1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -basic-aa -memcpyopt -dse -S | FileCheck %s 3 4target 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" 5target triple = "i686-apple-darwin9" 6 7%0 = type { x86_fp80, x86_fp80 } 8%1 = type { i32, i32 } 9 10define void @test1(%0* sret %agg.result, x86_fp80 %z.0, x86_fp80 %z.1) nounwind { 11; CHECK-LABEL: @test1( 12; CHECK-NEXT: entry: 13; CHECK-NEXT: [[TMP2:%.*]] = alloca [[TMP0:%.*]], align 16 14; CHECK-NEXT: [[TMP5:%.*]] = fsub x86_fp80 0xK80000000000000000000, [[Z_1:%.*]] 15; CHECK-NEXT: call void @ccoshl(%0* sret [[TMP2]], x86_fp80 [[TMP5]], x86_fp80 [[Z_0:%.*]]) [[ATTR0:#.*]] 16; CHECK-NEXT: [[TMP219:%.*]] = bitcast %0* [[TMP2]] to i8* 17; CHECK-NEXT: [[AGG_RESULT21:%.*]] = bitcast %0* [[AGG_RESULT:%.*]] to i8* 18; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 [[AGG_RESULT21]], i8* align 16 [[TMP219]], i32 32, i1 false) 19; CHECK-NEXT: ret void 20; 21entry: 22 %tmp2 = alloca %0 23 %memtmp = alloca %0, align 16 24 %tmp5 = fsub x86_fp80 0xK80000000000000000000, %z.1 25 call void @ccoshl(%0* sret %memtmp, x86_fp80 %tmp5, x86_fp80 %z.0) nounwind 26 %tmp219 = bitcast %0* %tmp2 to i8* 27 %memtmp20 = bitcast %0* %memtmp to i8* 28 call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 %tmp219, i8* align 16 %memtmp20, i32 32, i1 false) 29 %agg.result21 = bitcast %0* %agg.result to i8* 30 call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 %agg.result21, i8* align 16 %tmp219, i32 32, i1 false) 31 ret void 32 33; Check that one of the memcpy's are removed. 34;; FIXME: PR 8643 We should be able to eliminate the last memcpy here. 35 36} 37 38declare void @ccoshl(%0* nocapture sret, x86_fp80, x86_fp80) nounwind 39 40 41; The intermediate alloca and one of the memcpy's should be eliminated, the 42; other should be related with a memmove. 43define void @test2(i8* %P, i8* %Q) nounwind { 44; CHECK-LABEL: @test2( 45; CHECK-NEXT: call void @llvm.memmove.p0i8.p0i8.i32(i8* align 16 [[Q:%.*]], i8* align 16 [[P:%.*]], i32 32, i1 false) 46; CHECK-NEXT: ret void 47; 48 %memtmp = alloca %0, align 16 49 %R = bitcast %0* %memtmp to i8* 50 call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 %R, i8* align 16 %P, i32 32, i1 false) 51 call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 %Q, i8* align 16 %R, i32 32, i1 false) 52 ret void 53 54} 55 56; The intermediate alloca and one of the memcpy's should be eliminated, the 57; other should be related with a memcpy. 58define void @test2_memcpy(i8* noalias %P, i8* noalias %Q) nounwind { 59; CHECK-LABEL: @test2_memcpy( 60; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 [[Q:%.*]], i8* align 16 [[P:%.*]], i32 32, i1 false) 61; CHECK-NEXT: ret void 62; 63 %memtmp = alloca %0, align 16 64 %R = bitcast %0* %memtmp to i8* 65 call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 %R, i8* align 16 %P, i32 32, i1 false) 66 call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 %Q, i8* align 16 %R, i32 32, i1 false) 67 ret void 68 69} 70 71 72 73 74@x = external global %0 75 76define void @test3(%0* noalias sret %agg.result) nounwind { 77; CHECK-LABEL: @test3( 78; CHECK-NEXT: [[AGG_RESULT1:%.*]] = bitcast %0* [[AGG_RESULT:%.*]] to i8* 79; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 [[AGG_RESULT1]], i8* align 16 bitcast (%0* @x to i8*), i32 32, i1 false) 80; CHECK-NEXT: ret void 81; 82 %x.0 = alloca %0 83 %x.01 = bitcast %0* %x.0 to i8* 84 call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 %x.01, i8* align 16 bitcast (%0* @x to i8*), i32 32, i1 false) 85 %agg.result2 = bitcast %0* %agg.result to i8* 86 call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 %agg.result2, i8* align 16 %x.01, i32 32, i1 false) 87 ret void 88} 89 90 91; PR8644 92define void @test4(i8 *%P) { 93; CHECK-LABEL: @test4( 94; CHECK-NEXT: call void @test4a(i8* byval align 1 [[P:%.*]]) 95; CHECK-NEXT: ret void 96; 97 %A = alloca %1 98 %a = bitcast %1* %A to i8* 99 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %a, i8* align 4 %P, i64 8, i1 false) 100 call void @test4a(i8* align 1 byval %a) 101 ret void 102} 103 104; Make sure we don't remove the memcpy if the source address space doesn't match the byval argument 105define void @test4_addrspace(i8 addrspace(1)* %P) { 106; CHECK-LABEL: @test4_addrspace( 107; CHECK-NEXT: [[A1:%.*]] = alloca [[TMP1:%.*]], align 8 108; CHECK-NEXT: [[A2:%.*]] = bitcast %1* [[A1]] to i8* 109; CHECK-NEXT: call void @llvm.memcpy.p0i8.p1i8.i64(i8* align 4 [[A2]], i8 addrspace(1)* align 4 [[P:%.*]], i64 8, i1 false) 110; CHECK-NEXT: call void @test4a(i8* byval align 1 [[A2]]) 111; CHECK-NEXT: ret void 112; 113 %a1 = alloca %1 114 %a2 = bitcast %1* %a1 to i8* 115 call void @llvm.memcpy.p0i8.p1i8.i64(i8* align 4 %a2, i8 addrspace(1)* align 4 %P, i64 8, i1 false) 116 call void @test4a(i8* align 1 byval %a2) 117 ret void 118} 119 120declare void @test4a(i8* align 1 byval) 121declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture, i64, i1) nounwind 122declare void @llvm.memcpy.p0i8.p1i8.i64(i8* nocapture, i8 addrspace(1)* nocapture, i64, i1) nounwind 123declare void @llvm.memcpy.p1i8.p1i8.i64(i8 addrspace(1)* nocapture, i8 addrspace(1)* nocapture, i64, i1) nounwind 124 125%struct.S = type { i128, [4 x i8]} 126 127@sS = external global %struct.S, align 16 128 129declare void @test5a(%struct.S* align 16 byval) nounwind ssp 130 131 132; rdar://8713376 - This memcpy can't be eliminated. 133define i32 @test5(i32 %x) nounwind ssp { 134; CHECK-LABEL: @test5( 135; CHECK-NEXT: entry: 136; CHECK-NEXT: [[Y:%.*]] = alloca [[STRUCT_S:%.*]], align 16 137; CHECK-NEXT: [[TMP:%.*]] = bitcast %struct.S* [[Y]] to i8* 138; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 16 [[TMP]], i8* align 16 bitcast (%struct.S* @sS to i8*), i64 32, i1 false) 139; CHECK-NEXT: [[A:%.*]] = getelementptr [[STRUCT_S]], %struct.S* [[Y]], i64 0, i32 1, i64 0 140; CHECK-NEXT: store i8 4, i8* [[A]], align 1 141; CHECK-NEXT: call void @test5a(%struct.S* byval align 16 [[Y]]) 142; CHECK-NEXT: ret i32 0 143; 144entry: 145 %y = alloca %struct.S, align 16 146 %tmp = bitcast %struct.S* %y to i8* 147 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 16 %tmp, i8* align 16 bitcast (%struct.S* @sS to i8*), i64 32, i1 false) 148 %a = getelementptr %struct.S, %struct.S* %y, i64 0, i32 1, i64 0 149 store i8 4, i8* %a 150 call void @test5a(%struct.S* align 16 byval %y) 151 ret i32 0 152} 153 154;; Noop memcpy should be zapped. 155define void @test6(i8 *%P) { 156; CHECK-LABEL: @test6( 157; CHECK-NEXT: ret void 158; 159 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %P, i8* align 4 %P, i64 8, i1 false) 160 ret void 161} 162 163 164; PR9794 - Should forward memcpy into byval argument even though the memcpy 165; isn't itself 8 byte aligned. 166%struct.p = type { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } 167 168define i32 @test7(%struct.p* nocapture align 8 byval %q) nounwind ssp { 169; CHECK-LABEL: @test7( 170; CHECK-NEXT: entry: 171; CHECK-NEXT: [[CALL:%.*]] = call i32 @g(%struct.p* byval align 8 [[Q:%.*]]) [[ATTR0]] 172; CHECK-NEXT: ret i32 [[CALL]] 173; 174entry: 175 %agg.tmp = alloca %struct.p, align 4 176 %tmp = bitcast %struct.p* %agg.tmp to i8* 177 %tmp1 = bitcast %struct.p* %q to i8* 178 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %tmp, i8* align 4 %tmp1, i64 48, i1 false) 179 %call = call i32 @g(%struct.p* align 8 byval %agg.tmp) nounwind 180 ret i32 %call 181} 182 183declare i32 @g(%struct.p* align 8 byval) 184 185declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture, i8* nocapture, i32, i1) nounwind 186 187; PR11142 - When looking for a memcpy-memcpy dependency, don't get stuck on 188; instructions between the memcpy's that only affect the destination pointer. 189@test8.str = internal constant [7 x i8] c"ABCDEF\00" 190 191define void @test8() { 192; CHECK-LABEL: @test8( 193; CHECK-NEXT: ret void 194; 195 %A = tail call i8* @malloc(i32 10) 196 %B = getelementptr inbounds i8, i8* %A, i64 2 197 tail call void @llvm.memcpy.p0i8.p0i8.i32(i8* %B, i8* getelementptr inbounds ([7 x i8], [7 x i8]* @test8.str, i64 0, i64 0), i32 7, i1 false) 198 %C = tail call i8* @malloc(i32 10) 199 %D = getelementptr inbounds i8, i8* %C, i64 2 200 tail call void @llvm.memcpy.p0i8.p0i8.i32(i8* %D, i8* %B, i32 7, i1 false) 201 ret void 202} 203 204declare noalias i8* @malloc(i32) 205 206; rdar://11341081 207%struct.big = type { [50 x i32] } 208 209define void @test9_addrspacecast() nounwind ssp uwtable { 210; CHECK-LABEL: @test9_addrspacecast( 211; CHECK-NEXT: entry: 212; CHECK-NEXT: [[B:%.*]] = alloca [[STRUCT_BIG:%.*]], align 4 213; CHECK-NEXT: call void @f1(%struct.big* sret [[B]]) 214; CHECK-NEXT: call void @f2(%struct.big* [[B]]) 215; CHECK-NEXT: ret void 216; 217entry: 218 %b = alloca %struct.big, align 4 219 %tmp = alloca %struct.big, align 4 220 call void @f1(%struct.big* sret %tmp) 221 %0 = addrspacecast %struct.big* %b to i8 addrspace(1)* 222 %1 = addrspacecast %struct.big* %tmp to i8 addrspace(1)* 223 call void @llvm.memcpy.p1i8.p1i8.i64(i8 addrspace(1)* align 4 %0, i8 addrspace(1)* align 4 %1, i64 200, i1 false) 224 call void @f2(%struct.big* %b) 225 ret void 226} 227 228define void @test9() nounwind ssp uwtable { 229; CHECK-LABEL: @test9( 230; CHECK-NEXT: entry: 231; CHECK-NEXT: [[B:%.*]] = alloca [[STRUCT_BIG:%.*]], align 4 232; CHECK-NEXT: call void @f1(%struct.big* sret [[B]]) 233; CHECK-NEXT: call void @f2(%struct.big* [[B]]) 234; CHECK-NEXT: ret void 235; 236entry: 237 %b = alloca %struct.big, align 4 238 %tmp = alloca %struct.big, align 4 239 call void @f1(%struct.big* sret %tmp) 240 %0 = bitcast %struct.big* %b to i8* 241 %1 = bitcast %struct.big* %tmp to i8* 242 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %0, i8* align 4 %1, i64 200, i1 false) 243 call void @f2(%struct.big* %b) 244 ret void 245} 246 247; rdar://14073661. 248; Test10 triggered assertion when the compiler try to get the size of the 249; opaque type of *x, where the x is the formal argument with attribute 'sret'. 250 251%opaque = type opaque 252declare void @foo(i32* noalias nocapture) 253 254define void @test10(%opaque* noalias nocapture sret %x, i32 %y) { 255; CHECK-LABEL: @test10( 256; CHECK-NEXT: [[A:%.*]] = alloca i32, align 4 257; CHECK-NEXT: store i32 [[Y:%.*]], i32* [[A]], align 4 258; CHECK-NEXT: call void @foo(i32* noalias nocapture [[A]]) 259; CHECK-NEXT: [[C:%.*]] = load i32, i32* [[A]], align 4 260; CHECK-NEXT: [[D:%.*]] = bitcast %opaque* [[X:%.*]] to i32* 261; CHECK-NEXT: store i32 [[C]], i32* [[D]], align 4 262; CHECK-NEXT: ret void 263; 264 %a = alloca i32, align 4 265 store i32 %y, i32* %a 266 call void @foo(i32* noalias nocapture %a) 267 %c = load i32, i32* %a 268 %d = bitcast %opaque* %x to i32* 269 store i32 %c, i32* %d 270 ret void 271} 272 273; don't create new addressspacecasts when we don't know they're safe for the target 274define void @test11([20 x i32] addrspace(1)* nocapture dereferenceable(80) %P) { 275; CHECK-LABEL: @test11( 276; CHECK-NEXT: [[B:%.*]] = bitcast [20 x i32] addrspace(1)* [[P:%.*]] to i8 addrspace(1)* 277; CHECK-NEXT: call void @llvm.memset.p1i8.i64(i8 addrspace(1)* align 4 [[B]], i8 0, i64 80, i1 false) 278; CHECK-NEXT: ret void 279; 280 %A = alloca [20 x i32], align 4 281 %a = bitcast [20 x i32]* %A to i8* 282 %b = bitcast [20 x i32] addrspace(1)* %P to i8 addrspace(1)* 283 call void @llvm.memset.p0i8.i64(i8* align 4 %a, i8 0, i64 80, i1 false) 284 call void @llvm.memcpy.p1i8.p0i8.i64(i8 addrspace(1)* align 4 %b, i8* align 4 %a, i64 80, i1 false) 285 ret void 286} 287 288declare void @llvm.memset.p0i8.i64(i8* nocapture, i8, i64, i1) nounwind 289declare void @llvm.memcpy.p1i8.p0i8.i64(i8 addrspace(1)* nocapture, i8* nocapture, i64, i1) nounwind 290 291declare void @f1(%struct.big* nocapture sret) 292declare void @f2(%struct.big*) 293 294; CHECK: attributes [[ATTR0]] = { nounwind } 295; CHECK: attributes #1 = { argmemonly nounwind willreturn } 296; CHECK: attributes #2 = { nounwind ssp } 297; CHECK: attributes #3 = { nounwind ssp uwtable } 298