1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -passes=loop-vectorize,instcombine -S -mattr=avx512f < %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(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 i32 11; CHECK-NEXT: [[SMAX2:%.*]] = call i64 @llvm.smax.i64(i64 [[N]], i64 1) 12; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp slt i64 [[N]], 8 13; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]] 14; CHECK: vector.memcheck: 15; CHECK-NEXT: [[TMP0:%.*]] = shl i64 [[N]], 2 16; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[B:%.*]], i64 [[TMP0]] 17; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[A:%.*]], i64 4 18; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[B]], [[SCEVGEP1]] 19; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[A]], [[SCEVGEP]] 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_CHECK3:%.*]] = icmp slt i64 [[N]], 16 24; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK3]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]] 25; CHECK: vector.ph: 26; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[SMAX2]], 9223372036854775792 27; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x ptr> poison, ptr [[A]], i64 0 28; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <16 x ptr> [[BROADCAST_SPLATINSERT]], <16 x ptr> poison, <16 x i32> zeroinitializer 29; CHECK-NEXT: [[TMP1:%.*]] = icmp ne <16 x ptr> [[BROADCAST_SPLAT]], zeroinitializer 30; CHECK-NEXT: [[BROADCAST_SPLATINSERT4:%.*]] = insertelement <16 x i32> poison, i32 [[NTRUNC]], i64 0 31; CHECK-NEXT: [[BROADCAST_SPLAT5:%.*]] = shufflevector <16 x i32> [[BROADCAST_SPLATINSERT4]], <16 x i32> poison, <16 x i32> zeroinitializer 32; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 33; CHECK: vector.body: 34; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 35; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[INDEX]] 36; CHECK-NEXT: store <16 x i32> [[BROADCAST_SPLAT5]], ptr [[TMP2]], align 4, !alias.scope [[META0:![0-9]+]], !noalias [[META3:![0-9]+]] 37; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16 38; CHECK-NEXT: [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 39; CHECK-NEXT: br i1 [[TMP3]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] 40; CHECK: middle.block: 41; CHECK-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0(<16 x ptr> [[BROADCAST_SPLAT]], i32 4, <16 x i1> [[TMP1]], <16 x i32> poison), !alias.scope [[META3]] 42; CHECK-NEXT: [[PREDPHI:%.*]] = select <16 x i1> [[TMP1]], <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> 43; CHECK-NEXT: [[TMP4:%.*]] = extractelement <16 x i32> [[PREDPHI]], i64 15 44; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[SMAX2]], [[N_VEC]] 45; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]] 46; CHECK: vec.epilog.iter.check: 47; CHECK-NEXT: [[N_VEC_REMAINING:%.*]] = and i64 [[SMAX2]], 8 48; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK_NOT_NOT:%.*]] = icmp eq i64 [[N_VEC_REMAINING]], 0 49; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK_NOT_NOT]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]] 50; CHECK: vec.epilog.ph: 51; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ] 52; CHECK-NEXT: [[N_VEC7:%.*]] = and i64 [[SMAX2]], 9223372036854775800 53; CHECK-NEXT: [[BROADCAST_SPLATINSERT8:%.*]] = insertelement <8 x ptr> poison, ptr [[A]], i64 0 54; CHECK-NEXT: [[BROADCAST_SPLAT9:%.*]] = shufflevector <8 x ptr> [[BROADCAST_SPLATINSERT8]], <8 x ptr> poison, <8 x i32> zeroinitializer 55; CHECK-NEXT: [[TMP5:%.*]] = icmp ne <8 x ptr> [[BROADCAST_SPLAT9]], zeroinitializer 56; CHECK-NEXT: [[BROADCAST_SPLATINSERT11:%.*]] = insertelement <8 x i32> poison, i32 [[NTRUNC]], i64 0 57; CHECK-NEXT: [[BROADCAST_SPLAT12:%.*]] = shufflevector <8 x i32> [[BROADCAST_SPLATINSERT11]], <8 x i32> poison, <8 x i32> zeroinitializer 58; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]] 59; CHECK: vec.epilog.vector.body: 60; CHECK-NEXT: [[INDEX10:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT15:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ] 61; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[INDEX10]] 62; CHECK-NEXT: store <8 x i32> [[BROADCAST_SPLAT12]], ptr [[TMP6]], align 4, !alias.scope [[META8:![0-9]+]], !noalias [[META11:![0-9]+]] 63; CHECK-NEXT: [[INDEX_NEXT15]] = add nuw i64 [[INDEX10]], 8 64; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT15]], [[N_VEC7]] 65; CHECK-NEXT: br i1 [[TMP7]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]] 66; CHECK: vec.epilog.middle.block: 67; CHECK-NEXT: [[WIDE_MASKED_GATHER13:%.*]] = call <8 x i32> @llvm.masked.gather.v8i32.v8p0(<8 x ptr> [[BROADCAST_SPLAT9]], i32 4, <8 x i1> [[TMP5]], <8 x i32> poison), !alias.scope [[META11]] 68; CHECK-NEXT: [[PREDPHI14:%.*]] = select <8 x i1> [[TMP5]], <8 x i32> [[WIDE_MASKED_GATHER13]], <8 x i32> <i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 1> 69; CHECK-NEXT: [[TMP8:%.*]] = extractelement <8 x i32> [[PREDPHI14]], i64 7 70; CHECK-NEXT: [[CMP_N16:%.*]] = icmp eq i64 [[SMAX2]], [[N_VEC7]] 71; CHECK-NEXT: br i1 [[CMP_N16]], label [[FOR_END]], label [[VEC_EPILOG_SCALAR_PH]] 72; CHECK: vec.epilog.scalar.ph: 73; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC7]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[ITER_CHECK:%.*]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ] 74; CHECK-NEXT: br label [[FOR_BODY:%.*]] 75; CHECK: for.body: 76; CHECK-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[LATCH:%.*]] ], [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ] 77; CHECK-NEXT: [[I1:%.*]] = getelementptr inbounds nuw i32, ptr [[B]], i64 [[I]] 78; CHECK-NEXT: [[CMP_NOT:%.*]] = icmp eq ptr [[A]], null 79; CHECK-NEXT: store i32 [[NTRUNC]], ptr [[I1]], align 4 80; CHECK-NEXT: br i1 [[CMP_NOT]], label [[LATCH]], label [[COND_LOAD:%.*]] 81; CHECK: cond_load: 82; CHECK-NEXT: [[ALOAD:%.*]] = load i32, ptr [[A]], align 4 83; CHECK-NEXT: br label [[LATCH]] 84; CHECK: latch: 85; CHECK-NEXT: [[A_LCSSA:%.*]] = phi i32 [ [[ALOAD]], [[COND_LOAD]] ], [ 1, [[FOR_BODY]] ] 86; CHECK-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 87; CHECK-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]] 88; CHECK-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP14:![0-9]+]] 89; CHECK: for.end: 90; CHECK-NEXT: [[A_LCSSA_LCSSA:%.*]] = phi i32 [ [[A_LCSSA]], [[LATCH]] ], [ [[TMP4]], [[MIDDLE_BLOCK]] ], [ [[TMP8]], [[VEC_EPILOG_MIDDLE_BLOCK]] ] 91; CHECK-NEXT: ret i32 [[A_LCSSA_LCSSA]] 92; 93entry: 94 %ntrunc = trunc i64 %n to i32 95 br label %for.body 96 97for.body: ; preds = %for.body, %entry 98 %i = phi i64 [ %i.next, %latch ], [ 0, %entry ] 99 %i1 = getelementptr inbounds i32, ptr %b, i64 %i 100 %i2 = load i32, ptr %i1, align 8 101 %cmp = icmp ne ptr %a, null 102 store i32 %ntrunc, ptr %i1 103 br i1 %cmp, label %cond_load, label %latch 104 105cond_load: 106 %aload = load i32, ptr %a, align 4 107 br label %latch 108 109latch: 110 %a.lcssa = phi i32 [ %aload, %cond_load ], [ 1, %for.body ] 111 %i.next = add nuw nsw i64 %i, 1 112 %cond = icmp slt i64 %i.next, %n 113 br i1 %cond, label %for.body, label %for.end 114 115for.end: ; preds = %for.body 116 ret i32 %a.lcssa 117} 118