| Dehao Chen | cc76344 | 2016-12-30 00:50:28 +0000 | [diff] [blame^] | 1 | ; RUN: opt < %s -S -loop-unroll -unroll-runtime -unroll-threshold=40 -unroll-max-percent-threshold-boost=100 | FileCheck %s |
| Dehao Chen | 41d72a8 | 2016-11-17 01:17:02 +0000 | [diff] [blame] | 2 | |
| 3 | @known_constant = internal unnamed_addr constant [9 x i32] [i32 0, i32 -1, i32 0, i32 -1, i32 5, i32 -1, i32 0, i32 -1, i32 0], align 16 |
| 4 | |
| 5 | ; CHECK-LABEL: @bar_prof |
| 6 | ; CHECK: loop.prol: |
| 7 | ; CHECK: loop: |
| 8 | ; CHECK: %mul = mul |
| 9 | ; CHECK: %mul.1 = mul |
| 10 | ; CHECK: %mul.2 = mul |
| 11 | ; CHECK: %mul.3 = mul |
| 12 | define i32 @bar_prof(i32* noalias nocapture readonly %src, i64 %c) !prof !1 { |
| 13 | entry: |
| 14 | br label %loop |
| 15 | |
| 16 | loop: |
| 17 | %iv = phi i64 [ 0, %entry ], [ %inc, %loop ] |
| 18 | %r = phi i32 [ 0, %entry ], [ %add, %loop ] |
| 19 | %arrayidx = getelementptr inbounds i32, i32* %src, i64 %iv |
| 20 | %src_element = load i32, i32* %arrayidx, align 4 |
| 21 | %array_const_idx = getelementptr inbounds [9 x i32], [9 x i32]* @known_constant, i64 0, i64 %iv |
| 22 | %const_array_element = load i32, i32* %array_const_idx, align 4 |
| 23 | %mul = mul nsw i32 %src_element, %const_array_element |
| 24 | %add = add nsw i32 %mul, %r |
| 25 | %inc = add nuw nsw i64 %iv, 1 |
| 26 | %exitcond86.i = icmp eq i64 %inc, %c |
| 27 | br i1 %exitcond86.i, label %loop.end, label %loop, !prof !2 |
| 28 | |
| 29 | loop.end: |
| 30 | %r.lcssa = phi i32 [ %r, %loop ] |
| 31 | ret i32 %r.lcssa |
| 32 | } |
| 33 | |
| 34 | ; CHECK-LABEL: @bar_prof_flat |
| 35 | ; CHECK-NOT: loop.prol |
| 36 | define i32 @bar_prof_flat(i32* noalias nocapture readonly %src, i64 %c) !prof !1 { |
| 37 | entry: |
| 38 | br label %loop |
| 39 | |
| 40 | loop: |
| 41 | %iv = phi i64 [ 0, %entry ], [ %inc, %loop ] |
| 42 | %r = phi i32 [ 0, %entry ], [ %add, %loop ] |
| 43 | %arrayidx = getelementptr inbounds i32, i32* %src, i64 %iv |
| 44 | %src_element = load i32, i32* %arrayidx, align 4 |
| 45 | %array_const_idx = getelementptr inbounds [9 x i32], [9 x i32]* @known_constant, i64 0, i64 %iv |
| 46 | %const_array_element = load i32, i32* %array_const_idx, align 4 |
| 47 | %mul = mul nsw i32 %src_element, %const_array_element |
| 48 | %add = add nsw i32 %mul, %r |
| 49 | %inc = add nuw nsw i64 %iv, 1 |
| 50 | %exitcond86.i = icmp eq i64 %inc, %c |
| 51 | br i1 %exitcond86.i, label %loop, label %loop.end, !prof !2 |
| 52 | |
| 53 | loop.end: |
| 54 | %r.lcssa = phi i32 [ %r, %loop ] |
| 55 | ret i32 %r.lcssa |
| 56 | } |
| 57 | |
| 58 | !1 = !{!"function_entry_count", i64 1} |
| 59 | !2 = !{!"branch_weights", i32 1, i32 1000} |