1 ; RUN: opt < %s -S -loop-unroll -unroll-max-iteration-count-to-analyze=1000 -unroll-threshold=10 -unroll-max-percent-threshold-boost=200 | FileCheck %s 2 ; RUN: opt < %s -S -passes='require<opt-remark-emit>,loop(unroll-full)' -unroll-max-iteration-count-to-analyze=1000 -unroll-threshold=10 -unroll-max-percent-threshold-boost=200 | FileCheck %s 3 target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128" 4 5 @unknown_global = internal unnamed_addr global [9 x i32] [i32 0, i32 -1, i32 0, i32 -1, i32 5, i32 -1, i32 0, i32 -1, i32 0], align 16 6 @weak_constant = weak 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 7 8 ; Though @unknown_global is initialized with constant values, we can't consider 9 ; it as a constant, so we shouldn't unroll the loop. 10 ; CHECK-LABEL: @foo 11 ; CHECK: %array_const_idx = getelementptr inbounds [9 x i32], [9 x i32]* @unknown_global, i64 0, i64 %iv 12 define i32 @foo(i32* noalias nocapture readonly %src) { 13 entry: 14 br label %loop 15 16 loop: ; preds = %loop, %entry 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]* @unknown_global, 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, 9 27 br i1 %exitcond86.i, label %loop.end, label %loop 28 29 loop.end: ; preds = %loop 30 %r.lcssa = phi i32 [ %r, %loop ] 31 ret i32 %r.lcssa 32 } 33 34 ; Similarly, we can't consider 'weak' symbols as a known constant value, so we 35 ; shouldn't unroll the loop. 36 ; CHECK-LABEL: @foo2 37 ; CHECK: %array_const_idx = getelementptr inbounds [9 x i32], [9 x i32]* @weak_constant, i64 0, i64 %iv 38 define i32 @foo2(i32* noalias nocapture readonly %src) { 39 entry: 40 br label %loop 41 42 loop: ; preds = %loop, %entry 43 %iv = phi i64 [ 0, %entry ], [ %inc, %loop ] 44 %r = phi i32 [ 0, %entry ], [ %add, %loop ] 45 %arrayidx = getelementptr inbounds i32, i32* %src, i64 %iv 46 %src_element = load i32, i32* %arrayidx, align 4 47 %array_const_idx = getelementptr inbounds [9 x i32], [9 x i32]* @weak_constant, i64 0, i64 %iv 48 %const_array_element = load i32, i32* %array_const_idx, align 4 49 %mul = mul nsw i32 %src_element, %const_array_element 50 %add = add nsw i32 %mul, %r 51 %inc = add nuw nsw i64 %iv, 1 52 %exitcond86.i = icmp eq i64 %inc, 9 53 br i1 %exitcond86.i, label %loop.end, label %loop 54 55 loop.end: ; preds = %loop 56 %r.lcssa = phi i32 [ %r, %loop ] 57 ret i32 %r.lcssa 58 } 59 60 ; In this case the loaded value is used only to control branch. 61 ; If we missed that, we could've thought that it's unused and unrolling would 62 ; clean up almost entire loop. Make sure that we do not unroll such loop. 63 ; CHECK-LABEL: @foo3 64 ; CHECK: br i1 %exitcond, label %loop.end, label %loop.header 65 define i32 @foo3(i32* noalias nocapture readonly %src) { 66 entry: 67 br label %loop.header 68 69 loop.header: 70 %iv = phi i64 [ 0, %entry ], [ %inc, %loop.latch ] 71 %r1 = phi i32 [ 0, %entry ], [ %r3, %loop.latch ] 72 %arrayidx = getelementptr inbounds i32, i32* %src, i64 %iv 73 %src_element = load i32, i32* %arrayidx, align 4 74 %cmp = icmp eq i32 0, %src_element 75 br i1 %cmp, label %loop.if, label %loop.latch 76 77 loop.if: 78 %r2 = add i32 %r1, 1 79 br label %loop.latch 80 81 loop.latch: 82 %r3 = phi i32 [%r1, %loop.header], [%r2, %loop.if] 83 %inc = add nuw nsw i64 %iv, 1 84 %exitcond = icmp eq i64 %inc, 9 85 br i1 %exitcond, label %loop.end, label %loop.header 86 87 loop.end: 88 %r.lcssa = phi i32 [ %r3, %loop.latch ] 89 ret i32 %r.lcssa 90 } 91
