1; RUN: llc < %s -x86-discriminate-memops -prefetch-hints-file=%S/insert-prefetch-inline.afdo | FileCheck %s 2; 3; Verify we can insert prefetch instructions in code belonging to inlined 4; functions. 5; 6; ModuleID = 'test.cc' 7 8target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128" 9target triple = "x86_64-unknown-linux-gnu" 10 11; Function Attrs: norecurse nounwind readonly uwtable 12define dso_local i32 @sum(i32* nocapture readonly %arr, i32 %pos1, i32 %pos2) local_unnamed_addr #0 !dbg !7 { 13entry: 14 %idxprom = sext i32 %pos1 to i64, !dbg !10 15 %arrayidx = getelementptr inbounds i32, i32* %arr, i64 %idxprom, !dbg !10 16 %0 = load i32, i32* %arrayidx, align 4, !dbg !10, !tbaa !11 17 %idxprom1 = sext i32 %pos2 to i64, !dbg !15 18 %arrayidx2 = getelementptr inbounds i32, i32* %arr, i64 %idxprom1, !dbg !15 19 %1 = load i32, i32* %arrayidx2, align 4, !dbg !15, !tbaa !11 20 %add = add nsw i32 %1, %0, !dbg !16 21 ret i32 %add, !dbg !17 22} 23 24; "caller" inlines "sum". The associated .afdo file references instructions 25; in "caller" that came from "sum"'s inlining. 26; 27; Function Attrs: norecurse nounwind readonly uwtable 28define dso_local i32 @caller(i32* nocapture readonly %arr) local_unnamed_addr #0 !dbg !18 { 29entry: 30 %0 = load i32, i32* %arr, align 4, !dbg !19, !tbaa !11 31 %arrayidx2.i = getelementptr inbounds i32, i32* %arr, i64 2, !dbg !21 32 %1 = load i32, i32* %arrayidx2.i, align 4, !dbg !21, !tbaa !11 33 %add.i = add nsw i32 %1, %0, !dbg !22 34 ret i32 %add.i, !dbg !23 35} 36 37attributes #0 = { "target-cpu"="x86-64" } 38 39!llvm.dbg.cu = !{!0} 40!llvm.module.flags = !{!3, !4, !5} 41!llvm.ident = !{!6} 42 43!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus, file: !1, producer: "clang version 7.0.0 (trunk 324940) (llvm/trunk 324941)", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly, enums: !2, debugInfoForProfiling: true) 44!1 = !DIFile(filename: "test.cc", directory: "/tmp") 45!2 = !{} 46!3 = !{i32 2, !"Dwarf Version", i32 4} 47!4 = !{i32 2, !"Debug Info Version", i32 3} 48!5 = !{i32 1, !"wchar_size", i32 4} 49!6 = !{!"clang version 7.0.0 (trunk 324940) (llvm/trunk 324941)"} 50!7 = distinct !DISubprogram(name: "sum", linkageName: "sum", scope: !8, file: !8, line: 3, type: !9, isLocal: false, isDefinition: true, scopeLine: 3, flags: DIFlagPrototyped, isOptimized: true, unit: !0) 51!8 = !DIFile(filename: "./test.h", directory: "/tmp") 52!9 = !DISubroutineType(types: !2) 53!10 = !DILocation(line: 6, column: 10, scope: !7) 54!11 = !{!12, !12, i64 0} 55!12 = !{!"int", !13, i64 0} 56!13 = !{!"omnipotent char", !14, i64 0} 57!14 = !{!"Simple C++ TBAA"} 58!15 = !DILocation(line: 6, column: 22, scope: !7) 59!16 = !DILocation(line: 6, column: 20, scope: !7) 60!17 = !DILocation(line: 6, column: 3, scope: !7) 61!18 = distinct !DISubprogram(name: "caller", linkageName: "caller", scope: !1, file: !1, line: 4, type: !9, isLocal: false, isDefinition: true, scopeLine: 4, flags: DIFlagPrototyped, isOptimized: true, unit: !0) 62!19 = !DILocation(line: 6, column: 10, scope: !7, inlinedAt: !20) 63!20 = distinct !DILocation(line: 6, column: 10, scope: !18) 64!21 = !DILocation(line: 6, column: 22, scope: !7, inlinedAt: !20) 65!22 = !DILocation(line: 6, column: 20, scope: !7, inlinedAt: !20) 66!23 = !DILocation(line: 6, column: 3, scope: !18) 67 68; CHECK-LABEL: caller: 69; CHECK-LABEL: # %bb.0: 70; CHECK-NEXT: .loc 1 6 22 prologue_end 71; CHECK-NEXT: prefetchnta 23464(%rdi) 72; CHECK-NEXT: movl 8(%rdi), %eax 73; CHECK-NEXT: .loc 1 6 20 is_stmt 0 discriminator 2 74; CHECK-NEXT: prefetchnta 8764(%rdi) 75; CHECK-NEXT: prefetchnta 64(%rdi) 76; CHECK-NEXT: addl (%rdi), %eax 77