1; RUN: opt -O2 -S < %s | FileCheck %s 2; 3; This test tries to ensure that simplifycfg hoisting common instructions 4; of then/else branch indeed happens. BPF target has added an IR pass 5; before loop optimizations as Commit 1d51dc38d89b 6; ([SimplifyCFG][LoopRotate] SimplifyCFG: disable common instruction 7; hoisting by default, enable late in pipeline) 8; disabled common instruction hoisting. Due to optimization triggered 9; code changes, later SimplifyCFG may not be able to perform optimization 10; even common inst hoisting is enabled. 11; 12; Source: 13; typedef struct { 14; void *f_back; 15; } FrameData; 16; extern int get_data(void *, void *); 17; extern void get_frame_ptr(void *); 18; int test() { 19; void *frame_ptr; 20; FrameData frame; 21; 22; get_frame_ptr(&frame_ptr); 23; 24; #pragma nounroll 25; for (int i = 0; i < 6; i++) { 26; if (frame_ptr && get_data(frame_ptr, &frame)) { 27; frame_ptr = frame.f_back; 28; } 29; } 30; return frame_ptr == 0; 31; } 32; Compilation flag: 33; clang -target bpf -O2 -Xclang -disable-llvm-passes -S -emit-llvm t.c -o t.ll 34 35target datalayout = "e-m:e-p:64:64-i64:64-i128:128-n32:64-S128" 36target triple = "bpf" 37 38%struct.FrameData = type { i8* } 39 40; Function Attrs: nounwind 41define dso_local i32 @test() #0 { 42entry: 43 %frame_ptr = alloca i8*, align 8 44 %frame = alloca %struct.FrameData, align 8 45 %i = alloca i32, align 4 46 %0 = bitcast i8** %frame_ptr to i8* 47 call void @llvm.lifetime.start.p0i8(i64 8, i8* %0) #3 48 %1 = bitcast %struct.FrameData* %frame to i8* 49 call void @llvm.lifetime.start.p0i8(i64 8, i8* %1) #3 50 %2 = bitcast i8** %frame_ptr to i8* 51 call void @get_frame_ptr(i8* %2) 52 %3 = bitcast i32* %i to i8* 53 call void @llvm.lifetime.start.p0i8(i64 4, i8* %3) #3 54 store i32 0, i32* %i, align 4, !tbaa !2 55 br label %for.cond 56 57; CHECK-LABEL: entry 58; CHECK: %{{[0-9]+}} = load i8*, i8** %frame_ptr, align 8 59; CHECK: %{{[0-9a-z.]+}} = icmp eq i8* %2, null 60; CHECK: br label 61 62for.cond: ; preds = %for.inc, %entry 63 %4 = load i32, i32* %i, align 4, !tbaa !2 64 %cmp = icmp slt i32 %4, 6 65 br i1 %cmp, label %for.body, label %for.cond.cleanup 66 67for.cond.cleanup: ; preds = %for.cond 68 %5 = bitcast i32* %i to i8* 69 call void @llvm.lifetime.end.p0i8(i64 4, i8* %5) #3 70 br label %for.end 71 72for.body: ; preds = %for.cond 73 %6 = load i8*, i8** %frame_ptr, align 8, !tbaa !6 74 %tobool = icmp ne i8* %6, null 75 br i1 %tobool, label %land.lhs.true, label %if.end 76 77land.lhs.true: ; preds = %for.body 78 %7 = load i8*, i8** %frame_ptr, align 8, !tbaa !6 79 %8 = bitcast %struct.FrameData* %frame to i8* 80 %call = call i32 @get_data(i8* %7, i8* %8) 81 %tobool1 = icmp ne i32 %call, 0 82 br i1 %tobool1, label %if.then, label %if.end 83 84if.then: ; preds = %land.lhs.true 85 %f_back = getelementptr inbounds %struct.FrameData, %struct.FrameData* %frame, i32 0, i32 0 86 %9 = load i8*, i8** %f_back, align 8, !tbaa !8 87 store i8* %9, i8** %frame_ptr, align 8, !tbaa !6 88 br label %if.end 89 90if.end: ; preds = %if.then, %land.lhs.true, %for.body 91 br label %for.inc 92 93for.inc: ; preds = %if.end 94 %10 = load i32, i32* %i, align 4, !tbaa !2 95 %inc = add nsw i32 %10, 1 96 store i32 %inc, i32* %i, align 4, !tbaa !2 97 br label %for.cond, !llvm.loop !10 98 99for.end: ; preds = %for.cond.cleanup 100 %11 = load i8*, i8** %frame_ptr, align 8, !tbaa !6 101 %cmp2 = icmp eq i8* %11, null 102 %conv = zext i1 %cmp2 to i32 103 %12 = bitcast %struct.FrameData* %frame to i8* 104 call void @llvm.lifetime.end.p0i8(i64 8, i8* %12) #3 105 %13 = bitcast i8** %frame_ptr to i8* 106 call void @llvm.lifetime.end.p0i8(i64 8, i8* %13) #3 107 ret i32 %conv 108} 109 110; Function Attrs: argmemonly nounwind willreturn 111declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #1 112 113declare dso_local void @get_frame_ptr(i8*) #2 114 115declare dso_local i32 @get_data(i8*, i8*) #2 116 117; Function Attrs: argmemonly nounwind willreturn 118declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #1 119 120attributes #0 = { nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } 121attributes #1 = { argmemonly nounwind willreturn } 122attributes #2 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } 123attributes #3 = { nounwind } 124 125!llvm.module.flags = !{!0} 126!llvm.ident = !{!1} 127 128!0 = !{i32 1, !"wchar_size", i32 4} 129!1 = !{!"clang version 12.0.0 (https://github.com/llvm/llvm-project.git 1b3c1c543269da36ae41ab84f646cf98d2e5b1e5)"} 130!2 = !{!3, !3, i64 0} 131!3 = !{!"int", !4, i64 0} 132!4 = !{!"omnipotent char", !5, i64 0} 133!5 = !{!"Simple C/C++ TBAA"} 134!6 = !{!7, !7, i64 0} 135!7 = !{!"any pointer", !4, i64 0} 136!8 = !{!9, !7, i64 0} 137!9 = !{!"", !7, i64 0} 138!10 = distinct !{!10, !11} 139!11 = !{!"llvm.loop.unroll.disable"} 140