1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4 2; RUN: opt -mtriple=x86_64 -mattr=-avx,-avx2,-avx512f,+sse,-sse2,-sse3,-sse4.2 -passes=loop-vectorize -S %s | FileCheck --check-prefix=NOVEC %s 3; RUN: opt -mtriple=x86_64 -mattr=-avx,-avx2,-avx512f,+sse,-sse2,-sse3,-sse4.2 -passes=loop-vectorize -force-vector-width=4 -S %s | FileCheck --check-prefix=VEC %s 4 5@h = global i64 0 6 7define void @test(ptr %p) { 8; NOVEC-LABEL: define void @test( 9; NOVEC-SAME: ptr [[P:%.*]]) #[[ATTR0:[0-9]+]] { 10; NOVEC-NEXT: entry: 11; NOVEC-NEXT: br label [[FOR_BODY:%.*]] 12; NOVEC: for.body: 13; NOVEC-NEXT: [[IDX_EXT_MERGE:%.*]] = phi i64 [ 1, [[ENTRY:%.*]] ], [ [[IDX:%.*]], [[FOR_BODY]] ] 14; NOVEC-NEXT: [[INC_MERGE:%.*]] = phi i16 [ 1, [[ENTRY]] ], [ [[INC:%.*]], [[FOR_BODY]] ] 15; NOVEC-NEXT: [[IDX_MERGE:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ [[IDX_EXT_MERGE]], [[FOR_BODY]] ] 16; NOVEC-NEXT: [[ADD:%.*]] = shl i64 [[IDX_MERGE]], 1 17; NOVEC-NEXT: [[ARRAYIDX:%.*]] = getelementptr i64, ptr [[P]], i64 [[ADD]] 18; NOVEC-NEXT: store i64 0, ptr [[ARRAYIDX]], align 8 19; NOVEC-NEXT: [[INC]] = add i16 [[INC_MERGE]], 1 20; NOVEC-NEXT: [[IDX]] = zext i16 [[INC]] to i64 21; NOVEC-NEXT: [[GEP:%.*]] = getelementptr i64, ptr [[P]], i64 [[IDX]] 22; NOVEC-NEXT: [[CMP:%.*]] = icmp ugt ptr [[GEP]], @h 23; NOVEC-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[FOR_BODY]] 24; NOVEC: exit: 25; NOVEC-NEXT: ret void 26; 27; VEC-LABEL: define void @test( 28; VEC-SAME: ptr [[P:%.*]]) #[[ATTR0:[0-9]+]] { 29; VEC-NEXT: entry: 30; VEC-NEXT: [[P1:%.*]] = ptrtoint ptr [[P]] to i64 31; VEC-NEXT: [[TMP0:%.*]] = add i64 [[P1]], 16 32; VEC-NEXT: [[UMAX2:%.*]] = call i64 @llvm.umax.i64(i64 [[TMP0]], i64 add (i64 ptrtoint (ptr @h to i64), i64 1)) 33; VEC-NEXT: [[TMP1:%.*]] = add i64 [[UMAX2]], -9 34; VEC-NEXT: [[TMP2:%.*]] = sub i64 [[TMP1]], [[P1]] 35; VEC-NEXT: [[TMP3:%.*]] = lshr i64 [[TMP2]], 3 36; VEC-NEXT: [[TMP4:%.*]] = add nuw nsw i64 [[TMP3]], 1 37; VEC-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP4]], 4 38; VEC-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]] 39; VEC: vector.scevcheck: 40; VEC-NEXT: [[TMP5:%.*]] = add i64 [[P1]], 16 41; VEC-NEXT: [[UMAX:%.*]] = call i64 @llvm.umax.i64(i64 [[TMP5]], i64 add (i64 ptrtoint (ptr @h to i64), i64 1)) 42; VEC-NEXT: [[TMP6:%.*]] = add i64 [[UMAX]], -9 43; VEC-NEXT: [[TMP7:%.*]] = sub i64 [[TMP6]], [[P1]] 44; VEC-NEXT: [[TMP8:%.*]] = lshr i64 [[TMP7]], 3 45; VEC-NEXT: [[TMP10:%.*]] = trunc i64 [[TMP8]] to i16 46; VEC-NEXT: [[TMP11:%.*]] = add i16 2, [[TMP10]] 47; VEC-NEXT: [[TMP12:%.*]] = icmp ult i16 [[TMP11]], 2 48; VEC-NEXT: [[TMP13:%.*]] = icmp ugt i64 [[TMP8]], 65535 49; VEC-NEXT: [[TMP14:%.*]] = or i1 [[TMP12]], [[TMP13]] 50; VEC-NEXT: br i1 [[TMP14]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]] 51; VEC: vector.ph: 52; VEC-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP4]], 4 53; VEC-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP4]], [[N_MOD_VF]] 54; VEC-NEXT: [[DOTCAST:%.*]] = trunc i64 [[N_VEC]] to i16 55; VEC-NEXT: [[IND_END:%.*]] = add i16 1, [[DOTCAST]] 56; VEC-NEXT: br label [[VECTOR_BODY:%.*]] 57; VEC: vector.body: 58; VEC-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 59; VEC-NEXT: [[VECTOR_RECUR:%.*]] = phi <4 x i64> [ <i64 poison, i64 poison, i64 poison, i64 1>, [[VECTOR_PH]] ], [ [[TMP28:%.*]], [[VECTOR_BODY]] ] 60; VEC-NEXT: [[VEC_IND:%.*]] = phi <4 x i16> [ <i16 1, i16 2, i16 3, i16 4>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] 61; VEC-NEXT: [[TMP15:%.*]] = add i64 [[INDEX]], 0 62; VEC-NEXT: [[TMP16:%.*]] = add i64 [[INDEX]], 1 63; VEC-NEXT: [[TMP17:%.*]] = add i64 [[INDEX]], 2 64; VEC-NEXT: [[TMP18:%.*]] = add i64 [[INDEX]], 3 65; VEC-NEXT: [[TMP19:%.*]] = shl i64 [[TMP15]], 1 66; VEC-NEXT: [[TMP20:%.*]] = shl i64 [[TMP16]], 1 67; VEC-NEXT: [[TMP21:%.*]] = shl i64 [[TMP17]], 1 68; VEC-NEXT: [[TMP22:%.*]] = shl i64 [[TMP18]], 1 69; VEC-NEXT: [[TMP23:%.*]] = getelementptr i64, ptr [[P]], i64 [[TMP19]] 70; VEC-NEXT: [[TMP24:%.*]] = getelementptr i64, ptr [[P]], i64 [[TMP20]] 71; VEC-NEXT: [[TMP25:%.*]] = getelementptr i64, ptr [[P]], i64 [[TMP21]] 72; VEC-NEXT: [[TMP26:%.*]] = getelementptr i64, ptr [[P]], i64 [[TMP22]] 73; VEC-NEXT: store i64 0, ptr [[TMP23]], align 8 74; VEC-NEXT: store i64 0, ptr [[TMP24]], align 8 75; VEC-NEXT: store i64 0, ptr [[TMP25]], align 8 76; VEC-NEXT: store i64 0, ptr [[TMP26]], align 8 77; VEC-NEXT: [[TMP27:%.*]] = add <4 x i16> [[VEC_IND]], splat (i16 1) 78; VEC-NEXT: [[TMP28]] = zext <4 x i16> [[TMP27]] to <4 x i64> 79; VEC-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 80; VEC-NEXT: [[VEC_IND_NEXT]] = add <4 x i16> [[VEC_IND]], splat (i16 4) 81; VEC-NEXT: [[TMP30:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 82; VEC-NEXT: br i1 [[TMP30]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 83; VEC: middle.block: 84; VEC-NEXT: [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <4 x i64> [[TMP28]], i32 3 85; VEC-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP4]], [[N_VEC]] 86; VEC-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]] 87; VEC: scalar.ph: 88; VEC-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi i64 [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ], [ 1, [[VECTOR_SCEVCHECK]] ], [ 1, [[ENTRY:%.*]] ] 89; VEC-NEXT: [[BC_RESUME_VAL:%.*]] = phi i16 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 1, [[VECTOR_SCEVCHECK]] ], [ 1, [[ENTRY]] ] 90; VEC-NEXT: [[BC_RESUME_VAL3:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[VECTOR_SCEVCHECK]] ], [ 0, [[ENTRY]] ] 91; VEC-NEXT: br label [[FOR_BODY:%.*]] 92; VEC: for.body: 93; VEC-NEXT: [[SCALAR_RECUR:%.*]] = phi i64 [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ], [ [[IDX:%.*]], [[FOR_BODY]] ] 94; VEC-NEXT: [[INC_MERGE:%.*]] = phi i16 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ] 95; VEC-NEXT: [[IDX_MERGE:%.*]] = phi i64 [ [[BC_RESUME_VAL3]], [[SCALAR_PH]] ], [ [[SCALAR_RECUR]], [[FOR_BODY]] ] 96; VEC-NEXT: [[ADD:%.*]] = shl i64 [[IDX_MERGE]], 1 97; VEC-NEXT: [[ARRAYIDX:%.*]] = getelementptr i64, ptr [[P]], i64 [[ADD]] 98; VEC-NEXT: store i64 0, ptr [[ARRAYIDX]], align 8 99; VEC-NEXT: [[INC]] = add i16 [[INC_MERGE]], 1 100; VEC-NEXT: [[IDX]] = zext i16 [[INC]] to i64 101; VEC-NEXT: [[GEP:%.*]] = getelementptr i64, ptr [[P]], i64 [[IDX]] 102; VEC-NEXT: [[CMP:%.*]] = icmp ugt ptr [[GEP]], @h 103; VEC-NEXT: br i1 [[CMP]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] 104; VEC: exit: 105; VEC-NEXT: ret void 106; 107entry: 108 br label %for.body 109 110for.body: 111 %idx.ext.merge = phi i64 [ 1, %entry ], [ %idx, %for.body ] 112 %inc.merge = phi i16 [ 1, %entry ], [ %inc, %for.body ] 113 %idx.merge = phi i64 [ 0, %entry ], [ %idx.ext.merge, %for.body ] 114 %add = shl i64 %idx.merge, 1 115 %arrayidx = getelementptr i64, ptr %p, i64 %add 116 store i64 0, ptr %arrayidx 117 %inc = add i16 %inc.merge, 1 118 %idx = zext i16 %inc to i64 119 %gep = getelementptr i64, ptr %p, i64 %idx 120 %cmp = icmp ugt ptr %gep, @h 121 br i1 %cmp, label %exit, label %for.body 122 123exit: 124 ret void 125} 126;. 127; VEC: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]} 128; VEC: [[META1]] = !{!"llvm.loop.isvectorized", i32 1} 129; VEC: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"} 130; VEC: [[LOOP3]] = distinct !{[[LOOP3]], [[META1]]} 131;. 132