1 ; RUN: opt < %s -basicaa -tbaa -licm -S | FileCheck %s 2 ; RUN: opt -aa-pipeline=type-based-aa,basic-aa -passes='require<aa>,require<targetir>,require<scalar-evolution>,require<opt-remark-emit>,loop(licm)' -S %s | FileCheck %s 3 target datalayout = "E-p:64:64:64-a0:0:8-f32:32:32-f64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-v64:64:64-v128:128:128" 4 5 @X = global i32 7 ; <i32*> [#uses=4] 6 7 define void @test1(i32 %i) { 8 Entry: 9 br label %Loop 10 ; CHECK-LABEL: @test1( 11 ; CHECK: Entry: 12 ; CHECK-NEXT: load i32, i32* @X 13 ; CHECK-NEXT: br label %Loop 14 15 16 Loop: ; preds = %Loop, %0 17 %j = phi i32 [ 0, %Entry ], [ %Next, %Loop ] ; <i32> [#uses=1] 18 %x = load i32, i32* @X ; <i32> [#uses=1] 19 %x2 = add i32 %x, 1 ; <i32> [#uses=1] 20 store i32 %x2, i32* @X 21 %Next = add i32 %j, 1 ; <i32> [#uses=2] 22 %cond = icmp eq i32 %Next, 0 ; <i1> [#uses=1] 23 br i1 %cond, label %Out, label %Loop 24 25 Out: 26 ret void 27 ; CHECK: Out: 28 ; CHECK-NEXT: %[[LCSSAPHI:.*]] = phi i32 [ %x2 29 ; CHECK-NEXT: store i32 %[[LCSSAPHI]], i32* @X 30 ; CHECK-NEXT: ret void 31 32 } 33 34 define void @test2(i32 %i) { 35 Entry: 36 br label %Loop 37 ; CHECK-LABEL: @test2( 38 ; CHECK: Entry: 39 ; CHECK-NEXT: %.promoted = load i32, i32* getelementptr inbounds (i32, i32* @X, i64 1) 40 ; CHECK-NEXT: br label %Loop 41 42 Loop: ; preds = %Loop, %0 43 %X1 = getelementptr i32, i32* @X, i64 1 ; <i32*> [#uses=1] 44 %A = load i32, i32* %X1 ; <i32> [#uses=1] 45 %V = add i32 %A, 1 ; <i32> [#uses=1] 46 %X2 = getelementptr i32, i32* @X, i64 1 ; <i32*> [#uses=1] 47 store i32 %V, i32* %X2 48 br i1 false, label %Loop, label %Exit 49 50 Exit: ; preds = %Loop 51 ret void 52 ; CHECK: Exit: 53 ; CHECK-NEXT: %[[LCSSAPHI:.*]] = phi i32 [ %V 54 ; CHECK-NEXT: store i32 %[[LCSSAPHI]], i32* getelementptr inbounds (i32, i32* @X, i64 1) 55 ; CHECK-NEXT: ret void 56 } 57 58 59 60 define void @test3(i32 %i) { 61 ; CHECK-LABEL: @test3( 62 br label %Loop 63 Loop: 64 ; Should not promote this to a register 65 %x = load volatile i32, i32* @X 66 %x2 = add i32 %x, 1 67 store i32 %x2, i32* @X 68 br i1 true, label %Out, label %Loop 69 70 ; CHECK: Loop: 71 ; CHECK-NEXT: load volatile 72 73 Out: ; preds = %Loop 74 ret void 75 } 76 77 ; PR8041 78 define void @test4(i8* %x, i8 %n) { 79 ; CHECK-LABEL: @test4( 80 %handle1 = alloca i8* 81 %handle2 = alloca i8* 82 store i8* %x, i8** %handle1 83 br label %loop 84 85 loop: 86 %tmp = getelementptr i8, i8* %x, i64 8 87 store i8* %tmp, i8** %handle2 88 br label %subloop 89 90 subloop: 91 %count = phi i8 [ 0, %loop ], [ %nextcount, %subloop ] 92 %offsetx2 = load i8*, i8** %handle2 93 store i8 %n, i8* %offsetx2 94 %newoffsetx2 = getelementptr i8, i8* %offsetx2, i64 -1 95 store i8* %newoffsetx2, i8** %handle2 96 %nextcount = add i8 %count, 1 97 %innerexitcond = icmp sge i8 %nextcount, 8 98 br i1 %innerexitcond, label %innerexit, label %subloop 99 100 ; Should have promoted 'handle2' accesses. 101 ; CHECK: subloop: 102 ; CHECK-NEXT: phi i8* [ 103 ; CHECK-NEXT: %count = phi i8 [ 104 ; CHECK-NEXT: store i8 %n 105 ; CHECK-NOT: store 106 ; CHECK: br i1 107 108 innerexit: 109 %offsetx1 = load i8*, i8** %handle1 110 %val = load i8, i8* %offsetx1 111 %cond = icmp eq i8 %val, %n 112 br i1 %cond, label %exit, label %loop 113 114 ; Should not have promoted offsetx1 loads. 115 ; CHECK: innerexit: 116 ; CHECK: %val = load i8, i8* %offsetx1 117 ; CHECK: %cond = icmp eq i8 %val, %n 118 ; CHECK: br i1 %cond, label %exit, label %loop 119 120 exit: 121 ret void 122 } 123 124 define void @test5(i32 %i, i32** noalias %P2) { 125 Entry: 126 br label %Loop 127 ; CHECK-LABEL: @test5( 128 ; CHECK: Entry: 129 ; CHECK-NEXT: load i32, i32* @X 130 ; CHECK-NEXT: br label %Loop 131 132 133 Loop: ; preds = %Loop, %0 134 %j = phi i32 [ 0, %Entry ], [ %Next, %Loop ] ; <i32> [#uses=1] 135 %x = load i32, i32* @X ; <i32> [#uses=1] 136 %x2 = add i32 %x, 1 ; <i32> [#uses=1] 137 store i32 %x2, i32* @X 138 139 store atomic i32* @X, i32** %P2 monotonic, align 8 140 141 %Next = add i32 %j, 1 ; <i32> [#uses=2] 142 %cond = icmp eq i32 %Next, 0 ; <i1> [#uses=1] 143 br i1 %cond, label %Out, label %Loop 144 145 Out: 146 ret void 147 ; CHECK: Out: 148 ; CHECK-NEXT: %[[LCSSAPHI:.*]] = phi i32 [ %x2 149 ; CHECK-NEXT: store i32 %[[LCSSAPHI]], i32* @X 150 ; CHECK-NEXT: ret void 151 152 } 153 154 155 ; PR14753 - Preserve TBAA tags when promoting values in a loop. 156 define void @test6(i32 %n, float* nocapture %a, i32* %gi) { 157 entry: 158 store i32 0, i32* %gi, align 4, !tbaa !0 159 %cmp1 = icmp slt i32 0, %n 160 br i1 %cmp1, label %for.body.lr.ph, label %for.end 161 162 for.body.lr.ph: ; preds = %entry 163 br label %for.body 164 165 for.body: ; preds = %for.body.lr.ph, %for.body 166 %storemerge2 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.body ] 167 %idxprom = sext i32 %storemerge2 to i64 168 %arrayidx = getelementptr inbounds float, float* %a, i64 %idxprom 169 store float 0.000000e+00, float* %arrayidx, align 4, !tbaa !3 170 %0 = load i32, i32* %gi, align 4, !tbaa !0 171 %inc = add nsw i32 %0, 1 172 store i32 %inc, i32* %gi, align 4, !tbaa !0 173 %cmp = icmp slt i32 %inc, %n 174 br i1 %cmp, label %for.body, label %for.cond.for.end_crit_edge 175 176 for.cond.for.end_crit_edge: ; preds = %for.body 177 br label %for.end 178 179 for.end: ; preds = %for.cond.for.end_crit_edge, %entry 180 ret void 181 182 ; CHECK: for.body.lr.ph: 183 ; CHECK-NEXT: %gi.promoted = load i32, i32* %gi, align 4, !tbaa !0 184 ; CHECK: for.cond.for.end_crit_edge: 185 ; CHECK-NEXT: %[[LCSSAPHI:.*]] = phi i32 [ %inc 186 ; CHECK-NEXT: store i32 %[[LCSSAPHI]], i32* %gi, align 4, !tbaa !0 187 } 188 189 declare i32 @opaque(i32) argmemonly 190 declare void @capture(i32*) 191 192 ; We can promote even if opaque may throw. 193 define i32 @test7() { 194 ; CHECK-LABEL: @test7( 195 ; CHECK: entry: 196 ; CHECK-NEXT: %local = alloca 197 ; CHECK-NEXT: call void @capture(i32* %local) 198 ; CHECK-NEXT: load i32, i32* %local 199 ; CHECK-NEXT: br label %loop 200 ; CHECK: exit: 201 ; CHECK-NEXT: %[[LCSSAPHI:.*]] = phi i32 [ %x2, %loop ] 202 ; CHECK-NEXT: store i32 %[[LCSSAPHI]], i32* %local 203 ; CHECK-NEXT: %ret = load i32, i32* %local 204 ; CHECK-NEXT: ret i32 %ret 205 entry: 206 %local = alloca i32 207 call void @capture(i32* %local) 208 br label %loop 209 210 loop: 211 %j = phi i32 [ 0, %entry ], [ %next, %loop ] 212 %x = load i32, i32* %local 213 %x2 = call i32 @opaque(i32 %x) ; Note this does not capture %local 214 store i32 %x2, i32* %local 215 %next = add i32 %j, 1 216 %cond = icmp eq i32 %next, 0 217 br i1 %cond, label %exit, label %loop 218 219 exit: 220 %ret = load i32, i32* %local 221 ret i32 %ret 222 } 223 224 ; Make sure we don't promote if the store is really control-flow dependent. 225 define i32 @test7bad() { 226 ; CHECK-LABEL: @test7bad( 227 ; CHECK: entry: 228 ; CHECK-NEXT: %local = alloca 229 ; CHECK-NEXT: call void @capture(i32* %local) 230 ; CHECK-NEXT: br label %loop 231 ; CHECK: if: 232 ; CHECK-NEXT: store i32 %x2, i32* %local 233 ; CHECK-NEXT: br label %else 234 ; CHECK: exit: 235 ; CHECK-NEXT: %ret = load i32, i32* %local 236 ; CHECK-NEXT: ret i32 %ret 237 entry: 238 %local = alloca i32 239 call void @capture(i32* %local) 240 br label %loop 241 loop: 242 %j = phi i32 [ 0, %entry ], [ %next, %else ] 243 %x = load i32, i32* %local 244 %x2 = call i32 @opaque(i32 %x) ; Note this does not capture %local 245 %cmp = icmp eq i32 %x2, 0 246 br i1 %cmp, label %if, label %else 247 248 if: 249 store i32 %x2, i32* %local 250 br label %else 251 252 else: 253 %next = add i32 %j, 1 254 %cond = icmp eq i32 %next, 0 255 br i1 %cond, label %exit, label %loop 256 257 exit: 258 %ret = load i32, i32* %local 259 ret i32 %ret 260 } 261 262 ; Even if neither the load nor the store or guaranteed to execute because 263 ; opaque() may throw, we can still promote - the load not being guaranteed 264 ; doesn't block us, because %local is always dereferenceable. 265 define i32 @test8() { 266 ; CHECK-LABEL: @test8( 267 ; CHECK: entry: 268 ; CHECK-NEXT: %local = alloca 269 ; CHECK-NEXT: call void @capture(i32* %local) 270 ; CHECK-NEXT: load i32, i32* %local 271 ; CHECK-NEXT: br label %loop 272 ; CHECK: exit: 273 ; CHECK-NEXT: %[[LCSSAPHI:.*]] = phi i32 [ %x2, %loop ] 274 ; CHECK-NEXT: store i32 %[[LCSSAPHI]], i32* %local 275 ; CHECK-NEXT: %ret = load i32, i32* %local 276 ; CHECK-NEXT: ret i32 %ret 277 entry: 278 %local = alloca i32 279 call void @capture(i32* %local) 280 br label %loop 281 282 loop: 283 %j = phi i32 [ 0, %entry ], [ %next, %loop ] 284 %throwaway = call i32 @opaque(i32 %j) 285 %x = load i32, i32* %local 286 %x2 = call i32 @opaque(i32 %x) 287 store i32 %x2, i32* %local 288 %next = add i32 %j, 1 289 %cond = icmp eq i32 %next, 0 290 br i1 %cond, label %exit, label %loop 291 292 exit: 293 %ret = load i32, i32* %local 294 ret i32 %ret 295 } 296 297 298 ; If the store is "guaranteed modulo exceptions", and the load depends on 299 ; control flow, we can only promote if the pointer is otherwise known to be 300 ; dereferenceable 301 define i32 @test9() { 302 ; CHECK-LABEL: @test9( 303 ; CHECK: entry: 304 ; CHECK-NEXT: %local = alloca 305 ; CHECK-NEXT: call void @capture(i32* %local) 306 ; CHECK-NEXT: load i32, i32* %local 307 ; CHECK-NEXT: br label %loop 308 ; CHECK: exit: 309 ; CHECK-NEXT: %[[LCSSAPHI:.*]] = phi i32 [ %x2, %else ] 310 ; CHECK-NEXT: store i32 %[[LCSSAPHI]], i32* %local 311 ; CHECK-NEXT: %ret = load i32, i32* %local 312 ; CHECK-NEXT: ret i32 %ret 313 entry: 314 %local = alloca i32 315 call void @capture(i32* %local) 316 br label %loop 317 318 loop: 319 %j = phi i32 [ 0, %entry ], [ %next, %else ] 320 %j2 = call i32 @opaque(i32 %j) 321 %cmp = icmp eq i32 %j2, 0 322 br i1 %cmp, label %if, label %else 323 324 if: 325 %x = load i32, i32* %local 326 br label %else 327 328 else: 329 %x2 = phi i32 [ 0, %loop ], [ %x, %if] 330 store i32 %x2, i32* %local 331 %next = add i32 %j, 1 332 %cond = icmp eq i32 %next, 0 333 br i1 %cond, label %exit, label %loop 334 335 exit: 336 %ret = load i32, i32* %local 337 ret i32 %ret 338 } 339 340 define i32 @test9bad(i32 %i) { 341 ; CHECK-LABEL: @test9bad( 342 ; CHECK: entry: 343 ; CHECK-NEXT: %local = alloca 344 ; CHECK-NEXT: call void @capture(i32* %local) 345 ; CHECK-NEXT: %notderef = getelementptr 346 ; CHECK-NEXT: br label %loop 347 ; CHECK: if: 348 ; CHECK-NEXT: load i32, i32* %notderef 349 ; CHECK-NEXT: br label %else 350 ; CHECK: exit: 351 ; CHECK-NEXT: %ret = load i32, i32* %notderef 352 ; CHECK-NEXT: ret i32 %ret 353 entry: 354 %local = alloca i32 355 call void @capture(i32* %local) 356 %notderef = getelementptr i32, i32* %local, i32 %i 357 br label %loop 358 359 loop: 360 %j = phi i32 [ 0, %entry ], [ %next, %else ] 361 %j2 = call i32 @opaque(i32 %j) 362 %cmp = icmp eq i32 %j2, 0 363 br i1 %cmp, label %if, label %else 364 365 if: 366 %x = load i32, i32* %notderef 367 br label %else 368 369 else: 370 %x2 = phi i32 [ 0, %loop ], [ %x, %if] 371 store i32 %x2, i32* %notderef 372 %next = add i32 %j, 1 373 %cond = icmp eq i32 %next, 0 374 br i1 %cond, label %exit, label %loop 375 376 exit: 377 %ret = load i32, i32* %notderef 378 ret i32 %ret 379 } 380 381 define void @test10(i32 %i) { 382 Entry: 383 br label %Loop 384 ; CHECK-LABEL: @test10( 385 ; CHECK: Entry: 386 ; CHECK-NEXT: load atomic i32, i32* @X unordered, align 4 387 ; CHECK-NEXT: br label %Loop 388 389 390 Loop: ; preds = %Loop, %0 391 %j = phi i32 [ 0, %Entry ], [ %Next, %Loop ] ; <i32> [#uses=1] 392 %x = load atomic i32, i32* @X unordered, align 4 393 %x2 = add i32 %x, 1 394 store atomic i32 %x2, i32* @X unordered, align 4 395 %Next = add i32 %j, 1 396 %cond = icmp eq i32 %Next, 0 397 br i1 %cond, label %Out, label %Loop 398 399 Out: 400 ret void 401 ; CHECK: Out: 402 ; CHECK-NEXT: %[[LCSSAPHI:.*]] = phi i32 [ %x2 403 ; CHECK-NEXT: store atomic i32 %[[LCSSAPHI]], i32* @X unordered, align 4 404 ; CHECK-NEXT: ret void 405 406 } 407 408 ; Early exit is known not to be taken on first iteration and thus doesn't 409 ; effect whether load is known to execute. 410 define void @test11(i32 %i) { 411 Entry: 412 br label %Loop 413 ; CHECK-LABEL: @test11( 414 ; CHECK: Entry: 415 ; CHECK-NEXT: load i32, i32* @X 416 ; CHECK-NEXT: br label %Loop 417 418 419 Loop: ; preds = %Loop, %0 420 %j = phi i32 [ 0, %Entry ], [ %Next, %body ] ; <i32> [#uses=1] 421 %early.test = icmp ult i32 %j, 32 422 br i1 %early.test, label %body, label %Early 423 body: 424 %x = load i32, i32* @X ; <i32> [#uses=1] 425 %x2 = add i32 %x, 1 ; <i32> [#uses=1] 426 store i32 %x2, i32* @X 427 %Next = add i32 %j, 1 ; <i32> [#uses=2] 428 %cond = icmp eq i32 %Next, 0 ; <i1> [#uses=1] 429 br i1 %cond, label %Out, label %Loop 430 431 Early: 432 ; CHECK: Early: 433 ; CHECK-NEXT: %[[LCSSAPHI:.*]] = phi i32 [ %x2 434 ; CHECK-NEXT: store i32 %[[LCSSAPHI]], i32* @X 435 ; CHECK-NEXT: ret void 436 ret void 437 Out: 438 ret void 439 ; CHECK: Out: 440 ; CHECK-NEXT: %[[LCSSAPHI:.*]] = phi i32 [ %x2 441 ; CHECK-NEXT: store i32 %[[LCSSAPHI]], i32* @X 442 ; CHECK-NEXT: ret void 443 444 } 445 446 !0 = !{!4, !4, i64 0} 447 !1 = !{!"omnipotent char", !2} 448 !2 = !{!"Simple C/C++ TBAA"} 449 !3 = !{!5, !5, i64 0} 450 !4 = !{!"int", !1} 451 !5 = !{!"float", !1} 452
