[llvm] API for encoding/decoding DWARF discriminators.

Summary:
Added a pair of APIs for encoding/decoding the 3 components of a DWARF discriminator described in http://lists.llvm.org/pipermail/llvm-dev/2016-October/106532.html: the base discriminator, the duplication factor (useful in profile-guided optimization) and the copy index (used to identify copies of code in cases like loop unrolling)

The encoding packs 3 unsigned values in 32 bits. This CL addresses 2 issues:
- communicates overflow back to the user
- supports encoding all 3 components together. Current APIs assume a sequencing of events. For example, creating a new discriminator based on an existing one by changing the base discriminator was not supported.

Reviewers: davidxl, danielcdh, wmi, dblaikie

Reviewed By: dblaikie

Subscribers: zzheng, dmgreen, aprantl, JDevlieghere, llvm-commits

Differential Revision: https://reviews.llvm.org/D55681

llvm-svn: 349973
This commit is contained in:
Mircea Trofin 2018-12-21 22:48:50 +00:00
parent 8d20cfdfc6
commit b53eeb6f4c
8 changed files with 284 additions and 56 deletions

View File

@ -1430,6 +1430,9 @@ class DILocation : public MDNode {
/// Reverse transformation as getPrefixEncodingFromUnsigned.
static unsigned getUnsignedFromPrefixEncoding(unsigned U) {
if (U & 1)
return 0;
U >>= 1;
return (U & 0x20) ? (((U >> 1) & 0xfe0) | (U & 0x1f)) : (U & 0x1f);
}
@ -1448,6 +1451,14 @@ class DILocation : public MDNode {
getRawInlinedAt(), isImplicitCode());
}
static unsigned encodeComponent(unsigned C) {
return (C == 0) ? 1U : (getPrefixEncodingFromUnsigned(C) << 1);
}
static unsigned encodingBits(unsigned C) {
return (C == 0) ? 1 : (C > 0x1f ? 14 : 7);
}
public:
// Disallow replacing operands.
void replaceOperandWith(unsigned I, Metadata *New) = delete;
@ -1518,20 +1529,35 @@ public:
/// order. If the lowest bit is 1, the current component is empty, and the
/// next component will start in the next bit. Otherwise, the current
/// component is non-empty, and its content starts in the next bit. The
/// length of each components is either 5 bit or 12 bit: if the 7th bit
/// value of each components is either 5 bit or 12 bit: if the 7th bit
/// is 0, the bit 2~6 (5 bits) are used to represent the component; if the
/// 7th bit is 1, the bit 2~6 (5 bits) and 8~14 (7 bits) are combined to
/// represent the component.
/// represent the component. Thus, the number of bits used for a component
/// is either 0 (if it and all the next components are empty); 1 - if it is
/// empty; 7 - if its value is up to and including 0x1f (lsb and msb are both
/// 0); or 14, if its value is up to and including 0x1ff. Note that the last
/// component is also capped at 0x1ff, even in the case when both first
/// components are 0, and we'd technically have 29 bits available.
///
/// For precise control over the data being encoded in the discriminator,
/// use encodeDiscriminator/decodeDiscriminator.
///
/// Use {get|set}BaseDiscriminator and cloneWithDuplicationFactor after reading
/// their documentation, as their behavior has side-effects.
inline unsigned getDiscriminator() const;
/// Returns a new DILocation with updated \p Discriminator.
inline const DILocation *cloneWithDiscriminator(unsigned Discriminator) const;
/// Returns a new DILocation with updated base discriminator \p BD.
inline const DILocation *setBaseDiscriminator(unsigned BD) const;
/// Returns a new DILocation with updated base discriminator \p BD. Only the
/// base discriminator is set in the new DILocation, the other encoded values
/// are elided.
/// If the discriminator cannot be encoded, the function returns None.
inline Optional<const DILocation *> setBaseDiscriminator(unsigned BD) const;
/// Returns the duplication factor stored in the discriminator.
/// Returns the duplication factor stored in the discriminator, or 1 if no
/// duplication factor (or 0) is encoded.
inline unsigned getDuplicationFactor() const;
/// Returns the copy identifier stored in the discriminator.
@ -1540,9 +1566,11 @@ public:
/// Returns the base discriminator stored in the discriminator.
inline unsigned getBaseDiscriminator() const;
/// Returns a new DILocation with duplication factor \p DF encoded in the
/// discriminator.
inline const DILocation *cloneWithDuplicationFactor(unsigned DF) const;
/// Returns a new DILocation with duplication factor \p DF * current
/// duplication factor encoded in the discriminator. The current duplication
/// factor is as defined by getDuplicationFactor().
/// Returns None if encoding failed.
inline Optional<const DILocation *> cloneWithDuplicationFactor(unsigned DF) const;
/// When two instructions are combined into a single instruction we also
/// need to combine the original locations into a single location.
@ -1563,19 +1591,31 @@ public:
/// Returns the base discriminator for a given encoded discriminator \p D.
static unsigned getBaseDiscriminatorFromDiscriminator(unsigned D) {
if ((D & 1) == 0)
return getUnsignedFromPrefixEncoding(D >> 1);
else
return 0;
return getUnsignedFromPrefixEncoding(D);
}
/// Returns the duplication factor for a given encoded discriminator \p D.
/// Raw encoding of the discriminator. APIs such as setBaseDiscriminator or
/// cloneWithDuplicationFactor have certain side-effects. This API, in
/// conjunction with cloneWithDiscriminator, may be used to encode precisely
/// the values provided. \p BD: base discriminator \p DF: duplication factor
/// \p CI: copy index
/// The return is None if the values cannot be encoded in 32 bits - for
/// example, values for BD or DF larger than 12 bits. Otherwise, the return
/// is the encoded value.
static Optional<unsigned> encodeDiscriminator(unsigned BD, unsigned DF, unsigned CI);
/// Raw decoder for values in an encoded discriminator D.
static void decodeDiscriminator(unsigned D, unsigned &BD, unsigned &DF,
unsigned &CI);
/// Returns the duplication factor for a given encoded discriminator \p D, or
/// 1 if no value or 0 is encoded.
static unsigned getDuplicationFactorFromDiscriminator(unsigned D) {
D = getNextComponentInDiscriminator(D);
if (D == 0 || (D & 1))
unsigned Ret = getUnsignedFromPrefixEncoding(D);
if (Ret == 0)
return 1;
else
return getUnsignedFromPrefixEncoding(D >> 1);
return Ret;
}
/// Returns the copy identifier for a given encoded discriminator \p D.
@ -1999,28 +2039,24 @@ unsigned DILocation::getCopyIdentifier() const {
return getCopyIdentifierFromDiscriminator(getDiscriminator());
}
const DILocation *DILocation::setBaseDiscriminator(unsigned D) const {
Optional<const DILocation *> DILocation::setBaseDiscriminator(unsigned D) const {
if (D == 0)
return this;
else
return cloneWithDiscriminator(getPrefixEncodingFromUnsigned(D) << 1);
if (D > 0xfff)
return None;
return cloneWithDiscriminator(encodeComponent(D));
}
const DILocation *DILocation::cloneWithDuplicationFactor(unsigned DF) const {
Optional<const DILocation *> DILocation::cloneWithDuplicationFactor(unsigned DF) const {
DF *= getDuplicationFactor();
if (DF <= 1)
return this;
unsigned BD = getBaseDiscriminator();
unsigned CI = getCopyIdentifier() << (DF > 0x1f ? 14 : 7);
unsigned D = CI | (getPrefixEncodingFromUnsigned(DF) << 1);
if (BD == 0)
D = (D << 1) | 1;
else
D = (D << (BD > 0x1f ? 14 : 7)) | (getPrefixEncodingFromUnsigned(BD) << 1);
return cloneWithDiscriminator(D);
unsigned CI = getCopyIdentifier();
if (Optional<unsigned> D = encodeDiscriminator(BD, DF, CI))
return cloneWithDiscriminator(*D);
return None;
}
class DINamespace : public DIScope {

View File

@ -20,6 +20,8 @@
#include "llvm/IR/Function.h"
#include "llvm/IR/Instructions.h"
#include <numeric>
using namespace llvm;
DILocation::DILocation(LLVMContext &C, StorageType Storage, unsigned Line,
@ -113,6 +115,47 @@ const DILocation *DILocation::getMergedLocation(const DILocation *LocA,
return DILocation::get(Result->getContext(), 0, 0, S, L);
}
Optional<unsigned> DILocation::encodeDiscriminator(unsigned BD, unsigned DF, unsigned CI) {
SmallVector<unsigned, 3> Components = {BD, DF, CI};
uint64_t RemainingWork = 0U;
// We use RemainingWork to figure out if we have no remaining components to
// encode. For example: if BD != 0 but DF == 0 && CI == 0, we don't need to
// encode anything for the latter 2.
// Since any of the input components is at most 32 bits, their sum will be
// less than 34 bits, and thus RemainingWork won't overflow.
RemainingWork = std::accumulate(Components.begin(), Components.end(), RemainingWork);
int I = 0;
unsigned Ret = 0;
unsigned NextBitInsertionIndex = 0;
while (RemainingWork > 0) {
unsigned C = Components[I++];
RemainingWork -= C;
unsigned EC = encodeComponent(C);
Ret |= (EC << NextBitInsertionIndex);
NextBitInsertionIndex += encodingBits(C);
}
// Encoding may be unsuccessful because of overflow. We determine success by
// checking equivalence of components before & after encoding. Alternatively,
// we could determine Success during encoding, but the current alternative is
// simpler.
unsigned TBD, TDF, TCI = 0;
decodeDiscriminator(Ret, TBD, TDF, TCI);
if (TBD == BD && TDF == DF && TCI == CI)
return Ret;
return None;
}
void DILocation::decodeDiscriminator(unsigned D, unsigned &BD, unsigned &DF,
unsigned &CI) {
BD = getUnsignedFromPrefixEncoding(D);
DF = getUnsignedFromPrefixEncoding(getNextComponentInDiscriminator(D));
CI = getUnsignedFromPrefixEncoding(
getNextComponentInDiscriminator(getNextComponentInDiscriminator(D)));
}
DINode::DIFlags DINode::getFlag(StringRef Flag) {
return StringSwitch<DIFlags>(Flag)
#define HANDLE_DI_FLAG(ID, NAME) .Case("DIFlag" #NAME, Flag##NAME)

View File

@ -21,6 +21,7 @@
#include "llvm/IR/DebugInfoMetadata.h"
#include "llvm/ProfileData/SampleProf.h"
#include "llvm/ProfileData/SampleProfReader.h"
#include "llvm/Support/Debug.h"
#include "llvm/Transforms/IPO/SampleProfile.h"
using namespace llvm;
@ -107,27 +108,36 @@ bool X86DiscriminateMemOps::runOnMachineFunction(MachineFunction &MF) {
if (!DI) {
DI = ReferenceDI;
}
DenseSet<unsigned> &Set = Seen[diToLocation(DI)];
Location L = diToLocation(DI);
DenseSet<unsigned> &Set = Seen[L];
const std::pair<DenseSet<unsigned>::iterator, bool> TryInsert =
Set.insert(DI->getBaseDiscriminator());
if (!TryInsert.second) {
DI = DI->setBaseDiscriminator(++MemOpDiscriminators[diToLocation(DI)]);
updateDebugInfo(&MI, DI);
Changed = true;
const std::pair<DenseSet<unsigned>::iterator, bool> MustInsert =
Set.insert(DI->getBaseDiscriminator());
// FIXME (mtrofin): check if the to-be inserted base discriminator can
// be added. This requires a new API on DILocation.
// The assumption is that this scenario is infrequent/OK not to support.
// If evidence points otherwise, we can explore synthesize unique DIs by
// adding fake line numbers.
if (!MustInsert.second) {
LLVM_DEBUG(dbgs()
<< "Unable to create a unique discriminator in "
unsigned BF, DF, CI = 0;
DILocation::decodeDiscriminator(DI->getDiscriminator(), BF, DF, CI);
Optional<unsigned> EncodedDiscriminator = DILocation::encodeDiscriminator(
MemOpDiscriminators[L] + 1, DF, CI);
if (!EncodedDiscriminator) {
// FIXME(mtrofin): The assumption is that this scenario is infrequent/OK
// not to support. If evidence points otherwise, we can explore synthesizeing
// unique DIs by adding fake line numbers, or by constructing 64 bit
// discriminators.
LLVM_DEBUG(dbgs() << "Unable to create a unique discriminator "
"for instruction with memory operand in: "
<< DI->getFilename() << " Line: " << DI->getLine()
<< " Column: " << DI->getColumn()
<< ". This is likely due to a large macro expansion.\n");
<< ". This is likely due to a large macro expansion. \n");
continue;
}
// Since we were able to encode, bump the MemOpDiscriminators.
++MemOpDiscriminators[L];
DI = DI->cloneWithDiscriminator(EncodedDiscriminator.getValue());
updateDebugInfo(&MI, DI);
Changed = true;
std::pair<DenseSet<unsigned>::iterator, bool> MustInsert =
Set.insert(DI->getBaseDiscriminator());
assert(MustInsert.second && "New discriminator shouldn't be present in set");
}
// Bump the reference DI to avoid cramming discriminators on line 0.

View File

@ -209,10 +209,18 @@ static bool addDiscriminators(Function &F) {
// Only the lowest 7 bits are used to represent a discriminator to fit
// it in 1 byte ULEB128 representation.
unsigned Discriminator = R.second ? ++LDM[L] : LDM[L];
I.setDebugLoc(DIL->setBaseDiscriminator(Discriminator));
LLVM_DEBUG(dbgs() << DIL->getFilename() << ":" << DIL->getLine() << ":"
<< DIL->getColumn() << ":" << Discriminator << " " << I
<< "\n");
auto NewDIL = DIL->setBaseDiscriminator(Discriminator);
if (!NewDIL) {
LLVM_DEBUG(dbgs() << "Could not encode discriminator: "
<< DIL->getFilename() << ":" << DIL->getLine() << ":"
<< DIL->getColumn() << ":" << Discriminator << " "
<< I << "\n");
} else {
I.setDebugLoc(NewDIL.getValue());
LLVM_DEBUG(dbgs() << DIL->getFilename() << ":" << DIL->getLine() << ":"
<< DIL->getColumn() << ":" << Discriminator << " " << I
<< "\n");
}
Changed = true;
}
}
@ -239,8 +247,17 @@ static bool addDiscriminators(Function &F) {
std::make_pair(CurrentDIL->getFilename(), CurrentDIL->getLine());
if (!CallLocations.insert(L).second) {
unsigned Discriminator = ++LDM[L];
Current->setDebugLoc(CurrentDIL->setBaseDiscriminator(Discriminator));
Changed = true;
auto NewDIL = CurrentDIL->setBaseDiscriminator(Discriminator);
if (!NewDIL) {
LLVM_DEBUG(dbgs()
<< "Could not encode discriminator: "
<< CurrentDIL->getFilename() << ":"
<< CurrentDIL->getLine() << ":" << CurrentDIL->getColumn()
<< ":" << Discriminator << " " << I << "\n");
} else {
Current->setDebugLoc(NewDIL.getValue());
Changed = true;
}
}
}
}

View File

@ -598,8 +598,15 @@ LoopUnrollResult llvm::UnrollLoop(
for (BasicBlock *BB : L->getBlocks())
for (Instruction &I : *BB)
if (!isa<DbgInfoIntrinsic>(&I))
if (const DILocation *DIL = I.getDebugLoc())
I.setDebugLoc(DIL->cloneWithDuplicationFactor(Count));
if (const DILocation *DIL = I.getDebugLoc()) {
auto NewDIL = DIL->cloneWithDuplicationFactor(Count);
if (NewDIL)
I.setDebugLoc(NewDIL.getValue());
else
LLVM_DEBUG(dbgs()
<< "Failed to create new discriminator: "
<< DIL->getFilename() << " Line: " << DIL->getLine());
}
for (unsigned It = 1; It != Count; ++It) {
std::vector<BasicBlock*> NewBlocks;

View File

@ -300,8 +300,15 @@ LoopUnrollResult llvm::UnrollAndJamLoop(
for (BasicBlock *BB : L->getBlocks())
for (Instruction &I : *BB)
if (!isa<DbgInfoIntrinsic>(&I))
if (const DILocation *DIL = I.getDebugLoc())
I.setDebugLoc(DIL->cloneWithDuplicationFactor(Count));
if (const DILocation *DIL = I.getDebugLoc()) {
auto NewDIL = DIL->cloneWithDuplicationFactor(Count);
if (NewDIL)
I.setDebugLoc(NewDIL.getValue());
else
LLVM_DEBUG(dbgs()
<< "Failed to create new discriminator: "
<< DIL->getFilename() << " Line: " << DIL->getLine());
}
// Copy all blocks
for (unsigned It = 1; It != Count; ++It) {

View File

@ -759,8 +759,15 @@ void InnerLoopVectorizer::setDebugLocFromInst(IRBuilder<> &B, const Value *Ptr)
if (const Instruction *Inst = dyn_cast_or_null<Instruction>(Ptr)) {
const DILocation *DIL = Inst->getDebugLoc();
if (DIL && Inst->getFunction()->isDebugInfoForProfiling() &&
!isa<DbgInfoIntrinsic>(Inst))
B.SetCurrentDebugLocation(DIL->cloneWithDuplicationFactor(UF * VF));
!isa<DbgInfoIntrinsic>(Inst)) {
auto NewDIL = DIL->cloneWithDuplicationFactor(UF * VF);
if (NewDIL)
B.SetCurrentDebugLocation(NewDIL.getValue());
else
LLVM_DEBUG(dbgs()
<< "Failed to create new discriminator: "
<< DIL->getFilename() << " Line: " << DIL->getLine());
}
else
B.SetCurrentDebugLocation(DIL);
} else

View File

@ -981,6 +981,107 @@ TEST_F(DILocationTest, cloneTemporary) {
EXPECT_TRUE(L2->isTemporary());
}
TEST_F(DILocationTest, discriminatorEncoding) {
EXPECT_EQ(0U, DILocation::encodeDiscriminator(0, 0, 0).getValue());
// Encode base discriminator as a component: lsb is 0, then the value.
// The other components are all absent, so we leave all the other bits 0.
EXPECT_EQ(2U, DILocation::encodeDiscriminator(1, 0, 0).getValue());
// Base discriminator component is empty, so lsb is 1. Next component is not
// empty, so its lsb is 0, then its value (1). Next component is empty.
// So the bit pattern is 101.
EXPECT_EQ(5U, DILocation::encodeDiscriminator(0, 1, 0).getValue());
// First 2 components are empty, so the bit pattern is 11. Then the
// next component - ending up with 1011.
EXPECT_EQ(0xbU, DILocation::encodeDiscriminator(0, 0, 1).getValue());
// The bit pattern for the first 2 components is 11. The next bit is 0,
// because the last component is not empty. We have 29 bits usable for
// encoding, but we cap it at 12 bits uniformously for all components. We
// encode the last component over 14 bits.
EXPECT_EQ(0xfffbU, DILocation::encodeDiscriminator(0, 0, 0xfff).getValue());
EXPECT_EQ(0x102U, DILocation::encodeDiscriminator(1, 1, 0).getValue());
EXPECT_EQ(0x13eU, DILocation::encodeDiscriminator(0x1f, 1, 0).getValue());
EXPECT_EQ(0x87feU, DILocation::encodeDiscriminator(0x1ff, 1, 0).getValue());
EXPECT_EQ(0x1f3eU, DILocation::encodeDiscriminator(0x1f, 0x1f, 0).getValue());
EXPECT_EQ(0x3ff3eU,
DILocation::encodeDiscriminator(0x1f, 0x1ff, 0).getValue());
EXPECT_EQ(0x1ff87feU,
DILocation::encodeDiscriminator(0x1ff, 0x1ff, 0).getValue());
EXPECT_EQ(0xfff9f3eU,
DILocation::encodeDiscriminator(0x1f, 0x1f, 0xfff).getValue());
EXPECT_EQ(0xffc3ff3eU,
DILocation::encodeDiscriminator(0x1f, 0x1ff, 0x1ff).getValue());
EXPECT_EQ(0xffcf87feU,
DILocation::encodeDiscriminator(0x1ff, 0x1f, 0x1ff).getValue());
EXPECT_EQ(0xe1ff87feU,
DILocation::encodeDiscriminator(0x1ff, 0x1ff, 7).getValue());
}
TEST_F(DILocationTest, discriminatorEncodingNegativeTests) {
EXPECT_EQ(None, DILocation::encodeDiscriminator(0, 0, 0x1000));
EXPECT_EQ(None, DILocation::encodeDiscriminator(0x1000, 0, 0));
EXPECT_EQ(None, DILocation::encodeDiscriminator(0, 0x1000, 0));
EXPECT_EQ(None, DILocation::encodeDiscriminator(0, 0, 0x1000));
EXPECT_EQ(None, DILocation::encodeDiscriminator(0x1ff, 0x1ff, 8));
EXPECT_EQ(None,
DILocation::encodeDiscriminator(std::numeric_limits<uint32_t>::max(),
std::numeric_limits<uint32_t>::max(),
0));
}
TEST_F(DILocationTest, discriminatorSpecialCases) {
// We don't test getCopyIdentifier here because the only way
// to set it is by constructing an encoded discriminator using
// encodeDiscriminator, which is already tested.
auto L1 = DILocation::get(Context, 1, 2, getSubprogram());
EXPECT_EQ(0U, L1->getBaseDiscriminator());
EXPECT_EQ(1U, L1->getDuplicationFactor());
auto L2 = L1->setBaseDiscriminator(1).getValue();
EXPECT_EQ(0U, L1->getBaseDiscriminator());
EXPECT_EQ(1U, L1->getDuplicationFactor());
EXPECT_EQ(1U, L2->getBaseDiscriminator());
EXPECT_EQ(1U, L2->getDuplicationFactor());
auto L3 = L2->cloneWithDuplicationFactor(2).getValue();
EXPECT_EQ(1U, L3->getBaseDiscriminator());
EXPECT_EQ(2U, L3->getDuplicationFactor());
auto L4 = L3->cloneWithDuplicationFactor(4).getValue();
EXPECT_EQ(1U, L4->getBaseDiscriminator());
EXPECT_EQ(8U, L4->getDuplicationFactor());
auto L5 = L4->setBaseDiscriminator(2).getValue();
EXPECT_EQ(2U, L5->getBaseDiscriminator());
EXPECT_EQ(1U, L5->getDuplicationFactor());
// Check extreme cases
auto L6 = L1->setBaseDiscriminator(0xfff).getValue();
EXPECT_EQ(0xfffU, L6->getBaseDiscriminator());
EXPECT_EQ(
0xfffU,
L6->cloneWithDuplicationFactor(0xfff).getValue()->getDuplicationFactor());
// Check we return None for unencodable cases.
EXPECT_EQ(None, L4->setBaseDiscriminator(0x1000));
EXPECT_EQ(None, L4->cloneWithDuplicationFactor(0x1000));
}
typedef MetadataTest GenericDINodeTest;
TEST_F(GenericDINodeTest, get) {