-GLOBL initcgo(SB), $8