; RUN: llc < %s -mtriple=xcore > %t1.s
;; This caused a compilation failure since the
;; address arithmetic was folded into the LDWSP instruction,
;; resulting in a negative offset which eliminateFrameIndex was
;; unable to eliminate.
define i32 @test(i32 %bar) nounwind readnone {
entry:
        %bar_addr = alloca i32
        %0 = getelementptr i32, ptr %bar_addr, i32 -1
        %1 = load i32, ptr %0, align 4
        ret i32 %1
}