mirror/tesseract
1
0
Fork 0
mirror of https://github.com/tesseract-ocr/tesseract.git synced 2025-06-10 20:23:12 +08:00
Code Issues Actions 10 Packages Projects Releases Wiki Activity

Use POSIX data types for external interfaces (#1358)

Browse Source 
Replace the Tesseract specific data types in header files which are
part of Debian package libtesseract-dev by POSIX data types.

Update also matching cpp files.

Signed-off-by: Stefan Weil <sw@weilnetz.de>
This commit is contained in:
Stefan Weil 2018-03-13 19:01:40 +01:00 committed by zdenop
parent c6afad03b2
commit 47a326b02d
34 changed files with 167 additions and 167 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

38
ccutil/genericvector.h
View File

@ -346,8 +346,8 @@ class GenericVector {
// vector are small enough that for efficiency it makes sense // vector are small enough that for efficiency it makes sense
// to start with a larger initial size. // to start with a larger initial size.
static const int kDefaultVectorSize = 4; static const int kDefaultVectorSize = 4;
inT32 size_used_; int32_t size_used_;
inT32 size_reserved_; int32_t size_reserved_;
T* data_; T* data_;
TessCallback1<T>* clear_cb_; TessCallback1<T>* clear_cb_;
// Mutable because Run method is not const // Mutable because Run method is not const
@ -536,20 +536,20 @@ class PointerVector : public GenericVector<T*> {
// normal GenericVector of those. // normal GenericVector of those.
// Returns false in case of error. // Returns false in case of error.
bool Serialize(FILE* fp) const { bool Serialize(FILE* fp) const {
inT32 used = GenericVector<T*>::size_used_; int32_t used = GenericVector<T*>::size_used_;
if (fwrite(&used, sizeof(used), 1, fp) != 1) return false; if (fwrite(&used, sizeof(used), 1, fp) != 1) return false;
for (int i = 0; i < used; ++i) { for (int i = 0; i < used; ++i) {
inT8 non_null = GenericVector<T*>::data_[i] != NULL; int8_t non_null = GenericVector<T*>::data_[i] != NULL;
if (fwrite(&non_null, sizeof(non_null), 1, fp) != 1) return false; if (fwrite(&non_null, sizeof(non_null), 1, fp) != 1) return false;
if (non_null && !GenericVector<T*>::data_[i]->Serialize(fp)) return false; if (non_null && !GenericVector<T*>::data_[i]->Serialize(fp)) return false;
} }
return true; return true;
} }
bool Serialize(TFile* fp) const { bool Serialize(TFile* fp) const {
inT32 used = GenericVector<T*>::size_used_; int32_t used = GenericVector<T*>::size_used_;
if (fp->FWrite(&used, sizeof(used), 1) != 1) return false; if (fp->FWrite(&used, sizeof(used), 1) != 1) return false;
for (int i = 0; i < used; ++i) { for (int i = 0; i < used; ++i) {
inT8 non_null = GenericVector<T*>::data_[i] != NULL; int8_t non_null = GenericVector<T*>::data_[i] != NULL;
if (fp->FWrite(&non_null, sizeof(non_null), 1) != 1) return false; if (fp->FWrite(&non_null, sizeof(non_null), 1) != 1) return false;
if (non_null && !GenericVector<T*>::data_[i]->Serialize(fp)) return false; if (non_null && !GenericVector<T*>::data_[i]->Serialize(fp)) return false;
} }
@ -563,13 +563,13 @@ class PointerVector : public GenericVector<T*> {
// Also needs T::T(), as new T is used in this function. // Also needs T::T(), as new T is used in this function.
// Returns false in case of error. // Returns false in case of error.
bool DeSerialize(bool swap, FILE* fp) { bool DeSerialize(bool swap, FILE* fp) {
inT32 reserved; int32_t reserved;
if (fread(&reserved, sizeof(reserved), 1, fp) != 1) return false; if (fread(&reserved, sizeof(reserved), 1, fp) != 1) return false;
if (swap) Reverse32(&reserved); if (swap) Reverse32(&reserved);
GenericVector<T*>::reserve(reserved); GenericVector<T*>::reserve(reserved);
truncate(0); truncate(0);
for (int i = 0; i < reserved; ++i) { for (int i = 0; i < reserved; ++i) {
inT8 non_null; int8_t non_null;
if (fread(&non_null, sizeof(non_null), 1, fp) != 1) return false; if (fread(&non_null, sizeof(non_null), 1, fp) != 1) return false;
T* item = NULL; T* item = NULL;
if (non_null) { if (non_null) {
@ -587,7 +587,7 @@ class PointerVector : public GenericVector<T*> {
return true; return true;
} }
bool DeSerialize(TFile* fp) { bool DeSerialize(TFile* fp) {
inT32 reserved; int32_t reserved;
if (!DeSerializeSize(fp, &reserved)) return false; if (!DeSerializeSize(fp, &reserved)) return false;
GenericVector<T*>::reserve(reserved); GenericVector<T*>::reserve(reserved);
truncate(0); truncate(0);
@ -600,12 +600,12 @@ class PointerVector : public GenericVector<T*> {
// retain the integrity of the stream, the caller must call some combination // retain the integrity of the stream, the caller must call some combination
// of DeSerializeElement and DeSerializeSkip of the exact number returned in // of DeSerializeElement and DeSerializeSkip of the exact number returned in
// *size, assuming a true return. // *size, assuming a true return.
static bool DeSerializeSize(TFile* fp, inT32* size) { static bool DeSerializeSize(TFile* fp, int32_t* size) {
return fp->FReadEndian(size, sizeof(*size), 1) == 1; return fp->FReadEndian(size, sizeof(*size), 1) == 1;
} }
// Reads and appends to the vector the next element of the serialization. // Reads and appends to the vector the next element of the serialization.
bool DeSerializeElement(TFile* fp) { bool DeSerializeElement(TFile* fp) {
inT8 non_null; int8_t non_null;
if (fp->FRead(&non_null, sizeof(non_null), 1) != 1) return false; if (fp->FRead(&non_null, sizeof(non_null), 1) != 1) return false;
T* item = NULL; T* item = NULL;
if (non_null) { if (non_null) {
@ -623,7 +623,7 @@ class PointerVector : public GenericVector<T*> {
} }
// Skips the next element of the serialization. // Skips the next element of the serialization.
static bool DeSerializeSkip(TFile* fp) { static bool DeSerializeSkip(TFile* fp) {
inT8 non_null; int8_t non_null;
if (fp->FRead(&non_null, sizeof(non_null), 1) != 1) return false; if (fp->FRead(&non_null, sizeof(non_null), 1) != 1) return false;
if (non_null) { if (non_null) {
if (!T::SkipDeSerialize(fp)) return false; if (!T::SkipDeSerialize(fp)) return false;
@ -906,7 +906,7 @@ bool GenericVector<T>::write(
template <typename T> template <typename T>
bool GenericVector<T>::read( bool GenericVector<T>::read(
tesseract::TFile* f, TessResultCallback2<bool, tesseract::TFile*, T*>* cb) { tesseract::TFile* f, TessResultCallback2<bool, tesseract::TFile*, T*>* cb) {
inT32 reserved; int32_t reserved;
if (f->FReadEndian(&reserved, sizeof(reserved), 1) != 1) return false; if (f->FReadEndian(&reserved, sizeof(reserved), 1) != 1) return false;
reserve(reserved); reserve(reserved);
if (f->FReadEndian(&size_used_, sizeof(size_used_), 1) != 1) return false; if (f->FReadEndian(&size_used_, sizeof(size_used_), 1) != 1) return false;
@ -947,7 +947,7 @@ bool GenericVector<T>::Serialize(tesseract::TFile* fp) const {
// If swap is true, assumes a big/little-endian swap is needed. // If swap is true, assumes a big/little-endian swap is needed.
template <typename T> template <typename T>
bool GenericVector<T>::DeSerialize(bool swap, FILE* fp) { bool GenericVector<T>::DeSerialize(bool swap, FILE* fp) {
inT32 reserved; int32_t reserved;
if (fread(&reserved, sizeof(reserved), 1, fp) != 1) return false; if (fread(&reserved, sizeof(reserved), 1, fp) != 1) return false;
if (swap) Reverse32(&reserved); if (swap) Reverse32(&reserved);
reserve(reserved); reserve(reserved);
@ -961,7 +961,7 @@ bool GenericVector<T>::DeSerialize(bool swap, FILE* fp) {
} }
template <typename T> template <typename T>
bool GenericVector<T>::DeSerialize(tesseract::TFile* fp) { bool GenericVector<T>::DeSerialize(tesseract::TFile* fp) {
inT32 reserved; int32_t reserved;
if (fp->FReadEndian(&reserved, sizeof(reserved), 1) != 1) return false; if (fp->FReadEndian(&reserved, sizeof(reserved), 1) != 1) return false;
reserve(reserved); reserve(reserved);
size_used_ = reserved; size_used_ = reserved;
@ -969,7 +969,7 @@ bool GenericVector<T>::DeSerialize(tesseract::TFile* fp) {
} }
template <typename T> template <typename T>
bool GenericVector<T>::SkipDeSerialize(tesseract::TFile* fp) { bool GenericVector<T>::SkipDeSerialize(tesseract::TFile* fp) {
inT32 reserved; int32_t reserved;
if (fp->FReadEndian(&reserved, sizeof(reserved), 1) != 1) return false; if (fp->FReadEndian(&reserved, sizeof(reserved), 1) != 1) return false;
return fp->FRead(NULL, sizeof(T), reserved) == reserved; return fp->FRead(NULL, sizeof(T), reserved) == reserved;
} }
@ -1001,7 +1001,7 @@ bool GenericVector<T>::SerializeClasses(tesseract::TFile* fp) const {
// If swap is true, assumes a big/little-endian swap is needed. // If swap is true, assumes a big/little-endian swap is needed.
template <typename T> template <typename T>
bool GenericVector<T>::DeSerializeClasses(bool swap, FILE* fp) { bool GenericVector<T>::DeSerializeClasses(bool swap, FILE* fp) {
inT32 reserved; int32_t reserved;
if (fread(&reserved, sizeof(reserved), 1, fp) != 1) return false; if (fread(&reserved, sizeof(reserved), 1, fp) != 1) return false;
if (swap) Reverse32(&reserved); if (swap) Reverse32(&reserved);
T empty; T empty;
@ -1013,7 +1013,7 @@ bool GenericVector<T>::DeSerializeClasses(bool swap, FILE* fp) {
} }
template <typename T> template <typename T>
bool GenericVector<T>::DeSerializeClasses(tesseract::TFile* fp) { bool GenericVector<T>::DeSerializeClasses(tesseract::TFile* fp) {
inT32 reserved; int32_t reserved;
if (fp->FReadEndian(&reserved, sizeof(reserved), 1) != 1) return false; if (fp->FReadEndian(&reserved, sizeof(reserved), 1) != 1) return false;
T empty; T empty;
init_to_size(reserved, empty); init_to_size(reserved, empty);
@ -1024,7 +1024,7 @@ bool GenericVector<T>::DeSerializeClasses(tesseract::TFile* fp) {
} }
template <typename T> template <typename T>
bool GenericVector<T>::SkipDeSerializeClasses(tesseract::TFile* fp) { bool GenericVector<T>::SkipDeSerializeClasses(tesseract::TFile* fp) {
inT32 reserved; int32_t reserved;
if (fp->FReadEndian(&reserved, sizeof(reserved), 1) != 1) return false; if (fp->FReadEndian(&reserved, sizeof(reserved), 1) != 1) return false;
for (int i = 0; i < reserved; ++i) { for (int i = 0; i < reserved; ++i) {
if (!T::SkipDeSerialize(fp)) return false; if (!T::SkipDeSerialize(fp)) return false;

8
ccutil/helpers.h
View File

@ -42,17 +42,17 @@ class TRand {
public: public:
TRand() : seed_(1) {} TRand() : seed_(1) {}
// Sets the seed to the given value. // Sets the seed to the given value.
void set_seed(uinT64 seed) { void set_seed(uint64_t seed) {
seed_ = seed; seed_ = seed;
} }
// Sets the seed using a hash of a string. // Sets the seed using a hash of a string.
void set_seed(const std::string& str) { void set_seed(const std::string& str) {
std::hash<std::string> hasher; std::hash<std::string> hasher;
set_seed(static_cast<uinT64>(hasher(str))); set_seed(static_cast<uint64_t>(hasher(str)));
} }
// Returns an integer in the range 0 to MAX_INT32. // Returns an integer in the range 0 to MAX_INT32.
inT32 IntRand() { int32_t IntRand() {
Iterate(); Iterate();
return seed_ >> 33; return seed_ >> 33;
} }
@ -73,7 +73,7 @@ class TRand {
} }
// The current value of the seed. // The current value of the seed.
uinT64 seed_; uint64_t seed_;
}; };
} // namespace tesseract } // namespace tesseract

6
ccutil/host.h
View File

@ -65,9 +65,9 @@ typedef unsigned char BOOL8;
#define MAX_UINT32 0xffffffff #define MAX_UINT32 0xffffffff
#define MAX_FLOAT32 std::numeric_limits<float>::max() #define MAX_FLOAT32 std::numeric_limits<float>::max()
#define MIN_INT8 static_cast<inT8>(0x80) #define MIN_INT8 static_cast<int8_t>(0x80)
#define MIN_INT16 static_cast<inT16>(0x8000) #define MIN_INT16 static_cast<int16_t>(0x8000)
#define MIN_INT32 static_cast<inT32>(0x80000000) #define MIN_INT32 static_cast<int32_t>(0x80000000)
#define MIN_UINT8 0x00 #define MIN_UINT8 0x00
#define MIN_UINT16 0x0000 #define MIN_UINT16 0x0000
#define MIN_UINT32 0x00000000 #define MIN_UINT32 0x00000000

4
ccutil/memry.cpp
View File

@ -27,7 +27,7 @@
// TODO(rays) further cleanup by redirecting calls to new and creating proper // TODO(rays) further cleanup by redirecting calls to new and creating proper
// constructors. // constructors.
char *alloc_string(inT32 count) { char *alloc_string(int32_t count) {
// Round up the amount allocated to a multiple of 4 // Round up the amount allocated to a multiple of 4
return static_cast<char*>(malloc((count + 3) & ~3)); return static_cast<char*>(malloc((count + 3) & ~3));
} }
@ -36,7 +36,7 @@ void free_string(char *string) {
free(string); free(string);
} }
void *alloc_mem(inT32 count) { void *alloc_mem(int32_t count) {
return malloc(static_cast<size_t>(count)); return malloc(static_cast<size_t>(count));
} }

4
ccutil/memry.h
View File

@ -24,11 +24,11 @@
#include "host.h" #include "host.h"
// allocate string // allocate string
extern char *alloc_string(inT32 count); extern char *alloc_string(int32_t count);
// free a string. // free a string.
extern void free_string(char *string); extern void free_string(char *string);
// get some memory // get some memory
extern void *alloc_mem(inT32 count); extern void *alloc_mem(int32_t count);
// free mem from alloc_mem // free mem from alloc_mem
extern void free_mem(void *oldchunk); extern void free_mem(void *oldchunk);

34
ccutil/ocrclass.h
View File

@ -80,16 +80,16 @@ typedef struct { /*single character */
// characters sets will need to handle extended characters appropriately, but // characters sets will need to handle extended characters appropriately, but
// *all* code needs to be prepared to receive UTF8 coded characters for // *all* code needs to be prepared to receive UTF8 coded characters for
// characters such as bullet and fancy quotes. // characters such as bullet and fancy quotes.
uinT16 char_code; /*character itself */ uint16_t char_code; /*character itself */
inT16 left; /*of char (-1) */ int16_t left; /*of char (-1) */
inT16 right; /*of char (-1) */ int16_t right; /*of char (-1) */
inT16 top; /*of char (-1) */ int16_t top; /*of char (-1) */
inT16 bottom; /*of char (-1) */ int16_t bottom; /*of char (-1) */
inT16 font_index; /*what font (0) */ int16_t font_index; /*what font (0) */
uinT8 confidence; /*0=perfect, 100=reject (0/100) */ uint8_t confidence; /*0=perfect, 100=reject (0/100) */
uinT8 point_size; /*of char, 72=i inch, (10) */ uint8_t point_size; /*of char, 72=i inch, (10) */
inT8 blanks; /*no of spaces before this char (1) */ int8_t blanks; /*no of spaces before this char (1) */
uinT8 formatting; /*char formatting (0) */ uint8_t formatting; /*char formatting (0) */
} EANYCODE_CHAR; /*single character */ } EANYCODE_CHAR; /*single character */
/********************************************************************** /**********************************************************************
@ -114,14 +114,14 @@ typedef bool (*PROGRESS_FUNC)(int progress, int left, int right, int top,
class ETEXT_DESC { // output header class ETEXT_DESC { // output header
public: public:
inT16 count; /// chars in this buffer(0) int16_t count; /// chars in this buffer(0)
inT16 progress; /// percent complete increasing (0-100) int16_t progress; /// percent complete increasing (0-100)
/** Progress monitor covers word recognition and it does not cover layout /** Progress monitor covers word recognition and it does not cover layout
* analysis. * analysis.
* See Ray comment in https://github.com/tesseract-ocr/tesseract/pull/27 */ * See Ray comment in https://github.com/tesseract-ocr/tesseract/pull/27 */
inT8 more_to_come; /// true if not last int8_t more_to_come; /// true if not last
volatile inT8 ocr_alive; /// ocr sets to 1, HP 0 volatile int8_t ocr_alive; /// ocr sets to 1, HP 0
inT8 err_code; /// for errcode use int8_t err_code; /// for errcode use
CANCEL_FUNC cancel; /// returns true to cancel CANCEL_FUNC cancel; /// returns true to cancel
PROGRESS_FUNC progress_callback; /// called whenever progress increases PROGRESS_FUNC progress_callback; /// called whenever progress increases
void* cancel_this; /// this or other data for cancel void* cancel_this; /// this or other data for cancel
@ -143,9 +143,9 @@ class ETEXT_DESC { // output header
} }
// Sets the end time to be deadline_msecs milliseconds from now. // Sets the end time to be deadline_msecs milliseconds from now.
void set_deadline_msecs(inT32 deadline_msecs) { void set_deadline_msecs(int32_t deadline_msecs) {
gettimeofday(&end_time, NULL); gettimeofday(&end_time, NULL);
inT32 deadline_secs = deadline_msecs / 1000; int32_t deadline_secs = deadline_msecs / 1000;
end_time.tv_sec += deadline_secs; end_time.tv_sec += deadline_secs;
end_time.tv_usec += (deadline_msecs - deadline_secs * 1000) * 1000; end_time.tv_usec += (deadline_msecs - deadline_secs * 1000) * 1000;
if (end_time.tv_usec > 1000000) { if (end_time.tv_usec > 1000000) {

4
ccutil/params.cpp
View File

@ -147,7 +147,7 @@ bool ParamUtils::GetParamAsString(const char *name,
member_params->int_params); member_params->int_params);
if (ip) { if (ip) {
char buf[128]; char buf[128];
snprintf(buf, sizeof(buf), "%d", inT32(*ip)); snprintf(buf, sizeof(buf), "%d", int32_t(*ip));
*value = buf; *value = buf;
return true; return true;
} }
@ -177,7 +177,7 @@ void ParamUtils::PrintParams(FILE *fp, const ParamsVectors *member_params) {
const ParamsVectors *vec = (v == 0) ? GlobalParams() : member_params; const ParamsVectors *vec = (v == 0) ? GlobalParams() : member_params;
for (i = 0; i < vec->int_params.size(); ++i) { for (i = 0; i < vec->int_params.size(); ++i) {
fprintf(fp, "%s\t%d\t%s\n", vec->int_params[i]->name_str(), fprintf(fp, "%s\t%d\t%s\n", vec->int_params[i]->name_str(),
(inT32)(*vec->int_params[i]), vec->int_params[i]->info_str()); (int32_t)(*vec->int_params[i]), vec->int_params[i]->info_str());
} }
for (i = 0; i < vec->bool_params.size(); ++i) { for (i = 0; i < vec->bool_params.size(); ++i) {
fprintf(fp, "%s\t%d\t%s\n", vec->bool_params[i]->name_str(), fprintf(fp, "%s\t%d\t%s\n", vec->bool_params[i]->name_str(),

12
ccutil/params.h
View File

@ -141,7 +141,7 @@ class Param {
class IntParam : public Param { class IntParam : public Param {
public: public:
IntParam(inT32 value, const char *name, const char *comment, bool init, IntParam(int32_t value, const char *name, const char *comment, bool init,
ParamsVectors *vec) : Param(name, comment, init) { ParamsVectors *vec) : Param(name, comment, init) {
value_ = value; value_ = value;
default_ = value; default_ = value;
@ -149,16 +149,16 @@ class IntParam : public Param {
vec->int_params.push_back(this); vec->int_params.push_back(this);
} }
~IntParam() { ParamUtils::RemoveParam<IntParam>(this, params_vec_); } ~IntParam() { ParamUtils::RemoveParam<IntParam>(this, params_vec_); }
operator inT32() const { return value_; } operator int32_t() const { return value_; }
void operator=(inT32 value) { value_ = value; } void operator=(int32_t value) { value_ = value; }
void set_value(inT32 value) { value_ = value; } void set_value(int32_t value) { value_ = value; }
void ResetToDefault() { void ResetToDefault() {
value_ = default_; value_ = default_;
} }
private: private:
inT32 value_; int32_t value_;
inT32 default_; int32_t default_;
// Pointer to the vector that contains this param (not owened by this class). // Pointer to the vector that contains this param (not owened by this class).
GenericVector<IntParam *> *params_vec_; GenericVector<IntParam *> *params_vec_;
}; };

4
ccutil/serialis.cpp
View File

@ -62,9 +62,9 @@ bool TFile::Open(const char* data, int size) {
return true; return true;
} }
bool TFile::Open(FILE* fp, inT64 end_offset) { bool TFile::Open(FILE* fp, int64_t end_offset) {
offset_ = 0; offset_ = 0;
inT64 current_pos = ftell(fp); int64_t current_pos = ftell(fp);
if (end_offset < 0) { if (end_offset < 0) {
if (fseek(fp, 0, SEEK_END)) if (fseek(fp, 0, SEEK_END))
return false; return false;

2
ccutil/serialis.h
View File

@ -60,7 +60,7 @@ class TFile {
// From an existing memory buffer. // From an existing memory buffer.
bool Open(const char* data, int size); bool Open(const char* data, int size);
// From an open file and an end offset. // From an open file and an end offset.
bool Open(FILE* fp, inT64 end_offset); bool Open(FILE* fp, int64_t end_offset);
// Sets the value of the swap flag, so that FReadEndian does the right thing. // Sets the value of the swap flag, so that FReadEndian does the right thing.
void set_swap(bool value) { swap_ = value; } void set_swap(bool value) { swap_ = value; }

26
ccutil/strngs.cpp
View File

@ -67,7 +67,7 @@ void STRING::DiscardData() {
// This is a private method; ensure FixHeader is called (or used_ is well defined) // This is a private method; ensure FixHeader is called (or used_ is well defined)
// beforehand // beforehand
char* STRING::ensure_cstr(inT32 min_capacity) { char* STRING::ensure_cstr(int32_t min_capacity) {
STRING_HEADER* orig_header = GetHeader(); STRING_HEADER* orig_header = GetHeader();
if (min_capacity <= orig_header->capacity_) if (min_capacity <= orig_header->capacity_)
return ((char *)this->data_) + sizeof(STRING_HEADER); return ((char *)this->data_) + sizeof(STRING_HEADER);
@ -146,14 +146,14 @@ STRING::~STRING() {
// TODO(rays) Change all callers to use TFile and remove the old functions. // TODO(rays) Change all callers to use TFile and remove the old functions.
// Writes to the given file. Returns false in case of error. // Writes to the given file. Returns false in case of error.
bool STRING::Serialize(FILE* fp) const { bool STRING::Serialize(FILE* fp) const {
inT32 len = length(); int32_t len = length();
if (fwrite(&len, sizeof(len), 1, fp) != 1) return false; if (fwrite(&len, sizeof(len), 1, fp) != 1) return false;
if (static_cast<int>(fwrite(GetCStr(), 1, len, fp)) != len) return false; if (static_cast<int>(fwrite(GetCStr(), 1, len, fp)) != len) return false;
return true; return true;
} }
// Writes to the given file. Returns false in case of error. // Writes to the given file. Returns false in case of error.
bool STRING::Serialize(TFile* fp) const { bool STRING::Serialize(TFile* fp) const {
inT32 len = length(); int32_t len = length();
if (fp->FWrite(&len, sizeof(len), 1) != 1) return false; if (fp->FWrite(&len, sizeof(len), 1) != 1) return false;
if (fp->FWrite(GetCStr(), 1, len) != len) return false; if (fp->FWrite(GetCStr(), 1, len) != len) return false;
return true; return true;
@ -161,7 +161,7 @@ bool STRING::Serialize(TFile* fp) const {
// Reads from the given file. Returns false in case of error. // Reads from the given file. Returns false in case of error.
// If swap is true, assumes a big/little-endian swap is needed. // If swap is true, assumes a big/little-endian swap is needed.
bool STRING::DeSerialize(bool swap, FILE* fp) { bool STRING::DeSerialize(bool swap, FILE* fp) {
inT32 len; int32_t len;
if (fread(&len, sizeof(len), 1, fp) != 1) return false; if (fread(&len, sizeof(len), 1, fp) != 1) return false;
if (swap) if (swap)
ReverseN(&len, sizeof(len)); ReverseN(&len, sizeof(len));
@ -172,7 +172,7 @@ bool STRING::DeSerialize(bool swap, FILE* fp) {
// Reads from the given file. Returns false in case of error. // Reads from the given file. Returns false in case of error.
// If swap is true, assumes a big/little-endian swap is needed. // If swap is true, assumes a big/little-endian swap is needed.
bool STRING::DeSerialize(TFile* fp) { bool STRING::DeSerialize(TFile* fp) {
inT32 len; int32_t len;
if (fp->FReadEndian(&len, sizeof(len), 1) != 1) return false; if (fp->FReadEndian(&len, sizeof(len), 1) != 1) return false;
truncate_at(len); truncate_at(len);
if (fp->FRead(GetCStr(), 1, len) != len) return false; if (fp->FRead(GetCStr(), 1, len) != len) return false;
@ -181,7 +181,7 @@ bool STRING::DeSerialize(TFile* fp) {
// As DeSerialize, but only seeks past the data - hence a static method. // As DeSerialize, but only seeks past the data - hence a static method.
bool STRING::SkipDeSerialize(tesseract::TFile* fp) { bool STRING::SkipDeSerialize(tesseract::TFile* fp) {
inT32 len; int32_t len;
if (fp->FReadEndian(&len, sizeof(len), 1) != 1) return false; if (fp->FReadEndian(&len, sizeof(len), 1) != 1) return false;
return fp->FRead(NULL, 1, len) == len; return fp->FRead(NULL, 1, len) == len;
} }
@ -190,7 +190,7 @@ BOOL8 STRING::contains(const char c) const {
return (c != '\0') && (strchr (GetCStr(), c) != NULL); return (c != '\0') && (strchr (GetCStr(), c) != NULL);
} }
inT32 STRING::length() const { int32_t STRING::length() const {
FixHeader(); FixHeader();
return GetHeader()->used_ - 1; return GetHeader()->used_ - 1;
} }
@ -218,11 +218,11 @@ const char* STRING::c_str() const {
* Also makes the [] operator return a const so it is immutable * Also makes the [] operator return a const so it is immutable
*/ */
#if STRING_IS_PROTECTED #if STRING_IS_PROTECTED
const char& STRING::operator[](inT32 index) const { const char& STRING::operator[](int32_t index) const {
return GetCStr()[index]; return GetCStr()[index];
} }
void STRING::insert_range(inT32 index, const char* str, int len) { void STRING::insert_range(int32_t index, const char* str, int len) {
// if index is outside current range, then also grow size of string // if index is outside current range, then also grow size of string
// to accmodate the requested range. // to accmodate the requested range.
STRING_HEADER* this_header = GetHeader(); STRING_HEADER* this_header = GetHeader();
@ -255,7 +255,7 @@ void STRING::insert_range(inT32 index, const char* str, int len) {
assert(InvariantOk()); assert(InvariantOk());
} }
void STRING::erase_range(inT32 index, int len) { void STRING::erase_range(int32_t index, int len) {
char* this_cstr = GetCStr(); char* this_cstr = GetCStr();
STRING_HEADER* this_header = GetHeader(); STRING_HEADER* this_header = GetHeader();
@ -266,7 +266,7 @@ void STRING::erase_range(inT32 index, int len) {
} }
#else #else
void STRING::truncate_at(inT32 index) { void STRING::truncate_at(int32_t index) {
ASSERT_HOST(index >= 0); ASSERT_HOST(index >= 0);
FixHeader(); FixHeader();
char* this_cstr = ensure_cstr(index + 1); char* this_cstr = ensure_cstr(index + 1);
@ -275,7 +275,7 @@ void STRING::truncate_at(inT32 index) {
assert(InvariantOk()); assert(InvariantOk());
} }
char& STRING::operator[](inT32 index) const { char& STRING::operator[](int32_t index) const {
// Code is casting away this const and mutating the string, // Code is casting away this const and mutating the string,
// so mark used_ as -1 to flag it unreliable. // so mark used_ as -1 to flag it unreliable.
GetHeader()->used_ = -1; GetHeader()->used_ = -1;
@ -333,7 +333,7 @@ BOOL8 STRING::operator!=(const char* cstr) const {
if (cstr == NULL) if (cstr == NULL)
return this_header->used_ > 1; // either '\0' or NULL return this_header->used_ > 1; // either '\0' or NULL
else { else {
inT32 length = strlen(cstr) + 1; int32_t length = strlen(cstr) + 1;
return (this_header->used_ != length) return (this_header->used_ != length)
|| (memcmp(GetCStr(), cstr, length) != 0); || (memcmp(GetCStr(), cstr, length) != 0);
} }

26
ccutil/strngs.h
View File

@ -65,32 +65,32 @@ class TESS_API STRING
static bool SkipDeSerialize(tesseract::TFile* fp); static bool SkipDeSerialize(tesseract::TFile* fp);
BOOL8 contains(const char c) const; BOOL8 contains(const char c) const;
inT32 length() const; int32_t length() const;
inT32 size() const { return length(); } int32_t size() const { return length(); }
// Workaround to avoid g++ -Wsign-compare warnings. // Workaround to avoid g++ -Wsign-compare warnings.
uinT32 unsigned_size() const { uint32_t unsigned_size() const {
const inT32 len = length(); const int32_t len = length();
assert(0 <= len); assert(0 <= len);
return static_cast<uinT32>(len); return static_cast<uint32_t>(len);
} }
const char *string() const; const char *string() const;
const char *c_str() const; const char *c_str() const;
inline char* strdup() const { inline char* strdup() const {
inT32 len = length() + 1; int32_t len = length() + 1;
return strncpy(new char[len], GetCStr(), len); return strncpy(new char[len], GetCStr(), len);
} }
#if STRING_IS_PROTECTED #if STRING_IS_PROTECTED
const char &operator[] (inT32 index) const; const char &operator[] (int32_t index) const;
// len is number of chars in s to insert starting at index in this string // len is number of chars in s to insert starting at index in this string
void insert_range(inT32 index, const char*s, int len); void insert_range(int32_t index, const char*s, int len);
void erase_range(inT32 index, int len); void erase_range(int32_t index, int len);
#else #else
char &operator[] (inT32 index) const; char &operator[] (int32_t index) const;
#endif #endif
void split(const char c, GenericVector<STRING> *splited); void split(const char c, GenericVector<STRING> *splited);
void truncate_at(inT32 index); void truncate_at(int32_t index);
BOOL8 operator== (const STRING & string) const; BOOL8 operator== (const STRING & string) const;
BOOL8 operator!= (const STRING & string) const; BOOL8 operator!= (const STRING & string) const;
@ -118,7 +118,7 @@ class TESS_API STRING
void add_str_double(const char* str, double number); void add_str_double(const char* str, double number);
// ensure capacity but keep pointer encapsulated // ensure capacity but keep pointer encapsulated
inline void ensure(inT32 min_capacity) { ensure_cstr(min_capacity); } inline void ensure(int32_t min_capacity) { ensure_cstr(min_capacity); }
private: private:
typedef struct STRING_HEADER { typedef struct STRING_HEADER {
@ -171,7 +171,7 @@ class TESS_API STRING
// Ensure string has requested capacity as optimization // Ensure string has requested capacity as optimization
// to avoid unnecessary reallocations. // to avoid unnecessary reallocations.
// The return value is a cstr buffer with at least requested capacity // The return value is a cstr buffer with at least requested capacity
char* ensure_cstr(inT32 min_capacity); char* ensure_cstr(int32_t min_capacity);
void FixHeader() const; // make used_ non-negative, even if const void FixHeader() const; // make used_ non-negative, even if const

8
ccutil/unicharcompress.h
View File

@ -95,12 +95,12 @@ class RecodedCharID {
private: private:
// True if this code is self-normalizing, ie is the master entry for indices // True if this code is self-normalizing, ie is the master entry for indices
// that map to the same code. Has boolean value, but inT8 for serialization. // that map to the same code. Has boolean value, but int8_t for serialization.
inT8 self_normalized_; int8_t self_normalized_;
// The number of elements in use in code_; // The number of elements in use in code_;
inT32 length_; int32_t length_;
// The re-encoded form of the unichar-id to which this RecodedCharID relates. // The re-encoded form of the unichar-id to which this RecodedCharID relates.
inT32 code_[kMaxCodeLen]; int32_t code_[kMaxCodeLen];
}; };
// Class holds a "compression" of a unicharset to simplify the learning problem // Class holds a "compression" of a unicharset to simplify the learning problem

20
ccutil/unicharset.h
View File

@ -136,8 +136,8 @@ class CHAR_FRAGMENT {
// ie did not need chopping to be isolated, but may have been separated // ie did not need chopping to be isolated, but may have been separated
// out from a multi-outline blob. // out from a multi-outline blob.
bool natural; bool natural;
inT16 pos; // fragment position in the character int16_t pos; // fragment position in the character
inT16 total; // total number of fragments in the character int16_t total; // total number of fragments in the character
}; };
// The UNICHARSET class is an utility class for Tesseract that holds the // The UNICHARSET class is an utility class for Tesseract that holds the
@ -582,13 +582,13 @@ class UNICHARSET {
int min_bottom, int max_bottom, int min_bottom, int max_bottom,
int min_top, int max_top) { int min_top, int max_top) {
unichars[unichar_id].properties.min_bottom = unichars[unichar_id].properties.min_bottom =
static_cast<uinT8>(ClipToRange(min_bottom, 0, MAX_UINT8)); static_cast<uint8_t>(ClipToRange(min_bottom, 0, MAX_UINT8));
unichars[unichar_id].properties.max_bottom = unichars[unichar_id].properties.max_bottom =
static_cast<uinT8>(ClipToRange(max_bottom, 0, MAX_UINT8)); static_cast<uint8_t>(ClipToRange(max_bottom, 0, MAX_UINT8));
unichars[unichar_id].properties.min_top = unichars[unichar_id].properties.min_top =
static_cast<uinT8>(ClipToRange(min_top, 0, MAX_UINT8)); static_cast<uint8_t>(ClipToRange(min_top, 0, MAX_UINT8));
unichars[unichar_id].properties.max_top = unichars[unichar_id].properties.max_top =
static_cast<uinT8>(ClipToRange(max_top, 0, MAX_UINT8)); static_cast<uint8_t>(ClipToRange(max_top, 0, MAX_UINT8));
} }
// Returns the width stats (as mean, sd) of the given unichar relative to the // Returns the width stats (as mean, sd) of the given unichar relative to the
// median advance of all characters in the character set. // median advance of all characters in the character set.
@ -933,10 +933,10 @@ class UNICHARSET {
// baseline-normalized coordinates, ie, where the baseline is // baseline-normalized coordinates, ie, where the baseline is
// kBlnBaselineOffset and the meanline is kBlnBaselineOffset + kBlnXHeight // kBlnBaselineOffset and the meanline is kBlnBaselineOffset + kBlnXHeight
// (See normalis.h for the definitions). // (See normalis.h for the definitions).
uinT8 min_bottom; uint8_t min_bottom;
uinT8 max_bottom; uint8_t max_bottom;
uinT8 min_top; uint8_t min_top;
uinT8 max_top; uint8_t max_top;
// Statstics of the widths of bounding box, relative to the median advance. // Statstics of the widths of bounding box, relative to the median advance.
float width; float width;
float width_sd; float width_sd;

4
lstm/convolve.h
View File

@ -63,8 +63,8 @@ class Convolve : public Network {
protected: protected:
// Serialized data. // Serialized data.
inT32 half_x_; int32_t half_x_;
inT32 half_y_; int32_t half_y_;
}; };
} // namespace tesseract. } // namespace tesseract.

4
lstm/fullyconnected.cpp
View File

@ -148,7 +148,7 @@ void FullyConnected::Forward(bool debug, const NetworkIO& input,
#endif #endif
double* temp_line = temp_lines[thread_id]; double* temp_line = temp_lines[thread_id];
const double* d_input = NULL; const double* d_input = NULL;
const inT8* i_input = NULL; const int8_t* i_input = NULL;
if (input.int_mode()) { if (input.int_mode()) {
i_input = input.i(t); i_input = input.i(t);
} else { } else {
@ -188,7 +188,7 @@ void FullyConnected::SetupForward(const NetworkIO& input,
} }
} }
void FullyConnected::ForwardTimeStep(const double* d_input, const inT8* i_input, void FullyConnected::ForwardTimeStep(const double* d_input, const int8_t* i_input,
int t, double* output_line) { int t, double* output_line) {
// input is copied to source_ line-by-line for cache coherency. // input is copied to source_ line-by-line for cache coherency.
if (IsTraining() && external_source_ == NULL && d_input != NULL) if (IsTraining() && external_source_ == NULL && d_input != NULL)

2
lstm/fullyconnected.h
View File

@ -91,7 +91,7 @@ class FullyConnected : public Network {
// Components of Forward so FullyConnected can be reused inside LSTM. // Components of Forward so FullyConnected can be reused inside LSTM.
void SetupForward(const NetworkIO& input, void SetupForward(const NetworkIO& input,
const TransposedArray* input_transpose); const TransposedArray* input_transpose);
void ForwardTimeStep(const double* d_input, const inT8* i_input, int t, void ForwardTimeStep(const double* d_input, const int8_t* i_input, int t,
double* output_line); double* output_line);
// Runs backward propagation of errors on the deltas line. // Runs backward propagation of errors on the deltas line.

2
lstm/lstm.cpp
View File

@ -368,7 +368,7 @@ void LSTM::Forward(bool debug, const NetworkIO& input,
MultiplyVectorsInPlace(ns_, temp_lines[GF1], curr_state); MultiplyVectorsInPlace(ns_, temp_lines[GF1], curr_state);
if (Is2D()) { if (Is2D()) {
// Max-pool the forget gates (in 2-d) instead of blindly adding. // Max-pool the forget gates (in 2-d) instead of blindly adding.
inT8* which_fg_col = which_fg_[t]; int8_t* which_fg_col = which_fg_[t];
memset(which_fg_col, 1, ns_ * sizeof(which_fg_col[0])); memset(which_fg_col, 1, ns_ * sizeof(which_fg_col[0]));
if (valid_2d) { if (valid_2d) {
const double* stepped_state = states[mod_t]; const double* stepped_state = states[mod_t];

8
lstm/lstm.h
View File

@ -129,14 +129,14 @@ class LSTM : public Network {
// Size of padded input to weight matrices = ni_ + no_ for 1-D operation // Size of padded input to weight matrices = ni_ + no_ for 1-D operation
// and ni_ + 2 * no_ for 2-D operation. Note that there is a phantom 1 input // and ni_ + 2 * no_ for 2-D operation. Note that there is a phantom 1 input
// for the bias that makes the weight matrices of size [na + 1][no]. // for the bias that makes the weight matrices of size [na + 1][no].
inT32 na_; int32_t na_;
// Number of internal states. Equal to no_ except for a softmax LSTM. // Number of internal states. Equal to no_ except for a softmax LSTM.
// ns_ is NOT serialized, but is calculated from gate_weights_. // ns_ is NOT serialized, but is calculated from gate_weights_.
inT32 ns_; int32_t ns_;
// Number of additional feedback states. The softmax types feed back // Number of additional feedback states. The softmax types feed back
// additional output information on top of the ns_ internal states. // additional output information on top of the ns_ internal states.
// In the case of a binary-coded (EMBEDDED) softmax, nf_ < no_. // In the case of a binary-coded (EMBEDDED) softmax, nf_ < no_.
inT32 nf_; int32_t nf_;
// Flag indicating 2-D operation. // Flag indicating 2-D operation.
bool is_2d_; bool is_2d_;
@ -149,7 +149,7 @@ class LSTM : public Network {
// Internal state used during forward operation, of size [width, ns]. // Internal state used during forward operation, of size [width, ns].
NetworkIO state_; NetworkIO state_;
// State of the 2-d maxpool, generated during forward, used during backward. // State of the 2-d maxpool, generated during forward, used during backward.
GENERIC_2D_ARRAY<inT8> which_fg_; GENERIC_2D_ARRAY<int8_t> which_fg_;
// Internal state saved from forward, but used only during backward. // Internal state saved from forward, but used only during backward.
NetworkIO node_values_[WT_COUNT]; NetworkIO node_values_[WT_COUNT];
// Preserved input stride_map used for Backward when NT_LSTM_SQUASHED. // Preserved input stride_map used for Backward when NT_LSTM_SQUASHED.

10
lstm/lstmrecognizer.h
View File

@ -223,7 +223,7 @@ class LSTMRecognizer {
protected: protected:
// Sets the random seed from the sample_iteration_; // Sets the random seed from the sample_iteration_;
void SetRandomSeed() { void SetRandomSeed() {
inT64 seed = static_cast<inT64>(sample_iteration_) * 0x10000001; int64_t seed = static_cast<int64_t>(sample_iteration_) * 0x10000001;
randomizer_.set_seed(seed); randomizer_.set_seed(seed);
randomizer_.IntRand(); randomizer_.IntRand();
} }
@ -280,14 +280,14 @@ class LSTMRecognizer {
STRING network_str_; STRING network_str_;
// Flags used to determine the training method of the network. // Flags used to determine the training method of the network.
// See enum TrainingFlags above. // See enum TrainingFlags above.
inT32 training_flags_; int32_t training_flags_;
// Number of actual backward training steps used. // Number of actual backward training steps used.
inT32 training_iteration_; int32_t training_iteration_;
// Index into training sample set. sample_iteration >= training_iteration_. // Index into training sample set. sample_iteration >= training_iteration_.
inT32 sample_iteration_; int32_t sample_iteration_;
// Index in softmax of null character. May take the value UNICHAR_BROKEN or // Index in softmax of null character. May take the value UNICHAR_BROKEN or
// ccutil_.unicharset.size(). // ccutil_.unicharset.size().
inT32 null_char_; int32_t null_char_;
// Learning rate and momentum multipliers of deltas in backprop. // Learning rate and momentum multipliers of deltas in backprop.
float learning_rate_; float learning_rate_;
float momentum_; float momentum_;

6
lstm/lstmtrainer.cpp
View File

@ -88,7 +88,7 @@ LSTMTrainer::LSTMTrainer(FileReader file_reader, FileWriter file_writer,
CheckPointReader checkpoint_reader, CheckPointReader checkpoint_reader,
CheckPointWriter checkpoint_writer, CheckPointWriter checkpoint_writer,
const char* model_base, const char* checkpoint_name, const char* model_base, const char* checkpoint_name,
int debug_interval, inT64 max_memory) int debug_interval, int64_t max_memory)
: randomly_rotate_(false), : randomly_rotate_(false),
training_data_(max_memory), training_data_(max_memory),
file_reader_(file_reader), file_reader_(file_reader),
@ -446,7 +446,7 @@ bool LSTMTrainer::Serialize(SerializeAmount serialize_amount,
if (fp->FWrite(&error_rates_, sizeof(error_rates_), 1) != 1) return false; if (fp->FWrite(&error_rates_, sizeof(error_rates_), 1) != 1) return false;
if (fp->FWrite(&training_stage_, sizeof(training_stage_), 1) != 1) if (fp->FWrite(&training_stage_, sizeof(training_stage_), 1) != 1)
return false; return false;
uinT8 amount = serialize_amount; uint8_t amount = serialize_amount;
if (fp->FWrite(&amount, sizeof(amount), 1) != 1) return false; if (fp->FWrite(&amount, sizeof(amount), 1) != 1) return false;
if (serialize_amount == LIGHT) return true; // We are done. if (serialize_amount == LIGHT) return true; // We are done.
if (fp->FWrite(&best_error_rate_, sizeof(best_error_rate_), 1) != 1) if (fp->FWrite(&best_error_rate_, sizeof(best_error_rate_), 1) != 1)
@ -505,7 +505,7 @@ bool LSTMTrainer::DeSerialize(const TessdataManager* mgr, TFile* fp) {
if (fp->FRead(&error_rates_, sizeof(error_rates_), 1) != 1) return false; if (fp->FRead(&error_rates_, sizeof(error_rates_), 1) != 1) return false;
if (fp->FReadEndian(&training_stage_, sizeof(training_stage_), 1) != 1) if (fp->FReadEndian(&training_stage_, sizeof(training_stage_), 1) != 1)
return false; return false;
uinT8 amount; uint8_t amount;
if (fp->FRead(&amount, sizeof(amount), 1) != 1) return false; if (fp->FRead(&amount, sizeof(amount), 1) != 1) return false;
if (amount == LIGHT) return true; // Don't read the rest. if (amount == LIGHT) return true; // Don't read the rest.
if (fp->FReadEndian(&best_error_rate_, sizeof(best_error_rate_), 1) != 1) if (fp->FReadEndian(&best_error_rate_, sizeof(best_error_rate_), 1) != 1)

2
lstm/lstmtrainer.h
View File

@ -94,7 +94,7 @@ class LSTMTrainer : public LSTMRecognizer {
CheckPointReader checkpoint_reader, CheckPointReader checkpoint_reader,
CheckPointWriter checkpoint_writer, CheckPointWriter checkpoint_writer,
const char* model_base, const char* checkpoint_name, const char* model_base, const char* checkpoint_name,
int debug_interval, inT64 max_memory); int debug_interval, int64_t max_memory);
virtual ~LSTMTrainer(); virtual ~LSTMTrainer();
// Tries to deserialize a trainer from the given file and silently returns // Tries to deserialize a trainer from the given file and silently returns

6
lstm/network.cpp
View File

@ -123,7 +123,7 @@ void Network::SetEnableTraining(TrainingState state) {
// Sets flags that control the action of the network. See NetworkFlags enum // Sets flags that control the action of the network. See NetworkFlags enum
// for bit values. // for bit values.
void Network::SetNetworkFlags(uinT32 flags) { void Network::SetNetworkFlags(uint32_t flags) {
network_flags_ = flags; network_flags_ = flags;
} }
@ -151,7 +151,7 @@ bool Network::SetupNeedsBackprop(bool needs_backprop) {
// Writes to the given file. Returns false in case of error. // Writes to the given file. Returns false in case of error.
bool Network::Serialize(TFile* fp) const { bool Network::Serialize(TFile* fp) const {
inT8 data = NT_NONE; int8_t data = NT_NONE;
if (fp->FWrite(&data, sizeof(data), 1) != 1) return false; if (fp->FWrite(&data, sizeof(data), 1) != 1) return false;
STRING type_name = kTypeNames[type_]; STRING type_name = kTypeNames[type_];
if (!type_name.Serialize(fp)) return false; if (!type_name.Serialize(fp)) return false;
@ -170,7 +170,7 @@ bool Network::Serialize(TFile* fp) const {
// Reads from the given file. Returns false in case of error. // Reads from the given file. Returns false in case of error.
// Should be overridden by subclasses, but NOT called by their DeSerialize. // Should be overridden by subclasses, but NOT called by their DeSerialize.
bool Network::DeSerialize(TFile* fp) { bool Network::DeSerialize(TFile* fp) {
inT8 data = 0; int8_t data = 0;
if (fp->FRead(&data, sizeof(data), 1) != 1) return false; if (fp->FRead(&data, sizeof(data), 1) != 1) return false;
if (data == NT_NONE) { if (data == NT_NONE) {
STRING type_name; STRING type_name;

10
lstm/network.h
View File

@ -164,7 +164,7 @@ class Network {
// Sets flags that control the action of the network. See NetworkFlags enum // Sets flags that control the action of the network. See NetworkFlags enum
// for bit values. // for bit values.
virtual void SetNetworkFlags(uinT32 flags); virtual void SetNetworkFlags(uint32_t flags);
// Sets up the network for training. Initializes weights using weights of // Sets up the network for training. Initializes weights using weights of
// scale `range` picked according to the random number generator `randomizer`. // scale `range` picked according to the random number generator `randomizer`.
@ -299,10 +299,10 @@ class Network {
NetworkType type_; // Type of the derived network class. NetworkType type_; // Type of the derived network class.
TrainingState training_; // Are we currently training? TrainingState training_; // Are we currently training?
bool needs_to_backprop_; // This network needs to output back_deltas. bool needs_to_backprop_; // This network needs to output back_deltas.
inT32 network_flags_; // Behavior control flags in NetworkFlags. int32_t network_flags_; // Behavior control flags in NetworkFlags.
inT32 ni_; // Number of input values. int32_t ni_; // Number of input values.
inT32 no_; // Number of output values. int32_t no_; // Number of output values.
inT32 num_weights_; // Number of weights in this and sub-network. int32_t num_weights_; // Number of weights in this and sub-network.
STRING name_; // A unique name for this layer. STRING name_; // A unique name for this layer.
// NOT-serialized debug data. // NOT-serialized debug data.

28
lstm/networkio.cpp
View File

@ -223,7 +223,7 @@ void NetworkIO::Copy2DImage(int batch, Pix* pix, float black, float contrast,
int num_features = NumFeatures(); int num_features = NumFeatures();
bool color = num_features == 3; bool color = num_features == 3;
if (width > target_width) width = target_width; if (width > target_width) width = target_width;
uinT32* line = pixGetData(pix); uint32_t* line = pixGetData(pix);
for (int y = 0; y < target_height; ++y, line += wpl) { for (int y = 0; y < target_height; ++y, line += wpl) {
int x = 0; int x = 0;
if (y < height) { if (y < height) {
@ -262,7 +262,7 @@ void NetworkIO::Copy1DGreyImage(int batch, Pix* pix, float black,
int x; int x;
for (x = 0; x < width; ++x, ++t) { for (x = 0; x < width; ++x, ++t) {
for (int y = 0; y < height; ++y) { for (int y = 0; y < height; ++y) {
uinT32* line = pixGetData(pix) + wpl * y; uint32_t* line = pixGetData(pix) + wpl * y;
int pixel = GET_DATA_BYTE(line, x); int pixel = GET_DATA_BYTE(line, x);
SetPixel(t, y, pixel, black, contrast); SetPixel(t, y, pixel, black, contrast);
} }
@ -307,7 +307,7 @@ Pix* NetworkIO::ToPix() const {
int im_y = top_im_y; int im_y = top_im_y;
int t = index.t(); int t = index.t();
if (int_mode_) { if (int_mode_) {
const inT8* features = i_[t]; const int8_t* features = i_[t];
for (int y = 0; y < num_features; ++y, im_y += im_height) { for (int y = 0; y < num_features; ++y, im_y += im_height) {
int pixel = features[y * feature_factor]; int pixel = features[y * feature_factor];
// 1 or 2 features use greyscale. // 1 or 2 features use greyscale.
@ -421,7 +421,7 @@ void NetworkIO::ZeroTimeStepGeneral(int t, int offset, int num_features) {
void NetworkIO::Randomize(int t, int offset, int num_features, void NetworkIO::Randomize(int t, int offset, int num_features,
TRand* randomizer) { TRand* randomizer) {
if (int_mode_) { if (int_mode_) {
inT8* line = i_[t] + offset; int8_t* line = i_[t] + offset;
for (int i = 0; i < num_features; ++i) for (int i = 0; i < num_features; ++i)
line[i] = IntCastRounded(randomizer->SignedRand(MAX_INT8)); line[i] = IntCastRounded(randomizer->SignedRand(MAX_INT8));
} else { } else {
@ -602,7 +602,7 @@ bool NetworkIO::AnySuspiciousTruth(float confidence_thr) const {
// Reads a single timestep to floats in the range [-1, 1]. // Reads a single timestep to floats in the range [-1, 1].
void NetworkIO::ReadTimeStep(int t, double* output) const { void NetworkIO::ReadTimeStep(int t, double* output) const {
if (int_mode_) { if (int_mode_) {
const inT8* line = i_[t]; const int8_t* line = i_[t];
for (int i = 0; i < i_.dim2(); ++i) { for (int i = 0; i < i_.dim2(); ++i) {
output[i] = static_cast<double>(line[i]) / MAX_INT8; output[i] = static_cast<double>(line[i]) / MAX_INT8;
} }
@ -618,7 +618,7 @@ void NetworkIO::ReadTimeStep(int t, double* output) const {
void NetworkIO::AddTimeStep(int t, double* inout) const { void NetworkIO::AddTimeStep(int t, double* inout) const {
int num_features = NumFeatures(); int num_features = NumFeatures();
if (int_mode_) { if (int_mode_) {
const inT8* line = i_[t]; const int8_t* line = i_[t];
for (int i = 0; i < num_features; ++i) { for (int i = 0; i < num_features; ++i) {
inout[i] += static_cast<double>(line[i]) / MAX_INT8; inout[i] += static_cast<double>(line[i]) / MAX_INT8;
} }
@ -634,7 +634,7 @@ void NetworkIO::AddTimeStep(int t, double* inout) const {
void NetworkIO::AddTimeStepPart(int t, int offset, int num_features, void NetworkIO::AddTimeStepPart(int t, int offset, int num_features,
float* inout) const { float* inout) const {
if (int_mode_) { if (int_mode_) {
const inT8* line = i_[t] + offset; const int8_t* line = i_[t] + offset;
for (int i = 0; i < num_features; ++i) { for (int i = 0; i < num_features; ++i) {
inout[i] += static_cast<float>(line[i]) / MAX_INT8; inout[i] += static_cast<float>(line[i]) / MAX_INT8;
} }
@ -656,7 +656,7 @@ void NetworkIO::WriteTimeStep(int t, const double* input) {
void NetworkIO::WriteTimeStepPart(int t, int offset, int num_features, void NetworkIO::WriteTimeStepPart(int t, int offset, int num_features,
const double* input) { const double* input) {
if (int_mode_) { if (int_mode_) {
inT8* line = i_[t] + offset; int8_t* line = i_[t] + offset;
for (int i = 0; i < num_features; ++i) { for (int i = 0; i < num_features; ++i) {
line[i] = ClipToRange(IntCastRounded(input[i] * MAX_INT8), line[i] = ClipToRange(IntCastRounded(input[i] * MAX_INT8),
-MAX_INT8, MAX_INT8); -MAX_INT8, MAX_INT8);
@ -675,8 +675,8 @@ void NetworkIO::MaxpoolTimeStep(int dest_t, const NetworkIO& src, int src_t,
ASSERT_HOST(int_mode_ == src.int_mode_); ASSERT_HOST(int_mode_ == src.int_mode_);
if (int_mode_) { if (int_mode_) {
int dim = i_.dim2(); int dim = i_.dim2();
inT8* dest_line = i_[dest_t]; int8_t* dest_line = i_[dest_t];
const inT8* src_line = src.i_[src_t]; const int8_t* src_line = src.i_[src_t];
for (int i = 0; i < dim; ++i) { for (int i = 0; i < dim; ++i) {
if (dest_line[i] < src_line[i]) { if (dest_line[i] < src_line[i]) {
dest_line[i] = src_line[i]; dest_line[i] = src_line[i];
@ -721,7 +721,7 @@ float NetworkIO::MinOfMaxes() const {
for (int t = 0; t < width; ++t) { for (int t = 0; t < width; ++t) {
float max_value = -MAX_FLOAT32; float max_value = -MAX_FLOAT32;
if (int_mode_) { if (int_mode_) {
const inT8* column = i_[t]; const int8_t* column = i_[t];
for (int i = 0; i < num_features; ++i) { for (int i = 0; i < num_features; ++i) {
if (column[i] > max_value) max_value = column[i]; if (column[i] > max_value) max_value = column[i];
} }
@ -747,9 +747,9 @@ void NetworkIO::CombineOutputs(const NetworkIO& base_output,
if (int_mode_) { if (int_mode_) {
// Number of outputs from base and final result. // Number of outputs from base and final result.
for (int t = 0; t < width; ++t) { for (int t = 0; t < width; ++t) {
inT8* out_line = i_[t]; int8_t* out_line = i_[t];
const inT8* base_line = base_output.i_[t]; const int8_t* base_line = base_output.i_[t];
const inT8* comb_line = combiner_output.i_[t]; const int8_t* comb_line = combiner_output.i_[t];
float base_weight = static_cast<float>(comb_line[no]) / MAX_INT8; float base_weight = static_cast<float>(comb_line[no]) / MAX_INT8;
float boost_weight = 1.0f - base_weight; float boost_weight = 1.0f - base_weight;
for (int i = 0; i < no; ++i) { for (int i = 0; i < no; ++i) {

10
lstm/networkio.h
View File

@ -34,7 +34,7 @@ struct Pix;
namespace tesseract { namespace tesseract {
// Class to contain all the input/output of a network, allowing for fixed or // Class to contain all the input/output of a network, allowing for fixed or
// variable-strided 2d to 1d mapping, and float or inT8 values. Provides // variable-strided 2d to 1d mapping, and float or int8_t values. Provides
// enough calculating functions to hide the detail of the implementation. // enough calculating functions to hide the detail of the implementation.
class NetworkIO { class NetworkIO {
public: public:
@ -120,7 +120,7 @@ class NetworkIO {
ASSERT_HOST(!int_mode_); ASSERT_HOST(!int_mode_);
return f_[t]; return f_[t];
} }
const inT8* i(int t) const { const int8_t* i(int t) const {
ASSERT_HOST(int_mode_); ASSERT_HOST(int_mode_);
return i_[t]; return i_[t];
} }
@ -262,8 +262,8 @@ class NetworkIO {
ASSERT_HOST(!v_io.int_mode_); ASSERT_HOST(!v_io.int_mode_);
int dim = f_.dim2(); int dim = f_.dim2();
if (int_mode_) { if (int_mode_) {
const inT8* u = i_[t]; const int8_t* u = i_[t];
const inT8* v = v_io.i_[t]; const int8_t* v = v_io.i_[t];
for (int i = 0; i < dim; ++i) { for (int i = 0; i < dim; ++i) {
product[i] = f(u[i] / static_cast<double>(MAX_INT8)) * v[i] / product[i] = f(u[i] / static_cast<double>(MAX_INT8)) * v[i] /
static_cast<double>(MAX_INT8); static_cast<double>(MAX_INT8);
@ -333,7 +333,7 @@ class NetworkIO {
// Choice of float vs 8 bit int for data. // Choice of float vs 8 bit int for data.
GENERIC_2D_ARRAY<float> f_; GENERIC_2D_ARRAY<float> f_;
GENERIC_2D_ARRAY<inT8> i_; GENERIC_2D_ARRAY<int8_t> i_;
// Which of f_ and i_ are we actually using. // Which of f_ and i_ are we actually using.
bool int_mode_; bool int_mode_;
// Stride for 2d input data. // Stride for 2d input data.

2
lstm/networkscratch.h
View File

@ -242,7 +242,7 @@ class NetworkScratch {
}; // class Stack. }; // class Stack.
private: private:
// If true, the network weights are inT8, if false, float. // If true, the network weights are int8_t, if false, float.
bool int_mode_; bool int_mode_;
// Stacks of NetworkIO and GenericVector<float>. Once allocated, they are not // Stacks of NetworkIO and GenericVector<float>. Once allocated, they are not
// deleted until the NetworkScratch is deleted. // deleted until the NetworkScratch is deleted.

6
lstm/plumbing.cpp
View File

@ -39,7 +39,7 @@ void Plumbing::SetEnableTraining(TrainingState state) {
// Sets flags that control the action of the network. See NetworkFlags enum // Sets flags that control the action of the network. See NetworkFlags enum
// for bit values. // for bit values.
void Plumbing::SetNetworkFlags(uinT32 flags) { void Plumbing::SetNetworkFlags(uint32_t flags) {
Network::SetNetworkFlags(flags); Network::SetNetworkFlags(flags);
for (int i = 0; i < stack_.size(); ++i) for (int i = 0; i < stack_.size(); ++i)
stack_[i]->SetNetworkFlags(flags); stack_[i]->SetNetworkFlags(flags);
@ -184,7 +184,7 @@ float* Plumbing::LayerLearningRatePtr(const char* id) const {
// Writes to the given file. Returns false in case of error. // Writes to the given file. Returns false in case of error.
bool Plumbing::Serialize(TFile* fp) const { bool Plumbing::Serialize(TFile* fp) const {
if (!Network::Serialize(fp)) return false; if (!Network::Serialize(fp)) return false;
inT32 size = stack_.size(); int32_t size = stack_.size();
// Can't use PointerVector::Serialize here as we need a special DeSerialize. // Can't use PointerVector::Serialize here as we need a special DeSerialize.
if (fp->FWrite(&size, sizeof(size), 1) != 1) return false; if (fp->FWrite(&size, sizeof(size), 1) != 1) return false;
for (int i = 0; i < size; ++i) for (int i = 0; i < size; ++i)
@ -200,7 +200,7 @@ bool Plumbing::Serialize(TFile* fp) const {
bool Plumbing::DeSerialize(TFile* fp) { bool Plumbing::DeSerialize(TFile* fp) {
stack_.truncate(0); stack_.truncate(0);
no_ = 0; // We will be modifying this as we AddToStack. no_ = 0; // We will be modifying this as we AddToStack.
inT32 size; int32_t size;
if (fp->FReadEndian(&size, sizeof(size), 1) != 1) return false; if (fp->FReadEndian(&size, sizeof(size), 1) != 1) return false;
for (int i = 0; i < size; ++i) { for (int i = 0; i < size; ++i) {
Network* network = CreateFromFile(fp); Network* network = CreateFromFile(fp);

2
lstm/plumbing.h
View File

@ -49,7 +49,7 @@ class Plumbing : public Network {
// Sets flags that control the action of the network. See NetworkFlags enum // Sets flags that control the action of the network. See NetworkFlags enum
// for bit values. // for bit values.
virtual void SetNetworkFlags(uinT32 flags); virtual void SetNetworkFlags(uint32_t flags);
// Sets up the network for training. Initializes weights using weights of // Sets up the network for training. Initializes weights using weights of
// scale `range` picked according to the random number generator `randomizer`. // scale `range` picked according to the random number generator `randomizer`.

8
lstm/recodebeam.cpp
View File

@ -736,7 +736,7 @@ void RecodeBeamSearch::PushHeapIfBetter(int max_size, int code, int unichar_id,
float score = cert; float score = cert;
if (prev != NULL) score += prev->score; if (prev != NULL) score += prev->score;
if (heap->size() < max_size || score > heap->PeekTop().data.score) { if (heap->size() < max_size || score > heap->PeekTop().data.score) {
uinT64 hash = ComputeCodeHash(code, dup, prev); uint64_t hash = ComputeCodeHash(code, dup, prev);
RecodeNode node(code, unichar_id, permuter, dawg_start, word_start, end, RecodeNode node(code, unichar_id, permuter, dawg_start, word_start, end,
dup, cert, score, prev, d, hash); dup, cert, score, prev, d, hash);
if (UpdateHeapIfMatched(&node, heap)) return; if (UpdateHeapIfMatched(&node, heap)) return;
@ -791,12 +791,12 @@ bool RecodeBeamSearch::UpdateHeapIfMatched(RecodeNode* new_node,
} }
// Computes and returns the code-hash for the given code and prev. // Computes and returns the code-hash for the given code and prev.
uinT64 RecodeBeamSearch::ComputeCodeHash(int code, bool dup, uint64_t RecodeBeamSearch::ComputeCodeHash(int code, bool dup,
const RecodeNode* prev) const { const RecodeNode* prev) const {
uinT64 hash = prev == nullptr ? 0 : prev->code_hash; uint64_t hash = prev == nullptr ? 0 : prev->code_hash;
if (!dup && code != null_char_) { if (!dup && code != null_char_) {
int num_classes = recoder_.code_range(); int num_classes = recoder_.code_range();
uinT64 carry = (((hash >> 32) * num_classes) >> 32); uint64_t carry = (((hash >> 32) * num_classes) >> 32);
hash *= num_classes; hash *= num_classes;
hash += carry; hash += carry;
hash += code; hash += code;

6
lstm/recodebeam.h
View File

@ -102,7 +102,7 @@ struct RecodeNode {
code_hash(0) {} code_hash(0) {}
RecodeNode(int c, int uni_id, PermuterType perm, bool dawg_start, RecodeNode(int c, int uni_id, PermuterType perm, bool dawg_start,
bool word_start, bool end, bool dup, float cert, float s, bool word_start, bool end, bool dup, float cert, float s,
const RecodeNode* p, DawgPositionVector* d, uinT64 hash) const RecodeNode* p, DawgPositionVector* d, uint64_t hash)
: code(c), : code(c),
unichar_id(uni_id), unichar_id(uni_id),
permuter(perm), permuter(perm),
@ -166,7 +166,7 @@ struct RecodeNode {
DawgPositionVector* dawgs; DawgPositionVector* dawgs;
// A hash of all codes in the prefix and this->code as well. Used for // A hash of all codes in the prefix and this->code as well. Used for
// duplicate path removal. // duplicate path removal.
uinT64 code_hash; uint64_t code_hash;
}; };
typedef KDPairInc<double, RecodeNode> RecodePair; typedef KDPairInc<double, RecodeNode> RecodePair;
@ -337,7 +337,7 @@ class RecodeBeamSearch {
// with reshuffle if needed. Returns true if there was a match. // with reshuffle if needed. Returns true if there was a match.
bool UpdateHeapIfMatched(RecodeNode* new_node, RecodeHeap* heap); bool UpdateHeapIfMatched(RecodeNode* new_node, RecodeHeap* heap);
// Computes and returns the code-hash for the given code and prev. // Computes and returns the code-hash for the given code and prev.
uinT64 ComputeCodeHash(int code, bool dup, const RecodeNode* prev) const; uint64_t ComputeCodeHash(int code, bool dup, const RecodeNode* prev) const;
// Backtracks to extract the best path through the lattice that was built // Backtracks to extract the best path through the lattice that was built
// during Decode. On return the best_nodes vector essentially contains the set // during Decode. On return the best_nodes vector essentially contains the set
// of code, score pairs that make the optimal path with the constraint that // of code, score pairs that make the optimal path with the constraint that

4
lstm/reconfig.h
View File

@ -75,8 +75,8 @@ class Reconfig : public Network {
// Non-serialized data used to store parameters between forward and back. // Non-serialized data used to store parameters between forward and back.
StrideMap back_map_; StrideMap back_map_;
// Serialized data. // Serialized data.
inT32 x_scale_; int32_t x_scale_;
inT32 y_scale_; int32_t y_scale_;
}; };
} // namespace tesseract. } // namespace tesseract.

10
lstm/weightmatrix.cpp
View File

@ -103,7 +103,7 @@ void WeightMatrix::ConvertToInt() {
int dim2 = wi_.dim2(); int dim2 = wi_.dim2();
for (int t = 0; t < wi_.dim1(); ++t) { for (int t = 0; t < wi_.dim1(); ++t) {
double* f_line = wf_[t]; double* f_line = wf_[t];
inT8* i_line = wi_[t]; int8_t* i_line = wi_[t];
double max_abs = 0.0; double max_abs = 0.0;
for (int f = 0; f < dim2; ++f) { for (int f = 0; f < dim2; ++f) {
double abs_val = fabs(f_line[f]); double abs_val = fabs(f_line[f]);
@ -133,7 +133,7 @@ void WeightMatrix::InitBackward() {
if (use_adam_) dw_sq_sum_.Resize(no, ni, 0.0); if (use_adam_) dw_sq_sum_.Resize(no, ni, 0.0);
} }
// Flag on mode to indicate that this weightmatrix uses inT8. // Flag on mode to indicate that this weightmatrix uses int8_t.
const int kInt8Flag = 1; const int kInt8Flag = 1;
// Flag on mode to indicate that this weightmatrix uses adam. // Flag on mode to indicate that this weightmatrix uses adam.
const int kAdamFlag = 4; const int kAdamFlag = 4;
@ -146,7 +146,7 @@ const int kDoubleFlag = 128;
bool WeightMatrix::Serialize(bool training, TFile* fp) const { bool WeightMatrix::Serialize(bool training, TFile* fp) const {
// For backward compatibility, add kDoubleFlag to mode to indicate the doubles // For backward compatibility, add kDoubleFlag to mode to indicate the doubles
// format, without errs, so we can detect and read old format weight matrices. // format, without errs, so we can detect and read old format weight matrices.
uinT8 mode = uint8_t mode =
(int_mode_ ? kInt8Flag : 0) | (use_adam_ ? kAdamFlag : 0) | kDoubleFlag; (int_mode_ ? kInt8Flag : 0) | (use_adam_ ? kAdamFlag : 0) | kDoubleFlag;
if (fp->FWrite(&mode, sizeof(mode), 1) != 1) return false; if (fp->FWrite(&mode, sizeof(mode), 1) != 1) return false;
if (int_mode_) { if (int_mode_) {
@ -163,7 +163,7 @@ bool WeightMatrix::Serialize(bool training, TFile* fp) const {
// Reads from the given file. Returns false in case of error. // Reads from the given file. Returns false in case of error.
bool WeightMatrix::DeSerialize(bool training, TFile* fp) { bool WeightMatrix::DeSerialize(bool training, TFile* fp) {
uinT8 mode = 0; uint8_t mode = 0;
if (fp->FRead(&mode, sizeof(mode), 1) != 1) return false; if (fp->FRead(&mode, sizeof(mode), 1) != 1) return false;
int_mode_ = (mode & kInt8Flag) != 0; int_mode_ = (mode & kInt8Flag) != 0;
use_adam_ = (mode & kAdamFlag) != 0; use_adam_ = (mode & kAdamFlag) != 0;
@ -218,7 +218,7 @@ void WeightMatrix::MatrixDotVector(const double* u, double* v) const {
MatrixDotVectorInternal(wf_, true, false, u, v); MatrixDotVectorInternal(wf_, true, false, u, v);
} }
void WeightMatrix::MatrixDotVector(const inT8* u, double* v) const { void WeightMatrix::MatrixDotVector(const int8_t* u, double* v) const {
ASSERT_HOST(int_mode_); ASSERT_HOST(int_mode_);
ASSERT_HOST(multiplier_ != nullptr); ASSERT_HOST(multiplier_ != nullptr);
multiplier_->MatrixDotVector(wi_, scales_, u, v); multiplier_->MatrixDotVector(wi_, scales_, u, v);

6
lstm/weightmatrix.h
View File

@ -60,7 +60,7 @@ class TransposedArray : public GENERIC_2D_ARRAY<double> {
}; // class TransposedArray }; // class TransposedArray
// Generic weight matrix for network layers. Can store the matrix as either // Generic weight matrix for network layers. Can store the matrix as either
// an array of floats or inT8. Provides functions to compute the forward and // an array of floats or int8_t. Provides functions to compute the forward and
// backward steps with the matrix and updates to the weights. // backward steps with the matrix and updates to the weights.
class WeightMatrix { class WeightMatrix {
public: public:
@ -122,7 +122,7 @@ class WeightMatrix {
// implement the bias, but it doesn't actually have it. // implement the bias, but it doesn't actually have it.
// Asserts that the call matches what we have. // Asserts that the call matches what we have.
void MatrixDotVector(const double* u, double* v) const; void MatrixDotVector(const double* u, double* v) const;
void MatrixDotVector(const inT8* u, double* v) const; void MatrixDotVector(const int8_t* u, double* v) const;
// MatrixDotVector for peep weights, MultiplyAccumulate adds the // MatrixDotVector for peep weights, MultiplyAccumulate adds the
// component-wise products of *this[0] and v to inout. // component-wise products of *this[0] and v to inout.
void MultiplyAccumulate(const double* v, double* inout); void MultiplyAccumulate(const double* v, double* inout);
@ -172,7 +172,7 @@ class WeightMatrix {
private: private:
// Choice between float and 8 bit int implementations. // Choice between float and 8 bit int implementations.
GENERIC_2D_ARRAY<double> wf_; GENERIC_2D_ARRAY<double> wf_;
GENERIC_2D_ARRAY<inT8> wi_; GENERIC_2D_ARRAY<int8_t> wi_;
// Transposed copy of wf_, used only for Backward, and set with each Update. // Transposed copy of wf_, used only for Backward, and set with each Update.
TransposedArray wf_t_; TransposedArray wf_t_;
// Which of wf_ and wi_ are we actually using. // Which of wf_ and wi_ are we actually using.