Skip to content

Commit c3f9697

Browse files
author
Kai Luo
committed
[PowerPC] Fix wrong codegen when stack pointer has to realign performing dynalloc
Current powerpc backend generates wrong code sequence if stack pointer has to realign if `-fstack-clash-protection` enabled. When probing dynamic stack allocation, current `PREPARE_PROBED_ALLOCA` takes `NegSizeReg` as input and returns `FinalStackPtr`. `FinalStackPtr=StackPtr+ActualNegSize` is calculated correctly, however code following `PREPARE_PROBED_ALLOCA` still uses value of `NegSizeReg`, which does not contain `ActualNegSize` if `MaxAlign > TargetAlign`, to calculate loop trip count and residual number of bytes. This patch is part of fix of https://bugs.llvm.org/show_bug.cgi?id=46759. Differential Revision: https://reviews.llvm.org/D84152
1 parent 8912252 commit c3f9697

File tree

6 files changed

+198
-160
lines changed

6 files changed

+198
-160
lines changed

‎llvm/lib/Target/PowerPC/PPCISelLowering.cpp

Lines changed: 26 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -11954,18 +11954,34 @@ PPCTargetLowering::emitProbedAlloca(MachineInstr &MI,
1195411954
Register SPReg = isPPC64 ? PPC::X1 : PPC::R1;
1195511955
Register FinalStackPtr = MRI.createVirtualRegister(isPPC64 ? G8RC : GPRC);
1195611956
Register FramePointer = MRI.createVirtualRegister(isPPC64 ? G8RC : GPRC);
11957-
11958-
// Get the canonical FinalStackPtr like what
11959-
// PPCRegisterInfo::lowerDynamicAlloc does.
11960-
BuildMI(*MBB, {MI}, DL,
11961-
TII->get(isPPC64 ? PPC::PREPARE_PROBED_ALLOCA_64
11962-
: PPC::PREPARE_PROBED_ALLOCA_32),
11963-
FramePointer)
11964-
.addDef(FinalStackPtr)
11957+
Register ActualNegSizeReg = MRI.createVirtualRegister(isPPC64 ? G8RC : GPRC);
11958+
11959+
// Since value of NegSizeReg might be realigned in prologepilog, insert a
11960+
// PREPARE_PROBED_ALLOCA pseudo instruction to get actual FramePointer and
11961+
// NegSize.
11962+
unsigned ProbeOpc;
11963+
if (!MRI.hasOneNonDBGUse(NegSizeReg))
11964+
ProbeOpc =
11965+
isPPC64 ? PPC::PREPARE_PROBED_ALLOCA_64 : PPC::PREPARE_PROBED_ALLOCA_32;
11966+
else
11967+
// By introducing PREPARE_PROBED_ALLOCA_NEGSIZE_OPT, ActualNegSizeReg
11968+
// and NegSizeReg will be allocated in the same phyreg to avoid
11969+
// redundant copy when NegSizeReg has only one use which is current MI and
11970+
// will be replaced by PREPARE_PROBED_ALLOCA then.
11971+
ProbeOpc = isPPC64 ? PPC::PREPARE_PROBED_ALLOCA_NEGSIZE_SAME_REG_64
11972+
: PPC::PREPARE_PROBED_ALLOCA_NEGSIZE_SAME_REG_32;
11973+
BuildMI(*MBB, {MI}, DL, TII->get(ProbeOpc), FramePointer)
11974+
.addDef(ActualNegSizeReg)
1196511975
.addReg(NegSizeReg)
1196611976
.add(MI.getOperand(2))
1196711977
.add(MI.getOperand(3));
1196811978

11979+
// Calculate final stack pointer, which equals to SP + ActualNegSize.
11980+
BuildMI(*MBB, {MI}, DL, TII->get(isPPC64 ? PPC::ADD8 : PPC::ADD4),
11981+
FinalStackPtr)
11982+
.addReg(SPReg)
11983+
.addReg(ActualNegSizeReg);
11984+
1196911985
// Materialize a scratch register for update.
1197011986
int64_t NegProbeSize = -(int64_t)ProbeSize;
1197111987
assert(isInt<32>(NegProbeSize) && "Unhandled probe size!");
@@ -11986,7 +12002,7 @@ PPCTargetLowering::emitProbedAlloca(MachineInstr &MI,
1198612002
// Probing leading residual part.
1198712003
Register Div = MRI.createVirtualRegister(isPPC64 ? G8RC : GPRC);
1198812004
BuildMI(*MBB, {MI}, DL, TII->get(isPPC64 ? PPC::DIVD : PPC::DIVW), Div)
11989-
.addReg(NegSizeReg)
12005+
.addReg(ActualNegSizeReg)
1199012006
.addReg(ScratchReg);
1199112007
Register Mul = MRI.createVirtualRegister(isPPC64 ? G8RC : GPRC);
1199212008
BuildMI(*MBB, {MI}, DL, TII->get(isPPC64 ? PPC::MULLD : PPC::MULLW), Mul)
@@ -11995,7 +12011,7 @@ PPCTargetLowering::emitProbedAlloca(MachineInstr &MI,
1199512011
Register NegMod = MRI.createVirtualRegister(isPPC64 ? G8RC : GPRC);
1199612012
BuildMI(*MBB, {MI}, DL, TII->get(isPPC64 ? PPC::SUBF8 : PPC::SUBF), NegMod)
1199712013
.addReg(Mul)
11998-
.addReg(NegSizeReg);
12014+
.addReg(ActualNegSizeReg);
1199912015
BuildMI(*MBB, {MI}, DL, TII->get(isPPC64 ? PPC::STDUX : PPC::STWUX), SPReg)
1200012016
.addReg(FramePointer)
1200112017
.addReg(SPReg)

‎llvm/lib/Target/PowerPC/PPCInstr64Bit.td

Lines changed: 7 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -431,9 +431,14 @@ def PROBED_ALLOCA_64 : PPCCustomInserterPseudo<(outs g8rc:$result),
431431
(ins g8rc:$negsize, memri:$fpsi), "#PROBED_ALLOCA_64",
432432
[(set i64:$result,
433433
(PPCprobedalloca i64:$negsize, iaddr:$fpsi))]>;
434-
def PREPARE_PROBED_ALLOCA_64 : PPCEmitTimePseudo<(outs g8rc:$fp,
435-
g8rc:$sp),
434+
def PREPARE_PROBED_ALLOCA_64 : PPCEmitTimePseudo<(outs
435+
g8rc:$fp, g8rc:$actual_negsize),
436436
(ins g8rc:$negsize, memri:$fpsi), "#PREPARE_PROBED_ALLOCA_64", []>;
437+
def PREPARE_PROBED_ALLOCA_NEGSIZE_SAME_REG_64 : PPCEmitTimePseudo<(outs
438+
g8rc:$fp, g8rc:$actual_negsize),
439+
(ins g8rc:$negsize, memri:$fpsi),
440+
"#PREPARE_PROBED_ALLOCA_NEGSIZE_SAME_REG_64", []>,
441+
RegConstraint<"$actual_negsize = $negsize">;
437442
def PROBED_STACKALLOC_64 : PPCEmitTimePseudo<(outs g8rc:$scratch, g8rc:$temp),
438443
(ins i64imm:$stacksize),
439444
"#PROBED_STACKALLOC_64", []>;

‎llvm/lib/Target/PowerPC/PPCInstrInfo.td

Lines changed: 7 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1406,9 +1406,14 @@ def PROBED_ALLOCA_32 : PPCCustomInserterPseudo<(outs gprc:$result),
14061406
(ins gprc:$negsize, memri:$fpsi), "#PROBED_ALLOCA_32",
14071407
[(set i32:$result,
14081408
(PPCprobedalloca i32:$negsize, iaddr:$fpsi))]>;
1409-
def PREPARE_PROBED_ALLOCA_32 : PPCEmitTimePseudo<(outs gprc:$fp,
1410-
gprc:$sp),
1409+
def PREPARE_PROBED_ALLOCA_32 : PPCEmitTimePseudo<(outs
1410+
gprc:$fp, gprc:$actual_negsize),
14111411
(ins gprc:$negsize, memri:$fpsi), "#PREPARE_PROBED_ALLOCA_32", []>;
1412+
def PREPARE_PROBED_ALLOCA_NEGSIZE_SAME_REG_32 : PPCEmitTimePseudo<(outs
1413+
gprc:$fp, gprc:$actual_negsize),
1414+
(ins gprc:$negsize, memri:$fpsi),
1415+
"#PREPARE_PROBED_ALLOCA_NEGSIZE_SAME_REG_32", []>,
1416+
RegConstraint<"$actual_negsize = $negsize">;
14121417
def PROBED_STACKALLOC_32 : PPCEmitTimePseudo<(outs gprc:$scratch, gprc:$temp),
14131418
(ins i64imm:$stacksize),
14141419
"#PROBED_STACKALLOC_32", []>;

‎llvm/lib/Target/PowerPC/PPCRegisterInfo.cpp

Lines changed: 24 additions & 13 deletions
Original file line numberDiff line numberDiff line change
@@ -624,21 +624,30 @@ void PPCRegisterInfo::lowerPrepareProbedAlloca(
624624
bool LP64 = TM.isPPC64();
625625
DebugLoc dl = MI.getDebugLoc();
626626
Register FramePointer = MI.getOperand(0).getReg();
627-
Register FinalStackPtr = MI.getOperand(1).getReg();
627+
const Register ActualNegSizeReg = MI.getOperand(1).getReg();
628628
bool KillNegSizeReg = MI.getOperand(2).isKill();
629629
Register NegSizeReg = MI.getOperand(2).getReg();
630-
prepareDynamicAlloca(II, NegSizeReg, KillNegSizeReg, FramePointer);
631-
if (LP64) {
632-
BuildMI(MBB, II, dl, TII.get(PPC::ADD8), FinalStackPtr)
633-
.addReg(PPC::X1)
634-
.addReg(NegSizeReg, getKillRegState(KillNegSizeReg));
635-
636-
} else {
637-
BuildMI(MBB, II, dl, TII.get(PPC::ADD4), FinalStackPtr)
638-
.addReg(PPC::R1)
639-
.addReg(NegSizeReg, getKillRegState(KillNegSizeReg));
630+
const MCInstrDesc &CopyInst = TII.get(LP64 ? PPC::OR8 : PPC::OR);
631+
// RegAllocator might allocate FramePointer and NegSizeReg in the same phyreg.
632+
if (FramePointer == NegSizeReg) {
633+
assert(KillNegSizeReg && "FramePointer is a def and NegSizeReg is an use, "
634+
"NegSizeReg should be killed");
635+
// FramePointer is clobbered earlier than the use of NegSizeReg in
636+
// prepareDynamicAlloca, save NegSizeReg in ActualNegSizeReg to avoid
637+
// misuse.
638+
BuildMI(MBB, II, dl, CopyInst, ActualNegSizeReg)
639+
.addReg(NegSizeReg)
640+
.addReg(NegSizeReg);
641+
NegSizeReg = ActualNegSizeReg;
642+
KillNegSizeReg = false;
640643
}
641-
644+
prepareDynamicAlloca(II, NegSizeReg, KillNegSizeReg, FramePointer);
645+
// NegSizeReg might be updated in prepareDynamicAlloca if MaxAlign >
646+
// TargetAlign.
647+
if (NegSizeReg != ActualNegSizeReg)
648+
BuildMI(MBB, II, dl, CopyInst, ActualNegSizeReg)
649+
.addReg(NegSizeReg)
650+
.addReg(NegSizeReg);
642651
MBB.erase(II);
643652
}
644653

@@ -1084,7 +1093,9 @@ PPCRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
10841093

10851094
if (FPSI && FrameIndex == FPSI &&
10861095
(OpC == PPC::PREPARE_PROBED_ALLOCA_64 ||
1087-
OpC == PPC::PREPARE_PROBED_ALLOCA_32)) {
1096+
OpC == PPC::PREPARE_PROBED_ALLOCA_32 ||
1097+
OpC == PPC::PREPARE_PROBED_ALLOCA_NEGSIZE_SAME_REG_64 ||
1098+
OpC == PPC::PREPARE_PROBED_ALLOCA_NEGSIZE_SAME_REG_32)) {
10881099
lowerPrepareProbedAlloca(II);
10891100
return;
10901101
}

‎llvm/test/CodeGen/PowerPC/pr46759.ll

Lines changed: 14 additions & 13 deletions
Original file line numberDiff line numberDiff line change
@@ -20,26 +20,27 @@ define void @foo(i32 %vla_size) #0 {
2020
; CHECK-LE-NEXT: .cfi_offset r31, -8
2121
; CHECK-LE-NEXT: .cfi_offset r30, -16
2222
; CHECK-LE-NEXT: clrldi r3, r3, 32
23-
; CHECK-LE-NEXT: li r6, -4096
24-
; CHECK-LE-NEXT: ld r4, 0(r1)
23+
; CHECK-LE-NEXT: li r5, -2048
2524
; CHECK-LE-NEXT: mr r31, r1
2625
; CHECK-LE-NEXT: addi r3, r3, 15
2726
; CHECK-LE-NEXT: rldicl r3, r3, 60, 4
2827
; CHECK-LE-NEXT: rldicl r3, r3, 4, 31
29-
; CHECK-LE-NEXT: neg r5, r3
30-
; CHECK-LE-NEXT: li r3, -2048
31-
; CHECK-LE-NEXT: divd r7, r5, r6
32-
; CHECK-LE-NEXT: and r3, r5, r3
33-
; CHECK-LE-NEXT: add r3, r1, r3
34-
; CHECK-LE-NEXT: mulld r6, r7, r6
35-
; CHECK-LE-NEXT: sub r5, r5, r6
36-
; CHECK-LE-NEXT: stdux r4, r1, r5
37-
; CHECK-LE-NEXT: cmpd r1, r3
28+
; CHECK-LE-NEXT: neg r4, r3
29+
; CHECK-LE-NEXT: ld r3, 0(r1)
30+
; CHECK-LE-NEXT: and r5, r4, r5
31+
; CHECK-LE-NEXT: mr r4, r5
32+
; CHECK-LE-NEXT: li r5, -4096
33+
; CHECK-LE-NEXT: divd r6, r4, r5
34+
; CHECK-LE-NEXT: mulld r5, r6, r5
35+
; CHECK-LE-NEXT: sub r5, r4, r5
36+
; CHECK-LE-NEXT: add r4, r1, r4
37+
; CHECK-LE-NEXT: stdux r3, r1, r5
38+
; CHECK-LE-NEXT: cmpd r1, r4
3839
; CHECK-LE-NEXT: beq cr0, .LBB0_2
3940
; CHECK-LE-NEXT: .LBB0_1: # %entry
4041
; CHECK-LE-NEXT: #
41-
; CHECK-LE-NEXT: stdu r4, -4096(r1)
42-
; CHECK-LE-NEXT: cmpd r1, r3
42+
; CHECK-LE-NEXT: stdu r3, -4096(r1)
43+
; CHECK-LE-NEXT: cmpd r1, r4
4344
; CHECK-LE-NEXT: bne cr0, .LBB0_1
4445
; CHECK-LE-NEXT: .LBB0_2: # %entry
4546
; CHECK-LE-NEXT: addi r3, r1, 2048

0 commit comments

Comments
 (0)