1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5 2; RUN: opt -p loop-vectorize -S %s | FileCheck %s 3 4target datalayout = "e-m:e-p:64:64-i64:64-i128:128-n32:64-S128" 5target triple = "riscv64-unknown-linux-gnu" 6 7; Test case for https://github.com/llvm/llvm-project/issues/106417. 8define void @skip_free_iv_truncate(i16 %x, ptr %A) #0 { 9; CHECK-LABEL: define void @skip_free_iv_truncate( 10; CHECK-SAME: i16 [[X:%.*]], ptr [[A:%.*]]) #[[ATTR0:[0-9]+]] { 11; CHECK-NEXT: [[ENTRY:.*]]: 12; CHECK-NEXT: [[X_I32:%.*]] = sext i16 [[X]] to i32 13; CHECK-NEXT: [[X_I64:%.*]] = sext i16 [[X]] to i64 14; CHECK-NEXT: [[INVARIANT_GEP:%.*]] = getelementptr i8, ptr [[A]], i64 -8 15; CHECK-NEXT: [[SMAX20:%.*]] = call i64 @llvm.smax.i64(i64 [[X_I64]], i64 99) 16; CHECK-NEXT: [[TMP0:%.*]] = sub i64 [[SMAX20]], [[X_I64]] 17; CHECK-NEXT: [[UMIN21:%.*]] = call i64 @llvm.umin.i64(i64 [[TMP0]], i64 1) 18; CHECK-NEXT: [[TMP1:%.*]] = sub i64 [[SMAX20]], [[UMIN21]] 19; CHECK-NEXT: [[TMP2:%.*]] = sub i64 [[TMP1]], [[X_I64]] 20; CHECK-NEXT: [[TMP3:%.*]] = udiv i64 [[TMP2]], 3 21; CHECK-NEXT: [[TMP4:%.*]] = add i64 [[UMIN21]], [[TMP3]] 22; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[TMP4]], 1 23; CHECK-NEXT: [[TMP6:%.*]] = call i64 @llvm.vscale.i64() 24; CHECK-NEXT: [[TMP7:%.*]] = mul i64 [[TMP6]], 8 25; CHECK-NEXT: [[TMP8:%.*]] = call i64 @llvm.umax.i64(i64 128, i64 [[TMP7]]) 26; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ule i64 [[TMP5]], [[TMP8]] 27; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_MEMCHECK:.*]] 28; CHECK: [[VECTOR_MEMCHECK]]: 29; CHECK-NEXT: [[TMP31:%.*]] = shl nsw i64 [[X_I64]], 1 30; CHECK-NEXT: [[SCEVGEP9:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP31]] 31; CHECK-NEXT: [[SMAX10:%.*]] = call i64 @llvm.smax.i64(i64 [[X_I64]], i64 99) 32; CHECK-NEXT: [[TMP32:%.*]] = sub i64 [[SMAX10]], [[X_I64]] 33; CHECK-NEXT: [[UMIN11:%.*]] = call i64 @llvm.umin.i64(i64 [[TMP32]], i64 1) 34; CHECK-NEXT: [[TMP33:%.*]] = sub i64 [[SMAX10]], [[UMIN11]] 35; CHECK-NEXT: [[TMP34:%.*]] = sub i64 [[TMP33]], [[X_I64]] 36; CHECK-NEXT: [[TMP35:%.*]] = udiv i64 [[TMP34]], 3 37; CHECK-NEXT: [[TMP36:%.*]] = add i64 [[UMIN11]], [[TMP35]] 38; CHECK-NEXT: [[TMP37:%.*]] = mul i64 [[TMP36]], 6 39; CHECK-NEXT: [[TMP38:%.*]] = add i64 [[TMP37]], [[TMP31]] 40; CHECK-NEXT: [[TMP39:%.*]] = add i64 [[TMP38]], 2 41; CHECK-NEXT: [[SCEVGEP12:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP39]] 42; CHECK-NEXT: [[TMP40:%.*]] = shl nsw i64 [[X_I64]], 3 43; CHECK-NEXT: [[SCEVGEP13:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP40]] 44; CHECK-NEXT: [[TMP41:%.*]] = mul i64 [[TMP36]], 24 45; CHECK-NEXT: [[TMP42:%.*]] = add i64 [[TMP41]], [[TMP40]] 46; CHECK-NEXT: [[TMP43:%.*]] = add i64 [[TMP42]], 8 47; CHECK-NEXT: [[SCEVGEP14:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP43]] 48; CHECK-NEXT: [[TMP44:%.*]] = add nsw i64 [[TMP40]], -8 49; CHECK-NEXT: [[SCEVGEP15:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP44]] 50; CHECK-NEXT: [[SCEVGEP16:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP42]] 51; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[SCEVGEP9]], [[SCEVGEP14]] 52; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[SCEVGEP13]], [[SCEVGEP12]] 53; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] 54; CHECK-NEXT: [[BOUND017:%.*]] = icmp ult ptr [[SCEVGEP9]], [[SCEVGEP16]] 55; CHECK-NEXT: [[BOUND118:%.*]] = icmp ult ptr [[SCEVGEP15]], [[SCEVGEP12]] 56; CHECK-NEXT: [[FOUND_CONFLICT19:%.*]] = and i1 [[BOUND017]], [[BOUND118]] 57; CHECK-NEXT: [[CONFLICT_RDX:%.*]] = or i1 [[FOUND_CONFLICT]], [[FOUND_CONFLICT19]] 58; CHECK-NEXT: br i1 [[CONFLICT_RDX]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]] 59; CHECK: [[VECTOR_PH]]: 60; CHECK-NEXT: [[TMP45:%.*]] = call i64 @llvm.vscale.i64() 61; CHECK-NEXT: [[TMP46:%.*]] = mul i64 [[TMP45]], 8 62; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP5]], [[TMP46]] 63; CHECK-NEXT: [[TMP47:%.*]] = icmp eq i64 [[N_MOD_VF]], 0 64; CHECK-NEXT: [[TMP48:%.*]] = select i1 [[TMP47]], i64 [[TMP46]], i64 [[N_MOD_VF]] 65; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP5]], [[TMP48]] 66; CHECK-NEXT: [[TMP51:%.*]] = call i64 @llvm.vscale.i64() 67; CHECK-NEXT: [[TMP52:%.*]] = mul i64 [[TMP51]], 8 68; CHECK-NEXT: [[TMP49:%.*]] = mul i64 [[N_VEC]], 3 69; CHECK-NEXT: [[IND_END:%.*]] = add i64 [[X_I64]], [[TMP49]] 70; CHECK-NEXT: [[DOTCAST:%.*]] = trunc i64 [[N_VEC]] to i32 71; CHECK-NEXT: [[TMP50:%.*]] = mul i32 [[DOTCAST]], 3 72; CHECK-NEXT: [[IND_END22:%.*]] = add i32 [[X_I32]], [[TMP50]] 73; CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 8 x i64> poison, i64 [[X_I64]], i64 0 74; CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <vscale x 8 x i64> [[DOTSPLATINSERT]], <vscale x 8 x i64> poison, <vscale x 8 x i32> zeroinitializer 75; CHECK-NEXT: [[TMP53:%.*]] = call <vscale x 8 x i64> @llvm.stepvector.nxv8i64() 76; CHECK-NEXT: [[TMP55:%.*]] = mul <vscale x 8 x i64> [[TMP53]], splat (i64 3) 77; CHECK-NEXT: [[INDUCTION:%.*]] = add <vscale x 8 x i64> [[DOTSPLAT]], [[TMP55]] 78; CHECK-NEXT: [[TMP58:%.*]] = mul i64 3, [[TMP52]] 79; CHECK-NEXT: [[DOTSPLATINSERT24:%.*]] = insertelement <vscale x 8 x i64> poison, i64 [[TMP58]], i64 0 80; CHECK-NEXT: [[DOTSPLAT25:%.*]] = shufflevector <vscale x 8 x i64> [[DOTSPLATINSERT24]], <vscale x 8 x i64> poison, <vscale x 8 x i32> zeroinitializer 81; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] 82; CHECK: [[VECTOR_BODY]]: 83; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 84; CHECK-NEXT: [[VEC_IND:%.*]] = phi <vscale x 8 x i64> [ [[INDUCTION]], %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ] 85; CHECK-NEXT: [[TMP59:%.*]] = getelementptr i16, ptr [[A]], <vscale x 8 x i64> [[VEC_IND]] 86; CHECK-NEXT: call void @llvm.masked.scatter.nxv8i16.nxv8p0(<vscale x 8 x i16> zeroinitializer, <vscale x 8 x ptr> [[TMP59]], i32 2, <vscale x 8 x i1> splat (i1 true)), !alias.scope [[META0:![0-9]+]], !noalias [[META3:![0-9]+]] 87; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP52]] 88; CHECK-NEXT: [[VEC_IND_NEXT]] = add <vscale x 8 x i64> [[VEC_IND]], [[DOTSPLAT25]] 89; CHECK-NEXT: [[TMP60:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 90; CHECK-NEXT: br i1 [[TMP60]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] 91; CHECK: [[MIDDLE_BLOCK]]: 92; CHECK-NEXT: br label %[[SCALAR_PH]] 93; CHECK: [[SCALAR_PH]]: 94; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], %[[MIDDLE_BLOCK]] ], [ [[X_I64]], %[[VECTOR_MEMCHECK]] ], [ [[X_I64]], %[[ENTRY]] ] 95; CHECK-NEXT: [[BC_RESUME_VAL13:%.*]] = phi i32 [ [[IND_END22]], %[[MIDDLE_BLOCK]] ], [ [[X_I32]], %[[VECTOR_MEMCHECK]] ], [ [[X_I32]], %[[ENTRY]] ] 96; CHECK-NEXT: br label %[[LOOP:.*]] 97; CHECK: [[LOOP]]: 98; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] 99; CHECK-NEXT: [[IV_CONV:%.*]] = phi i32 [ [[BC_RESUME_VAL13]], %[[SCALAR_PH]] ], [ [[TMP64:%.*]], %[[LOOP]] ] 100; CHECK-NEXT: [[GEP_I64:%.*]] = getelementptr i64, ptr [[A]], i64 [[IV]] 101; CHECK-NEXT: [[TMP61:%.*]] = load i64, ptr [[GEP_I64]], align 8 102; CHECK-NEXT: [[TMP62:%.*]] = sext i32 [[IV_CONV]] to i64 103; CHECK-NEXT: [[GEP_CONV:%.*]] = getelementptr i64, ptr [[INVARIANT_GEP]], i64 [[TMP62]] 104; CHECK-NEXT: [[TMP63:%.*]] = load i64, ptr [[GEP_CONV]], align 8 105; CHECK-NEXT: [[GEP_I16:%.*]] = getelementptr i16, ptr [[A]], i64 [[IV]] 106; CHECK-NEXT: store i16 0, ptr [[GEP_I16]], align 2 107; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 3 108; CHECK-NEXT: [[TMP64]] = trunc i64 [[IV_NEXT]] to i32 109; CHECK-NEXT: [[C:%.*]] = icmp slt i64 [[IV]], 99 110; CHECK-NEXT: br i1 [[C]], label %[[LOOP]], label %[[EXIT:.*]], !llvm.loop [[LOOP9:![0-9]+]] 111; CHECK: [[EXIT]]: 112; CHECK-NEXT: ret void 113; 114entry: 115 %x.i32 = sext i16 %x to i32 116 %x.i64 = sext i16 %x to i64 117 %invariant.gep = getelementptr i8, ptr %A, i64 -8 118 br label %loop 119 120loop: 121 %iv = phi i64 [ %x.i64, %entry ], [ %iv.next, %loop ] 122 %iv.conv = phi i32 [ %x.i32, %entry ], [ %5, %loop ] 123 %gep.i64 = getelementptr i64, ptr %A, i64 %iv 124 %2 = load i64, ptr %gep.i64, align 8 125 %3 = sext i32 %iv.conv to i64 126 %gep.conv = getelementptr i64, ptr %invariant.gep, i64 %3 127 %4 = load i64, ptr %gep.conv, align 8 128 %gep.i16 = getelementptr i16, ptr %A, i64 %iv 129 store i16 0, ptr %gep.i16, align 2 130 %iv.next = add i64 %iv, 3 131 %5 = trunc i64 %iv.next to i32 132 %c = icmp slt i64 %iv, 99 133 br i1 %c, label %loop, label %exit 134 135exit: 136 ret void 137} 138 139attributes #0 = { "target-features"="+64bit,+v,+zvl256b" } 140;. 141; CHECK: [[META0]] = !{[[META1:![0-9]+]]} 142; CHECK: [[META1]] = distinct !{[[META1]], [[META2:![0-9]+]]} 143; CHECK: [[META2]] = distinct !{[[META2]], !"LVerDomain"} 144; CHECK: [[META3]] = !{[[META4:![0-9]+]], [[META5:![0-9]+]]} 145; CHECK: [[META4]] = distinct !{[[META4]], [[META2]]} 146; CHECK: [[META5]] = distinct !{[[META5]], [[META2]]} 147; CHECK: [[LOOP6]] = distinct !{[[LOOP6]], [[META7:![0-9]+]], [[META8:![0-9]+]]} 148; CHECK: [[META7]] = !{!"llvm.loop.isvectorized", i32 1} 149; CHECK: [[META8]] = !{!"llvm.loop.unroll.runtime.disable"} 150; CHECK: [[LOOP9]] = distinct !{[[LOOP9]], [[META7]]} 151;. 152