175 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -passes=loop-vectorize,instcombine -force-vector-width=4 -S < %s 2>&1 | FileCheck %s3 4target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"5 6;;;; Derived from the following C code7;; void forked_ptrs_different_base_same_offset(ptr A, ptr B, ptr C, int *D) {8;; for (int i=0; i<100; i++) {9;; if (D[i] != 0) {10;; C[i] = A[i];11;; } else {12;; C[i] = B[i];13;; }14;; }15;; }16 17define dso_local void @forked_ptrs_different_base_same_offset(ptr nocapture readonly %Base1, ptr nocapture readonly %Base2, ptr nocapture %Dest, ptr nocapture readonly %Preds) {18; CHECK-LABEL: @forked_ptrs_different_base_same_offset(19; CHECK-NEXT: entry:20; CHECK-NEXT: [[BASE1:%.*]] = freeze ptr [[BASE3:%.*]]21; CHECK-NEXT: [[BASE2:%.*]] = freeze ptr [[BASE4:%.*]]22; CHECK-NEXT: [[DEST:%.*]] = freeze ptr [[DEST2:%.*]]23; CHECK-NEXT: br label [[VECTOR_MEMCHECK:%.*]]24; CHECK: vector.memcheck:25; CHECK-NEXT: [[DEST1:%.*]] = ptrtoint ptr [[DEST]] to i6426; CHECK-NEXT: [[PREDS2:%.*]] = ptrtoint ptr [[PREDS:%.*]] to i6427; CHECK-NEXT: [[BASE23:%.*]] = ptrtoint ptr [[BASE2]] to i6428; CHECK-NEXT: [[BASE15:%.*]] = ptrtoint ptr [[BASE1]] to i6429; CHECK-NEXT: [[TMP0:%.*]] = sub i64 [[DEST1]], [[PREDS2]]30; CHECK-NEXT: [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP0]], 1631; CHECK-NEXT: [[TMP1:%.*]] = sub i64 [[DEST1]], [[BASE23]]32; CHECK-NEXT: [[DIFF_CHECK4:%.*]] = icmp ult i64 [[TMP1]], 1633; CHECK-NEXT: [[CONFLICT_RDX:%.*]] = or i1 [[DIFF_CHECK]], [[DIFF_CHECK4]]34; CHECK-NEXT: [[TMP2:%.*]] = sub i64 [[DEST1]], [[BASE15]]35; CHECK-NEXT: [[DIFF_CHECK6:%.*]] = icmp ult i64 [[TMP2]], 1636; CHECK-NEXT: [[CONFLICT_RDX7:%.*]] = or i1 [[CONFLICT_RDX]], [[DIFF_CHECK6]]37; CHECK-NEXT: br i1 [[CONFLICT_RDX7]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]38; CHECK: vector.ph:39; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x ptr> poison, ptr [[BASE2]], i64 040; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x ptr> [[BROADCAST_SPLATINSERT]], <4 x ptr> poison, <4 x i32> zeroinitializer41; CHECK-NEXT: [[BROADCAST_SPLATINSERT8:%.*]] = insertelement <4 x ptr> poison, ptr [[BASE1]], i64 042; CHECK-NEXT: [[BROADCAST_SPLAT9:%.*]] = shufflevector <4 x ptr> [[BROADCAST_SPLATINSERT8]], <4 x ptr> poison, <4 x i32> zeroinitializer43; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]44; CHECK: vector.body:45; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]46; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[PREDS]], i64 [[INDEX]]47; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP6]], align 448; CHECK-NEXT: [[TMP7:%.*]] = icmp eq <4 x i32> [[WIDE_LOAD]], zeroinitializer49; CHECK-NEXT: [[TMP8:%.*]] = select <4 x i1> [[TMP7]], <4 x ptr> [[BROADCAST_SPLAT]], <4 x ptr> [[BROADCAST_SPLAT9]]50; CHECK-NEXT: [[TMP9:%.*]] = extractelement <4 x ptr> [[TMP8]], i64 051; CHECK-NEXT: [[TMP11:%.*]] = extractelement <4 x ptr> [[TMP8]], i64 152; CHECK-NEXT: [[TMP13:%.*]] = extractelement <4 x ptr> [[TMP8]], i64 253; CHECK-NEXT: [[TMP15:%.*]] = extractelement <4 x ptr> [[TMP8]], i64 354; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds float, ptr [[TMP9]], i64 [[INDEX]]55; CHECK-NEXT: [[TMP29:%.*]] = getelementptr float, ptr [[TMP11]], i64 [[INDEX]]56; CHECK-NEXT: [[TMP12:%.*]] = getelementptr i8, ptr [[TMP29]], i64 457; CHECK-NEXT: [[TMP30:%.*]] = getelementptr float, ptr [[TMP13]], i64 [[INDEX]]58; CHECK-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[TMP30]], i64 859; CHECK-NEXT: [[TMP31:%.*]] = getelementptr float, ptr [[TMP15]], i64 [[INDEX]]60; CHECK-NEXT: [[TMP16:%.*]] = getelementptr i8, ptr [[TMP31]], i64 1261; CHECK-NEXT: [[TMP17:%.*]] = load float, ptr [[TMP10]], align 462; CHECK-NEXT: [[TMP18:%.*]] = load float, ptr [[TMP12]], align 463; CHECK-NEXT: [[TMP19:%.*]] = load float, ptr [[TMP14]], align 464; CHECK-NEXT: [[TMP20:%.*]] = load float, ptr [[TMP16]], align 465; CHECK-NEXT: [[TMP21:%.*]] = insertelement <4 x float> poison, float [[TMP17]], i64 066; CHECK-NEXT: [[TMP22:%.*]] = insertelement <4 x float> [[TMP21]], float [[TMP18]], i64 167; CHECK-NEXT: [[TMP23:%.*]] = insertelement <4 x float> [[TMP22]], float [[TMP19]], i64 268; CHECK-NEXT: [[TMP24:%.*]] = insertelement <4 x float> [[TMP23]], float [[TMP20]], i64 369; CHECK-NEXT: [[TMP25:%.*]] = getelementptr inbounds float, ptr [[DEST]], i64 [[INDEX]]70; CHECK-NEXT: store <4 x float> [[TMP24]], ptr [[TMP25]], align 471; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 472; CHECK-NEXT: [[TMP26:%.*]] = icmp eq i64 [[INDEX_NEXT]], 10073; CHECK-NEXT: br i1 [[TMP26]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]74; CHECK: middle.block:75; CHECK-NEXT: br label [[FOR_COND_CLEANUP:%.*]]76; CHECK: scalar.ph:77; CHECK-NEXT: br label [[FOR_BODY:%.*]]78; CHECK: for.cond.cleanup:79; CHECK-NEXT: ret void80; CHECK: for.body:81; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]82; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds nuw i32, ptr [[PREDS]], i64 [[INDVARS_IV]]83; CHECK-NEXT: [[TMP27:%.*]] = load i32, ptr [[ARRAYIDX]], align 484; CHECK-NEXT: [[CMP1_NOT:%.*]] = icmp eq i32 [[TMP27]], 085; CHECK-NEXT: [[SPEC_SELECT:%.*]] = select i1 [[CMP1_NOT]], ptr [[BASE2]], ptr [[BASE1]]86; CHECK-NEXT: [[DOTSINK_IN:%.*]] = getelementptr inbounds nuw float, ptr [[SPEC_SELECT]], i64 [[INDVARS_IV]]87; CHECK-NEXT: [[DOTSINK:%.*]] = load float, ptr [[DOTSINK_IN]], align 488; CHECK-NEXT: [[TMP28:%.*]] = getelementptr inbounds nuw float, ptr [[DEST]], i64 [[INDVARS_IV]]89; CHECK-NEXT: store float [[DOTSINK]], ptr [[TMP28]], align 490; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 191; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], 10092; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]93;94entry:95 br label %for.body96 97for.cond.cleanup:98 ret void99 100for.body:101 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]102 %arrayidx = getelementptr inbounds i32, ptr %Preds, i64 %indvars.iv103 %0 = load i32, ptr %arrayidx, align 4104 %cmp1.not = icmp eq i32 %0, 0105 %spec.select = select i1 %cmp1.not, ptr %Base2, ptr %Base1106 %.sink.in = getelementptr inbounds float, ptr %spec.select, i64 %indvars.iv107 %.sink = load float, ptr %.sink.in, align 4108 %1 = getelementptr inbounds float, ptr %Dest, i64 %indvars.iv109 store float %.sink, ptr %1, align 4110 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1111 %exitcond.not = icmp eq i64 %indvars.iv.next, 100112 br i1 %exitcond.not, label %for.cond.cleanup, label %for.body113}114 115;;;; Derived from the following C code116;; void forked_ptrs_same_base_different_offset(ptr A, ptr B, int *C) {117;; int offset;118;; for (int i = 0; i < 100; i++) {119;; if (C[i] != 0)120;; offset = i;121;; else122;; offset = i+1;123;; B[i] = A[offset];124;; }125;; }126 127define dso_local void @forked_ptrs_same_base_different_offset(ptr nocapture readonly %Base, ptr nocapture %Dest, ptr nocapture readonly %Preds) {128; CHECK-LABEL: @forked_ptrs_same_base_different_offset(129; CHECK-NEXT: entry:130; CHECK-NEXT: br label [[FOR_BODY:%.*]]131; CHECK: for.cond.cleanup:132; CHECK-NEXT: ret void133; CHECK: for.body:134; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]135; CHECK-NEXT: [[I_014:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[ADD:%.*]], [[FOR_BODY]] ]136; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds nuw i32, ptr [[PREDS:%.*]], i64 [[INDVARS_IV]]137; CHECK-NEXT: [[TMP0:%.*]] = load i32, ptr [[ARRAYIDX]], align 4138; CHECK-NEXT: [[CMP1_NOT:%.*]] = icmp eq i32 [[TMP0]], 0139; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1140; CHECK-NEXT: [[ADD]] = add nuw nsw i32 [[I_014]], 1141; CHECK-NEXT: [[TMP1:%.*]] = trunc i64 [[INDVARS_IV]] to i32142; CHECK-NEXT: [[OFFSET_0:%.*]] = select i1 [[CMP1_NOT]], i32 [[ADD]], i32 [[TMP1]]143; CHECK-NEXT: [[IDXPROM213:%.*]] = zext i32 [[OFFSET_0]] to i64144; CHECK-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds nuw float, ptr [[BASE:%.*]], i64 [[IDXPROM213]]145; CHECK-NEXT: [[TMP2:%.*]] = load float, ptr [[ARRAYIDX3]], align 4146; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds nuw float, ptr [[DEST:%.*]], i64 [[INDVARS_IV]]147; CHECK-NEXT: store float [[TMP2]], ptr [[ARRAYIDX5]], align 4148; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], 100149; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP:%.*]], label [[FOR_BODY]]150;151entry:152 br label %for.body153 154for.cond.cleanup: ; preds = %for.body155 ret void156 157for.body: ; preds = %entry, %for.body158 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]159 %i.014 = phi i32 [ 0, %entry ], [ %add, %for.body ]160 %arrayidx = getelementptr inbounds i32, ptr %Preds, i64 %indvars.iv161 %0 = load i32, ptr %arrayidx, align 4162 %cmp1.not = icmp eq i32 %0, 0163 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1164 %add = add nuw nsw i32 %i.014, 1165 %1 = trunc i64 %indvars.iv to i32166 %offset.0 = select i1 %cmp1.not, i32 %add, i32 %1167 %idxprom213 = zext i32 %offset.0 to i64168 %arrayidx3 = getelementptr inbounds float, ptr %Base, i64 %idxprom213169 %2 = load float, ptr %arrayidx3, align 4170 %arrayidx5 = getelementptr inbounds float, ptr %Dest, i64 %indvars.iv171 store float %2, ptr %arrayidx5, align 4172 %exitcond.not = icmp eq i64 %indvars.iv.next, 100173 br i1 %exitcond.not, label %for.cond.cleanup, label %for.body174}175