105 lines · plain
1; RUN: opt -passes=loop-vectorize -force-vector-width=8 -force-vector-interleave=2 -disable-output -debug -S %s 2>&1 | FileCheck --check-prefixes=CHECK %s2 3target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128"4 5; REQUIRES: asserts6 7; Check if the vector loop condition can be simplified to true for a given8; VF/IC combination.9define void @test_tc_less_than_16(ptr %A, i64 %N) {10; CHECK: LV: Scalarizing: %cmp =11; CHECK: VPlan 'Initial VPlan for VF={8},UF>=1' {12; CHECK-NEXT: Live-in vp<[[VF:%.+]]> = VF13; CHECK-NEXT: Live-in vp<[[VFxUF:%.+]]> = VF * UF14; CHECK-NEXT: Live-in vp<[[VTC:%.+]]> = vector-trip-count15; CHECK-NEXT: ir<16> = original trip-count16; CHECK-EMPTY:17; CHECK-NEXT: ir-bb<entry>:18; CHECK-NEXT: Successor(s): scalar.ph, vector.ph19; CHECK-EMPTY:20; CHECK-NEXT: vector.ph:21; CHECK-NEXT: vp<[[END1:%.+]]> = DERIVED-IV ir<16> + vp<[[VTC]]> * ir<-1>22; CHECK-NEXT: vp<[[END2:%.+]]> = DERIVED-IV ir<%A> + vp<[[VTC]]> * ir<1>23; CHECK-NEXT: Successor(s): vector loop24; CHECK-EMPTY:25; CHECK-NEXT: <x1> vector loop: {26; CHECK-NEXT: vector.body:27; CHECK-NEXT: EMIT vp<[[CAN_IV:%.+]]> = CANONICAL-INDUCTION ir<0>, vp<[[CAN_IV_NEXT:%.+]]>28; CHECK-NEXT: vp<[[STEPS:%.+]]> = SCALAR-STEPS vp<[[CAN_IV]]>, ir<1>, vp<[[VF]]>29; CHECK-NEXT: EMIT vp<[[PADD:%.+]]> = ptradd ir<%A>, vp<[[STEPS]]>30; CHECK-NEXT: vp<[[VPTR:%.]]> = vector-pointer vp<[[PADD]]>31; CHECK-NEXT: WIDEN ir<%l> = load vp<[[VPTR]]>32; CHECK-NEXT: WIDEN ir<%add> = add nsw ir<%l>, ir<10>33; CHECK-NEXT: vp<[[VPTR2:%.+]]> = vector-pointer vp<[[PADD]]>34; CHECK-NEXT: WIDEN store vp<[[VPTR2]]>, ir<%add>35; CHECK-NEXT: EMIT vp<[[CAN_IV_NEXT]]> = add nuw vp<[[CAN_IV:%.+]]>, vp<[[VFxUF]]>36; CHECK-NEXT: EMIT branch-on-count vp<[[CAN_IV_NEXT]]>, vp<[[VTC]]>37; CHECK-NEXT: No successors38; CHECK-NEXT: }39; CHECK-NEXT: Successor(s): middle.block40; CHECK-EMPTY:41; CHECK-NEXT: middle.block:42; CHECK-NEXT: EMIT vp<[[C:%.+]]> = icmp eq ir<16>, vp<[[VTC]]>43; CHECK-NEXT: EMIT branch-on-cond vp<[[C]]>44; CHECK-NEXT: Successor(s): ir-bb<exit>, scalar.ph45; CHECK-EMPTY:46; CHECK-NEXT: ir-bb<exit>:47; CHECK-NEXT: No successors48; CHECK-EMPTY:49; CHECK-NEXT: scalar.ph:50; CHECK-NEXT: EMIT-SCALAR vp<[[RESUME1:%.+]]> = phi [ vp<[[END1]]>, middle.block ], [ ir<16>, ir-bb<entry> ]51; CHECK-NEXT: EMIT-SCALAR vp<[[RESUME2:%.+]]>.1 = phi [ vp<[[END2]]>, middle.block ], [ ir<%A>, ir-bb<entry> ]52; CHECK-NEXT: Successor(s): ir-bb<loop>53; CHECK-EMPTY:54; CHECK-NEXT: ir-bb<loop>:55; CHECK-NEXT: IR %iv = phi i64 [ 16, %entry ], [ %iv.next, %loop ] (extra operand: vp<[[RESUME1]]> from scalar.ph)56; CHECK-NEXT: IR %p.src = phi ptr [ %A, %entry ], [ %p.src.next, %loop ] (extra operand: vp<[[RESUME2]]>.1 from scalar.ph)57; CHECK: IR %cmp = icmp eq i64 %iv.next, 058; CHECK-NEXT: No successors59; CHECK-NEXT: }60;61; CHECK: Executing best plan with VF=8, UF=262; CHECK-NEXT: VPlan 'Final VPlan for VF={8},UF={2}' {63; CHECK-NEXT: Live-in ir<16> = original trip-count64; CHECK-EMPTY:65; CHECK-NEXT: ir-bb<entry>:66; CHECK-NEXT: Successor(s): vector.ph67; CHECK-EMPTY:68; CHECK-NEXT: vector.ph:69; CHECK-NEXT: Successor(s): vector.body70; CHECK-EMPTY:71; CHECK-NEXT: vector.body:72; CHECK-NEXT: vp<[[VPTR2:%.]]> = vector-pointer ir<%A>, ir<1>73; CHECK-NEXT: WIDEN ir<%l> = load ir<%A>74; CHECK-NEXT: WIDEN ir<%l>.1 = load vp<[[VPTR2]]>75; CHECK-NEXT: WIDEN ir<%add> = add nsw ir<%l>, ir<10>76; CHECK-NEXT: WIDEN ir<%add>.1 = add nsw ir<%l>.1, ir<10>77; CHECK-NEXT: WIDEN store ir<%A>, ir<%add>78; CHECK-NEXT: WIDEN store vp<[[VPTR2]]>, ir<%add>.179; CHECK-NEXT: Successor(s): middle.block80; CHECK-EMPTY:81; CHECK-NEXT: middle.block:82; CHECK-NEXT: Successor(s): ir-bb<exit>83; CHECK-EMPTY:84; CHECK-NEXT: ir-bb<exit>:85; CHECK-NEXT: No successors86; CHECK-NEXT: }87;88entry:89 br label %loop90 91loop:92 %iv = phi i64 [ 16, %entry ], [ %iv.next, %loop ]93 %p.src = phi ptr [ %A, %entry ], [ %p.src.next, %loop ]94 %p.src.next = getelementptr inbounds i8, ptr %p.src, i64 195 %l = load i8, ptr %p.src, align 196 %add = add nsw i8 %l, 1097 store i8 %add, ptr %p.src98 %iv.next = add nsw i64 %iv, -199 %cmp = icmp eq i64 %iv.next, 0100 br i1 %cmp, label %exit, label %loop101 102exit:103 ret void104}105