source: icGREP/icgrep-devel/icgrep/grep_engine.cpp @ 5659

Last change on this file since 5659 was 5646, checked in by nmedfort, 2 years ago

Minor clean up. Bug fix for object cache when the same cached kernel is used twice in a single run. Improvement to RE Minimizer.

File size: 29.4 KB
Line 
1/*
2 *  Copyright (c) 2017 International Characters.
3 *  This software is licensed to the public under the Open Software License 3.0.
4 *  icgrep is a trademark of International Characters.
5 */
6
7#include "grep_engine.h"
8#include "grep_interface.h"
9#include <llvm/IR/Module.h>
10#include <boost/filesystem.hpp>
11#include <UCD/UnicodeNameData.h>
12#include <UCD/resolve_properties.h>
13#include <kernels/charclasses.h>
14#include <kernels/cc_kernel.h>
15#include <kernels/grep_kernel.h>
16#include <kernels/linebreak_kernel.h>
17#include <kernels/streams_merge.h>
18#include <kernels/source_kernel.h>
19#include <kernels/s2p_kernel.h>
20#include <kernels/scanmatchgen.h>
21#include <kernels/streamset.h>
22#include <kernels/until_n.h>
23#include <kernels/kernel_builder.h>
24#include <pablo/pablo_kernel.h>
25#include <re/re_cc.h>
26#include <re/re_toolchain.h>
27#include <toolchain/toolchain.h>
28#include <re/re_name_resolve.h>   
29#include <re/re_collect_unicodesets.h>
30#include <re/re_multiplex.h>
31#include <toolchain/cpudriver.h>
32#include <toolchain/NVPTXDriver.h>
33#include <iostream>
34#include <sstream>
35#include <cc/multiplex_CCs.h>
36#include <llvm/Support/raw_ostream.h>
37#include <util/aligned_allocator.h>
38#include <sys/stat.h>
39#include <fcntl.h>
40#include <errno.h>
41#include <mutex>
42#ifdef CUDA_ENABLED
43#include <preprocess.cpp>
44#include <IR_Gen/CudaDriver.h>
45#endif
46
47using namespace parabix;
48using namespace llvm;
49
50namespace grep {
51
52static std::stringstream * resultStrs = nullptr;
53static std::vector<std::string> inputFiles;
54static std::vector<std::string> linePrefix;
55static bool grepMatchFound;
56
57size_t * startPoints = nullptr;
58size_t * accumBytes = nullptr;
59
60
61std::mutex count_mutex;
62size_t fileCount;
63
64// DoGrep thread function.
65void *DoGrepThreadFunction(void *args)
66{
67    size_t fileIdx;
68    grep::GrepEngine * grepEngine = (grep::GrepEngine *)args;
69
70    count_mutex.lock();
71    fileIdx = fileCount;
72    fileCount++;
73    count_mutex.unlock();
74
75    while (fileIdx < inputFiles.size()) {
76        size_t grepResult = grepEngine->doGrep(inputFiles[fileIdx], fileIdx);
77       
78        count_mutex.lock();
79        if (grepResult > 0) grepMatchFound = true;
80        fileIdx = fileCount;
81        fileCount++;
82        count_mutex.unlock();
83        if (QuietMode && grepMatchFound) pthread_exit(nullptr);
84    }
85
86    pthread_exit(nullptr);
87}
88
89bool matchesNeedToBeMovedToEOL() {
90    if ((Mode == QuietMode) | (Mode == FilesWithMatch) | (Mode == FilesWithoutMatch)) {
91        return false;
92    }
93    else if (LineRegexpFlag) {
94        return false;
95    }
96    // TODO: return false for other cases based on regexp analysis, e.g., regexp ends with $.
97    return true;
98}
99   
100void GrepEngine::doGrep(const std::string & fileName, std::string & PTXFilename) const{
101#ifdef CUDA_ENABLED
102    const bool CountOnly = true;
103    boost::filesystem::path file(fileName);
104    if (exists(file)) {
105        if (is_directory(file)) {
106            return;
107        }
108    } else {
109        if (!NoMessagesFlag) {
110            std::cerr << "Error: cannot open " << fileName << " for processing. Skipped.\n";
111            return;
112        }
113    }
114
115    const auto fileSize = file_size(file);
116   
117    if (fileSize > 0) {
118        try {
119            boost::iostreams::mapped_file_source source(fileName, fileSize, 0);
120            char * fileBuffer = const_cast<char *>(source.data());
121           
122            codegen::BlockSize = 128;
123            std::vector<size_t> LFPositions = preprocess(fileBuffer, fileSize);
124           
125            const unsigned numOfGroups = codegen::GroupNum;
126            if (posix_memalign((void**)&startPoints, 8, (numOfGroups+1)*sizeof(size_t)) ||
127                posix_memalign((void**)&accumBytes, 8, (numOfGroups+1)*sizeof(size_t))) {
128                std::cerr << "Cannot allocate memory for startPoints or accumBytes.\n";
129                exit(-1);
130            }
131            if(PTXFilename=="")
132                PTXFilename = mGrepDriver->getBuilder()->getModule()->getModuleIdentifier() + ".ptx";
133            RunPTX(PTXFilename, fileBuffer, fileSize, CountOnly, LFPositions, startPoints, accumBytes);
134            source.close();
135        } catch (std::exception & e) {
136            if (!NoMessagesFlag) {
137                std::cerr << "Boost mmap error: " + fileName + ": " + e.what() + " Skipped.\n";
138                return;
139            }
140        }
141    } else {
142        std::cout << 0 << std::endl;
143    }
144#endif
145}
146
147uint64_t GrepEngine::doGrep(const std::string & fileName, const uint32_t fileIdx) const {
148    struct stat sb;
149    const int32_t fd = open(fileName.c_str(), O_RDONLY);
150    if (LLVM_UNLIKELY(fd == -1)) {
151        if (!NoMessagesFlag  && !(Mode == QuietMode)) {
152            if (errno == EACCES) {
153                resultStrs[fileIdx] << "icgrep: " << fileName << ": Permission denied.\n";
154            }
155            else if (errno == ENOENT) {
156                resultStrs[fileIdx] << "icgrep: " << fileName << ": No such file.\n";
157            }
158            else {
159                resultStrs[fileIdx] << "icgrep: " << fileName << ": Failed.\n";
160            }
161        }
162        return 0;
163    }
164    if (stat(fileName.c_str(), &sb) == 0 && S_ISDIR(sb.st_mode)) {
165        if (!NoMessagesFlag  && !(Mode == QuietMode)) {
166            resultStrs[fileIdx] << "icgrep: " << fileName << ": Is a directory.\n";
167        }
168        close(fd);
169        return 0;
170    }
171    const auto result = doGrep(fd, fileIdx);
172    close(fd);
173    return result;
174}
175
176uint64_t GrepEngine::doGrep(const int32_t fileDescriptor, const uint32_t fileIdx) const {
177    assert (mGrepDriver);
178    typedef uint64_t (*GrepFunctionType)(int32_t fileDescriptor, const uint32_t fileIdx);
179    auto f = reinterpret_cast<GrepFunctionType>(mGrepDriver->getMain());
180   
181    uint64_t grepResult = f(fileDescriptor, fileIdx);
182    if (grepResult > 0) grepMatchFound = true;
183    else if ((Mode == NormalMode) && !resultStrs[fileIdx].str().empty()) grepMatchFound = true;
184   
185    if (Mode == CountOnly) {
186        resultStrs[fileIdx] << linePrefix[fileIdx] << grepResult << "\n";
187    }
188    else if (Mode == FilesWithMatch || Mode == FilesWithoutMatch ) {
189        size_t requiredCount = Mode == FilesWithMatch ? 1 : 0;
190        if (grepResult == requiredCount) {
191            resultStrs[fileIdx] << linePrefix[fileIdx];
192        }
193    }
194    else if (Mode == QuietMode) {
195        if (grepMatchFound) exit(MatchFoundExitCode);
196    }
197    return grepResult;
198}
199
200void initFileResult(std::vector<std::string> filenames){
201    grepMatchFound = false;
202    const int n = filenames.size();
203    linePrefix.resize(n);
204    if ((n > 1) && !NoFilenameFlag) {
205        WithFilenameFlag = true;
206    }
207    std::string fileSuffix = "";
208    bool setLinePrefix = WithFilenameFlag || (Mode == FilesWithMatch) || (Mode == FilesWithoutMatch);
209    if (setLinePrefix) {
210        if (NullFlag) {
211            fileSuffix = std::string("\0", 1);
212        }
213        else if ((Mode == NormalMode) && InitialTabFlag && !(LineNumberFlag || ByteOffsetFlag)) {
214            fileSuffix = "\t:";
215        }
216        else if ((Mode == NormalMode) || (Mode == CountOnly)) {
217            fileSuffix = ":";
218        }
219        else if ((Mode == FilesWithMatch) || (Mode == FilesWithoutMatch)) {
220            fileSuffix = "\n";
221        }
222    }
223    inputFiles = filenames;
224    resultStrs = new std::stringstream[n];
225    for (unsigned i = 0; i < inputFiles.size(); ++i) {
226        if (setLinePrefix) {
227            if (inputFiles[i] == "-") {
228                linePrefix[i] = LabelFlag + fileSuffix;
229            }
230            else {
231                linePrefix[i] = inputFiles[i] + fileSuffix;
232            }
233        }
234    }
235}
236
237template<typename CodeUnit>
238void wrapped_report_match(const size_t lineNum, size_t line_start, size_t line_end, const CodeUnit * const buffer, const size_t filesize, const size_t fileIdx) {
239
240//    errs().write_hex((size_t)buffer) << " : " << lineNum << " (" << line_start << ", " << line_end << ", " << filesize << ")\n";
241
242    assert (buffer);
243    assert (line_start <= line_end);
244    assert (line_end <= filesize);
245
246    if (WithFilenameFlag) {
247        resultStrs[fileIdx] << linePrefix[fileIdx];
248    }
249    if (LineNumberFlag) {
250        // Internally line numbers are counted from 0.  For display, adjust
251        // the line number so that lines are numbered from 1.
252        if (InitialTabFlag) {
253            resultStrs[fileIdx] << lineNum+1 << "\t:";
254        }
255        else {
256            resultStrs[fileIdx] << lineNum+1 << ":";
257        }
258    }
259
260    // If the line "starts" on the LF of a CRLF, it is actually the end of the last line.
261    if ((buffer[line_start] == 0xA) && (line_start != line_end)) {
262        ++line_start;
263    }
264
265    if (LLVM_UNLIKELY(line_end == filesize)) {
266        // The match position is at end-of-file.   We have a final unterminated line.
267        resultStrs[fileIdx].write((char *)&buffer[line_start], (line_end - line_start) * sizeof(CodeUnit));
268        if (NormalizeLineBreaksFlag) {
269            resultStrs[fileIdx] << '\n';  // terminate it
270        }
271    } else {
272        const auto end_byte = buffer[line_end];
273        if (grep::NormalizeLineBreaksFlag) {
274            if (LLVM_UNLIKELY(end_byte == 0x85)) {
275                // Line terminated with NEL, on the second byte.  Back up 1.
276                line_end -= 1;
277            } else if (LLVM_UNLIKELY(end_byte > 0xD)) {
278                // Line terminated with PS or LS, on the third byte.  Back up 2.
279                line_end -= 2;
280            }
281            resultStrs[fileIdx].write((char *)&buffer[line_start], (line_end - line_start) * sizeof(CodeUnit));
282            resultStrs[fileIdx] << '\n';
283        } else {
284            if (end_byte == 0x0D) {
285                // Check for line_end on first byte of CRLF; we don't want to access past the end of buffer.
286                if ((line_end + 1) < filesize) {
287                    if (buffer[line_end + 1] == 0x0A) {
288                        // Found CRLF; preserve both bytes.
289                        ++line_end;
290                    }
291                }
292            }
293            resultStrs[fileIdx].write((char *)&buffer[line_start], (line_end - line_start + 1) * sizeof(CodeUnit));
294        }
295    }
296}
297
298void PrintResults(){
299   
300    for (unsigned i = 0; i < inputFiles.size(); ++i){
301        std::cout << resultStrs[i].str();
302    }
303    exit(grepMatchFound ? MatchFoundExitCode : MatchNotFoundExitCode);
304}
305
306   
307std::pair<StreamSetBuffer *, StreamSetBuffer *> grepPipeline(Driver * grepDriver, std::vector<re::RE *> & REs, const GrepModeType grepMode, unsigned encodingBits, StreamSetBuffer * ByteStream) {
308    auto & idb = grepDriver->getBuilder();
309    const unsigned segmentSize = codegen::SegmentSize;
310    const unsigned bufferSegments = codegen::BufferSegments * codegen::ThreadNum;
311    size_t MatchLimit = ((grepMode == QuietMode) | (grepMode == FilesWithMatch) | (grepMode == FilesWithoutMatch)) ? 1 : MaxCountFlag;
312   
313
314    StreamSetBuffer * BasisBits = grepDriver->addBuffer(make_unique<CircularBuffer>(idb, idb->getStreamSetTy(encodingBits, 1), segmentSize * bufferSegments));
315    kernel::Kernel * s2pk = grepDriver->addKernelInstance(make_unique<kernel::S2PKernel>(idb));
316    grepDriver->makeKernelCall(s2pk, {ByteStream}, {BasisBits});
317   
318    StreamSetBuffer * LineBreakStream = grepDriver->addBuffer(make_unique<CircularBuffer>(idb, idb->getStreamSetTy(1, 1), segmentSize * bufferSegments));
319    kernel::Kernel * linebreakK = grepDriver->addKernelInstance(make_unique<kernel::LineBreakKernelBuilder>(idb, encodingBits));
320    grepDriver->makeKernelCall(linebreakK, {BasisBits}, {LineBreakStream});
321   
322    kernel::Kernel * requiredStreamsK = grepDriver->addKernelInstance(make_unique<kernel::RequiredStreams_UTF8>(idb));
323    StreamSetBuffer * RequiredStreams = grepDriver->addBuffer(make_unique<CircularBuffer>(idb, idb->getStreamSetTy(4, 1), segmentSize * bufferSegments));
324    grepDriver->makeKernelCall(requiredStreamsK, {BasisBits}, {RequiredStreams});
325   
326    const auto n = REs.size();
327   
328    std::vector<std::vector<UCD::UnicodeSet>> charclasses(n);
329
330    for (unsigned i = 0; i < n; i++) {
331        REs[i] = re::resolveNames(REs[i]);
332        std::vector<UCD::UnicodeSet> UnicodeSets = re::collect_UnicodeSets(REs[i]);
333        std::vector<std::vector<unsigned>> exclusiveSetIDs;
334        doMultiplexCCs(UnicodeSets, exclusiveSetIDs, charclasses[i]);
335        REs[i] = multiplex(REs[i], UnicodeSets, exclusiveSetIDs);
336    } 
337
338    std::vector<StreamSetBuffer *> MatchResultsBufs(n);
339
340    for(unsigned i = 0; i < n; ++i){
341        const auto numOfCharacterClasses = charclasses[i].size();
342        StreamSetBuffer * CharClasses = grepDriver->addBuffer(make_unique<CircularBuffer>(idb, idb->getStreamSetTy(numOfCharacterClasses), segmentSize * bufferSegments));
343        kernel::Kernel * ccK = grepDriver->addKernelInstance(make_unique<kernel::CharClassesKernel>(idb, std::move(charclasses[i])));
344        grepDriver->makeKernelCall(ccK, {BasisBits}, {CharClasses});
345        StreamSetBuffer * MatchResults = grepDriver->addBuffer(make_unique<CircularBuffer>(idb, idb->getStreamSetTy(1, 1), segmentSize * bufferSegments));
346        kernel::Kernel * icgrepK = grepDriver->addKernelInstance(make_unique<kernel::ICGrepKernel>(idb, REs[i], numOfCharacterClasses));
347        grepDriver->makeKernelCall(icgrepK, {CharClasses, LineBreakStream, RequiredStreams}, {MatchResults});
348        MatchResultsBufs[i] = MatchResults;
349    }
350    StreamSetBuffer * MergedResults = MatchResultsBufs[0];
351    if (REs.size() > 1) {
352        MergedResults = grepDriver->addBuffer(make_unique<CircularBuffer>(idb, idb->getStreamSetTy(1, 1), segmentSize * bufferSegments));
353        kernel::Kernel * streamsMergeK = grepDriver->addKernelInstance(make_unique<kernel::StreamsMerge>(idb, 1, REs.size()));
354        grepDriver->makeKernelCall(streamsMergeK, MatchResultsBufs, {MergedResults});
355    }
356    StreamSetBuffer * Matches = MergedResults;
357   
358    if (matchesNeedToBeMovedToEOL()) {
359        StreamSetBuffer * OriginalMatches = Matches;
360        kernel::Kernel * matchedLinesK = grepDriver->addKernelInstance(make_unique<kernel::MatchedLinesKernel>(idb));
361        Matches = grepDriver->addBuffer(make_unique<CircularBuffer>(idb, idb->getStreamSetTy(1, 1), segmentSize * bufferSegments));
362        grepDriver->makeKernelCall(matchedLinesK, {OriginalMatches, LineBreakStream}, {Matches});
363    }
364   
365    if (InvertMatchFlag) {
366        kernel::Kernel * invertK = grepDriver->addKernelInstance(make_unique<kernel::InvertMatchesKernel>(idb));
367        StreamSetBuffer * OriginalMatches = Matches;
368        Matches = grepDriver->addBuffer(make_unique<CircularBuffer>(idb, idb->getStreamSetTy(1, 1), segmentSize * bufferSegments));
369        grepDriver->makeKernelCall(invertK, {OriginalMatches, LineBreakStream}, {Matches});
370    }
371    if (MatchLimit > 0) {
372        kernel::Kernel * untilK = grepDriver->addKernelInstance(make_unique<kernel::UntilNkernel>(idb));
373        untilK->setInitialArguments({idb->getSize(MatchLimit)});
374        StreamSetBuffer * AllMatches = Matches;
375        Matches = grepDriver->addBuffer(make_unique<CircularBuffer>(idb, idb->getStreamSetTy(1, 1), segmentSize * bufferSegments));
376        grepDriver->makeKernelCall(untilK, {AllMatches}, {Matches});
377    }
378    return std::pair<StreamSetBuffer *, StreamSetBuffer *>(LineBreakStream, Matches);
379}
380
381
382   
383void GrepEngine::grepCodeGen_nvptx(std::vector<re::RE *> REs, const GrepModeType grepMode, const bool UTF_16) {
384
385    assert (mGrepDriver == nullptr);
386
387    mGrepDriver = new NVPTXDriver("engine");
388    auto & idb = mGrepDriver->getBuilder();
389    Module * M = idb->getModule();
390
391    const unsigned segmentSize = codegen::SegmentSize;
392    const unsigned encodingBits = UTF_16 ? 16 : 8;
393
394    Type * const int64Ty = idb->getInt64Ty();
395    Type * const int32Ty = idb->getInt32Ty();
396    Type * const size_ty = idb->getSizeTy();
397    Type * const sizeTyPtr = PointerType::get(size_ty, 1);
398    Type * const int64tyPtr = PointerType::get(int64Ty, 1);
399    Type * const voidTy = idb->getVoidTy();
400   
401    Function * mainFunc = cast<Function>(M->getOrInsertFunction("Main", voidTy, int64tyPtr, sizeTyPtr, sizeTyPtr, int64tyPtr, nullptr));
402    mainFunc->setCallingConv(CallingConv::C);
403    idb->SetInsertPoint(BasicBlock::Create(M->getContext(), "entry", mainFunc, 0));
404    auto args = mainFunc->arg_begin();
405
406    Value * const inputPtr = &*(args++);
407    inputPtr->setName("inputPtr");
408    Value * const startPointsPtr = &*(args++);
409    startPointsPtr->setName("startPointsPtr");
410    Value * const bufferSizesPtr = &*(args++);
411    bufferSizesPtr->setName("bufferSizesPtr");
412    Value * const outputPtr = &*(args++);
413    outputPtr->setName("outputPtr");
414
415    Function * tidFunc = M->getFunction("llvm.nvvm.read.ptx.sreg.tid.x");
416    Value * tid = idb->CreateCall(tidFunc);
417    Function * bidFunc = cast<Function>(M->getOrInsertFunction("llvm.nvvm.read.ptx.sreg.ctaid.x", int32Ty, nullptr));
418    Value * bid = idb->CreateCall(bidFunc);
419
420    Value * startPoint = idb->CreateLoad(idb->CreateGEP(startPointsPtr, bid));
421    Value * startBlock = idb->CreateUDiv(startPoint, ConstantInt::get(int64Ty, idb->getBitBlockWidth()));
422    Type * const inputStreamType = PointerType::get(ArrayType::get(ArrayType::get(idb->getBitBlockType(), 8), 1), 1);   
423    Value * inputStreamPtr = idb->CreateGEP(idb->CreateBitCast(inputPtr, inputStreamType), startBlock);
424    Value * inputStream = idb->CreateGEP(inputStreamPtr, tid);
425    Value * bufferSize = idb->CreateLoad(idb->CreateGEP(bufferSizesPtr, bid));
426
427    StreamSetBuffer * ByteStream = mGrepDriver->addBuffer(make_unique<SourceBuffer>(idb, idb->getStreamSetTy(1, 8), 1));
428    kernel::Kernel * sourceK = mGrepDriver->addKernelInstance(make_unique<kernel::MemorySourceKernel>(idb, inputStreamType, segmentSize));
429    sourceK->setInitialArguments({inputStream, bufferSize});
430    mGrepDriver->makeKernelCall(sourceK, {}, {ByteStream});
431
432    StreamSetBuffer * Matches = std::get<1>(grepPipeline(mGrepDriver, REs, grepMode, encodingBits, ByteStream));
433   
434    kernel::Kernel * matchCountK = mGrepDriver->addKernelInstance(make_unique<kernel::PopcountKernel>(idb));
435    mGrepDriver->makeKernelCall(matchCountK, {Matches}, {});
436    mGrepDriver->generatePipelineIR();
437    idb->setKernel(matchCountK);
438    Value * matchedLineCount = idb->getAccumulator("countResult");
439    matchedLineCount = idb->CreateZExt(matchedLineCount, int64Ty);
440   
441    Value * strideBlocks = ConstantInt::get(int32Ty, idb->getStride() / idb->getBitBlockWidth());
442    Value * outputThreadPtr = idb->CreateGEP(outputPtr, idb->CreateAdd(idb->CreateMul(bid, strideBlocks), tid));
443    idb->CreateStore(matchedLineCount, outputThreadPtr);
444    mGrepDriver->deallocateBuffers();
445    idb->CreateRetVoid();
446
447    mGrepDriver->finalizeObject();
448}
449
450void GrepEngine::grepCodeGen(std::vector<re::RE *> REs, const GrepModeType grepMode, const bool UTF_16, GrepSource grepSource) {
451
452    assert (mGrepDriver == nullptr);
453    mGrepDriver = new ParabixDriver("engine");
454    auto & idb = mGrepDriver->getBuilder();
455    Module * M = idb->getModule();
456
457    const unsigned segmentSize = codegen::SegmentSize;
458    const unsigned encodingBits = UTF_16 ? 16 : 8;
459
460    Type * const int64Ty = idb->getInt64Ty();
461    Type * const int32Ty = idb->getInt32Ty();
462
463    kernel::Kernel * sourceK = nullptr;
464   
465    Function * mainFunc = cast<Function>(M->getOrInsertFunction("Main", int64Ty, idb->getInt32Ty(), int32Ty, nullptr));
466    mainFunc->setCallingConv(CallingConv::C);
467    idb->SetInsertPoint(BasicBlock::Create(M->getContext(), "entry", mainFunc, 0));
468    auto args = mainFunc->arg_begin();
469
470    Value * const fileDescriptor = &*(args++);
471    fileDescriptor->setName("fileDescriptor");
472    Value * fileIdx = &*(args++);
473    fileIdx->setName("fileIdx");
474
475    StreamSetBuffer * ByteStream = mGrepDriver->addBuffer(make_unique<SourceBuffer>(idb, idb->getStreamSetTy(1, 8)));
476
477    if (grepSource == GrepSource::File) {
478        sourceK = mGrepDriver->addKernelInstance(make_unique<kernel::MMapSourceKernel>(idb, segmentSize));
479    } else {
480        sourceK = mGrepDriver->addKernelInstance(make_unique<kernel::ReadSourceKernel>(idb, segmentSize));
481    }
482    sourceK->setInitialArguments({fileDescriptor});
483
484    mGrepDriver->makeKernelCall(sourceK, {}, {ByteStream});
485   
486    StreamSetBuffer * LineBreakStream;
487    StreamSetBuffer * Matches;
488    std::tie(LineBreakStream, Matches) = grepPipeline(mGrepDriver, REs, grepMode, encodingBits, ByteStream);
489   
490    if (grepMode == NormalMode) {
491        kernel::Kernel * scanMatchK = mGrepDriver->addKernelInstance(make_unique<kernel::ScanMatchKernel>(idb, GrepType::Normal, encodingBits));
492        scanMatchK->setInitialArguments({fileIdx});
493        mGrepDriver->makeKernelCall(scanMatchK, {Matches, LineBreakStream, ByteStream}, {});
494        if (UTF_16) {
495            mGrepDriver->LinkFunction(*scanMatchK, "matcher", &wrapped_report_match<uint16_t>);
496        } else {
497            mGrepDriver->LinkFunction(*scanMatchK, "matcher", &wrapped_report_match<uint8_t>);
498        }
499        mGrepDriver->generatePipelineIR();
500        mGrepDriver->deallocateBuffers();
501
502        idb->CreateRet(idb->getInt64(0));
503    } else {
504        kernel::Kernel * matchCountK = mGrepDriver->addKernelInstance(make_unique<kernel::PopcountKernel>(idb));
505        mGrepDriver->makeKernelCall(matchCountK, {Matches}, {});
506        mGrepDriver->generatePipelineIR();
507        idb->setKernel(matchCountK);
508        Value * matchedLineCount = idb->getAccumulator("countResult");
509        matchedLineCount = idb->CreateZExt(matchedLineCount, int64Ty);
510        mGrepDriver->deallocateBuffers();
511        idb->CreateRet(matchedLineCount);
512    }
513    mGrepDriver->finalizeObject();
514}
515
516GrepEngine::GrepEngine()
517: mGrepDriver(nullptr) {
518
519}
520
521GrepEngine::~GrepEngine() {
522    delete mGrepDriver;
523}
524
525
526   
527static re::CC * parsedCodePointSet = nullptr;
528
529void insert_codepoints(const size_t lineNum, const size_t line_start, const size_t line_end, const char * const buffer) {
530    assert (buffer);
531    assert (line_start <= line_end);
532    re::codepoint_t c = 0;
533    size_t line_pos = line_start;
534    while (isxdigit(buffer[line_pos])) {
535        assert (line_pos < line_end);
536        if (isdigit(buffer[line_pos])) {
537            c = (c << 4) | (buffer[line_pos] - '0');
538        }
539        else {
540            c = (c << 4) | (tolower(buffer[line_pos]) - 'a' + 10);
541        }
542        line_pos++;
543    }
544    assert(((line_pos - line_start) >= 4) && ((line_pos - line_start) <= 6)); // UCD format 4 to 6 hex digits.
545    parsedCodePointSet->insert(c);
546}
547
548re::CC * grepCodepoints(re::RE * pattern, char * UnicodeDataBuffer, size_t bufferLength) {
549    parsedCodePointSet = re::makeCC();       
550    const unsigned segmentSize = 8;
551
552    ParabixDriver pxDriver("codepointEngine");
553    auto & idb = pxDriver.getBuilder();
554    Module * M = idb->getModule();
555   
556    Function * mainFunc = cast<Function>(M->getOrInsertFunction("Main", idb->getVoidTy(), idb->getInt8PtrTy(), idb->getSizeTy(), nullptr));
557    mainFunc->setCallingConv(CallingConv::C);
558    auto args = mainFunc->arg_begin();
559    Value * const buffer = &*(args++);
560    buffer->setName("buffer");
561    Value * length = &*(args++);
562    length->setName("length");
563   
564    idb->SetInsertPoint(BasicBlock::Create(M->getContext(), "entry", mainFunc, 0));
565   
566    StreamSetBuffer * ByteStream = pxDriver.addBuffer(make_unique<SourceBuffer>(idb, idb->getStreamSetTy(1, 8)));
567    kernel::Kernel * sourceK = pxDriver.addKernelInstance(make_unique<kernel::MemorySourceKernel>(idb, idb->getInt8PtrTy(), segmentSize));
568    sourceK->setInitialArguments({buffer, length});
569    pxDriver.makeKernelCall(sourceK, {}, {ByteStream});
570   
571    StreamSetBuffer * BasisBits = pxDriver.addBuffer(make_unique<CircularBuffer>(idb, idb->getStreamSetTy(8, 1), segmentSize));
572   
573    kernel::Kernel * s2pk = pxDriver.addKernelInstance(make_unique<kernel::S2PKernel>(idb));
574    pxDriver.makeKernelCall(s2pk, {ByteStream}, {BasisBits});
575   
576    kernel::Kernel * linebreakK = pxDriver.addKernelInstance(make_unique<kernel::LineBreakKernelBuilder>(idb, 8));
577    StreamSetBuffer * LineBreakStream = pxDriver.addBuffer(make_unique<CircularBuffer>(idb, idb->getStreamSetTy(1, 1), segmentSize));
578    pxDriver.makeKernelCall(linebreakK, {BasisBits}, {LineBreakStream});
579   
580    kernel::Kernel * requiredStreamsK = pxDriver.addKernelInstance(make_unique<kernel::RequiredStreams_UTF8>(idb));
581    StreamSetBuffer * RequiredStreams = pxDriver.addBuffer(make_unique<CircularBuffer>(idb, idb->getStreamSetTy(4, 1), segmentSize));
582    pxDriver.makeKernelCall(requiredStreamsK, {BasisBits}, {RequiredStreams});
583   
584    StreamSetBuffer * MatchResults = pxDriver.addBuffer(make_unique<CircularBuffer>(idb, idb->getStreamSetTy(1, 1), segmentSize));
585    kernel::Kernel * icgrepK = pxDriver.addKernelInstance(make_unique<kernel::ICGrepKernel>(idb, pattern));
586    pxDriver.makeKernelCall(icgrepK, {BasisBits, LineBreakStream, RequiredStreams}, {MatchResults});
587   
588    StreamSetBuffer * MatchedLines = pxDriver.addBuffer(make_unique<CircularBuffer>(idb, idb->getStreamSetTy(1, 1), segmentSize));
589    kernel::Kernel * matchedLinesK = pxDriver.addKernelInstance(make_unique<kernel::MatchedLinesKernel>(idb));
590    pxDriver.makeKernelCall(matchedLinesK, {MatchResults, LineBreakStream}, {MatchedLines});
591   
592    kernel::Kernel * scanMatchK = pxDriver.addKernelInstance(make_unique<kernel::ScanMatchKernel>(idb, GrepType::NameExpression, 8));
593    scanMatchK->setInitialArguments({idb->getInt32(0)});
594    pxDriver.makeKernelCall(scanMatchK, {MatchedLines, LineBreakStream, ByteStream}, {});
595    pxDriver.LinkFunction(*scanMatchK, "matcher", &insert_codepoints);
596    pxDriver.generatePipelineIR();
597    pxDriver.deallocateBuffers();
598    idb->CreateRetVoid();
599    pxDriver.finalizeObject();
600   
601    typedef void (*GrepFunctionType)(const char * buffer, const size_t length);
602    auto f = reinterpret_cast<GrepFunctionType>(pxDriver.getMain());
603    f(UnicodeDataBuffer, bufferLength);
604   
605    return parsedCodePointSet;   
606}
607
608   
609static std::vector<std::string> parsedPropertyValues;
610
611void insert_property_values(size_t lineNum, size_t line_start, size_t line_end, const char * buffer) {
612    assert (line_start <= line_end);
613    parsedPropertyValues.emplace_back(buffer + line_start, buffer + line_end);
614}
615
616
617const std::vector<std::string> & grepPropertyValues(const std::string& propertyName, re::RE * propertyValuePattern) {
618    ParabixDriver pxDriver("propertyValueEngine");
619    AlignedAllocator<char, 32> alloc;
620
621    parsedPropertyValues.clear();
622
623    const std::string & str = UCD::getPropertyValueGrepString(propertyName);
624
625    auto & idb = pxDriver.getBuilder();
626
627    const unsigned segmentSize = 8;
628    const auto n = str.length();
629    const auto w = idb->getBitBlockWidth() * segmentSize;
630    const auto m = w - (n % w);
631
632    char * aligned = alloc.allocate(n + m, 0);
633    std::memcpy(aligned, str.data(), n);
634    std::memset(aligned + n, 0, m);
635
636    Module * M = idb->getModule();
637   
638    Function * mainFunc = cast<Function>(M->getOrInsertFunction("Main", idb->getVoidTy(), idb->getInt8PtrTy(), idb->getSizeTy(), nullptr));
639    mainFunc->setCallingConv(CallingConv::C);
640    auto args = mainFunc->arg_begin();
641    Value * const buffer = &*(args++);
642    buffer->setName("buffer");
643    Value * length = &*(args++);
644    length->setName("length");
645   
646    idb->SetInsertPoint(BasicBlock::Create(M->getContext(), "entry", mainFunc, 0));
647   
648    StreamSetBuffer * ByteStream = pxDriver.addBuffer(make_unique<SourceBuffer>(idb, idb->getStreamSetTy(1, 8)));
649    kernel::Kernel * sourceK = pxDriver.addKernelInstance(make_unique<kernel::MemorySourceKernel>(idb, idb->getInt8PtrTy(), segmentSize));
650    sourceK->setInitialArguments({buffer, length});
651    pxDriver.makeKernelCall(sourceK, {}, {ByteStream});
652   
653    StreamSetBuffer * BasisBits = pxDriver.addBuffer(make_unique<CircularBuffer>(idb, idb->getStreamSetTy(8, 1), segmentSize));
654   
655    kernel::Kernel * s2pk = pxDriver.addKernelInstance(make_unique<kernel::S2PKernel>(idb));
656    pxDriver.makeKernelCall(s2pk, {ByteStream}, {BasisBits});
657   
658    kernel::Kernel * linebreakK = pxDriver.addKernelInstance(make_unique<kernel::LineBreakKernelBuilder>(idb, 8));
659    StreamSetBuffer * LineBreakStream = pxDriver.addBuffer(make_unique<CircularBuffer>(idb, idb->getStreamSetTy(1, 1), segmentSize));
660    pxDriver.makeKernelCall(linebreakK, {BasisBits}, {LineBreakStream});
661   
662    kernel::Kernel * requiredStreamsK = pxDriver.addKernelInstance(make_unique<kernel::RequiredStreams_UTF8>(idb));
663    StreamSetBuffer * RequiredStreams = pxDriver.addBuffer(make_unique<CircularBuffer>(idb, idb->getStreamSetTy(4, 1), segmentSize));
664    pxDriver.makeKernelCall(requiredStreamsK, {BasisBits}, {RequiredStreams});
665   
666    StreamSetBuffer * MatchResults = pxDriver.addBuffer(make_unique<CircularBuffer>(idb, idb->getStreamSetTy(1, 1), segmentSize));
667    kernel::Kernel * icgrepK = pxDriver.addKernelInstance(make_unique<kernel::ICGrepKernel>(idb, propertyValuePattern));
668    pxDriver.makeKernelCall(icgrepK, {BasisBits, LineBreakStream, RequiredStreams}, {MatchResults});
669
670    StreamSetBuffer * MatchedLines = pxDriver.addBuffer(make_unique<CircularBuffer>(idb, idb->getStreamSetTy(1, 1), segmentSize));
671    kernel::Kernel * matchedLinesK = pxDriver.addKernelInstance(make_unique<kernel::MatchedLinesKernel>(idb));
672    pxDriver.makeKernelCall(matchedLinesK, {MatchResults, LineBreakStream}, {MatchedLines});
673
674    kernel::Kernel * scanMatchK = pxDriver.addKernelInstance(make_unique<kernel::ScanMatchKernel>(idb, GrepType::PropertyValue, 8));
675    scanMatchK->setInitialArguments({idb->getInt32(0)});
676    pxDriver.makeKernelCall(scanMatchK, {MatchedLines, LineBreakStream, ByteStream}, {});
677    pxDriver.LinkFunction(*scanMatchK, "matcher", &insert_property_values);
678    pxDriver.generatePipelineIR();
679    pxDriver.deallocateBuffers();
680    idb->CreateRetVoid();
681    pxDriver.finalizeObject();
682
683    typedef void (*GrepFunctionType)(const char * buffer, const size_t length);
684    auto f = reinterpret_cast<GrepFunctionType>(pxDriver.getMain());
685    f(aligned, n);
686   
687    alloc.deallocate(aligned, 0);
688    return parsedPropertyValues;
689}
690
691   
692}
Note: See TracBrowser for help on using the repository browser.