441 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -passes=instcombine -S | FileCheck %s3target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64"4@C.0.1248 = internal constant [128 x float] [ float -1.000000e+00, float -1.000000e+00, float -1.000000e+00, float 0.000000e+00, float -1.000000e+00, float -1.000000e+00, float 0.000000e+00, float -1.000000e+00, float -1.000000e+00, float -1.000000e+00, float 0.000000e+00, float 1.000000e+00, float -1.000000e+00, float -1.000000e+00, float 1.000000e+00, float 0.000000e+00, float -1.000000e+00, float 0.000000e+00, float -1.000000e+00, float -1.000000e+00, float -1.000000e+00, float 0.000000e+00, float -1.000000e+00, float 1.000000e+00, float -1.000000e+00, float 0.000000e+00, float 1.000000e+00, float -1.000000e+00, float -1.000000e+00, float 0.000000e+00, float 1.000000e+00, float 1.000000e+00, float -1.000000e+00, float 1.000000e+00, float -1.000000e+00, float 0.000000e+00, float -1.000000e+00, float 1.000000e+00, float 0.000000e+00, float -1.000000e+00, float -1.000000e+00, float 1.000000e+00, float 0.000000e+00, float 1.000000e+00, float -1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 0.000000e+00, float 0.000000e+00, float -1.000000e+00, float -1.000000e+00, float -1.000000e+00, float 0.000000e+00, float -1.000000e+00, float -1.000000e+00, float 1.000000e+00, float 0.000000e+00, float -1.000000e+00, float 1.000000e+00, float -1.000000e+00, float 0.000000e+00, float -1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float -1.000000e+00, float -1.000000e+00, float 0.000000e+00, float 1.000000e+00, float -1.000000e+00, float 0.000000e+00, float -1.000000e+00, float 1.000000e+00, float -1.000000e+00, float 0.000000e+00, float 1.000000e+00, float 1.000000e+00, float -1.000000e+00, float 1.000000e+00, float 0.000000e+00, float 1.000000e+00, float 0.000000e+00, float -1.000000e+00, float -1.000000e+00, float 1.000000e+00, float 0.000000e+00, float -1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 0.000000e+00, float 1.000000e+00, float -1.000000e+00, float 1.000000e+00, float 0.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float -1.000000e+00, float 0.000000e+00, float 1.000000e+00, float 1.000000e+00, float 0.000000e+00, float -1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 0.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 0.000000e+00, float 0.000000e+00, float 1.000000e+00, float -1.000000e+00, float -1.000000e+00, float 0.000000e+00, float 1.000000e+00, float -1.000000e+00, float 1.000000e+00, float 0.000000e+00, float 1.000000e+00, float 1.000000e+00, float -1.000000e+00, float 0.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00 ], align 32 ; <ptr> [#uses=1]5 6define float @test1(i32 %hash, float %x, float %y, float %z, float %w) {7; CHECK-LABEL: @test1(8; CHECK-NEXT: entry:9; CHECK-NEXT: [[T3:%.*]] = shl i32 [[HASH:%.*]], 210; CHECK-NEXT: [[T5:%.*]] = and i32 [[T3]], 12411; CHECK-NEXT: [[TMP0:%.*]] = zext nneg i32 [[T5]] to i6412; CHECK-NEXT: [[T753:%.*]] = getelementptr float, ptr @C.0.1248, i64 [[TMP0]]13; CHECK-NEXT: [[T9:%.*]] = load float, ptr [[T753]], align 414; CHECK-NEXT: [[T11:%.*]] = fmul float [[T9]], [[X:%.*]]15; CHECK-NEXT: [[T13:%.*]] = fadd float [[T11]], 0.000000e+0016; CHECK-NEXT: [[TMP1:%.*]] = zext nneg i32 [[T5]] to i6417; CHECK-NEXT: [[TMP2:%.*]] = getelementptr float, ptr @C.0.1248, i64 [[TMP1]]18; CHECK-NEXT: [[T1851:%.*]] = getelementptr i8, ptr [[TMP2]], i64 419; CHECK-NEXT: [[T19:%.*]] = load float, ptr [[T1851]], align 420; CHECK-NEXT: [[T21:%.*]] = fmul float [[T19]], [[Y:%.*]]21; CHECK-NEXT: [[T23:%.*]] = fadd float [[T21]], [[T13]]22; CHECK-NEXT: [[TMP3:%.*]] = zext nneg i32 [[T5]] to i6423; CHECK-NEXT: [[TMP4:%.*]] = getelementptr float, ptr @C.0.1248, i64 [[TMP3]]24; CHECK-NEXT: [[T2849:%.*]] = getelementptr i8, ptr [[TMP4]], i64 825; CHECK-NEXT: [[T29:%.*]] = load float, ptr [[T2849]], align 426; CHECK-NEXT: [[T31:%.*]] = fmul float [[T29]], [[Z:%.*]]27; CHECK-NEXT: [[T33:%.*]] = fadd float [[T31]], [[T23]]28; CHECK-NEXT: [[TMP5:%.*]] = zext nneg i32 [[T5]] to i6429; CHECK-NEXT: [[TMP6:%.*]] = getelementptr float, ptr @C.0.1248, i64 [[TMP5]]30; CHECK-NEXT: [[T3847:%.*]] = getelementptr i8, ptr [[TMP6]], i64 1231; CHECK-NEXT: [[T39:%.*]] = load float, ptr [[T3847]], align 432; CHECK-NEXT: [[T41:%.*]] = fmul float [[T39]], [[W:%.*]]33; CHECK-NEXT: [[T43:%.*]] = fadd float [[T41]], [[T33]]34; CHECK-NEXT: ret float [[T43]]35;36entry:37 %lookupTable = alloca [128 x float], align 1638 call void @llvm.memcpy.p0.p0.i64(ptr align 16 %lookupTable, ptr align 16 @C.0.1248, i64 512, i1 false)39 40 41 %t3 = shl i32 %hash, 242 %t5 = and i32 %t3, 12443 %t753 = getelementptr [128 x float], ptr %lookupTable, i32 0, i32 %t544 %t9 = load float, ptr %t75345 %t11 = fmul float %t9, %x46 %t13 = fadd float %t11, 0.000000e+0047 %t17.sum52 = or i32 %t5, 148 %t1851 = getelementptr [128 x float], ptr %lookupTable, i32 0, i32 %t17.sum5249 %t19 = load float, ptr %t185150 %t21 = fmul float %t19, %y51 %t23 = fadd float %t21, %t1352 %t27.sum50 = or i32 %t5, 253 %t2849 = getelementptr [128 x float], ptr %lookupTable, i32 0, i32 %t27.sum5054 %t29 = load float, ptr %t284955 %t31 = fmul float %t29, %z56 %t33 = fadd float %t31, %t2357 %t37.sum48 = or i32 %t5, 358 %t3847 = getelementptr [128 x float], ptr %lookupTable, i32 0, i32 %t37.sum4859 %t39 = load float, ptr %t384760 %t41 = fmul float %t39, %w61 %t43 = fadd float %t41, %t3362 ret float %t4363}64 65declare void @llvm.memcpy.p0.p0.i64(ptr nocapture, ptr nocapture, i64, i1) nounwind66declare void @llvm.memcpy.p1.p0.i64(ptr addrspace(1) nocapture, ptr nocapture, i64, i1) nounwind67declare void @llvm.memcpy.p0.p1.i64(ptr nocapture, ptr addrspace(1) nocapture, i64, i1) nounwind68declare void @llvm.memcpy.p1.p1.i64(ptr addrspace(1) nocapture, ptr addrspace(1) nocapture, i64, i1) nounwind69 70%T = type { i8, [123 x i8] }71%U = type { i32, i32, i32, i32, i32 }72 73@G = constant %T {i8 1, [123 x i8] zeroinitializer }74@H = constant [2 x %U] zeroinitializer, align 1675@I = internal addrspace(1) constant [4 x float] zeroinitializer , align 476 77define void @test2() {78; CHECK-LABEL: @test2(79; CHECK-NEXT: [[B:%.*]] = alloca [[T:%.*]], align 880; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr noundef nonnull align 8 dereferenceable(124) [[B]], ptr noundef nonnull align 16 dereferenceable(124) @G, i64 124, i1 false)81; CHECK-NEXT: call void @bar(ptr nonnull [[B]])82; CHECK-NEXT: ret void83;84 %A = alloca %T85 %B = alloca %T86 87 88; %A alloca is deleted89 90; use @G instead of %A91 call void @llvm.memcpy.p0.p0.i64(ptr align 4 %A, ptr align 4 @G, i64 124, i1 false)92 call void @llvm.memcpy.p0.p0.i64(ptr align 4 %B, ptr align 4 %A, i64 124, i1 false)93 call void @bar(ptr %B)94 ret void95}96 97define void @test2_no_null_opt() #0 {98; CHECK-LABEL: @test2_no_null_opt(99; CHECK-NEXT: [[B:%.*]] = alloca [[T:%.*]], align 8100; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr noundef nonnull align 8 dereferenceable(124) [[B]], ptr noundef nonnull align 16 dereferenceable(124) @G, i64 124, i1 false)101; CHECK-NEXT: call void @bar(ptr nonnull [[B]])102; CHECK-NEXT: ret void103;104 %A = alloca %T105 %B = alloca %T106 107 108; %A alloca is deleted109 110; use @G instead of %A111 call void @llvm.memcpy.p0.p0.i64(ptr align 4 %A, ptr align 4 @G, i64 124, i1 false)112 call void @llvm.memcpy.p0.p0.i64(ptr align 4 %B, ptr align 4 %A, i64 124, i1 false)113 call void @bar(ptr %B)114 ret void115}116 117define void @test2_addrspacecast() {118; CHECK-LABEL: @test2_addrspacecast(119; CHECK-NEXT: [[B:%.*]] = alloca [[T:%.*]], align 8120; CHECK-NEXT: [[B_CAST:%.*]] = addrspacecast ptr [[B]] to ptr addrspace(1)121; CHECK-NEXT: call void @llvm.memcpy.p1.p1.i64(ptr addrspace(1) noundef align 4 dereferenceable(124) [[B_CAST]], ptr addrspace(1) noundef align 4 dereferenceable(124) addrspacecast (ptr @G to ptr addrspace(1)), i64 124, i1 false)122; CHECK-NEXT: call void @bar_as1(ptr addrspace(1) [[B_CAST]])123; CHECK-NEXT: ret void124;125 %A = alloca %T126 %B = alloca %T127 %a.cast = addrspacecast ptr %A to ptr addrspace(1)128 %b.cast = addrspacecast ptr %B to ptr addrspace(1)129 130 131; %A alloca is deleted132; This doesn't exactly match what test2 does, because folding the type133; cast into the alloca doesn't work for the addrspacecast yet.134 135; use @G instead of %A136 call void @llvm.memcpy.p1.p0.i64(ptr addrspace(1) align 4 %a.cast, ptr align 4 @G, i64 124, i1 false)137 call void @llvm.memcpy.p1.p1.i64(ptr addrspace(1) align 4 %b.cast, ptr addrspace(1) align 4 %a.cast, i64 124, i1 false)138 call void @bar_as1(ptr addrspace(1) %b.cast)139 ret void140}141 142declare void @bar(ptr nocapture)143declare void @bar_may_capture(ptr)144declare void @bar_as1(ptr addrspace(1) nocapture)145 146 147;; Should be able to eliminate the alloca.148define void @test3_nocapture() {149; CHECK-LABEL: @test3_nocapture(150; CHECK-NEXT: call void @bar(ptr nonnull @G) #[[ATTR3:[0-9]+]]151; CHECK-NEXT: ret void152;153 %A = alloca %T154 call void @llvm.memcpy.p0.p0.i64(ptr align 4 %A, ptr align 4 @G, i64 124, i1 false)155 call void @bar(ptr %A) readonly156 ret void157}158 159; Can not eliminate the alloca, as the function may capture its address.160define void @test3_may_capture() {161; CHECK-LABEL: @test3_may_capture(162; CHECK-NEXT: [[A:%.*]] = alloca [[T:%.*]], align 8163; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr noundef nonnull align 8 dereferenceable(124) [[A]], ptr noundef nonnull align 16 dereferenceable(124) @G, i64 124, i1 false)164; CHECK-NEXT: call void @bar_may_capture(ptr nonnull [[A]]) #[[ATTR3]]165; CHECK-NEXT: ret void166;167 %A = alloca %T168 call void @llvm.memcpy.p0.p0.i64(ptr align 4 %A, ptr align 4 @G, i64 124, i1 false)169 call void @bar_may_capture(ptr %A) readonly170 ret void171}172 173define void @test3_addrspacecast() {174; CHECK-LABEL: @test3_addrspacecast(175; CHECK-NEXT: call void @bar(ptr nonnull @G) #[[ATTR3]]176; CHECK-NEXT: ret void177;178 %A = alloca %T179 call void @llvm.memcpy.p0.p1.i64(ptr align 4 %A, ptr addrspace(1) align 4 addrspacecast (ptr @G to ptr addrspace(1)), i64 124, i1 false)180 call void @bar(ptr %A) readonly181 ret void182}183 184 185define void @test4() {186; CHECK-LABEL: @test4(187; CHECK-NEXT: call void @baz(ptr nonnull byval(i8) @G)188; CHECK-NEXT: ret void189;190 %A = alloca %T191 call void @llvm.memcpy.p0.p0.i64(ptr align 4 %A, ptr align 4 @G, i64 124, i1 false)192 call void @baz(ptr byval(i8) %A)193 ret void194}195 196declare void @llvm.lifetime.start.p0(ptr)197define void @test5() {198; CHECK-LABEL: @test5(199; CHECK-NEXT: call void @baz(ptr nonnull byval(i8) @G)200; CHECK-NEXT: ret void201;202 %A = alloca %T203 call void @llvm.lifetime.start.p0(ptr %A)204 call void @llvm.memcpy.p0.p0.i64(ptr align 4 %A, ptr align 4 @G, i64 124, i1 false)205 call void @baz(ptr byval(i8) %A)206 ret void207}208 209 210declare void @baz(ptr byval(i8))211 212 213define void @test6() {214; CHECK-LABEL: @test6(215; CHECK-NEXT: call void @bar(ptr nonnull @H) #[[ATTR3]]216; CHECK-NEXT: ret void217;218 %A = alloca %U, align 16219 call void @llvm.memcpy.p0.p0.i64(ptr align 16 %A, ptr align 16 @H, i64 20, i1 false)220 call void @bar(ptr %A) readonly221 ret void222}223 224define void @test7() {225; CHECK-LABEL: @test7(226; CHECK-NEXT: call void @bar(ptr nonnull @H) #[[ATTR3]]227; CHECK-NEXT: ret void228;229 %A = alloca %U, align 16230 call void @llvm.memcpy.p0.p0.i64(ptr align 4 %A, ptr align 4 @H, i64 20, i1 false)231 call void @bar(ptr %A) readonly232 ret void233}234 235define void @test8() {236; CHECK-LABEL: @test8(237; CHECK-NEXT: [[AL:%.*]] = alloca [[U:%.*]], align 16238; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr noundef nonnull align 16 dereferenceable(20) [[AL]], ptr noundef nonnull align 4 dereferenceable(20) getelementptr inbounds nuw (i8, ptr @H, i64 20), i64 20, i1 false)239; CHECK-NEXT: call void @bar(ptr nonnull [[AL]]) #[[ATTR3]]240; CHECK-NEXT: ret void241;242 %al = alloca %U, align 16243 call void @llvm.memcpy.p0.p0.i64(ptr align 4 %al, ptr align 4 getelementptr ([2 x %U], ptr @H, i64 0, i32 1), i64 20, i1 false)244 call void @bar(ptr %al) readonly245 ret void246}247 248 249define void @test8_addrspacecast() {250; CHECK-LABEL: @test8_addrspacecast(251; CHECK-NEXT: [[AL:%.*]] = alloca [[U:%.*]], align 16252; CHECK-NEXT: call void @llvm.memcpy.p0.p1.i64(ptr noundef nonnull align 16 dereferenceable(20) [[AL]], ptr addrspace(1) noundef align 4 dereferenceable(20) addrspacecast (ptr getelementptr inbounds nuw (i8, ptr @H, i64 20) to ptr addrspace(1)), i64 20, i1 false)253; CHECK-NEXT: call void @bar(ptr nonnull [[AL]]) #[[ATTR3]]254; CHECK-NEXT: ret void255;256 %Al = alloca %U, align 16257 call void @llvm.memcpy.p0.p1.i64(ptr align 4 %Al, ptr addrspace(1) align 4 addrspacecast (ptr getelementptr ([2 x %U], ptr @H, i64 0, i32 1) to ptr addrspace(1)), i64 20, i1 false)258 call void @bar(ptr %Al) readonly259 ret void260}261 262define void @test9() {263; CHECK-LABEL: @test9(264; CHECK-NEXT: call void @bar(ptr nonnull getelementptr inbounds nuw (i8, ptr @H, i64 20)) #[[ATTR3]]265; CHECK-NEXT: ret void266;267 %A = alloca %U, align 4268 call void @llvm.memcpy.p0.p0.i64(ptr align 4 %A, ptr align 4 getelementptr ([2 x %U], ptr @H, i64 0, i32 1), i64 20, i1 false)269 call void @bar(ptr %A) readonly270 ret void271}272 273define void @test9_addrspacecast() {274; CHECK-LABEL: @test9_addrspacecast(275; CHECK-NEXT: call void @bar(ptr nonnull getelementptr inbounds nuw (i8, ptr @H, i64 20)) #[[ATTR3]]276; CHECK-NEXT: ret void277;278 %A = alloca %U, align 4279 call void @llvm.memcpy.p0.p1.i64(ptr align 4 %A, ptr addrspace(1) align 4 addrspacecast (ptr getelementptr ([2 x %U], ptr @H, i64 0, i32 1) to ptr addrspace(1)), i64 20, i1 false)280 call void @bar(ptr %A) readonly281 ret void282}283 284@bbb = local_unnamed_addr global [1000000 x i8] zeroinitializer, align 16285@_ZL3KKK = internal unnamed_addr constant [3 x i8] c"\01\01\02", align 1286 287; Should not replace alloca with global because of size mismatch.288define void @test9_small_global() {289; CHECK-LABEL: @test9_small_global(290; CHECK-NEXT: entry:291; CHECK-NEXT: [[CC:%.*]] = alloca [1000000 x i8], align 16292; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr noundef nonnull align 16 dereferenceable(3) [[CC]], ptr noundef nonnull align 16 dereferenceable(3) @_ZL3KKK, i64 3, i1 false)293; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr noundef nonnull align 16 dereferenceable(1000000) @bbb, ptr noundef nonnull align 16 dereferenceable(1000000) [[CC]], i64 1000000, i1 false)294; CHECK-NEXT: ret void295;296entry:297 %cc = alloca [1000000 x i8], align 16298 call void @llvm.memcpy.p0.p0.i64(ptr %cc, ptr @_ZL3KKK, i64 3, i1 false)299 call void @llvm.memcpy.p0.p0.i64(ptr align 16 @bbb, ptr align 16 %cc, i64 1000000, i1 false)300 ret void301}302 303; Should replace alloca with global as they have exactly the same size.304define void @test10_same_global() {305; CHECK-LABEL: @test10_same_global(306; CHECK-NEXT: entry:307; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr noundef nonnull align 16 dereferenceable(3) @bbb, ptr noundef nonnull align 16 dereferenceable(3) @_ZL3KKK, i64 3, i1 false)308; CHECK-NEXT: ret void309;310entry:311 %cc = alloca [3 x i8], align 1312 call void @llvm.memcpy.p0.p0.i64(ptr %cc, ptr @_ZL3KKK, i64 3, i1 false)313 call void @llvm.memcpy.p0.p0.i64(ptr @bbb, ptr %cc, i64 3, i1 false)314 ret void315}316 317; Should replace alloca with global even when the global is in a different address space318define float @test11(i64 %i) {319; CHECK-LABEL: @test11(320; CHECK-NEXT: entry:321; CHECK-NEXT: ret float 0.000000e+00322;323 324entry:325 %a = alloca [4 x float], align 4326 call void @llvm.lifetime.start.p0(ptr %a)327 call void @llvm.memcpy.p0.p1.i64(ptr align 4 %a, ptr addrspace(1) align 4 @I, i64 16, i1 false)328 %g = getelementptr inbounds [4 x float], ptr %a, i64 0, i64 %i329 %r = load float, ptr %g, align 4330 ret float %r331}332 333; If the memcpy is volatile, it should not be removed334define float @test11_volatile(i64 %i) {335; CHECK-LABEL: @test11_volatile(336; CHECK-NEXT: entry:337; CHECK-NEXT: [[A:%.*]] = alloca [4 x float], align 4338; CHECK-NEXT: call void @llvm.lifetime.start.p0(ptr nonnull [[A]])339; CHECK-NEXT: call void @llvm.memcpy.p0.p1.i64(ptr align 4 [[A]], ptr addrspace(1) align 4 @I, i64 16, i1 true)340; CHECK-NEXT: [[G:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[I:%.*]]341; CHECK-NEXT: [[R:%.*]] = load float, ptr [[G]], align 4342; CHECK-NEXT: ret float [[R]]343;344 345entry:346 %a = alloca [4 x float], align 4347 call void @llvm.lifetime.start.p0(ptr %a)348 call void @llvm.memcpy.p0.p1.i64(ptr align 4 %a, ptr addrspace(1) align 4 @I, i64 16, i1 true)349 %g = getelementptr inbounds [4 x float], ptr %a, i64 0, i64 %i350 %r = load float, ptr %g, align 4351 ret float %r352}353 354; Tests that we can eliminate allocas copied from readonly noalias pointers.355define void @memcpy_from_readonly_noalias(ptr readonly noalias align 8 dereferenceable(124) %arg) {356; CHECK-LABEL: @memcpy_from_readonly_noalias(357; CHECK-NEXT: call void @bar(ptr nonnull [[ARG:%.*]]) #[[ATTR3]]358; CHECK-NEXT: ret void359;360 %alloca = alloca %T, align 8361 call void @llvm.memcpy.p0.p0.i64(ptr %alloca, ptr %arg, i64 124, i1 false)362 call void @bar(ptr %alloca) readonly363 ret void364}365 366; Tests that we don't eliminate allocas copied from readonly pointers without noalias.367define void @memcpy_from_just_readonly(ptr readonly align 8 dereferenceable(124) %arg) {368; CHECK-LABEL: @memcpy_from_just_readonly(369; CHECK-NEXT: [[ALLOCA:%.*]] = alloca [[T:%.*]], align 8370; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr noundef nonnull align 8 dereferenceable(124) [[ALLOCA]], ptr noundef nonnull align 8 dereferenceable(124) [[ARG:%.*]], i64 124, i1 false)371; CHECK-NEXT: call void @bar(ptr nonnull [[ALLOCA]]) #[[ATTR3]]372; CHECK-NEXT: ret void373;374 %alloca = alloca %T, align 8375 call void @llvm.memcpy.p0.p0.i64(ptr %alloca, ptr %arg, i64 124, i1 false)376 call void @bar(ptr %alloca) readonly377 ret void378}379 380; Test that we don't elide a volatile memcpy.381define void @volatile_memcpy() {382; CHECK-LABEL: @volatile_memcpy(383; CHECK-NEXT: [[A:%.*]] = alloca [[U:%.*]], align 16384; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 4 [[A]], ptr align 4 @H, i64 20, i1 true)385; CHECK-NEXT: call void @bar(ptr nonnull [[A]]) #[[ATTR3]]386; CHECK-NEXT: ret void387;388 %A = alloca %U, align 16389 call void @llvm.memcpy.p0.p0.i64(ptr align 4 %A, ptr align 4 @H, i64 20, i1 true)390 call void @bar(ptr %A) readonly391 ret void392}393 394; Test that we can elide a memcpy when copying a constant value onto the stack395; and then forwarding it by readonly nocapture reference.396define void @memcpy_to_nocapture_readonly() {397; CHECK-LABEL: @memcpy_to_nocapture_readonly(398; CHECK-NEXT: call void @bar(ptr nonnull readonly captures(none) @H)399; CHECK-NEXT: ret void400;401 %A = alloca %U, align 16402 call void @llvm.memcpy.p0.p0.i64(ptr align 4 %A, ptr align 4 @H, i64 20, i1 false)403 call void @bar(ptr nocapture readonly %A)404 ret void405}406 407; Test that we don't elide the memcpy when copying a constant value onto the408; stack and then forwarding it by readonly, but capturing, reference.409define void @memcpy_to_capturing_readonly() {410; CHECK-LABEL: @memcpy_to_capturing_readonly(411; CHECK-NEXT: [[A:%.*]] = alloca [[U:%.*]], align 16412; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr noundef nonnull align 16 dereferenceable(20) [[A]], ptr noundef nonnull align 16 dereferenceable(20) @H, i64 20, i1 false)413; CHECK-NEXT: call void @bar_may_capture(ptr nonnull readonly [[A]])414; CHECK-NEXT: ret void415;416 %A = alloca %U, align 16417 call void @llvm.memcpy.p0.p0.i64(ptr align 4 %A, ptr align 4 @H, i64 20, i1 false)418 call void @bar_may_capture(ptr readonly %A)419 ret void420}421 422; Test that we don't elide the memcpy when copying a constant value onto the423; stack and then forwarding it by read-write, nocapture reference, even if it's424; also forwarded by readonly nocapture reference to the same function.425define void @memcpy_to_aliased_nocapture_readonly() {426; CHECK-LABEL: @memcpy_to_aliased_nocapture_readonly(427; CHECK-NEXT: [[A:%.*]] = alloca [[U:%.*]], align 16428; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr noundef nonnull align 16 dereferenceable(20) [[A]], ptr noundef nonnull align 16 dereferenceable(20) @H, i64 20, i1 false)429; CHECK-NEXT: call void @two_params(ptr nonnull readonly captures(none) [[A]], ptr nonnull captures(none) [[A]])430; CHECK-NEXT: ret void431;432 %A = alloca %U, align 16433 call void @llvm.memcpy.p0.p0.i64(ptr align 4 %A, ptr align 4 @H, i64 20, i1 false)434 call void @two_params(ptr nocapture readonly %A, ptr nocapture %A)435 ret void436}437 438declare void @two_params(ptr nocapture readonly, ptr nocapture)439 440attributes #0 = { null_pointer_is_valid }441