| Krzysztof Parzyszek | 046090d | 2018-03-12 14:01:28 +0000 | [diff] [blame] | 1 | ; RUN: llc -march=hexagon -enable-pipeliner -stats \ |
| 2 | ; RUN: -pipeliner-prune-loop-carried=false -fp-contract=fast \ |
| 3 | ; RUN: -o /dev/null < %s 2>&1 | FileCheck %s --check-prefix=STATS |
| Krzysztof Parzyszek | ea2324f | 2018-03-12 15:20:36 +0000 | [diff] [blame] | 4 | ; REQUIRES: asserts |
| Krzysztof Parzyszek | 046090d | 2018-03-12 14:01:28 +0000 | [diff] [blame] | 5 | |
| 6 | ; That that we do not pipeline this loop. The recurrence is too large. If |
| 7 | ; we pipeline this example, that means we're not checking the complete |
| 8 | ; chain of dependences. |
| 9 | |
| 10 | ; STATS-NOT: 1 pipeliner - Number of loops software pipelined |
| 11 | |
| 12 | ; Function Attrs: nounwind |
| 13 | define void @f0(i32 %a0, i32 %a1, double %a2, double %a3, i8* %a4, i8* %a5, i8* %a6, i8* %a7, i8* %a8, [1000 x i8]* %a9, [1000 x i8]* %a10, [1000 x i8]* %a11) #0 { |
| 14 | b0: |
| 15 | br i1 undef, label %b1, label %b4 |
| 16 | |
| 17 | b1: ; preds = %b3, %b0 |
| 18 | br label %b2 |
| 19 | |
| 20 | b2: ; preds = %b2, %b1 |
| 21 | %v0 = phi i8* [ %v22, %b2 ], [ %a5, %b1 ] |
| 22 | %v1 = phi i8* [ %v23, %b2 ], [ %a6, %b1 ] |
| 23 | %v2 = phi i8* [ %v24, %b2 ], [ %a7, %b1 ] |
| 24 | %v3 = phi i32 [ %v20, %b2 ], [ 0, %b1 ] |
| 25 | %v4 = load i8, i8* %v0, align 1, !tbaa !0 |
| 26 | %v5 = zext i8 %v4 to i32 |
| 27 | %v6 = load i8, i8* %v1, align 1, !tbaa !0 |
| 28 | %v7 = sext i8 %v6 to i32 |
| 29 | %v8 = load i8, i8* %v2, align 1, !tbaa !0 |
| 30 | %v9 = sext i8 %v8 to i32 |
| 31 | %v10 = mul nsw i32 %v9, %v7 |
| 32 | %v11 = add nsw i32 %v10, %v5 |
| 33 | %v12 = trunc i32 %v11 to i8 |
| 34 | store i8 %v12, i8* undef, align 1, !tbaa !0 |
| 35 | %v13 = load i8, i8* %v2, align 1, !tbaa !0 |
| 36 | %v14 = sext i8 %v13 to i32 |
| 37 | %v15 = load i8, i8* undef, align 1, !tbaa !0 |
| 38 | %v16 = sext i8 %v15 to i32 |
| 39 | %v17 = mul nsw i32 %v16, %v14 |
| 40 | %v18 = add i32 %v17, %v11 |
| 41 | %v19 = trunc i32 %v18 to i8 |
| 42 | store i8 %v19, i8* %v0, align 1, !tbaa !0 |
| 43 | %v20 = add nsw i32 %v3, 1 |
| 44 | store i8 0, i8* undef, align 1, !tbaa !0 |
| 45 | %v21 = icmp eq i32 %v20, undef |
| 46 | %v22 = getelementptr i8, i8* %v0, i32 1 |
| 47 | %v23 = getelementptr i8, i8* %v1, i32 1 |
| 48 | %v24 = getelementptr i8, i8* %v2, i32 1 |
| 49 | br i1 %v21, label %b3, label %b2 |
| 50 | |
| 51 | b3: ; preds = %b2 |
| 52 | tail call void @f1(i32 %a1, i8* %a4, i8* %a5, i8* %a6, i8* %a7, i8* %a8, [1000 x i8]* %a9, [1000 x i8]* %a10, [1000 x i8]* %a11, i8 signext 1) #2 |
| 53 | br i1 undef, label %b4, label %b1 |
| 54 | |
| 55 | b4: ; preds = %b3, %b0 |
| 56 | ret void |
| 57 | } |
| 58 | |
| 59 | declare void @f1(i32, i8*, i8*, i8*, i8*, i8*, [1000 x i8]*, [1000 x i8]*, [1000 x i8]*, i8 signext) #1 |
| 60 | |
| 61 | attributes #0 = { nounwind "target-cpu"="hexagonv55" } |
| 62 | attributes #1 = { "target-cpu"="hexagonv55" } |
| 63 | attributes #2 = { nounwind } |
| 64 | |
| 65 | !0 = !{!1, !1, i64 0} |
| 66 | !1 = !{!"omnipotent char", !2, i64 0} |
| 67 | !2 = !{!"Simple C/C++ TBAA"} |