1; RUN: llc < %s -regalloc=fast -optimize-regalloc=0 -march=x86 -mattr=+mmx | grep esi 2; PR2082 3; Local register allocator was refusing to use ESI, EDI, and EBP so it ran out of 4; registers. 5define void @transpose4x4(i8* %dst, i8* %src, i32 %dst_stride, i32 %src_stride) { 6entry: 7 %dst_addr = alloca i8* ; <i8**> [#uses=5] 8 %src_addr = alloca i8* ; <i8**> [#uses=5] 9 %dst_stride_addr = alloca i32 ; <i32*> [#uses=4] 10 %src_stride_addr = alloca i32 ; <i32*> [#uses=4] 11 %"alloca point" = bitcast i32 0 to i32 ; <i32> [#uses=0] 12 store i8* %dst, i8** %dst_addr 13 store i8* %src, i8** %src_addr 14 store i32 %dst_stride, i32* %dst_stride_addr 15 store i32 %src_stride, i32* %src_stride_addr 16 %tmp = load i8*, i8** %dst_addr, align 4 ; <i8*> [#uses=1] 17 %tmp1 = getelementptr i8, i8* %tmp, i32 0 ; <i8*> [#uses=1] 18 %tmp12 = bitcast i8* %tmp1 to i32* ; <i32*> [#uses=1] 19 %tmp3 = load i8*, i8** %dst_addr, align 4 ; <i8*> [#uses=1] 20 %tmp4 = load i32, i32* %dst_stride_addr, align 4 ; <i32> [#uses=1] 21 %tmp5 = getelementptr i8, i8* %tmp3, i32 %tmp4 ; <i8*> [#uses=1] 22 %tmp56 = bitcast i8* %tmp5 to i32* ; <i32*> [#uses=1] 23 %tmp7 = load i32, i32* %dst_stride_addr, align 4 ; <i32> [#uses=1] 24 %tmp8 = mul i32 %tmp7, 2 ; <i32> [#uses=1] 25 %tmp9 = load i8*, i8** %dst_addr, align 4 ; <i8*> [#uses=1] 26 %tmp10 = getelementptr i8, i8* %tmp9, i32 %tmp8 ; <i8*> [#uses=1] 27 %tmp1011 = bitcast i8* %tmp10 to i32* ; <i32*> [#uses=1] 28 %tmp13 = load i32, i32* %dst_stride_addr, align 4 ; <i32> [#uses=1] 29 %tmp14 = mul i32 %tmp13, 3 ; <i32> [#uses=1] 30 %tmp15 = load i8*, i8** %dst_addr, align 4 ; <i8*> [#uses=1] 31 %tmp16 = getelementptr i8, i8* %tmp15, i32 %tmp14 ; <i8*> [#uses=1] 32 %tmp1617 = bitcast i8* %tmp16 to i32* ; <i32*> [#uses=1] 33 %tmp18 = load i8*, i8** %src_addr, align 4 ; <i8*> [#uses=1] 34 %tmp19 = getelementptr i8, i8* %tmp18, i32 0 ; <i8*> [#uses=1] 35 %tmp1920 = bitcast i8* %tmp19 to i32* ; <i32*> [#uses=1] 36 %tmp21 = load i8*, i8** %src_addr, align 4 ; <i8*> [#uses=1] 37 %tmp22 = load i32, i32* %src_stride_addr, align 4 ; <i32> [#uses=1] 38 %tmp23 = getelementptr i8, i8* %tmp21, i32 %tmp22 ; <i8*> [#uses=1] 39 %tmp2324 = bitcast i8* %tmp23 to i32* ; <i32*> [#uses=1] 40 %tmp25 = load i32, i32* %src_stride_addr, align 4 ; <i32> [#uses=1] 41 %tmp26 = mul i32 %tmp25, 2 ; <i32> [#uses=1] 42 %tmp27 = load i8*, i8** %src_addr, align 4 ; <i8*> [#uses=1] 43 %tmp28 = getelementptr i8, i8* %tmp27, i32 %tmp26 ; <i8*> [#uses=1] 44 %tmp2829 = bitcast i8* %tmp28 to i32* ; <i32*> [#uses=1] 45 %tmp30 = load i32, i32* %src_stride_addr, align 4 ; <i32> [#uses=1] 46 %tmp31 = mul i32 %tmp30, 3 ; <i32> [#uses=1] 47 %tmp32 = load i8*, i8** %src_addr, align 4 ; <i8*> [#uses=1] 48 %tmp33 = getelementptr i8, i8* %tmp32, i32 %tmp31 ; <i8*> [#uses=1] 49 %tmp3334 = bitcast i8* %tmp33 to i32* ; <i32*> [#uses=1] 50 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 51 br label %return 52 53return: ; preds = %entry 54 ret void 55} 56