1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -basic-aa -memcpyopt -dse -S -enable-memcpyopt-memoryssa=0 | FileCheck %s 3; RUN: opt < %s -basic-aa -memcpyopt -dse -S -enable-memcpyopt-memoryssa=1 -verify-memoryssa | FileCheck %s 4 5target 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" 6target triple = "i686-apple-darwin9" 7 8%0 = type { x86_fp80, x86_fp80 } 9%1 = type { i32, i32 } 10 11define void @test1(%0* sret(%0) %agg.result, x86_fp80 %z.0, x86_fp80 %z.1) nounwind { 12; CHECK-LABEL: @test1( 13; CHECK-NEXT: entry: 14; CHECK-NEXT: [[TMP2:%.*]] = alloca [[TMP0:%.*]], align 16 15; CHECK-NEXT: [[MEMTMP:%.*]] = alloca [[TMP0]], align 16 16; CHECK-NEXT: [[TMP5:%.*]] = fsub x86_fp80 0xK80000000000000000000, [[Z_1:%.*]] 17; CHECK-NEXT: call void @ccoshl(%0* sret([[TMP0]]) [[TMP2]], x86_fp80 [[TMP5]], x86_fp80 [[Z_0:%.*]]) #[[ATTR0:[0-9]+]] 18; CHECK-NEXT: [[TMP219:%.*]] = bitcast %0* [[TMP2]] to i8* 19; CHECK-NEXT: [[MEMTMP20:%.*]] = bitcast %0* [[MEMTMP]] to i8* 20; CHECK-NEXT: [[AGG_RESULT21:%.*]] = bitcast %0* [[AGG_RESULT:%.*]] to i8* 21; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 [[AGG_RESULT21]], i8* align 16 [[TMP219]], i32 32, i1 false) 22; CHECK-NEXT: ret void 23; 24entry: 25 %tmp2 = alloca %0 26 %memtmp = alloca %0, align 16 27 %tmp5 = fsub x86_fp80 0xK80000000000000000000, %z.1 28 call void @ccoshl(%0* sret(%0) %memtmp, x86_fp80 %tmp5, x86_fp80 %z.0) nounwind 29 %tmp219 = bitcast %0* %tmp2 to i8* 30 %memtmp20 = bitcast %0* %memtmp to i8* 31 call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 %tmp219, i8* align 16 %memtmp20, i32 32, i1 false) 32 %agg.result21 = bitcast %0* %agg.result to i8* 33 call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 %agg.result21, i8* align 16 %tmp219, i32 32, i1 false) 34 ret void 35 36; Check that one of the memcpy's are removed. 37;; FIXME: PR 8643 We should be able to eliminate the last memcpy here. 38 39} 40 41declare void @ccoshl(%0* nocapture sret(%0), x86_fp80, x86_fp80) nounwind 42 43 44; The intermediate alloca and one of the memcpy's should be eliminated, the 45; other should be related with a memmove. 46define void @test2(i8* %P, i8* %Q) nounwind { 47; CHECK-LABEL: @test2( 48; CHECK-NEXT: call void @llvm.memmove.p0i8.p0i8.i32(i8* align 16 [[Q:%.*]], i8* align 16 [[P:%.*]], i32 32, i1 false) 49; CHECK-NEXT: ret void 50; 51 %memtmp = alloca %0, align 16 52 %R = bitcast %0* %memtmp to i8* 53 call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 %R, i8* align 16 %P, i32 32, i1 false) 54 call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 %Q, i8* align 16 %R, i32 32, i1 false) 55 ret void 56 57} 58 59; The intermediate alloca and one of the memcpy's should be eliminated, the 60; other should be related with a memcpy. 61define void @test2_memcpy(i8* noalias %P, i8* noalias %Q) nounwind { 62; CHECK-LABEL: @test2_memcpy( 63; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 [[Q:%.*]], i8* align 16 [[P:%.*]], i32 32, i1 false) 64; CHECK-NEXT: ret void 65; 66 %memtmp = alloca %0, align 16 67 %R = bitcast %0* %memtmp to i8* 68 call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 %R, i8* align 16 %P, i32 32, i1 false) 69 call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 %Q, i8* align 16 %R, i32 32, i1 false) 70 ret void 71 72} 73 74 75 76 77@x = external global %0 78 79define void @test3(%0* noalias sret(%0) %agg.result) nounwind { 80; CHECK-LABEL: @test3( 81; CHECK-NEXT: [[X_0:%.*]] = alloca [[TMP0:%.*]], align 16 82; CHECK-NEXT: [[X_01:%.*]] = bitcast %0* [[X_0]] to i8* 83; CHECK-NEXT: [[AGG_RESULT1:%.*]] = bitcast %0* [[AGG_RESULT:%.*]] to i8* 84; 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) 85; CHECK-NEXT: [[AGG_RESULT2:%.*]] = bitcast %0* [[AGG_RESULT]] to i8* 86; CHECK-NEXT: ret void 87; 88 %x.0 = alloca %0 89 %x.01 = bitcast %0* %x.0 to i8* 90 call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 %x.01, i8* align 16 bitcast (%0* @x to i8*), i32 32, i1 false) 91 %agg.result2 = bitcast %0* %agg.result to i8* 92 call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 %agg.result2, i8* align 16 %x.01, i32 32, i1 false) 93 ret void 94} 95 96 97; PR8644 98define void @test4(i8 *%P) { 99; CHECK-LABEL: @test4( 100; CHECK-NEXT: call void @test4a(i8* byval(i8) align 1 [[P:%.*]]) 101; CHECK-NEXT: ret void 102; 103 %A = alloca %1 104 %a = bitcast %1* %A to i8* 105 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %a, i8* align 4 %P, i64 8, i1 false) 106 call void @test4a(i8* align 1 byval(i8) %a) 107 ret void 108} 109 110; Make sure we don't remove the memcpy if the source address space doesn't match the byval argument 111define void @test4_addrspace(i8 addrspace(1)* %P) { 112; CHECK-LABEL: @test4_addrspace( 113; CHECK-NEXT: [[A1:%.*]] = alloca [[TMP1:%.*]], align 8 114; CHECK-NEXT: [[A2:%.*]] = bitcast %1* [[A1]] to i8* 115; CHECK-NEXT: call void @llvm.memcpy.p0i8.p1i8.i64(i8* align 4 [[A2]], i8 addrspace(1)* align 4 [[P:%.*]], i64 8, i1 false) 116; CHECK-NEXT: call void @test4a(i8* byval(i8) align 1 [[A2]]) 117; CHECK-NEXT: ret void 118; 119 %a1 = alloca %1 120 %a2 = bitcast %1* %a1 to i8* 121 call void @llvm.memcpy.p0i8.p1i8.i64(i8* align 4 %a2, i8 addrspace(1)* align 4 %P, i64 8, i1 false) 122 call void @test4a(i8* align 1 byval(i8) %a2) 123 ret void 124} 125 126define void @test4_write_between(i8 *%P) { 127; CHECK-LABEL: @test4_write_between( 128; CHECK-NEXT: [[A1:%.*]] = alloca [[TMP1:%.*]], align 8 129; CHECK-NEXT: [[A2:%.*]] = bitcast %1* [[A1]] to i8* 130; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[A2]], i8* align 4 [[P:%.*]], i64 8, i1 false) 131; CHECK-NEXT: store i8 0, i8* [[A2]], align 1 132; CHECK-NEXT: call void @test4a(i8* byval(i8) align 1 [[A2]]) 133; CHECK-NEXT: ret void 134; 135 %a1 = alloca %1 136 %a2 = bitcast %1* %a1 to i8* 137 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %a2, i8* align 4 %P, i64 8, i1 false) 138 store i8 0, i8* %a2 139 call void @test4a(i8* align 1 byval(i8) %a2) 140 ret void 141} 142 143define i8 @test4_read_between(i8 *%P) { 144; NO_MSSA-LABEL: @test4_read_between( 145; NO_MSSA-NEXT: [[A1:%.*]] = alloca [[TMP1:%.*]], align 8 146; NO_MSSA-NEXT: [[A2:%.*]] = bitcast %1* [[A1]] to i8* 147; NO_MSSA-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[A2]], i8* align 4 [[P:%.*]], i64 8, i1 false) 148; NO_MSSA-NEXT: [[X:%.*]] = load i8, i8* [[A2]], align 1 149; NO_MSSA-NEXT: call void @test4a(i8* byval align 1 [[A2]]) 150; NO_MSSA-NEXT: ret i8 [[X]] 151; 152; MSSA-LABEL: @test4_read_between( 153; MSSA-NEXT: [[A1:%.*]] = alloca [[TMP1:%.*]], align 8 154; MSSA-NEXT: [[A2:%.*]] = bitcast %1* [[A1]] to i8* 155; MSSA-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[A2]], i8* align 4 [[P:%.*]], i64 8, i1 false) 156; MSSA-NEXT: [[X:%.*]] = load i8, i8* [[A2]], align 1 157; MSSA-NEXT: call void @test4a(i8* byval align 1 [[P]]) 158; MSSA-NEXT: ret i8 [[X]] 159; 160 %a1 = alloca %1 161 %a2 = bitcast %1* %a1 to i8* 162 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %a2, i8* align 4 %P, i64 8, i1 false) 163 %x = load i8, i8* %a2 164 call void @test4a(i8* align 1 byval(i8) %a2) 165 ret i8 %x 166} 167 168define void @test4_non_local(i8 *%P, i1 %c) { 169; NO_MSSA-LABEL: @test4_non_local( 170; NO_MSSA-NEXT: [[A1:%.*]] = alloca [[TMP1:%.*]], align 8 171; NO_MSSA-NEXT: [[A2:%.*]] = bitcast %1* [[A1]] to i8* 172; NO_MSSA-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[A2]], i8* align 4 [[P:%.*]], i64 8, i1 false) 173; NO_MSSA-NEXT: br i1 [[C:%.*]], label [[CALL:%.*]], label [[EXIT:%.*]] 174; NO_MSSA: call: 175; NO_MSSA-NEXT: call void @test4a(i8* byval align 1 [[A2]]) 176; NO_MSSA-NEXT: br label [[EXIT]] 177; NO_MSSA: exit: 178; NO_MSSA-NEXT: ret void 179; 180; MSSA-LABEL: @test4_non_local( 181; MSSA-NEXT: [[A1:%.*]] = alloca [[TMP1:%.*]], align 8 182; MSSA-NEXT: [[A2:%.*]] = bitcast %1* [[A1]] to i8* 183; MSSA-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[A2]], i8* align 4 [[P:%.*]], i64 8, i1 false) 184; MSSA-NEXT: br i1 [[C:%.*]], label [[CALL:%.*]], label [[EXIT:%.*]] 185; MSSA: call: 186; MSSA-NEXT: call void @test4a(i8* byval align 1 [[P]]) 187; MSSA-NEXT: br label [[EXIT]] 188; MSSA: exit: 189; MSSA-NEXT: ret void 190; 191 %a1 = alloca %1 192 %a2 = bitcast %1* %a1 to i8* 193 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %a2, i8* align 4 %P, i64 8, i1 false) 194 br i1 %c, label %call, label %exit 195 196call: 197 call void @test4a(i8* align 1 byval(i8) %a2) 198 br label %exit 199 200exit: 201 ret void 202} 203 204declare void @test4a(i8* align 1 byval(i8)) 205declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture, i64, i1) nounwind 206declare void @llvm.memcpy.p0i8.p1i8.i64(i8* nocapture, i8 addrspace(1)* nocapture, i64, i1) nounwind 207declare void @llvm.memcpy.p1i8.p1i8.i64(i8 addrspace(1)* nocapture, i8 addrspace(1)* nocapture, i64, i1) nounwind 208 209%struct.S = type { i128, [4 x i8]} 210 211@sS = external global %struct.S, align 16 212 213declare void @test5a(%struct.S* align 16 byval(%struct.S)) nounwind ssp 214 215 216; rdar://8713376 - This memcpy can't be eliminated. 217define i32 @test5(i32 %x) nounwind ssp { 218; CHECK-LABEL: @test5( 219; CHECK-NEXT: entry: 220; CHECK-NEXT: [[Y:%.*]] = alloca [[STRUCT_S:%.*]], align 16 221; CHECK-NEXT: [[TMP:%.*]] = bitcast %struct.S* [[Y]] to i8* 222; 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) 223; CHECK-NEXT: [[A:%.*]] = getelementptr [[STRUCT_S]], %struct.S* [[Y]], i64 0, i32 1, i64 0 224; CHECK-NEXT: store i8 4, i8* [[A]], align 1 225; CHECK-NEXT: call void @test5a(%struct.S* byval([[STRUCT_S]]) align 16 [[Y]]) 226; CHECK-NEXT: ret i32 0 227; 228entry: 229 %y = alloca %struct.S, align 16 230 %tmp = bitcast %struct.S* %y to i8* 231 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 16 %tmp, i8* align 16 bitcast (%struct.S* @sS to i8*), i64 32, i1 false) 232 %a = getelementptr %struct.S, %struct.S* %y, i64 0, i32 1, i64 0 233 store i8 4, i8* %a 234 call void @test5a(%struct.S* align 16 byval(%struct.S) %y) 235 ret i32 0 236} 237 238;; Noop memcpy should be zapped. 239define void @test6(i8 *%P) { 240; CHECK-LABEL: @test6( 241; CHECK-NEXT: ret void 242; 243 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %P, i8* align 4 %P, i64 8, i1 false) 244 ret void 245} 246 247 248; PR9794 - Should forward memcpy into byval argument even though the memcpy 249; isn't itself 8 byte aligned. 250%struct.p = type { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } 251 252define i32 @test7(%struct.p* nocapture align 8 byval(%struct.p) %q) nounwind ssp { 253; CHECK-LABEL: @test7( 254; CHECK-NEXT: entry: 255; CHECK-NEXT: [[CALL:%.*]] = call i32 @g(%struct.p* byval([[STRUCT_P:%.*]]) align 8 [[Q:%.*]]) #[[ATTR0]] 256; CHECK-NEXT: ret i32 [[CALL]] 257; 258entry: 259 %agg.tmp = alloca %struct.p, align 4 260 %tmp = bitcast %struct.p* %agg.tmp to i8* 261 %tmp1 = bitcast %struct.p* %q to i8* 262 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %tmp, i8* align 4 %tmp1, i64 48, i1 false) 263 %call = call i32 @g(%struct.p* align 8 byval(%struct.p) %agg.tmp) nounwind 264 ret i32 %call 265} 266 267declare i32 @g(%struct.p* align 8 byval(%struct.p)) 268 269declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture, i8* nocapture, i32, i1) nounwind 270 271; PR11142 - When looking for a memcpy-memcpy dependency, don't get stuck on 272; instructions between the memcpy's that only affect the destination pointer. 273@test8.str = internal constant [7 x i8] c"ABCDEF\00" 274 275define void @test8() { 276; CHECK-LABEL: @test8( 277; CHECK-NEXT: ret void 278; 279 %A = tail call i8* @malloc(i32 10) 280 %B = getelementptr inbounds i8, i8* %A, i64 2 281 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) 282 %C = tail call i8* @malloc(i32 10) 283 %D = getelementptr inbounds i8, i8* %C, i64 2 284 tail call void @llvm.memcpy.p0i8.p0i8.i32(i8* %D, i8* %B, i32 7, i1 false) 285 ret void 286} 287 288declare noalias i8* @malloc(i32) willreturn 289 290; rdar://11341081 291%struct.big = type { [50 x i32] } 292 293define void @test9_addrspacecast() nounwind ssp uwtable { 294; CHECK-LABEL: @test9_addrspacecast( 295; CHECK-NEXT: entry: 296; CHECK-NEXT: [[B:%.*]] = alloca [[STRUCT_BIG:%.*]], align 4 297; CHECK-NEXT: [[TMP:%.*]] = alloca [[STRUCT_BIG]], align 4 298; CHECK-NEXT: call void @f1(%struct.big* sret([[STRUCT_BIG]]) [[B]]) 299; CHECK-NEXT: [[TMP0:%.*]] = addrspacecast %struct.big* [[B]] to i8 addrspace(1)* 300; CHECK-NEXT: [[TMP1:%.*]] = addrspacecast %struct.big* [[TMP]] to i8 addrspace(1)* 301; CHECK-NEXT: call void @f2(%struct.big* [[B]]) 302; CHECK-NEXT: ret void 303; 304entry: 305 %b = alloca %struct.big, align 4 306 %tmp = alloca %struct.big, align 4 307 call void @f1(%struct.big* sret(%struct.big) %tmp) 308 %0 = addrspacecast %struct.big* %b to i8 addrspace(1)* 309 %1 = addrspacecast %struct.big* %tmp to i8 addrspace(1)* 310 call void @llvm.memcpy.p1i8.p1i8.i64(i8 addrspace(1)* align 4 %0, i8 addrspace(1)* align 4 %1, i64 200, i1 false) 311 call void @f2(%struct.big* %b) 312 ret void 313} 314 315define void @test9() nounwind ssp uwtable { 316; CHECK-LABEL: @test9( 317; CHECK-NEXT: entry: 318; CHECK-NEXT: [[B:%.*]] = alloca [[STRUCT_BIG:%.*]], align 4 319; CHECK-NEXT: [[TMP:%.*]] = alloca [[STRUCT_BIG]], align 4 320; CHECK-NEXT: call void @f1(%struct.big* sret([[STRUCT_BIG]]) [[B]]) 321; CHECK-NEXT: [[TMP0:%.*]] = bitcast %struct.big* [[B]] to i8* 322; CHECK-NEXT: [[TMP1:%.*]] = bitcast %struct.big* [[TMP]] to i8* 323; CHECK-NEXT: call void @f2(%struct.big* [[B]]) 324; CHECK-NEXT: ret void 325; 326entry: 327 %b = alloca %struct.big, align 4 328 %tmp = alloca %struct.big, align 4 329 call void @f1(%struct.big* sret(%struct.big) %tmp) 330 %0 = bitcast %struct.big* %b to i8* 331 %1 = bitcast %struct.big* %tmp to i8* 332 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %0, i8* align 4 %1, i64 200, i1 false) 333 call void @f2(%struct.big* %b) 334 ret void 335} 336 337; rdar://14073661. 338; Test10 triggered assertion when the compiler try to get the size of the 339; opaque type of *x, where the x is the formal argument with attribute 'sret'. 340 341%opaque = type opaque 342declare void @foo(i32* noalias nocapture) 343 344define void @test10(%opaque* noalias nocapture sret(%opaque) %x, i32 %y) { 345; CHECK-LABEL: @test10( 346; CHECK-NEXT: [[A:%.*]] = alloca i32, align 4 347; CHECK-NEXT: store i32 [[Y:%.*]], i32* [[A]], align 4 348; CHECK-NEXT: call void @foo(i32* noalias nocapture [[A]]) 349; CHECK-NEXT: [[C:%.*]] = load i32, i32* [[A]], align 4 350; CHECK-NEXT: [[D:%.*]] = bitcast %opaque* [[X:%.*]] to i32* 351; CHECK-NEXT: store i32 [[C]], i32* [[D]], align 4 352; CHECK-NEXT: ret void 353; 354 %a = alloca i32, align 4 355 store i32 %y, i32* %a 356 call void @foo(i32* noalias nocapture %a) 357 %c = load i32, i32* %a 358 %d = bitcast %opaque* %x to i32* 359 store i32 %c, i32* %d 360 ret void 361} 362 363; don't create new addressspacecasts when we don't know they're safe for the target 364define void @test11([20 x i32] addrspace(1)* nocapture dereferenceable(80) %P) { 365; CHECK-LABEL: @test11( 366; CHECK-NEXT: [[B:%.*]] = bitcast [20 x i32] addrspace(1)* [[P:%.*]] to i8 addrspace(1)* 367; CHECK-NEXT: call void @llvm.memset.p1i8.i64(i8 addrspace(1)* align 4 [[B]], i8 0, i64 80, i1 false) 368; CHECK-NEXT: ret void 369; 370 %A = alloca [20 x i32], align 4 371 %a = bitcast [20 x i32]* %A to i8* 372 %b = bitcast [20 x i32] addrspace(1)* %P to i8 addrspace(1)* 373 call void @llvm.memset.p0i8.i64(i8* align 4 %a, i8 0, i64 80, i1 false) 374 call void @llvm.memcpy.p1i8.p0i8.i64(i8 addrspace(1)* align 4 %b, i8* align 4 %a, i64 80, i1 false) 375 ret void 376} 377 378declare void @llvm.memset.p0i8.i64(i8* nocapture, i8, i64, i1) nounwind 379declare void @llvm.memcpy.p1i8.p0i8.i64(i8 addrspace(1)* nocapture, i8* nocapture, i64, i1) nounwind 380 381declare void @f1(%struct.big* nocapture sret(%struct.big)) 382declare void @f2(%struct.big*) 383 384; CHECK: attributes #1 = { argmemonly nofree nosync nounwind willreturn } 385; CHECK: attributes #2 = { nounwind ssp } 386; CHECK: attributes #3 = { willreturn } 387; CHECK: attributes #4 = { nounwind ssp uwtable } 388; CHECK: attributes #5 = { argmemonly nofree nosync nounwind willreturn writeonly } 389