703 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -passes=loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -S | FileCheck --check-prefixes=CHECK,DEFAULT %s3; RUN: opt < %s -passes=loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -lv-strided-pointer-ivs=true -S | FileCheck --check-prefixes=CHECK,STRIDED %s4target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"5 6; Note: Most scalar pointer induction GEPs could be sunk into the conditional7; blocks.8; Function Attrs: nofree norecurse nounwind9define void @a(ptr readnone %b) {10; CHECK-LABEL: @a(11; CHECK-NEXT: entry:12; CHECK-NEXT: [[B1:%.*]] = ptrtoint ptr [[B:%.*]] to i6413; CHECK-NEXT: [[CMP_NOT4:%.*]] = icmp eq ptr [[B]], null14; CHECK-NEXT: br i1 [[CMP_NOT4]], label [[FOR_COND_CLEANUP:%.*]], label [[FOR_BODY_PREHEADER:%.*]]15; CHECK: for.body.preheader:16; CHECK-NEXT: [[TMP0:%.*]] = sub i64 0, [[B1]]17; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 418; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]19; CHECK: vector.ph:20; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 421; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]22; CHECK-NEXT: [[TMP1:%.*]] = mul i64 [[N_VEC]], -123; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr null, i64 [[TMP1]]24; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]25; CHECK: vector.body:26; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE10:%.*]] ]27; CHECK-NEXT: [[OFFSET_IDX:%.*]] = sub i64 0, [[INDEX]]28; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[OFFSET_IDX]], 029; CHECK-NEXT: [[TMP11:%.*]] = add i64 [[OFFSET_IDX]], -130; CHECK-NEXT: [[TMP14:%.*]] = add i64 [[OFFSET_IDX]], -231; CHECK-NEXT: [[TMP17:%.*]] = add i64 [[OFFSET_IDX]], -332; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr null, i64 [[TMP2]]33; CHECK-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr null, i64 [[TMP11]]34; CHECK-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr null, i64 [[TMP14]]35; CHECK-NEXT: [[NEXT_GEP4:%.*]] = getelementptr i8, ptr null, i64 [[TMP17]]36; CHECK-NEXT: [[TMP21:%.*]] = insertelement <4 x ptr> poison, ptr [[NEXT_GEP]], i32 037; CHECK-NEXT: [[TMP22:%.*]] = insertelement <4 x ptr> [[TMP21]], ptr [[NEXT_GEP2]], i32 138; CHECK-NEXT: [[TMP23:%.*]] = insertelement <4 x ptr> [[TMP22]], ptr [[NEXT_GEP3]], i32 239; CHECK-NEXT: [[TMP24:%.*]] = insertelement <4 x ptr> [[TMP23]], ptr [[NEXT_GEP4]], i32 340; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i8, ptr [[NEXT_GEP]], i64 -141; CHECK-NEXT: [[TMP25:%.*]] = getelementptr inbounds i8, ptr [[TMP3]], i64 042; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i8, ptr [[TMP25]], i64 -343; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i8>, ptr [[TMP5]], align 144; CHECK-NEXT: [[REVERSE:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>45; CHECK-NEXT: [[TMP7:%.*]] = icmp ne <4 x i8> [[REVERSE]], zeroinitializer46; CHECK-NEXT: [[TMP8:%.*]] = extractelement <4 x i1> [[TMP7]], i32 047; CHECK-NEXT: br i1 [[TMP8]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]48; CHECK: pred.store.if:49; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds i8, ptr [[NEXT_GEP]], i64 -150; CHECK-NEXT: store i8 95, ptr [[TMP9]], align 151; CHECK-NEXT: br label [[PRED_STORE_CONTINUE]]52; CHECK: pred.store.continue:53; CHECK-NEXT: [[TMP10:%.*]] = extractelement <4 x i1> [[TMP7]], i32 154; CHECK-NEXT: br i1 [[TMP10]], label [[PRED_STORE_IF5:%.*]], label [[PRED_STORE_CONTINUE6:%.*]]55; CHECK: pred.store.if5:56; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds i8, ptr [[NEXT_GEP2]], i64 -157; CHECK-NEXT: store i8 95, ptr [[TMP12]], align 158; CHECK-NEXT: br label [[PRED_STORE_CONTINUE6]]59; CHECK: pred.store.continue6:60; CHECK-NEXT: [[TMP13:%.*]] = extractelement <4 x i1> [[TMP7]], i32 261; CHECK-NEXT: br i1 [[TMP13]], label [[PRED_STORE_IF7:%.*]], label [[PRED_STORE_CONTINUE8:%.*]]62; CHECK: pred.store.if7:63; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds i8, ptr [[NEXT_GEP3]], i64 -164; CHECK-NEXT: store i8 95, ptr [[TMP15]], align 165; CHECK-NEXT: br label [[PRED_STORE_CONTINUE8]]66; CHECK: pred.store.continue8:67; CHECK-NEXT: [[TMP16:%.*]] = extractelement <4 x i1> [[TMP7]], i32 368; CHECK-NEXT: br i1 [[TMP16]], label [[PRED_STORE_IF9:%.*]], label [[PRED_STORE_CONTINUE10]]69; CHECK: pred.store.if9:70; CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds i8, ptr [[NEXT_GEP4]], i64 -171; CHECK-NEXT: store i8 95, ptr [[TMP18]], align 172; CHECK-NEXT: br label [[PRED_STORE_CONTINUE10]]73; CHECK: pred.store.continue10:74; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 475; CHECK-NEXT: [[TMP19:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]76; CHECK-NEXT: br i1 [[TMP19]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]77; CHECK: middle.block:78; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]79; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]], label [[SCALAR_PH]]80; CHECK: scalar.ph:81; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ null, [[FOR_BODY_PREHEADER]] ]82; CHECK-NEXT: br label [[FOR_BODY:%.*]]83; CHECK: for.cond.cleanup.loopexit:84; CHECK-NEXT: br label [[FOR_COND_CLEANUP]]85; CHECK: for.cond.cleanup:86; CHECK-NEXT: ret void87; CHECK: for.body:88; CHECK-NEXT: [[C_05:%.*]] = phi ptr [ [[INCDEC_PTR:%.*]], [[IF_END:%.*]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]89; CHECK-NEXT: [[INCDEC_PTR]] = getelementptr inbounds i8, ptr [[C_05]], i64 -190; CHECK-NEXT: [[TMP20:%.*]] = load i8, ptr [[INCDEC_PTR]], align 191; CHECK-NEXT: [[TOBOOL_NOT:%.*]] = icmp eq i8 [[TMP20]], 092; CHECK-NEXT: br i1 [[TOBOOL_NOT]], label [[IF_END]], label [[IF_THEN:%.*]]93; CHECK: if.then:94; CHECK-NEXT: store i8 95, ptr [[INCDEC_PTR]], align 195; CHECK-NEXT: br label [[IF_END]]96; CHECK: if.end:97; CHECK-NEXT: [[CMP_NOT:%.*]] = icmp eq ptr [[INCDEC_PTR]], [[B]]98; CHECK-NEXT: br i1 [[CMP_NOT]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]99;100 101entry:102 %cmp.not4 = icmp eq ptr %b, null103 br i1 %cmp.not4, label %for.cond.cleanup, label %for.body.preheader104 105for.body.preheader: ; preds = %entry106 br label %for.body107 108for.cond.cleanup.loopexit: ; preds = %if.end109 br label %for.cond.cleanup110 111for.cond.cleanup: ; preds = %for.cond.cleanup.loopexit, %entry112 ret void113 114for.body: ; preds = %for.body.preheader, %if.end115 %c.05 = phi ptr [ %incdec.ptr, %if.end ], [ null, %for.body.preheader ]116 %incdec.ptr = getelementptr inbounds i8, ptr %c.05, i64 -1117 %0 = load i8, ptr %incdec.ptr, align 1118 %tobool.not = icmp eq i8 %0, 0119 br i1 %tobool.not, label %if.end, label %if.then120 121if.then: ; preds = %for.body122 store i8 95, ptr %incdec.ptr, align 1123 br label %if.end124 125if.end: ; preds = %for.body, %if.then126 %cmp.not = icmp eq ptr %incdec.ptr, %b127 br i1 %cmp.not, label %for.cond.cleanup.loopexit, label %for.body128}129 130; In the test below the pointer phi %ptr.iv.2 is used as131; 1. As a uniform address for the load, and132; 2. Non-uniform use by the getelementptr which is stored. This requires the133; vector value.134define void @pointer_induction_used_as_vector(ptr noalias %start.1, ptr noalias %start.2, i64 %N) {135; CHECK-LABEL: @pointer_induction_used_as_vector(136; CHECK-NEXT: entry:137; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 4138; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]139; CHECK: vector.ph:140; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 4141; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]142; CHECK-NEXT: [[TMP0:%.*]] = mul i64 [[N_VEC]], 8143; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[START_1:%.*]], i64 [[TMP0]]144; CHECK-NEXT: [[IND_END2:%.*]] = getelementptr i8, ptr [[START_2:%.*]], i64 [[N_VEC]]145; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]146; CHECK: vector.body:147; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]148; CHECK-NEXT: [[POINTER_PHI:%.*]] = phi ptr [ [[START_2]], [[VECTOR_PH]] ], [ [[PTR_IND:%.*]], [[VECTOR_BODY]] ]149; CHECK-NEXT: [[VECTOR_GEP:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <4 x i64> <i64 0, i64 1, i64 2, i64 3>150; CHECK-NEXT: [[TMP4:%.*]] = extractelement <4 x ptr> [[VECTOR_GEP]], i32 0151; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 8152; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[START_1]], i64 [[OFFSET_IDX]]153; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, <4 x ptr> [[VECTOR_GEP]], i64 1154; CHECK-NEXT: store <4 x ptr> [[TMP2]], ptr [[NEXT_GEP]], align 8155; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i8>, ptr [[TMP4]], align 1156; CHECK-NEXT: [[TMP6:%.*]] = add <4 x i8> [[WIDE_LOAD]], splat (i8 1)157; CHECK-NEXT: store <4 x i8> [[TMP6]], ptr [[TMP4]], align 1158; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4159; CHECK-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 4160; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]161; CHECK-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]162; CHECK: middle.block:163; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]164; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]165; CHECK: scalar.ph:166; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]167; CHECK-NEXT: [[BC_RESUME_VAL1:%.*]] = phi ptr [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[START_1]], [[ENTRY]] ]168; CHECK-NEXT: [[BC_RESUME_VAL3:%.*]] = phi ptr [ [[IND_END2]], [[MIDDLE_BLOCK]] ], [ [[START_2]], [[ENTRY]] ]169; CHECK-NEXT: br label [[LOOP_BODY:%.*]]170; CHECK: loop.body:171; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_BODY]] ]172; CHECK-NEXT: [[PTR_IV_1:%.*]] = phi ptr [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[PTR_IV_1_NEXT:%.*]], [[LOOP_BODY]] ]173; CHECK-NEXT: [[PTR_IV_2:%.*]] = phi ptr [ [[BC_RESUME_VAL3]], [[SCALAR_PH]] ], [ [[PTR_IV_2_NEXT:%.*]], [[LOOP_BODY]] ]174; CHECK-NEXT: [[PTR_IV_1_NEXT]] = getelementptr inbounds ptr, ptr [[PTR_IV_1]], i64 1175; CHECK-NEXT: [[PTR_IV_2_NEXT]] = getelementptr inbounds i8, ptr [[PTR_IV_2]], i64 1176; CHECK-NEXT: store ptr [[PTR_IV_2_NEXT]], ptr [[PTR_IV_1]], align 8177; CHECK-NEXT: [[LV:%.*]] = load i8, ptr [[PTR_IV_2]], align 1178; CHECK-NEXT: [[ADD:%.*]] = add i8 [[LV]], 1179; CHECK-NEXT: store i8 [[ADD]], ptr [[PTR_IV_2]], align 1180; CHECK-NEXT: [[IV_NEXT]] = add nuw i64 [[IV]], 1181; CHECK-NEXT: [[C:%.*]] = icmp ne i64 [[IV_NEXT]], [[N]]182; CHECK-NEXT: br i1 [[C]], label [[LOOP_BODY]], label [[EXIT]], !llvm.loop [[LOOP5:![0-9]+]]183; CHECK: exit:184; CHECK-NEXT: ret void185;186 187entry:188 br label %loop.body189 190loop.body: ; preds = %loop.body, %entry191 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.body ]192 %ptr.iv.1 = phi ptr [ %start.1, %entry ], [ %ptr.iv.1.next, %loop.body ]193 %ptr.iv.2 = phi ptr [ %start.2, %entry ], [ %ptr.iv.2.next, %loop.body ]194 %ptr.iv.1.next = getelementptr inbounds ptr, ptr %ptr.iv.1, i64 1195 %ptr.iv.2.next = getelementptr inbounds i8, ptr %ptr.iv.2, i64 1196 store ptr %ptr.iv.2.next, ptr %ptr.iv.1, align 8197 %lv = load i8, ptr %ptr.iv.2, align 1198 %add = add i8 %lv, 1199 store i8 %add, ptr %ptr.iv.2, align 1200 %iv.next = add nuw i64 %iv, 1201 %c = icmp ne i64 %iv.next, %N202 br i1 %c, label %loop.body, label %exit203 204exit: ; preds = %loop.body205 ret void206}207 208; Test the vector expansion of a non-constant stride pointer IV209define void @non_constant_vector_expansion(i32 %0, ptr %call) {210; DEFAULT-LABEL: @non_constant_vector_expansion(211; DEFAULT-NEXT: entry:212; DEFAULT-NEXT: [[MUL:%.*]] = shl i32 [[TMP0:%.*]], 1213; DEFAULT-NEXT: br label [[FOR_COND:%.*]]214; DEFAULT: for.cond:215; DEFAULT-NEXT: [[TMP1:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[INC:%.*]], [[FOR_COND]] ]216; DEFAULT-NEXT: [[P_0:%.*]] = phi ptr [ null, [[ENTRY]] ], [ [[ADD_PTR:%.*]], [[FOR_COND]] ]217; DEFAULT-NEXT: [[ADD_PTR]] = getelementptr i8, ptr [[P_0]], i32 [[MUL]]218; DEFAULT-NEXT: [[ARRAYIDX:%.*]] = getelementptr ptr, ptr [[CALL:%.*]], i32 [[TMP1]]219; DEFAULT-NEXT: store ptr [[P_0]], ptr [[ARRAYIDX]], align 4220; DEFAULT-NEXT: [[INC]] = add i32 [[TMP1]], 1221; DEFAULT-NEXT: [[TOBOOL_NOT:%.*]] = icmp eq i32 [[TMP1]], 100222; DEFAULT-NEXT: br i1 [[TOBOOL_NOT]], label [[FOR_END:%.*]], label [[FOR_COND]]223; DEFAULT: for.end:224; DEFAULT-NEXT: ret void225;226; STRIDED-LABEL: @non_constant_vector_expansion(227; STRIDED-NEXT: entry:228; STRIDED-NEXT: [[MUL:%.*]] = shl i32 [[TMP0:%.*]], 1229; STRIDED-NEXT: [[TMP1:%.*]] = sext i32 [[MUL]] to i64230; STRIDED-NEXT: br label [[VECTOR_PH:%.*]]231; STRIDED: vector.ph:232; STRIDED-NEXT: [[TMP2:%.*]] = mul i64 100, [[TMP1]]233; STRIDED-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr null, i64 [[TMP2]]234; STRIDED-NEXT: br label [[VECTOR_BODY:%.*]]235; STRIDED: vector.body:236; STRIDED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]237; STRIDED-NEXT: [[POINTER_PHI:%.*]] = phi ptr [ null, [[VECTOR_PH]] ], [ [[PTR_IND:%.*]], [[VECTOR_BODY]] ]238; STRIDED-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[TMP1]], i64 0239; STRIDED-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x i64> [[DOTSPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer240; STRIDED-NEXT: [[TMP4:%.*]] = mul <4 x i64> <i64 0, i64 1, i64 2, i64 3>, [[DOTSPLAT]]241; STRIDED-NEXT: [[VECTOR_GEP:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <4 x i64> [[TMP4]]242; STRIDED-NEXT: [[OFFSET_IDX:%.*]] = trunc i64 [[INDEX]] to i32243; STRIDED-NEXT: [[TMP6:%.*]] = getelementptr ptr, ptr [[CALL:%.*]], i32 [[OFFSET_IDX]]244; STRIDED-NEXT: store <4 x ptr> [[VECTOR_GEP]], ptr [[TMP6]], align 4245; STRIDED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4246; STRIDED-NEXT: [[TMP3:%.*]] = mul i64 [[TMP1]], 4247; STRIDED-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 [[TMP3]]248; STRIDED-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], 100249; STRIDED-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]250; STRIDED: middle.block:251; STRIDED-NEXT: br label [[SCALAR_PH:%.*]]252; STRIDED: scalar.ph:253; STRIDED-NEXT: br label [[FOR_COND:%.*]]254; STRIDED: for.cond:255; STRIDED-NEXT: [[TMP9:%.*]] = phi i32 [ 100, [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_COND]] ]256; STRIDED-NEXT: [[P_0:%.*]] = phi ptr [ [[IND_END]], [[SCALAR_PH]] ], [ [[ADD_PTR:%.*]], [[FOR_COND]] ]257; STRIDED-NEXT: [[ADD_PTR]] = getelementptr i8, ptr [[P_0]], i32 [[MUL]]258; STRIDED-NEXT: [[ARRAYIDX:%.*]] = getelementptr ptr, ptr [[CALL]], i32 [[TMP9]]259; STRIDED-NEXT: store ptr [[P_0]], ptr [[ARRAYIDX]], align 4260; STRIDED-NEXT: [[INC]] = add i32 [[TMP9]], 1261; STRIDED-NEXT: [[TOBOOL_NOT:%.*]] = icmp eq i32 [[TMP9]], 100262; STRIDED-NEXT: br i1 [[TOBOOL_NOT]], label [[FOR_END:%.*]], label [[FOR_COND]], !llvm.loop [[LOOP7:![0-9]+]]263; STRIDED: for.end:264; STRIDED-NEXT: ret void265;266entry:267 %mul = shl i32 %0, 1268 br label %for.cond269 270for.cond: ; preds = %for.body, %entry271 %1 = phi i32 [ 0, %entry ], [ %inc, %for.cond ]272 %p.0 = phi ptr [ null, %entry ], [ %add.ptr, %for.cond ]273 %add.ptr = getelementptr i8, ptr %p.0, i32 %mul274 %arrayidx = getelementptr ptr, ptr %call, i32 %1275 store ptr %p.0, ptr %arrayidx, align 4276 %inc = add i32 %1, 1277 %tobool.not = icmp eq i32 %1, 100278 br i1 %tobool.not, label %for.end, label %for.cond279 280 281for.end: ; preds = %for.cond282 ret void283}284 285; Test that when WidenPointerInductionRecipes are ordered before other286; WidenIntOrFpInductionRecipes that their PHIs are emitted in the right place.287define void @outside_lattice(ptr noalias %p, ptr noalias %q, i32 %n) {288; DEFAULT-LABEL: @outside_lattice(289; DEFAULT-NEXT: entry:290; DEFAULT-NEXT: [[TMP0:%.*]] = zext i32 [[N:%.*]] to i64291; DEFAULT-NEXT: [[UMAX1:%.*]] = call i64 @llvm.umax.i64(i64 [[TMP0]], i64 1)292; DEFAULT-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[UMAX1]], 4293; DEFAULT-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]294; DEFAULT: vector.scevcheck:295; DEFAULT-NEXT: [[UMAX:%.*]] = call i32 @llvm.umax.i32(i32 [[N]], i32 1)296; DEFAULT-NEXT: [[TMP1:%.*]] = add i32 [[UMAX]], -1297; DEFAULT-NEXT: [[TMP2:%.*]] = icmp slt i32 [[TMP1]], 0298; DEFAULT-NEXT: br i1 [[TMP2]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]299; DEFAULT: vector.ph:300; DEFAULT-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[UMAX1]], 4301; DEFAULT-NEXT: [[N_VEC:%.*]] = sub i64 [[UMAX1]], [[N_MOD_VF]]302; DEFAULT-NEXT: [[TMP3:%.*]] = mul i64 [[N_VEC]], 4303; DEFAULT-NEXT: [[TMP4:%.*]] = getelementptr i8, ptr [[P:%.*]], i64 [[TMP3]]304; DEFAULT-NEXT: [[IND_END2:%.*]] = trunc i64 [[N_VEC]] to i32305; DEFAULT-NEXT: br label [[VECTOR_BODY:%.*]]306; DEFAULT: vector.body:307; DEFAULT-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]308; DEFAULT-NEXT: [[POINTER_PHI:%.*]] = phi ptr [ [[P]], [[VECTOR_PH]] ], [ [[PTR_IND:%.*]], [[VECTOR_BODY]] ]309; DEFAULT-NEXT: [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]310; DEFAULT-NEXT: [[VECTOR_GEP:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <4 x i64> <i64 0, i64 4, i64 8, i64 12>311; DEFAULT-NEXT: [[OFFSET_IDX:%.*]] = trunc i64 [[INDEX]] to i32312; DEFAULT-NEXT: [[TMP5:%.*]] = getelementptr inbounds ptr, ptr [[P]], i32 [[OFFSET_IDX]]313; DEFAULT-NEXT: store <4 x ptr> [[VECTOR_GEP]], ptr [[TMP5]], align 8314; DEFAULT-NEXT: [[TMP7:%.*]] = getelementptr inbounds i32, ptr [[Q:%.*]], i32 [[OFFSET_IDX]]315; DEFAULT-NEXT: store <4 x i32> [[VEC_IND]], ptr [[TMP7]], align 4316; DEFAULT-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4317; DEFAULT-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 16318; DEFAULT-NEXT: [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], splat (i32 4)319; DEFAULT-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]320; DEFAULT-NEXT: br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]321; DEFAULT: middle.block:322; DEFAULT-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[UMAX1]], [[N_VEC]]323; DEFAULT-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]324; DEFAULT: scalar.ph:325; DEFAULT-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[TMP4]], [[MIDDLE_BLOCK]] ], [ [[P]], [[ENTRY:%.*]] ], [ [[P]], [[VECTOR_SCEVCHECK]] ]326; DEFAULT-NEXT: [[BC_RESUME_VAL3:%.*]] = phi i32 [ [[IND_END2]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ], [ 0, [[VECTOR_SCEVCHECK]] ]327; DEFAULT-NEXT: br label [[FOR_BODY:%.*]]328; DEFAULT: for.body:329; DEFAULT-NEXT: [[IV_PTR:%.*]] = phi ptr [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_PTR_NEXT:%.*]], [[FOR_BODY]] ]330; DEFAULT-NEXT: [[IV_INT:%.*]] = phi i32 [ [[BC_RESUME_VAL3]], [[SCALAR_PH]] ], [ [[IV_INT_NEXT:%.*]], [[FOR_BODY]] ]331; DEFAULT-NEXT: [[P_GEP:%.*]] = getelementptr inbounds ptr, ptr [[P]], i32 [[IV_INT]]332; DEFAULT-NEXT: store ptr [[IV_PTR]], ptr [[P_GEP]], align 8333; DEFAULT-NEXT: [[Q_GEP:%.*]] = getelementptr inbounds i32, ptr [[Q]], i32 [[IV_INT]]334; DEFAULT-NEXT: store i32 [[IV_INT]], ptr [[Q_GEP]], align 4335; DEFAULT-NEXT: [[IV_INT_NEXT]] = add i32 [[IV_INT]], 1336; DEFAULT-NEXT: [[IV_PTR_NEXT]] = getelementptr inbounds i32, ptr [[IV_PTR]], i32 1337; DEFAULT-NEXT: [[DONE:%.*]] = icmp ult i32 [[IV_INT_NEXT]], [[N]]338; DEFAULT-NEXT: br i1 [[DONE]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP7:![0-9]+]]339; DEFAULT: for.end:340; DEFAULT-NEXT: ret void341;342; STRIDED-LABEL: @outside_lattice(343; STRIDED-NEXT: entry:344; STRIDED-NEXT: [[TMP0:%.*]] = zext i32 [[N:%.*]] to i64345; STRIDED-NEXT: [[UMAX1:%.*]] = call i64 @llvm.umax.i64(i64 [[TMP0]], i64 1)346; STRIDED-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[UMAX1]], 4347; STRIDED-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]348; STRIDED: vector.scevcheck:349; STRIDED-NEXT: [[UMAX:%.*]] = call i32 @llvm.umax.i32(i32 [[N]], i32 1)350; STRIDED-NEXT: [[TMP1:%.*]] = add i32 [[UMAX]], -1351; STRIDED-NEXT: [[TMP2:%.*]] = icmp slt i32 [[TMP1]], 0352; STRIDED-NEXT: br i1 [[TMP2]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]353; STRIDED: vector.ph:354; STRIDED-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[UMAX1]], 4355; STRIDED-NEXT: [[N_VEC:%.*]] = sub i64 [[UMAX1]], [[N_MOD_VF]]356; STRIDED-NEXT: [[TMP3:%.*]] = mul i64 [[N_VEC]], 4357; STRIDED-NEXT: [[TMP4:%.*]] = getelementptr i8, ptr [[P:%.*]], i64 [[TMP3]]358; STRIDED-NEXT: [[IND_END2:%.*]] = trunc i64 [[N_VEC]] to i32359; STRIDED-NEXT: br label [[VECTOR_BODY:%.*]]360; STRIDED: vector.body:361; STRIDED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]362; STRIDED-NEXT: [[POINTER_PHI:%.*]] = phi ptr [ [[P]], [[VECTOR_PH]] ], [ [[PTR_IND:%.*]], [[VECTOR_BODY]] ]363; STRIDED-NEXT: [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]364; STRIDED-NEXT: [[VECTOR_GEP:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <4 x i64> <i64 0, i64 4, i64 8, i64 12>365; STRIDED-NEXT: [[OFFSET_IDX:%.*]] = trunc i64 [[INDEX]] to i32366; STRIDED-NEXT: [[TMP5:%.*]] = getelementptr inbounds ptr, ptr [[P]], i32 [[OFFSET_IDX]]367; STRIDED-NEXT: store <4 x ptr> [[VECTOR_GEP]], ptr [[TMP5]], align 8368; STRIDED-NEXT: [[TMP7:%.*]] = getelementptr inbounds i32, ptr [[Q:%.*]], i32 [[OFFSET_IDX]]369; STRIDED-NEXT: store <4 x i32> [[VEC_IND]], ptr [[TMP7]], align 4370; STRIDED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4371; STRIDED-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 16372; STRIDED-NEXT: [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], splat (i32 4)373; STRIDED-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]374; STRIDED-NEXT: br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]375; STRIDED: middle.block:376; STRIDED-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[UMAX1]], [[N_VEC]]377; STRIDED-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]378; STRIDED: scalar.ph:379; STRIDED-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[TMP4]], [[MIDDLE_BLOCK]] ], [ [[P]], [[ENTRY:%.*]] ], [ [[P]], [[VECTOR_SCEVCHECK]] ]380; STRIDED-NEXT: [[BC_RESUME_VAL3:%.*]] = phi i32 [ [[IND_END2]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ], [ 0, [[VECTOR_SCEVCHECK]] ]381; STRIDED-NEXT: br label [[FOR_BODY:%.*]]382; STRIDED: for.body:383; STRIDED-NEXT: [[IV_PTR:%.*]] = phi ptr [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_PTR_NEXT:%.*]], [[FOR_BODY]] ]384; STRIDED-NEXT: [[IV_INT:%.*]] = phi i32 [ [[BC_RESUME_VAL3]], [[SCALAR_PH]] ], [ [[IV_INT_NEXT:%.*]], [[FOR_BODY]] ]385; STRIDED-NEXT: [[P_GEP:%.*]] = getelementptr inbounds ptr, ptr [[P]], i32 [[IV_INT]]386; STRIDED-NEXT: store ptr [[IV_PTR]], ptr [[P_GEP]], align 8387; STRIDED-NEXT: [[Q_GEP:%.*]] = getelementptr inbounds i32, ptr [[Q]], i32 [[IV_INT]]388; STRIDED-NEXT: store i32 [[IV_INT]], ptr [[Q_GEP]], align 4389; STRIDED-NEXT: [[IV_INT_NEXT]] = add i32 [[IV_INT]], 1390; STRIDED-NEXT: [[IV_PTR_NEXT]] = getelementptr inbounds i32, ptr [[IV_PTR]], i32 1391; STRIDED-NEXT: [[DONE:%.*]] = icmp ult i32 [[IV_INT_NEXT]], [[N]]392; STRIDED-NEXT: br i1 [[DONE]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP9:![0-9]+]]393; STRIDED: for.end:394; STRIDED-NEXT: ret void395;396entry:397 br label %for.body398 399for.body: ; preds = %for.body, %entry400 %iv.ptr = phi ptr [ %p, %entry ], [ %iv.ptr.next, %for.body ]401 %iv.int = phi i32 [ 0, %entry ], [ %iv.int.next, %for.body ]402 403 %p.gep = getelementptr inbounds ptr, ptr %p, i32 %iv.int404 store ptr %iv.ptr, ptr %p.gep405 406 %q.gep = getelementptr inbounds i32, ptr %q, i32 %iv.int407 store i32 %iv.int, ptr %q.gep408 409 %iv.int.next = add i32 %iv.int, 1410 %iv.ptr.next = getelementptr inbounds i32, ptr %iv.ptr, i32 1411 412 %done = icmp ult i32 %iv.int.next, %n413 br i1 %done, label %for.body, label %for.end414 415for.end: ; preds = %for.body416 ret void417}418 419define i64 @ivopt_widen_ptr_indvar_1(ptr noalias %a, i64 %stride, i64 %n) {420; DEFAULT-LABEL: @ivopt_widen_ptr_indvar_1(421; DEFAULT-NEXT: entry:422; DEFAULT-NEXT: br label [[FOR_BODY:%.*]]423; DEFAULT: for.body:424; DEFAULT-NEXT: [[INDVAR:%.*]] = phi i64 [ [[INDVAR_NEXT:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ]425; DEFAULT-NEXT: [[PTR_IV:%.*]] = phi ptr [ [[PTR_IV_NEXT:%.*]], [[FOR_BODY]] ], [ null, [[ENTRY]] ]426; DEFAULT-NEXT: [[ARRAYIDX:%.*]] = getelementptr i64, ptr [[A:%.*]], i64 [[INDVAR]]427; DEFAULT-NEXT: [[TMP0:%.*]] = load i64, ptr [[ARRAYIDX]], align 8428; DEFAULT-NEXT: [[PTR_IV_NEXT]] = getelementptr i64, ptr [[PTR_IV]], i64 [[STRIDE:%.*]]429; DEFAULT-NEXT: [[INDVAR_NEXT]] = add i64 [[INDVAR]], 1430; DEFAULT-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVAR]], [[N:%.*]]431; DEFAULT-NEXT: br i1 [[EXITCOND_NOT]], label [[EXIT:%.*]], label [[FOR_BODY]]432; DEFAULT: exit:433; DEFAULT-NEXT: [[PTR_IV_LCSSA:%.*]] = phi ptr [ [[PTR_IV]], [[FOR_BODY]] ]434; DEFAULT-NEXT: [[DOTLCSSA:%.*]] = phi i64 [ [[TMP0]], [[FOR_BODY]] ]435; DEFAULT-NEXT: [[CAST_PTR:%.*]] = ptrtoint ptr [[PTR_IV_LCSSA]] to i64436; DEFAULT-NEXT: [[RESULT:%.*]] = add i64 [[CAST_PTR]], [[DOTLCSSA]]437; DEFAULT-NEXT: ret i64 [[RESULT]]438;439; STRIDED-LABEL: @ivopt_widen_ptr_indvar_1(440; STRIDED-NEXT: entry:441; STRIDED-NEXT: [[TMP0:%.*]] = add i64 [[N:%.*]], 1442; STRIDED-NEXT: [[TMP1:%.*]] = shl i64 [[STRIDE:%.*]], 3443; STRIDED-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 4444; STRIDED-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]445; STRIDED: vector.ph:446; STRIDED-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 4447; STRIDED-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]448; STRIDED-NEXT: [[TMP2:%.*]] = mul i64 [[N_VEC]], [[TMP1]]449; STRIDED-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr null, i64 [[TMP2]]450; STRIDED-NEXT: br label [[VECTOR_BODY:%.*]]451; STRIDED: vector.body:452; STRIDED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]453; STRIDED-NEXT: [[TMP4:%.*]] = getelementptr i64, ptr [[A:%.*]], i64 [[INDEX]]454; STRIDED-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP4]], align 8455; STRIDED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4456; STRIDED-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]457; STRIDED-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]458; STRIDED: middle.block:459; STRIDED-NEXT: [[TMP7:%.*]] = extractelement <4 x i64> [[WIDE_LOAD]], i32 3460; STRIDED-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]461; STRIDED-NEXT: [[TMP8:%.*]] = sub i64 0, [[TMP1]]462; STRIDED-NEXT: [[IND_ESCAPE:%.*]] = getelementptr i8, ptr [[TMP3]], i64 [[TMP8]]463; STRIDED-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]464; STRIDED: scalar.ph:465; STRIDED-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]466; STRIDED-NEXT: [[BC_RESUME_VAL1:%.*]] = phi ptr [ [[TMP3]], [[MIDDLE_BLOCK]] ], [ null, [[ENTRY]] ]467; STRIDED-NEXT: br label [[FOR_BODY:%.*]]468; STRIDED: for.body:469; STRIDED-NEXT: [[INDVAR:%.*]] = phi i64 [ [[INDVAR_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]470; STRIDED-NEXT: [[PTR_IV:%.*]] = phi ptr [ [[PTR_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ]471; STRIDED-NEXT: [[ARRAYIDX:%.*]] = getelementptr i64, ptr [[A]], i64 [[INDVAR]]472; STRIDED-NEXT: [[TMP9:%.*]] = load i64, ptr [[ARRAYIDX]], align 8473; STRIDED-NEXT: [[PTR_IV_NEXT]] = getelementptr i64, ptr [[PTR_IV]], i64 [[STRIDE]]474; STRIDED-NEXT: [[INDVAR_NEXT]] = add i64 [[INDVAR]], 1475; STRIDED-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVAR]], [[N]]476; STRIDED-NEXT: br i1 [[EXITCOND_NOT]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]]477; STRIDED: exit:478; STRIDED-NEXT: [[PTR_IV_LCSSA:%.*]] = phi ptr [ [[PTR_IV]], [[FOR_BODY]] ], [ [[IND_ESCAPE]], [[MIDDLE_BLOCK]] ]479; STRIDED-NEXT: [[DOTLCSSA:%.*]] = phi i64 [ [[TMP9]], [[FOR_BODY]] ], [ [[TMP7]], [[MIDDLE_BLOCK]] ]480; STRIDED-NEXT: [[CAST_PTR:%.*]] = ptrtoint ptr [[PTR_IV_LCSSA]] to i64481; STRIDED-NEXT: [[RESULT:%.*]] = add i64 [[CAST_PTR]], [[DOTLCSSA]]482; STRIDED-NEXT: ret i64 [[RESULT]]483;484entry:485 br label %for.body486 487for.body:488 %indvar = phi i64 [ %indvar.next, %for.body ], [ 0, %entry ]489 %ptr.iv = phi ptr [ %ptr.iv.next, %for.body ], [ null, %entry ]490 %arrayidx = getelementptr i64, ptr %a, i64 %indvar491 %0 = load i64, ptr %arrayidx, align 8492 %ptr.iv.next = getelementptr i64, ptr %ptr.iv, i64 %stride493 %indvar.next = add i64 %indvar, 1494 %exitcond.not = icmp eq i64 %indvar, %n495 br i1 %exitcond.not, label %exit, label %for.body496 497exit:498 %cast.ptr = ptrtoint ptr %ptr.iv to i64499 %result = add i64 %cast.ptr, %0500 ret i64 %result501}502 503define i64 @ivopt_widen_ptr_indvar_2(ptr noalias %a, i64 %stride, i64 %n) {504; DEFAULT-LABEL: @ivopt_widen_ptr_indvar_2(505; DEFAULT-NEXT: entry:506; DEFAULT-NEXT: br label [[FOR_BODY:%.*]]507; DEFAULT: for.body:508; DEFAULT-NEXT: [[INDVAR:%.*]] = phi i64 [ [[INDVAR_NEXT:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ]509; DEFAULT-NEXT: [[PTR_IV:%.*]] = phi ptr [ [[PTR_IV_NEXT:%.*]], [[FOR_BODY]] ], [ null, [[ENTRY]] ]510; DEFAULT-NEXT: [[ARRAYIDX:%.*]] = getelementptr i64, ptr [[A:%.*]], i64 [[INDVAR]]511; DEFAULT-NEXT: [[TMP0:%.*]] = load i64, ptr [[ARRAYIDX]], align 8512; DEFAULT-NEXT: store i64 [[TMP0]], ptr [[PTR_IV]], align 8513; DEFAULT-NEXT: [[PTR_IV_NEXT]] = getelementptr i64, ptr [[PTR_IV]], i64 [[STRIDE:%.*]]514; DEFAULT-NEXT: [[INDVAR_NEXT]] = add i64 [[INDVAR]], 1515; DEFAULT-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVAR]], [[N:%.*]]516; DEFAULT-NEXT: br i1 [[EXITCOND_NOT]], label [[EXIT:%.*]], label [[FOR_BODY]]517; DEFAULT: exit:518; DEFAULT-NEXT: [[PTR_IV_LCSSA:%.*]] = phi ptr [ [[PTR_IV]], [[FOR_BODY]] ]519; DEFAULT-NEXT: [[DOTLCSSA:%.*]] = phi i64 [ [[TMP0]], [[FOR_BODY]] ]520; DEFAULT-NEXT: [[CAST_PTR:%.*]] = ptrtoint ptr [[PTR_IV_LCSSA]] to i64521; DEFAULT-NEXT: [[RESULT:%.*]] = add i64 [[CAST_PTR]], [[DOTLCSSA]]522; DEFAULT-NEXT: ret i64 [[RESULT]]523;524; STRIDED-LABEL: @ivopt_widen_ptr_indvar_2(525; STRIDED-NEXT: entry:526; STRIDED-NEXT: [[TMP0:%.*]] = add i64 [[N:%.*]], 1527; STRIDED-NEXT: [[TMP1:%.*]] = shl i64 [[STRIDE:%.*]], 3528; STRIDED-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 4529; STRIDED-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]530; STRIDED: vector.ph:531; STRIDED-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 4532; STRIDED-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]533; STRIDED-NEXT: [[TMP2:%.*]] = mul i64 [[N_VEC]], [[TMP1]]534; STRIDED-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr null, i64 [[TMP2]]535; STRIDED-NEXT: br label [[VECTOR_BODY:%.*]]536; STRIDED: vector.body:537; STRIDED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]538; STRIDED-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], [[TMP1]]539; STRIDED-NEXT: [[TMP8:%.*]] = mul i64 0, [[TMP1]]540; STRIDED-NEXT: [[TMP9:%.*]] = add i64 [[OFFSET_IDX]], [[TMP8]]541; STRIDED-NEXT: [[TMP10:%.*]] = mul i64 1, [[TMP1]]542; STRIDED-NEXT: [[TMP11:%.*]] = add i64 [[OFFSET_IDX]], [[TMP10]]543; STRIDED-NEXT: [[TMP12:%.*]] = mul i64 2, [[TMP1]]544; STRIDED-NEXT: [[TMP13:%.*]] = add i64 [[OFFSET_IDX]], [[TMP12]]545; STRIDED-NEXT: [[TMP14:%.*]] = mul i64 3, [[TMP1]]546; STRIDED-NEXT: [[TMP15:%.*]] = add i64 [[OFFSET_IDX]], [[TMP14]]547; STRIDED-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr null, i64 [[TMP9]]548; STRIDED-NEXT: [[NEXT_GEP1:%.*]] = getelementptr i8, ptr null, i64 [[TMP11]]549; STRIDED-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr null, i64 [[TMP13]]550; STRIDED-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr null, i64 [[TMP15]]551; STRIDED-NEXT: [[TMP21:%.*]] = getelementptr i64, ptr [[A:%.*]], i64 [[INDEX]]552; STRIDED-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP21]], align 8553; STRIDED-NEXT: [[TMP23:%.*]] = extractelement <4 x i64> [[WIDE_LOAD]], i32 0554; STRIDED-NEXT: [[TMP24:%.*]] = extractelement <4 x i64> [[WIDE_LOAD]], i32 1555; STRIDED-NEXT: [[TMP16:%.*]] = extractelement <4 x i64> [[WIDE_LOAD]], i32 2556; STRIDED-NEXT: [[TMP25:%.*]] = extractelement <4 x i64> [[WIDE_LOAD]], i32 3557; STRIDED-NEXT: store i64 [[TMP23]], ptr [[NEXT_GEP]], align 8558; STRIDED-NEXT: store i64 [[TMP24]], ptr [[NEXT_GEP1]], align 8559; STRIDED-NEXT: store i64 [[TMP16]], ptr [[NEXT_GEP2]], align 8560; STRIDED-NEXT: store i64 [[TMP25]], ptr [[NEXT_GEP3]], align 8561; STRIDED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4562; STRIDED-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]563; STRIDED-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]]564; STRIDED: middle.block:565; STRIDED-NEXT: [[TMP17:%.*]] = extractelement <4 x i64> [[WIDE_LOAD]], i32 3566; STRIDED-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]567; STRIDED-NEXT: [[TMP19:%.*]] = sub i64 0, [[TMP1]]568; STRIDED-NEXT: [[IND_ESCAPE:%.*]] = getelementptr i8, ptr [[TMP3]], i64 [[TMP19]]569; STRIDED-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]570; STRIDED: scalar.ph:571; STRIDED-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]572; STRIDED-NEXT: [[BC_RESUME_VAL4:%.*]] = phi ptr [ [[TMP3]], [[MIDDLE_BLOCK]] ], [ null, [[ENTRY]] ]573; STRIDED-NEXT: br label [[FOR_BODY:%.*]]574; STRIDED: for.body:575; STRIDED-NEXT: [[INDVAR:%.*]] = phi i64 [ [[INDVAR_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]576; STRIDED-NEXT: [[PTR_IV:%.*]] = phi ptr [ [[PTR_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL4]], [[SCALAR_PH]] ]577; STRIDED-NEXT: [[ARRAYIDX:%.*]] = getelementptr i64, ptr [[A]], i64 [[INDVAR]]578; STRIDED-NEXT: [[TMP20:%.*]] = load i64, ptr [[ARRAYIDX]], align 8579; STRIDED-NEXT: store i64 [[TMP20]], ptr [[PTR_IV]], align 8580; STRIDED-NEXT: [[PTR_IV_NEXT]] = getelementptr i64, ptr [[PTR_IV]], i64 [[STRIDE]]581; STRIDED-NEXT: [[INDVAR_NEXT]] = add i64 [[INDVAR]], 1582; STRIDED-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVAR]], [[N]]583; STRIDED-NEXT: br i1 [[EXITCOND_NOT]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]]584; STRIDED: exit:585; STRIDED-NEXT: [[PTR_IV_LCSSA:%.*]] = phi ptr [ [[PTR_IV]], [[FOR_BODY]] ], [ [[IND_ESCAPE]], [[MIDDLE_BLOCK]] ]586; STRIDED-NEXT: [[DOTLCSSA:%.*]] = phi i64 [ [[TMP20]], [[FOR_BODY]] ], [ [[TMP17]], [[MIDDLE_BLOCK]] ]587; STRIDED-NEXT: [[CAST_PTR:%.*]] = ptrtoint ptr [[PTR_IV_LCSSA]] to i64588; STRIDED-NEXT: [[RESULT:%.*]] = add i64 [[CAST_PTR]], [[DOTLCSSA]]589; STRIDED-NEXT: ret i64 [[RESULT]]590;591entry:592 br label %for.body593 594for.body:595 %indvar = phi i64 [ %indvar.next, %for.body ], [ 0, %entry ]596 %ptr.iv = phi ptr [ %ptr.iv.next, %for.body ], [ null, %entry ]597 %arrayidx = getelementptr i64, ptr %a, i64 %indvar598 %0 = load i64, ptr %arrayidx, align 8599 store i64 %0, ptr %ptr.iv, align 8600 %ptr.iv.next = getelementptr i64, ptr %ptr.iv, i64 %stride601 %indvar.next = add i64 %indvar, 1602 %exitcond.not = icmp eq i64 %indvar, %n603 br i1 %exitcond.not, label %exit, label %for.body604 605exit:606 %cast.ptr = ptrtoint ptr %ptr.iv to i64607 %result = add i64 %cast.ptr, %0608 ret i64 %result609}610 611define i64 @ivopt_widen_ptr_indvar_3(ptr noalias %a, i64 %stride, i64 %n) {612; DEFAULT-LABEL: @ivopt_widen_ptr_indvar_3(613; DEFAULT-NEXT: entry:614; DEFAULT-NEXT: br label [[FOR_BODY:%.*]]615; DEFAULT: for.body:616; DEFAULT-NEXT: [[INDVAR:%.*]] = phi i64 [ [[INDVAR_NEXT:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ]617; DEFAULT-NEXT: [[PTR_IV:%.*]] = phi ptr [ [[PTR_IV_NEXT:%.*]], [[FOR_BODY]] ], [ null, [[ENTRY]] ]618; DEFAULT-NEXT: [[ARRAYIDX:%.*]] = getelementptr i64, ptr [[A:%.*]], i64 [[INDVAR]]619; DEFAULT-NEXT: [[TMP0:%.*]] = load i64, ptr [[ARRAYIDX]], align 8620; DEFAULT-NEXT: [[PTR_IV_NEXT]] = getelementptr i64, ptr [[PTR_IV]], i64 [[STRIDE:%.*]]621; DEFAULT-NEXT: [[INDVAR_NEXT]] = add i64 [[INDVAR]], 1622; DEFAULT-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVAR]], [[N:%.*]]623; DEFAULT-NEXT: br i1 [[EXITCOND_NOT]], label [[EXIT:%.*]], label [[FOR_BODY]]624; DEFAULT: exit:625; DEFAULT-NEXT: [[DOTLCSSA:%.*]] = phi i64 [ [[TMP0]], [[FOR_BODY]] ]626; DEFAULT-NEXT: [[PTR_IV_NEXT_LCSSA:%.*]] = phi ptr [ [[PTR_IV_NEXT]], [[FOR_BODY]] ]627; DEFAULT-NEXT: [[CAST_PTR:%.*]] = ptrtoint ptr [[PTR_IV_NEXT_LCSSA]] to i64628; DEFAULT-NEXT: [[RESULT:%.*]] = add i64 [[CAST_PTR]], [[DOTLCSSA]]629; DEFAULT-NEXT: ret i64 [[RESULT]]630;631; STRIDED-LABEL: @ivopt_widen_ptr_indvar_3(632; STRIDED-NEXT: entry:633; STRIDED-NEXT: [[TMP0:%.*]] = add i64 [[N:%.*]], 1634; STRIDED-NEXT: [[TMP1:%.*]] = shl i64 [[STRIDE:%.*]], 3635; STRIDED-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 4636; STRIDED-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]637; STRIDED: vector.ph:638; STRIDED-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 4639; STRIDED-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]640; STRIDED-NEXT: [[TMP2:%.*]] = mul i64 [[N_VEC]], [[TMP1]]641; STRIDED-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr null, i64 [[TMP2]]642; STRIDED-NEXT: br label [[VECTOR_BODY:%.*]]643; STRIDED: vector.body:644; STRIDED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]645; STRIDED-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], [[TMP1]]646; STRIDED-NEXT: [[TMP4:%.*]] = mul i64 0, [[TMP1]]647; STRIDED-NEXT: [[TMP5:%.*]] = add i64 [[OFFSET_IDX]], [[TMP4]]648; STRIDED-NEXT: [[TMP6:%.*]] = mul i64 1, [[TMP1]]649; STRIDED-NEXT: [[TMP7:%.*]] = add i64 [[OFFSET_IDX]], [[TMP6]]650; STRIDED-NEXT: [[TMP8:%.*]] = mul i64 2, [[TMP1]]651; STRIDED-NEXT: [[TMP9:%.*]] = add i64 [[OFFSET_IDX]], [[TMP8]]652; STRIDED-NEXT: [[TMP10:%.*]] = mul i64 3, [[TMP1]]653; STRIDED-NEXT: [[TMP11:%.*]] = add i64 [[OFFSET_IDX]], [[TMP10]]654; STRIDED-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr null, i64 [[TMP11]]655; STRIDED-NEXT: [[TMP12:%.*]] = getelementptr i64, ptr [[A:%.*]], i64 [[INDEX]]656; STRIDED-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP12]], align 8657; STRIDED-NEXT: [[TMP14:%.*]] = getelementptr i64, ptr [[NEXT_GEP3]], i64 [[STRIDE]]658; STRIDED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4659; STRIDED-NEXT: [[TMP15:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]660; STRIDED-NEXT: br i1 [[TMP15]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP14:![0-9]+]]661; STRIDED: middle.block:662; STRIDED-NEXT: [[TMP16:%.*]] = extractelement <4 x i64> [[WIDE_LOAD]], i32 3663; STRIDED-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]664; STRIDED-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]665; STRIDED: scalar.ph:666; STRIDED-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]667; STRIDED-NEXT: [[BC_RESUME_VAL4:%.*]] = phi ptr [ [[TMP3]], [[MIDDLE_BLOCK]] ], [ null, [[ENTRY]] ]668; STRIDED-NEXT: br label [[FOR_BODY:%.*]]669; STRIDED: for.body:670; STRIDED-NEXT: [[INDVAR:%.*]] = phi i64 [ [[INDVAR_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]671; STRIDED-NEXT: [[PTR_IV:%.*]] = phi ptr [ [[PTR_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL4]], [[SCALAR_PH]] ]672; STRIDED-NEXT: [[ARRAYIDX:%.*]] = getelementptr i64, ptr [[A]], i64 [[INDVAR]]673; STRIDED-NEXT: [[TMP17:%.*]] = load i64, ptr [[ARRAYIDX]], align 8674; STRIDED-NEXT: [[PTR_IV_NEXT]] = getelementptr i64, ptr [[PTR_IV]], i64 [[STRIDE]]675; STRIDED-NEXT: [[INDVAR_NEXT]] = add i64 [[INDVAR]], 1676; STRIDED-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVAR]], [[N]]677; STRIDED-NEXT: br i1 [[EXITCOND_NOT]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]]678; STRIDED: exit:679; STRIDED-NEXT: [[DOTLCSSA:%.*]] = phi i64 [ [[TMP17]], [[FOR_BODY]] ], [ [[TMP16]], [[MIDDLE_BLOCK]] ]680; STRIDED-NEXT: [[PTR_IV_NEXT_LCSSA:%.*]] = phi ptr [ [[PTR_IV_NEXT]], [[FOR_BODY]] ], [ [[TMP14]], [[MIDDLE_BLOCK]] ]681; STRIDED-NEXT: [[CAST_PTR:%.*]] = ptrtoint ptr [[PTR_IV_NEXT_LCSSA]] to i64682; STRIDED-NEXT: [[RESULT:%.*]] = add i64 [[CAST_PTR]], [[DOTLCSSA]]683; STRIDED-NEXT: ret i64 [[RESULT]]684;685entry:686 br label %for.body687 688for.body:689 %indvar = phi i64 [ %indvar.next, %for.body ], [ 0, %entry ]690 %ptr.iv = phi ptr [ %ptr.iv.next, %for.body ], [ null, %entry ]691 %arrayidx = getelementptr i64, ptr %a, i64 %indvar692 %0 = load i64, ptr %arrayidx, align 8693 %ptr.iv.next = getelementptr i64, ptr %ptr.iv, i64 %stride694 %indvar.next = add i64 %indvar, 1695 %exitcond.not = icmp eq i64 %indvar, %n696 br i1 %exitcond.not, label %exit, label %for.body697 698exit:699 %cast.ptr = ptrtoint ptr %ptr.iv.next to i64700 %result = add i64 %cast.ptr, %0701 ret i64 %result702}703