| Krzysztof Parzyszek | 046090d | 2018-03-12 14:01:28 +0000 | [diff] [blame] | 1 | ; RUN: llc -march=hexagon < %s | FileCheck %s |
| 2 | |
| 3 | ; Test that we generate multiple using upper result. |
| 4 | |
| 5 | ; CHECK: = mpy(r{{[0-9]+}},r{{[0-9]+}}) |
| 6 | ; CHECK: = mpy(r{{[0-9]+}},r{{[0-9]+}}) |
| 7 | ; CHECK: = mpy(r{{[0-9]+}},r{{[0-9]+}}) |
| 8 | ; CHECK: = mpy(r{{[0-9]+}},r{{[0-9]+}}) |
| 9 | |
| 10 | @g0 = external constant [1152 x i32], align 8 |
| 11 | @g1 = external constant [2 x i32], align 8 |
| 12 | |
| 13 | ; Function Attrs: nounwind |
| 14 | define void @f0(i32* nocapture readonly %a0, i32* %a1, i32* nocapture %a2, i32 %a3, i32 %a4) #0 { |
| 15 | b0: |
| 16 | %v0 = getelementptr inbounds i32, i32* %a0, i32 512 |
| 17 | %v1 = getelementptr inbounds i32, i32* %a0, i32 511 |
| 18 | %v2 = getelementptr inbounds i32, i32* %a2, i32 1023 |
| 19 | %v3 = getelementptr inbounds i32, i32* %a1, i32 1023 |
| 20 | br label %b1 |
| 21 | |
| 22 | b1: ; preds = %b0 |
| 23 | %v4 = load i32, i32* getelementptr inbounds ([2 x i32], [2 x i32]* @g1, i32 0, i32 1), align 4 |
| 24 | %v5 = getelementptr inbounds [1152 x i32], [1152 x i32]* @g0, i32 0, i32 %v4 |
| 25 | br label %b2 |
| 26 | |
| 27 | b2: ; preds = %b1 |
| 28 | br label %b3 |
| 29 | |
| 30 | b3: ; preds = %b3, %b2 |
| 31 | %v6 = phi i32* [ %v30, %b3 ], [ %a2, %b2 ] |
| 32 | %v7 = phi i32* [ %v44, %b3 ], [ %a1, %b2 ] |
| 33 | %v8 = phi i32* [ %v17, %b3 ], [ %v0, %b2 ] |
| 34 | %v9 = phi i32* [ %v34, %b3 ], [ %v1, %b2 ] |
| 35 | %v10 = phi i32* [ %v40, %b3 ], [ %v3, %b2 ] |
| 36 | %v11 = phi i32* [ %v33, %b3 ], [ %v2, %b2 ] |
| 37 | %v12 = phi i32* [ %v15, %b3 ], [ %v5, %b2 ] |
| 38 | %v13 = getelementptr inbounds i32, i32* %v12, i32 1 |
| 39 | %v14 = load i32, i32* %v12, align 4 |
| 40 | %v15 = getelementptr inbounds i32, i32* %v12, i32 2 |
| 41 | %v16 = load i32, i32* %v13, align 4 |
| 42 | %v17 = getelementptr inbounds i32, i32* %v8, i32 1 |
| 43 | %v18 = load i32, i32* %v8, align 4 |
| 44 | %v19 = sext i32 %v14 to i64 |
| 45 | %v20 = sext i32 %v18 to i64 |
| 46 | %v21 = mul nsw i64 %v20, %v19 |
| 47 | %v22 = lshr i64 %v21, 32 |
| 48 | %v23 = trunc i64 %v22 to i32 |
| 49 | %v24 = sext i32 %v16 to i64 |
| 50 | %v25 = mul nsw i64 %v20, %v24 |
| 51 | %v26 = lshr i64 %v25, 32 |
| 52 | %v27 = trunc i64 %v26 to i32 |
| 53 | %v28 = load i32, i32* %v7, align 4 |
| 54 | %v29 = sub nsw i32 %v28, %v23 |
| 55 | %v30 = getelementptr inbounds i32, i32* %v6, i32 1 |
| 56 | store i32 %v29, i32* %v6, align 4 |
| 57 | %v31 = load i32, i32* %v10, align 4 |
| 58 | %v32 = add nsw i32 %v27, %v31 |
| 59 | %v33 = getelementptr inbounds i32, i32* %v11, i32 -1 |
| 60 | store i32 %v32, i32* %v11, align 4 |
| 61 | %v34 = getelementptr inbounds i32, i32* %v9, i32 -1 |
| 62 | %v35 = load i32, i32* %v9, align 4 |
| 63 | %v36 = sext i32 %v35 to i64 |
| 64 | %v37 = mul nsw i64 %v36, %v19 |
| 65 | %v38 = lshr i64 %v37, 32 |
| 66 | %v39 = trunc i64 %v38 to i32 |
| 67 | %v40 = getelementptr inbounds i32, i32* %v10, i32 -1 |
| 68 | store i32 %v39, i32* %v10, align 4 |
| 69 | %v41 = mul nsw i64 %v36, %v24 |
| 70 | %v42 = lshr i64 %v41, 32 |
| 71 | %v43 = trunc i64 %v42 to i32 |
| 72 | %v44 = getelementptr inbounds i32, i32* %v7, i32 1 |
| 73 | store i32 %v43, i32* %v7, align 4 |
| 74 | %v45 = icmp ult i32* %v44, %v40 |
| 75 | br i1 %v45, label %b3, label %b4 |
| 76 | |
| 77 | b4: ; preds = %b3 |
| 78 | br label %b6 |
| 79 | |
| 80 | b5: ; No predecessors! |
| 81 | br label %b6 |
| 82 | |
| 83 | b6: ; preds = %b5, %b4 |
| 84 | ret void |
| 85 | } |
| 86 | |
| 87 | attributes #0 = { nounwind "target-cpu"="hexagonv55" } |