1; 2; There should be just a single copy of each loop when strictest mutiplier 3; candidates formula (unscaled candidates == 0) is enforced: 4 5; RUN: opt < %s -enable-unswitch-cost-multiplier=true \ 6; RUN: -unswitch-num-initial-unscaled-candidates=0 -unswitch-siblings-toplevel-div=1 \ 7; RUN: -passes='loop-mssa(simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | FileCheck %s --check-prefixes=LOOP1 8; 9; RUN: opt < %s -enable-unswitch-cost-multiplier=true \ 10; RUN: -unswitch-num-initial-unscaled-candidates=0 -unswitch-siblings-toplevel-div=16 \ 11; RUN: -passes='loop(simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | FileCheck %s --check-prefixes=LOOP1 12; 13; RUN: opt < %s -enable-unswitch-cost-multiplier=true \ 14; RUN: -unswitch-num-initial-unscaled-candidates=0 -unswitch-siblings-toplevel-div=1 \ 15; RUN: -passes='loop-mssa(simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | FileCheck %s --check-prefixes=LOOP1 16; 17; RUN: opt < %s -enable-unswitch-cost-multiplier=true \ 18; RUN: -unswitch-num-initial-unscaled-candidates=0 -unswitch-siblings-toplevel-div=16 \ 19; RUN: -passes='loop-mssa(simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | FileCheck %s --check-prefixes=LOOP1 20; 21; When we relax the candidates part of a multiplier formula 22; (unscaled candidates == 4) we start getting some unswitches, 23; which leads to siblings multiplier kicking in. 24; 25; The tests below also run licm, because it is needed to hoist out 26; loop-invariant freeze instructions, which otherwise may block further 27; unswitching. 28; 29; RUN: opt < %s -enable-unswitch-cost-multiplier=true \ 30; RUN: -unswitch-num-initial-unscaled-candidates=4 -unswitch-siblings-toplevel-div=1 \ 31; RUN: -passes='loop-mssa(licm,simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | \ 32; RUN: sort -b -k 1 | FileCheck %s --check-prefixes=LOOP-UNSCALE4-DIV1 33; 34; NB: sort -b is essential here and below, otherwise blanks might lead to different 35; order depending on locale. 36; 37; RUN: opt < %s -enable-unswitch-cost-multiplier=true \ 38; RUN: -unswitch-num-initial-unscaled-candidates=4 -unswitch-siblings-toplevel-div=2 \ 39; RUN: -passes='loop-mssa(licm,simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | \ 40; RUN: sort -b -k 1 | FileCheck %s --check-prefixes=LOOP-UNSCALE4-DIV2 41; 42; Get 43; 2^(num conds) == 2^5 = 32 44; loop nests when cost multiplier is disabled: 45; 46; RUN: opt < %s -enable-unswitch-cost-multiplier=false \ 47; RUN: -passes='loop-mssa(licm,simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | \ 48; RUN: sort -b -k 1 | FileCheck %s --check-prefixes=LOOP32 49; 50; Single loop nest, not unswitched 51; LOOP1: Loop at depth 1 containing: 52; LOOP1: Loop at depth 2 containing: 53; LOOP1: Loop at depth 3 containing: 54; LOOP1-NOT: Loop at depth {{[0-9]+}} containing: 55; 56; Half unswitched loop nests, with unscaled4 and div1 it gets less depth1 loops unswitched 57; since they have more cost. 58; LOOP-UNSCALE4-DIV1-COUNT-6: Loop at depth 1 containing: 59; LOOP-UNSCALE4-DIV1-COUNT-19: Loop at depth 2 containing: 60; LOOP-UNSCALE4-DIV1-COUNT-29: Loop at depth 3 containing: 61; LOOP-UNSCALE4-DIV1-NOT: Loop at depth {{[0-9]+}} containing: 62; 63; Half unswitched loop nests, with unscaled4 and div2 it gets more depth1 loops unswitched 64; as div2 kicks in. 65; LOOP-UNSCALE4-DIV2-COUNT-11: Loop at depth 1 containing: 66; LOOP-UNSCALE4-DIV2-COUNT-22: Loop at depth 2 containing: 67; LOOP-UNSCALE4-DIV2-COUNT-29: Loop at depth 3 containing: 68; LOOP-UNSCALE4-DIV2-NOT: Loop at depth {{[0-9]+}} containing: 69; 70; 32 loop nests, fully unswitched 71; LOOP32-COUNT-32: Loop at depth 1 containing: 72; LOOP32-COUNT-32: Loop at depth 2 containing: 73; LOOP32-COUNT-32: Loop at depth 3 containing: 74; LOOP32-NOT: Loop at depth {{[0-9]+}} containing: 75 76declare void @bar() 77 78define void @loop_nested3_conds5(i32* %addr, i1 %c1, i1 %c2, i1 %c3, i1 %c4, i1 %c5) { 79entry: 80 %addr1 = getelementptr i32, i32* %addr, i64 0 81 %addr2 = getelementptr i32, i32* %addr, i64 1 82 %addr3 = getelementptr i32, i32* %addr, i64 2 83 br label %outer 84outer: 85 %iv1 = phi i32 [0, %entry], [%iv1.next, %outer_latch] 86 %iv1.next = add i32 %iv1, 1 87 ;; skip nontrivial unswitch 88 call void @bar() 89 br label %middle 90middle: 91 %iv2 = phi i32 [0, %outer], [%iv2.next, %middle_latch] 92 %iv2.next = add i32 %iv2, 1 93 ;; skip nontrivial unswitch 94 call void @bar() 95 br label %loop 96loop: 97 %iv3 = phi i32 [0, %middle], [%iv3.next, %loop_latch] 98 %iv3.next = add i32 %iv3, 1 99 ;; skip nontrivial unswitch 100 call void @bar() 101 br i1 %c1, label %loop_next1_left, label %loop_next1_right 102loop_next1_left: 103 br label %loop_next1 104loop_next1_right: 105 br label %loop_next1 106 107loop_next1: 108 br i1 %c2, label %loop_next2_left, label %loop_next2_right 109loop_next2_left: 110 br label %loop_next2 111loop_next2_right: 112 br label %loop_next2 113 114loop_next2: 115 br i1 %c3, label %loop_next3_left, label %loop_next3_right 116loop_next3_left: 117 br label %loop_next3 118loop_next3_right: 119 br label %loop_next3 120 121loop_next3: 122 br i1 %c4, label %loop_next4_left, label %loop_next4_right 123loop_next4_left: 124 br label %loop_next4 125loop_next4_right: 126 br label %loop_next4 127 128loop_next4: 129 br i1 %c5, label %loop_latch_left, label %loop_latch_right 130loop_latch_left: 131 br label %loop_latch 132loop_latch_right: 133 br label %loop_latch 134 135loop_latch: 136 store volatile i32 0, i32* %addr1 137 %test_loop = icmp slt i32 %iv3, 50 138 br i1 %test_loop, label %loop, label %middle_latch 139middle_latch: 140 store volatile i32 0, i32* %addr2 141 %test_middle = icmp slt i32 %iv2, 50 142 br i1 %test_middle, label %middle, label %outer_latch 143outer_latch: 144 store volatile i32 0, i32* %addr3 145 %test_outer = icmp slt i32 %iv1, 50 146 br i1 %test_outer, label %outer, label %exit 147exit: 148 ret void 149} 150