Krzysztof Parzyszek | 046090d | 2018-03-12 14:01:28 +0000 | [diff] [blame] | 1 | ; RUN: llc -march=hexagon -O2 < %s | FileCheck %s |
| 2 | |
| 3 | ; Generate REG_SEQUENCE instead of combine |
| 4 | ; CHECK-NOT: combine(#0 |
| 5 | |
| 6 | ; Function Attrs: nounwind |
| 7 | define void @f0(i16* nocapture readonly %a0, i16* nocapture readonly %a1, i16* nocapture %a2, i16* nocapture readonly %a3, i32 %a4) #0 { |
| 8 | b0: |
| 9 | %v0 = lshr i32 %a4, 1 |
| 10 | %v1 = icmp eq i32 %v0, 0 |
| 11 | br i1 %v1, label %b3, label %b1 |
| 12 | |
| 13 | b1: ; preds = %b0 |
| 14 | %v2 = bitcast i16* %a2 to i64* |
| 15 | %v3 = bitcast i16* %a1 to i64* |
| 16 | %v4 = bitcast i16* %a0 to i64* |
| 17 | br label %b2 |
| 18 | |
| 19 | b2: ; preds = %b2, %b1 |
| 20 | %v5 = phi i32 [ 0, %b1 ], [ %v71, %b2 ] |
| 21 | %v6 = phi i64* [ %v4, %b1 ], [ %v9, %b2 ] |
| 22 | %v7 = phi i64* [ %v3, %b1 ], [ %v11, %b2 ] |
| 23 | %v8 = phi i64* [ %v2, %b1 ], [ %v70, %b2 ] |
| 24 | %v9 = getelementptr inbounds i64, i64* %v6, i32 1 |
| 25 | %v10 = load i64, i64* %v6, align 8, !tbaa !0 |
| 26 | %v11 = getelementptr inbounds i64, i64* %v7, i32 1 |
| 27 | %v12 = load i64, i64* %v7, align 8, !tbaa !0 |
| 28 | %v13 = trunc i64 %v10 to i32 |
| 29 | %v14 = lshr i64 %v10, 32 |
| 30 | %v15 = tail call i64 @llvm.hexagon.S2.vzxthw(i32 %v13) |
| 31 | %v16 = trunc i64 %v12 to i32 |
| 32 | %v17 = lshr i64 %v12, 32 |
| 33 | %v18 = tail call i64 @llvm.hexagon.S2.vzxthw(i32 %v16) |
| 34 | %v19 = trunc i64 %v15 to i32 |
| 35 | %v20 = lshr i64 %v15, 32 |
| 36 | %v21 = getelementptr inbounds i16, i16* %a3, i32 %v19 |
| 37 | %v22 = load i16, i16* %v21, align 2, !tbaa !3 |
| 38 | %v23 = trunc i64 %v20 to i32 |
| 39 | %v24 = getelementptr inbounds i16, i16* %a3, i32 %v23 |
| 40 | %v25 = load i16, i16* %v24, align 2, !tbaa !3 |
| 41 | %v26 = trunc i64 %v18 to i32 |
| 42 | %v27 = lshr i64 %v18, 32 |
| 43 | %v28 = getelementptr inbounds i16, i16* %a3, i32 %v26 |
| 44 | %v29 = load i16, i16* %v28, align 2, !tbaa !3 |
| 45 | %v30 = trunc i64 %v27 to i32 |
| 46 | %v31 = getelementptr inbounds i16, i16* %a3, i32 %v30 |
| 47 | %v32 = load i16, i16* %v31, align 2, !tbaa !3 |
| 48 | %v33 = zext i16 %v32 to i64 |
| 49 | %v34 = shl nuw nsw i64 %v33, 32 |
| 50 | %v35 = zext i16 %v29 to i64 |
| 51 | %v36 = or i64 %v35, %v34 |
| 52 | %v37 = zext i16 %v25 to i64 |
| 53 | %v38 = shl nuw nsw i64 %v37, 32 |
| 54 | %v39 = zext i16 %v22 to i64 |
| 55 | %v40 = or i64 %v39, %v38 |
| 56 | %v41 = tail call i64 @llvm.hexagon.S2.vtrunewh(i64 %v36, i64 %v40) |
| 57 | %v42 = getelementptr inbounds i64, i64* %v8, i32 1 |
| 58 | store i64 %v41, i64* %v8, align 8, !tbaa !0 |
| 59 | %v43 = trunc i64 %v14 to i32 |
| 60 | %v44 = tail call i64 @llvm.hexagon.S2.vzxthw(i32 %v43) |
| 61 | %v45 = trunc i64 %v17 to i32 |
| 62 | %v46 = tail call i64 @llvm.hexagon.S2.vzxthw(i32 %v45) |
| 63 | %v47 = trunc i64 %v44 to i32 |
| 64 | %v48 = lshr i64 %v44, 32 |
| 65 | %v49 = getelementptr inbounds i16, i16* %a3, i32 %v47 |
| 66 | %v50 = load i16, i16* %v49, align 2, !tbaa !3 |
| 67 | %v51 = trunc i64 %v48 to i32 |
| 68 | %v52 = getelementptr inbounds i16, i16* %a3, i32 %v51 |
| 69 | %v53 = load i16, i16* %v52, align 2, !tbaa !3 |
| 70 | %v54 = trunc i64 %v46 to i32 |
| 71 | %v55 = lshr i64 %v46, 32 |
| 72 | %v56 = getelementptr inbounds i16, i16* %a3, i32 %v54 |
| 73 | %v57 = load i16, i16* %v56, align 2, !tbaa !3 |
| 74 | %v58 = trunc i64 %v55 to i32 |
| 75 | %v59 = getelementptr inbounds i16, i16* %a3, i32 %v58 |
| 76 | %v60 = load i16, i16* %v59, align 2, !tbaa !3 |
| 77 | %v61 = zext i16 %v60 to i64 |
| 78 | %v62 = shl nuw nsw i64 %v61, 32 |
| 79 | %v63 = zext i16 %v57 to i64 |
| 80 | %v64 = or i64 %v63, %v62 |
| 81 | %v65 = zext i16 %v53 to i64 |
| 82 | %v66 = shl nuw nsw i64 %v65, 32 |
| 83 | %v67 = zext i16 %v50 to i64 |
| 84 | %v68 = or i64 %v67, %v66 |
| 85 | %v69 = tail call i64 @llvm.hexagon.S2.vtrunewh(i64 %v64, i64 %v68) |
| 86 | %v70 = getelementptr inbounds i64, i64* %v8, i32 2 |
| 87 | store i64 %v69, i64* %v42, align 8, !tbaa !0 |
| 88 | %v71 = add nsw i32 %v5, 1 |
| 89 | %v72 = icmp ult i32 %v71, %v0 |
| 90 | br i1 %v72, label %b2, label %b3 |
| 91 | |
| 92 | b3: ; preds = %b2, %b0 |
| 93 | ret void |
| 94 | } |
| 95 | |
| 96 | ; Function Attrs: nounwind readnone |
| 97 | declare i64 @llvm.hexagon.S2.vzxthw(i32) #1 |
| 98 | |
| 99 | ; Function Attrs: nounwind readnone |
| 100 | declare i64 @llvm.hexagon.S2.vtrunewh(i64, i64) #1 |
| 101 | |
| 102 | attributes #0 = { nounwind "target-cpu"="hexagonv60" } |
| 103 | attributes #1 = { nounwind readnone } |
| 104 | |
| 105 | !0 = !{!1, !1, i64 0} |
| 106 | !1 = !{!"omnipotent char", !2, i64 0} |
| 107 | !2 = !{!"Simple C/C++ TBAA"} |
| 108 | !3 = !{!4, !4, i64 0} |
| 109 | !4 = !{!"short", !1, i64 0} |