149 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --version 42; RUN: opt -p loop-vectorize -force-vector-interleave=4 -S %s | FileCheck %s3 4target triple = "aarch64-unknown-linux-gnu"5 6declare void @init_mem(ptr, i64);7 8define i64 @same_exit_block_pre_inc_use1() #0 {9; CHECK-LABEL: define i64 @same_exit_block_pre_inc_use1(10; CHECK-SAME: ) #[[ATTR0:[0-9]+]] {11; CHECK-NEXT: entry:12; CHECK-NEXT: [[P1:%.*]] = alloca [1024 x i8], align 413; CHECK-NEXT: [[P2:%.*]] = alloca [1024 x i8], align 414; CHECK-NEXT: call void @init_mem(ptr [[P1]], i64 1024)15; CHECK-NEXT: call void @init_mem(ptr [[P2]], i64 1024)16; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()17; CHECK-NEXT: [[TMP1:%.*]] = shl nuw nsw i64 [[TMP0]], 618; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 510, [[TMP1]]19; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]20; CHECK: vector.ph:21; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()22; CHECK-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 6423; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 510, [[TMP3]]24; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 510, [[N_MOD_VF]]25; CHECK-NEXT: [[INDEX_NEXT:%.*]] = add i64 3, [[N_VEC]]26; CHECK-NEXT: br label [[LOOP:%.*]]27; CHECK: vector.body:28; CHECK-NEXT: [[INDEX1:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT3:%.*]], [[LOOP]] ]29; CHECK-NEXT: [[OFFSET_IDX:%.*]] = add i64 3, [[INDEX1]]30; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i8, ptr [[P1]], i64 [[OFFSET_IDX]]31; CHECK-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()32; CHECK-NEXT: [[TMP9:%.*]] = shl nuw i64 [[TMP8]], 433; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i8, ptr [[TMP7]], i64 [[TMP9]]34; CHECK-NEXT: [[TMP11:%.*]] = call i64 @llvm.vscale.i64()35; CHECK-NEXT: [[TMP27:%.*]] = shl nuw i64 [[TMP11]], 536; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds i8, ptr [[TMP7]], i64 [[TMP27]]37; CHECK-NEXT: [[TMP14:%.*]] = call i64 @llvm.vscale.i64()38; CHECK-NEXT: [[TMP15:%.*]] = mul nuw i64 [[TMP14]], 4839; CHECK-NEXT: [[TMP28:%.*]] = getelementptr inbounds i8, ptr [[TMP7]], i64 [[TMP15]]40; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 16 x i8>, ptr [[TMP7]], align 141; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 16 x i8>, ptr [[TMP10]], align 142; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 16 x i8>, ptr [[TMP13]], align 143; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 16 x i8>, ptr [[TMP28]], align 144; CHECK-NEXT: [[TMP29:%.*]] = getelementptr inbounds i8, ptr [[P2]], i64 [[OFFSET_IDX]]45; CHECK-NEXT: [[TMP18:%.*]] = call i64 @llvm.vscale.i64()46; CHECK-NEXT: [[TMP19:%.*]] = shl nuw i64 [[TMP18]], 447; CHECK-NEXT: [[TMP20:%.*]] = getelementptr inbounds i8, ptr [[TMP29]], i64 [[TMP19]]48; CHECK-NEXT: [[TMP21:%.*]] = call i64 @llvm.vscale.i64()49; CHECK-NEXT: [[TMP22:%.*]] = shl nuw i64 [[TMP21]], 550; CHECK-NEXT: [[TMP23:%.*]] = getelementptr inbounds i8, ptr [[TMP29]], i64 [[TMP22]]51; CHECK-NEXT: [[TMP24:%.*]] = call i64 @llvm.vscale.i64()52; CHECK-NEXT: [[TMP25:%.*]] = mul nuw i64 [[TMP24]], 4853; CHECK-NEXT: [[TMP26:%.*]] = getelementptr inbounds i8, ptr [[TMP29]], i64 [[TMP25]]54; CHECK-NEXT: [[WIDE_LOAD5:%.*]] = load <vscale x 16 x i8>, ptr [[TMP29]], align 155; CHECK-NEXT: [[WIDE_LOAD6:%.*]] = load <vscale x 16 x i8>, ptr [[TMP20]], align 156; CHECK-NEXT: [[WIDE_LOAD7:%.*]] = load <vscale x 16 x i8>, ptr [[TMP23]], align 157; CHECK-NEXT: [[WIDE_LOAD8:%.*]] = load <vscale x 16 x i8>, ptr [[TMP26]], align 158; CHECK-NEXT: [[TMP32:%.*]] = icmp ne <vscale x 16 x i8> [[WIDE_LOAD]], [[WIDE_LOAD5]]59; CHECK-NEXT: [[TMP30:%.*]] = icmp ne <vscale x 16 x i8> [[WIDE_LOAD2]], [[WIDE_LOAD6]]60; CHECK-NEXT: [[TMP31:%.*]] = icmp ne <vscale x 16 x i8> [[WIDE_LOAD3]], [[WIDE_LOAD7]]61; CHECK-NEXT: [[TMP59:%.*]] = icmp ne <vscale x 16 x i8> [[WIDE_LOAD4]], [[WIDE_LOAD8]]62; CHECK-NEXT: [[INDEX_NEXT3]] = add nuw i64 [[INDEX1]], [[TMP3]]63; CHECK-NEXT: [[TMP37:%.*]] = freeze <vscale x 16 x i1> [[TMP32]]64; CHECK-NEXT: [[TMP38:%.*]] = freeze <vscale x 16 x i1> [[TMP30]]65; CHECK-NEXT: [[TMP54:%.*]] = or <vscale x 16 x i1> [[TMP37]], [[TMP38]]66; CHECK-NEXT: [[TMP60:%.*]] = freeze <vscale x 16 x i1> [[TMP31]]67; CHECK-NEXT: [[TMP62:%.*]] = or <vscale x 16 x i1> [[TMP54]], [[TMP60]]68; CHECK-NEXT: [[TMP34:%.*]] = freeze <vscale x 16 x i1> [[TMP59]]69; CHECK-NEXT: [[TMP33:%.*]] = or <vscale x 16 x i1> [[TMP62]], [[TMP34]]70; CHECK-NEXT: [[TMP12:%.*]] = call i1 @llvm.vector.reduce.or.nxv16i1(<vscale x 16 x i1> [[TMP33]])71; CHECK-NEXT: [[TMP35:%.*]] = icmp eq i64 [[INDEX_NEXT3]], [[N_VEC]]72; CHECK-NEXT: [[TMP36:%.*]] = or i1 [[TMP12]], [[TMP35]]73; CHECK-NEXT: br i1 [[TMP36]], label [[MIDDLE_SPLIT:%.*]], label [[LOOP]], !llvm.loop [[LOOP0:![0-9]+]]74; CHECK: middle.split:75; CHECK-NEXT: br i1 [[TMP12]], label [[VECTOR_EARLY_EXIT:%.*]], label [[LOOP_INC:%.*]]76; CHECK: middle.block:77; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 510, [[N_VEC]]78; CHECK-NEXT: br i1 [[CMP_N]], label [[LOOP_END:%.*]], label [[SCALAR_PH]]79; CHECK: vector.early.exit:80; CHECK-NEXT: [[TMP39:%.*]] = call i64 @llvm.vscale.i64()81; CHECK-NEXT: [[TMP40:%.*]] = mul nuw i64 [[TMP39]], 1682; CHECK-NEXT: [[TMP41:%.*]] = call i64 @llvm.experimental.cttz.elts.i64.nxv16i1(<vscale x 16 x i1> [[TMP59]], i1 false)83; CHECK-NEXT: [[TMP42:%.*]] = mul i64 [[TMP40]], 384; CHECK-NEXT: [[TMP43:%.*]] = add i64 [[TMP42]], [[TMP41]]85; CHECK-NEXT: [[TMP44:%.*]] = call i64 @llvm.experimental.cttz.elts.i64.nxv16i1(<vscale x 16 x i1> [[TMP31]], i1 false)86; CHECK-NEXT: [[TMP45:%.*]] = mul i64 [[TMP40]], 287; CHECK-NEXT: [[TMP46:%.*]] = add i64 [[TMP45]], [[TMP44]]88; CHECK-NEXT: [[TMP47:%.*]] = icmp ne i64 [[TMP44]], [[TMP40]]89; CHECK-NEXT: [[TMP48:%.*]] = select i1 [[TMP47]], i64 [[TMP46]], i64 [[TMP43]]90; CHECK-NEXT: [[TMP49:%.*]] = call i64 @llvm.experimental.cttz.elts.i64.nxv16i1(<vscale x 16 x i1> [[TMP30]], i1 false)91; CHECK-NEXT: [[TMP50:%.*]] = mul i64 [[TMP40]], 192; CHECK-NEXT: [[TMP51:%.*]] = add i64 [[TMP50]], [[TMP49]]93; CHECK-NEXT: [[TMP52:%.*]] = icmp ne i64 [[TMP49]], [[TMP40]]94; CHECK-NEXT: [[TMP53:%.*]] = select i1 [[TMP52]], i64 [[TMP51]], i64 [[TMP48]]95; CHECK-NEXT: [[TMP61:%.*]] = call i64 @llvm.experimental.cttz.elts.i64.nxv16i1(<vscale x 16 x i1> [[TMP32]], i1 false)96; CHECK-NEXT: [[TMP55:%.*]] = mul i64 [[TMP40]], 097; CHECK-NEXT: [[TMP56:%.*]] = add i64 [[TMP55]], [[TMP61]]98; CHECK-NEXT: [[TMP57:%.*]] = icmp ne i64 [[TMP61]], [[TMP40]]99; CHECK-NEXT: [[TMP58:%.*]] = select i1 [[TMP57]], i64 [[TMP56]], i64 [[TMP53]]100; CHECK-NEXT: [[TMP16:%.*]] = add i64 [[INDEX1]], [[TMP58]]101; CHECK-NEXT: [[TMP17:%.*]] = add i64 3, [[TMP16]]102; CHECK-NEXT: br label [[LOOP_END]]103; CHECK: scalar.ph:104; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ [[INDEX_NEXT]], [[LOOP_INC]] ], [ 3, [[ENTRY:%.*]] ]105; CHECK-NEXT: br label [[LOOP1:%.*]]106; CHECK: loop:107; CHECK-NEXT: [[INDEX2:%.*]] = phi i64 [ [[INDEX_NEXT1:%.*]], [[LOOP_INC1:%.*]] ], [ [[INDEX]], [[SCALAR_PH]] ]108; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i8, ptr [[P1]], i64 [[INDEX2]]109; CHECK-NEXT: [[LD1:%.*]] = load i8, ptr [[ARRAYIDX]], align 1110; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds i8, ptr [[P2]], i64 [[INDEX2]]111; CHECK-NEXT: [[LD2:%.*]] = load i8, ptr [[ARRAYIDX1]], align 1112; CHECK-NEXT: [[CMP3:%.*]] = icmp eq i8 [[LD1]], [[LD2]]113; CHECK-NEXT: br i1 [[CMP3]], label [[LOOP_INC1]], label [[LOOP_END]]114; CHECK: loop.inc:115; CHECK-NEXT: [[INDEX_NEXT1]] = add i64 [[INDEX2]], 1116; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDEX_NEXT1]], 513117; CHECK-NEXT: br i1 [[EXITCOND]], label [[LOOP1]], label [[LOOP_END]], !llvm.loop [[LOOP3:![0-9]+]]118; CHECK: loop.end:119; CHECK-NEXT: [[RETVAL:%.*]] = phi i64 [ [[INDEX2]], [[LOOP1]] ], [ 67, [[LOOP_INC1]] ], [ 67, [[LOOP_INC]] ], [ [[TMP17]], [[VECTOR_EARLY_EXIT]] ]120; CHECK-NEXT: ret i64 [[RETVAL]]121;122entry:123 %p1 = alloca [1024 x i8]124 %p2 = alloca [1024 x i8]125 call void @init_mem(ptr %p1, i64 1024)126 call void @init_mem(ptr %p2, i64 1024)127 br label %loop128 129loop:130 %index = phi i64 [ %index.next, %loop.inc ], [ 3, %entry ]131 %arrayidx = getelementptr inbounds i8, ptr %p1, i64 %index132 %ld1 = load i8, ptr %arrayidx, align 1133 %arrayidx1 = getelementptr inbounds i8, ptr %p2, i64 %index134 %ld2 = load i8, ptr %arrayidx1, align 1135 %cmp3 = icmp eq i8 %ld1, %ld2136 br i1 %cmp3, label %loop.inc, label %loop.end137 138loop.inc:139 %index.next = add i64 %index, 1140 %exitcond = icmp ne i64 %index.next, 513141 br i1 %exitcond, label %loop, label %loop.end142 143loop.end:144 %retval = phi i64 [ %index, %loop ], [ 67, %loop.inc ]145 ret i64 %retval146}147 148attributes #0 = { "target-features"="+sve" vscale_range(1,16) }149