[adjust lea offsets to simply addressing modes
peter@cordes.ca**20080319044041] {
hunk ./rshift.asm 254
-	leaq	(%rsi,%rdx,8), %rsi     C rsi = rsi + rdx*8 = last limb + 1
+	leaq	-8(%rsi,%rdx,8), %rsi     C rsi = rsi + rdx*8 = last limb.
hunk ./rshift.asm 265
-	movq	-8(%rsi,%rdx,8), %xmm6	C %6 = limb1
+	movq	(%rsi,%rdx,8), %xmm6	C %6 = limb1
hunk ./rshift.asm 270
-	movq	(%rsi,%rdx,8), %xmm7	C %7 = limb2
+	movq	8(%rsi,%rdx,8), %xmm7	C %7 = limb2
hunk ./rshift.asm 277
-C	movq	8(%rsi,%rdx,8),	%xmm6	C %6 = limb3  (about to become limb1)
+C	movq	16(%rsi,%rdx,8),	%xmm6	C %6 = limb3  (about to become limb1)
hunk ./rshift.asm 285
-	cmovng	-8(%rsi,%rdx,8), %rcx
+	cmovng	(%rsi), %rcx		C (%rsi,%rdx,8), but the move happens only when rdx is zero.
}