212 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -passes=dse -enable-dse-partial-store-merging -S < %s | FileCheck %s3target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-f128:128:128-n8:16:32:64"4 5define void @byte_by_byte_replacement(ptr %ptr) {6; CHECK-LABEL: @byte_by_byte_replacement(7; CHECK-NEXT: entry:8; CHECK-NEXT: store i32 202050057, ptr [[PTR:%.*]]9; CHECK-NEXT: ret void10;11entry:12 ;; This store's value should be modified as it should be better to use one13 ;; larger store than several smaller ones.14 ;; store will turn into 0x0C0B0A09 == 20205005715 store i32 305419896, ptr %ptr ; 0x1234567816 %bptr1 = getelementptr inbounds i8, ptr %ptr, i64 117 %bptr2 = getelementptr inbounds i8, ptr %ptr, i64 218 %bptr3 = getelementptr inbounds i8, ptr %ptr, i64 319 20 ;; We should be able to merge these four stores with the i32 above21 ; value (and bytes) stored before ; 0x1234567822 store i8 9, ptr %ptr ; 0923 store i8 10, ptr %bptr1 ; 0A24 store i8 11, ptr %bptr2 ; 0B25 store i8 12, ptr %bptr3 ; 0C26 ; 0x0C0B0A0927 ret void28}29 30define void @word_replacement(ptr %ptr) {31; CHECK-LABEL: @word_replacement(32; CHECK-NEXT: entry:33; CHECK-NEXT: store i64 8106482645252179720, ptr [[PTR:%.*]]34; CHECK-NEXT: ret void35;36entry:37 store i64 72623859790382856, ptr %ptr ; 0x010203040506070838 39 %wptr1 = getelementptr inbounds i16, ptr %ptr, i64 140 %wptr3 = getelementptr inbounds i16, ptr %ptr, i64 341 42 ;; We should be able to merge these two stores with the i64 one above43 ; value (not bytes) stored before ; 0x010203040506070844 store i16 4128, ptr %wptr1 ; 102045 store i16 28800, ptr %wptr3 ; 708046 ; 0x708003041020070847 ret void48}49 50 51define void @differently_sized_replacements(ptr %ptr) {52; CHECK-LABEL: @differently_sized_replacements(53; CHECK-NEXT: entry:54; CHECK-NEXT: store i64 578437695752307201, ptr [[PTR:%.*]]55; CHECK-NEXT: ret void56;57entry:58 store i64 579005069656919567, ptr %ptr ; 0x08090a0b0c0d0e0f59 60 %bptr6 = getelementptr inbounds i8, ptr %ptr, i64 661 %wptr2 = getelementptr inbounds i16, ptr %ptr, i64 262 63 ;; We should be able to merge all these stores with the i64 one above64 ; value (not bytes) stored before ; 0x08090a0b0c0d0e0f65 store i8 7, ptr %bptr6 ; 0766 store i16 1541, ptr %wptr2 ; 060567 store i32 67305985, ptr %ptr ; 0403020168 ; 0x080706050403020169 ret void70}71 72 73define void @multiple_replacements_to_same_byte(ptr %ptr) {74; CHECK-LABEL: @multiple_replacements_to_same_byte(75; CHECK-NEXT: entry:76; CHECK-NEXT: store i64 579005069522043393, ptr [[PTR:%.*]]77; CHECK-NEXT: ret void78;79entry:80 store i64 579005069656919567, ptr %ptr ; 0x08090a0b0c0d0e0f81 82 %bptr3 = getelementptr inbounds i8, ptr %ptr, i64 383 %wptr1 = getelementptr inbounds i16, ptr %ptr, i64 184 85 ;; We should be able to merge all these stores with the i64 one above86 ; value (not bytes) stored before ; 0x08090a0b0c0d0e0f87 store i8 7, ptr %bptr3 ; 0788 store i16 1541, ptr %wptr1 ; 060589 store i32 67305985, ptr %ptr ; 0403020190 ; 0x08090a0b0403020191 ret void92}93 94define void @merged_merges(ptr %ptr) {95; CHECK-LABEL: @merged_merges(96; CHECK-NEXT: entry:97; CHECK-NEXT: store i64 579005069572506113, ptr [[PTR:%.*]]98; CHECK-NEXT: ret void99;100entry:101 store i64 579005069656919567, ptr %ptr ; 0x08090a0b0c0d0e0f102 103 %bptr3 = getelementptr inbounds i8, ptr %ptr, i64 3104 %wptr1 = getelementptr inbounds i16, ptr %ptr, i64 1105 106 ;; We should be able to merge all these stores with the i64 one above107 ; value (not bytes) stored before ; 0x08090a0b0c0d0e0f108 store i32 67305985, ptr %ptr ; 04030201109 store i16 1541, ptr %wptr1 ; 0605110 store i8 7, ptr %bptr3 ; 07111 ; 0x08090a0b07050201112 ret void113}114 115define signext i8 @shouldnt_merge_since_theres_a_full_overlap(ptr %ptr) {116; CHECK-LABEL: @shouldnt_merge_since_theres_a_full_overlap(117; CHECK-NEXT: entry:118; CHECK-NEXT: [[BPTRM1:%.*]] = getelementptr inbounds i8, ptr [[PTR:%.*]], i64 -1119; CHECK-NEXT: [[BPTR3:%.*]] = getelementptr inbounds i8, ptr [[PTR]], i64 3120; CHECK-NEXT: store i32 1234, ptr [[BPTRM1]], align 1121; CHECK-NEXT: store i64 5678, ptr [[BPTR3]], align 1122; CHECK-NEXT: ret i8 0123;124entry:125 126 ; Also check that alias.scope metadata doesn't get dropped127 store i64 0, ptr %ptr, !alias.scope !32128 129 %bptrm1 = getelementptr inbounds i8, ptr %ptr, i64 -1130 %bptr3 = getelementptr inbounds i8, ptr %ptr, i64 3131 132 store i32 1234, ptr %bptrm1, align 1133 store i64 5678, ptr %bptr3, align 1134 135 ret i8 0136}137 138;; Test case from PR31777139%union.U = type { i64 }140 141define void @foo(ptr nocapture %u) {142; CHECK-LABEL: @foo(143; CHECK-NEXT: entry:144; CHECK-NEXT: store i64 42, ptr [[U:%.*]], align 8145; CHECK-NEXT: ret void146;147entry:148 store i64 0, ptr %u, align 8, !dbg !22, !tbaa !26, !noalias !32, !nontemporal !29149 store i16 42, ptr %u, align 8150 ret void151}152 153; Don't crash by operating on stale data if we merge (kill) the last 2 stores.154 155define void @PR34074(ptr %x, ptr %y) {156; CHECK-LABEL: @PR34074(157; CHECK-NEXT: store i64 42, ptr %y158; CHECK-NEXT: store i32 4, ptr %x159; CHECK-NEXT: ret void160;161 store i64 42, ptr %y ; independent store162 store i32 0, ptr %x ; big store of constant163 store i8 4, ptr %x ; small store with mergeable constant164 ret void165}166 167; We can't eliminate the last store because P and Q may alias.168 169define void @PR36129(ptr %P, ptr %Q) {170; CHECK-LABEL: @PR36129(171; CHECK-NEXT: store i32 1, ptr [[P:%.*]], align 4172; CHECK-NEXT: store i32 2, ptr [[Q:%.*]], align 4173; CHECK-NEXT: store i8 3, ptr [[P]], align 1174; CHECK-NEXT: ret void175;176 store i32 1, ptr %P, align 4177 store i32 2, ptr %Q, align 4178 store i8 3, ptr %P, align 1179 ret void180}181 182!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus, file: !1, producer: "clang version 5.0.0 (trunk 306512)", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug, enums: !2)183!1 = !DIFile(filename: "me.cpp", directory: "/compiler-explorer")184!2 = !{}185!7 = distinct !DISubprogram(name: "foo", linkageName: "foo(U*)", scope: !1, file: !1, line: 9, type: !8, isLocal: false, isDefinition: true, scopeLine: 9, flags: DIFlagPrototyped, isOptimized: true, unit: !0, retainedNodes: !20)186!8 = !DISubroutineType(types: !9)187!9 = !{null, !10}188!10 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !11, size: 64)189!11 = distinct !DICompositeType(tag: DW_TAG_union_type, name: "U", file: !1, line: 4, size: 64, elements: !12, identifier: "typeinfo name for U")190!12 = !{!13, !17}191!13 = !DIDerivedType(tag: DW_TAG_member, name: "i", scope: !11, file: !1, line: 5, baseType: !14, size: 64)192!14 = !DIDerivedType(tag: DW_TAG_typedef, name: "uint64_t", file: !15, line: 55, baseType: !16)193!15 = !DIFile(filename: "/usr/include/stdint.h", directory: "/compiler-explorer")194!16 = !DIBasicType(name: "long unsigned int", size: 64, encoding: DW_ATE_unsigned)195!17 = !DIDerivedType(tag: DW_TAG_member, name: "s", scope: !11, file: !1, line: 6, baseType: !18, size: 16)196!18 = !DIDerivedType(tag: DW_TAG_typedef, name: "uint16_t", file: !15, line: 49, baseType: !19)197!19 = !DIBasicType(name: "unsigned short", size: 16, encoding: DW_ATE_unsigned)198!20 = !{!21}199!21 = !DILocalVariable(name: "u", arg: 1, scope: !7, file: !1, line: 9, type: !10)200!22 = !DILocation(line: 10, column: 8, scope: !7)201 202!26 = !{!27, !27, i64 0}203!27 = !{!"omnipotent char", !28, i64 0}204!28 = !{!"Simple C++ TBAA"}205 206!29 = !{i32 1}207 208; Domains and scopes which might alias209!30 = !{!30}210!31 = !{!31, !30}211!32 = !{!31}212