1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -memcpyopt -S -enable-memcpyopt-memoryssa=0 | FileCheck %s 3; RUN: opt < %s -memcpyopt -S -enable-memcpyopt-memoryssa=1 -verify-memoryssa | FileCheck %s 4 5; All the stores in this example should be merged into a single memset. 6 7target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128" 8target triple = "i386-apple-darwin8" 9 10define void @test1(i8 signext %c) nounwind { 11; CHECK-LABEL: @test1( 12; CHECK-NEXT: entry: 13; CHECK-NEXT: [[X:%.*]] = alloca [19 x i8], align 1 14; CHECK-NEXT: [[TMP:%.*]] = getelementptr [19 x i8], [19 x i8]* [[X]], i32 0, i32 0 15; CHECK-NEXT: [[TMP5:%.*]] = getelementptr [19 x i8], [19 x i8]* [[X]], i32 0, i32 1 16; CHECK-NEXT: [[TMP9:%.*]] = getelementptr [19 x i8], [19 x i8]* [[X]], i32 0, i32 2 17; CHECK-NEXT: [[TMP13:%.*]] = getelementptr [19 x i8], [19 x i8]* [[X]], i32 0, i32 3 18; CHECK-NEXT: [[TMP17:%.*]] = getelementptr [19 x i8], [19 x i8]* [[X]], i32 0, i32 4 19; CHECK-NEXT: [[TMP21:%.*]] = getelementptr [19 x i8], [19 x i8]* [[X]], i32 0, i32 5 20; CHECK-NEXT: [[TMP25:%.*]] = getelementptr [19 x i8], [19 x i8]* [[X]], i32 0, i32 6 21; CHECK-NEXT: [[TMP29:%.*]] = getelementptr [19 x i8], [19 x i8]* [[X]], i32 0, i32 7 22; CHECK-NEXT: [[TMP33:%.*]] = getelementptr [19 x i8], [19 x i8]* [[X]], i32 0, i32 8 23; CHECK-NEXT: [[TMP37:%.*]] = getelementptr [19 x i8], [19 x i8]* [[X]], i32 0, i32 9 24; CHECK-NEXT: [[TMP41:%.*]] = getelementptr [19 x i8], [19 x i8]* [[X]], i32 0, i32 10 25; CHECK-NEXT: [[TMP45:%.*]] = getelementptr [19 x i8], [19 x i8]* [[X]], i32 0, i32 11 26; CHECK-NEXT: [[TMP49:%.*]] = getelementptr [19 x i8], [19 x i8]* [[X]], i32 0, i32 12 27; CHECK-NEXT: [[TMP53:%.*]] = getelementptr [19 x i8], [19 x i8]* [[X]], i32 0, i32 13 28; CHECK-NEXT: [[TMP57:%.*]] = getelementptr [19 x i8], [19 x i8]* [[X]], i32 0, i32 14 29; CHECK-NEXT: [[TMP61:%.*]] = getelementptr [19 x i8], [19 x i8]* [[X]], i32 0, i32 15 30; CHECK-NEXT: [[TMP65:%.*]] = getelementptr [19 x i8], [19 x i8]* [[X]], i32 0, i32 16 31; CHECK-NEXT: [[TMP69:%.*]] = getelementptr [19 x i8], [19 x i8]* [[X]], i32 0, i32 17 32; CHECK-NEXT: [[TMP73:%.*]] = getelementptr [19 x i8], [19 x i8]* [[X]], i32 0, i32 18 33; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* align 1 [[TMP]], i8 [[C:%.*]], i64 19, i1 false) 34; CHECK-NEXT: [[TMP76:%.*]] = call i32 (...) @bar([19 x i8]* [[X]]) [[ATTR0:#.*]] 35; CHECK-NEXT: ret void 36; 37entry: 38 %x = alloca [19 x i8] ; <[19 x i8]*> [#uses=20] 39 %tmp = getelementptr [19 x i8], [19 x i8]* %x, i32 0, i32 0 ; <i8*> [#uses=1] 40 store i8 %c, i8* %tmp, align 1 41 %tmp5 = getelementptr [19 x i8], [19 x i8]* %x, i32 0, i32 1 ; <i8*> [#uses=1] 42 store i8 %c, i8* %tmp5, align 1 43 %tmp9 = getelementptr [19 x i8], [19 x i8]* %x, i32 0, i32 2 ; <i8*> [#uses=1] 44 store i8 %c, i8* %tmp9, align 1 45 %tmp13 = getelementptr [19 x i8], [19 x i8]* %x, i32 0, i32 3 ; <i8*> [#uses=1] 46 store i8 %c, i8* %tmp13, align 1 47 %tmp17 = getelementptr [19 x i8], [19 x i8]* %x, i32 0, i32 4 ; <i8*> [#uses=1] 48 store i8 %c, i8* %tmp17, align 1 49 %tmp21 = getelementptr [19 x i8], [19 x i8]* %x, i32 0, i32 5 ; <i8*> [#uses=1] 50 store i8 %c, i8* %tmp21, align 1 51 %tmp25 = getelementptr [19 x i8], [19 x i8]* %x, i32 0, i32 6 ; <i8*> [#uses=1] 52 store i8 %c, i8* %tmp25, align 1 53 %tmp29 = getelementptr [19 x i8], [19 x i8]* %x, i32 0, i32 7 ; <i8*> [#uses=1] 54 store i8 %c, i8* %tmp29, align 1 55 %tmp33 = getelementptr [19 x i8], [19 x i8]* %x, i32 0, i32 8 ; <i8*> [#uses=1] 56 store i8 %c, i8* %tmp33, align 1 57 %tmp37 = getelementptr [19 x i8], [19 x i8]* %x, i32 0, i32 9 ; <i8*> [#uses=1] 58 store i8 %c, i8* %tmp37, align 1 59 %tmp41 = getelementptr [19 x i8], [19 x i8]* %x, i32 0, i32 10 ; <i8*> [#uses=1] 60 store i8 %c, i8* %tmp41, align 1 61 %tmp45 = getelementptr [19 x i8], [19 x i8]* %x, i32 0, i32 11 ; <i8*> [#uses=1] 62 store i8 %c, i8* %tmp45, align 1 63 %tmp49 = getelementptr [19 x i8], [19 x i8]* %x, i32 0, i32 12 ; <i8*> [#uses=1] 64 store i8 %c, i8* %tmp49, align 1 65 %tmp53 = getelementptr [19 x i8], [19 x i8]* %x, i32 0, i32 13 ; <i8*> [#uses=1] 66 store i8 %c, i8* %tmp53, align 1 67 %tmp57 = getelementptr [19 x i8], [19 x i8]* %x, i32 0, i32 14 ; <i8*> [#uses=1] 68 store i8 %c, i8* %tmp57, align 1 69 %tmp61 = getelementptr [19 x i8], [19 x i8]* %x, i32 0, i32 15 ; <i8*> [#uses=1] 70 store i8 %c, i8* %tmp61, align 1 71 %tmp65 = getelementptr [19 x i8], [19 x i8]* %x, i32 0, i32 16 ; <i8*> [#uses=1] 72 store i8 %c, i8* %tmp65, align 1 73 %tmp69 = getelementptr [19 x i8], [19 x i8]* %x, i32 0, i32 17 ; <i8*> [#uses=1] 74 store i8 %c, i8* %tmp69, align 1 75 %tmp73 = getelementptr [19 x i8], [19 x i8]* %x, i32 0, i32 18 ; <i8*> [#uses=1] 76 store i8 %c, i8* %tmp73, align 1 77 %tmp76 = call i32 (...) @bar( [19 x i8]* %x ) nounwind 78 ret void 79} 80 81declare i32 @bar(...) 82 83%struct.MV = type { i16, i16 } 84 85 86define void @test2() nounwind { 87; CHECK-LABEL: @test2( 88; CHECK-NEXT: entry: 89; CHECK-NEXT: [[REF_IDX:%.*]] = alloca [8 x i8], align 1 90; CHECK-NEXT: [[LEFT_MVD:%.*]] = alloca [8 x %struct.MV], align 8 91; CHECK-NEXT: [[UP_MVD:%.*]] = alloca [8 x %struct.MV], align 8 92; CHECK-NEXT: [[TMP20:%.*]] = getelementptr [8 x i8], [8 x i8]* [[REF_IDX]], i32 0, i32 7 93; CHECK-NEXT: [[TMP23:%.*]] = getelementptr [8 x i8], [8 x i8]* [[REF_IDX]], i32 0, i32 6 94; CHECK-NEXT: [[TMP26:%.*]] = getelementptr [8 x i8], [8 x i8]* [[REF_IDX]], i32 0, i32 5 95; CHECK-NEXT: [[TMP29:%.*]] = getelementptr [8 x i8], [8 x i8]* [[REF_IDX]], i32 0, i32 4 96; CHECK-NEXT: [[TMP32:%.*]] = getelementptr [8 x i8], [8 x i8]* [[REF_IDX]], i32 0, i32 3 97; CHECK-NEXT: [[TMP35:%.*]] = getelementptr [8 x i8], [8 x i8]* [[REF_IDX]], i32 0, i32 2 98; CHECK-NEXT: [[TMP38:%.*]] = getelementptr [8 x i8], [8 x i8]* [[REF_IDX]], i32 0, i32 1 99; CHECK-NEXT: [[TMP41:%.*]] = getelementptr [8 x i8], [8 x i8]* [[REF_IDX]], i32 0, i32 0 100; CHECK-NEXT: [[TMP43:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[UP_MVD]], i32 0, i32 7, i32 0 101; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* align 1 [[TMP41]], i8 -1, i64 8, i1 false) 102; CHECK-NEXT: [[TMP46:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[UP_MVD]], i32 0, i32 7, i32 1 103; CHECK-NEXT: [[TMP57:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[UP_MVD]], i32 0, i32 6, i32 0 104; CHECK-NEXT: [[TMP60:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[UP_MVD]], i32 0, i32 6, i32 1 105; CHECK-NEXT: [[TMP71:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[UP_MVD]], i32 0, i32 5, i32 0 106; CHECK-NEXT: [[TMP74:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[UP_MVD]], i32 0, i32 5, i32 1 107; CHECK-NEXT: [[TMP85:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[UP_MVD]], i32 0, i32 4, i32 0 108; CHECK-NEXT: [[TMP88:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[UP_MVD]], i32 0, i32 4, i32 1 109; CHECK-NEXT: [[TMP99:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[UP_MVD]], i32 0, i32 3, i32 0 110; CHECK-NEXT: [[TMP102:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[UP_MVD]], i32 0, i32 3, i32 1 111; CHECK-NEXT: [[TMP113:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[UP_MVD]], i32 0, i32 2, i32 0 112; CHECK-NEXT: [[TMP116:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[UP_MVD]], i32 0, i32 2, i32 1 113; CHECK-NEXT: [[TMP127:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[UP_MVD]], i32 0, i32 1, i32 0 114; CHECK-NEXT: [[TMP130:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[UP_MVD]], i32 0, i32 1, i32 1 115; CHECK-NEXT: [[TMP141:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[UP_MVD]], i32 0, i32 0, i32 0 116; CHECK-NEXT: [[TMP144:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[UP_MVD]], i32 0, i32 0, i32 1 117; CHECK-NEXT: [[TMP148:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[LEFT_MVD]], i32 0, i32 7, i32 0 118; CHECK-NEXT: [[TMP0:%.*]] = bitcast i16* [[TMP141]] to i8* 119; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* align 8 [[TMP0]], i8 0, i64 32, i1 false) 120; CHECK-NEXT: [[TMP151:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[LEFT_MVD]], i32 0, i32 7, i32 1 121; CHECK-NEXT: [[TMP162:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[LEFT_MVD]], i32 0, i32 6, i32 0 122; CHECK-NEXT: [[TMP165:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[LEFT_MVD]], i32 0, i32 6, i32 1 123; CHECK-NEXT: [[TMP176:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[LEFT_MVD]], i32 0, i32 5, i32 0 124; CHECK-NEXT: [[TMP179:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[LEFT_MVD]], i32 0, i32 5, i32 1 125; CHECK-NEXT: [[TMP190:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[LEFT_MVD]], i32 0, i32 4, i32 0 126; CHECK-NEXT: [[TMP193:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[LEFT_MVD]], i32 0, i32 4, i32 1 127; CHECK-NEXT: [[TMP204:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[LEFT_MVD]], i32 0, i32 3, i32 0 128; CHECK-NEXT: [[TMP207:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[LEFT_MVD]], i32 0, i32 3, i32 1 129; CHECK-NEXT: [[TMP218:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[LEFT_MVD]], i32 0, i32 2, i32 0 130; CHECK-NEXT: [[TMP221:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[LEFT_MVD]], i32 0, i32 2, i32 1 131; CHECK-NEXT: [[TMP232:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[LEFT_MVD]], i32 0, i32 1, i32 0 132; CHECK-NEXT: [[TMP235:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[LEFT_MVD]], i32 0, i32 1, i32 1 133; CHECK-NEXT: [[TMP246:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[LEFT_MVD]], i32 0, i32 0, i32 0 134; CHECK-NEXT: [[TMP249:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[LEFT_MVD]], i32 0, i32 0, i32 1 135; CHECK-NEXT: [[UP_MVD252:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[UP_MVD]], i32 0, i32 0 136; CHECK-NEXT: [[LEFT_MVD253:%.*]] = getelementptr [8 x %struct.MV], [8 x %struct.MV]* [[LEFT_MVD]], i32 0, i32 0 137; CHECK-NEXT: [[TMP1:%.*]] = bitcast i16* [[TMP246]] to i8* 138; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* align 8 [[TMP1]], i8 0, i64 32, i1 false) 139; CHECK-NEXT: call void @foo(%struct.MV* [[UP_MVD252]], %struct.MV* [[LEFT_MVD253]], i8* [[TMP41]]) [[ATTR0]] 140; CHECK-NEXT: ret void 141; 142entry: 143 %ref_idx = alloca [8 x i8] ; <[8 x i8]*> [#uses=8] 144 %left_mvd = alloca [8 x %struct.MV] ; <[8 x %struct.MV]*> [#uses=17] 145 %up_mvd = alloca [8 x %struct.MV] ; <[8 x %struct.MV]*> [#uses=17] 146 %tmp20 = getelementptr [8 x i8], [8 x i8]* %ref_idx, i32 0, i32 7 ; <i8*> [#uses=1] 147 store i8 -1, i8* %tmp20, align 1 148 %tmp23 = getelementptr [8 x i8], [8 x i8]* %ref_idx, i32 0, i32 6 ; <i8*> [#uses=1] 149 store i8 -1, i8* %tmp23, align 1 150 %tmp26 = getelementptr [8 x i8], [8 x i8]* %ref_idx, i32 0, i32 5 ; <i8*> [#uses=1] 151 store i8 -1, i8* %tmp26, align 1 152 %tmp29 = getelementptr [8 x i8], [8 x i8]* %ref_idx, i32 0, i32 4 ; <i8*> [#uses=1] 153 store i8 -1, i8* %tmp29, align 1 154 %tmp32 = getelementptr [8 x i8], [8 x i8]* %ref_idx, i32 0, i32 3 ; <i8*> [#uses=1] 155 store i8 -1, i8* %tmp32, align 1 156 %tmp35 = getelementptr [8 x i8], [8 x i8]* %ref_idx, i32 0, i32 2 ; <i8*> [#uses=1] 157 store i8 -1, i8* %tmp35, align 1 158 %tmp38 = getelementptr [8 x i8], [8 x i8]* %ref_idx, i32 0, i32 1 ; <i8*> [#uses=1] 159 store i8 -1, i8* %tmp38, align 1 160 %tmp41 = getelementptr [8 x i8], [8 x i8]* %ref_idx, i32 0, i32 0 ; <i8*> [#uses=2] 161 store i8 -1, i8* %tmp41, align 1 162 %tmp43 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %up_mvd, i32 0, i32 7, i32 0 ; <i16*> [#uses=1] 163 store i16 0, i16* %tmp43, align 2 164 %tmp46 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %up_mvd, i32 0, i32 7, i32 1 ; <i16*> [#uses=1] 165 store i16 0, i16* %tmp46, align 2 166 %tmp57 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %up_mvd, i32 0, i32 6, i32 0 ; <i16*> [#uses=1] 167 store i16 0, i16* %tmp57, align 2 168 %tmp60 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %up_mvd, i32 0, i32 6, i32 1 ; <i16*> [#uses=1] 169 store i16 0, i16* %tmp60, align 2 170 %tmp71 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %up_mvd, i32 0, i32 5, i32 0 ; <i16*> [#uses=1] 171 store i16 0, i16* %tmp71, align 2 172 %tmp74 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %up_mvd, i32 0, i32 5, i32 1 ; <i16*> [#uses=1] 173 store i16 0, i16* %tmp74, align 2 174 %tmp85 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %up_mvd, i32 0, i32 4, i32 0 ; <i16*> [#uses=1] 175 store i16 0, i16* %tmp85, align 2 176 %tmp88 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %up_mvd, i32 0, i32 4, i32 1 ; <i16*> [#uses=1] 177 store i16 0, i16* %tmp88, align 2 178 %tmp99 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %up_mvd, i32 0, i32 3, i32 0 ; <i16*> [#uses=1] 179 store i16 0, i16* %tmp99, align 2 180 %tmp102 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %up_mvd, i32 0, i32 3, i32 1 ; <i16*> [#uses=1] 181 store i16 0, i16* %tmp102, align 2 182 %tmp113 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %up_mvd, i32 0, i32 2, i32 0 ; <i16*> [#uses=1] 183 store i16 0, i16* %tmp113, align 2 184 %tmp116 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %up_mvd, i32 0, i32 2, i32 1 ; <i16*> [#uses=1] 185 store i16 0, i16* %tmp116, align 2 186 %tmp127 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %up_mvd, i32 0, i32 1, i32 0 ; <i16*> [#uses=1] 187 store i16 0, i16* %tmp127, align 2 188 %tmp130 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %up_mvd, i32 0, i32 1, i32 1 ; <i16*> [#uses=1] 189 store i16 0, i16* %tmp130, align 2 190 %tmp141 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %up_mvd, i32 0, i32 0, i32 0 ; <i16*> [#uses=1] 191 store i16 0, i16* %tmp141, align 8 192 %tmp144 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %up_mvd, i32 0, i32 0, i32 1 ; <i16*> [#uses=1] 193 store i16 0, i16* %tmp144, align 2 194 %tmp148 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %left_mvd, i32 0, i32 7, i32 0 ; <i16*> [#uses=1] 195 store i16 0, i16* %tmp148, align 2 196 %tmp151 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %left_mvd, i32 0, i32 7, i32 1 ; <i16*> [#uses=1] 197 store i16 0, i16* %tmp151, align 2 198 %tmp162 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %left_mvd, i32 0, i32 6, i32 0 ; <i16*> [#uses=1] 199 store i16 0, i16* %tmp162, align 2 200 %tmp165 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %left_mvd, i32 0, i32 6, i32 1 ; <i16*> [#uses=1] 201 store i16 0, i16* %tmp165, align 2 202 %tmp176 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %left_mvd, i32 0, i32 5, i32 0 ; <i16*> [#uses=1] 203 store i16 0, i16* %tmp176, align 2 204 %tmp179 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %left_mvd, i32 0, i32 5, i32 1 ; <i16*> [#uses=1] 205 store i16 0, i16* %tmp179, align 2 206 %tmp190 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %left_mvd, i32 0, i32 4, i32 0 ; <i16*> [#uses=1] 207 store i16 0, i16* %tmp190, align 2 208 %tmp193 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %left_mvd, i32 0, i32 4, i32 1 ; <i16*> [#uses=1] 209 store i16 0, i16* %tmp193, align 2 210 %tmp204 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %left_mvd, i32 0, i32 3, i32 0 ; <i16*> [#uses=1] 211 store i16 0, i16* %tmp204, align 2 212 %tmp207 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %left_mvd, i32 0, i32 3, i32 1 ; <i16*> [#uses=1] 213 store i16 0, i16* %tmp207, align 2 214 %tmp218 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %left_mvd, i32 0, i32 2, i32 0 ; <i16*> [#uses=1] 215 store i16 0, i16* %tmp218, align 2 216 %tmp221 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %left_mvd, i32 0, i32 2, i32 1 ; <i16*> [#uses=1] 217 store i16 0, i16* %tmp221, align 2 218 %tmp232 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %left_mvd, i32 0, i32 1, i32 0 ; <i16*> [#uses=1] 219 store i16 0, i16* %tmp232, align 2 220 %tmp235 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %left_mvd, i32 0, i32 1, i32 1 ; <i16*> [#uses=1] 221 store i16 0, i16* %tmp235, align 2 222 %tmp246 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %left_mvd, i32 0, i32 0, i32 0 ; <i16*> [#uses=1] 223 store i16 0, i16* %tmp246, align 8 224 %tmp249 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %left_mvd, i32 0, i32 0, i32 1 ; <i16*> [#uses=1] 225 store i16 0, i16* %tmp249, align 2 226 %up_mvd252 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %up_mvd, i32 0, i32 0 ; <%struct.MV*> [#uses=1] 227 %left_mvd253 = getelementptr [8 x %struct.MV], [8 x %struct.MV]* %left_mvd, i32 0, i32 0 ; <%struct.MV*> [#uses=1] 228 call void @foo( %struct.MV* %up_mvd252, %struct.MV* %left_mvd253, i8* %tmp41 ) nounwind 229 ret void 230 231} 232 233declare void @foo(%struct.MV*, %struct.MV*, i8*) 234 235 236; Store followed by memset. 237define void @test3(i32* nocapture %P) nounwind ssp { 238; CHECK-LABEL: @test3( 239; CHECK-NEXT: entry: 240; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[P:%.*]], i64 1 241; CHECK-NEXT: [[ADD_PTR:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 2 242; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[ADD_PTR]] to i8* 243; CHECK-NEXT: [[TMP1:%.*]] = bitcast i32* [[ARRAYIDX]] to i8* 244; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* align 4 [[TMP1]], i8 0, i64 15, i1 false) 245; CHECK-NEXT: ret void 246; 247entry: 248 %arrayidx = getelementptr inbounds i32, i32* %P, i64 1 249 store i32 0, i32* %arrayidx, align 4 250 %add.ptr = getelementptr inbounds i32, i32* %P, i64 2 251 %0 = bitcast i32* %add.ptr to i8* 252 tail call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 11, i1 false) 253 ret void 254} 255 256; store followed by memset, different offset scenario 257define void @test4(i32* nocapture %P) nounwind ssp { 258; CHECK-LABEL: @test4( 259; CHECK-NEXT: entry: 260; CHECK-NEXT: [[ADD_PTR:%.*]] = getelementptr inbounds i32, i32* [[P:%.*]], i64 1 261; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[ADD_PTR]] to i8* 262; CHECK-NEXT: [[TMP1:%.*]] = bitcast i32* [[P]] to i8* 263; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* align 4 [[TMP1]], i8 0, i64 15, i1 false) 264; CHECK-NEXT: ret void 265; 266entry: 267 store i32 0, i32* %P, align 4 268 %add.ptr = getelementptr inbounds i32, i32* %P, i64 1 269 %0 = bitcast i32* %add.ptr to i8* 270 tail call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 11, i1 false) 271 ret void 272} 273 274declare void @llvm.memset.p0i8.i64(i8* nocapture, i8, i64, i1) nounwind 275 276; Memset followed by store. 277define void @test5(i32* nocapture %P) nounwind ssp { 278; CHECK-LABEL: @test5( 279; CHECK-NEXT: entry: 280; CHECK-NEXT: [[ADD_PTR:%.*]] = getelementptr inbounds i32, i32* [[P:%.*]], i64 2 281; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[ADD_PTR]] to i8* 282; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 1 283; CHECK-NEXT: [[TMP1:%.*]] = bitcast i32* [[ARRAYIDX]] to i8* 284; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* align 4 [[TMP1]], i8 0, i64 15, i1 false) 285; CHECK-NEXT: ret void 286; 287entry: 288 %add.ptr = getelementptr inbounds i32, i32* %P, i64 2 289 %0 = bitcast i32* %add.ptr to i8* 290 tail call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 11, i1 false) 291 %arrayidx = getelementptr inbounds i32, i32* %P, i64 1 292 store i32 0, i32* %arrayidx, align 4 293 ret void 294} 295 296;; Memset followed by memset. 297define void @test6(i32* nocapture %P) nounwind ssp { 298; CHECK-LABEL: @test6( 299; CHECK-NEXT: entry: 300; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[P:%.*]] to i8* 301; CHECK-NEXT: [[ADD_PTR:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 3 302; CHECK-NEXT: [[TMP1:%.*]] = bitcast i32* [[ADD_PTR]] to i8* 303; CHECK-NEXT: [[TMP2:%.*]] = bitcast i32* [[P]] to i8* 304; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* [[TMP2]], i8 0, i64 24, i1 false) 305; CHECK-NEXT: ret void 306; 307entry: 308 %0 = bitcast i32* %P to i8* 309 tail call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 12, i1 false) 310 %add.ptr = getelementptr inbounds i32, i32* %P, i64 3 311 %1 = bitcast i32* %add.ptr to i8* 312 tail call void @llvm.memset.p0i8.i64(i8* %1, i8 0, i64 12, i1 false) 313 ret void 314} 315 316; More aggressive heuristic 317; rdar://9892684 318define void @test7(i32* nocapture %c) nounwind optsize { 319; CHECK-LABEL: @test7( 320; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32, i32* [[C:%.*]], i32 1 321; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[C]], i32 2 322; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, i32* [[C]], i32 3 323; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, i32* [[C]], i32 4 324; CHECK-NEXT: [[TMP5:%.*]] = bitcast i32* [[C]] to i8* 325; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* align 4 [[TMP5]], i8 -1, i64 20, i1 false) 326; CHECK-NEXT: ret void 327; 328 store i32 -1, i32* %c, align 4 329 %1 = getelementptr inbounds i32, i32* %c, i32 1 330 store i32 -1, i32* %1, align 4 331 %2 = getelementptr inbounds i32, i32* %c, i32 2 332 store i32 -1, i32* %2, align 4 333 %3 = getelementptr inbounds i32, i32* %c, i32 3 334 store i32 -1, i32* %3, align 4 335 %4 = getelementptr inbounds i32, i32* %c, i32 4 336 store i32 -1, i32* %4, align 4 337 ret void 338} 339 340%struct.test8 = type { [4 x i32] } 341 342define void @test8() { 343; CHECK-LABEL: @test8( 344; CHECK-NEXT: entry: 345; CHECK-NEXT: [[MEMTMP:%.*]] = alloca [[STRUCT_TEST8:%.*]], align 16 346; CHECK-NEXT: [[TMP0:%.*]] = bitcast %struct.test8* [[MEMTMP]] to <4 x i32>* 347; CHECK-NEXT: store <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1>, <4 x i32>* [[TMP0]], align 16 348; CHECK-NEXT: ret void 349; 350entry: 351 %memtmp = alloca %struct.test8, align 16 352 %0 = bitcast %struct.test8* %memtmp to <4 x i32>* 353 store <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1>, <4 x i32>* %0, align 16 354 ret void 355} 356 357@test9buf = internal unnamed_addr global [16 x i64] zeroinitializer, align 16 358 359define void @test9() nounwind { 360; CHECK-LABEL: @test9( 361; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* align 16 bitcast ([16 x i64]* @test9buf to i8*), i8 -1, i64 16, i1 false) 362; CHECK-NEXT: ret void 363; 364 store i8 -1, i8* bitcast ([16 x i64]* @test9buf to i8*), align 16 365 store i8 -1, i8* getelementptr (i8, i8* bitcast ([16 x i64]* @test9buf to i8*), i64 1), align 1 366 store i8 -1, i8* getelementptr (i8, i8* bitcast ([16 x i64]* @test9buf to i8*), i64 2), align 2 367 store i8 -1, i8* getelementptr (i8, i8* bitcast ([16 x i64]* @test9buf to i8*), i64 3), align 1 368 store i8 -1, i8* getelementptr (i8, i8* bitcast ([16 x i64]* @test9buf to i8*), i64 4), align 4 369 store i8 -1, i8* getelementptr (i8, i8* bitcast ([16 x i64]* @test9buf to i8*), i64 5), align 1 370 store i8 -1, i8* getelementptr (i8, i8* bitcast ([16 x i64]* @test9buf to i8*), i64 6), align 2 371 store i8 -1, i8* getelementptr (i8, i8* bitcast ([16 x i64]* @test9buf to i8*), i64 7), align 1 372 store i8 -1, i8* bitcast (i64* getelementptr inbounds ([16 x i64], [16 x i64]* @test9buf, i64 0, i64 1) to i8*), align 8 373 store i8 -1, i8* getelementptr (i8, i8* bitcast ([16 x i64]* @test9buf to i8*), i64 9), align 1 374 store i8 -1, i8* getelementptr (i8, i8* bitcast ([16 x i64]* @test9buf to i8*), i64 10), align 2 375 store i8 -1, i8* getelementptr (i8, i8* bitcast ([16 x i64]* @test9buf to i8*), i64 11), align 1 376 store i8 -1, i8* getelementptr (i8, i8* bitcast ([16 x i64]* @test9buf to i8*), i64 12), align 4 377 store i8 -1, i8* getelementptr (i8, i8* bitcast ([16 x i64]* @test9buf to i8*), i64 13), align 1 378 store i8 -1, i8* getelementptr (i8, i8* bitcast ([16 x i64]* @test9buf to i8*), i64 14), align 2 379 store i8 -1, i8* getelementptr (i8, i8* bitcast ([16 x i64]* @test9buf to i8*), i64 15), align 1 380 ret void 381} 382 383; PR19092 384define void @test10(i8* nocapture %P) nounwind { 385; CHECK-LABEL: @test10( 386; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* [[P:%.*]], i8 0, i64 42, i1 false) 387; CHECK-NEXT: ret void 388; 389 tail call void @llvm.memset.p0i8.i64(i8* %P, i8 0, i64 42, i1 false) 390 tail call void @llvm.memset.p0i8.i64(i8* %P, i8 0, i64 23, i1 false) 391 ret void 392} 393 394; Memset followed by odd store. 395define void @test11(i32* nocapture %P) nounwind ssp { 396; CHECK-LABEL: @test11( 397; CHECK-NEXT: entry: 398; CHECK-NEXT: [[ADD_PTR:%.*]] = getelementptr inbounds i32, i32* [[P:%.*]], i64 3 399; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[ADD_PTR]] to i8* 400; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 0 401; CHECK-NEXT: [[ARRAYIDX_CAST:%.*]] = bitcast i32* [[ARRAYIDX]] to i96* 402; CHECK-NEXT: [[TMP1:%.*]] = bitcast i96* [[ARRAYIDX_CAST]] to i8* 403; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* align 4 [[TMP1]], i8 1, i64 23, i1 false) 404; CHECK-NEXT: ret void 405; 406entry: 407 %add.ptr = getelementptr inbounds i32, i32* %P, i64 3 408 %0 = bitcast i32* %add.ptr to i8* 409 tail call void @llvm.memset.p0i8.i64(i8* %0, i8 1, i64 11, i1 false) 410 %arrayidx = getelementptr inbounds i32, i32* %P, i64 0 411 %arrayidx.cast = bitcast i32* %arrayidx to i96* 412 store i96 310698676526526814092329217, i96* %arrayidx.cast, align 4 413 ret void 414} 415 416; Alignment should be preserved when there is a store with default align 417define void @test12(i32* nocapture %P) nounwind ssp { 418; CHECK-LABEL: @test12( 419; CHECK-NEXT: entry: 420; CHECK-NEXT: [[ADD_PTR:%.*]] = getelementptr inbounds i32, i32* [[P:%.*]], i64 1 421; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[ADD_PTR]] to i8* 422; CHECK-NEXT: [[TMP1:%.*]] = bitcast i32* [[P]] to i8* 423; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* align 4 [[TMP1]], i8 0, i64 15, i1 false) 424; CHECK-NEXT: ret void 425; 426entry: 427 store i32 0, i32* %P 428 %add.ptr = getelementptr inbounds i32, i32* %P, i64 1 429 %0 = bitcast i32* %add.ptr to i8* 430 tail call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 11, i1 false) 431 ret void 432} 433