[llvm] r261364 - llvm-dwp: Improve performance (N^2 to amortized N) by using a MapVector instead of linear searches through a vector
David Blaikie via llvm-commits
llvm-commits at lists.llvm.org
Fri Feb 19 13:09:27 PST 2016
Author: dblaikie
Date: Fri Feb 19 15:09:26 2016
New Revision: 261364
URL: http://llvm.org/viewvc/llvm-project?rev=261364&view=rev
Log:
llvm-dwp: Improve performance (N^2 to amortized N) by using a MapVector instead of linear searches through a vector
Figured this would be a problem, but didn't want to jump the gun - large
inputs demonstrate it pretty easily (mostly for type units, but might as
well do the same for CUs too). A random sample 6m27s -> 27s change.
Also, by checking this up-front for CUs (rather than when building the
cu_index) we can probably provide better error messages (see FIXMEs),
hopefully providing the name of the CUs rather than just their
signature.
Modified:
llvm/trunk/tools/llvm-dwp/llvm-dwp.cpp
Modified: llvm/trunk/tools/llvm-dwp/llvm-dwp.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/llvm-dwp/llvm-dwp.cpp?rev=261364&r1=261363&r2=261364&view=diff
==============================================================================
--- llvm/trunk/tools/llvm-dwp/llvm-dwp.cpp (original)
+++ llvm/trunk/tools/llvm-dwp/llvm-dwp.cpp Fri Feb 19 15:09:26 2016
@@ -1,3 +1,4 @@
+#include "llvm/ADT/MapVector.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/StringSet.h"
#include "llvm/CodeGen/AsmPrinter.h"
@@ -24,8 +25,8 @@
#include "llvm/Target/TargetMachine.h"
#include <list>
#include <memory>
-#include <unordered_set>
#include <system_error>
+#include <unordered_set>
using namespace llvm;
using namespace llvm::object;
@@ -136,27 +137,22 @@ static uint64_t getCUSignature(StringRef
}
struct UnitIndexEntry {
- uint64_t Signature;
DWARFUnitIndex::Entry::SectionContribution Contributions[8];
};
-static void addAllTypesFromDWP(MCStreamer &Out,
- std::vector<UnitIndexEntry> &TypeIndexEntries,
- const DWARFUnitIndex &TUIndex,
- MCSection *OutputTypes, StringRef Types,
- const UnitIndexEntry &TUEntry,
- uint32_t &TypesOffset) {
+static void addAllTypesFromDWP(
+ MCStreamer &Out, MapVector<uint64_t, UnitIndexEntry> &TypeIndexEntries,
+ const DWARFUnitIndex &TUIndex, MCSection *OutputTypes, StringRef Types,
+ const UnitIndexEntry &TUEntry, uint32_t &TypesOffset) {
Out.SwitchSection(OutputTypes);
for (const DWARFUnitIndex::Entry &E : TUIndex.getRows()) {
auto *I = E.getOffsets();
if (!I)
continue;
- if (any_of(TypeIndexEntries, [&](const UnitIndexEntry &OldEntry) {
- return OldEntry.Signature == E.getSignature();
- }))
+ auto P = TypeIndexEntries.insert(std::make_pair(E.getSignature(), TUEntry));
+ if (!P.second)
continue;
- auto Entry = TUEntry;
- Entry.Signature = E.getSignature();
+ auto &Entry = P.first->second;
// Zero out the debug_info contribution
Entry.Contributions[0] = {};
for (auto Kind : TUIndex.getColumnKinds()) {
@@ -171,12 +167,11 @@ static void addAllTypesFromDWP(MCStreame
C.Length));
C.Offset = TypesOffset;
TypesOffset += C.Length;
- TypeIndexEntries.push_back(Entry);
}
}
static void addAllTypes(MCStreamer &Out,
- std::vector<UnitIndexEntry> &TypeIndexEntries,
+ MapVector<uint64_t, UnitIndexEntry> &TypeIndexEntries,
MCSection *OutputTypes, StringRef Types,
const UnitIndexEntry &CUEntry, uint32_t &TypesOffset) {
if (Types.empty())
@@ -198,34 +193,32 @@ static void addAllTypes(MCStreamer &Out,
Data.getU16(&Offset); // Version
Data.getU32(&Offset); // Abbrev offset
Data.getU8(&Offset); // Address size
- Entry.Signature = Data.getU64(&Offset);
+ auto Signature = Data.getU64(&Offset);
Offset = PrevOffset + C.Length;
- if (any_of(TypeIndexEntries, [&](const UnitIndexEntry &E) {
- return E.Signature == Entry.Signature;
- }))
+ auto P = TypeIndexEntries.insert(std::make_pair(Signature, Entry));
+ if (!P.second)
continue;
Out.EmitBytes(Types.substr(PrevOffset, C.Length));
TypesOffset += C.Length;
-
- TypeIndexEntries.push_back(Entry);
}
}
static void
writeIndexTable(MCStreamer &Out, ArrayRef<unsigned> ContributionOffsets,
- ArrayRef<UnitIndexEntry> IndexEntries,
+ const MapVector<uint64_t, UnitIndexEntry> &IndexEntries,
uint32_t DWARFUnitIndex::Entry::SectionContribution::*Field) {
for (const auto &E : IndexEntries)
- for (size_t i = 0; i != array_lengthof(E.Contributions); ++i)
+ for (size_t i = 0; i != array_lengthof(E.second.Contributions); ++i)
if (ContributionOffsets[i])
- Out.EmitIntValue(E.Contributions[i].*Field, 4);
+ Out.EmitIntValue(E.second.Contributions[i].*Field, 4);
}
-static void writeIndex(MCStreamer &Out, MCSection *Section,
- ArrayRef<unsigned> ContributionOffsets,
- ArrayRef<UnitIndexEntry> IndexEntries) {
+static void
+writeIndex(MCStreamer &Out, MCSection *Section,
+ ArrayRef<unsigned> ContributionOffsets,
+ const MapVector<uint64_t, UnitIndexEntry> &IndexEntries) {
unsigned Columns = 0;
for (auto &C : ContributionOffsets)
if (C)
@@ -233,15 +226,17 @@ static void writeIndex(MCStreamer &Out,
std::vector<unsigned> Buckets(NextPowerOf2(3 * IndexEntries.size() / 2));
uint64_t Mask = Buckets.size() - 1;
- for (size_t i = 0; i != IndexEntries.size(); ++i) {
- auto S = IndexEntries[i].Signature;
+ size_t i = 0;
+ for (const auto &P : IndexEntries) {
+ auto S = P.first;
auto H = S & Mask;
while (Buckets[H]) {
- assert(S != IndexEntries[Buckets[H] - 1].Signature &&
+ assert(S != IndexEntries.begin()[Buckets[H] - 1].first &&
"Duplicate unit");
H = (H + (((S >> 32) & Mask) | 1)) % Buckets.size();
}
Buckets[H] = i + 1;
+ ++i;
}
Out.SwitchSection(Section);
@@ -252,7 +247,7 @@ static void writeIndex(MCStreamer &Out,
// Write the signatures.
for (const auto &I : Buckets)
- Out.EmitIntValue(I ? IndexEntries[I - 1].Signature : 0, 8);
+ Out.EmitIntValue(I ? IndexEntries.begin()[I - 1].first : 0, 8);
// Write the indexes.
for (const auto &I : Buckets)
@@ -305,8 +300,8 @@ static std::error_code write(MCStreamer
{"debug_cu_index", {CUIndexSection, static_cast<DWARFSectionKind>(0)}},
{"debug_tu_index", {TUIndexSection, static_cast<DWARFSectionKind>(0)}}};
- std::vector<UnitIndexEntry> IndexEntries;
- std::vector<UnitIndexEntry> TypeIndexEntries;
+ MapVector<uint64_t, UnitIndexEntry> IndexEntries;
+ MapVector<uint64_t, UnitIndexEntry> TypeIndexEntries;
StringMap<uint32_t> Strings;
uint32_t StringOffset = 0;
@@ -412,18 +407,19 @@ static std::error_code write(MCStreamer
return make_error_code(std::errc::invalid_argument);
for (const DWARFUnitIndex::Entry &E : CUIndex.getRows()) {
- auto NewEntry = CurEntry;
auto *I = E.getOffsets();
if (!I)
continue;
- NewEntry.Signature = E.getSignature();
+ auto P =
+ IndexEntries.insert(std::make_pair(E.getSignature(), CurEntry));
+ // FIXME: Check P.second and error for duplicate CU signatures
+ auto &NewEntry = P.first->second;
for (auto Kind : CUIndex.getColumnKinds()) {
auto &C = NewEntry.Contributions[Kind - DW_SECT_INFO];
C.Offset += I->Offset;
C.Length = I->Length;
++I;
}
- IndexEntries.push_back(NewEntry);
}
if (!CurTypesSection.empty()) {
@@ -439,11 +435,11 @@ static std::error_code write(MCStreamer
ContributionOffsets[DW_SECT_TYPES - DW_SECT_INFO]);
}
} else {
- CurEntry.Signature = getCUSignature(AbbrevSection, InfoSection);
+ IndexEntries.insert(
+ std::make_pair(getCUSignature(AbbrevSection, InfoSection), CurEntry));
+ // FIXME: Check P.second and error for duplicate CU signatures
addAllTypes(Out, TypeIndexEntries, TypesSection, CurTypesSection,
CurEntry, ContributionOffsets[DW_SECT_TYPES - DW_SECT_INFO]);
-
- IndexEntries.push_back(CurEntry);
}
if (auto Err = writeStringsAndOffsets(Out, Strings, StringOffset,
More information about the llvm-commits
mailing list