diff options
Diffstat (limited to 'kaldi_io/src/kaldi/base')
-rw-r--r-- | kaldi_io/src/kaldi/base/io-funcs-inl.h | 219 | ||||
-rw-r--r-- | kaldi_io/src/kaldi/base/io-funcs.h | 231 | ||||
-rw-r--r-- | kaldi_io/src/kaldi/base/kaldi-common.h | 41 | ||||
-rw-r--r-- | kaldi_io/src/kaldi/base/kaldi-error.h | 153 | ||||
-rw-r--r-- | kaldi_io/src/kaldi/base/kaldi-math.h | 346 | ||||
-rw-r--r-- | kaldi_io/src/kaldi/base/kaldi-types.h | 64 | ||||
-rw-r--r-- | kaldi_io/src/kaldi/base/kaldi-utils.h | 157 | ||||
-rw-r--r-- | kaldi_io/src/kaldi/base/timer.h | 83 |
8 files changed, 1294 insertions, 0 deletions
diff --git a/kaldi_io/src/kaldi/base/io-funcs-inl.h b/kaldi_io/src/kaldi/base/io-funcs-inl.h new file mode 100644 index 0000000..e55458e --- /dev/null +++ b/kaldi_io/src/kaldi/base/io-funcs-inl.h @@ -0,0 +1,219 @@ +// base/io-funcs-inl.h + +// Copyright 2009-2011 Microsoft Corporation; Saarland University; +// Jan Silovsky; Yanmin Qian; Johns Hopkins University (Author: Daniel Povey) + +// See ../../COPYING for clarification regarding multiple authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at + +// http://www.apache.org/licenses/LICENSE-2.0 + +// THIS CODE IS PROVIDED *AS IS* BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, EITHER EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION ANY IMPLIED +// WARRANTIES OR CONDITIONS OF TITLE, FITNESS FOR A PARTICULAR PURPOSE, +// MERCHANTABLITY OR NON-INFRINGEMENT. +// See the Apache 2 License for the specific language governing permissions and +// limitations under the License. + +#ifndef KALDI_BASE_IO_FUNCS_INL_H_ +#define KALDI_BASE_IO_FUNCS_INL_H_ 1 + +// Do not include this file directly. It is included by base/io-funcs.h + +#include <limits> +#include <vector> + +namespace kaldi { + +// Template that covers integers. +template<class T> void WriteBasicType(std::ostream &os, + bool binary, T t) { + // Compile time assertion that this is not called with a wrong type. + KALDI_ASSERT_IS_INTEGER_TYPE(T); + if (binary) { + char len_c = (std::numeric_limits<T>::is_signed ? 1 : -1) + * static_cast<char>(sizeof(t)); + os.put(len_c); + os.write(reinterpret_cast<const char *>(&t), sizeof(t)); + } else { + if (sizeof(t) == 1) + os << static_cast<int16>(t) << " "; + else + os << t << " "; + } + if (os.fail()) { + throw std::runtime_error("Write failure in WriteBasicType."); + } +} + +// Template that covers integers. +template<class T> inline void ReadBasicType(std::istream &is, + bool binary, T *t) { + KALDI_PARANOID_ASSERT(t != NULL); + // Compile time assertion that this is not called with a wrong type. + KALDI_ASSERT_IS_INTEGER_TYPE(T); + if (binary) { + int len_c_in = is.get(); + if (len_c_in == -1) + KALDI_ERR << "ReadBasicType: encountered end of stream."; + char len_c = static_cast<char>(len_c_in), len_c_expected + = (std::numeric_limits<T>::is_signed ? 1 : -1) + * static_cast<char>(sizeof(*t)); + + if (len_c != len_c_expected) { + KALDI_ERR << "ReadBasicType: did not get expected integer type, " + << static_cast<int>(len_c) + << " vs. " << static_cast<int>(len_c_expected) + << ". You can change this code to successfully" + << " read it later, if needed."; + // insert code here to read "wrong" type. Might have a switch statement. + } + is.read(reinterpret_cast<char *>(t), sizeof(*t)); + } else { + if (sizeof(*t) == 1) { + int16 i; + is >> i; + *t = i; + } else { + is >> *t; + } + } + if (is.fail()) { + KALDI_ERR << "Read failure in ReadBasicType, file position is " + << is.tellg() << ", next char is " << is.peek(); + } +} + + +template<class T> inline void WriteIntegerVector(std::ostream &os, bool binary, + const std::vector<T> &v) { + // Compile time assertion that this is not called with a wrong type. + KALDI_ASSERT_IS_INTEGER_TYPE(T); + if (binary) { + char sz = sizeof(T); // this is currently just a check. + os.write(&sz, 1); + int32 vecsz = static_cast<int32>(v.size()); + KALDI_ASSERT((size_t)vecsz == v.size()); + os.write(reinterpret_cast<const char *>(&vecsz), sizeof(vecsz)); + if (vecsz != 0) { + os.write(reinterpret_cast<const char *>(&(v[0])), sizeof(T)*vecsz); + } + } else { + // focus here is on prettiness of text form rather than + // efficiency of reading-in. + // reading-in is dominated by low-level operations anyway: + // for efficiency use binary. + os << "[ "; + typename std::vector<T>::const_iterator iter = v.begin(), end = v.end(); + for (; iter != end; ++iter) { + if (sizeof(T) == 1) + os << static_cast<int16>(*iter) << " "; + else + os << *iter << " "; + } + os << "]\n"; + } + if (os.fail()) { + throw std::runtime_error("Write failure in WriteIntegerType."); + } +} + + +template<class T> inline void ReadIntegerVector(std::istream &is, + bool binary, + std::vector<T> *v) { + KALDI_ASSERT_IS_INTEGER_TYPE(T); + KALDI_ASSERT(v != NULL); + if (binary) { + int sz = is.peek(); + if (sz == sizeof(T)) { + is.get(); + } else { // this is currently just a check. + KALDI_ERR << "ReadIntegerVector: expected to see type of size " + << sizeof(T) << ", saw instead " << sz << ", at file position " + << is.tellg(); + } + int32 vecsz; + is.read(reinterpret_cast<char *>(&vecsz), sizeof(vecsz)); + if (is.fail() || vecsz < 0) goto bad; + v->resize(vecsz); + if (vecsz > 0) { + is.read(reinterpret_cast<char *>(&((*v)[0])), sizeof(T)*vecsz); + } + } else { + std::vector<T> tmp_v; // use temporary so v doesn't use extra memory + // due to resizing. + is >> std::ws; + if (is.peek() != static_cast<int>('[')) { + KALDI_ERR << "ReadIntegerVector: expected to see [, saw " + << is.peek() << ", at file position " << is.tellg(); + } + is.get(); // consume the '['. + is >> std::ws; // consume whitespace. + while (is.peek() != static_cast<int>(']')) { + if (sizeof(T) == 1) { // read/write chars as numbers. + int16 next_t; + is >> next_t >> std::ws; + if (is.fail()) goto bad; + else + tmp_v.push_back((T)next_t); + } else { + T next_t; + is >> next_t >> std::ws; + if (is.fail()) goto bad; + else + tmp_v.push_back(next_t); + } + } + is.get(); // get the final ']'. + *v = tmp_v; // could use std::swap to use less temporary memory, but this + // uses less permanent memory. + } + if (!is.fail()) return; + bad: + KALDI_ERR << "ReadIntegerVector: read failure at file position " + << is.tellg(); +} + +// Initialize an opened stream for writing by writing an optional binary +// header and modifying the floating-point precision. +inline void InitKaldiOutputStream(std::ostream &os, bool binary) { + // This does not throw exceptions (does not check for errors). + if (binary) { + os.put('\0'); + os.put('B'); + } + // Note, in non-binary mode we may at some point want to mess with + // the precision a bit. + // 7 is a bit more than the precision of float.. + if (os.precision() < 7) + os.precision(7); +} + +/// Initialize an opened stream for reading by detecting the binary header and +// setting the "binary" value appropriately. +inline bool InitKaldiInputStream(std::istream &is, bool *binary) { + // Sets the 'binary' variable. + // Throws exception in the very unusual situation that stream + // starts with '\0' but not then 'B'. + + if (is.peek() == '\0') { // seems to be binary + is.get(); + if (is.peek() != 'B') { + return false; + } + is.get(); + *binary = true; + return true; + } else { + *binary = false; + return true; + } +} + +} // end namespace kaldi. + +#endif // KALDI_BASE_IO_FUNCS_INL_H_ diff --git a/kaldi_io/src/kaldi/base/io-funcs.h b/kaldi_io/src/kaldi/base/io-funcs.h new file mode 100644 index 0000000..2bc9da8 --- /dev/null +++ b/kaldi_io/src/kaldi/base/io-funcs.h @@ -0,0 +1,231 @@ +// base/io-funcs.h + +// Copyright 2009-2011 Microsoft Corporation; Saarland University; +// Jan Silovsky; Yanmin Qian + +// See ../../COPYING for clarification regarding multiple authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at + +// http://www.apache.org/licenses/LICENSE-2.0 + +// THIS CODE IS PROVIDED *AS IS* BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, EITHER EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION ANY IMPLIED +// WARRANTIES OR CONDITIONS OF TITLE, FITNESS FOR A PARTICULAR PURPOSE, +// MERCHANTABLITY OR NON-INFRINGEMENT. +// See the Apache 2 License for the specific language governing permissions and +// limitations under the License. + +#ifndef KALDI_BASE_IO_FUNCS_H_ +#define KALDI_BASE_IO_FUNCS_H_ + +// This header only contains some relatively low-level I/O functions. +// The full Kaldi I/O declarations are in ../util/kaldi-io.h +// and ../util/kaldi-table.h +// They were put in util/ in order to avoid making the Matrix library +// dependent on them. + +#include <cctype> +#include <vector> +#include <string> +#include "base/kaldi-common.h" + +namespace kaldi { + + + +/* + This comment describes the Kaldi approach to I/O. All objects can be written + and read in two modes: binary and text. In addition we want to make the I/O + work if we redefine the typedef "BaseFloat" between floats and doubles. + We also want to have control over whitespace in text mode without affecting + the meaning of the file, for pretty-printing purposes. + + Errors are handled by throwing an exception (std::runtime_error). + + For integer and floating-point types (and boolean values): + + WriteBasicType(std::ostream &, bool binary, const T&); + ReadBasicType(std::istream &, bool binary, T*); + + and we expect these functions to be defined in such a way that they work when + the type T changes between float and double, so you can read float into double + and vice versa]. Note that for efficiency and space-saving reasons, the Vector + and Matrix classes do not use these functions [but they preserve the type + interchangeability in their own way] + + For a class (or struct) C: + class C { + .. + Write(std::ostream &, bool binary, [possibly extra optional args for specific classes]) const; + Read(std::istream &, bool binary, [possibly extra optional args for specific classes]); + .. + } + NOTE: The only actual optional args we used are the "add" arguments in + Vector/Matrix classes, which specify whether we should sum the data already + in the class with the data being read. + + For types which are typedef's involving stl classes, I/O is as follows: + typedef std::vector<std::pair<A, B> > MyTypedefName; + + The user should define something like: + + WriteMyTypedefName(std::ostream &, bool binary, const MyTypedefName &t); + ReadMyTypedefName(std::ostream &, bool binary, MyTypedefName *t); + + The user would have to write these functions. + + For a type std::vector<T>: + + void WriteIntegerVector(std::ostream &os, bool binary, const std::vector<T> &v); + void ReadIntegerVector(std::istream &is, bool binary, std::vector<T> *v); + + For other types, e.g. vectors of pairs, the user should create a routine of the + type WriteMyTypedefName. This is to avoid introducing confusing templated functions; + we could easily create templated functions to handle most of these cases but they + would have to share the same name. + + It also often happens that the user needs to write/read special tokens as part + of a file. These might be class headers, or separators/identifiers in the class. + We provide special functions for manipulating these. These special tokens must + be nonempty and must not contain any whitespace. + + void WriteToken(std::ostream &os, bool binary, const char*); + void WriteToken(std::ostream &os, bool binary, const std::string & token); + int Peek(std::istream &is, bool binary); + void ReadToken(std::istream &is, bool binary, std::string *str); + void PeekToken(std::istream &is, bool binary, std::string *str); + + + WriteToken writes the token and one space (whether in binary or text mode). + + Peek returns the first character of the next token, by consuming whitespace + (in text mode) and then returning the peek() character. It returns -1 at EOF; + it doesn't throw. It's useful if a class can have various forms based on + typedefs and virtual classes, and wants to know which version to read. + + ReadToken allow the caller to obtain the next token. PeekToken works just + like ReadToken, but seeks back to the beginning of the token. A subsequent + call to ReadToken will read the same token again. This is useful when + different object types are written to the same file; using PeekToken one can + decide which of the objects to read. + + There is currently no special functionality for writing/reading strings (where the strings + contain data rather than "special tokens" that are whitespace-free and nonempty). This is + because Kaldi is structured in such a way that strings don't appear, except as OpenFst symbol + table entries (and these have their own format). + + + NOTE: you should not call ReadIntegerType and WriteIntegerType with types, + such as int and size_t, that are machine-independent -- at least not + if you want your file formats to port between machines. Use int32 and + int64 where necessary. There is no way to detect this using compile-time + assertions because C++ only keeps track of the internal representation of + the type. +*/ + +/// \addtogroup io_funcs_basic +/// @{ + + +/// WriteBasicType is the name of the write function for bool, integer types, +/// and floating-point types. They all throw on error. +template<class T> void WriteBasicType(std::ostream &os, bool binary, T t); + +/// ReadBasicType is the name of the read function for bool, integer types, +/// and floating-point types. They all throw on error. +template<class T> void ReadBasicType(std::istream &is, bool binary, T *t); + + +// Declare specialization for bool. +template<> +void WriteBasicType<bool>(std::ostream &os, bool binary, bool b); + +template <> +void ReadBasicType<bool>(std::istream &is, bool binary, bool *b); + +// Declare specializations for float and double. +template<> +void WriteBasicType<float>(std::ostream &os, bool binary, float f); + +template<> +void WriteBasicType<double>(std::ostream &os, bool binary, double f); + +template<> +void ReadBasicType<float>(std::istream &is, bool binary, float *f); + +template<> +void ReadBasicType<double>(std::istream &is, bool binary, double *f); + +// Define ReadBasicType that accepts an "add" parameter to add to +// the destination. Caution: if used in Read functions, be careful +// to initialize the parameters concerned to zero in the default +// constructor. +template<class T> +inline void ReadBasicType(std::istream &is, bool binary, T *t, bool add) { + if (!add) { + ReadBasicType(is, binary, t); + } else { + T tmp = T(0); + ReadBasicType(is, binary, &tmp); + *t += tmp; + } +} + +/// Function for writing STL vectors of integer types. +template<class T> inline void WriteIntegerVector(std::ostream &os, bool binary, + const std::vector<T> &v); + +/// Function for reading STL vector of integer types. +template<class T> inline void ReadIntegerVector(std::istream &is, bool binary, + std::vector<T> *v); + +/// The WriteToken functions are for writing nonempty sequences of non-space +/// characters. They are not for general strings. +void WriteToken(std::ostream &os, bool binary, const char *token); +void WriteToken(std::ostream &os, bool binary, const std::string & token); + +/// Peek consumes whitespace (if binary == false) and then returns the peek() +/// value of the stream. +int Peek(std::istream &is, bool binary); + +/// ReadToken gets the next token and puts it in str (exception on failure). +void ReadToken(std::istream &is, bool binary, std::string *token); + +/// PeekToken will return the first character of the next token, or -1 if end of +/// file. It's the same as Peek(), except if the first character is '<' it will +/// skip over it and will return the next character. It will unget the '<' so +/// the stream is where it was before you did PeekToken(). +int PeekToken(std::istream &is, bool binary); + +/// ExpectToken tries to read in the given token, and throws an exception +/// on failure. +void ExpectToken(std::istream &is, bool binary, const char *token); +void ExpectToken(std::istream &is, bool binary, const std::string & token); + +/// ExpectPretty attempts to read the text in "token", but only in non-binary +/// mode. Throws exception on failure. It expects an exact match except that +/// arbitrary whitespace matches arbitrary whitespace. +void ExpectPretty(std::istream &is, bool binary, const char *token); +void ExpectPretty(std::istream &is, bool binary, const std::string & token); + +/// @} end "addtogroup io_funcs_basic" + + +/// InitKaldiOutputStream initializes an opened stream for writing by writing an +/// optional binary header and modifying the floating-point precision; it will +/// typically not be called by users directly. +inline void InitKaldiOutputStream(std::ostream &os, bool binary); + +/// InitKaldiInputStream initializes an opened stream for reading by detecting +/// the binary header and setting the "binary" value appropriately; +/// It will typically not be called by users directly. +inline bool InitKaldiInputStream(std::istream &is, bool *binary); + +} // end namespace kaldi. + +#include "base/io-funcs-inl.h" + +#endif // KALDI_BASE_IO_FUNCS_H_ diff --git a/kaldi_io/src/kaldi/base/kaldi-common.h b/kaldi_io/src/kaldi/base/kaldi-common.h new file mode 100644 index 0000000..33f6f31 --- /dev/null +++ b/kaldi_io/src/kaldi/base/kaldi-common.h @@ -0,0 +1,41 @@ +// base/kaldi-common.h + +// Copyright 2009-2011 Microsoft Corporation + +// See ../../COPYING for clarification regarding multiple authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// THIS CODE IS PROVIDED *AS IS* BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, EITHER EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION ANY IMPLIED +// WARRANTIES OR CONDITIONS OF TITLE, FITNESS FOR A PARTICULAR PURPOSE, +// MERCHANTABLITY OR NON-INFRINGEMENT. +// See the Apache 2 License for the specific language governing permissions and +// limitations under the License. + +#ifndef KALDI_BASE_KALDI_COMMON_H_ +#define KALDI_BASE_KALDI_COMMON_H_ 1 + +#include <cstddef> +#include <cstdlib> +#include <cstring> // C string stuff like strcpy +#include <string> +#include <sstream> +#include <stdexcept> +#include <cassert> +#include <vector> +#include <iostream> +#include <fstream> + +#include "base/kaldi-utils.h" +#include "base/kaldi-error.h" +#include "base/kaldi-types.h" +#include "base/io-funcs.h" +#include "base/kaldi-math.h" + +#endif // KALDI_BASE_KALDI_COMMON_H_ + diff --git a/kaldi_io/src/kaldi/base/kaldi-error.h b/kaldi_io/src/kaldi/base/kaldi-error.h new file mode 100644 index 0000000..8334e42 --- /dev/null +++ b/kaldi_io/src/kaldi/base/kaldi-error.h @@ -0,0 +1,153 @@ +// base/kaldi-error.h + +// Copyright 2009-2011 Microsoft Corporation; Ondrej Glembek; Lukas Burget; +// Saarland University + +// See ../../COPYING for clarification regarding multiple authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// THIS CODE IS PROVIDED *AS IS* BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, EITHER EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION ANY IMPLIED +// WARRANTIES OR CONDITIONS OF TITLE, FITNESS FOR A PARTICULAR PURPOSE, +// MERCHANTABLITY OR NON-INFRINGEMENT. +// See the Apache 2 License for the specific language governing permissions and +// limitations under the License. + +#ifndef KALDI_BASE_KALDI_ERROR_H_ +#define KALDI_BASE_KALDI_ERROR_H_ 1 + +#include <stdexcept> +#include <string> +#include <cstring> +#include <sstream> +#include <cstdio> + +#ifdef _MSC_VER +#define NOEXCEPT(Predicate) +#elif __cplusplus > 199711L || defined(__GXX_EXPERIMENTAL_CXX0X__) +#define NOEXCEPT(Predicate) noexcept((Predicate)) +#else +#define NOEXCEPT(Predicate) +#endif + +#include "base/kaldi-types.h" +#include "base/kaldi-utils.h" + +/* Important that this file does not depend on any other kaldi headers. */ + + +namespace kaldi { + +/// \addtogroup error_group +/// @{ + +/// This is set by util/parse-options.{h, cc} if you set --verbose = ? option +extern int32 g_kaldi_verbose_level; + +/// This is set by util/parse-options.{h, cc} (from argv[0]) and used (if set) +/// in error reporting code to display the name of the program (this is because +/// in our scripts, we often mix together the stderr of many programs). it is +/// the base-name of the program (no directory), followed by ':' We don't use +/// std::string, due to the static initialization order fiasco. +extern const char *g_program_name; + +inline int32 GetVerboseLevel() { return g_kaldi_verbose_level; } + +/// This should be rarely used; command-line programs set the verbose level +/// automatically from ParseOptions. +inline void SetVerboseLevel(int32 i) { g_kaldi_verbose_level = i; } + +// Class KaldiLogMessage is invoked from the KALDI_WARN, KALDI_VLOG and +// KALDI_LOG macros. It prints the message to stderr. Note: we avoid +// using cerr, due to problems with thread safety. fprintf is guaranteed +// thread-safe. + +// class KaldiWarnMessage is invoked from the KALDI_WARN macro. +class KaldiWarnMessage { + public: + inline std::ostream &stream() { return ss; } + KaldiWarnMessage(const char *func, const char *file, int32 line); + ~KaldiWarnMessage() { fprintf(stderr, "%s\n", ss.str().c_str()); } + private: + std::ostringstream ss; +}; + +// class KaldiLogMessage is invoked from the KALDI_LOG macro. +class KaldiLogMessage { + public: + inline std::ostream &stream() { return ss; } + KaldiLogMessage(const char *func, const char *file, int32 line); + ~KaldiLogMessage() { fprintf(stderr, "%s\n", ss.str().c_str()); } + private: + std::ostringstream ss; +}; + +// Class KaldiVlogMessage is invoked from the KALDI_VLOG macro. +class KaldiVlogMessage { + public: + KaldiVlogMessage(const char *func, const char *file, int32 line, + int32 verbose_level); + inline std::ostream &stream() { return ss; } + ~KaldiVlogMessage() { fprintf(stderr, "%s\n", ss.str().c_str()); } + private: + std::ostringstream ss; +}; + + +// class KaldiErrorMessage is invoked from the KALDI_ERROR macro. +// The destructor throws an exception. +class KaldiErrorMessage { + public: + KaldiErrorMessage(const char *func, const char *file, int32 line); + inline std::ostream &stream() { return ss; } + ~KaldiErrorMessage() NOEXCEPT(false); // defined in kaldi-error.cc + private: + std::ostringstream ss; +}; + + + +#ifdef _MSC_VER +#define __func__ __FUNCTION__ +#endif + +#ifndef NDEBUG +#define KALDI_ASSERT(cond) \ + if (!(cond)) kaldi::KaldiAssertFailure_(__func__, __FILE__, __LINE__, #cond); +#else +#define KALDI_ASSERT(cond) +#endif +// also see KALDI_COMPILE_TIME_ASSERT, defined in base/kaldi-utils.h, +// and KALDI_ASSERT_IS_INTEGER_TYPE and KALDI_ASSERT_IS_FLOATING_TYPE, +// also defined there. +#ifdef KALDI_PARANOID // some more expensive asserts only checked if this defined +#define KALDI_PARANOID_ASSERT(cond) \ + if (!(cond)) kaldi::KaldiAssertFailure_(__func__, __FILE__, __LINE__, #cond); +#else +#define KALDI_PARANOID_ASSERT(cond) +#endif + +#define KALDI_ERR kaldi::KaldiErrorMessage(__func__, __FILE__, __LINE__).stream() +#define KALDI_WARN kaldi::KaldiWarnMessage(__func__, __FILE__, __LINE__).stream() +#define KALDI_LOG kaldi::KaldiLogMessage(__func__, __FILE__, __LINE__).stream() + +#define KALDI_VLOG(v) if (v <= kaldi::g_kaldi_verbose_level) \ + kaldi::KaldiVlogMessage(__func__, __FILE__, __LINE__, v).stream() + +inline bool IsKaldiError(const std::string &str) { + return(!strncmp(str.c_str(), "ERROR ", 6)); +} + +void KaldiAssertFailure_(const char *func, const char *file, + int32 line, const char *cond_str); + +/// @} end "addtogroup error_group" + +} // namespace kaldi + +#endif // KALDI_BASE_KALDI_ERROR_H_ diff --git a/kaldi_io/src/kaldi/base/kaldi-math.h b/kaldi_io/src/kaldi/base/kaldi-math.h new file mode 100644 index 0000000..4f60d00 --- /dev/null +++ b/kaldi_io/src/kaldi/base/kaldi-math.h @@ -0,0 +1,346 @@ +// base/kaldi-math.h + +// Copyright 2009-2011 Ondrej Glembek; Microsoft Corporation; Yanmin Qian; +// Jan Silovsky; Saarland University +// +// See ../../COPYING for clarification regarding multiple authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// THIS CODE IS PROVIDED *AS IS* BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, EITHER EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION ANY IMPLIED +// WARRANTIES OR CONDITIONS OF TITLE, FITNESS FOR A PARTICULAR PURPOSE, +// MERCHANTABLITY OR NON-INFRINGEMENT. +// See the Apache 2 License for the specific language governing permissions and +// limitations under the License. + +#ifndef KALDI_BASE_KALDI_MATH_H_ +#define KALDI_BASE_KALDI_MATH_H_ 1 + +#ifdef _MSC_VER +#include <float.h> +#endif + +#include <cmath> +#include <limits> +#include <vector> + +#include "base/kaldi-types.h" +#include "base/kaldi-common.h" + + +#ifndef DBL_EPSILON +#define DBL_EPSILON 2.2204460492503131e-16 +#endif +#ifndef FLT_EPSILON +#define FLT_EPSILON 1.19209290e-7f +#endif + +#ifndef M_PI +# define M_PI 3.1415926535897932384626433832795 +#endif + +#ifndef M_SQRT2 +# define M_SQRT2 1.4142135623730950488016887 +#endif + + +#ifndef M_2PI +# define M_2PI 6.283185307179586476925286766559005 +#endif + +#ifndef M_SQRT1_2 +# define M_SQRT1_2 0.7071067811865475244008443621048490 +#endif + +#ifndef M_LOG_2PI +#define M_LOG_2PI 1.8378770664093454835606594728112 +#endif + +#ifndef M_LN2 +#define M_LN2 0.693147180559945309417232121458 +#endif + +#ifdef _MSC_VER +# define KALDI_ISNAN _isnan +# define KALDI_ISINF(x) (!_isnan(x) && _isnan(x-x)) +# define KALDI_ISFINITE _finite +#else +# define KALDI_ISNAN std::isnan +# define KALDI_ISINF std::isinf +# define KALDI_ISFINITE(x) std::isfinite(x) +#endif +#if !defined(KALDI_SQR) +# define KALDI_SQR(x) ((x) * (x)) +#endif + +namespace kaldi { + +// -infinity +const float kLogZeroFloat = -std::numeric_limits<float>::infinity(); +const double kLogZeroDouble = -std::numeric_limits<double>::infinity(); +const BaseFloat kLogZeroBaseFloat = -std::numeric_limits<BaseFloat>::infinity(); + +// Returns a random integer between 0 and RAND_MAX, inclusive +int Rand(struct RandomState* state=NULL); + +// State for thread-safe random number generator +struct RandomState { + RandomState(); + unsigned seed; +}; + +// Returns a random integer between min and max inclusive. +int32 RandInt(int32 min, int32 max, struct RandomState* state=NULL); + +bool WithProb(BaseFloat prob, struct RandomState* state=NULL); // Returns true with probability "prob", +// with 0 <= prob <= 1 [we check this]. +// Internally calls Rand(). This function is carefully implemented so +// that it should work even if prob is very small. + +/// Returns a random number strictly between 0 and 1. +inline float RandUniform(struct RandomState* state = NULL) { + return static_cast<float>((Rand(state) + 1.0) / (RAND_MAX+2.0)); +} + +inline float RandGauss(struct RandomState* state = NULL) { + return static_cast<float>(sqrtf (-2 * logf(RandUniform(state))) + * cosf(2*M_PI*RandUniform(state))); +} + +// Returns poisson-distributed random number. Uses Knuth's algorithm. +// Take care: this takes time proportinal +// to lambda. Faster algorithms exist but are more complex. +int32 RandPoisson(float lambda, struct RandomState* state=NULL); + +// Returns a pair of gaussian random numbers. Uses Box-Muller transform +void RandGauss2(float *a, float *b, RandomState *state = NULL); +void RandGauss2(double *a, double *b, RandomState *state = NULL); + +// Also see Vector<float,double>::RandCategorical(). + +// This is a randomized pruning mechanism that preserves expectations, +// that we typically use to prune posteriors. +template<class Float> +inline Float RandPrune(Float post, BaseFloat prune_thresh, struct RandomState* state=NULL) { + KALDI_ASSERT(prune_thresh >= 0.0); + if (post == 0.0 || std::abs(post) >= prune_thresh) + return post; + return (post >= 0 ? 1.0 : -1.0) * + (RandUniform(state) <= fabs(post)/prune_thresh ? prune_thresh : 0.0); +} + +static const double kMinLogDiffDouble = std::log(DBL_EPSILON); // negative! +static const float kMinLogDiffFloat = std::log(FLT_EPSILON); // negative! + +inline double LogAdd(double x, double y) { + double diff; + if (x < y) { + diff = x - y; + x = y; + } else { + diff = y - x; + } + // diff is negative. x is now the larger one. + + if (diff >= kMinLogDiffDouble) { + double res; +#ifdef _MSC_VER + res = x + log(1.0 + exp(diff)); +#else + res = x + log1p(exp(diff)); +#endif + return res; + } else { + return x; // return the larger one. + } +} + + +inline float LogAdd(float x, float y) { + float diff; + if (x < y) { + diff = x - y; + x = y; + } else { + diff = y - x; + } + // diff is negative. x is now the larger one. + + if (diff >= kMinLogDiffFloat) { + float res; +#ifdef _MSC_VER + res = x + logf(1.0 + expf(diff)); +#else + res = x + log1pf(expf(diff)); +#endif + return res; + } else { + return x; // return the larger one. + } +} + + +// returns exp(x) - exp(y). +inline double LogSub(double x, double y) { + if (y >= x) { // Throws exception if y>=x. + if (y == x) + return kLogZeroDouble; + else + KALDI_ERR << "Cannot subtract a larger from a smaller number."; + } + + double diff = y - x; // Will be negative. + double res = x + log(1.0 - exp(diff)); + + // res might be NAN if diff ~0.0, and 1.0-exp(diff) == 0 to machine precision + if (KALDI_ISNAN(res)) + return kLogZeroDouble; + return res; +} + + +// returns exp(x) - exp(y). +inline float LogSub(float x, float y) { + if (y >= x) { // Throws exception if y>=x. + if (y == x) + return kLogZeroDouble; + else + KALDI_ERR << "Cannot subtract a larger from a smaller number."; + } + + float diff = y - x; // Will be negative. + float res = x + logf(1.0 - expf(diff)); + + // res might be NAN if diff ~0.0, and 1.0-exp(diff) == 0 to machine precision + if (KALDI_ISNAN(res)) + return kLogZeroFloat; + return res; +} + +/// return abs(a - b) <= relative_tolerance * (abs(a)+abs(b)). +static inline bool ApproxEqual(float a, float b, + float relative_tolerance = 0.001) { + // a==b handles infinities. + if (a==b) return true; + float diff = std::abs(a-b); + if (diff == std::numeric_limits<float>::infinity() + || diff != diff) return false; // diff is +inf or nan. + return (diff <= relative_tolerance*(std::abs(a)+std::abs(b))); +} + +/// assert abs(a - b) <= relative_tolerance * (abs(a)+abs(b)) +static inline void AssertEqual(float a, float b, + float relative_tolerance = 0.001) { + // a==b handles infinities. + KALDI_ASSERT(ApproxEqual(a, b, relative_tolerance)); +} + + +// RoundUpToNearestPowerOfTwo does the obvious thing. It crashes if n <= 0. +int32 RoundUpToNearestPowerOfTwo(int32 n); + +template<class I> I Gcd(I m, I n) { + if (m == 0 || n == 0) { + if (m == 0 && n == 0) { // gcd not defined, as all integers are divisors. + KALDI_ERR << "Undefined GCD since m = 0, n = 0."; + } + return (m == 0 ? (n > 0 ? n : -n) : ( m > 0 ? m : -m)); + // return absolute value of whichever is nonzero + } + // could use compile-time assertion + // but involves messing with complex template stuff. + KALDI_ASSERT(std::numeric_limits<I>::is_integer); + while (1) { + m %= n; + if (m == 0) return (n > 0 ? n : -n); + n %= m; + if (n == 0) return (m > 0 ? m : -m); + } +} + +/// Returns the least common multiple of two integers. Will +/// crash unless the inputs are positive. +template<class I> I Lcm(I m, I n) { + KALDI_ASSERT(m > 0 && n > 0); + I gcd = Gcd(m, n); + return gcd * (m/gcd) * (n/gcd); +} + + +template<class I> void Factorize(I m, std::vector<I> *factors) { + // Splits a number into its prime factors, in sorted order from + // least to greatest, with duplication. A very inefficient + // algorithm, which is mainly intended for use in the + // mixed-radix FFT computation (where we assume most factors + // are small). + KALDI_ASSERT(factors != NULL); + KALDI_ASSERT(m >= 1); // Doesn't work for zero or negative numbers. + factors->clear(); + I small_factors[10] = { 2, 3, 5, 7, 11, 13, 17, 19, 23, 29 }; + + // First try small factors. + for (I i = 0; i < 10; i++) { + if (m == 1) return; // We're done. + while (m % small_factors[i] == 0) { + m /= small_factors[i]; + factors->push_back(small_factors[i]); + } + } + // Next try all odd numbers starting from 31. + for (I j = 31;; j += 2) { + if (m == 1) return; + while (m % j == 0) { + m /= j; + factors->push_back(j); + } + } +} + +inline double Hypot(double x, double y) { return hypot(x, y); } + +inline float Hypot(float x, float y) { return hypotf(x, y); } + +#if !defined(_MSC_VER) || (_MSC_VER >= 1800) +inline double Log1p(double x) { return log1p(x); } + +inline float Log1p(float x) { return log1pf(x); } +#else +inline double Log1p(double x) { + const double cutoff = 1.0e-08; + if (x < cutoff) + return x - 2 * x * x; + else + return log(1.0 + x); +} + +inline float Log1p(float x) { + const float cutoff = 1.0e-07; + if (x < cutoff) + return x - 2 * x * x; + else + return log(1.0 + x); +} +#endif + +inline double Exp(double x) { return exp(x); } + +#ifndef KALDI_NO_EXPF +inline float Exp(float x) { return expf(x); } +#else +inline float Exp(float x) { return exp(x); } +#endif + +inline double Log(double x) { return log(x); } + +inline float Log(float x) { return logf(x); } + + +} // namespace kaldi + + +#endif // KALDI_BASE_KALDI_MATH_H_ diff --git a/kaldi_io/src/kaldi/base/kaldi-types.h b/kaldi_io/src/kaldi/base/kaldi-types.h new file mode 100644 index 0000000..04354b2 --- /dev/null +++ b/kaldi_io/src/kaldi/base/kaldi-types.h @@ -0,0 +1,64 @@ +// base/kaldi-types.h + +// Copyright 2009-2011 Microsoft Corporation; Saarland University; +// Jan Silovsky; Yanmin Qian + +// See ../../COPYING for clarification regarding multiple authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// THIS CODE IS PROVIDED *AS IS* BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, EITHER EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION ANY IMPLIED +// WARRANTIES OR CONDITIONS OF TITLE, FITNESS FOR A PARTICULAR PURPOSE, +// MERCHANTABLITY OR NON-INFRINGEMENT. +// See the Apache 2 License for the specific language governing permissions and +// limitations under the License. + +#ifndef KALDI_BASE_KALDI_TYPES_H_ +#define KALDI_BASE_KALDI_TYPES_H_ 1 + +namespace kaldi { +// TYPEDEFS .................................................................. +#if (KALDI_DOUBLEPRECISION != 0) +typedef double BaseFloat; +#else +typedef float BaseFloat; +#endif +} + +#ifdef _MSC_VER +namespace kaldi { +typedef unsigned __int16 uint16; +typedef unsigned __int32 uint32; +typedef __int16 int16; +typedef __int32 int32; +typedef __int64 int64; +typedef unsigned __int64 uint64; +typedef float float32; +typedef double double64; +} +#include <basetsd.h> +#define ssize_t SSIZE_T + +#else +// we can do this a different way if some platform +// we find in the future lacks stdint.h +#include <stdint.h> + +namespace kaldi { +typedef uint16_t uint16; +typedef uint32_t uint32; +typedef uint64_t uint64; +typedef int16_t int16; +typedef int32_t int32; +typedef int64_t int64; +typedef float float32; +typedef double double64; +} // end namespace kaldi +#endif + +#endif // KALDI_BASE_KALDI_TYPES_H_ diff --git a/kaldi_io/src/kaldi/base/kaldi-utils.h b/kaldi_io/src/kaldi/base/kaldi-utils.h new file mode 100644 index 0000000..1b2c893 --- /dev/null +++ b/kaldi_io/src/kaldi/base/kaldi-utils.h @@ -0,0 +1,157 @@ +// base/kaldi-utils.h + +// Copyright 2009-2011 Ondrej Glembek; Microsoft Corporation; +// Saarland University; Karel Vesely; Yanmin Qian + +// See ../../COPYING for clarification regarding multiple authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// THIS CODE IS PROVIDED *AS IS* BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, EITHER EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION ANY IMPLIED +// WARRANTIES OR CONDITIONS OF TITLE, FITNESS FOR A PARTICULAR PURPOSE, +// MERCHANTABLITY OR NON-INFRINGEMENT. +// See the Apache 2 License for the specific language governing permissions and +// limitations under the License. + +#ifndef KALDI_BASE_KALDI_UTILS_H_ +#define KALDI_BASE_KALDI_UTILS_H_ 1 + +#include <limits> +#include <string> + +#if defined(_MSC_VER) +# define WIN32_LEAN_AND_MEAN +# define NOMINMAX +# include <windows.h> +#endif + +#if defined(_MSC_VER) +#pragma warning(disable: 4244 4056 4305 4800 4267 4996 4756 4661) +#define __restrict__ +#endif + +#ifdef HAVE_POSIX_MEMALIGN +# define KALDI_MEMALIGN(align, size, pp_orig) \ + (!posix_memalign(pp_orig, align, size) ? *(pp_orig) : NULL) +# define KALDI_MEMALIGN_FREE(x) free(x) +#elif defined(HAVE_MEMALIGN) + /* Some systems have memalign() but no declaration for it */ + void * memalign(size_t align, size_t size); +# define KALDI_MEMALIGN(align, size, pp_orig) \ + (*(pp_orig) = memalign(align, size)) +# define KALDI_MEMALIGN_FREE(x) free(x) +#elif defined(_MSC_VER) +# define KALDI_MEMALIGN(align, size, pp_orig) \ + (*(pp_orig) = _aligned_malloc(size, align)) +# define KALDI_MEMALIGN_FREE(x) _aligned_free(x) +#else +#error Manual memory alignment is no longer supported +#endif + +#ifdef __ICC +#pragma warning(disable: 383) // ICPC remark we don't want. +#pragma warning(disable: 810) // ICPC remark we don't want. +#pragma warning(disable: 981) // ICPC remark we don't want. +#pragma warning(disable: 1418) // ICPC remark we don't want. +#pragma warning(disable: 444) // ICPC remark we don't want. +#pragma warning(disable: 869) // ICPC remark we don't want. +#pragma warning(disable: 1287) // ICPC remark we don't want. +#pragma warning(disable: 279) // ICPC remark we don't want. +#pragma warning(disable: 981) // ICPC remark we don't want. +#endif + + +namespace kaldi { + + +// CharToString prints the character in a human-readable form, for debugging. +std::string CharToString(const char &c); + + +inline int MachineIsLittleEndian() { + int check = 1; + return (*reinterpret_cast<char*>(&check) != 0); +} + +// This function kaldi::Sleep() provides a portable way to sleep for a possibly fractional +// number of seconds. On Windows it's only accurate to microseconds. +void Sleep(float seconds); + +} + +#define KALDI_SWAP8(a) { \ + int t = ((char*)&a)[0]; ((char*)&a)[0]=((char*)&a)[7]; ((char*)&a)[7]=t;\ + t = ((char*)&a)[1]; ((char*)&a)[1]=((char*)&a)[6]; ((char*)&a)[6]=t;\ + t = ((char*)&a)[2]; ((char*)&a)[2]=((char*)&a)[5]; ((char*)&a)[5]=t;\ + t = ((char*)&a)[3]; ((char*)&a)[3]=((char*)&a)[4]; ((char*)&a)[4]=t;} +#define KALDI_SWAP4(a) { \ + int t = ((char*)&a)[0]; ((char*)&a)[0]=((char*)&a)[3]; ((char*)&a)[3]=t;\ + t = ((char*)&a)[1]; ((char*)&a)[1]=((char*)&a)[2]; ((char*)&a)[2]=t;} +#define KALDI_SWAP2(a) { \ + int t = ((char*)&a)[0]; ((char*)&a)[0]=((char*)&a)[1]; ((char*)&a)[1]=t;} + + +// Makes copy constructor and operator= private. Same as in compat.h of OpenFst +// toolkit. If using VS, for which this results in compilation errors, we +// do it differently. + +#if defined(_MSC_VER) +#define KALDI_DISALLOW_COPY_AND_ASSIGN(type) \ + void operator = (const type&) +#else +#define KALDI_DISALLOW_COPY_AND_ASSIGN(type) \ + type(const type&); \ + void operator = (const type&) +#endif + +template<bool B> class KaldiCompileTimeAssert { }; +template<> class KaldiCompileTimeAssert<true> { + public: + static inline void Check() { } +}; + +#define KALDI_COMPILE_TIME_ASSERT(b) KaldiCompileTimeAssert<(b)>::Check() + +#define KALDI_ASSERT_IS_INTEGER_TYPE(I) \ + KaldiCompileTimeAssert<std::numeric_limits<I>::is_specialized \ + && std::numeric_limits<I>::is_integer>::Check() + +#define KALDI_ASSERT_IS_FLOATING_TYPE(F) \ + KaldiCompileTimeAssert<std::numeric_limits<F>::is_specialized \ + && !std::numeric_limits<F>::is_integer>::Check() + +#ifdef _MSC_VER +#include <stdio.h> +#define unlink _unlink +#else +#include <unistd.h> +#endif + + +#ifdef _MSC_VER +#define KALDI_STRCASECMP _stricmp +#else +#define KALDI_STRCASECMP strcasecmp +#endif +#ifdef _MSC_VER +# define KALDI_STRTOLL(cur_cstr, end_cstr) _strtoi64(cur_cstr, end_cstr, 10); +#else +# define KALDI_STRTOLL(cur_cstr, end_cstr) strtoll(cur_cstr, end_cstr, 10); +#endif + +#define KALDI_STRTOD(cur_cstr, end_cstr) strtod(cur_cstr, end_cstr) + +#ifdef _MSC_VER +# define KALDI_STRTOF(cur_cstr, end_cstr) \ + static_cast<float>(strtod(cur_cstr, end_cstr)); +#else +# define KALDI_STRTOF(cur_cstr, end_cstr) strtof(cur_cstr, end_cstr); +#endif + +#endif // KALDI_BASE_KALDI_UTILS_H_ + diff --git a/kaldi_io/src/kaldi/base/timer.h b/kaldi_io/src/kaldi/base/timer.h new file mode 100644 index 0000000..d93a461 --- /dev/null +++ b/kaldi_io/src/kaldi/base/timer.h @@ -0,0 +1,83 @@ +// base/timer.h + +// Copyright 2009-2011 Ondrej Glembek; Microsoft Corporation + +// See ../../COPYING for clarification regarding multiple authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at + +// http://www.apache.org/licenses/LICENSE-2.0 + +// THIS CODE IS PROVIDED *AS IS* BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, EITHER EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION ANY IMPLIED +// WARRANTIES OR CONDITIONS OF TITLE, FITNESS FOR A PARTICULAR PURPOSE, +// MERCHANTABLITY OR NON-INFRINGEMENT. +// See the Apache 2 License for the specific language governing permissions and +// limitations under the License. +#ifndef KALDI_BASE_TIMER_H_ +#define KALDI_BASE_TIMER_H_ + +#include "base/kaldi-utils.h" +// Note: Sleep(float secs) is included in base/kaldi-utils.h. + + +#if defined(_MSC_VER) || defined(MINGW) + +namespace kaldi +{ + +class Timer { + public: + Timer() { Reset(); } + void Reset() { + QueryPerformanceCounter(&time_start_); + } + double Elapsed() { + LARGE_INTEGER time_end; + LARGE_INTEGER freq; + QueryPerformanceCounter(&time_end); + if (QueryPerformanceFrequency(&freq) == 0) return 0.0; // Hardware does not support this. + return ((double)time_end.QuadPart - (double)time_start_.QuadPart) / + ((double)freq.QuadPart); + } + private: + LARGE_INTEGER time_start_; +}; +} + +#else + +# include <sys/time.h> +# include <unistd.h> +namespace kaldi +{ +class Timer +{ + public: + Timer() { Reset(); } + + void Reset() { gettimeofday(&this->time_start_, &time_zone_); } + + /// Returns time in seconds. + double Elapsed() { + struct timeval time_end; + gettimeofday(&time_end, &time_zone_); + double t1, t2; + t1 = (double)time_start_.tv_sec + + (double)time_start_.tv_usec/(1000*1000); + t2 = (double)time_end.tv_sec + (double)time_end.tv_usec/(1000*1000); + return t2-t1; + } + + private: + struct timeval time_start_; + struct timezone time_zone_; +}; +} + +#endif + + +#endif |