1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -loop-vectorize -S -mattr=avx512f -instcombine < %s | FileCheck %s 3 4target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128" 5target triple = "x86_64-unknown-linux-gnu" 6 7define i32 @inv_load_conditional(i32* %a, i64 %n, i32* %b, i32 %k) { 8; CHECK-LABEL: @inv_load_conditional( 9; CHECK-NEXT: iter.check: 10; CHECK-NEXT: [[NTRUNC:%.*]] = trunc i64 [[N:%.*]] to i32 11; CHECK-NEXT: [[SMAX6:%.*]] = call i64 @llvm.smax.i64(i64 [[N]], i64 1) 12; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX6]], 8 13; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]] 14; CHECK: vector.memcheck: 15; CHECK-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N]], i64 1) 16; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i32, i32* [[B:%.*]], i64 [[SMAX]] 17; CHECK-NEXT: [[SCEVGEP4:%.*]] = getelementptr i32, i32* [[A:%.*]], i64 1 18; CHECK-NEXT: [[BOUND0:%.*]] = icmp ugt i32* [[SCEVGEP4]], [[B]] 19; CHECK-NEXT: [[BOUND1:%.*]] = icmp ugt i32* [[SCEVGEP]], [[A]] 20; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] 21; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label [[VEC_EPILOG_SCALAR_PH]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]] 22; CHECK: vector.main.loop.iter.check: 23; CHECK-NEXT: [[MIN_ITERS_CHECK7:%.*]] = icmp ult i64 [[SMAX6]], 16 24; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK7]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]] 25; CHECK: vector.ph: 26; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[SMAX6]], 9223372036854775792 27; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i32*> poison, i32* [[A]], i64 0 28; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <16 x i32*> [[BROADCAST_SPLATINSERT]], <16 x i32*> poison, <16 x i32> zeroinitializer 29; CHECK-NEXT: [[BROADCAST_SPLATINSERT8:%.*]] = insertelement <16 x i32> poison, i32 [[NTRUNC]], i64 0 30; CHECK-NEXT: [[BROADCAST_SPLAT9:%.*]] = shufflevector <16 x i32> [[BROADCAST_SPLATINSERT8]], <16 x i32> poison, <16 x i32> zeroinitializer 31; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 32; CHECK: vector.body: 33; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 34; CHECK-NEXT: [[TMP0:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[INDEX]] 35; CHECK-NEXT: [[TMP1:%.*]] = bitcast i32* [[TMP0]] to <16 x i32>* 36; CHECK-NEXT: store <16 x i32> [[BROADCAST_SPLAT9]], <16 x i32>* [[TMP1]], align 4, !alias.scope !0, !noalias !3 37; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16 38; CHECK-NEXT: [[TMP2:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 39; CHECK-NEXT: br i1 [[TMP2]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] 40; CHECK: middle.block: 41; CHECK-NEXT: [[TMP3:%.*]] = icmp ne <16 x i32*> [[BROADCAST_SPLAT]], zeroinitializer 42; CHECK-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[BROADCAST_SPLAT]], i32 4, <16 x i1> [[TMP3]], <16 x i32> undef), !alias.scope !3 43; CHECK-NEXT: [[PREDPHI:%.*]] = select <16 x i1> [[TMP3]], <16 x i32> [[WIDE_MASKED_GATHER]], <16 x i32> <i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 1> 44; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[SMAX6]], [[N_VEC]] 45; CHECK-NEXT: [[TMP4:%.*]] = extractelement <16 x i32> [[PREDPHI]], i64 15 46; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]] 47; CHECK: vec.epilog.iter.check: 48; CHECK-NEXT: [[N_VEC_REMAINING:%.*]] = and i64 [[SMAX6]], 8 49; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK_NOT_NOT:%.*]] = icmp eq i64 [[N_VEC_REMAINING]], 0 50; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK_NOT_NOT]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]] 51; CHECK: vec.epilog.ph: 52; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ] 53; CHECK-NEXT: [[SMAX11:%.*]] = call i64 @llvm.smax.i64(i64 [[N]], i64 1) 54; CHECK-NEXT: [[N_VEC13:%.*]] = and i64 [[SMAX11]], 9223372036854775800 55; CHECK-NEXT: [[BROADCAST_SPLATINSERT17:%.*]] = insertelement <8 x i32*> poison, i32* [[A]], i64 0 56; CHECK-NEXT: [[BROADCAST_SPLAT18:%.*]] = shufflevector <8 x i32*> [[BROADCAST_SPLATINSERT17]], <8 x i32*> poison, <8 x i32> zeroinitializer 57; CHECK-NEXT: [[BROADCAST_SPLATINSERT19:%.*]] = insertelement <8 x i32> poison, i32 [[NTRUNC]], i64 0 58; CHECK-NEXT: [[BROADCAST_SPLAT20:%.*]] = shufflevector <8 x i32> [[BROADCAST_SPLATINSERT19]], <8 x i32> poison, <8 x i32> zeroinitializer 59; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]] 60; CHECK: vec.epilog.vector.body: 61; CHECK-NEXT: [[OFFSET_IDX:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT23:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ] 62; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[OFFSET_IDX]] 63; CHECK-NEXT: [[TMP6:%.*]] = bitcast i32* [[TMP5]] to <8 x i32>* 64; CHECK-NEXT: store <8 x i32> [[BROADCAST_SPLAT20]], <8 x i32>* [[TMP6]], align 4 65; CHECK-NEXT: [[INDEX_NEXT23]] = add nuw i64 [[OFFSET_IDX]], 8 66; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT23]], [[N_VEC13]] 67; CHECK-NEXT: br i1 [[TMP7]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]] 68; CHECK: vec.epilog.middle.block: 69; CHECK-NEXT: [[TMP8:%.*]] = icmp ne <8 x i32*> [[BROADCAST_SPLAT18]], zeroinitializer 70; CHECK-NEXT: [[WIDE_MASKED_GATHER21:%.*]] = call <8 x i32> @llvm.masked.gather.v8i32.v8p0i32(<8 x i32*> [[BROADCAST_SPLAT18]], i32 4, <8 x i1> [[TMP8]], <8 x i32> undef) 71; CHECK-NEXT: [[PREDPHI22:%.*]] = select <8 x i1> [[TMP8]], <8 x i32> [[WIDE_MASKED_GATHER21]], <8 x i32> <i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 1> 72; CHECK-NEXT: [[CMP_N14:%.*]] = icmp eq i64 [[SMAX11]], [[N_VEC13]] 73; CHECK-NEXT: [[TMP9:%.*]] = extractelement <8 x i32> [[PREDPHI22]], i64 7 74; CHECK-NEXT: br i1 [[CMP_N14]], label [[FOR_END_LOOPEXIT:%.*]], label [[VEC_EPILOG_SCALAR_PH]] 75; CHECK: vec.epilog.scalar.ph: 76; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC13]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[ITER_CHECK:%.*]] ] 77; CHECK-NEXT: br label [[FOR_BODY:%.*]] 78; CHECK: for.body: 79; CHECK-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[LATCH:%.*]] ], [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ] 80; CHECK-NEXT: [[I1:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[I]] 81; CHECK-NEXT: [[CMP_NOT:%.*]] = icmp eq i32* [[A]], null 82; CHECK-NEXT: store i32 [[NTRUNC]], i32* [[I1]], align 4 83; CHECK-NEXT: br i1 [[CMP_NOT]], label [[LATCH]], label [[COND_LOAD:%.*]] 84; CHECK: cond_load: 85; CHECK-NEXT: [[ALOAD:%.*]] = load i32, i32* [[A]], align 4 86; CHECK-NEXT: br label [[LATCH]] 87; CHECK: latch: 88; CHECK-NEXT: [[A_LCSSA:%.*]] = phi i32 [ [[ALOAD]], [[COND_LOAD]] ], [ 1, [[FOR_BODY]] ] 89; CHECK-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 90; CHECK-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]] 91; CHECK-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END_LOOPEXIT]], !llvm.loop [[LOOP9:![0-9]+]] 92; CHECK: for.end.loopexit: 93; CHECK-NEXT: [[A_LCSSA_LCSSA10:%.*]] = phi i32 [ [[A_LCSSA]], [[LATCH]] ], [ [[TMP9]], [[VEC_EPILOG_MIDDLE_BLOCK]] ] 94; CHECK-NEXT: br label [[FOR_END]] 95; CHECK: for.end: 96; CHECK-NEXT: [[A_LCSSA_LCSSA:%.*]] = phi i32 [ [[TMP4]], [[MIDDLE_BLOCK]] ], [ [[A_LCSSA_LCSSA10]], [[FOR_END_LOOPEXIT]] ] 97; CHECK-NEXT: ret i32 [[A_LCSSA_LCSSA]] 98; 99entry: 100 %ntrunc = trunc i64 %n to i32 101 br label %for.body 102 103for.body: ; preds = %for.body, %entry 104 %i = phi i64 [ %i.next, %latch ], [ 0, %entry ] 105 %i1 = getelementptr inbounds i32, i32* %b, i64 %i 106 %i2 = load i32, i32* %i1, align 8 107 %cmp = icmp ne i32* %a, null 108 store i32 %ntrunc, i32* %i1 109 br i1 %cmp, label %cond_load, label %latch 110 111cond_load: 112 %aload = load i32, i32* %a, align 4 113 br label %latch 114 115latch: 116 %a.lcssa = phi i32 [ %aload, %cond_load ], [ 1, %for.body ] 117 %i.next = add nuw nsw i64 %i, 1 118 %cond = icmp slt i64 %i.next, %n 119 br i1 %cond, label %for.body, label %for.end 120 121for.end: ; preds = %for.body 122 ret i32 %a.lcssa 123} 124