1 ; RUN: llc < %s -o - -march=x86 -mattr=+mmx | FileCheck %s
2 ; There are no MMX instructions here. We use add+adcl for the adds.
4 define <1 x i64> @unsigned_add3(<1 x i64>* %a, <1 x i64>* %b, i32 %count) nounwind {
6 %tmp2942 = icmp eq i32 %count, 0 ; <i1> [#uses=1]
7 br i1 %tmp2942, label %bb31, label %bb26
9 bb26: ; preds = %bb26, %entry
14 %i.037.0 = phi i32 [ 0, %entry ], [ %tmp25, %bb26 ] ; <i32> [#uses=3]
15 %sum.035.0 = phi <1 x i64> [ zeroinitializer, %entry ], [ %tmp22, %bb26 ] ; <<1 x i64>> [#uses=1]
16 %tmp13 = getelementptr <1 x i64>* %b, i32 %i.037.0 ; <<1 x i64>*> [#uses=1]
17 %tmp14 = load <1 x i64>* %tmp13 ; <<1 x i64>> [#uses=1]
18 %tmp18 = getelementptr <1 x i64>* %a, i32 %i.037.0 ; <<1 x i64>*> [#uses=1]
19 %tmp19 = load <1 x i64>* %tmp18 ; <<1 x i64>> [#uses=1]
20 %tmp21 = add <1 x i64> %tmp19, %tmp14 ; <<1 x i64>> [#uses=1]
21 %tmp22 = add <1 x i64> %tmp21, %sum.035.0 ; <<1 x i64>> [#uses=2]
22 %tmp25 = add i32 %i.037.0, 1 ; <i32> [#uses=2]
23 %tmp29 = icmp ult i32 %tmp25, %count ; <i1> [#uses=1]
24 br i1 %tmp29, label %bb26, label %bb31
26 bb31: ; preds = %bb26, %entry
27 %sum.035.1 = phi <1 x i64> [ zeroinitializer, %entry ], [ %tmp22, %bb26 ] ; <<1 x i64>> [#uses=1]
28 ret <1 x i64> %sum.035.1
32 ; This is the original test converted to use MMX intrinsics.
34 define <1 x i64> @unsigned_add3a(x86_mmx* %a, x86_mmx* %b, i32 %count) nounwind {
36 %tmp2943 = bitcast <1 x i64><i64 0> to x86_mmx
37 %tmp2942 = icmp eq i32 %count, 0 ; <i1> [#uses=1]
38 br i1 %tmp2942, label %bb31, label %bb26
40 bb26: ; preds = %bb26, %entry
42 ; CHECK: movq ({{.*}},8), %mm
43 ; CHECK: paddq ({{.*}},8), %mm
44 ; CHECK: paddq %mm{{[0-7]}}, %mm
46 %i.037.0 = phi i32 [ 0, %entry ], [ %tmp25, %bb26 ] ; <i32> [#uses=3]
47 %sum.035.0 = phi x86_mmx [ %tmp2943, %entry ], [ %tmp22, %bb26 ] ; <x86_mmx> [#uses=1]
48 %tmp13 = getelementptr x86_mmx* %b, i32 %i.037.0 ; <x86_mmx*> [#uses=1]
49 %tmp14 = load x86_mmx* %tmp13 ; <x86_mmx> [#uses=1]
50 %tmp18 = getelementptr x86_mmx* %a, i32 %i.037.0 ; <x86_mmx*> [#uses=1]
51 %tmp19 = load x86_mmx* %tmp18 ; <x86_mmx> [#uses=1]
52 %tmp21 = call x86_mmx @llvm.x86.mmx.padd.q (x86_mmx %tmp19, x86_mmx %tmp14) ; <x86_mmx> [#uses=1]
53 %tmp22 = call x86_mmx @llvm.x86.mmx.padd.q (x86_mmx %tmp21, x86_mmx %sum.035.0) ; <x86_mmx> [#uses=2]
54 %tmp25 = add i32 %i.037.0, 1 ; <i32> [#uses=2]
55 %tmp29 = icmp ult i32 %tmp25, %count ; <i1> [#uses=1]
56 br i1 %tmp29, label %bb26, label %bb31
58 bb31: ; preds = %bb26, %entry
59 %sum.035.1 = phi x86_mmx [ %tmp2943, %entry ], [ %tmp22, %bb26 ] ; <x86_mmx> [#uses=1]
60 %t = bitcast x86_mmx %sum.035.1 to <1 x i64>
64 declare x86_mmx @llvm.x86.mmx.padd.q(x86_mmx, x86_mmx)