maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[GlobalISel] fix widenScalarUnmerge if widen type is not a multiple of destination type 

Fix creation of illegal unmerge when widen was requested to a type which
is not a multiple of the destination type. E.g. when trying to widen
an s48 unmerge to s64 the existing code would create an illegal unmerge
from s64 to s48.

Instead, create further unmerges to a GCD type, then use this to remerge
these intermediate results to the actual destinations.

Reviewed By: arsenm

Differential Revision: https://reviews.llvm.org/D88422
This commit is contained in:
Dominik Montada 2020-09-28 16:38:35 +02:00
parent 074ab233ed
commit 113114a5da
2 changed files with 91 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

69
llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
View File

@ -1605,35 +1605,60 @@ LegalizerHelper::widenScalarUnmergeValues(MachineInstr &MI, unsigned TypeIdx,
auto Unmerge = MIRBuilder.buildUnmerge(WideTy, WideSrc);
// Create a sequence of unmerges to the original results. since we may have
// widened the source, we will need to pad the results with dead defs to cover
// the source register.
// e.g. widen s16 to s32:
// %1:_(s16), %2:_(s16), %3:_(s16) = G_UNMERGE_VALUES %0:_(s48)
// Create a sequence of unmerges and merges to the original results. Since we
// may have widened the source, we will need to pad the results with dead defs
// to cover the source register.
// e.g. widen s48 to s64:
// %1:_(s48), %2:_(s48) = G_UNMERGE_VALUES %0:_(s96)
//
// =>
// %4:_(s64) = G_ANYEXT %0:_(s48)
// %5:_(s32), %6:_(s32) = G_UNMERGE_VALUES %4 ; Requested unmerge
// %1:_(s16), %2:_(s16) = G_UNMERGE_VALUES %5 ; unpack to original regs
// %3:_(s16), dead %7 = G_UNMERGE_VALUES %6 ; original reg + extra dead def
// %4:_(s192) = G_ANYEXT %0:_(s96)
// %5:_(s64), %6, %7 = G_UNMERGE_VALUES %4 ; Requested unmerge
// ; unpack to GCD type, with extra dead defs
// %8:_(s16), %9, %10, %11 = G_UNMERGE_VALUES %5:_(s64)
// %12:_(s16), %13, dead %14, dead %15 = G_UNMERGE_VALUES %6:_(s64)
// dead %16:_(s16), dead %17, dead %18, dead %18 = G_UNMERGE_VALUES %7:_(s64)
// %1:_(s48) = G_MERGE_VALUES %8:_(s16), %9, %10 ; Remerge to destination
// %2:_(s48) = G_MERGE_VALUES %11:_(s16), %12, %13 ; Remerge to destination
const LLT GCDTy = getGCDType(WideTy, DstTy);
const int NumUnmerge = Unmerge->getNumOperands() - 1;
const int PartsPerUnmerge = WideTy.getSizeInBits() / DstTy.getSizeInBits();
const int PartsPerRemerge = DstTy.getSizeInBits() / GCDTy.getSizeInBits();
for (int I = 0; I != NumUnmerge; ++I) {
auto MIB = MIRBuilder.buildInstr(TargetOpcode::G_UNMERGE_VALUES);
// Directly unmerge to the destination without going through a GCD type
// if possible
if (PartsPerRemerge == 1) {
const int PartsPerUnmerge = WideTy.getSizeInBits() / DstTy.getSizeInBits();
for (int J = 0; J != PartsPerUnmerge; ++J) {
int Idx = I * PartsPerUnmerge + J;
if (Idx < NumDst)
MIB.addDef(MI.getOperand(Idx).getReg());
else {
// Create dead def for excess components.
MIB.addDef(MRI.createGenericVirtualRegister(DstTy));
for (int I = 0; I != NumUnmerge; ++I) {
auto MIB = MIRBuilder.buildInstr(TargetOpcode::G_UNMERGE_VALUES);
for (int J = 0; J != PartsPerUnmerge; ++J) {
int Idx = I * PartsPerUnmerge + J;
if (Idx < NumDst)
MIB.addDef(MI.getOperand(Idx).getReg());
else {
// Create dead def for excess components.
MIB.addDef(MRI.createGenericVirtualRegister(DstTy));
}
}
}
MIB.addUse(Unmerge.getReg(I));
MIB.addUse(Unmerge.getReg(I));
}
} else {
SmallVector<Register, 16> Parts;
for (int J = 0; J != NumUnmerge; ++J)
extractGCDType(Parts, GCDTy, Unmerge.getReg(J));
SmallVector<Register, 8> RemergeParts;
for (int I = 0; I != NumDst; ++I) {
for (int J = 0; J < PartsPerRemerge; ++J) {
const int Idx = I * PartsPerRemerge + J;
RemergeParts.emplace_back(Parts[Idx]);
}
MIRBuilder.buildMerge(MI.getOperand(I).getReg(), RemergeParts);
RemergeParts.clear();
}
}
MI.eraseFromParent();

44
llvm/unittests/CodeGen/GlobalISel/LegalizerHelperTest.cpp
View File

@ -3135,4 +3135,48 @@ TEST_F(AArch64GISelMITest, FewerElementsInsertVectorElt) {
EXPECT_TRUE(CheckMachineFunction(*MF, CheckStr)) << *MF;
}
// Test widen scalar of G_UNMERGE_VALUES
TEST_F(AArch64GISelMITest, widenScalarUnmerge) {
setUp();
if (!TM)
return;
DefineLegalizerInfo(A, {});
LLT S96{LLT::scalar(96)};
LLT S64{LLT::scalar(64)};
LLT S48{LLT::scalar(48)};
auto Src = B.buildAnyExt(S96, Copies[0]);
auto Unmerge = B.buildUnmerge(S48, Src);
AInfo Info(MF->getSubtarget());
DummyGISelObserver Observer;
LegalizerHelper Helper(*MF, Info, Observer, B);
// Perform Legalization
B.setInsertPt(*EntryMBB, Unmerge->getIterator());
// This should create unmerges to a GCD type (S16), then remerge to S48
EXPECT_EQ(LegalizerHelper::LegalizeResult::Legalized,
Helper.widenScalar(*Unmerge, 0, S64));
const auto *CheckStr = R"(
CHECK: [[COPY0:%[0-9]+]]:_(s64) = COPY
CHECK: [[COPY1:%[0-9]+]]:_(s64) = COPY
CHECK: [[COPY2:%[0-9]+]]:_(s64) = COPY
CHECK: [[ANYEXT:%[0-9]+]]:_(s96) = G_ANYEXT [[COPY0]]
CHECK: [[ANYEXT1:%[0-9]+]]:_(s192) = G_ANYEXT [[ANYEXT]]
CHECK: [[UNMERGE:%[0-9]+]]:_(s64), [[UNMERGE1:%[0-9]+]]:_(s64), [[UNMERGE2:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[ANYEXT1]]
CHECK: [[UNMERGE3:%[0-9]+]]:_(s16), [[UNMERGE4:%[0-9]+]]:_(s16), [[UNMERGE5:%[0-9]+]]:_(s16), [[UNMERGE6:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UNMERGE]]
CHECK: [[UNMERGE7:%[0-9]+]]:_(s16), [[UNMERGE8:%[0-9]+]]:_(s16), [[UNMERGE9:%[0-9]+]]:_(s16), [[UNMERGE10:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UNMERGE1]]
CHECK: [[UNMERGE11:%[0-9]+]]:_(s16), [[UNMERGE12:%[0-9]+]]:_(s16), [[UNMERGE13:%[0-9]+]]:_(s16), [[UNMERGE14:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UNMERGE2]]
CHECK: [[MERGE:%[0-9]+]]:_(s48) = G_MERGE_VALUES [[UNMERGE3]]:_(s16), [[UNMERGE4]]:_(s16), [[UNMERGE5]]:_(s16)
CHECK: [[MERGE1:%[0-9]+]]:_(s48) = G_MERGE_VALUES [[UNMERGE6]]:_(s16), [[UNMERGE7]]:_(s16), [[UNMERGE8]]:_(s16)
)";
// Check
EXPECT_TRUE(CheckMachineFunction(*MF, CheckStr)) << *MF;
}
} // namespace