big: assembly routines for 386 long shifts

R=rsc
CC=golang-dev
https://golang.org/cl/974043

src/pkg/big/arith_386.s

@@ -100,17 +100,52 @@ E4:	CMPL BX, BP         // i < n
 
 
 // func shlVW(z, x *Word, s Word, n int) (c Word)
-// TODO(gri) implement this routine
 TEXT ·shlVW(SB),7,$0
-	NOP			// work around bug in linker
-	JMP ·shlVW_g(SB)
+	MOVL z+0(FP), DI
+	MOVL x+4(FP), SI
+	MOVL s+8(FP), CX
+	MOVL n+12(FP), BP
+	MOVL $0, AX		// c = 0
+	MOVL $0, BX		// i = 0
+	JMP E8
+
+L8:	MOVL (SI)(BX*8), DX
+	SHLL CX, DX:AX
+	MOVL DX, (DI)(BX*8)
+	MOVL (SI)(BX*8), AX	// reload (not enough regs to save original DX)
+	ADDL $1, BX		// i++
+
+E8:	CMPL BX, BP		// i < n
+	JL L8
+
+	MOVL $0, DX
+	SHLL CX, DX:AX
+	MOVL DX, c+16(FP)
+	RET
 
 
 // func shrVW(z, x *Word, s Word, n int) (c Word)
-// TODO(gri) implement this routine
 TEXT ·shrVW(SB),7,$0
-	NOP			// work around bug in linker
-	JMP ·shrVW_g(SB)
+	MOVL z+0(FP), DI
+	MOVL x+4(FP), SI
+	MOVL s+8(FP), CX
+	MOVL n+12(FP), BX	// i = n
+	MOVL $0, AX		// c = 0
+	JMP E9
+
+L9:	MOVL (SI)(BX*8), DX
+	MOVL DX, BP
+	SHRL CX, DX:AX
+	MOVL DX, (DI)(BX*8)
+	MOVL BP, AX
+
+E9:	SUBL $1, BX		// i--
+	JGE L9
+
+	MOVL $0, DX
+	SHRL CX, DX:AX
+	MOVL DX, c+16(FP)
+	RET
 
 
 // func mulAddVWW(z, x *Word, y, r Word, n int) (c Word)