596 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -aa-pipeline=basic-aa -passes=loop-vectorize,dce,instcombine -force-vector-interleave=1 -force-vector-width=4 -S | FileCheck %s3; RUN: opt < %s -aa-pipeline= -passes=loop-vectorize -S -pass-remarks-analysis='loop-vectorize' 2>&1 | FileCheck %s -check-prefix=FORCED_OPTSIZE4 5target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"6 7; Make sure we vectorize this loop:8; int foo(float *a, float *b, int n) {9; for (int i=0; i<n; ++i)10; a[i] = b[i] * 3;11; }12 13define void @foo(ptr nocapture %a, ptr nocapture %b, i32 %n) nounwind uwtable ssp {14; CHECK-LABEL: @foo(15; CHECK-NEXT: entry:16; CHECK-NEXT: [[B2:%.*]] = ptrtoint ptr [[B:%.*]] to i64, !dbg [[DBG4:![0-9]+]]17; CHECK-NEXT: [[A1:%.*]] = ptrtoint ptr [[A:%.*]] to i64, !dbg [[DBG4]]18; CHECK-NEXT: [[CMP6:%.*]] = icmp sgt i32 [[N:%.*]], 0, !dbg [[DBG4]]19; CHECK-NEXT: br i1 [[CMP6]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_END:%.*]], !dbg [[DBG4]]20; CHECK: for.body.preheader:21; CHECK-NEXT: [[TMP0:%.*]] = zext nneg i32 [[N]] to i64, !dbg [[DBG9:![0-9]+]]22; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[N]], 4, !dbg [[DBG9]]23; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]], !dbg [[DBG9]]24; CHECK: vector.memcheck:25; CHECK-NEXT: [[TMP1:%.*]] = sub i64 [[A1]], [[B2]], !dbg [[DBG9]]26; CHECK-NEXT: [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP1]], 16, !dbg [[DBG9]]27; CHECK-NEXT: br i1 [[DIFF_CHECK]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]], !dbg [[DBG9]]28; CHECK: vector.ph:29; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[TMP0]], 214748364430; CHECK-NEXT: br label [[VECTOR_BODY:%.*]], !dbg [[DBG9]]31; CHECK: vector.body:32; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ], !dbg [[DBG9]]33; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[INDEX]], !dbg [[DBG9]]34; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x float>, ptr [[TMP2]], align 4, !dbg [[DBG9]]35; CHECK-NEXT: [[TMP3:%.*]] = fmul <4 x float> [[WIDE_LOAD]], splat (float 3.000000e+00), !dbg [[DBG9]]36; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[INDEX]], !dbg [[DBG9]]37; CHECK-NEXT: store <4 x float> [[TMP3]], ptr [[TMP4]], align 4, !dbg [[DBG9]]38; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4, !dbg [[DBG9]]39; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]], !dbg [[DBG9]]40; CHECK-NEXT: br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !dbg [[DBG9]], !llvm.loop [[LOOP10:![0-9]+]]41; CHECK: middle.block:42; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N_VEC]], [[TMP0]], !dbg [[DBG9]]43; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END_LOOPEXIT:%.*]], label [[SCALAR_PH]], !dbg [[DBG9]]44; CHECK: scalar.ph:45; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_PREHEADER]] ], [ 0, [[VECTOR_MEMCHECK]] ], !dbg [[DBG9]]46; CHECK-NEXT: br label [[FOR_BODY:%.*]], !dbg [[DBG9]]47; CHECK: for.body:48; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], !dbg [[DBG9]]49; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[INDVARS_IV]], !dbg [[DBG9]]50; CHECK-NEXT: [[TMP6:%.*]] = load float, ptr [[ARRAYIDX]], align 4, !dbg [[DBG9]]51; CHECK-NEXT: [[MUL:%.*]] = fmul float [[TMP6]], 3.000000e+00, !dbg [[DBG9]]52; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[INDVARS_IV]], !dbg [[DBG9]]53; CHECK-NEXT: store float [[MUL]], ptr [[ARRAYIDX2]], align 4, !dbg [[DBG9]]54; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1, !dbg [[DBG9]]55; CHECK-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32, !dbg [[DBG9]]56; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[N]], [[LFTR_WIDEIV]], !dbg [[DBG9]]57; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END_LOOPEXIT]], label [[FOR_BODY]], !dbg [[DBG9]], !llvm.loop [[LOOP13:![0-9]+]]58; CHECK: for.end.loopexit:59; CHECK-NEXT: br label [[FOR_END]], !dbg [[DBG14:![0-9]+]]60; CHECK: for.end:61; CHECK-NEXT: ret void, !dbg [[DBG14]]62;63; FORCED_OPTSIZE-LABEL: @foo(64; FORCED_OPTSIZE-NEXT: entry:65; FORCED_OPTSIZE-NEXT: [[CMP6:%.*]] = icmp sgt i32 [[N:%.*]], 0, !dbg [[DBG4:![0-9]+]]66; FORCED_OPTSIZE-NEXT: br i1 [[CMP6]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_END:%.*]], !dbg [[DBG4]]67; FORCED_OPTSIZE: for.body.preheader:68; FORCED_OPTSIZE-NEXT: br label [[FOR_BODY:%.*]], !dbg [[DBG9:![0-9]+]]69; FORCED_OPTSIZE: for.body:70; FORCED_OPTSIZE-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ 0, [[FOR_BODY_PREHEADER]] ], !dbg [[DBG9]]71; FORCED_OPTSIZE-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, ptr [[B:%.*]], i64 [[INDVARS_IV]], !dbg [[DBG9]]72; FORCED_OPTSIZE-NEXT: [[TMP0:%.*]] = load float, ptr [[ARRAYIDX]], align 4, !dbg [[DBG9]]73; FORCED_OPTSIZE-NEXT: [[MUL:%.*]] = fmul float [[TMP0]], 3.000000e+00, !dbg [[DBG9]]74; FORCED_OPTSIZE-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i64 [[INDVARS_IV]], !dbg [[DBG9]]75; FORCED_OPTSIZE-NEXT: store float [[MUL]], ptr [[ARRAYIDX2]], align 4, !dbg [[DBG9]]76; FORCED_OPTSIZE-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1, !dbg [[DBG9]]77; FORCED_OPTSIZE-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32, !dbg [[DBG9]]78; FORCED_OPTSIZE-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], [[N]], !dbg [[DBG9]]79; FORCED_OPTSIZE-NEXT: br i1 [[EXITCOND]], label [[FOR_END_LOOPEXIT:%.*]], label [[FOR_BODY]], !dbg [[DBG9]]80; FORCED_OPTSIZE: for.end.loopexit:81; FORCED_OPTSIZE-NEXT: br label [[FOR_END]], !dbg [[DBG10:![0-9]+]]82; FORCED_OPTSIZE: for.end:83; FORCED_OPTSIZE-NEXT: ret void, !dbg [[DBG10]]84;85entry:86 %cmp6 = icmp sgt i32 %n, 0, !dbg !687 br i1 %cmp6, label %for.body, label %for.end, !dbg !688 89for.body: ; preds = %entry, %for.body90 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ], !dbg !791 %arrayidx = getelementptr inbounds float, ptr %b, i64 %indvars.iv, !dbg !792 %0 = load float, ptr %arrayidx, align 4, !dbg !793 %mul = fmul float %0, 3.000000e+00, !dbg !794 %arrayidx2 = getelementptr inbounds float, ptr %a, i64 %indvars.iv, !dbg !795 store float %mul, ptr %arrayidx2, align 4, !dbg !796 %indvars.iv.next = add i64 %indvars.iv, 1, !dbg !797 %lftr.wideiv = trunc i64 %indvars.iv.next to i32, !dbg !798 %exitcond = icmp eq i32 %lftr.wideiv, %n, !dbg !799 br i1 %exitcond, label %for.end, label %for.body, !dbg !7100 101for.end: ; preds = %for.body, %entry102 ret void, !dbg !8103}104 105; Make sure that we try to vectorize loops with a runtime check if the106; dependency check fails.107 108define void @test_runtime_check(ptr %a, float %b, i64 %offset, i64 %offset2, i64 %n) {109; CHECK-LABEL: @test_runtime_check(110; CHECK-NEXT: entry:111; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 4112; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]113; CHECK: vector.memcheck:114; CHECK-NEXT: [[TMP0:%.*]] = shl i64 [[OFFSET:%.*]], 2115; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[A:%.*]], i64 [[TMP0]]116; CHECK-NEXT: [[TMP1:%.*]] = shl i64 [[N]], 2117; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP1]]118; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[TMP2]], i64 [[TMP0]]119; CHECK-NEXT: [[TMP3:%.*]] = shl i64 [[OFFSET2:%.*]], 2120; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP3]]121; CHECK-NEXT: [[TMP4:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP1]]122; CHECK-NEXT: [[SCEVGEP3:%.*]] = getelementptr i8, ptr [[TMP4]], i64 [[TMP3]]123; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[SCEVGEP]], [[SCEVGEP3]]124; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[SCEVGEP2]], [[SCEVGEP1]]125; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]126; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]127; CHECK: vector.ph:128; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[N]], -4129; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x float> poison, float [[B:%.*]], i64 0130; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x float> [[BROADCAST_SPLATINSERT]], <4 x float> poison, <4 x i32> zeroinitializer131; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]132; CHECK: vector.body:133; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]134; CHECK-NEXT: [[TMP5:%.*]] = getelementptr float, ptr [[A]], i64 [[INDEX]]135; CHECK-NEXT: [[TMP6:%.*]] = getelementptr float, ptr [[TMP5]], i64 [[OFFSET]]136; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x float>, ptr [[TMP6]], align 4, !alias.scope [[META15:![0-9]+]], !noalias [[META18:![0-9]+]]137; CHECK-NEXT: [[TMP7:%.*]] = getelementptr float, ptr [[A]], i64 [[INDEX]]138; CHECK-NEXT: [[TMP8:%.*]] = getelementptr float, ptr [[TMP7]], i64 [[OFFSET2]]139; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <4 x float>, ptr [[TMP8]], align 4, !alias.scope [[META18]]140; CHECK-NEXT: [[TMP9:%.*]] = fmul fast <4 x float> [[BROADCAST_SPLAT]], [[WIDE_LOAD4]]141; CHECK-NEXT: [[TMP10:%.*]] = fadd fast <4 x float> [[WIDE_LOAD]], [[TMP9]]142; CHECK-NEXT: store <4 x float> [[TMP10]], ptr [[TMP6]], align 4, !alias.scope [[META15]], !noalias [[META18]]143; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4144; CHECK-NEXT: [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]145; CHECK-NEXT: br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP20:![0-9]+]]146; CHECK: middle.block:147; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]148; CHECK-NEXT: br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]]149; CHECK: scalar.ph:150; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ], [ 0, [[VECTOR_MEMCHECK]] ]151; CHECK-NEXT: br label [[FOR_BODY:%.*]]152; CHECK: for.body:153; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ]154; CHECK-NEXT: [[TMP12:%.*]] = getelementptr float, ptr [[A]], i64 [[IV]]155; CHECK-NEXT: [[ARR_IDX:%.*]] = getelementptr float, ptr [[TMP12]], i64 [[OFFSET]]156; CHECK-NEXT: [[L1:%.*]] = load float, ptr [[ARR_IDX]], align 4157; CHECK-NEXT: [[TMP13:%.*]] = getelementptr float, ptr [[A]], i64 [[IV]]158; CHECK-NEXT: [[ARR_IDX2:%.*]] = getelementptr float, ptr [[TMP13]], i64 [[OFFSET2]]159; CHECK-NEXT: [[L2:%.*]] = load float, ptr [[ARR_IDX2]], align 4160; CHECK-NEXT: [[M:%.*]] = fmul fast float [[B]], [[L2]]161; CHECK-NEXT: [[AD:%.*]] = fadd fast float [[L1]], [[M]]162; CHECK-NEXT: store float [[AD]], ptr [[ARR_IDX]], align 4163; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1164; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]165; CHECK-NEXT: br i1 [[EXITCOND]], label [[LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP21:![0-9]+]]166; CHECK: loopexit:167; CHECK-NEXT: ret void168;169; FORCED_OPTSIZE-LABEL: @test_runtime_check(170; FORCED_OPTSIZE-NEXT: entry:171; FORCED_OPTSIZE-NEXT: br label [[FOR_BODY:%.*]]172; FORCED_OPTSIZE: for.body:173; FORCED_OPTSIZE-NEXT: [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ]174; FORCED_OPTSIZE-NEXT: [[IND_SUM:%.*]] = add i64 [[IV]], [[OFFSET:%.*]]175; FORCED_OPTSIZE-NEXT: [[ARR_IDX:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i64 [[IND_SUM]]176; FORCED_OPTSIZE-NEXT: [[L1:%.*]] = load float, ptr [[ARR_IDX]], align 4177; FORCED_OPTSIZE-NEXT: [[IND_SUM2:%.*]] = add i64 [[IV]], [[OFFSET2:%.*]]178; FORCED_OPTSIZE-NEXT: [[ARR_IDX2:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IND_SUM2]]179; FORCED_OPTSIZE-NEXT: [[L2:%.*]] = load float, ptr [[ARR_IDX2]], align 4180; FORCED_OPTSIZE-NEXT: [[M:%.*]] = fmul fast float [[B:%.*]], [[L2]]181; FORCED_OPTSIZE-NEXT: [[AD:%.*]] = fadd fast float [[L1]], [[M]]182; FORCED_OPTSIZE-NEXT: store float [[AD]], ptr [[ARR_IDX]], align 4183; FORCED_OPTSIZE-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1184; FORCED_OPTSIZE-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[IV_NEXT]], [[N:%.*]]185; FORCED_OPTSIZE-NEXT: br i1 [[EXITCOND]], label [[LOOPEXIT:%.*]], label [[FOR_BODY]]186; FORCED_OPTSIZE: loopexit:187; FORCED_OPTSIZE-NEXT: ret void188;189entry:190 br label %for.body191 192for.body:193 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]194 %ind.sum = add i64 %iv, %offset195 %arr.idx = getelementptr inbounds float, ptr %a, i64 %ind.sum196 %l1 = load float, ptr %arr.idx, align 4197 %ind.sum2 = add i64 %iv, %offset2198 %arr.idx2 = getelementptr inbounds float, ptr %a, i64 %ind.sum2199 %l2 = load float, ptr %arr.idx2, align 4200 %m = fmul fast float %b, %l2201 %ad = fadd fast float %l1, %m202 store float %ad, ptr %arr.idx, align 4203 %iv.next = add nuw nsw i64 %iv, 1204 %exitcond = icmp eq i64 %iv.next, %n205 br i1 %exitcond, label %loopexit, label %for.body206 207loopexit:208 ret void209}210 211; Check we do not generate runtime checks if we found a known dependence preventing212; vectorization. In this case, it is a read of c[i-1] followed by a write of c[i].213; The runtime checks would always fail.214 215; void test_runtime_check2(float *a, float b, unsigned offset, unsigned offset2, unsigned n, float *c) {216; for (unsigned i = 1; i < n; i++) {217; a[i+o1] += a[i+o2] + b;218; c[i] = c[i-1] + b;219; }220; }221 222define void @test_runtime_check2(ptr %a, float %b, i64 %offset, i64 %offset2, i64 %n, ptr %c) {223; CHECK-LABEL: @test_runtime_check2(224; CHECK-NEXT: entry:225; CHECK-NEXT: br label [[FOR_BODY:%.*]]226; CHECK: for.body:227; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ]228; CHECK-NEXT: [[TMP0:%.*]] = getelementptr float, ptr [[A:%.*]], i64 [[IV]]229; CHECK-NEXT: [[ARR_IDX:%.*]] = getelementptr float, ptr [[TMP0]], i64 [[OFFSET:%.*]]230; CHECK-NEXT: [[L1:%.*]] = load float, ptr [[ARR_IDX]], align 4231; CHECK-NEXT: [[TMP1:%.*]] = getelementptr float, ptr [[A]], i64 [[IV]]232; CHECK-NEXT: [[ARR_IDX2:%.*]] = getelementptr float, ptr [[TMP1]], i64 [[OFFSET2:%.*]]233; CHECK-NEXT: [[L2:%.*]] = load float, ptr [[ARR_IDX2]], align 4234; CHECK-NEXT: [[M:%.*]] = fmul fast float [[B:%.*]], [[L2]]235; CHECK-NEXT: [[AD:%.*]] = fadd fast float [[L1]], [[M]]236; CHECK-NEXT: store float [[AD]], ptr [[ARR_IDX]], align 4237; CHECK-NEXT: [[TMP2:%.*]] = getelementptr float, ptr [[C:%.*]], i64 [[IV]]238; CHECK-NEXT: [[C_IDX:%.*]] = getelementptr i8, ptr [[TMP2]], i64 -4239; CHECK-NEXT: [[LC:%.*]] = load float, ptr [[C_IDX]], align 4240; CHECK-NEXT: [[VC:%.*]] = fadd float [[LC]], 1.000000e+00241; CHECK-NEXT: [[C_IDX2:%.*]] = getelementptr inbounds nuw float, ptr [[C]], i64 [[IV]]242; CHECK-NEXT: store float [[VC]], ptr [[C_IDX2]], align 4243; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1244; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[IV_NEXT]], [[N:%.*]]245; CHECK-NEXT: br i1 [[EXITCOND]], label [[LOOPEXIT:%.*]], label [[FOR_BODY]]246; CHECK: loopexit:247; CHECK-NEXT: ret void248;249; FORCED_OPTSIZE-LABEL: @test_runtime_check2(250; FORCED_OPTSIZE-NEXT: entry:251; FORCED_OPTSIZE-NEXT: br label [[FOR_BODY:%.*]]252; FORCED_OPTSIZE: for.body:253; FORCED_OPTSIZE-NEXT: [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ]254; FORCED_OPTSIZE-NEXT: [[IND_SUM:%.*]] = add i64 [[IV]], [[OFFSET:%.*]]255; FORCED_OPTSIZE-NEXT: [[ARR_IDX:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i64 [[IND_SUM]]256; FORCED_OPTSIZE-NEXT: [[L1:%.*]] = load float, ptr [[ARR_IDX]], align 4257; FORCED_OPTSIZE-NEXT: [[IND_SUM2:%.*]] = add i64 [[IV]], [[OFFSET2:%.*]]258; FORCED_OPTSIZE-NEXT: [[ARR_IDX2:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IND_SUM2]]259; FORCED_OPTSIZE-NEXT: [[L2:%.*]] = load float, ptr [[ARR_IDX2]], align 4260; FORCED_OPTSIZE-NEXT: [[M:%.*]] = fmul fast float [[B:%.*]], [[L2]]261; FORCED_OPTSIZE-NEXT: [[AD:%.*]] = fadd fast float [[L1]], [[M]]262; FORCED_OPTSIZE-NEXT: store float [[AD]], ptr [[ARR_IDX]], align 4263; FORCED_OPTSIZE-NEXT: [[C_IND:%.*]] = add i64 [[IV]], -1264; FORCED_OPTSIZE-NEXT: [[C_IDX:%.*]] = getelementptr inbounds float, ptr [[C:%.*]], i64 [[C_IND]]265; FORCED_OPTSIZE-NEXT: [[LC:%.*]] = load float, ptr [[C_IDX]], align 4266; FORCED_OPTSIZE-NEXT: [[VC:%.*]] = fadd float [[LC]], 1.000000e+00267; FORCED_OPTSIZE-NEXT: [[C_IDX2:%.*]] = getelementptr inbounds float, ptr [[C]], i64 [[IV]]268; FORCED_OPTSIZE-NEXT: store float [[VC]], ptr [[C_IDX2]], align 4269; FORCED_OPTSIZE-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1270; FORCED_OPTSIZE-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[IV_NEXT]], [[N:%.*]]271; FORCED_OPTSIZE-NEXT: br i1 [[EXITCOND]], label [[LOOPEXIT:%.*]], label [[FOR_BODY]]272; FORCED_OPTSIZE: loopexit:273; FORCED_OPTSIZE-NEXT: ret void274;275entry:276 br label %for.body277 278for.body:279 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]280 %ind.sum = add i64 %iv, %offset281 %arr.idx = getelementptr inbounds float, ptr %a, i64 %ind.sum282 %l1 = load float, ptr %arr.idx, align 4283 %ind.sum2 = add i64 %iv, %offset2284 %arr.idx2 = getelementptr inbounds float, ptr %a, i64 %ind.sum2285 %l2 = load float, ptr %arr.idx2, align 4286 %m = fmul fast float %b, %l2287 %ad = fadd fast float %l1, %m288 store float %ad, ptr %arr.idx, align 4289 %c.ind = add i64 %iv, -1290 %c.idx = getelementptr inbounds float, ptr %c, i64 %c.ind291 %lc = load float, ptr %c.idx, align 4292 %vc = fadd float %lc, 1.0293 %c.idx2 = getelementptr inbounds float, ptr %c, i64 %iv294 store float %vc, ptr %c.idx2295 %iv.next = add nuw nsw i64 %iv, 1296 %exitcond = icmp eq i64 %iv.next, %n297 br i1 %exitcond, label %loopexit, label %for.body298 299loopexit:300 ret void301}302 303define void @different_load_store_pairs(ptr %src.1, ptr %src.2, ptr %dst.1, ptr %dst.2, i64 %n) {304; CHECK-LABEL: @different_load_store_pairs(305; CHECK-NEXT: entry:306; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[UMAX19:%.*]], 4307; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]308; CHECK: vector.memcheck:309; CHECK-NEXT: [[TMP0:%.*]] = shl i64 [[UMAX19]], 2310; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[DST_1:%.*]], i64 [[TMP0]]311; CHECK-NEXT: [[TMP1:%.*]] = shl i64 [[UMAX19]], 3312; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[DST_2:%.*]], i64 [[TMP1]]313; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[SRC_1:%.*]], i64 [[TMP0]]314; CHECK-NEXT: [[SCEVGEP3:%.*]] = getelementptr i8, ptr [[SRC_2:%.*]], i64 [[TMP1]]315; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[DST_1]], [[SCEVGEP1]]316; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[DST_2]], [[SCEVGEP]]317; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]318; CHECK-NEXT: [[BOUND04:%.*]] = icmp ult ptr [[DST_1]], [[SCEVGEP2]]319; CHECK-NEXT: [[BOUND15:%.*]] = icmp ult ptr [[SRC_1]], [[SCEVGEP]]320; CHECK-NEXT: [[FOUND_CONFLICT6:%.*]] = and i1 [[BOUND04]], [[BOUND15]]321; CHECK-NEXT: [[CONFLICT_RDX:%.*]] = or i1 [[FOUND_CONFLICT]], [[FOUND_CONFLICT6]]322; CHECK-NEXT: [[BOUND07:%.*]] = icmp ult ptr [[DST_1]], [[SCEVGEP3]]323; CHECK-NEXT: [[BOUND18:%.*]] = icmp ult ptr [[SRC_2]], [[SCEVGEP]]324; CHECK-NEXT: [[FOUND_CONFLICT9:%.*]] = and i1 [[BOUND07]], [[BOUND18]]325; CHECK-NEXT: [[CONFLICT_RDX10:%.*]] = or i1 [[CONFLICT_RDX]], [[FOUND_CONFLICT9]]326; CHECK-NEXT: [[BOUND011:%.*]] = icmp ult ptr [[DST_2]], [[SCEVGEP2]]327; CHECK-NEXT: [[BOUND112:%.*]] = icmp ult ptr [[SRC_1]], [[SCEVGEP1]]328; CHECK-NEXT: [[FOUND_CONFLICT13:%.*]] = and i1 [[BOUND011]], [[BOUND112]]329; CHECK-NEXT: [[CONFLICT_RDX14:%.*]] = or i1 [[CONFLICT_RDX10]], [[FOUND_CONFLICT13]]330; CHECK-NEXT: [[BOUND015:%.*]] = icmp ult ptr [[DST_2]], [[SCEVGEP3]]331; CHECK-NEXT: [[BOUND116:%.*]] = icmp ult ptr [[SRC_2]], [[SCEVGEP1]]332; CHECK-NEXT: [[FOUND_CONFLICT17:%.*]] = and i1 [[BOUND015]], [[BOUND116]]333; CHECK-NEXT: [[CONFLICT_RDX18:%.*]] = or i1 [[CONFLICT_RDX14]], [[FOUND_CONFLICT17]]334; CHECK-NEXT: br i1 [[CONFLICT_RDX18]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]335; CHECK: vector.ph:336; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[UMAX19]], -4337; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]338; CHECK: vector.body:339; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]340; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i32, ptr [[SRC_1]], i64 [[INDEX]]341; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP2]], align 4, !alias.scope [[META22:![0-9]+]]342; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i64, ptr [[SRC_2]], i64 [[INDEX]]343; CHECK-NEXT: [[WIDE_LOAD20:%.*]] = load <4 x i64>, ptr [[TMP3]], align 8, !alias.scope [[META25:![0-9]+]]344; CHECK-NEXT: [[TMP4:%.*]] = getelementptr nusw i32, ptr [[DST_1]], i64 [[INDEX]]345; CHECK-NEXT: store <4 x i32> [[WIDE_LOAD]], ptr [[TMP4]], align 4, !alias.scope [[META27:![0-9]+]], !noalias [[META29:![0-9]+]]346; CHECK-NEXT: [[TMP5:%.*]] = getelementptr nusw i64, ptr [[DST_2]], i64 [[INDEX]]347; CHECK-NEXT: store <4 x i64> [[WIDE_LOAD20]], ptr [[TMP5]], align 8, !alias.scope [[META31:![0-9]+]], !noalias [[META32:![0-9]+]]348; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4349; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]350; CHECK-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP33:![0-9]+]]351; CHECK: middle.block:352; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[UMAX19]], [[N_VEC]]353; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]354; CHECK: scalar.ph:355; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ], [ 0, [[VECTOR_MEMCHECK]] ]356; CHECK-NEXT: br label [[LOOP:%.*]]357; CHECK: loop:358; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]359; CHECK-NEXT: [[GEP_SRC_1:%.*]] = getelementptr i32, ptr [[SRC_1]], i64 [[IV]]360; CHECK-NEXT: [[LD_SRC_1:%.*]] = load i32, ptr [[GEP_SRC_1]], align 4361; CHECK-NEXT: [[GEP_SRC_2:%.*]] = getelementptr i64, ptr [[SRC_2]], i64 [[IV]]362; CHECK-NEXT: [[LD_SRC_2:%.*]] = load i64, ptr [[GEP_SRC_2]], align 8363; CHECK-NEXT: [[GEP_DST_1:%.*]] = getelementptr nusw i32, ptr [[DST_1]], i64 [[IV]]364; CHECK-NEXT: store i32 [[LD_SRC_1]], ptr [[GEP_DST_1]], align 4365; CHECK-NEXT: [[GEP_DST_2:%.*]] = getelementptr nusw i64, ptr [[DST_2]], i64 [[IV]]366; CHECK-NEXT: store i64 [[LD_SRC_2]], ptr [[GEP_DST_2]], align 8367; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1368; CHECK-NEXT: [[COND:%.*]] = icmp ult i64 [[IV_NEXT]], [[UMAX19]]369; CHECK-NEXT: br i1 [[COND]], label [[LOOP]], label [[EXIT]], !llvm.loop [[LOOP34:![0-9]+]]370; CHECK: exit:371; CHECK-NEXT: ret void372;373; FORCED_OPTSIZE-LABEL: @different_load_store_pairs(374; FORCED_OPTSIZE-NEXT: entry:375; FORCED_OPTSIZE-NEXT: br label [[LOOP:%.*]]376; FORCED_OPTSIZE: loop:377; FORCED_OPTSIZE-NEXT: [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]378; FORCED_OPTSIZE-NEXT: [[GEP_SRC_1:%.*]] = getelementptr i32, ptr [[SRC_1:%.*]], i64 [[IV]]379; FORCED_OPTSIZE-NEXT: [[LD_SRC_1:%.*]] = load i32, ptr [[GEP_SRC_1]], align 4380; FORCED_OPTSIZE-NEXT: [[GEP_SRC_2:%.*]] = getelementptr i64, ptr [[SRC_2:%.*]], i64 [[IV]]381; FORCED_OPTSIZE-NEXT: [[LD_SRC_2:%.*]] = load i64, ptr [[GEP_SRC_2]], align 8382; FORCED_OPTSIZE-NEXT: [[GEP_DST_1:%.*]] = getelementptr nusw i32, ptr [[DST_1:%.*]], i64 [[IV]]383; FORCED_OPTSIZE-NEXT: store i32 [[LD_SRC_1]], ptr [[GEP_DST_1]], align 4384; FORCED_OPTSIZE-NEXT: [[GEP_DST_2:%.*]] = getelementptr nusw i64, ptr [[DST_2:%.*]], i64 [[IV]]385; FORCED_OPTSIZE-NEXT: store i64 [[LD_SRC_2]], ptr [[GEP_DST_2]], align 8386; FORCED_OPTSIZE-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1387; FORCED_OPTSIZE-NEXT: [[COND:%.*]] = icmp ult i64 [[IV_NEXT]], [[N:%.*]]388; FORCED_OPTSIZE-NEXT: br i1 [[COND]], label [[LOOP]], label [[EXIT:%.*]]389; FORCED_OPTSIZE: exit:390; FORCED_OPTSIZE-NEXT: ret void391;392entry:393 br label %loop394 395loop:396 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]397 %gep.src.1 = getelementptr i32, ptr %src.1, i64 %iv398 %ld.src.1 = load i32, ptr %gep.src.1399 %gep.src.2 = getelementptr i64, ptr %src.2, i64 %iv400 %ld.src.2 = load i64, ptr %gep.src.2401 %gep.dst.1 = getelementptr nusw i32, ptr %dst.1, i64 %iv402 store i32 %ld.src.1, ptr %gep.dst.1403 %gep.dst.2 = getelementptr nusw i64, ptr %dst.2, i64 %iv404 store i64 %ld.src.2, ptr %gep.dst.2405 %iv.next = add nuw nsw i64 %iv, 1406 %cond = icmp ult i64 %iv.next, %n407 br i1 %cond, label %loop, label %exit408 409exit:410 ret void411}412 413define dso_local void @forced_optsize(ptr noalias nocapture readonly %x_p, ptr noalias nocapture readonly %y_p, ptr noalias nocapture %z_p) minsize optsize {414; CHECK-LABEL: @forced_optsize(415; CHECK-NEXT: entry:416; CHECK-NEXT: br label [[VECTOR_PH:%.*]]417; CHECK: vector.ph:418; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]419; CHECK: vector.body:420; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]421; CHECK-NEXT: [[TMP0:%.*]] = getelementptr inbounds i64, ptr [[X_P:%.*]], i64 [[INDEX]]422; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i64>, ptr [[TMP0]], align 8423; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i64, ptr [[Y_P:%.*]], i64 [[INDEX]]424; CHECK-NEXT: [[WIDE_LOAD1:%.*]] = load <2 x i64>, ptr [[TMP1]], align 8425; CHECK-NEXT: [[TMP2:%.*]] = add nsw <2 x i64> [[WIDE_LOAD1]], [[WIDE_LOAD]]426; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[Z_P:%.*]], i64 [[INDEX]]427; CHECK-NEXT: store <2 x i64> [[TMP2]], ptr [[TMP3]], align 8428; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2429; CHECK-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128430; CHECK-NEXT: br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP35:![0-9]+]]431; CHECK: middle.block:432; CHECK-NEXT: br label [[FOR_BODY:%.*]]433; CHECK: for.cond.cleanup:434; CHECK-NEXT: ret void435;436; FORCED_OPTSIZE-LABEL: @forced_optsize(437; FORCED_OPTSIZE-NEXT: entry:438; FORCED_OPTSIZE-NEXT: br label [[FOR_BODY:%.*]]439; FORCED_OPTSIZE: for.cond.cleanup:440; FORCED_OPTSIZE-NEXT: ret void441; FORCED_OPTSIZE: for.body:442; FORCED_OPTSIZE-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]443; FORCED_OPTSIZE-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[X_P:%.*]], i64 [[INDVARS_IV]]444; FORCED_OPTSIZE-NEXT: [[TMP0:%.*]] = load i64, ptr [[ARRAYIDX]], align 8445; FORCED_OPTSIZE-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i64, ptr [[Y_P:%.*]], i64 [[INDVARS_IV]]446; FORCED_OPTSIZE-NEXT: [[TMP1:%.*]] = load i64, ptr [[ARRAYIDX2]], align 8447; FORCED_OPTSIZE-NEXT: [[ADD:%.*]] = add nsw i64 [[TMP1]], [[TMP0]]448; FORCED_OPTSIZE-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds i64, ptr [[Z_P:%.*]], i64 [[INDVARS_IV]]449; FORCED_OPTSIZE-NEXT: store i64 [[ADD]], ptr [[ARRAYIDX4]], align 8450; FORCED_OPTSIZE-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1451; FORCED_OPTSIZE-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], 128452; FORCED_OPTSIZE-NEXT: br i1 [[EXITCOND]], label [[FOR_COND_CLEANUP:%.*]], label [[FOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]]453;454entry:455 br label %for.body456 457for.cond.cleanup:458 ret void459 460for.body:461 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]462 %arrayidx = getelementptr inbounds i64, ptr %x_p, i64 %indvars.iv463 %0 = load i64, ptr %arrayidx, align 8464 %arrayidx2 = getelementptr inbounds i64, ptr %y_p, i64 %indvars.iv465 %1 = load i64, ptr %arrayidx2, align 8466 %add = add nsw i64 %1, %0467 %arrayidx4 = getelementptr inbounds i64, ptr %z_p, i64 %indvars.iv468 store i64 %add, ptr %arrayidx4, align 8469 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1470 %exitcond = icmp eq i64 %indvars.iv.next, 128471 br i1 %exitcond, label %for.cond.cleanup, label %for.body, !llvm.loop !12472}473 474declare i1 @cond()475 476define void @test_scev_check_mul_add_expansion(ptr %out, ptr %in, i32 %len, i32 %d) {477; CHECK-LABEL: @test_scev_check_mul_add_expansion(478; CHECK-NEXT: entry:479; CHECK-NEXT: [[PRE_1:%.*]] = icmp samesign ugt i32 [[D:%.*]], 5480; CHECK-NEXT: tail call void @llvm.assume(i1 [[PRE_1]])481; CHECK-NEXT: [[PRE_2:%.*]] = icmp ult i32 [[D]], 7482; CHECK-NEXT: tail call void @llvm.assume(i1 [[PRE_2]])483; CHECK-NEXT: [[PRE_3:%.*]] = icmp slt i32 [[D]], [[LEN:%.*]]484; CHECK-NEXT: tail call void @llvm.assume(i1 [[PRE_3]])485; CHECK-NEXT: [[SMAX3:%.*]] = call i32 @llvm.smax.i32(i32 [[LEN]], i32 7)486; CHECK-NEXT: [[TMP0:%.*]] = add nsw i32 [[SMAX3]], -6487; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp slt i32 [[LEN]], 10488; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]489; CHECK: vector.memcheck:490; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[OUT:%.*]], i64 12491; CHECK-NEXT: [[TMP1:%.*]] = shl nuw i32 [[LEN]], 1492; CHECK-NEXT: [[TMP3:%.*]] = zext i32 [[TMP1]] to i64493; CHECK-NEXT: [[TMP4:%.*]] = getelementptr i8, ptr [[OUT]], i64 [[TMP3]]494; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[IN:%.*]], i64 4495; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[SCEVGEP]], [[SCEVGEP2]]496; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[IN]], [[TMP4]]497; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]498; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]499; CHECK: vector.ph:500; CHECK-NEXT: [[N_VEC:%.*]] = and i32 [[TMP0]], -4501; CHECK-NEXT: [[TMP5:%.*]] = add i32 [[N_VEC]], 6502; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]503; CHECK: vector.body:504; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]505; CHECK-NEXT: [[OFFSET_IDX:%.*]] = add i32 [[INDEX]], 6506; CHECK-NEXT: [[TMP6:%.*]] = sext i32 [[OFFSET_IDX]] to i64507; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i16, ptr [[OUT]], i64 [[TMP6]]508; CHECK-NEXT: store <4 x i16> zeroinitializer, ptr [[TMP7]], align 2, !alias.scope [[META36:![0-9]+]], !noalias [[META39:![0-9]+]]509; CHECK-NEXT: store i32 0, ptr [[IN]], align 4, !alias.scope [[META39]]510; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4511; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]512; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP41:![0-9]+]]513; CHECK: middle.block:514; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]515; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]516; CHECK: scalar.ph:517; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[TMP5]], [[MIDDLE_BLOCK]] ], [ 6, [[ENTRY:%.*]] ], [ 6, [[VECTOR_MEMCHECK]] ]518; CHECK-NEXT: br label [[LOOP:%.*]]519; CHECK: loop:520; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]521; CHECK-NEXT: [[TMP9:%.*]] = sext i32 [[IV]] to i64522; CHECK-NEXT: [[ARRAYIDX80:%.*]] = getelementptr i16, ptr [[OUT]], i64 [[TMP9]]523; CHECK-NEXT: store i16 0, ptr [[ARRAYIDX80]], align 2524; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1525; CHECK-NEXT: store i32 0, ptr [[IN]], align 4526; CHECK-NEXT: [[CMP7_NOT:%.*]] = icmp sgt i32 [[LEN]], [[IV_NEXT]]527; CHECK-NEXT: br i1 [[CMP7_NOT]], label [[LOOP]], label [[EXIT]], !llvm.loop [[LOOP42:![0-9]+]]528; CHECK: exit:529; CHECK-NEXT: ret void530;531; FORCED_OPTSIZE-LABEL: @test_scev_check_mul_add_expansion(532; FORCED_OPTSIZE-NEXT: entry:533; FORCED_OPTSIZE-NEXT: [[PRE_1:%.*]] = icmp sgt i32 [[D:%.*]], 5534; FORCED_OPTSIZE-NEXT: tail call void @llvm.assume(i1 [[PRE_1]])535; FORCED_OPTSIZE-NEXT: [[PRE_2:%.*]] = icmp samesign ule i32 [[D]], 6536; FORCED_OPTSIZE-NEXT: tail call void @llvm.assume(i1 [[PRE_2]])537; FORCED_OPTSIZE-NEXT: [[PRE_3:%.*]] = icmp slt i32 [[D]], [[LEN:%.*]]538; FORCED_OPTSIZE-NEXT: tail call void @llvm.assume(i1 [[PRE_3]])539; FORCED_OPTSIZE-NEXT: br label [[LOOP:%.*]]540; FORCED_OPTSIZE: loop:541; FORCED_OPTSIZE-NEXT: [[IV:%.*]] = phi i32 [ 6, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]542; FORCED_OPTSIZE-NEXT: [[ARRAYIDX80:%.*]] = getelementptr i16, ptr [[OUT:%.*]], i32 [[IV]]543; FORCED_OPTSIZE-NEXT: store i16 0, ptr [[ARRAYIDX80]], align 2544; FORCED_OPTSIZE-NEXT: [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1545; FORCED_OPTSIZE-NEXT: store i32 0, ptr [[IN:%.*]], align 4546; FORCED_OPTSIZE-NEXT: [[CMP7_NOT:%.*]] = icmp sgt i32 [[LEN]], [[IV_NEXT]]547; FORCED_OPTSIZE-NEXT: br i1 [[CMP7_NOT]], label [[LOOP]], label [[EXIT:%.*]]548; FORCED_OPTSIZE: exit:549; FORCED_OPTSIZE-NEXT: ret void550;551entry:552 %pre.1 = icmp sgt i32 %d, 5553 tail call void @llvm.assume(i1 %pre.1)554 %pre.2 = icmp samesign ule i32 %d, 6555 tail call void @llvm.assume(i1 %pre.2)556 %pre.3 = icmp slt i32 %d, %len557 tail call void @llvm.assume(i1 %pre.3)558 br label %loop559 560loop:561 %iv = phi i32 [ 6, %entry ], [ %iv.next, %loop ]562 %arrayidx80 = getelementptr i16, ptr %out, i32 %iv563 store i16 0, ptr %arrayidx80, align 2564 %iv.next = add nuw nsw i32 %iv, 1565 store i32 0, ptr %in, align 4566 %cmp7.not = icmp sgt i32 %len, %iv.next567 br i1 %cmp7.not, label %loop, label %exit568 569exit:570 ret void571}572 573; CHECK: !9 = !DILocation(line: 101, column: 1, scope: !{{.*}})574 575!llvm.module.flags = !{!0, !1}576!llvm.dbg.cu = !{!9}577!0 = !{i32 2, !"Dwarf Version", i32 4}578!1 = !{i32 2, !"Debug Info Version", i32 3}579 580!2 = !{}581!3 = !DISubroutineType(types: !2)582!4 = !DIFile(filename: "test.cpp", directory: "/tmp")583!5 = distinct !DISubprogram(name: "foo", scope: !4, file: !4, line: 99, type: !3, isLocal: false, isDefinition: true, scopeLine: 100, flags: DIFlagPrototyped, isOptimized: false, unit: !9, retainedNodes: !2)584!6 = !DILocation(line: 100, column: 1, scope: !5)585!7 = !DILocation(line: 101, column: 1, scope: !5)586!8 = !DILocation(line: 102, column: 1, scope: !5)587!9 = distinct !DICompileUnit(language: DW_LANG_C99, producer: "clang",588 file: !10,589 isOptimized: true, flags: "-O2",590 splitDebugFilename: "abc.debug", emissionKind: 2)591!10 = !DIFile(filename: "path/to/file", directory: "/path/to/dir")592!11 = !{i32 2, !"Debug Info Version", i32 3}593!12 = distinct !{!12, !13, !14}594!13 = !{!"llvm.loop.vectorize.width", i32 2}595!14 = !{!"llvm.loop.vectorize.enable", i1 true}596