Eric Christopher | cee313d | 2019-04-17 04:52:47 +0000 | [diff] [blame^] | 1 | ; RUN: opt < %s -S -loop-unroll -unroll-runtime=true -unroll-count=16 | FileCheck --check-prefix=UNROLL-16 %s |
| 2 | ; RUN: opt < %s -S -loop-unroll -unroll-runtime=true -unroll-count=4 | FileCheck --check-prefix=UNROLL-4 %s |
| 3 | |
| 4 | ; RUN: opt < %s -S -passes='require<opt-remark-emit>,unroll' -unroll-runtime=true -unroll-count=16 | FileCheck --check-prefix=UNROLL-16 %s |
| 5 | ; RUN: opt < %s -S -passes='require<opt-remark-emit>,unroll' -unroll-runtime=true -unroll-count=4 | FileCheck --check-prefix=UNROLL-4 %s |
| 6 | |
| 7 | ; Given that the trip-count of this loop is a 3-bit value, we cannot |
| 8 | ; safely unroll it with a count of anything more than 8. |
| 9 | |
| 10 | define i3 @test(i3* %a, i3 %n) { |
| 11 | ; UNROLL-16-LABEL: @test( |
| 12 | ; UNROLL-4-LABEL: @test( |
| 13 | entry: |
| 14 | %cmp1 = icmp eq i3 %n, 0 |
| 15 | br i1 %cmp1, label %for.end, label %for.body |
| 16 | |
| 17 | for.body: ; preds = %for.body, %entry |
| 18 | ; UNROLL-16-LABEL: for.body: |
| 19 | ; UNROLL-4-LABEL: for.body: |
| 20 | %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ] |
| 21 | %sum.02 = phi i3 [ %add, %for.body ], [ 0, %entry ] |
| 22 | %arrayidx = getelementptr inbounds i3, i3* %a, i64 %indvars.iv |
| 23 | |
| 24 | ; UNROLL-16-LABEL: for.body |
| 25 | ; UNROLL-16-LABEL: getelementptr |
| 26 | ; UNROLL-16-LABEL-NOT: getelementptr |
| 27 | |
| 28 | ; UNROLL-4-LABEL: getelementptr |
| 29 | ; UNROLL-4-LABEL: getelementptr |
| 30 | ; UNROLL-4-LABEL: getelementptr |
| 31 | ; UNROLL-4-LABEL: getelementptr |
| 32 | |
| 33 | %0 = load i3, i3* %arrayidx |
| 34 | %add = add nsw i3 %0, %sum.02 |
| 35 | %indvars.iv.next = add i64 %indvars.iv, 1 |
| 36 | %lftr.wideiv = trunc i64 %indvars.iv.next to i3 |
| 37 | %exitcond = icmp eq i3 %lftr.wideiv, %n |
| 38 | br i1 %exitcond, label %for.end, label %for.body |
| 39 | |
| 40 | ; UNROLL-16-LABEL: for.end |
| 41 | ; UNROLL-4-LABEL: for.end |
| 42 | |
| 43 | ; UNROLL-16-NOT: for.body.epil: |
| 44 | ; UNROLL-4: for.body.epil: |
| 45 | |
| 46 | for.end: ; preds = %for.body, %entry |
| 47 | %sum.0.lcssa = phi i3 [ 0, %entry ], [ %add, %for.body ] |
| 48 | ret i3 %sum.0.lcssa |
| 49 | } |