source: icGREP/icgrep-devel/icgrep/u8u16.cpp @ 5208

Last change on this file since 5208 was 5202, checked in by nmedfort, 3 years ago

Initial work on adding types to PabloAST and mutable Var objects.

File size: 20.0 KB
Line 
1/*
2 *  Copyright (c) 2016 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 <string>
8#include <iostream>
9#include <fstream>
10#include <sstream>
11#include <stdlib.h>
12
13#include <llvm/IR/Function.h>
14#include <llvm/IR/Module.h>
15#include <llvm/ExecutionEngine/ExecutionEngine.h>
16#include <llvm/ExecutionEngine/MCJIT.h>
17#include <llvm/IRReader/IRReader.h>
18#include <llvm/IR/Verifier.h>
19#include <llvm/Support/Debug.h>
20
21#include <llvm/Support/CommandLine.h>
22
23#include <toolchain.h>
24#include <re/re_cc.h>
25#include <cc/cc_compiler.h>
26#include <pablo/pablo_toolchain.h>
27#include <pablo/pablo_kernel.h>
28#include <pablo/function.h>
29#include <IDISA/idisa_builder.h>
30#include <IDISA/idisa_target.h>
31#include <kernels/pipeline.h>
32#include <kernels/interface.h>
33#include <kernels/kernel.h>
34#include <kernels/s2p_kernel.h>
35#include <kernels/p2s_kernel.h>
36#include <kernels/deletion.h>
37#include <kernels/stdout_kernel.h>
38#include <llvm/IR/TypeBuilder.h>
39
40
41// mmap system
42#include <boost/filesystem.hpp>
43#include <boost/iostreams/device/mapped_file.hpp>
44#include <boost/interprocess/anonymous_shared_memory.hpp>
45#include <boost/interprocess/mapped_region.hpp>
46#include <fcntl.h>
47static cl::OptionCategory u8u16Options("u8u16 Options",
48                                            "Transcoding control options.");
49
50static cl::list<std::string> inputFiles(cl::Positional, cl::desc("<input file ...>"), cl::OneOrMore, cl::cat(u8u16Options));
51
52static cl::opt<bool> segmentPipelineParallel("enable-segment-pipeline-parallel", cl::desc("Enable multithreading with segment pipeline parallelism."), cl::cat(u8u16Options));
53static cl::opt<bool> mMapBuffering("mmap-buffering", cl::desc("Enable mmap buffering."), cl::cat(u8u16Options));
54static cl::opt<bool> memAlignBuffering("memalign-buffering", cl::desc("Enable posix_memalign buffering."), cl::cat(u8u16Options));
55
56//
57//
58//
59namespace pablo {
60
61PabloFunction * u8u16_pablo() {
62    //  input: 8 basis bit streams
63    //  output: 16 u8-indexed streams, + delmask stream + error stream
64    PabloFunction * function = PabloFunction::Create("u8u16"); // , 1, 18
65    cc::CC_Compiler ccc(*function);
66   
67    PabloBuilder & main = ccc.getBuilder();
68    const auto u8_bits = ccc.getBasisBits();
69
70    PabloAST * zeroes = main.createZeroes();
71
72    // Outputs
73    Var * u16_hi[8];
74    for (int i = 0; i < 8; ++i) {
75        u16_hi[i] = main.createVar("u16_hi" + std::to_string(i), zeroes);
76    }
77    Var * u16_lo[8];
78    for (int i = 0; i < 8; ++i) {
79        u16_lo[i] = main.createVar("u16_lo" + std::to_string(i), zeroes);
80    }
81    Var * delmask = main.createVar("delmask", zeroes);
82    Var * error_mask = main.createVar("error_mask", zeroes);
83
84
85
86    // The logic for processing non-ASCII bytes will be embedded within an if-hierarchy.
87    PabloAST * nonASCII = ccc.compileCC(re::makeCC(0x80, 0xFF));
88   
89    // Builder for the if statement handling all non-ASCII logic
90    PabloBuilder nAb = PabloBuilder::Create(main);
91    // Bits 3 through 7 of a 2-byte prefix are data bits, needed to
92    // produce the UTF-16 code unit data ...,
93    PabloAST * bit3a1 = nAb.createAdvance(u8_bits[3], 1);
94    PabloAST * bit4a1 = nAb.createAdvance(u8_bits[4], 1);
95    PabloAST * bit5a1 = nAb.createAdvance(u8_bits[5], 1);
96    PabloAST * bit6a1 = nAb.createAdvance(u8_bits[6], 1);
97    PabloAST * bit7a1 = nAb.createAdvance(u8_bits[7], 1);
98   
99    // Entry condition for 3 or 4 byte sequences: we have a prefix byte in the range 0xE0-0xFF.
100    PabloAST * pfx34 = ccc.compileCC(re::makeCC(0xE0, 0xFF), nAb);
101    // Builder for the if statement handling all logic for 3- and 4-byte sequences.
102    PabloBuilder p34b = PabloBuilder::Create(nAb);
103    // Bits 4 through 7 of a 3-byte prefix are data bits.  They must be moved
104    // to the final position of the 3-byte sequence.
105    PabloAST * bit2a1 = p34b.createAdvance(u8_bits[2], 1);
106    PabloAST * bit4a2 = p34b.createAdvance(bit4a1, 1);
107    PabloAST * bit5a2 = p34b.createAdvance(bit5a1, 1);
108    PabloAST * bit6a2 = p34b.createAdvance(bit6a1, 1);
109    PabloAST * bit7a2 = p34b.createAdvance(bit7a1, 1);
110
111
112    Var * const u8scope32 = nAb.createVar("u8scope32", zeroes);
113    Var * const u8scope33 = nAb.createVar("u8scope33", zeroes);
114    Var * const u8scope44 = nAb.createVar("u8scope44", zeroes);
115
116    //
117    // Logic for 4-byte UTF-8 sequences
118    //
119    // Entry condition  or 4 byte sequences: we have a prefix byte in the range 0xF0-0xFF.
120    PabloAST * pfx4 = ccc.compileCC(re::makeCC(0xF0, 0xFF), p34b);
121    // Builder for the if statement handling all logic for 4-byte sequences only.
122    PabloBuilder p4b = PabloBuilder::Create(p34b);
123    // Illegal 4-byte sequences
124    PabloAST * F0 = ccc.compileCC(re::makeCC(0xF0), p4b);
125    PabloAST * F4 = ccc.compileCC(re::makeCC(0xF4), p4b);
126    PabloAST * F0_err = p4b.createAnd(p4b.createAdvance(F0, 1), ccc.compileCC(re::makeCC(0x80, 0x8F), p4b));
127    PabloAST * F4_err = p4b.createAnd(p4b.createAdvance(F4, 1), ccc.compileCC(re::makeCC(0x90, 0xBF), p4b));
128    PabloAST * F5_FF = ccc.compileCC(re::makeCC(0xF5, 0xFF), p4b);
129
130    Var * FX_err = p34b.createVar("FX_err", zeroes);
131    p4b.createAssign(FX_err, p4b.createOr(F5_FF, p4b.createOr(F0_err, F4_err)));
132    //
133    // 4-byte prefixes have a scope that extends over the next 3 bytes.
134
135    Var * u8scope42 = p34b.createVar("u8scope42", zeroes);
136    Var * u8scope43 = p34b.createVar("u8scope43", zeroes);
137
138    p4b.createAssign(u8scope42, p4b.createAdvance(pfx4, 1));
139    p4b.createAssign(u8scope43, p4b.createAdvance(u8scope42, 1));
140    p4b.createAssign(u8scope44, p4b.createAdvance(u8scope43, 1));
141    //
142   
143    //  From the 4-byte sequence 11110abc 10defghi 10jklmno 10pqrstu,
144    //  we must calculate the value abcde - 1 to produce the bit values
145    //  for u16_hi6, hi7, lo0, lo1 at the scope43 position.
146    Var * s43_lo0 = nAb.createVar("scope43_lo0", zeroes);
147    Var * s43_lo1 = nAb.createVar("scope43_lo1", zeroes);
148    Var * s43_hi6 = nAb.createVar("scope43_hi6", zeroes);
149    Var * s43_hi7 = nAb.createVar("scope43_hi7", zeroes);
150
151    Var * s43_lo2 = main.createVar("scope43_lo2", zeroes);
152    Var * s43_lo3 = main.createVar("scope43_lo3", zeroes);
153    Var * s43_lo4 = main.createVar("scope43_lo4", zeroes);
154    Var * s43_lo5 = main.createVar("scope43_lo5", zeroes);
155    Var * s43_lo6 = main.createVar("scope43_lo6", zeroes);
156    Var * s43_lo7 = main.createVar("scope43_lo7", zeroes);
157
158    p4b.createAssign(s43_lo1, p4b.createAnd(u8scope43, p4b.createNot(bit3a1)));           // e - 1
159    p4b.createAssign(s43_lo0, p4b.createAnd(u8scope43, p4b.createXor(bit2a1, s43_lo1)));  // d - borrow
160    PabloAST * brw1 = p4b.createAnd(s43_lo1, p4b.createNot(bit2a1));
161    p4b.createAssign(s43_hi7, p4b.createAnd(u8scope43, p4b.createXor(bit7a2, brw1)));     // c - borrow
162    PabloAST * brw2 = p4b.createAnd(brw1, p4b.createNot(bit7a2));
163    p4b.createAssign(s43_hi6, p4b.createAnd(u8scope43, p4b.createXor(bit6a2, brw2)));     // b - borrow
164    //
165    p4b.createAssign(s43_lo2, p4b.createAnd(u8scope43, bit4a1));
166    p4b.createAssign(s43_lo3, p4b.createAnd(u8scope43, bit5a1));
167    p4b.createAssign(s43_lo4, p4b.createAnd(u8scope43, bit6a1));
168    p4b.createAssign(s43_lo5, p4b.createAnd(u8scope43, bit7a1));
169    p4b.createAssign(s43_lo6, p4b.createAnd(u8scope43, u8_bits[2]));
170    p4b.createAssign(s43_lo7, p4b.createAnd(u8scope43, u8_bits[3]));
171    //
172    //
173    p34b.createIf(pfx4, p4b);
174    //
175    // Combined logic for 3 and 4 byte sequences
176    //
177    PabloAST * pfx3 = ccc.compileCC(re::makeCC(0xE0, 0xEF), p34b);
178
179    p34b.createAssign(u8scope32, p34b.createAdvance(pfx3, 1));
180    p34b.createAssign(u8scope33, p34b.createAdvance(u8scope32, 1));
181
182    // Illegal 3-byte sequences
183    PabloAST * E0 = ccc.compileCC(re::makeCC(0xE0), p34b);
184    PabloAST * ED = ccc.compileCC(re::makeCC(0xED), p34b);
185    PabloAST * E0_err = p34b.createAnd(p34b.createAdvance(E0, 1), ccc.compileCC(re::makeCC(0x80, 0x9F), p34b));
186    PabloAST * ED_err = p34b.createAnd(p34b.createAdvance(ED, 1), ccc.compileCC(re::makeCC(0xA0, 0xBF), p34b));
187    Var * EX_FX_err = nAb.createVar("EX_FX_err", zeroes);
188
189    p34b.createAssign(EX_FX_err, p34b.createOr(p34b.createOr(E0_err, ED_err), FX_err));
190    // Two surrogate UTF-16 units are computed at the 3rd and 4th positions of 4-byte sequences.
191    PabloAST * surrogate = p34b.createOr(u8scope43, u8scope44);
192   
193    Var * p34del = nAb.createVar("p34del", zeroes);
194    p34b.createAssign(p34del, p34b.createOr(u8scope32, u8scope42));
195
196
197    // The high 5 bits of the UTF-16 code unit are only nonzero for 3 and 4-byte
198    // UTF-8 sequences.
199    p34b.createAssign(u16_hi[0], p34b.createOr(p34b.createAnd(u8scope33, bit4a2), surrogate));
200    p34b.createAssign(u16_hi[1], p34b.createOr(p34b.createAnd(u8scope33, bit5a2), surrogate));
201    p34b.createAssign(u16_hi[2], p34b.createAnd(u8scope33, bit6a2));
202    p34b.createAssign(u16_hi[3], p34b.createOr(p34b.createAnd(u8scope33, bit7a2), surrogate));
203    p34b.createAssign(u16_hi[4], p34b.createOr(p34b.createAnd(u8scope33, bit2a1), surrogate));
204   
205    //
206    nAb.createIf(pfx34, p34b);
207    //
208    // Combined logic for 2, 3 and 4 byte sequences
209    //
210
211    Var * u8lastscope = main.createVar("u8lastscope", zeroes);
212
213    PabloAST * pfx2 = ccc.compileCC(re::makeCC(0xC0, 0xDF), nAb);
214    PabloAST * u8scope22 = nAb.createAdvance(pfx2, 1);
215    nAb.createAssign(u8lastscope, nAb.createOr(u8scope22, nAb.createOr(u8scope33, u8scope44)));
216    PabloAST * u8anyscope = nAb.createOr(u8lastscope, p34del);
217
218    PabloAST * C0_C1_err = ccc.compileCC(re::makeCC(0xC0, 0xC1), nAb);
219    PabloAST * scope_suffix_mismatch = nAb.createXor(u8anyscope, ccc.compileCC(re::makeCC(0x80, 0xBF), nAb));
220    nAb.createAssign(error_mask, nAb.createOr(scope_suffix_mismatch, nAb.createOr(C0_C1_err, EX_FX_err)));
221    nAb.createAssign(delmask, nAb.createOr(p34del, ccc.compileCC(re::makeCC(0xC0, 0xFF), nAb)));
222   
223    // The low 3 bits of the high byte of the UTF-16 code unit as well as the high bit of the
224    // low byte are only nonzero for 2, 3 and 4 byte sequences.
225    nAb.createAssign(u16_hi[5], nAb.createOr(nAb.createAnd(u8lastscope, bit3a1), u8scope44));
226    nAb.createAssign(u16_hi[6], nAb.createOr(nAb.createAnd(u8lastscope, bit4a1), s43_hi6));
227    nAb.createAssign(u16_hi[7], nAb.createOr(nAb.createAnd(u8lastscope, bit5a1), s43_hi7));
228    nAb.createAssign(u16_lo[0], nAb.createOr(nAb.createAnd(u8lastscope, bit6a1), s43_lo0));
229
230    Var * p234_lo1 = main.createVar("p234_lo1", zeroes);
231
232    nAb.createAssign(p234_lo1, nAb.createOr(nAb.createAnd(u8lastscope, bit7a1), s43_lo1));
233
234    main.createIf(nonASCII, nAb);
235    //
236    //
237    PabloAST * ASCII = ccc.compileCC(re::makeCC(0x0, 0x7F));
238    PabloAST * last_byte = main.createOr(ASCII, u8lastscope);
239    main.createAssign(u16_lo[1], main.createOr(main.createAnd(ASCII, u8_bits[1]), p234_lo1));
240    main.createAssign(u16_lo[2], main.createOr(main.createAnd(last_byte, u8_bits[2]), s43_lo2));
241    main.createAssign(u16_lo[3], main.createOr(main.createAnd(last_byte, u8_bits[3]), s43_lo3));
242    main.createAssign(u16_lo[4], main.createOr(main.createAnd(last_byte, u8_bits[4]), s43_lo4));
243    main.createAssign(u16_lo[5], main.createOr(main.createAnd(last_byte, u8_bits[5]), s43_lo5));
244    main.createAssign(u16_lo[6], main.createOr(main.createAnd(last_byte, u8_bits[6]), s43_lo6));
245    main.createAssign(u16_lo[7], main.createOr(main.createAnd(last_byte, u8_bits[7]), s43_lo7));
246   
247    Var * output = function->addResult("output", getStreamTy(1, 18));
248    for (unsigned i = 0; i < 8; i++) {
249        main.createAssign(main.createExtract(output, i), u16_hi[i]);
250    }
251    for (unsigned i = 0; i < 8; i++) {
252        main.createAssign(main.createExtract(output, i + 8), u16_lo[i]);
253    }
254    main.createAssign(main.createExtract(output, 16), delmask);
255    main.createAssign(main.createExtract(output, 17), error_mask);
256
257    return function;
258}
259}
260
261
262
263using namespace kernel;
264using namespace parabix;
265
266Function * u8u16Pipeline(Module * mMod, IDISA::IDISA_Builder * iBuilder, pablo::PabloFunction * function) {
267    Type * mBitBlockType = iBuilder->getBitBlockType();
268
269    const unsigned segmentSize = codegen::SegmentSize;
270    const unsigned bufferSegments = codegen::BufferSegments;
271   
272    ExternalFileBuffer ByteStream(iBuilder, StreamSetType(iBuilder,1, 8));
273
274    CircularBuffer BasisBits(iBuilder, StreamSetType(iBuilder,8, 1), segmentSize * bufferSegments);
275
276    CircularBuffer U8u16Bits(iBuilder, StreamSetType(iBuilder, 18, 1), segmentSize * bufferSegments);
277
278    CircularBuffer U16Bits(iBuilder, StreamSetType(iBuilder,16, 1), segmentSize * bufferSegments);
279   
280    CircularBuffer DeletionCounts(iBuilder, StreamSetType(iBuilder, 1, 1), segmentSize * bufferSegments);
281
282    // Different choices for the output buffer depending on chosen option.
283    ExternalFileBuffer U16external(iBuilder, StreamSetType(iBuilder,1, 16));
284    LinearCopybackBuffer U16out(iBuilder, StreamSetType(iBuilder,1, 16), segmentSize * bufferSegments + 2);
285
286    s2pKernel  s2pk(iBuilder);
287    s2pk.generateKernel({&ByteStream}, {&BasisBits});
288
289    pablo_function_passes(function);
290    pablo::PabloKernel u8u16k(iBuilder, "u8u16", function);
291    u8u16k.generateKernel({&BasisBits}, {&U8u16Bits});
292   
293    DeletionKernel delK(iBuilder, iBuilder->getBitBlockWidth()/16, 16);
294    delK.generateKernel({&U8u16Bits}, {&U16Bits, &DeletionCounts});
295   
296    p2s_16Kernel_withCompressedOutput p2sk(iBuilder);
297   
298    stdOutKernel stdoutK(iBuilder, 16);
299   
300    if (mMapBuffering || memAlignBuffering) {
301        p2sk.generateKernel({&U16Bits, &DeletionCounts}, {&U16external});
302        stdoutK.generateKernel({&U16external}, {});
303    }
304    else {
305        p2sk.generateKernel({&U16Bits, &DeletionCounts}, {&U16out});
306        stdoutK.generateKernel({&U16out}, {});
307    }
308   
309    Type * const size_ty = iBuilder->getSizeTy();
310    Type * const voidTy = Type::getVoidTy(mMod->getContext());
311    Type * const inputType = PointerType::get(ArrayType::get(ArrayType::get(mBitBlockType, 8), 1), 0);
312    Type * const outputType = PointerType::get(ArrayType::get(ArrayType::get(mBitBlockType, 16), 1), 0);
313    Type * const int32ty = iBuilder->getInt32Ty();
314    Type * const int8PtrTy = iBuilder->getInt8PtrTy();
315    Type * const voidPtrTy = TypeBuilder<void *, false>::get(mMod->getContext());
316
317   
318    Function * const main = cast<Function>(mMod->getOrInsertFunction("Main", voidTy, inputType, outputType, size_ty, nullptr));
319    main->setCallingConv(CallingConv::C);
320    Function::arg_iterator args = main->arg_begin();
321   
322    Value * const inputStream = &*(args++);
323    inputStream->setName("inputStream");
324    Value * const outputStream = &*(args++);
325    outputStream->setName("outputStream");
326    Value * const fileSize = &*(args++);
327    fileSize->setName("fileSize");
328   
329    iBuilder->SetInsertPoint(BasicBlock::Create(mMod->getContext(), "entry", main,0));
330       
331
332    ByteStream.setStreamSetBuffer(inputStream, fileSize);
333    BasisBits.allocateBuffer();
334    U8u16Bits.allocateBuffer();
335    U16Bits.allocateBuffer();
336    DeletionCounts.allocateBuffer();
337    if (mMapBuffering || memAlignBuffering) {
338        U16external.setEmptyBuffer(outputStream);
339    }
340    else {
341        U16out.allocateBuffer();
342    }
343    Value * s2pInstance = s2pk.createInstance({});
344    Value * u8u16Instance = u8u16k.createInstance({});
345    Value * delInstance = delK.createInstance({});
346    Value * p2sInstance = p2sk.createInstance({});
347    Value * stdoutInstance = stdoutK.createInstance({});
348   
349    Type * pthreadTy = size_ty;
350    FunctionType * funVoidPtrVoidTy = FunctionType::get(voidTy, int8PtrTy, false);
351   
352    Function * pthreadCreateFunc = cast<Function>(mMod->getOrInsertFunction("pthread_create",
353                                                                         int32ty,
354                                                                         pthreadTy->getPointerTo(),
355                                                                         voidPtrTy,
356                                                                         static_cast<Type *>(funVoidPtrVoidTy)->getPointerTo(),
357                                                                         voidPtrTy, nullptr));
358    pthreadCreateFunc->setCallingConv(llvm::CallingConv::C);
359    Function * pthreadJoinFunc = cast<Function>(mMod->getOrInsertFunction("pthread_join",
360                                                                       int32ty,
361                                                                       pthreadTy,
362                                                                       PointerType::get(int8PtrTy, 0), nullptr));
363    pthreadJoinFunc->setCallingConv(llvm::CallingConv::C);
364   
365    Function * pthreadExitFunc = cast<Function>(mMod->getOrInsertFunction("pthread_exit",
366                                                                       voidTy, 
367                                                                       voidPtrTy, nullptr));
368    pthreadExitFunc->addFnAttr(llvm::Attribute::NoReturn);
369    pthreadExitFunc->setCallingConv(llvm::CallingConv::C);
370
371    if (segmentPipelineParallel){
372        generateSegmentParallelPipeline(iBuilder, {&s2pk, &u8u16k, &delK, &p2sk, &stdoutK}, {s2pInstance, u8u16Instance, delInstance, p2sInstance, stdoutInstance}, fileSize);
373    }
374    else{
375        generatePipelineLoop(iBuilder, {&s2pk, &u8u16k, &delK, &p2sk, &stdoutK}, {s2pInstance, u8u16Instance, delInstance, p2sInstance, stdoutInstance}, fileSize);
376    }
377
378    iBuilder->CreateRetVoid();
379    return main;
380}
381
382
383
384
385
386typedef void (*u8u16FunctionType)(char * byte_data, char * output_data, size_t filesize);
387
388static ExecutionEngine * u8u16Engine = nullptr;
389
390u8u16FunctionType u8u16CodeGen(void) {
391    LLVMContext TheContext;                           
392    Module * M = new Module("u8u16", TheContext);
393    IDISA::IDISA_Builder * idb = IDISA::GetIDISA_Builder(M);
394
395    pablo::PabloFunction * function = pablo::u8u16_pablo();
396   
397    llvm::Function * main_IR = u8u16Pipeline(M, idb, function);
398   
399    verifyModule(*M, &dbgs());
400    //std::cerr << "ExecuteKernels(); done\n";
401    u8u16Engine = JIT_to_ExecutionEngine(M);
402   
403    u8u16Engine->finalizeObject();
404    //std::cerr << "finalizeObject(); done\n";
405
406    delete idb;
407    return reinterpret_cast<u8u16FunctionType>(u8u16Engine->getPointerToFunction(main_IR));
408}
409
410void u8u16(u8u16FunctionType fn_ptr, const std::string & fileName) {
411    std::string mFileName = fileName;
412    size_t mFileSize;
413    char * mFileBuffer;
414   
415    const boost::filesystem::path file(mFileName);
416    if (exists(file)) {
417        if (is_directory(file)) {
418            return;
419        }
420    } else {
421        std::cerr << "Error: cannot open " << mFileName << " for processing. Skipped.\n";
422        return;
423    }
424   
425    mFileSize = file_size(file);
426    boost::iostreams::mapped_file_source mFile;
427    if (mFileSize == 0) {
428        mFileBuffer = nullptr;
429    }
430    else {
431        try {
432            mFile.open(mFileName);
433        } catch (std::exception &e) {
434            std::cerr << "Error: Boost mmap of " << mFileName << ": " << e.what() << std::endl;
435            return;
436        }
437        mFileBuffer = const_cast<char *>(mFile.data());
438    }
439
440    if (mMapBuffering) {
441        boost::interprocess::mapped_region outputBuffer(boost::interprocess::anonymous_shared_memory(2*mFileSize));
442        outputBuffer.advise(boost::interprocess::mapped_region::advice_willneed);
443        outputBuffer.advise(boost::interprocess::mapped_region::advice_sequential);
444        fn_ptr(mFileBuffer, static_cast<char*>(outputBuffer.get_address()), mFileSize);
445    }
446    else if (memAlignBuffering) {
447        char * outputBuffer;
448        posix_memalign(reinterpret_cast<void **>(&outputBuffer), 32, 2*mFileSize);
449        fn_ptr(mFileBuffer, outputBuffer, mFileSize);
450        free(reinterpret_cast<void *>(outputBuffer));
451    }
452    else {
453        /* No external output buffer */
454        fn_ptr(mFileBuffer, nullptr, mFileSize);
455    }
456    mFile.close();
457   
458}
459
460
461int main(int argc, char *argv[]) {
462    cl::HideUnrelatedOptions(ArrayRef<const cl::OptionCategory *>{&u8u16Options, pablo::pablo_toolchain_flags(), codegen::codegen_flags()});
463    cl::ParseCommandLineOptions(argc, argv);
464
465    u8u16FunctionType fn_ptr = u8u16CodeGen();
466
467    for (unsigned i = 0; i != inputFiles.size(); ++i) {
468        u8u16(fn_ptr, inputFiles[i]);
469    }
470
471    return 0;
472}
473
474                       
Note: See TracBrowser for help on using the repository browser.