[IR2Vec] Scale embeddings once in vocab analysis instead of repetitive scaling #143986
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This was referenced Jun 12, 2025
|
Warning This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
This stack of pull requests is managed by Graphite. Learn more about stacking. |
svkeerthy
changed the title
svkeerthy
requested review from
boomanaiden154,
jdoerfert,
kazutakahirata,
mtrofin and
snehasish
|
@llvm/pr-subscribers-mlgo @llvm/pr-subscribers-llvm-analysis Author: S. VenkataKeerthy (svkeerthy) ChangesChanges to scale opcodes, types and args once in (Tracking issue - #141817 ; partly fixes - #141832) Patch is 149.98 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/143986.diff 16 Files Affected:
diff --git a/llvm/include/llvm/Analysis/IR2Vec.h b/llvm/include/llvm/Analysis/IR2Vec.h
index de67955d85d7c..f1aaf4cd2e013 100644
--- a/llvm/include/llvm/Analysis/IR2Vec.h
+++ b/llvm/include/llvm/Analysis/IR2Vec.h
@@ -108,6 +108,7 @@ struct Embedding {
/// Arithmetic operators
Embedding &operator+=(const Embedding &RHS);
Embedding &operator-=(const Embedding &RHS);
+ Embedding &operator*=(double Factor);
/// Adds Src Embedding scaled by Factor with the called Embedding.
/// Called_Embedding += Src * Factor
@@ -116,6 +117,8 @@ struct Embedding {
/// Returns true if the embedding is approximately equal to the RHS embedding
/// within the specified tolerance.
bool approximatelyEquals(const Embedding &RHS, double Tolerance = 1e-6) const;
+
+ void print(raw_ostream &OS) const;
};
using InstEmbeddingsMap = DenseMap<const Instruction *, Embedding>;
@@ -234,6 +237,8 @@ class IR2VecVocabResult {
class IR2VecVocabAnalysis : public AnalysisInfoMixin<IR2VecVocabAnalysis> {
ir2vec::Vocab Vocabulary;
Error readVocabulary();
+ Error parseVocabSection(const char *Key, const json::Value ParsedVocabValue,
+ ir2vec::Vocab &TargetVocab, unsigned &Dim);
void emitError(Error Err, LLVMContext &Ctx);
public:
@@ -249,7 +254,6 @@ class IR2VecVocabAnalysis : public AnalysisInfoMixin<IR2VecVocabAnalysis> {
/// functions.
class IR2VecPrinterPass : public PassInfoMixin<IR2VecPrinterPass> {
raw_ostream &OS;
- void printVector(const ir2vec::Embedding &Vec) const;
public:
explicit IR2VecPrinterPass(raw_ostream &OS) : OS(OS) {}
@@ -257,6 +261,16 @@ class IR2VecPrinterPass : public PassInfoMixin<IR2VecPrinterPass> {
static bool isRequired() { return true; }
};
+/// This pass prints the embeddings in the vocabulary
+class IR2VecVocabPrinterPass : public PassInfoMixin<IR2VecVocabPrinterPass> {
+ raw_ostream &OS;
+
+public:
+ explicit IR2VecVocabPrinterPass(raw_ostream &OS) : OS(OS) {}
+ PreservedAnalyses run(Module &M, ModuleAnalysisManager &MAM);
+ static bool isRequired() { return true; }
+};
+
} // namespace llvm
#endif // LLVM_ANALYSIS_IR2VEC_H
diff --git a/llvm/lib/Analysis/IR2Vec.cpp b/llvm/lib/Analysis/IR2Vec.cpp
index fa38c35796a0e..f51d3252d6606 100644
--- a/llvm/lib/Analysis/IR2Vec.cpp
+++ b/llvm/lib/Analysis/IR2Vec.cpp
@@ -85,6 +85,12 @@ Embedding &Embedding::operator-=(const Embedding &RHS) {
return *this;
}
+Embedding &Embedding::operator*=(double Factor) {
+ std::transform(this->begin(), this->end(), this->begin(),
+ [Factor](double Elem) { return Elem * Factor; });
+ return *this;
+}
+
Embedding &Embedding::scaleAndAdd(const Embedding &Src, float Factor) {
assert(this->size() == Src.size() && "Vectors must have the same dimension");
for (size_t Itr = 0; Itr < this->size(); ++Itr)
@@ -101,6 +107,13 @@ bool Embedding::approximatelyEquals(const Embedding &RHS,
return true;
}
+void Embedding::print(raw_ostream &OS) const {
+ OS << " [";
+ for (const auto &Elem : Data)
+ OS << " " << format("%.2f", Elem) << " ";
+ OS << "]\n";
+}
+
// ==----------------------------------------------------------------------===//
// Embedder and its subclasses
//===----------------------------------------------------------------------===//
@@ -196,18 +209,12 @@ void SymbolicEmbedder::computeEmbeddings(const BasicBlock &BB) const {
for (const auto &I : BB.instructionsWithoutDebug()) {
Embedding InstVector(Dimension, 0);
- const auto OpcVec = lookupVocab(I.getOpcodeName());
- InstVector.scaleAndAdd(OpcVec, OpcWeight);
-
// FIXME: Currently lookups are string based. Use numeric Keys
// for efficiency.
- const auto Type = I.getType();
- const auto TypeVec = getTypeEmbedding(Type);
- InstVector.scaleAndAdd(TypeVec, TypeWeight);
-
+ InstVector += lookupVocab(I.getOpcodeName());
+ InstVector += getTypeEmbedding(I.getType());
for (const auto &Op : I.operands()) {
- const auto OperandVec = getOperandEmbedding(Op.get());
- InstVector.scaleAndAdd(OperandVec, ArgWeight);
+ InstVector += getOperandEmbedding(Op.get());
}
InstVecMap[&I] = InstVector;
BBVector += InstVector;
@@ -251,6 +258,47 @@ bool IR2VecVocabResult::invalidate(
return !(PAC.preservedWhenStateless());
}
+Error IR2VecVocabAnalysis::parseVocabSection(const char *Key,
+ const json::Value ParsedVocabValue,
+ ir2vec::Vocab &TargetVocab,
+ unsigned &Dim) {
+ assert(Key && "Key cannot be null");
+
+ json::Path::Root Path("");
+ const json::Object *RootObj = ParsedVocabValue.getAsObject();
+ if (!RootObj) {
+ return createStringError(errc::invalid_argument,
+ "JSON root is not an object");
+ }
+
+ const json::Value *SectionValue = RootObj->get(Key);
+ if (!SectionValue)
+ return createStringError(errc::invalid_argument,
+ "Missing '" + std::string(Key) +
+ "' section in vocabulary file");
+ if (!json::fromJSON(*SectionValue, TargetVocab, Path))
+ return createStringError(errc::illegal_byte_sequence,
+ "Unable to parse '" + std::string(Key) +
+ "' section from vocabulary");
+
+ Dim = TargetVocab.begin()->second.size();
+ if (Dim == 0)
+ return createStringError(errc::illegal_byte_sequence,
+ "Dimension of '" + std::string(Key) +
+ "' section of the vocabulary is zero");
+
+ if (!std::all_of(TargetVocab.begin(), TargetVocab.end(),
+ [Dim](const std::pair<StringRef, Embedding> &Entry) {
+ return Entry.second.size() == Dim;
+ }))
+ return createStringError(
+ errc::illegal_byte_sequence,
+ "All vectors in the '" + std::string(Key) +
+ "' section of the vocabulary are not of the same dimension");
+
+ return Error::success();
+};
+
// FIXME: Make this optional. We can avoid file reads
// by auto-generating a default vocabulary during the build time.
Error IR2VecVocabAnalysis::readVocabulary() {
@@ -259,32 +307,40 @@ Error IR2VecVocabAnalysis::readVocabulary() {
return createFileError(VocabFile, BufOrError.getError());
auto Content = BufOrError.get()->getBuffer();
- json::Path::Root Path("");
+
Expected<json::Value> ParsedVocabValue = json::parse(Content);
if (!ParsedVocabValue)
return ParsedVocabValue.takeError();
- bool Res = json::fromJSON(*ParsedVocabValue, Vocabulary, Path);
- if (!Res)
- return createStringError(errc::illegal_byte_sequence,
- "Unable to parse the vocabulary");
+ ir2vec::Vocab OpcodeVocab, TypeVocab, ArgVocab;
+ unsigned OpcodeDim, TypeDim, ArgDim;
+ if (auto Err = parseVocabSection("Opcodes", *ParsedVocabValue, OpcodeVocab,
+ OpcodeDim))
+ return Err;
- if (Vocabulary.empty())
- return createStringError(errc::illegal_byte_sequence,
- "Vocabulary is empty");
+ if (auto Err =
+ parseVocabSection("Types", *ParsedVocabValue, TypeVocab, TypeDim))
+ return Err;
- unsigned Dim = Vocabulary.begin()->second.size();
- if (Dim == 0)
+ if (auto Err =
+ parseVocabSection("Arguments", *ParsedVocabValue, ArgVocab, ArgDim))
+ return Err;
+
+ if (!(OpcodeDim == TypeDim && TypeDim == ArgDim))
return createStringError(errc::illegal_byte_sequence,
- "Dimension of vocabulary is zero");
+ "Vocabulary sections have different dimensions");
- if (!std::all_of(Vocabulary.begin(), Vocabulary.end(),
- [Dim](const std::pair<StringRef, Embedding> &Entry) {
- return Entry.second.size() == Dim;
- }))
- return createStringError(
- errc::illegal_byte_sequence,
- "All vectors in the vocabulary are not of the same dimension");
+ auto scaleVocabSection = [](ir2vec::Vocab &Vocab, double Weight) {
+ for (auto &Entry : Vocab)
+ Entry.second *= Weight;
+ };
+ scaleVocabSection(OpcodeVocab, OpcWeight);
+ scaleVocabSection(TypeVocab, TypeWeight);
+ scaleVocabSection(ArgVocab, ArgWeight);
+
+ Vocabulary.insert(OpcodeVocab.begin(), OpcodeVocab.end());
+ Vocabulary.insert(TypeVocab.begin(), TypeVocab.end());
+ Vocabulary.insert(ArgVocab.begin(), ArgVocab.end());
return Error::success();
}
@@ -304,7 +360,7 @@ void IR2VecVocabAnalysis::emitError(Error Err, LLVMContext &Ctx) {
IR2VecVocabAnalysis::Result
IR2VecVocabAnalysis::run(Module &M, ModuleAnalysisManager &AM) {
auto Ctx = &M.getContext();
- // FIXME: Scale the vocabulary once. This would avoid scaling per use later.
+
// If vocabulary is already populated by the constructor, use it.
if (!Vocabulary.empty())
return IR2VecVocabResult(std::move(Vocabulary));
@@ -323,16 +379,9 @@ IR2VecVocabAnalysis::run(Module &M, ModuleAnalysisManager &AM) {
}
// ==----------------------------------------------------------------------===//
-// IR2VecPrinterPass
+// Printer Passes
//===----------------------------------------------------------------------===//
-void IR2VecPrinterPass::printVector(const Embedding &Vec) const {
- OS << " [";
- for (const auto &Elem : Vec)
- OS << " " << format("%.2f", Elem) << " ";
- OS << "]\n";
-}
-
PreservedAnalyses IR2VecPrinterPass::run(Module &M,
ModuleAnalysisManager &MAM) {
auto IR2VecVocabResult = MAM.getResult<IR2VecVocabAnalysis>(M);
@@ -353,7 +402,7 @@ PreservedAnalyses IR2VecPrinterPass::run(Module &M,
OS << "IR2Vec embeddings for function " << F.getName() << ":\n";
OS << "Function vector: ";
- printVector(Emb->getFunctionVector());
+ Emb->getFunctionVector().print(OS);
OS << "Basic block vectors:\n";
const auto &BBMap = Emb->getBBVecMap();
@@ -361,7 +410,7 @@ PreservedAnalyses IR2VecPrinterPass::run(Module &M,
auto It = BBMap.find(&BB);
if (It != BBMap.end()) {
OS << "Basic block: " << BB.getName() << ":\n";
- printVector(It->second);
+ It->second.print(OS);
}
}
@@ -373,10 +422,24 @@ PreservedAnalyses IR2VecPrinterPass::run(Module &M,
if (It != InstMap.end()) {
OS << "Instruction: ";
I.print(OS);
- printVector(It->second);
+ It->second.print(OS);
}
}
}
}
return PreservedAnalyses::all();
}
+
+PreservedAnalyses IR2VecVocabPrinterPass::run(Module &M,
+ ModuleAnalysisManager &MAM) {
+ auto IR2VecVocabResult = MAM.getResult<IR2VecVocabAnalysis>(M);
+ assert(IR2VecVocabResult.isValid() && "IR2Vec Vocabulary is invalid");
+
+ auto Vocab = IR2VecVocabResult.getVocabulary();
+ for (const auto &Entry : Vocab) {
+ OS << "Key: " << Entry.first << ": ";
+ Entry.second.print(OS);
+ }
+
+ return PreservedAnalyses::all();
+}
\ No newline at end of file
diff --git a/llvm/lib/Analysis/models/seedEmbeddingVocab75D.json b/llvm/lib/Analysis/models/seedEmbeddingVocab75D.json
index 65ca240deba13..a0d539afd5294 100644
--- a/llvm/lib/Analysis/models/seedEmbeddingVocab75D.json
+++ b/llvm/lib/Analysis/models/seedEmbeddingVocab75D.json
@@ -1,65 +1,72 @@
{
-"add":[0.09571152, 0.08334279, 0.07070262, 0.14084256, 0.04825167, -0.12893851, -0.12180215, -0.07194936, -0.05329844, -0.28276998, 0.00202254, 0.07488817, -0.16139749, 0.07618549, -0.06084673, 0.10536429, -0.03031596, 0.16402271, 0.22730531, -0.0015432, 0.01986631, -0.15808797, 0.03030382, -0.04201249, 0.01689876, -0.30041003, -0.06479812, 0.1282431, -0.09032831, 0.14007935, -0.23540203, 0.00552223, 0.15325953, 0.13147154, -0.15204638, 0.1616615, 0.05758473, -0.1385157, 0.16438901, -0.14482859, 0.07895052, 0.18407233, 0.15283448, -0.00081216, -0.17934133, -0.16779658, 0.09044863, -0.18453072, -0.00552684, 0.01191218, 0.18504514, 0.13140059, -0.0174882, -0.18127315, 0.02269725, -0.02048657, 0.10858727, 0.0074029, 0.09485064, -0.13431476, 0.07491371, -0.17498694, -0.32914585, -0.14159656, -0.03594542, 0.04091231, 0.00298631, -0.08933277, -0.07178984, 0.04144038, 0.12151367, -0.09504163, 0.13691336, 0.07825345, 0.13958281],
-"alloca":[0.12644756, -0.20912014, 0.07298578, 0.13963142, 0.03932726, -0.12778169, -0.13084207, -0.09090705, -0.01497083, 0.0646746, 0.14847252, 0.08546949, -0.11579383, 0.07812478, -0.16431178, 0.08578802, 0.06094746, -0.1239017, -0.02789196, -0.01074964, -0.1738938, -0.1629063, -0.07137732, -0.02845656, 0.1728318, 0.13779363, -0.06250989, 0.03479109, -0.08423121, 0.14009888, 0.09620709, -0.02287545, -0.04616966, -0.19591278, -0.19636007, 0.21825573, 0.07949521, -0.14614704, 0.17752613, -0.15099092, -0.04518029, 0.17774306, 0.18947133, -0.00197501, -0.12619424, -0.18458585, -0.09960615, 0.01162648, 0.21306573, 0.0468024, 0.201505, 0.09970672, 0.06599383, 0.17757463, -0.11899275, 0.10299123, 0.06038174, -0.07116731, -0.15743989, -0.1258558, 0.1535076, 0.01872367, -0.14336765, -0.19708565, -0.26796287, 0.02628843, -0.0009325, -0.08809827, -0.12970725, 0.17223337, -0.10651242, -0.09162157, 0.03264611, 0.0911452, 0.09506542],
-"and":[-0.12902537, 0.07848564, 0.07048505, 0.13578993, 0.03570583, -0.1242408, -0.12619697, -0.06914084, -0.0515872, 0.13225996, 0.01673339, 0.0763608, 0.18325369, 0.07690141, -0.09591792, 0.10353354, -0.02401061, 0.16493416, 0.21942355, -0.00400911, 0.03811587, -0.15406959, -0.07134877, -0.02946596, -0.03854588, 0.28656727, -0.06625008, 0.12780443, -0.04631379, 0.13919763, -0.15808977, -0.00312698, 0.14692454, 0.18495218, -0.14863448, 0.13449706, 0.06471325, -0.13883992, 0.17630532, -0.16833898, 0.07391811, 0.17909151, 0.18229873, -0.00381102, -0.17680968, -0.1645554, 0.10016466, -0.07963493, 0.00130218, 0.05646244, 0.18222143, 0.10511146, -0.0191175, -0.08713559, 0.25423968, -0.02557301, 0.04319789, 0.03259414, 0.07209402, -0.13169754, 0.07424775, -0.17216511, -0.32057068, -0.13833733, -0.0658454, 0.02420194, -0.04393166, -0.08238692, -0.07023077, 0.04014502, 0.20101993, -0.09093616, 0.13076238, 0.09114857, -0.06483845],
-"ashr":[0.16414718, -0.02078147, 0.07353837, 0.14210321, 0.08068018, -0.13586298, -0.15728961, -0.10365791, 0.00359197, 0.04608767, 0.35770142, 0.08003625, 0.00944533, 0.08471741, 0.05809571, 0.09202059, -0.18349345, 0.22169511, 0.24418135, 0.19192688, -0.1956347, -0.17401719, -0.07583863, -0.02781139, 0.0952112, -0.01444751, -0.27259794, 0.14392436, -0.13541143, 0.1410963, 0.04314299, -0.01641751, -0.05448177, -0.22287542, -0.2131822, 0.25112826, 0.06918604, -0.14414005, 0.060288, -0.09658266, -0.09122665, -0.02779044, 0.20349248, 0.0041391, 0.10610974, -0.20890503, -0.09881835, -0.07368057, 0.22467837, 0.00075097, 0.2147348, -0.02612463, -0.01696278, -0.0649786, 0.15041149, 0.02361087, 0.0211603, -0.03706622, 0.18296233, -0.14298625, 0.14614436, 0.02273145, 0.0209446, -0.21062987, 0.16911499, -0.03668665, 0.00197532, -0.09607943, -0.08398084, 0.02006913, 0.05739584, -0.07919859, 0.19634944, 0.11082727, -0.06584227],
-"bitcast":[0.1675692, -0.12870269, 0.04048132, -0.1670965, -0.1279611, 0.02615386, -0.16829294, -0.09034907, 0.10913523, -0.07819421, 0.23986322, -0.05966561, 0.08410738, 0.19072439, 0.06047394, 0.02999627, -0.16747619, -0.06076627, -0.02673951, -0.1619169, 0.06443421, -0.13788716, -0.05644303, 0.01361013, -0.06858975, -0.06005004, 0.10011288, -0.05508338, -0.10613093, -0.11281271, -0.00758647, -0.12425531, 0.05333719, -0.16881412, -0.20088236, 0.06015657, -0.16405901, 0.06226884, 0.09171099, -0.09500738, 0.07907875, 0.0776544, -0.18457057, -0.11278627, -0.12111131, -0.10180638, 0.18328871, 0.18770072, 0.20346186, 0.10305139, -0.18344335, 0.10693213, -0.10920919, -0.05994263, 0.20354497, -0.03093485, 0.14214055, 0.00580597, 0.10480052, -0.09955201, -0.134185, -0.02904563, 0.00175069, 0.17646657, 0.01348841, -0.02030338, -0.06537742, 0.10032237, 0.15315783, 0.0102667, 0.07717734, -0.01060431, 0.14727928, -0.16261302, -0.06770234],
-"br":[0.11499645, 0.0824301, 0.07218035, 0.13258332, -0.13419574, -0.12916718, -0.12626709, -0.06741403, -0.02857593, -0.2543893, 0.02193022, 0.0760875, -0.1562702, 0.07712954, 0.3149361, 0.10217083, -0.041038, 0.16601022, 0.01906607, -0.02043359, 0.05471838, -0.15233372, -0.06945753, -0.02313732, -0.07342829, 0.3331809, -0.05246306, -0.0269839, -0.05435036, 0.13908924, 0.32604694, 0.00170966, 0.14997493, 0.13026518, -0.14908995, 0.12238151, -0.06773318, -0.13566032, 0.16068587, -0.12842499, 0.04970508, 0.17827724, 0.09729939, -0.00447832, -0.1739753, -0.16429187, 0.09886666, -0.08058207, 0.00714044, 0.04585538, 0.13424252, 0.11376464, -0.01675582, 0.17901348, -0.00653374, -0.01570439, 0.13032894, 0.01734108, 0.16833901, -0.1173776, 0.07662185, -0.15942436, -0.21173944, -0.10505079, 0.0597497, 0.03491669, 0.00338842, -0.04969047, -0.07644061, 0.04528612, 0.254365, -0.09514527, 0.12015092, 0.08262096, -0.02352029],
-"call":[0.10815012, -0.12419116, 0.18759736, -0.1905027, 0.01619313, -0.13483052, -0.12278763, -0.07051246, -0.0083437, 0.25107145, -0.16601063, 0.08127163, -0.17432374, -0.18380919, 0.24335551, 0.07208319, -0.04401246, 0.11606008, -0.02733191, 0.02098145, 0.019888, -0.13705409, -0.07569158, -0.03072285, 0.16870692, -0.09787013, -0.09340432, 0.01931342, -0.03557841, 0.14359893, -0.1592094, -0.00055867, 0.159316, 0.1099042, -0.11837319, 0.08741318, 0.03364393, -0.12831019, 0.10450637, -0.12699029, -0.20213994, 0.18390144, 0.11092624, -0.00209971, -0.13063665, 0.19996215, 0.09006448, -0.07840014, 0.22549215, 0.02587176, 0.13374338, 0.11009877, -0.01874998, -0.21446206, 0.02377797, -0.01036531, 0.05427047, 0.01418843, 0.00771817, -0.12639529, -0.10334941, -0.12244401, 0.30014148, -0.09857437, 0.21212636, 0.03429029, -0.04947309, 0.1023307, -0.07743628, 0.03006962, -0.24868701, -0.02357339, 0.11574048, 0.06895301, -0.363474],
-"constant":[-0.2850312, -0.22839, 0.12669143, -0.0674703, -0.12639391, -0.00477266, 0.04786542, 0.06336267, 0.08660185, 0.12805316, -0.07146342, 0.21539183, -0.0624397, -0.02638953, -0.28688517, 0.28374302, 0.05338082, 0.05559688, -0.13133128, 0.12440272, -0.03583231, 0.29848817, 0.13930812, 0.15453401, 0.0538353, -0.06874479, 0.00262802, 0.27964258, 0.19028014, -0.16371843, -0.05762961, 0.20059372, -0.20804578, -0.06549844, 0.09732475, -0.01551855, 0.21226783, 0.05889762, -0.07560658, 0.11312829, -0.04594622, -0.27309528, -0.05293005, 0.18953343, 0.05463868, -0.31045213, -0.04364616, 0.11005993, 0.12489324, -0.05413342, -0.05814561, -0.26131225, -0.18863814, 0.31165487, -0.08796364, -0.19958755, -0.10849535, 0.14899114, -0.01385941, 0.29359573, -0.01349372, 0.0562498, 0.10977754, 0.08993197, 0.06231657, -0.13509868, -0.20968516, 0.03578237, 0.15356435, -0.17766887, -0.11509016, 0.06898113, -0.06665445, -0.14065051, 0.34711906],
-"extractelement":[-1.62098512e-01, -2.00751424e-02, 1.71776459e-01, 1.38723463e-01, -1.60769299e-01, 9.30800363e-02, -1.24304645e-01, 1.45934001e-01, -5.04420400e-02, -7.54220188e-02, -5.30924797e-02, 8.55294541e-02, 1.17488159e-02, -1.82809234e-01, 9.37484950e-02, 8.38199258e-02, 1.26266479e-01, 9.31843743e-02, -2.26742271e-02, 6.50950940e-03, 5.02749532e-03, -1.14133619e-01, 1.80842891e-01, -9.63811725e-02, 1.67923152e-01, -5.84629439e-02, 1.04362026e-01, 8.92093012e-05, 7.93750137e-02, 1.39107972e-01, -8.40696543e-02, -1.59506593e-02, 1.99361086e-01, 1.93857521e-01, -1.46850392e-01, -1.78695560e-01, 9.88712162e-03, -1.40836969e-01, 1.77736521e-01, -1.61047727e-01, 3.51648539e-01, 1.79638579e-01, 1.49385989e-01, -6.06128015e-03, -1.72102928e-01, -1.81016047e-02, 1.01466164e-01, -1.28312945e-01, -8.05212185e-02, 6.28385469e-02, 1.15654446e-01, 1.91848025e-01, 3.03963851e-02, 3.55794169e-02, 1.40873834e-01, -1.44319711e-02, 1.85423180e-01, 1.16111919e-01, -7.47816712e-02, -1.14719503e-01, 1.02934733e-01, -1.83810964e-01, -2.64400076e-02, -8.99282843e-02, -1.90383971e-01, 7.31386840e-02, -4.36249487e-02, -4.71482053e-02, 1.07486300e-01, 1.09736443e-01, 4.15226035e-02, -1.42309800e-01, 8.96709636e-02, 6.64985999e-02, -6.13647103e-02],
-"extractvalue":[0.08820406, -0.0182279, 0.01493346, -0.17219813, 0.02927338, -0.17379265, -0.05663937, -0.06805898, -0.21235467, -0.01969833, -0.15152055, -0.18049374, 0.01062911, 0.07935719, -0.01993761, 0.12405304, -0.03198355, 0.13872959, 0.18017697, -0.0100...
[truncated]
|
svkeerthy
force-pushed
the
users/svkeerthy/06-12-_ir2vec_scale_vocab
branch
from
8e397f7 to
ac378a9
Compare
svkeerthy
commented
Jun 12, 2025
There was a problem hiding this comment.
@albertcohen - Please have a look. I am not able to add you as reviewer.
svkeerthy
force-pushed
the
users/svkeerthy/06-12-_ir2vec_scale_vocab
branch
from
ac378a9 to
2657262
Compare
mtrofin
reviewed
Jun 13, 2025
mtrofin
reviewed
Jun 13, 2025
mtrofin
reviewed
Jun 13, 2025
| "Unable to parse the vocabulary"); | ||
| ir2vec::Vocab OpcodeVocab, TypeVocab, ArgVocab; | ||
| unsigned OpcodeDim, TypeDim, ArgDim; | ||
| if (auto Err = parseVocabSection("Opcodes", *ParsedVocabValue, OpcodeVocab, |
There was a problem hiding this comment.
This changes the format, best to also update the doc.
Also, this means the sections must all be present (in any order), even if empty, correct? SGTM, just something worth spelling out.
There was a problem hiding this comment.
Correct. Will put it in the doc.
mtrofin
reviewed
Jun 13, 2025
llvm/lib/Passes/PassRegistry.def
Outdated
| @@ -104,7 +106,10 @@ MODULE_PASS("lower-ifunc", LowerIFuncPass()) | |||
| MODULE_PASS("simplify-type-tests", SimplifyTypeTestsPass()) | |||
| MODULE_PASS("lowertypetests", LowerTypeTestsPass()) | |||
| MODULE_PASS("fatlto-cleanup", FatLtoCleanup()) | |||
| MODULE_PASS("pgo-force-function-attrs", PGOForceFunctionAttrsPass(PGOOpt ? PGOOpt->ColdOptType : PGOOptions::ColdFuncOpt::Default)) | |||
| MODULE_PASS("pgo-force-function-attrs", | |||
| PGOForceFunctionAttrsPass(PGOOpt | |||
There was a problem hiding this comment.
can you make the unrelated stylistic changes to this file in a separate patch?
There was a problem hiding this comment.
Yeah, will do. Missed the unrelated formatting changes.
svkeerthy
force-pushed
the
users/svkeerthy/06-10-_mlininer_ir2vec_integrating_ir2vec_with_mlinliner
branch
from
b7ec652 to
d151083
Compare
svkeerthy
force-pushed
the
users/svkeerthy/06-12-_ir2vec_scale_vocab
branch
from
2657262 to
730ab91
Compare
svkeerthy
force-pushed
the
users/svkeerthy/06-10-_mlininer_ir2vec_integrating_ir2vec_with_mlinliner
branch
from
d151083 to
a2bec77
Compare
svkeerthy
force-pushed
the
users/svkeerthy/06-12-_ir2vec_scale_vocab
branch
from
730ab91 to
d31d756
Compare
svkeerthy
force-pushed
the
users/svkeerthy/06-10-_mlininer_ir2vec_integrating_ir2vec_with_mlinliner
branch
from
a2bec77 to
9124e83
Compare
svkeerthy
force-pushed
the
users/svkeerthy/06-12-_ir2vec_scale_vocab
branch
from
d31d756 to
32d16aa
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Changes to scale opcodes, types and args once in
IR2VecVocabAnalysisso that we can avoid scaling each time while computing embeddings. This PR refactors the vocabulary to explicitly define 3 sections---Opcodes, Types, and Arguments---used for computing Embeddings.(Tracking issue - #141817 ; partly fixes - #141832)