1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -S -o - -passes=gvn %s | FileCheck %s 3 4target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128" 5 6%struct.S1 = type { i32, i32 } 7 8@a2 = common global ptr null, align 8 9@a = common global ptr null, align 8 10@s1 = common global %struct.S1 zeroinitializer, align 8 11 12; Check that GVN doesn't determine %2 is partially redundant. 13 14define i32 @volatile_load(i32 %n) { 15; CHECK-LABEL: @volatile_load( 16; CHECK-NEXT: entry: 17; CHECK-NEXT: [[CMP6:%.*]] = icmp sgt i32 [[N:%.*]], 0 18; CHECK-NEXT: br i1 [[CMP6]], label [[FOR_BODY_LR_PH:%.*]], label [[FOR_END:%.*]] 19; CHECK: for.body.lr.ph: 20; CHECK-NEXT: [[TMP0:%.*]] = load ptr, ptr @a2, align 8, !tbaa [[TBAA5:![0-9]+]] 21; CHECK-NEXT: [[TMP1:%.*]] = load ptr, ptr @a, align 8, !tbaa [[TBAA5]] 22; CHECK-NEXT: br label [[FOR_BODY:%.*]] 23; CHECK: for.body: 24; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[FOR_BODY_LR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ] 25; CHECK-NEXT: [[S_09:%.*]] = phi i32 [ 0, [[FOR_BODY_LR_PH]] ], [ [[ADD:%.*]], [[FOR_BODY]] ] 26; CHECK-NEXT: [[P_08:%.*]] = phi ptr [ [[TMP0]], [[FOR_BODY_LR_PH]] ], [ [[INCDEC_PTR:%.*]], [[FOR_BODY]] ] 27; CHECK-NEXT: [[TMP2:%.*]] = load i32, ptr [[P_08]], align 4, !tbaa [[TBAA9:![0-9]+]] 28; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i64 [[INDVARS_IV]] 29; CHECK-NEXT: store i32 [[TMP2]], ptr [[ARRAYIDX]], align 4, !tbaa [[TBAA9]] 30; CHECK-NEXT: [[TMP3:%.*]] = load volatile i32, ptr [[P_08]], align 4, !tbaa [[TBAA9]] 31; CHECK-NEXT: [[ADD]] = add nsw i32 [[TMP3]], [[S_09]] 32; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1 33; CHECK-NEXT: [[INCDEC_PTR]] = getelementptr inbounds i32, ptr [[P_08]], i64 1 34; CHECK-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32 35; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i32 [[LFTR_WIDEIV]], [[N]] 36; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_BODY]], label [[FOR_COND_FOR_END_CRIT_EDGE:%.*]] 37; CHECK: for.cond.for.end_crit_edge: 38; CHECK-NEXT: br label [[FOR_END]] 39; CHECK: for.end: 40; CHECK-NEXT: [[S_0_LCSSA:%.*]] = phi i32 [ [[ADD]], [[FOR_COND_FOR_END_CRIT_EDGE]] ], [ 0, [[ENTRY:%.*]] ] 41; CHECK-NEXT: ret i32 [[S_0_LCSSA]] 42; 43entry: 44 %cmp6 = icmp sgt i32 %n, 0 45 br i1 %cmp6, label %for.body.lr.ph, label %for.end 46 47for.body.lr.ph: 48 %0 = load ptr, ptr @a2, align 8, !tbaa !1 49 %1 = load ptr, ptr @a, align 8, !tbaa !1 50 br label %for.body 51 52for.body: 53 %indvars.iv = phi i64 [ 0, %for.body.lr.ph ], [ %indvars.iv.next, %for.body ] 54 %s.09 = phi i32 [ 0, %for.body.lr.ph ], [ %add, %for.body ] 55 %p.08 = phi ptr [ %0, %for.body.lr.ph ], [ %incdec.ptr, %for.body ] 56 %2 = load i32, ptr %p.08, align 4, !tbaa !5 57 %arrayidx = getelementptr inbounds i32, ptr %1, i64 %indvars.iv 58 store i32 %2, ptr %arrayidx, align 4, !tbaa !5 59 %3 = load volatile i32, ptr %p.08, align 4, !tbaa !5 60 %add = add nsw i32 %3, %s.09 61 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 62 %incdec.ptr = getelementptr inbounds i32, ptr %p.08, i64 1 63 %lftr.wideiv = trunc i64 %indvars.iv.next to i32 64 %exitcond = icmp ne i32 %lftr.wideiv, %n 65 br i1 %exitcond, label %for.body, label %for.cond.for.end_crit_edge 66 67for.cond.for.end_crit_edge: 68 %add.lcssa = phi i32 [ %add, %for.body ] 69 br label %for.end 70 71for.end: 72 %s.0.lcssa = phi i32 [ %add.lcssa, %for.cond.for.end_crit_edge ], [ 0, %entry ] 73 ret i32 %s.0.lcssa 74} 75 76; %1 is partially redundant if %0 can be widened to a 64-bit load. 77; But we should not widen %0 to 64-bit load. 78 79define i32 @overaligned_load(i32 %a, ptr nocapture %b) !dbg !13 { 80; CHECK-LABEL: @overaligned_load( 81; CHECK-NEXT: entry: 82; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i32 [[A:%.*]], 0, !dbg [[DBG14:![0-9]+]] 83; CHECK-NEXT: br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_ELSE:%.*]], !dbg [[DBG14]] 84; CHECK: if.then: 85; CHECK-NEXT: [[TMP0:%.*]] = load i32, ptr @s1, align 8, !dbg [[DBG15:![0-9]+]], !tbaa [[TBAA9]] 86; CHECK-NEXT: br label [[IF_END:%.*]], !dbg [[DBG15]] 87; CHECK: if.else: 88; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[B:%.*]], i64 2, !dbg [[DBG16:![0-9]+]] 89; CHECK-NEXT: store i32 10, ptr [[ARRAYIDX]], align 4, !dbg [[DBG16]], !tbaa [[TBAA9]] 90; CHECK-NEXT: br label [[IF_END]], !dbg [[DBG16]] 91; CHECK: if.end: 92; CHECK-NEXT: [[I_0:%.*]] = phi i32 [ [[TMP0]], [[IF_THEN]] ], [ 0, [[IF_ELSE]] ] 93; CHECK-NEXT: [[P_0:%.*]] = phi ptr [ @s1, [[IF_THEN]] ], [ [[B]], [[IF_ELSE]] ] 94; CHECK-NEXT: [[ADD_PTR:%.*]] = getelementptr inbounds i32, ptr [[P_0]], i64 1, !dbg [[DBG17:![0-9]+]] 95; CHECK-NEXT: [[TMP1:%.*]] = load i32, ptr [[ADD_PTR]], align 4, !dbg [[DBG17]], !tbaa [[TBAA9]] 96; CHECK-NEXT: [[ADD1:%.*]] = add nsw i32 [[TMP1]], [[I_0]], !dbg [[DBG17]] 97; CHECK-NEXT: ret i32 [[ADD1]], !dbg [[DBG17]] 98; 99entry: 100 %cmp = icmp sgt i32 %a, 0, !dbg !14 101 br i1 %cmp, label %if.then, label %if.else, !dbg !14 102 103if.then: 104 %0 = load i32, ptr @s1, align 8, !tbaa !5, !dbg !15 105 br label %if.end, !dbg !15 106 107if.else: 108 %arrayidx = getelementptr inbounds i32, ptr %b, i64 2, !dbg !16 109 store i32 10, ptr %arrayidx, align 4, !tbaa !5, !dbg !16 110 br label %if.end, !dbg !16 111 112if.end: 113 %i.0 = phi i32 [ %0, %if.then ], [ 0, %if.else ] 114 %p.0 = phi ptr [ @s1, %if.then ], [ %b, %if.else ] 115 %add.ptr = getelementptr inbounds i32, ptr %p.0, i64 1, !dbg !17 116 %1 = load i32, ptr %add.ptr, align 4, !tbaa !5, !dbg !17 117 %add1 = add nsw i32 %1, %i.0, !dbg !17 118 ret i32 %add1, !dbg !17 119} 120 121!1 = !{!2, !2, i64 0} 122!2 = !{!"any pointer", !3, i64 0} 123!3 = !{!"omnipotent char", !4, i64 0} 124!4 = !{!"Simple C/C++ TBAA"} 125!5 = !{!6, !6, i64 0} 126!6 = !{!"int", !3, i64 0} 127 128!llvm.module.flags = !{!7, !8, !9} 129!llvm.dbg.cu = !{!18} 130!7 = !{i32 2, !"Dwarf Version", i32 4} 131!8 = !{i32 2, !"Debug Info Version", i32 3} 132!9 = !{i32 1, !"PIC Level", i32 2} 133 134!10 = !{} 135!11 = !DISubroutineType(types: !10) 136!12 = !DIFile(filename: "test.cpp", directory: "/tmp") 137!13 = distinct !DISubprogram(name: "test", scope: !12, file: !12, line: 99, type: !11, isLocal: false, isDefinition: true, scopeLine: 100, flags: DIFlagPrototyped, isOptimized: false, unit: !18, retainedNodes: !10) 138!14 = !DILocation(line: 100, column: 1, scope: !13) 139!15 = !DILocation(line: 101, column: 1, scope: !13) 140!16 = !DILocation(line: 102, column: 1, scope: !13) 141!17 = !DILocation(line: 103, column: 1, scope: !13) 142!18 = distinct !DICompileUnit(language: DW_LANG_C99, producer: "clang", 143 file: !12, 144 isOptimized: true, flags: "-O2", 145 splitDebugFilename: "abc.debug", emissionKind: 2) 146