1; RUN: llc -mtriple=powerpc64le-unknown-linux-gnu -mcpu=pwr8 %s -o - -verify-machineinstrs | FileCheck %s --check-prefix=CHECK --check-prefix=ENABLE 2; RUN: llc -mtriple=powerpc64le-unknown-linux-gnu %s -o - -enable-shrink-wrap=false -verify-machineinstrs | FileCheck %s --check-prefix=CHECK --check-prefix=DISABLE 3; 4; Note: Lots of tests use inline asm instead of regular calls. 5; This allows to have a better control on what the allocation will do. 6; Otherwise, we may have spill right in the entry block, defeating 7; shrink-wrapping. Moreover, some of the inline asm statement (nop) 8; are here to ensure that the related paths do not end up as critical 9; edges. 10 11 12; Initial motivating example: Simple diamond with a call just on one side. 13; CHECK-LABEL: foo: 14; 15; Compare the arguments and return 16; No prologue needed. 17; ENABLE: cmpw 3, 4 18; ENABLE-NEXT: bgelr 0 19; 20; Prologue code. 21; At a minimum, we save/restore the link register. Other registers may be saved 22; as well. 23; CHECK: mflr 24; 25; Compare the arguments and jump to exit. 26; After the prologue is set. 27; DISABLE: cmpw 3, 4 28; DISABLE-NEXT: bge 0, .[[EXIT_LABEL:LBB[0-9_]+]] 29; 30; Store %a on the stack 31; CHECK: stw 3, {{[0-9]+([0-9]+)}} 32; Set the alloca address in the second argument. 33; CHECK-NEXT: addi 4, 1, {{[0-9]+}} 34; Set the first argument to zero. 35; CHECK-NEXT: li 3, 0 36; CHECK-NEXT: bl doSomething 37; 38; With shrink-wrapping, epilogue is just after the call. 39; Restore the link register and return. 40; Note that there could be other epilog code before the link register is 41; restored but we will not check for it here. 42; ENABLE: mtlr 43; ENABLE-NEXT: blr 44; 45; DISABLE: [[EXIT_LABEL]]: 46; 47; Without shrink-wrapping, epilogue is in the exit block. 48; Epilogue code. (What we pop does not matter.) 49; DISABLE: mtlr {{[0-9]+}} 50; DISABLE-NEXT: blr 51; 52 53define i32 @foo(i32 %a, i32 %b) { 54 %tmp = alloca i32, align 4 55 %tmp2 = icmp slt i32 %a, %b 56 br i1 %tmp2, label %true, label %false 57 58true: 59 store i32 %a, i32* %tmp, align 4 60 %tmp4 = call i32 @doSomething(i32 0, i32* %tmp) 61 br label %false 62 63false: 64 %tmp.0 = phi i32 [ %tmp4, %true ], [ %a, %0 ] 65 ret i32 %tmp.0 66} 67 68; Function Attrs: optsize 69declare i32 @doSomething(i32, i32*) 70 71 72 73; Check that we do not perform the restore inside the loop whereas the save 74; is outside. 75; CHECK-LABEL: freqSaveAndRestoreOutsideLoop: 76; 77; Shrink-wrapping allows to skip the prologue in the else case. 78; ENABLE: cmplwi 3, 0 79; ENABLE: beq 0, .[[ELSE_LABEL:LBB[0-9_]+]] 80; 81; Prologue code. 82; Make sure we save the link register 83; CHECK: mflr {{[0-9]+}} 84; 85; DISABLE: cmplwi 3, 0 86; DISABLE: beq 0, .[[ELSE_LABEL:LBB[0-9_]+]] 87; 88; Loop preheader 89; CHECK-DAG: li [[SUM:[0-9]+]], 0 90; CHECK-DAG: li [[IV:[0-9]+]], 10 91; 92; Loop body 93; CHECK: .[[LOOP:LBB[0-9_]+]]: # %for.body 94; CHECK: bl something 95; CHECK-DAG: addi [[IV]], [[IV]], -1 96; CHECK-DAG: add [[SUM]], 3, [[SUM]] 97; CHECK-NEXT: cmplwi [[IV]], 0 98; CHECK-NEXT: bne 0, .[[LOOP]] 99; 100; Next BB. 101; CHECK: slwi 3, [[SUM]], 3 102; 103; Jump to epilogue. 104; DISABLE: b .[[EPILOG_BB:LBB[0-9_]+]] 105; 106; DISABLE: .[[ELSE_LABEL]]: # %if.else 107; Shift second argument by one and store into returned register. 108; DISABLE: slwi 3, 4, 1 109; DISABLE: .[[EPILOG_BB]]: # %if.end 110; 111; Epilogue code. 112; CHECK: mtlr {{[0-9]+}} 113; CHECK: blr 114; 115; ENABLE: .[[ELSE_LABEL]]: # %if.else 116; Shift second argument by one and store into returned register. 117; ENABLE: slwi 3, 4, 1 118; ENABLE-NEXT: blr 119define i32 @freqSaveAndRestoreOutsideLoop(i32 %cond, i32 %N) { 120entry: 121 %tobool = icmp eq i32 %cond, 0 122 br i1 %tobool, label %if.else, label %for.preheader 123 124for.preheader: 125 tail call void asm "nop", ""() 126 br label %for.body 127 128for.body: ; preds = %entry, %for.body 129 %i.05 = phi i32 [ %inc, %for.body ], [ 0, %for.preheader ] 130 %sum.04 = phi i32 [ %add, %for.body ], [ 0, %for.preheader ] 131 %call = tail call i32 bitcast (i32 (...)* @something to i32 ()*)() 132 %add = add nsw i32 %call, %sum.04 133 %inc = add nuw nsw i32 %i.05, 1 134 %exitcond = icmp eq i32 %inc, 10 135 br i1 %exitcond, label %for.end, label %for.body 136 137for.end: ; preds = %for.body 138 %shl = shl i32 %add, 3 139 br label %if.end 140 141if.else: ; preds = %entry 142 %mul = shl nsw i32 %N, 1 143 br label %if.end 144 145if.end: ; preds = %if.else, %for.end 146 %sum.1 = phi i32 [ %shl, %for.end ], [ %mul, %if.else ] 147 ret i32 %sum.1 148} 149 150declare i32 @something(...) 151 152; Check that we do not perform the shrink-wrapping inside the loop even 153; though that would be legal. The cost model must prevent that. 154; CHECK-LABEL: freqSaveAndRestoreOutsideLoop2: 155; Prologue code. 156; Make sure we save the link register before the call 157; CHECK: mflr {{[0-9]+}} 158; 159; Loop preheader 160; CHECK-DAG: li [[SUM:[0-9]+]], 0 161; CHECK-DAG: li [[IV:[0-9]+]], 10 162; 163; Loop body 164; CHECK: .[[LOOP:LBB[0-9_]+]]: # %for.body 165; CHECK: bl something 166; CHECK-DAG: addi [[IV]], [[IV]], -1 167; CHECK-DAG: add [[SUM]], 3, [[SUM]] 168; CHECK-NEXT: cmplwi [[IV]], 0 169; CHECK-NEXT: bne 0, .[[LOOP]] 170; 171; Next BB 172; CHECK: %for.exit 173; CHECK: mtlr {{[0-9]+}} 174; CHECK: blr 175define i32 @freqSaveAndRestoreOutsideLoop2(i32 %cond) { 176entry: 177 br label %for.preheader 178 179for.preheader: 180 tail call void asm "nop", ""() 181 br label %for.body 182 183for.body: ; preds = %for.body, %entry 184 %i.04 = phi i32 [ 0, %for.preheader ], [ %inc, %for.body ] 185 %sum.03 = phi i32 [ 0, %for.preheader ], [ %add, %for.body ] 186 %call = tail call i32 bitcast (i32 (...)* @something to i32 ()*)() 187 %add = add nsw i32 %call, %sum.03 188 %inc = add nuw nsw i32 %i.04, 1 189 %exitcond = icmp eq i32 %inc, 10 190 br i1 %exitcond, label %for.exit, label %for.body 191 192for.exit: 193 tail call void asm "nop", ""() 194 br label %for.end 195 196for.end: ; preds = %for.body 197 ret i32 %add 198} 199 200 201; Check with a more complex case that we do not have save within the loop and 202; restore outside. 203; CHECK-LABEL: loopInfoSaveOutsideLoop: 204; 205; ENABLE: cmplwi 3, 0 206; ENABLE-NEXT: beq 0, .[[ELSE_LABEL:LBB[0-9_]+]] 207; 208; Prologue code. 209; Make sure we save the link register 210; CHECK: mflr {{[0-9]+}} 211; 212; DISABLE: std 213; DISABLE-NEXT: std 214; DISABLE: cmplwi 3, 0 215; DISABLE-NEXT: beq 0, .[[ELSE_LABEL:LBB[0-9_]+]] 216; 217; Loop preheader 218; CHECK-DAG: li [[SUM:[0-9]+]], 0 219; CHECK-DAG: li [[IV:[0-9]+]], 10 220; 221; Loop body 222; CHECK: .[[LOOP:LBB[0-9_]+]]: # %for.body 223; CHECK: bl something 224; CHECK-DAG: addi [[IV]], [[IV]], -1 225; CHECK-DAG: add [[SUM]], 3, [[SUM]] 226; CHECK-NEXT: cmplwi [[IV]], 0 227; CHECK-NEXT: bne 0, .[[LOOP]] 228; 229; Next BB 230; CHECK: bl somethingElse 231; CHECK: slwi 3, [[SUM]], 3 232; 233; Jump to epilogue 234; DISABLE: b .[[EPILOG_BB:LBB[0-9_]+]] 235; 236; DISABLE: .[[ELSE_LABEL]]: # %if.else 237; Shift second argument by one and store into returned register. 238; DISABLE: slwi 3, 4, 1 239; 240; DISABLE: .[[EPILOG_BB]]: # %if.end 241; Epilog code 242; CHECK: mtlr {{[0-9]+}} 243; CHECK: blr 244; 245; ENABLE: .[[ELSE_LABEL]]: # %if.else 246; Shift second argument by one and store into returned register. 247; ENABLE: slwi 3, 4, 1 248; ENABLE-NEXT: blr 249define i32 @loopInfoSaveOutsideLoop(i32 %cond, i32 %N) { 250entry: 251 %tobool = icmp eq i32 %cond, 0 252 br i1 %tobool, label %if.else, label %for.preheader 253 254for.preheader: 255 tail call void asm "nop", ""() 256 br label %for.body 257 258for.body: ; preds = %entry, %for.body 259 %i.05 = phi i32 [ %inc, %for.body ], [ 0, %for.preheader ] 260 %sum.04 = phi i32 [ %add, %for.body ], [ 0, %for.preheader ] 261 %call = tail call i32 bitcast (i32 (...)* @something to i32 ()*)() 262 %add = add nsw i32 %call, %sum.04 263 %inc = add nuw nsw i32 %i.05, 1 264 %exitcond = icmp eq i32 %inc, 10 265 br i1 %exitcond, label %for.end, label %for.body 266 267for.end: ; preds = %for.body 268 tail call void bitcast (void (...)* @somethingElse to void ()*)() 269 %shl = shl i32 %add, 3 270 br label %if.end 271 272if.else: ; preds = %entry 273 %mul = shl nsw i32 %N, 1 274 br label %if.end 275 276if.end: ; preds = %if.else, %for.end 277 %sum.1 = phi i32 [ %shl, %for.end ], [ %mul, %if.else ] 278 ret i32 %sum.1 279} 280 281declare void @somethingElse(...) 282 283; Check with a more complex case that we do not have restore within the loop and 284; save outside. 285; CHECK-LABEL: loopInfoRestoreOutsideLoop: 286; 287; ENABLE: cmplwi 3, 0 288; ENABLE-NEXT: beq 0, .[[ELSE_LABEL:LBB[0-9_]+]] 289; 290; Prologue code. 291; Make sure we save the link register 292; CHECK: mflr {{[0-9]+}} 293; 294; DISABLE: std 295; DISABLE-NEXT: std 296; DISABLE: cmplwi 3, 0 297; DISABLE-NEXT: beq 0, .[[ELSE_LABEL:LBB[0-9_]+]] 298; 299; CHECK: bl somethingElse 300; 301; Loop preheader 302; CHECK-DAG: li [[SUM:[0-9]+]], 0 303; CHECK-DAG: li [[IV:[0-9]+]], 10 304; 305; Loop body 306; CHECK: .[[LOOP:LBB[0-9_]+]]: # %for.body 307; CHECK: bl something 308; CHECK-DAG: addi [[IV]], [[IV]], -1 309; CHECK-DAG: add [[SUM]], 3, [[SUM]] 310; CHECK-NEXT: cmplwi [[IV]], 0 311; CHECK-NEXT: bne 0, .[[LOOP]] 312; 313; Next BB. 314; slwi 3, [[SUM]], 3 315; 316; DISABLE: b .[[EPILOG_BB:LBB[0-9_]+]] 317; 318; DISABLE: .[[ELSE_LABEL]]: # %if.else 319; Shift second argument by one and store into returned register. 320; DISABLE: slwi 3, 4, 1 321; DISABLE: .[[EPILOG_BB]]: # %if.end 322; 323; Epilogue code. 324; CHECK: mtlr {{[0-9]+}} 325; CHECK: blr 326; 327; ENABLE: .[[ELSE_LABEL]]: # %if.else 328; Shift second argument by one and store into returned register. 329; ENABLE: slwi 3, 4, 1 330; ENABLE-NEXT: blr 331define i32 @loopInfoRestoreOutsideLoop(i32 %cond, i32 %N) nounwind { 332entry: 333 %tobool = icmp eq i32 %cond, 0 334 br i1 %tobool, label %if.else, label %if.then 335 336if.then: ; preds = %entry 337 tail call void bitcast (void (...)* @somethingElse to void ()*)() 338 br label %for.body 339 340for.body: ; preds = %for.body, %if.then 341 %i.05 = phi i32 [ 0, %if.then ], [ %inc, %for.body ] 342 %sum.04 = phi i32 [ 0, %if.then ], [ %add, %for.body ] 343 %call = tail call i32 bitcast (i32 (...)* @something to i32 ()*)() 344 %add = add nsw i32 %call, %sum.04 345 %inc = add nuw nsw i32 %i.05, 1 346 %exitcond = icmp eq i32 %inc, 10 347 br i1 %exitcond, label %for.end, label %for.body 348 349for.end: ; preds = %for.body 350 %shl = shl i32 %add, 3 351 br label %if.end 352 353if.else: ; preds = %entry 354 %mul = shl nsw i32 %N, 1 355 br label %if.end 356 357if.end: ; preds = %if.else, %for.end 358 %sum.1 = phi i32 [ %shl, %for.end ], [ %mul, %if.else ] 359 ret i32 %sum.1 360} 361 362; Check that we handle function with no frame information correctly. 363; CHECK-LABEL: emptyFrame: 364; CHECK: # %entry 365; CHECK-NEXT: li 3, 0 366; CHECK-NEXT: blr 367define i32 @emptyFrame() { 368entry: 369 ret i32 0 370} 371 372 373; Check that we handle inline asm correctly. 374; CHECK-LABEL: inlineAsm: 375; 376; ENABLE: cmplwi 3, 0 377; ENABLE-NEXT: beq 0, .[[ELSE_LABEL:LBB[0-9_]+]] 378; 379; Prologue code. 380; Make sure we save the CSR used in the inline asm: r14 381; ENABLE-DAG: li [[IV:[0-9]+]], 10 382; ENABLE-DAG: std 14, -[[STACK_OFFSET:[0-9]+]](1) # 8-byte Folded Spill 383; 384; DISABLE: cmplwi 3, 0 385; DISABLE-NEXT: std 14, -[[STACK_OFFSET:[0-9]+]](1) # 8-byte Folded Spill 386; DISABLE-NEXT: beq 0, .[[ELSE_LABEL:LBB[0-9_]+]] 387; DISABLE: li [[IV:[0-9]+]], 10 388; 389; CHECK: nop 390; CHECK: mtctr [[IV]] 391; 392; CHECK: .[[LOOP_LABEL:LBB[0-9_]+]]: # %for.body 393; Inline asm statement. 394; CHECK: addi 14, 14, 1 395; CHECK: bdnz .[[LOOP_LABEL]] 396; 397; Epilogue code. 398; CHECK: li 3, 0 399; CHECK-DAG: ld 14, -[[STACK_OFFSET]](1) # 8-byte Folded Reload 400; CHECK: nop 401; CHECK: blr 402; 403; CHECK: [[ELSE_LABEL]] 404; CHECK-NEXT: slwi 3, 4, 1 405; DISABLE: ld 14, -[[STACK_OFFSET]](1) # 8-byte Folded Reload 406; CHECK-NEXT: blr 407; 408define i32 @inlineAsm(i32 %cond, i32 %N) { 409entry: 410 %tobool = icmp eq i32 %cond, 0 411 br i1 %tobool, label %if.else, label %for.preheader 412 413for.preheader: 414 tail call void asm "nop", ""() 415 br label %for.body 416 417for.body: ; preds = %entry, %for.body 418 %i.03 = phi i32 [ %inc, %for.body ], [ 0, %for.preheader ] 419 tail call void asm "addi 14, 14, 1", "~{r14}"() 420 %inc = add nuw nsw i32 %i.03, 1 421 %exitcond = icmp eq i32 %inc, 10 422 br i1 %exitcond, label %for.exit, label %for.body 423 424for.exit: 425 tail call void asm "nop", ""() 426 br label %if.end 427 428if.else: ; preds = %entry 429 %mul = shl nsw i32 %N, 1 430 br label %if.end 431 432if.end: ; preds = %for.body, %if.else 433 %sum.0 = phi i32 [ %mul, %if.else ], [ 0, %for.exit ] 434 ret i32 %sum.0 435} 436 437 438; Check that we handle calls to variadic functions correctly. 439; CHECK-LABEL: callVariadicFunc: 440; 441; ENABLE: cmplwi 3, 0 442; ENABLE-NEXT: beq 0, .[[ELSE_LABEL:LBB[0-9_]+]] 443; 444; Prologue code. 445; CHECK: mflr {{[0-9]+}} 446; 447; DISABLE: cmplwi 3, 0 448; DISABLE-NEXT: beq 0, .[[ELSE_LABEL:LBB[0-9_]+]] 449; 450; Setup of the varags. 451; CHECK: mr 4, 3 452; CHECK-NEXT: mr 5, 3 453; CHECK-NEXT: mr 6, 3 454; CHECK-NEXT: mr 7, 3 455; CHECK-NEXT: mr 8, 3 456; CHECK-NEXT: mr 9, 3 457; CHECK-NEXT: bl someVariadicFunc 458; CHECK: slwi 3, 3, 3 459; DISABLE: b .[[EPILOGUE_BB:LBB[0-9_]+]] 460; 461; ENABLE: mtlr {{[0-9]+}} 462; ENABLE-NEXT: blr 463; 464; CHECK: .[[ELSE_LABEL]]: # %if.else 465; CHECK-NEXT: slwi 3, 4, 1 466; 467; DISABLE: .[[EPILOGUE_BB]]: # %if.end 468; DISABLE: mtlr 469; CHECK: blr 470define i32 @callVariadicFunc(i32 %cond, i32 %N) { 471entry: 472 %tobool = icmp eq i32 %cond, 0 473 br i1 %tobool, label %if.else, label %if.then 474 475if.then: ; preds = %entry 476 %call = tail call i32 (i32, ...) @someVariadicFunc(i32 %N, i32 %N, i32 %N, i32 %N, i32 %N, i32 %N, i32 %N) 477 %shl = shl i32 %call, 3 478 br label %if.end 479 480if.else: ; preds = %entry 481 %mul = shl nsw i32 %N, 1 482 br label %if.end 483 484if.end: ; preds = %if.else, %if.then 485 %sum.0 = phi i32 [ %shl, %if.then ], [ %mul, %if.else ] 486 ret i32 %sum.0 487} 488 489declare i32 @someVariadicFunc(i32, ...) 490 491 492 493; Make sure we do not insert unreachable code after noreturn function. 494; Although this is not incorrect to insert such code, it is useless 495; and it hurts the binary size. 496; 497; CHECK-LABEL: noreturn: 498; DISABLE: mflr {{[0-9]+}} 499; 500; CHECK: cmplwi 3, 0 501; CHECK-NEXT: bne{{[-]?}} 0, .[[ABORT:LBB[0-9_]+]] 502; 503; CHECK: li 3, 42 504; 505; DISABLE: mtlr {{[0-9]+}} 506; 507; CHECK-NEXT: blr 508; 509; CHECK: .[[ABORT]]: # %if.abort 510; 511; ENABLE: mflr {{[0-9]+}} 512; 513; CHECK: bl abort 514; ENABLE-NOT: mtlr {{[0-9]+}} 515define i32 @noreturn(i8 signext %bad_thing) { 516entry: 517 %tobool = icmp eq i8 %bad_thing, 0 518 br i1 %tobool, label %if.end, label %if.abort 519 520if.abort: 521 tail call void @abort() #0 522 unreachable 523 524if.end: 525 ret i32 42 526} 527 528declare void @abort() #0 529 530attributes #0 = { noreturn nounwind } 531 532 533; Make sure that we handle infinite loops properly When checking that the Save 534; and Restore blocks are control flow equivalent, the loop searches for the 535; immediate (post) dominator for the (restore) save blocks. When either the Save 536; or Restore block is located in an infinite loop the only immediate (post) 537; dominator is itself. In this case, we cannot perform shrink wrapping, but we 538; should return gracefully and continue compilation. 539; The only condition for this test is the compilation finishes correctly. 540; 541; CHECK-LABEL: infiniteloop 542; CHECK: blr 543define void @infiniteloop() { 544entry: 545 br i1 undef, label %if.then, label %if.end 546 547if.then: 548 %ptr = alloca i32, i32 4 549 br label %for.body 550 551for.body: ; preds = %for.body, %entry 552 %sum.03 = phi i32 [ 0, %if.then ], [ %add, %for.body ] 553 %call = tail call i32 bitcast (i32 (...)* @something to i32 ()*)() 554 %add = add nsw i32 %call, %sum.03 555 store i32 %add, i32* %ptr 556 br label %for.body 557 558if.end: 559 ret void 560} 561 562; Another infinite loop test this time with a body bigger than just one block. 563; CHECK-LABEL: infiniteloop2 564; CHECK: blr 565define void @infiniteloop2() { 566entry: 567 br i1 undef, label %if.then, label %if.end 568 569if.then: 570 %ptr = alloca i32, i32 4 571 br label %for.body 572 573for.body: ; preds = %for.body, %entry 574 %sum.03 = phi i32 [ 0, %if.then ], [ %add, %body1 ], [ 1, %body2] 575 %call = tail call i32 asm "mftb $0, 268", "=r,~{r14}"() 576 %add = add nsw i32 %call, %sum.03 577 store i32 %add, i32* %ptr 578 br i1 undef, label %body1, label %body2 579 580body1: 581 tail call void asm sideeffect "nop", "~{r14}"() 582 br label %for.body 583 584body2: 585 tail call void asm sideeffect "nop", "~{r14}"() 586 br label %for.body 587 588if.end: 589 ret void 590} 591 592; Another infinite loop test this time with two nested infinite loop. 593; CHECK-LABEL: infiniteloop3 594; CHECK: Lfunc_begin[[FUNCNUM:[0-9]+]] 595; CHECK: bclr 596; CHECK: Lfunc_end[[FUNCNUM]] 597define void @infiniteloop3() { 598entry: 599 br i1 undef, label %loop2a, label %body 600 601body: ; preds = %entry 602 br i1 undef, label %loop2a, label %end 603 604loop1: ; preds = %loop2a, %loop2b 605 %var.phi = phi i32* [ %next.phi, %loop2b ], [ %var, %loop2a ] 606 %next.phi = phi i32* [ %next.load, %loop2b ], [ %next.var, %loop2a ] 607 %0 = icmp eq i32* %var, null 608 %next.load = load i32*, i32** undef 609 br i1 %0, label %loop2a, label %loop2b 610 611loop2a: ; preds = %loop1, %body, %entry 612 %var = phi i32* [ null, %body ], [ null, %entry ], [ %next.phi, %loop1 ] 613 %next.var = phi i32* [ undef, %body ], [ null, %entry ], [ %next.load, %loop1 ] 614 br label %loop1 615 616loop2b: ; preds = %loop1 617 %gep1 = bitcast i32* %var.phi to i32* 618 %next.ptr = bitcast i32* %gep1 to i32** 619 store i32* %next.phi, i32** %next.ptr 620 br label %loop1 621 622end: 623 ret void 624} 625 626@columns = external global [0 x i32], align 4 627@lock = common global i32 0, align 4 628@htindex = common global i32 0, align 4 629@stride = common global i32 0, align 4 630@ht = common global i32* null, align 8 631@he = common global i8* null, align 8 632 633; Test for a bug that was caused when save point was equal to restore point. 634; Function Attrs: nounwind 635; CHECK-LABEL: transpose 636; 637; Store of callee-save register saved by shrink wrapping 638; FIXME: Test disabled: Improved scheduling needs no spills/reloads any longer! 639; CHECKXX: std [[CSR:[0-9]+]], -[[STACK_OFFSET:[0-9]+]](1) # 8-byte Folded Spill 640; 641; Reload of callee-save register 642; CHECKXX: ld [[CSR]], -[[STACK_OFFSET]](1) # 8-byte Folded Reload 643; 644; Ensure no subsequent uses of callee-save register before end of function 645; CHECK-NOT: {{[a-z]+}} [[CSR]] 646; CHECK: blr 647define signext i32 @transpose() { 648entry: 649 %0 = load i32, i32* getelementptr inbounds ([0 x i32], [0 x i32]* @columns, i64 0, i64 1), align 4 650 %shl.i = shl i32 %0, 7 651 %1 = load i32, i32* getelementptr inbounds ([0 x i32], [0 x i32]* @columns, i64 0, i64 2), align 4 652 %or.i = or i32 %shl.i, %1 653 %shl1.i = shl i32 %or.i, 7 654 %2 = load i32, i32* getelementptr inbounds ([0 x i32], [0 x i32]* @columns, i64 0, i64 3), align 4 655 %or2.i = or i32 %shl1.i, %2 656 %3 = load i32, i32* getelementptr inbounds ([0 x i32], [0 x i32]* @columns, i64 0, i64 7), align 4 657 %shl3.i = shl i32 %3, 7 658 %4 = load i32, i32* getelementptr inbounds ([0 x i32], [0 x i32]* @columns, i64 0, i64 6), align 4 659 %or4.i = or i32 %shl3.i, %4 660 %shl5.i = shl i32 %or4.i, 7 661 %5 = load i32, i32* getelementptr inbounds ([0 x i32], [0 x i32]* @columns, i64 0, i64 5), align 4 662 %or6.i = or i32 %shl5.i, %5 663 %cmp.i = icmp ugt i32 %or2.i, %or6.i 664 br i1 %cmp.i, label %cond.true.i, label %cond.false.i 665 666cond.true.i: 667 %shl7.i = shl i32 %or2.i, 7 668 %6 = load i32, i32* getelementptr inbounds ([0 x i32], [0 x i32]* @columns, i64 0, i64 4), align 4 669 %or8.i = or i32 %6, %shl7.i 670 %conv.i = zext i32 %or8.i to i64 671 %shl9.i = shl nuw nsw i64 %conv.i, 21 672 %conv10.i = zext i32 %or6.i to i64 673 %or11.i = or i64 %shl9.i, %conv10.i 674 br label %hash.exit 675 676cond.false.i: 677 %shl12.i = shl i32 %or6.i, 7 678 %7 = load i32, i32* getelementptr inbounds ([0 x i32], [0 x i32]* @columns, i64 0, i64 4), align 4 679 %or13.i = or i32 %7, %shl12.i 680 %conv14.i = zext i32 %or13.i to i64 681 %shl15.i = shl nuw nsw i64 %conv14.i, 21 682 %conv16.i = zext i32 %or2.i to i64 683 %or17.i = or i64 %shl15.i, %conv16.i 684 br label %hash.exit 685 686hash.exit: 687 %cond.i = phi i64 [ %or11.i, %cond.true.i ], [ %or17.i, %cond.false.i ] 688 %shr.29.i = lshr i64 %cond.i, 17 689 %conv18.i = trunc i64 %shr.29.i to i32 690 store i32 %conv18.i, i32* @lock, align 4 691 %rem.i = srem i64 %cond.i, 1050011 692 %conv19.i = trunc i64 %rem.i to i32 693 store i32 %conv19.i, i32* @htindex, align 4 694 %rem20.i = urem i32 %conv18.i, 179 695 %add.i = or i32 %rem20.i, 131072 696 store i32 %add.i, i32* @stride, align 4 697 %8 = load i32*, i32** @ht, align 8 698 %arrayidx = getelementptr inbounds i32, i32* %8, i64 %rem.i 699 %9 = load i32, i32* %arrayidx, align 4 700 %cmp1 = icmp eq i32 %9, %conv18.i 701 br i1 %cmp1, label %if.then, label %if.end 702 703if.then: 704 %idxprom.lcssa = phi i64 [ %rem.i, %hash.exit ], [ %idxprom.1, %if.end ], [ %idxprom.2, %if.end.1 ], [ %idxprom.3, %if.end.2 ], [ %idxprom.4, %if.end.3 ], [ %idxprom.5, %if.end.4 ], [ %idxprom.6, %if.end.5 ], [ %idxprom.7, %if.end.6 ] 705 %10 = load i8*, i8** @he, align 8 706 %arrayidx3 = getelementptr inbounds i8, i8* %10, i64 %idxprom.lcssa 707 %11 = load i8, i8* %arrayidx3, align 1 708 %conv = sext i8 %11 to i32 709 br label %cleanup 710 711if.end: 712 %add = add nsw i32 %add.i, %conv19.i 713 %cmp4 = icmp sgt i32 %add, 1050010 714 %sub = add nsw i32 %add, -1050011 715 %sub.add = select i1 %cmp4, i32 %sub, i32 %add 716 %idxprom.1 = sext i32 %sub.add to i64 717 %arrayidx.1 = getelementptr inbounds i32, i32* %8, i64 %idxprom.1 718 %12 = load i32, i32* %arrayidx.1, align 4 719 %cmp1.1 = icmp eq i32 %12, %conv18.i 720 br i1 %cmp1.1, label %if.then, label %if.end.1 721 722cleanup: 723 %retval.0 = phi i32 [ %conv, %if.then ], [ -128, %if.end.6 ] 724 ret i32 %retval.0 725 726if.end.1: 727 %add.1 = add nsw i32 %add.i, %sub.add 728 %cmp4.1 = icmp sgt i32 %add.1, 1050010 729 %sub.1 = add nsw i32 %add.1, -1050011 730 %sub.add.1 = select i1 %cmp4.1, i32 %sub.1, i32 %add.1 731 %idxprom.2 = sext i32 %sub.add.1 to i64 732 %arrayidx.2 = getelementptr inbounds i32, i32* %8, i64 %idxprom.2 733 %13 = load i32, i32* %arrayidx.2, align 4 734 %cmp1.2 = icmp eq i32 %13, %conv18.i 735 br i1 %cmp1.2, label %if.then, label %if.end.2 736 737if.end.2: 738 %add.2 = add nsw i32 %add.i, %sub.add.1 739 %cmp4.2 = icmp sgt i32 %add.2, 1050010 740 %sub.2 = add nsw i32 %add.2, -1050011 741 %sub.add.2 = select i1 %cmp4.2, i32 %sub.2, i32 %add.2 742 %idxprom.3 = sext i32 %sub.add.2 to i64 743 %arrayidx.3 = getelementptr inbounds i32, i32* %8, i64 %idxprom.3 744 %14 = load i32, i32* %arrayidx.3, align 4 745 %cmp1.3 = icmp eq i32 %14, %conv18.i 746 br i1 %cmp1.3, label %if.then, label %if.end.3 747 748if.end.3: 749 %add.3 = add nsw i32 %add.i, %sub.add.2 750 %cmp4.3 = icmp sgt i32 %add.3, 1050010 751 %sub.3 = add nsw i32 %add.3, -1050011 752 %sub.add.3 = select i1 %cmp4.3, i32 %sub.3, i32 %add.3 753 %idxprom.4 = sext i32 %sub.add.3 to i64 754 %arrayidx.4 = getelementptr inbounds i32, i32* %8, i64 %idxprom.4 755 %15 = load i32, i32* %arrayidx.4, align 4 756 %cmp1.4 = icmp eq i32 %15, %conv18.i 757 br i1 %cmp1.4, label %if.then, label %if.end.4 758 759if.end.4: 760 %add.4 = add nsw i32 %add.i, %sub.add.3 761 %cmp4.4 = icmp sgt i32 %add.4, 1050010 762 %sub.4 = add nsw i32 %add.4, -1050011 763 %sub.add.4 = select i1 %cmp4.4, i32 %sub.4, i32 %add.4 764 %idxprom.5 = sext i32 %sub.add.4 to i64 765 %arrayidx.5 = getelementptr inbounds i32, i32* %8, i64 %idxprom.5 766 %16 = load i32, i32* %arrayidx.5, align 4 767 %cmp1.5 = icmp eq i32 %16, %conv18.i 768 br i1 %cmp1.5, label %if.then, label %if.end.5 769 770if.end.5: 771 %add.5 = add nsw i32 %add.i, %sub.add.4 772 %cmp4.5 = icmp sgt i32 %add.5, 1050010 773 %sub.5 = add nsw i32 %add.5, -1050011 774 %sub.add.5 = select i1 %cmp4.5, i32 %sub.5, i32 %add.5 775 %idxprom.6 = sext i32 %sub.add.5 to i64 776 %arrayidx.6 = getelementptr inbounds i32, i32* %8, i64 %idxprom.6 777 %17 = load i32, i32* %arrayidx.6, align 4 778 %cmp1.6 = icmp eq i32 %17, %conv18.i 779 br i1 %cmp1.6, label %if.then, label %if.end.6 780 781if.end.6: 782 %add.6 = add nsw i32 %add.i, %sub.add.5 783 %cmp4.6 = icmp sgt i32 %add.6, 1050010 784 %sub.6 = add nsw i32 %add.6, -1050011 785 %sub.add.6 = select i1 %cmp4.6, i32 %sub.6, i32 %add.6 786 %idxprom.7 = sext i32 %sub.add.6 to i64 787 %arrayidx.7 = getelementptr inbounds i32, i32* %8, i64 %idxprom.7 788 %18 = load i32, i32* %arrayidx.7, align 4 789 %cmp1.7 = icmp eq i32 %18, %conv18.i 790 br i1 %cmp1.7, label %if.then, label %cleanup 791} 792