231 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=cortex-a57 -S %s | FileCheck --check-prefix=INTERLEAVE-4 %s3; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=cortex-a75 -S %s | FileCheck --check-prefix=INTERLEAVE-2 %s4; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=apple-m1 -S %s | FileCheck --check-prefix=INTERLEAVE-4 %s5; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=apple-a14 -S %s | FileCheck --check-prefix=INTERLEAVE-4 %s6; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=apple-a15 -S %s | FileCheck --check-prefix=INTERLEAVE-4 %s7; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=apple-a16 -S %s | FileCheck --check-prefix=INTERLEAVE-4 %s8; RUN: opt -passes=loop-vectorize -mtriple=arm64 -mcpu=neoverse-v2 -S %s | FileCheck --check-prefix=INTERLEAVE-4 %s9; RUN: opt -passes=loop-vectorize -mtriple=arm64 -mcpu=neoverse-v3 -S %s | FileCheck --check-prefix=INTERLEAVE-4-VLA %s10 11; Tests for selecting interleave counts for loops with loads and stores.12 13define void @interleave_single_load_store(ptr %src, ptr %dst, i64 %N, i8 %a, i8 %b) {14; INTERLEAVE-4-LABEL: @interleave_single_load_store(15; INTERLEAVE-4-NEXT: iter.check:16; INTERLEAVE-4-NEXT: [[SRC2:%.*]] = ptrtoint ptr [[SRC:%.*]] to i6417; INTERLEAVE-4-NEXT: [[DST1:%.*]] = ptrtoint ptr [[DST:%.*]] to i6418; INTERLEAVE-4-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 819; INTERLEAVE-4-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]20; INTERLEAVE-4: vector.memcheck:21; INTERLEAVE-4-NEXT: [[TMP0:%.*]] = sub i64 [[DST1]], [[SRC2]]22; INTERLEAVE-4-NEXT: [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP0]], 6423; INTERLEAVE-4-NEXT: br i1 [[DIFF_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]24; INTERLEAVE-4: vector.main.loop.iter.check:25; INTERLEAVE-4-NEXT: [[MIN_ITERS_CHECK3:%.*]] = icmp ult i64 [[N]], 6426; INTERLEAVE-4-NEXT: br i1 [[MIN_ITERS_CHECK3]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]27; INTERLEAVE-4: vector.ph:28; INTERLEAVE-4-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 6429; INTERLEAVE-4-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]30; INTERLEAVE-4-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i8> poison, i8 [[B:%.*]], i64 031; INTERLEAVE-4-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <16 x i8> [[BROADCAST_SPLATINSERT]], <16 x i8> poison, <16 x i32> zeroinitializer32; INTERLEAVE-4-NEXT: [[BROADCAST_SPLATINSERT7:%.*]] = insertelement <16 x i8> poison, i8 [[A:%.*]], i64 033; INTERLEAVE-4-NEXT: [[BROADCAST_SPLAT8:%.*]] = shufflevector <16 x i8> [[BROADCAST_SPLATINSERT7]], <16 x i8> poison, <16 x i32> zeroinitializer34; INTERLEAVE-4-NEXT: br label [[VECTOR_BODY:%.*]]35; INTERLEAVE-4: vector.body:36; INTERLEAVE-4-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]37; INTERLEAVE-4-NEXT: [[TMP5:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 [[INDEX]]38; INTERLEAVE-4-NEXT: [[TMP10:%.*]] = getelementptr inbounds i8, ptr [[TMP5]], i64 1639; INTERLEAVE-4-NEXT: [[TMP11:%.*]] = getelementptr inbounds i8, ptr [[TMP5]], i64 3240; INTERLEAVE-4-NEXT: [[TMP12:%.*]] = getelementptr inbounds i8, ptr [[TMP5]], i64 4841; INTERLEAVE-4-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP5]], align 142; INTERLEAVE-4-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP10]], align 143; INTERLEAVE-4-NEXT: [[WIDE_LOAD5:%.*]] = load <16 x i8>, ptr [[TMP11]], align 144; INTERLEAVE-4-NEXT: [[WIDE_LOAD6:%.*]] = load <16 x i8>, ptr [[TMP12]], align 145; INTERLEAVE-4-NEXT: [[TMP13:%.*]] = icmp sgt <16 x i8> [[WIDE_LOAD]], [[BROADCAST_SPLAT]]46; INTERLEAVE-4-NEXT: [[TMP14:%.*]] = icmp sgt <16 x i8> [[WIDE_LOAD4]], [[BROADCAST_SPLAT]]47; INTERLEAVE-4-NEXT: [[TMP15:%.*]] = icmp sgt <16 x i8> [[WIDE_LOAD5]], [[BROADCAST_SPLAT]]48; INTERLEAVE-4-NEXT: [[TMP16:%.*]] = icmp sgt <16 x i8> [[WIDE_LOAD6]], [[BROADCAST_SPLAT]]49; INTERLEAVE-4-NEXT: [[TMP17:%.*]] = call <16 x i8> @llvm.smax.v16i8(<16 x i8> [[WIDE_LOAD]], <16 x i8> [[BROADCAST_SPLAT8]])50; INTERLEAVE-4-NEXT: [[TMP18:%.*]] = call <16 x i8> @llvm.smax.v16i8(<16 x i8> [[WIDE_LOAD4]], <16 x i8> [[BROADCAST_SPLAT8]])51; INTERLEAVE-4-NEXT: [[TMP19:%.*]] = call <16 x i8> @llvm.smax.v16i8(<16 x i8> [[WIDE_LOAD5]], <16 x i8> [[BROADCAST_SPLAT8]])52; INTERLEAVE-4-NEXT: [[TMP20:%.*]] = call <16 x i8> @llvm.smax.v16i8(<16 x i8> [[WIDE_LOAD6]], <16 x i8> [[BROADCAST_SPLAT8]])53; INTERLEAVE-4-NEXT: [[TMP21:%.*]] = select <16 x i1> [[TMP13]], <16 x i8> [[BROADCAST_SPLAT]], <16 x i8> [[TMP17]]54; INTERLEAVE-4-NEXT: [[TMP22:%.*]] = select <16 x i1> [[TMP14]], <16 x i8> [[BROADCAST_SPLAT]], <16 x i8> [[TMP18]]55; INTERLEAVE-4-NEXT: [[TMP23:%.*]] = select <16 x i1> [[TMP15]], <16 x i8> [[BROADCAST_SPLAT]], <16 x i8> [[TMP19]]56; INTERLEAVE-4-NEXT: [[TMP24:%.*]] = select <16 x i1> [[TMP16]], <16 x i8> [[BROADCAST_SPLAT]], <16 x i8> [[TMP20]]57; INTERLEAVE-4-NEXT: [[TMP25:%.*]] = getelementptr inbounds i8, ptr [[DST]], i64 [[INDEX]]58; INTERLEAVE-4-NEXT: [[TMP30:%.*]] = getelementptr inbounds i8, ptr [[TMP25]], i64 1659; INTERLEAVE-4-NEXT: [[TMP31:%.*]] = getelementptr inbounds i8, ptr [[TMP25]], i64 3260; INTERLEAVE-4-NEXT: [[TMP32:%.*]] = getelementptr inbounds i8, ptr [[TMP25]], i64 4861; INTERLEAVE-4-NEXT: store <16 x i8> [[TMP21]], ptr [[TMP25]], align 162; INTERLEAVE-4-NEXT: store <16 x i8> [[TMP22]], ptr [[TMP30]], align 163; INTERLEAVE-4-NEXT: store <16 x i8> [[TMP23]], ptr [[TMP31]], align 164; INTERLEAVE-4-NEXT: store <16 x i8> [[TMP24]], ptr [[TMP32]], align 165; INTERLEAVE-4-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 6466; INTERLEAVE-4-NEXT: [[TMP33:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]67; INTERLEAVE-4-NEXT: br i1 [[TMP33]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]68; INTERLEAVE-4: middle.block:69; INTERLEAVE-4-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]70; INTERLEAVE-4-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]71; INTERLEAVE-4: vec.epilog.iter.check:72; INTERLEAVE-4-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], 873; INTERLEAVE-4-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF3:![0-9]+]]74; INTERLEAVE-4: vec.epilog.ph:75; INTERLEAVE-4-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]76; INTERLEAVE-4-NEXT: [[N_MOD_VF9:%.*]] = urem i64 [[N]], 877; INTERLEAVE-4-NEXT: [[N_VEC10:%.*]] = sub i64 [[N]], [[N_MOD_VF9]]78; INTERLEAVE-4-NEXT: [[BROADCAST_SPLATINSERT14:%.*]] = insertelement <8 x i8> poison, i8 [[B]], i64 079; INTERLEAVE-4-NEXT: [[BROADCAST_SPLAT15:%.*]] = shufflevector <8 x i8> [[BROADCAST_SPLATINSERT14]], <8 x i8> poison, <8 x i32> zeroinitializer80; INTERLEAVE-4-NEXT: [[BROADCAST_SPLATINSERT16:%.*]] = insertelement <8 x i8> poison, i8 [[A]], i64 081; INTERLEAVE-4-NEXT: [[BROADCAST_SPLAT17:%.*]] = shufflevector <8 x i8> [[BROADCAST_SPLATINSERT16]], <8 x i8> poison, <8 x i32> zeroinitializer82; INTERLEAVE-4-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]83; INTERLEAVE-4: vec.epilog.vector.body:84; INTERLEAVE-4-NEXT: [[INDEX12:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT18:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]85; INTERLEAVE-4-NEXT: [[TMP35:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 [[INDEX12]]86; INTERLEAVE-4-NEXT: [[WIDE_LOAD13:%.*]] = load <8 x i8>, ptr [[TMP35]], align 187; INTERLEAVE-4-NEXT: [[TMP37:%.*]] = icmp sgt <8 x i8> [[WIDE_LOAD13]], [[BROADCAST_SPLAT15]]88; INTERLEAVE-4-NEXT: [[TMP38:%.*]] = call <8 x i8> @llvm.smax.v8i8(<8 x i8> [[WIDE_LOAD13]], <8 x i8> [[BROADCAST_SPLAT17]])89; INTERLEAVE-4-NEXT: [[TMP39:%.*]] = select <8 x i1> [[TMP37]], <8 x i8> [[BROADCAST_SPLAT15]], <8 x i8> [[TMP38]]90; INTERLEAVE-4-NEXT: [[TMP40:%.*]] = getelementptr inbounds i8, ptr [[DST]], i64 [[INDEX12]]91; INTERLEAVE-4-NEXT: store <8 x i8> [[TMP39]], ptr [[TMP40]], align 192; INTERLEAVE-4-NEXT: [[INDEX_NEXT18]] = add nuw i64 [[INDEX12]], 893; INTERLEAVE-4-NEXT: [[TMP42:%.*]] = icmp eq i64 [[INDEX_NEXT18]], [[N_VEC10]]94; INTERLEAVE-4-NEXT: br i1 [[TMP42]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]95; INTERLEAVE-4: vec.epilog.middle.block:96; INTERLEAVE-4-NEXT: [[CMP_N11:%.*]] = icmp eq i64 [[N]], [[N_VEC10]]97; INTERLEAVE-4-NEXT: br i1 [[CMP_N11]], label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]98; INTERLEAVE-4: vec.epilog.scalar.ph:99; INTERLEAVE-4-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC10]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]100; INTERLEAVE-4-NEXT: br label [[LOOP:%.*]]101; INTERLEAVE-4: loop:102; INTERLEAVE-4-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]103; INTERLEAVE-4-NEXT: [[GEP_SRC:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 [[IV]]104; INTERLEAVE-4-NEXT: [[L:%.*]] = load i8, ptr [[GEP_SRC]], align 1105; INTERLEAVE-4-NEXT: [[CMP:%.*]] = icmp sgt i8 [[L]], [[B]]106; INTERLEAVE-4-NEXT: [[MAX:%.*]] = tail call i8 @llvm.smax.i8(i8 [[L]], i8 [[A]])107; INTERLEAVE-4-NEXT: [[SEL:%.*]] = select i1 [[CMP]], i8 [[B]], i8 [[MAX]]108; INTERLEAVE-4-NEXT: [[GEP_DST:%.*]] = getelementptr inbounds i8, ptr [[DST]], i64 [[IV]]109; INTERLEAVE-4-NEXT: store i8 [[SEL]], ptr [[GEP_DST]], align 1110; INTERLEAVE-4-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1111; INTERLEAVE-4-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]112; INTERLEAVE-4-NEXT: br i1 [[EC]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP5:![0-9]+]]113; INTERLEAVE-4: exit:114; INTERLEAVE-4-NEXT: ret void115;116; INTERLEAVE-2-LABEL: @interleave_single_load_store(117; INTERLEAVE-2-NEXT: iter.check:118; INTERLEAVE-2-NEXT: [[SRC2:%.*]] = ptrtoint ptr [[SRC:%.*]] to i64119; INTERLEAVE-2-NEXT: [[DST1:%.*]] = ptrtoint ptr [[DST:%.*]] to i64120; INTERLEAVE-2-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 8121; INTERLEAVE-2-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]122; INTERLEAVE-2: vector.memcheck:123; INTERLEAVE-2-NEXT: [[TMP0:%.*]] = sub i64 [[DST1]], [[SRC2]]124; INTERLEAVE-2-NEXT: [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP0]], 32125; INTERLEAVE-2-NEXT: br i1 [[DIFF_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]126; INTERLEAVE-2: vector.main.loop.iter.check:127; INTERLEAVE-2-NEXT: [[MIN_ITERS_CHECK3:%.*]] = icmp ult i64 [[N]], 32128; INTERLEAVE-2-NEXT: br i1 [[MIN_ITERS_CHECK3]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]129; INTERLEAVE-2: vector.ph:130; INTERLEAVE-2-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 32131; INTERLEAVE-2-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]132; INTERLEAVE-2-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i8> poison, i8 [[B:%.*]], i64 0133; INTERLEAVE-2-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <16 x i8> [[BROADCAST_SPLATINSERT]], <16 x i8> poison, <16 x i32> zeroinitializer134; INTERLEAVE-2-NEXT: [[BROADCAST_SPLATINSERT5:%.*]] = insertelement <16 x i8> poison, i8 [[A:%.*]], i64 0135; INTERLEAVE-2-NEXT: [[BROADCAST_SPLAT6:%.*]] = shufflevector <16 x i8> [[BROADCAST_SPLATINSERT5]], <16 x i8> poison, <16 x i32> zeroinitializer136; INTERLEAVE-2-NEXT: br label [[VECTOR_BODY:%.*]]137; INTERLEAVE-2: vector.body:138; INTERLEAVE-2-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]139; INTERLEAVE-2-NEXT: [[TMP3:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 [[INDEX]]140; INTERLEAVE-2-NEXT: [[TMP6:%.*]] = getelementptr inbounds i8, ptr [[TMP3]], i64 16141; INTERLEAVE-2-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP3]], align 1142; INTERLEAVE-2-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP6]], align 1143; INTERLEAVE-2-NEXT: [[TMP7:%.*]] = icmp sgt <16 x i8> [[WIDE_LOAD]], [[BROADCAST_SPLAT]]144; INTERLEAVE-2-NEXT: [[TMP8:%.*]] = icmp sgt <16 x i8> [[WIDE_LOAD4]], [[BROADCAST_SPLAT]]145; INTERLEAVE-2-NEXT: [[TMP9:%.*]] = call <16 x i8> @llvm.smax.v16i8(<16 x i8> [[WIDE_LOAD]], <16 x i8> [[BROADCAST_SPLAT6]])146; INTERLEAVE-2-NEXT: [[TMP10:%.*]] = call <16 x i8> @llvm.smax.v16i8(<16 x i8> [[WIDE_LOAD4]], <16 x i8> [[BROADCAST_SPLAT6]])147; INTERLEAVE-2-NEXT: [[TMP11:%.*]] = select <16 x i1> [[TMP7]], <16 x i8> [[BROADCAST_SPLAT]], <16 x i8> [[TMP9]]148; INTERLEAVE-2-NEXT: [[TMP12:%.*]] = select <16 x i1> [[TMP8]], <16 x i8> [[BROADCAST_SPLAT]], <16 x i8> [[TMP10]]149; INTERLEAVE-2-NEXT: [[TMP13:%.*]] = getelementptr inbounds i8, ptr [[DST]], i64 [[INDEX]]150; INTERLEAVE-2-NEXT: [[TMP16:%.*]] = getelementptr inbounds i8, ptr [[TMP13]], i64 16151; INTERLEAVE-2-NEXT: store <16 x i8> [[TMP11]], ptr [[TMP13]], align 1152; INTERLEAVE-2-NEXT: store <16 x i8> [[TMP12]], ptr [[TMP16]], align 1153; INTERLEAVE-2-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32154; INTERLEAVE-2-NEXT: [[TMP17:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]155; INTERLEAVE-2-NEXT: br i1 [[TMP17]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]156; INTERLEAVE-2: middle.block:157; INTERLEAVE-2-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]158; INTERLEAVE-2-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]159; INTERLEAVE-2: vec.epilog.iter.check:160; INTERLEAVE-2-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], 8161; INTERLEAVE-2-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF3:![0-9]+]]162; INTERLEAVE-2: vec.epilog.ph:163; INTERLEAVE-2-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]164; INTERLEAVE-2-NEXT: [[N_MOD_VF7:%.*]] = urem i64 [[N]], 8165; INTERLEAVE-2-NEXT: [[N_VEC8:%.*]] = sub i64 [[N]], [[N_MOD_VF7]]166; INTERLEAVE-2-NEXT: [[BROADCAST_SPLATINSERT12:%.*]] = insertelement <8 x i8> poison, i8 [[B]], i64 0167; INTERLEAVE-2-NEXT: [[BROADCAST_SPLAT13:%.*]] = shufflevector <8 x i8> [[BROADCAST_SPLATINSERT12]], <8 x i8> poison, <8 x i32> zeroinitializer168; INTERLEAVE-2-NEXT: [[BROADCAST_SPLATINSERT14:%.*]] = insertelement <8 x i8> poison, i8 [[A]], i64 0169; INTERLEAVE-2-NEXT: [[BROADCAST_SPLAT15:%.*]] = shufflevector <8 x i8> [[BROADCAST_SPLATINSERT14]], <8 x i8> poison, <8 x i32> zeroinitializer170; INTERLEAVE-2-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]171; INTERLEAVE-2: vec.epilog.vector.body:172; INTERLEAVE-2-NEXT: [[INDEX10:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT16:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]173; INTERLEAVE-2-NEXT: [[TMP19:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 [[INDEX10]]174; INTERLEAVE-2-NEXT: [[WIDE_LOAD11:%.*]] = load <8 x i8>, ptr [[TMP19]], align 1175; INTERLEAVE-2-NEXT: [[TMP21:%.*]] = icmp sgt <8 x i8> [[WIDE_LOAD11]], [[BROADCAST_SPLAT13]]176; INTERLEAVE-2-NEXT: [[TMP22:%.*]] = call <8 x i8> @llvm.smax.v8i8(<8 x i8> [[WIDE_LOAD11]], <8 x i8> [[BROADCAST_SPLAT15]])177; INTERLEAVE-2-NEXT: [[TMP23:%.*]] = select <8 x i1> [[TMP21]], <8 x i8> [[BROADCAST_SPLAT13]], <8 x i8> [[TMP22]]178; INTERLEAVE-2-NEXT: [[TMP24:%.*]] = getelementptr inbounds i8, ptr [[DST]], i64 [[INDEX10]]179; INTERLEAVE-2-NEXT: store <8 x i8> [[TMP23]], ptr [[TMP24]], align 1180; INTERLEAVE-2-NEXT: [[INDEX_NEXT16]] = add nuw i64 [[INDEX10]], 8181; INTERLEAVE-2-NEXT: [[TMP26:%.*]] = icmp eq i64 [[INDEX_NEXT16]], [[N_VEC8]]182; INTERLEAVE-2-NEXT: br i1 [[TMP26]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]183; INTERLEAVE-2: vec.epilog.middle.block:184; INTERLEAVE-2-NEXT: [[CMP_N9:%.*]] = icmp eq i64 [[N]], [[N_VEC8]]185; INTERLEAVE-2-NEXT: br i1 [[CMP_N9]], label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]186; INTERLEAVE-2: vec.epilog.scalar.ph:187; INTERLEAVE-2-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC8]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]188; INTERLEAVE-2-NEXT: br label [[LOOP:%.*]]189; INTERLEAVE-2: loop:190; INTERLEAVE-2-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]191; INTERLEAVE-2-NEXT: [[GEP_SRC:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 [[IV]]192; INTERLEAVE-2-NEXT: [[L:%.*]] = load i8, ptr [[GEP_SRC]], align 1193; INTERLEAVE-2-NEXT: [[CMP:%.*]] = icmp sgt i8 [[L]], [[B]]194; INTERLEAVE-2-NEXT: [[MAX:%.*]] = tail call i8 @llvm.smax.i8(i8 [[L]], i8 [[A]])195; INTERLEAVE-2-NEXT: [[SEL:%.*]] = select i1 [[CMP]], i8 [[B]], i8 [[MAX]]196; INTERLEAVE-2-NEXT: [[GEP_DST:%.*]] = getelementptr inbounds i8, ptr [[DST]], i64 [[IV]]197; INTERLEAVE-2-NEXT: store i8 [[SEL]], ptr [[GEP_DST]], align 1198; INTERLEAVE-2-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1199; INTERLEAVE-2-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]200; INTERLEAVE-2-NEXT: br i1 [[EC]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP4:![0-9]+]]201; INTERLEAVE-2: exit:202; INTERLEAVE-2-NEXT: ret void203;204; INTERLEAVE-4-VLA-LABEL: @interleave_single_load_store(205; INTERLEAVE-4-VLA: call <vscale x 16 x i8> @llvm.smax.nxv16i8(206; INTERLEAVE-4-VLA-NEXT: call <vscale x 16 x i8> @llvm.smax.nxv16i8(207; INTERLEAVE-4-VLA-NEXT: call <vscale x 16 x i8> @llvm.smax.nxv16i8(208; INTERLEAVE-4-VLA-NEXT: call <vscale x 16 x i8> @llvm.smax.nxv16i8(209;210entry:211 br label %loop212 213loop:214 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop]215 %gep.src = getelementptr inbounds i8, ptr %src, i64 %iv216 %l = load i8, ptr %gep.src, align 1217 %cmp = icmp sgt i8 %l, %b218 %max = tail call i8 @llvm.smax.i8(i8 %l, i8 %a)219 %sel = select i1 %cmp, i8 %b, i8 %max220 %gep.dst = getelementptr inbounds i8, ptr %dst, i64 %iv221 store i8 %sel, ptr %gep.dst, align 1222 %iv.next = add nuw nsw i64 %iv, 1223 %ec = icmp eq i64 %iv.next, %N224 br i1 %ec, label %exit, label %loop225 226exit:227 ret void228}229 230declare i8 @llvm.smax.i8(i8, i8)231