123 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -passes=loop-vectorize -mattr=avx512f -S %s | FileCheck %s3 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(ptr %a, i64 %n, ptr %b, i32 %k) {8; CHECK-LABEL: @inv_load_conditional(9; CHECK-NEXT: iter.check:10; CHECK-NEXT: [[NTRUNC:%.*]] = trunc i64 [[N:%.*]] to i3211; CHECK-NEXT: [[SMAX2:%.*]] = call i64 @llvm.smax.i64(i64 [[N]], i64 1)12; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX2]], 813; 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: [[TMP0:%.*]] = shl i64 [[SMAX]], 217; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[B:%.*]], i64 [[TMP0]]18; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[A:%.*]], i64 419; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[B]], [[SCEVGEP1]]20; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[A]], [[SCEVGEP]]21; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]22; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label [[VEC_EPILOG_SCALAR_PH]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]23; CHECK: vector.main.loop.iter.check:24; CHECK-NEXT: [[MIN_ITERS_CHECK3:%.*]] = icmp ult i64 [[SMAX2]], 1625; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK3]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]26; CHECK: vector.ph:27; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[SMAX2]], 1628; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[SMAX2]], [[N_MOD_VF]]29; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x ptr> poison, ptr [[A]], i64 030; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <16 x ptr> [[BROADCAST_SPLATINSERT]], <16 x ptr> poison, <16 x i32> zeroinitializer31; CHECK-NEXT: [[TMP1:%.*]] = icmp ne <16 x ptr> [[BROADCAST_SPLAT]], zeroinitializer32; CHECK-NEXT: [[BROADCAST_SPLATINSERT4:%.*]] = insertelement <16 x i32> poison, i32 [[NTRUNC]], i64 033; CHECK-NEXT: [[BROADCAST_SPLAT5:%.*]] = shufflevector <16 x i32> [[BROADCAST_SPLATINSERT4]], <16 x i32> poison, <16 x i32> zeroinitializer34; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]35; CHECK: vector.body:36; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]37; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[INDEX]]38; CHECK-NEXT: store <16 x i32> [[BROADCAST_SPLAT5]], ptr [[TMP2]], align 4, !alias.scope [[META0:![0-9]+]], !noalias [[META3:![0-9]+]]39; CHECK-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0(<16 x ptr> align 4 [[BROADCAST_SPLAT]], <16 x i1> [[TMP1]], <16 x i32> poison), !alias.scope [[META3]]40; CHECK-NEXT: [[TMP9:%.*]] = extractelement <16 x i1> [[TMP1]], i32 041; CHECK-NEXT: [[PREDPHI:%.*]] = select i1 [[TMP9]], <16 x i32> [[WIDE_MASKED_GATHER]], <16 x i32> splat (i32 1)42; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 1643; CHECK-NEXT: [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]44; CHECK-NEXT: br i1 [[TMP3]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]45; CHECK: middle.block:46; CHECK-NEXT: [[TMP4:%.*]] = extractelement <16 x i32> [[PREDPHI]], i32 1547; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[SMAX2]], [[N_VEC]]48; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]49; CHECK: vec.epilog.iter.check:50; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], 851; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF8:![0-9]+]]52; CHECK: vec.epilog.ph:53; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]54; CHECK-NEXT: [[N_MOD_VF6:%.*]] = urem i64 [[SMAX2]], 855; CHECK-NEXT: [[N_VEC7:%.*]] = sub i64 [[SMAX2]], [[N_MOD_VF6]]56; CHECK-NEXT: [[BROADCAST_SPLATINSERT8:%.*]] = insertelement <8 x ptr> poison, ptr [[A]], i64 057; CHECK-NEXT: [[BROADCAST_SPLAT9:%.*]] = shufflevector <8 x ptr> [[BROADCAST_SPLATINSERT8]], <8 x ptr> poison, <8 x i32> zeroinitializer58; CHECK-NEXT: [[TMP5:%.*]] = icmp ne <8 x ptr> [[BROADCAST_SPLAT9]], zeroinitializer59; CHECK-NEXT: [[BROADCAST_SPLATINSERT11:%.*]] = insertelement <8 x i32> poison, i32 [[NTRUNC]], i64 060; CHECK-NEXT: [[BROADCAST_SPLAT12:%.*]] = shufflevector <8 x i32> [[BROADCAST_SPLATINSERT11]], <8 x i32> poison, <8 x i32> zeroinitializer61; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]62; CHECK: vec.epilog.vector.body:63; CHECK-NEXT: [[INDEX10:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT15:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]64; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[INDEX10]]65; CHECK-NEXT: store <8 x i32> [[BROADCAST_SPLAT12]], ptr [[TMP6]], align 4, !alias.scope [[META0]], !noalias [[META3]]66; CHECK-NEXT: [[WIDE_MASKED_GATHER13:%.*]] = call <8 x i32> @llvm.masked.gather.v8i32.v8p0(<8 x ptr> align 4 [[BROADCAST_SPLAT9]], <8 x i1> [[TMP5]], <8 x i32> poison), !alias.scope [[META3]]67; CHECK-NEXT: [[TMP10:%.*]] = extractelement <8 x i1> [[TMP5]], i32 068; CHECK-NEXT: [[PREDPHI14:%.*]] = select i1 [[TMP10]], <8 x i32> [[WIDE_MASKED_GATHER13]], <8 x i32> splat (i32 1)69; CHECK-NEXT: [[INDEX_NEXT15]] = add nuw i64 [[INDEX10]], 870; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT15]], [[N_VEC7]]71; CHECK-NEXT: br i1 [[TMP7]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]72; CHECK: vec.epilog.middle.block:73; CHECK-NEXT: [[TMP8:%.*]] = extractelement <8 x i32> [[PREDPHI14]], i32 774; CHECK-NEXT: [[CMP_N16:%.*]] = icmp eq i64 [[SMAX2]], [[N_VEC7]]75; CHECK-NEXT: br i1 [[CMP_N16]], label [[FOR_END]], label [[VEC_EPILOG_SCALAR_PH]]76; CHECK: vec.epilog.scalar.ph:77; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC7]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]78; CHECK-NEXT: br label [[FOR_BODY:%.*]]79; CHECK: for.body:80; CHECK-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[LATCH:%.*]] ], [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ]81; CHECK-NEXT: [[I1:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[I]]82; CHECK-NEXT: [[I2:%.*]] = load i32, ptr [[I1]], align 883; CHECK-NEXT: [[CMP:%.*]] = icmp ne ptr [[A]], null84; CHECK-NEXT: store i32 [[NTRUNC]], ptr [[I1]], align 485; CHECK-NEXT: br i1 [[CMP]], label [[COND_LOAD:%.*]], label [[LATCH]]86; CHECK: cond_load:87; CHECK-NEXT: [[ALOAD:%.*]] = load i32, ptr [[A]], align 488; CHECK-NEXT: br label [[LATCH]]89; CHECK: latch:90; CHECK-NEXT: [[A_LCSSA:%.*]] = phi i32 [ [[ALOAD]], [[COND_LOAD]] ], [ 1, [[FOR_BODY]] ]91; CHECK-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 192; CHECK-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]93; CHECK-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP10:![0-9]+]]94; CHECK: for.end:95; CHECK-NEXT: [[A_LCSSA_LCSSA:%.*]] = phi i32 [ [[A_LCSSA]], [[LATCH]] ], [ [[TMP4]], [[MIDDLE_BLOCK]] ], [ [[TMP8]], [[VEC_EPILOG_MIDDLE_BLOCK]] ]96; CHECK-NEXT: ret i32 [[A_LCSSA_LCSSA]]97;98entry:99 %ntrunc = trunc i64 %n to i32100 br label %for.body101 102for.body: ; preds = %for.body, %entry103 %i = phi i64 [ %i.next, %latch ], [ 0, %entry ]104 %i1 = getelementptr inbounds i32, ptr %b, i64 %i105 %i2 = load i32, ptr %i1, align 8106 %cmp = icmp ne ptr %a, null107 store i32 %ntrunc, ptr %i1108 br i1 %cmp, label %cond_load, label %latch109 110cond_load:111 %aload = load i32, ptr %a, align 4112 br label %latch113 114latch:115 %a.lcssa = phi i32 [ %aload, %cond_load ], [ 1, %for.body ]116 %i.next = add nuw nsw i64 %i, 1117 %cond = icmp slt i64 %i.next, %n118 br i1 %cond, label %for.body, label %for.end119 120for.end: ; preds = %for.body121 ret i32 %a.lcssa122}123