--- /dev/null
+; RUN: llvm-as < %s | llc -regalloc=local -mattr=+mmx | grep esi
+; PR2082
+; Local register allocator was refusing to use ESI, EDI, and EBP so it ran out of
+; registers.
+define void @transpose4x4(i8* %dst, i8* %src, i32 %dst_stride, i32 %src_stride) {
+entry:
+ %dst_addr = alloca i8* ; <i8**> [#uses=5]
+ %src_addr = alloca i8* ; <i8**> [#uses=5]
+ %dst_stride_addr = alloca i32 ; <i32*> [#uses=4]
+ %src_stride_addr = alloca i32 ; <i32*> [#uses=4]
+ %"alloca point" = bitcast i32 0 to i32 ; <i32> [#uses=0]
+ store i8* %dst, i8** %dst_addr
+ store i8* %src, i8** %src_addr
+ store i32 %dst_stride, i32* %dst_stride_addr
+ store i32 %src_stride, i32* %src_stride_addr
+ %tmp = load i8** %dst_addr, align 4 ; <i8*> [#uses=1]
+ %tmp1 = getelementptr i8* %tmp, i32 0 ; <i8*> [#uses=1]
+ %tmp12 = bitcast i8* %tmp1 to i32* ; <i32*> [#uses=1]
+ %tmp3 = load i8** %dst_addr, align 4 ; <i8*> [#uses=1]
+ %tmp4 = load i32* %dst_stride_addr, align 4 ; <i32> [#uses=1]
+ %tmp5 = getelementptr i8* %tmp3, i32 %tmp4 ; <i8*> [#uses=1]
+ %tmp56 = bitcast i8* %tmp5 to i32* ; <i32*> [#uses=1]
+ %tmp7 = load i32* %dst_stride_addr, align 4 ; <i32> [#uses=1]
+ %tmp8 = mul i32 %tmp7, 2 ; <i32> [#uses=1]
+ %tmp9 = load i8** %dst_addr, align 4 ; <i8*> [#uses=1]
+ %tmp10 = getelementptr i8* %tmp9, i32 %tmp8 ; <i8*> [#uses=1]
+ %tmp1011 = bitcast i8* %tmp10 to i32* ; <i32*> [#uses=1]
+ %tmp13 = load i32* %dst_stride_addr, align 4 ; <i32> [#uses=1]
+ %tmp14 = mul i32 %tmp13, 3 ; <i32> [#uses=1]
+ %tmp15 = load i8** %dst_addr, align 4 ; <i8*> [#uses=1]
+ %tmp16 = getelementptr i8* %tmp15, i32 %tmp14 ; <i8*> [#uses=1]
+ %tmp1617 = bitcast i8* %tmp16 to i32* ; <i32*> [#uses=1]
+ %tmp18 = load i8** %src_addr, align 4 ; <i8*> [#uses=1]
+ %tmp19 = getelementptr i8* %tmp18, i32 0 ; <i8*> [#uses=1]
+ %tmp1920 = bitcast i8* %tmp19 to i32* ; <i32*> [#uses=1]
+ %tmp21 = load i8** %src_addr, align 4 ; <i8*> [#uses=1]
+ %tmp22 = load i32* %src_stride_addr, align 4 ; <i32> [#uses=1]
+ %tmp23 = getelementptr i8* %tmp21, i32 %tmp22 ; <i8*> [#uses=1]
+ %tmp2324 = bitcast i8* %tmp23 to i32* ; <i32*> [#uses=1]
+ %tmp25 = load i32* %src_stride_addr, align 4 ; <i32> [#uses=1]
+ %tmp26 = mul i32 %tmp25, 2 ; <i32> [#uses=1]
+ %tmp27 = load i8** %src_addr, align 4 ; <i8*> [#uses=1]
+ %tmp28 = getelementptr i8* %tmp27, i32 %tmp26 ; <i8*> [#uses=1]
+ %tmp2829 = bitcast i8* %tmp28 to i32* ; <i32*> [#uses=1]
+ %tmp30 = load i32* %src_stride_addr, align 4 ; <i32> [#uses=1]
+ %tmp31 = mul i32 %tmp30, 3 ; <i32> [#uses=1]
+ %tmp32 = load i8** %src_addr, align 4 ; <i8*> [#uses=1]
+ %tmp33 = getelementptr i8* %tmp32, i32 %tmp31 ; <i8*> [#uses=1]
+ %tmp3334 = bitcast i8* %tmp33 to i32* ; <i32*> [#uses=1]
+ call void asm sideeffect "movd $4, %mm0 \0A\09movd $5, %mm1 \0A\09movd $6, %mm2 \0A\09movd $7, %mm3 \0A\09punpcklbw %mm1, %mm0 \0A\09punpcklbw %mm3, %mm2 \0A\09movq %mm0, %mm1 \0A\09punpcklwd %mm2, %mm0 \0A\09punpckhwd %mm2, %mm1 \0A\09movd %mm0, $0 \0A\09punpckhdq %mm0, %mm0 \0A\09movd %mm0, $1 \0A\09movd %mm1, $2 \0A\09punpckhdq %mm1, %mm1 \0A\09movd %mm1, $3 \0A\09", "=*m,=*m,=*m,=*m,*m,*m,*m,*m,~{dirflag},~{fpsr},~{flags}"( i32* %tmp12, i32* %tmp56, i32* %tmp1011, i32* %tmp1617, i32* %tmp1920, i32* %tmp2324, i32* %tmp2829, i32* %tmp3334 ) nounwind
+ br label %return
+
+return: ; preds = %entry
+ ret void
+}