| /* |
| LodePNG version 20190210 |
| |
| Copyright (c) 2005-2019 Lode Vandevenne |
| |
| This software is provided 'as-is', without any express or implied |
| warranty. In no event will the authors be held liable for any damages |
| arising from the use of this software. |
| |
| Permission is granted to anyone to use this software for any purpose, |
| including commercial applications, and to alter it and redistribute it |
| freely, subject to the following restrictions: |
| |
| 1. The origin of this software must not be misrepresented; you must not |
| claim that you wrote the original software. If you use this software |
| in a product, an acknowledgment in the product documentation would be |
| appreciated but is not required. |
| |
| 2. Altered source versions must be plainly marked as such, and must not be |
| misrepresented as being the original software. |
| |
| 3. This notice may not be removed or altered from any source |
| distribution. |
| */ |
| |
| /* |
| The manual and changelog are in the header file "lodepng.h" |
| Rename this file to lodepng.cpp to use it for C++, or to lodepng.c to use it for C. |
| */ |
| |
| #ifdef _MSC_VER |
| #define _CRT_SECURE_NO_DEPRECATE |
| #pragma warning (disable : 4201) |
| |
| #ifndef BASISU_NO_ITERATOR_DEBUG_LEVEL |
| #if defined(_DEBUG) || defined(DEBUG) |
| #define _ITERATOR_DEBUG_LEVEL 1 |
| #define _SECURE_SCL 1 |
| #else |
| #define _SECURE_SCL 0 |
| #define _ITERATOR_DEBUG_LEVEL 0 |
| #endif |
| #endif |
| #endif |
| |
| #include "lodepng.h" |
| |
| #include <limits.h> /* LONG_MAX */ |
| #include <stdio.h> /* file handling */ |
| #include <stdlib.h> /* allocations */ |
| |
| #if defined(_MSC_VER) && (_MSC_VER >= 1310) /*Visual Studio: A few warning types are not desired here.*/ |
| #pragma warning( disable : 4244 ) /*implicit conversions: not warned by gcc -Wall -Wextra and requires too much casts*/ |
| #pragma warning( disable : 4996 ) /*VS does not like fopen, but fopen_s is not standard C so unusable here*/ |
| #endif /*_MSC_VER */ |
| |
| const char* LODEPNG_VERSION_STRING = "20190210"; |
| |
| /* |
| This source file is built up in the following large parts. The code sections |
| with the "LODEPNG_COMPILE_" #defines divide this up further in an intermixed way. |
| -Tools for C and common code for PNG and Zlib |
| -C Code for Zlib (huffman, deflate, ...) |
| -C Code for PNG (file format chunks, adam7, PNG filters, color conversions, ...) |
| -The C++ wrapper around all of the above |
| */ |
| |
| /*The malloc, realloc and free functions defined here with "lodepng_" in front |
| of the name, so that you can easily change them to others related to your |
| platform if needed. Everything else in the code calls these. Pass |
| -DLODEPNG_NO_COMPILE_ALLOCATORS to the compiler, or comment out |
| #define LODEPNG_COMPILE_ALLOCATORS in the header, to disable the ones here and |
| define them in your own project's source files without needing to change |
| lodepng source code. Don't forget to remove "static" if you copypaste them |
| from here.*/ |
| |
| #ifdef LODEPNG_COMPILE_ALLOCATORS |
| static void* lodepng_malloc(size_t size) { |
| #ifdef LODEPNG_MAX_ALLOC |
| if(size > LODEPNG_MAX_ALLOC) return 0; |
| #endif |
| return malloc(size); |
| } |
| |
| static void* lodepng_realloc(void* ptr, size_t new_size) { |
| #ifdef LODEPNG_MAX_ALLOC |
| if(new_size > LODEPNG_MAX_ALLOC) return 0; |
| #endif |
| return realloc(ptr, new_size); |
| } |
| |
| static void lodepng_free(void* ptr) { |
| free(ptr); |
| } |
| #else /*LODEPNG_COMPILE_ALLOCATORS*/ |
| void* lodepng_malloc(size_t size); |
| void* lodepng_realloc(void* ptr, size_t new_size); |
| void lodepng_free(void* ptr); |
| #endif /*LODEPNG_COMPILE_ALLOCATORS*/ |
| |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| /* // Tools for C, and common code for PNG and Zlib. // */ |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| |
| #define LODEPNG_MAX(a, b) (((a) > (b)) ? (a) : (b)) |
| #define LODEPNG_MIN(a, b) (((a) < (b)) ? (a) : (b)) |
| |
| /* |
| Often in case of an error a value is assigned to a variable and then it breaks |
| out of a loop (to go to the cleanup phase of a function). This macro does that. |
| It makes the error handling code shorter and more readable. |
| |
| Example: if(!uivector_resizev(&frequencies_ll, 286, 0)) ERROR_BREAK(83); |
| */ |
| #define CERROR_BREAK(errorvar, code){\ |
| errorvar = code;\ |
| break;\ |
| } |
| |
| /*version of CERROR_BREAK that assumes the common case where the error variable is named "error"*/ |
| #define ERROR_BREAK(code) CERROR_BREAK(error, code) |
| |
| /*Set error var to the error code, and return it.*/ |
| #define CERROR_RETURN_ERROR(errorvar, code){\ |
| errorvar = code;\ |
| return code;\ |
| } |
| |
| /*Try the code, if it returns error, also return the error.*/ |
| #define CERROR_TRY_RETURN(call){\ |
| unsigned error = call;\ |
| if(error) return error;\ |
| } |
| |
| /*Set error var to the error code, and return from the void function.*/ |
| #define CERROR_RETURN(errorvar, code){\ |
| errorvar = code;\ |
| return;\ |
| } |
| |
| /* |
| About uivector, ucvector and string: |
| -All of them wrap dynamic arrays or text strings in a similar way. |
| -LodePNG was originally written in C++. The vectors replace the std::vectors that were used in the C++ version. |
| -The string tools are made to avoid problems with compilers that declare things like strncat as deprecated. |
| -They're not used in the interface, only internally in this file as static functions. |
| -As with many other structs in this file, the init and cleanup functions serve as ctor and dtor. |
| */ |
| |
| #ifdef LODEPNG_COMPILE_ZLIB |
| /*dynamic vector of unsigned ints*/ |
| typedef struct uivector { |
| unsigned* data; |
| size_t size; /*size in number of unsigned longs*/ |
| size_t allocsize; /*allocated size in bytes*/ |
| } uivector; |
| |
| static void uivector_cleanup(void* p) { |
| ((uivector*)p)->size = ((uivector*)p)->allocsize = 0; |
| lodepng_free(((uivector*)p)->data); |
| ((uivector*)p)->data = NULL; |
| } |
| |
| /*returns 1 if success, 0 if failure ==> nothing done*/ |
| static unsigned uivector_reserve(uivector* p, size_t allocsize) { |
| if(allocsize > p->allocsize) { |
| size_t newsize = (allocsize > p->allocsize * 2) ? allocsize : (allocsize * 3 / 2); |
| void* data = lodepng_realloc(p->data, newsize); |
| if(data) { |
| p->allocsize = newsize; |
| p->data = (unsigned*)data; |
| } |
| else return 0; /*error: not enough memory*/ |
| } |
| return 1; |
| } |
| |
| /*returns 1 if success, 0 if failure ==> nothing done*/ |
| static unsigned uivector_resize(uivector* p, size_t size) { |
| if(!uivector_reserve(p, size * sizeof(unsigned))) return 0; |
| p->size = size; |
| return 1; /*success*/ |
| } |
| |
| /*resize and give all new elements the value*/ |
| static unsigned uivector_resizev(uivector* p, size_t size, unsigned value) { |
| size_t oldsize = p->size, i; |
| if(!uivector_resize(p, size)) return 0; |
| for(i = oldsize; i < size; ++i) p->data[i] = value; |
| return 1; |
| } |
| |
| static void uivector_init(uivector* p) { |
| p->data = NULL; |
| p->size = p->allocsize = 0; |
| } |
| |
| #ifdef LODEPNG_COMPILE_ENCODER |
| /*returns 1 if success, 0 if failure ==> nothing done*/ |
| static unsigned uivector_push_back(uivector* p, unsigned c) { |
| if(!uivector_resize(p, p->size + 1)) return 0; |
| if (!p->data) return 0; |
| p->data[p->size - 1] = c; |
| return 1; |
| } |
| #endif /*LODEPNG_COMPILE_ENCODER*/ |
| #endif /*LODEPNG_COMPILE_ZLIB*/ |
| |
| /* /////////////////////////////////////////////////////////////////////////// */ |
| |
| /*dynamic vector of unsigned chars*/ |
| typedef struct ucvector { |
| unsigned char* data; |
| size_t size; /*used size*/ |
| size_t allocsize; /*allocated size*/ |
| } ucvector; |
| |
| /*returns 1 if success, 0 if failure ==> nothing done*/ |
| static unsigned ucvector_reserve(ucvector* p, size_t allocsize) { |
| if(allocsize > p->allocsize) { |
| size_t newsize = (allocsize > p->allocsize * 2) ? allocsize : (allocsize * 3 / 2); |
| void* data = lodepng_realloc(p->data, newsize); |
| if(data) { |
| p->allocsize = newsize; |
| p->data = (unsigned char*)data; |
| } |
| else return 0; /*error: not enough memory*/ |
| } |
| return 1; |
| } |
| |
| /*returns 1 if success, 0 if failure ==> nothing done*/ |
| static unsigned ucvector_resize(ucvector* p, size_t size) { |
| if(!ucvector_reserve(p, size * sizeof(unsigned char))) return 0; |
| p->size = size; |
| return 1; /*success*/ |
| } |
| |
| #ifdef LODEPNG_COMPILE_PNG |
| |
| static void ucvector_cleanup(void* p) { |
| ((ucvector*)p)->size = ((ucvector*)p)->allocsize = 0; |
| lodepng_free(((ucvector*)p)->data); |
| ((ucvector*)p)->data = NULL; |
| } |
| |
| static void ucvector_init(ucvector* p) { |
| p->data = NULL; |
| p->size = p->allocsize = 0; |
| } |
| #endif /*LODEPNG_COMPILE_PNG*/ |
| |
| #ifdef LODEPNG_COMPILE_ZLIB |
| /*you can both convert from vector to buffer&size and vica versa. If you use |
| init_buffer to take over a buffer and size, it is not needed to use cleanup*/ |
| static void ucvector_init_buffer(ucvector* p, unsigned char* buffer, size_t size) { |
| p->data = buffer; |
| p->allocsize = p->size = size; |
| } |
| #endif /*LODEPNG_COMPILE_ZLIB*/ |
| |
| #if (defined(LODEPNG_COMPILE_PNG) && defined(LODEPNG_COMPILE_ANCILLARY_CHUNKS)) || defined(LODEPNG_COMPILE_ENCODER) |
| /*returns 1 if success, 0 if failure ==> nothing done*/ |
| static unsigned ucvector_push_back(ucvector* p, unsigned char c) { |
| if(!ucvector_resize(p, p->size + 1)) return 0; |
| p->data[p->size - 1] = c; |
| return 1; |
| } |
| #endif /*defined(LODEPNG_COMPILE_PNG) || defined(LODEPNG_COMPILE_ENCODER)*/ |
| |
| |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| |
| #ifdef LODEPNG_COMPILE_PNG |
| #ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS |
| |
| /*free string pointer and set it to NULL*/ |
| static void string_cleanup(char** out) { |
| lodepng_free(*out); |
| *out = NULL; |
| } |
| |
| /* dynamically allocates a new string with a copy of the null terminated input text */ |
| static char* alloc_string(const char* in) { |
| size_t insize = strlen(in); |
| char* out = (char*)lodepng_malloc(insize + 1); |
| if(out) { |
| size_t i; |
| for(i = 0; i != insize; ++i) { |
| out[i] = in[i]; |
| } |
| out[i] = 0; |
| } |
| return out; |
| } |
| #endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/ |
| #endif /*LODEPNG_COMPILE_PNG*/ |
| |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| |
| unsigned lodepng_read32bitInt(const unsigned char* buffer) { |
| return (unsigned)((buffer[0] << 24) | (buffer[1] << 16) | (buffer[2] << 8) | buffer[3]); |
| } |
| |
| #if defined(LODEPNG_COMPILE_PNG) || defined(LODEPNG_COMPILE_ENCODER) |
| /*buffer must have at least 4 allocated bytes available*/ |
| static void lodepng_set32bitInt(unsigned char* buffer, unsigned value) { |
| buffer[0] = (unsigned char)((value >> 24) & 0xff); |
| buffer[1] = (unsigned char)((value >> 16) & 0xff); |
| buffer[2] = (unsigned char)((value >> 8) & 0xff); |
| buffer[3] = (unsigned char)((value ) & 0xff); |
| } |
| #endif /*defined(LODEPNG_COMPILE_PNG) || defined(LODEPNG_COMPILE_ENCODER)*/ |
| |
| #ifdef LODEPNG_COMPILE_ENCODER |
| static void lodepng_add32bitInt(ucvector* buffer, unsigned value) { |
| ucvector_resize(buffer, buffer->size + 4); /*todo: give error if resize failed*/ |
| lodepng_set32bitInt(&buffer->data[buffer->size - 4], value); |
| } |
| #endif /*LODEPNG_COMPILE_ENCODER*/ |
| |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| /* / File IO / */ |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| |
| #ifdef LODEPNG_COMPILE_DISK |
| |
| /* returns negative value on error. This should be pure C compatible, so no fstat. */ |
| static long lodepng_filesize(const char* filename) { |
| FILE* file; |
| long size; |
| file = fopen(filename, "rb"); |
| if(!file) return -1; |
| |
| if(fseek(file, 0, SEEK_END) != 0) { |
| fclose(file); |
| return -1; |
| } |
| |
| size = ftell(file); |
| /* It may give LONG_MAX as directory size, this is invalid for us. */ |
| if(size == LONG_MAX) size = -1; |
| |
| fclose(file); |
| return size; |
| } |
| |
| /* load file into buffer that already has the correct allocated size. Returns error code.*/ |
| static unsigned lodepng_buffer_file(unsigned char* out, size_t size, const char* filename) { |
| FILE* file; |
| size_t readsize; |
| file = fopen(filename, "rb"); |
| if(!file) return 78; |
| |
| readsize = fread(out, 1, size, file); |
| fclose(file); |
| |
| if (readsize != size) return 78; |
| return 0; |
| } |
| |
| unsigned lodepng_load_file(unsigned char** out, size_t* outsize, const char* filename) { |
| long size = lodepng_filesize(filename); |
| if (size < 0) return 78; |
| *outsize = (size_t)size; |
| |
| *out = (unsigned char*)lodepng_malloc((size_t)size); |
| if(!(*out) && size > 0) return 83; /*the above malloc failed*/ |
| |
| return lodepng_buffer_file(*out, (size_t)size, filename); |
| } |
| |
| /*write given buffer to the file, overwriting the file, it doesn't append to it.*/ |
| unsigned lodepng_save_file(const unsigned char* buffer, size_t buffersize, const char* filename) { |
| FILE* file; |
| file = fopen(filename, "wb" ); |
| if(!file) return 79; |
| fwrite(buffer, 1, buffersize, file); |
| fclose(file); |
| return 0; |
| } |
| |
| #endif /*LODEPNG_COMPILE_DISK*/ |
| |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| /* // End of common code and tools. Begin of Zlib related code. // */ |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| |
| #ifdef LODEPNG_COMPILE_ZLIB |
| #ifdef LODEPNG_COMPILE_ENCODER |
| /*TODO: this ignores potential out of memory errors*/ |
| #define addBitToStream(/*size_t**/ bitpointer, /*ucvector**/ bitstream, /*unsigned char*/ bit){\ |
| /*add a new byte at the end*/\ |
| if(((*bitpointer) & 7) == 0) ucvector_push_back(bitstream, (unsigned char)0);\ |
| /*earlier bit of huffman code is in a lesser significant bit of an earlier byte*/\ |
| (bitstream->data[bitstream->size - 1]) |= (bit << ((*bitpointer) & 0x7));\ |
| ++(*bitpointer);\ |
| } |
| |
| static void addBitsToStream(size_t* bitpointer, ucvector* bitstream, unsigned value, size_t nbits) { |
| size_t i; |
| for(i = 0; i != nbits; ++i) addBitToStream(bitpointer, bitstream, (unsigned char)((value >> i) & 1)); |
| } |
| |
| static void addBitsToStreamReversed(size_t* bitpointer, ucvector* bitstream, unsigned value, size_t nbits) { |
| size_t i; |
| for(i = 0; i != nbits; ++i) addBitToStream(bitpointer, bitstream, (unsigned char)((value >> (nbits - 1 - i)) & 1)); |
| } |
| #endif /*LODEPNG_COMPILE_ENCODER*/ |
| |
| #ifdef LODEPNG_COMPILE_DECODER |
| |
| #define READBIT(bitpointer, bitstream) ((bitstream[bitpointer >> 3] >> (bitpointer & 0x7)) & (unsigned char)1) |
| |
| static unsigned char readBitFromStream(size_t* bitpointer, const unsigned char* bitstream) { |
| unsigned char result = (unsigned char)(READBIT(*bitpointer, bitstream)); |
| ++(*bitpointer); |
| return result; |
| } |
| |
| static unsigned readBitsFromStream(size_t* bitpointer, const unsigned char* bitstream, size_t nbits) { |
| unsigned result = 0, i; |
| for(i = 0; i != nbits; ++i) { |
| result += ((unsigned)READBIT(*bitpointer, bitstream)) << i; |
| ++(*bitpointer); |
| } |
| return result; |
| } |
| #endif /*LODEPNG_COMPILE_DECODER*/ |
| |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| /* / Deflate - Huffman / */ |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| |
| #define FIRST_LENGTH_CODE_INDEX 257 |
| #define LAST_LENGTH_CODE_INDEX 285 |
| /*256 literals, the end code, some length codes, and 2 unused codes*/ |
| #define NUM_DEFLATE_CODE_SYMBOLS 288 |
| /*the distance codes have their own symbols, 30 used, 2 unused*/ |
| #define NUM_DISTANCE_SYMBOLS 32 |
| /*the code length codes. 0-15: code lengths, 16: copy previous 3-6 times, 17: 3-10 zeros, 18: 11-138 zeros*/ |
| #define NUM_CODE_LENGTH_CODES 19 |
| |
| /*the base lengths represented by codes 257-285*/ |
| static const unsigned LENGTHBASE[29] |
| = {3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31, 35, 43, 51, 59, |
| 67, 83, 99, 115, 131, 163, 195, 227, 258}; |
| |
| /*the extra bits used by codes 257-285 (added to base length)*/ |
| static const unsigned LENGTHEXTRA[29] |
| = {0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, |
| 4, 4, 4, 4, 5, 5, 5, 5, 0}; |
| |
| /*the base backwards distances (the bits of distance codes appear after length codes and use their own huffman tree)*/ |
| static const unsigned DISTANCEBASE[30] |
| = {1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193, 257, 385, 513, |
| 769, 1025, 1537, 2049, 3073, 4097, 6145, 8193, 12289, 16385, 24577}; |
| |
| /*the extra bits of backwards distances (added to base)*/ |
| static const unsigned DISTANCEEXTRA[30] |
| = {0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, |
| 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13}; |
| |
| /*the order in which "code length alphabet code lengths" are stored, out of this |
| the huffman tree of the dynamic huffman tree lengths is generated*/ |
| static const unsigned CLCL_ORDER[NUM_CODE_LENGTH_CODES] |
| = {16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15}; |
| |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| |
| /* |
| Huffman tree struct, containing multiple representations of the tree |
| */ |
| typedef struct HuffmanTree { |
| unsigned* tree2d; |
| unsigned* tree1d; |
| unsigned* lengths; /*the lengths of the codes of the 1d-tree*/ |
| unsigned maxbitlen; /*maximum number of bits a single code can get*/ |
| unsigned numcodes; /*number of symbols in the alphabet = number of codes*/ |
| } HuffmanTree; |
| |
| /*function used for debug purposes to draw the tree in ascii art with C++*/ |
| /* |
| static void HuffmanTree_draw(HuffmanTree* tree) { |
| std::cout << "tree. length: " << tree->numcodes << " maxbitlen: " << tree->maxbitlen << std::endl; |
| for(size_t i = 0; i != tree->tree1d.size; ++i) { |
| if(tree->lengths.data[i]) |
| std::cout << i << " " << tree->tree1d.data[i] << " " << tree->lengths.data[i] << std::endl; |
| } |
| std::cout << std::endl; |
| }*/ |
| |
| static void HuffmanTree_init(HuffmanTree* tree) { |
| tree->tree2d = 0; |
| tree->tree1d = 0; |
| tree->lengths = 0; |
| } |
| |
| static void HuffmanTree_cleanup(HuffmanTree* tree) { |
| lodepng_free(tree->tree2d); |
| lodepng_free(tree->tree1d); |
| lodepng_free(tree->lengths); |
| } |
| |
| /*the tree representation used by the decoder. return value is error*/ |
| static unsigned HuffmanTree_make2DTree(HuffmanTree* tree) { |
| unsigned nodefilled = 0; /*up to which node it is filled*/ |
| unsigned treepos = 0; /*position in the tree (1 of the numcodes columns)*/ |
| unsigned n, i; |
| |
| tree->tree2d = (unsigned*)lodepng_malloc(tree->numcodes * 2 * sizeof(unsigned)); |
| if(!tree->tree2d) return 83; /*alloc fail*/ |
| |
| /* |
| convert tree1d[] to tree2d[][]. In the 2D array, a value of 32767 means |
| uninited, a value >= numcodes is an address to another bit, a value < numcodes |
| is a code. The 2 rows are the 2 possible bit values (0 or 1), there are as |
| many columns as codes - 1. |
| A good huffman tree has N * 2 - 1 nodes, of which N - 1 are internal nodes. |
| Here, the internal nodes are stored (what their 0 and 1 option point to). |
| There is only memory for such good tree currently, if there are more nodes |
| (due to too long length codes), error 55 will happen |
| */ |
| for(n = 0; n < tree->numcodes * 2; ++n) { |
| tree->tree2d[n] = 32767; /*32767 here means the tree2d isn't filled there yet*/ |
| } |
| |
| for(n = 0; n < tree->numcodes; ++n) /*the codes*/ { |
| for(i = 0; i != tree->lengths[n]; ++i) /*the bits for this code*/ { |
| unsigned char bit = (unsigned char)((tree->tree1d[n] >> (tree->lengths[n] - i - 1)) & 1); |
| /*oversubscribed, see comment in lodepng_error_text*/ |
| if(treepos > 2147483647 || treepos + 2 > tree->numcodes) return 55; |
| if(tree->tree2d[2 * treepos + bit] == 32767) /*not yet filled in*/ { |
| if(i + 1 == tree->lengths[n]) /*last bit*/ { |
| tree->tree2d[2 * treepos + bit] = n; /*put the current code in it*/ |
| treepos = 0; |
| } else { |
| /*put address of the next step in here, first that address has to be found of course |
| (it's just nodefilled + 1)...*/ |
| ++nodefilled; |
| /*addresses encoded with numcodes added to it*/ |
| tree->tree2d[2 * treepos + bit] = nodefilled + tree->numcodes; |
| treepos = nodefilled; |
| } |
| } |
| else treepos = tree->tree2d[2 * treepos + bit] - tree->numcodes; |
| } |
| } |
| |
| for(n = 0; n < tree->numcodes * 2; ++n) { |
| if(tree->tree2d[n] == 32767) tree->tree2d[n] = 0; /*remove possible remaining 32767's*/ |
| } |
| |
| return 0; |
| } |
| |
| /* |
| Second step for the ...makeFromLengths and ...makeFromFrequencies functions. |
| numcodes, lengths and maxbitlen must already be filled in correctly. return |
| value is error. |
| */ |
| static unsigned HuffmanTree_makeFromLengths2(HuffmanTree* tree) { |
| uivector blcount; |
| uivector nextcode; |
| unsigned error = 0; |
| unsigned bits, n; |
| |
| uivector_init(&blcount); |
| uivector_init(&nextcode); |
| |
| tree->tree1d = (unsigned*)lodepng_malloc(tree->numcodes * sizeof(unsigned)); |
| if(!tree->tree1d) error = 83; /*alloc fail*/ |
| |
| if(!uivector_resizev(&blcount, tree->maxbitlen + 1, 0) |
| || !uivector_resizev(&nextcode, tree->maxbitlen + 1, 0)) |
| error = 83; /*alloc fail*/ |
| |
| if(!error) { |
| /*step 1: count number of instances of each code length*/ |
| for(bits = 0; bits != tree->numcodes; ++bits) ++blcount.data[tree->lengths[bits]]; |
| /*step 2: generate the nextcode values*/ |
| for(bits = 1; bits <= tree->maxbitlen; ++bits) { |
| nextcode.data[bits] = (nextcode.data[bits - 1] + blcount.data[bits - 1]) << 1; |
| } |
| /*step 3: generate all the codes*/ |
| for(n = 0; n != tree->numcodes; ++n) { |
| if(tree->lengths[n] != 0) tree->tree1d[n] = nextcode.data[tree->lengths[n]]++; |
| } |
| } |
| |
| uivector_cleanup(&blcount); |
| uivector_cleanup(&nextcode); |
| |
| if(!error) return HuffmanTree_make2DTree(tree); |
| else return error; |
| } |
| |
| /* |
| given the code lengths (as stored in the PNG file), generate the tree as defined |
| by Deflate. maxbitlen is the maximum bits that a code in the tree can have. |
| return value is error. |
| */ |
| static unsigned HuffmanTree_makeFromLengths(HuffmanTree* tree, const unsigned* bitlen, |
| size_t numcodes, unsigned maxbitlen) { |
| unsigned i; |
| tree->lengths = (unsigned*)lodepng_malloc(numcodes * sizeof(unsigned)); |
| if(!tree->lengths) return 83; /*alloc fail*/ |
| for(i = 0; i != numcodes; ++i) tree->lengths[i] = bitlen[i]; |
| tree->numcodes = (unsigned)numcodes; /*number of symbols*/ |
| tree->maxbitlen = maxbitlen; |
| return HuffmanTree_makeFromLengths2(tree); |
| } |
| |
| #ifdef LODEPNG_COMPILE_ENCODER |
| |
| /*BPM: Boundary Package Merge, see "A Fast and Space-Economical Algorithm for Length-Limited Coding", |
| Jyrki Katajainen, Alistair Moffat, Andrew Turpin, 1995.*/ |
| |
| /*chain node for boundary package merge*/ |
| typedef struct BPMNode { |
| int weight; /*the sum of all weights in this chain*/ |
| unsigned index; /*index of this leaf node (called "count" in the paper)*/ |
| struct BPMNode* tail; /*the next nodes in this chain (null if last)*/ |
| int in_use; |
| } BPMNode; |
| |
| /*lists of chains*/ |
| typedef struct BPMLists { |
| /*memory pool*/ |
| unsigned memsize; |
| BPMNode* memory; |
| unsigned numfree; |
| unsigned nextfree; |
| BPMNode** freelist; |
| /*two heads of lookahead chains per list*/ |
| unsigned listsize; |
| BPMNode** chains0; |
| BPMNode** chains1; |
| } BPMLists; |
| |
| /*creates a new chain node with the given parameters, from the memory in the lists */ |
| static BPMNode* bpmnode_create(BPMLists* lists, int weight, unsigned index, BPMNode* tail) { |
| unsigned i; |
| BPMNode* result; |
| |
| /*memory full, so garbage collect*/ |
| if(lists->nextfree >= lists->numfree) { |
| /*mark only those that are in use*/ |
| for(i = 0; i != lists->memsize; ++i) lists->memory[i].in_use = 0; |
| for(i = 0; i != lists->listsize; ++i) { |
| BPMNode* node; |
| for(node = lists->chains0[i]; node != 0; node = node->tail) node->in_use = 1; |
| for(node = lists->chains1[i]; node != 0; node = node->tail) node->in_use = 1; |
| } |
| /*collect those that are free*/ |
| lists->numfree = 0; |
| for(i = 0; i != lists->memsize; ++i) { |
| if(!lists->memory[i].in_use) lists->freelist[lists->numfree++] = &lists->memory[i]; |
| } |
| lists->nextfree = 0; |
| } |
| |
| result = lists->freelist[lists->nextfree++]; |
| result->weight = weight; |
| result->index = index; |
| result->tail = tail; |
| return result; |
| } |
| |
| /*sort the leaves with stable mergesort*/ |
| static void bpmnode_sort(BPMNode* leaves, size_t num) { |
| BPMNode* mem = (BPMNode*)lodepng_malloc(sizeof(*leaves) * num); |
| size_t width, counter = 0; |
| for(width = 1; width < num; width *= 2) { |
| BPMNode* a = (counter & 1) ? mem : leaves; |
| BPMNode* b = (counter & 1) ? leaves : mem; |
| size_t p; |
| for(p = 0; p < num; p += 2 * width) { |
| size_t q = (p + width > num) ? num : (p + width); |
| size_t r = (p + 2 * width > num) ? num : (p + 2 * width); |
| size_t i = p, j = q, k; |
| for(k = p; k < r; k++) { |
| if(i < q && (j >= r || a[i].weight <= a[j].weight)) b[k] = a[i++]; |
| else b[k] = a[j++]; |
| } |
| } |
| counter++; |
| } |
| if(counter & 1) memcpy(leaves, mem, sizeof(*leaves) * num); |
| lodepng_free(mem); |
| } |
| |
| /*Boundary Package Merge step, numpresent is the amount of leaves, and c is the current chain.*/ |
| static void boundaryPM(BPMLists* lists, BPMNode* leaves, size_t numpresent, int c, int num) { |
| unsigned lastindex = lists->chains1[c]->index; |
| |
| if(c == 0) { |
| if(lastindex >= numpresent) return; |
| lists->chains0[c] = lists->chains1[c]; |
| lists->chains1[c] = bpmnode_create(lists, leaves[lastindex].weight, lastindex + 1, 0); |
| } else { |
| /*sum of the weights of the head nodes of the previous lookahead chains.*/ |
| int sum = lists->chains0[c - 1]->weight + lists->chains1[c - 1]->weight; |
| lists->chains0[c] = lists->chains1[c]; |
| if(lastindex < numpresent && sum > leaves[lastindex].weight) { |
| lists->chains1[c] = bpmnode_create(lists, leaves[lastindex].weight, lastindex + 1, lists->chains1[c]->tail); |
| return; |
| } |
| lists->chains1[c] = bpmnode_create(lists, sum, lastindex, lists->chains1[c - 1]); |
| /*in the end we are only interested in the chain of the last list, so no |
| need to recurse if we're at the last one (this gives measurable speedup)*/ |
| if(num + 1 < (int)(2 * numpresent - 2)) { |
| boundaryPM(lists, leaves, numpresent, c - 1, num); |
| boundaryPM(lists, leaves, numpresent, c - 1, num); |
| } |
| } |
| } |
| |
| unsigned lodepng_huffman_code_lengths(unsigned* lengths, const unsigned* frequencies, |
| size_t numcodes, unsigned maxbitlen) { |
| unsigned error = 0; |
| unsigned i; |
| size_t numpresent = 0; /*number of symbols with non-zero frequency*/ |
| BPMNode* leaves; /*the symbols, only those with > 0 frequency*/ |
| |
| if(numcodes == 0) return 80; /*error: a tree of 0 symbols is not supposed to be made*/ |
| if((1u << maxbitlen) < (unsigned)numcodes) return 80; /*error: represent all symbols*/ |
| |
| leaves = (BPMNode*)lodepng_malloc(numcodes * sizeof(*leaves)); |
| if(!leaves) return 83; /*alloc fail*/ |
| |
| for(i = 0; i != numcodes; ++i) { |
| if(frequencies[i] > 0) { |
| leaves[numpresent].weight = (int)frequencies[i]; |
| leaves[numpresent].index = i; |
| ++numpresent; |
| } |
| } |
| |
| for(i = 0; i != numcodes; ++i) lengths[i] = 0; |
| |
| /*ensure at least two present symbols. There should be at least one symbol |
| according to RFC 1951 section 3.2.7. Some decoders incorrectly require two. To |
| make these work as well ensure there are at least two symbols. The |
| Package-Merge code below also doesn't work correctly if there's only one |
| symbol, it'd give it the theoritical 0 bits but in practice zlib wants 1 bit*/ |
| if(numpresent == 0) { |
| lengths[0] = lengths[1] = 1; /*note that for RFC 1951 section 3.2.7, only lengths[0] = 1 is needed*/ |
| } else if(numpresent == 1) { |
| lengths[leaves[0].index] = 1; |
| lengths[leaves[0].index == 0 ? 1 : 0] = 1; |
| } else { |
| BPMLists lists; |
| BPMNode* node; |
| |
| bpmnode_sort(leaves, numpresent); |
| |
| lists.listsize = maxbitlen; |
| lists.memsize = 2 * maxbitlen * (maxbitlen + 1); |
| lists.nextfree = 0; |
| lists.numfree = lists.memsize; |
| lists.memory = (BPMNode*)lodepng_malloc(lists.memsize * sizeof(*lists.memory)); |
| lists.freelist = (BPMNode**)lodepng_malloc(lists.memsize * sizeof(BPMNode*)); |
| lists.chains0 = (BPMNode**)lodepng_malloc(lists.listsize * sizeof(BPMNode*)); |
| lists.chains1 = (BPMNode**)lodepng_malloc(lists.listsize * sizeof(BPMNode*)); |
| if(!lists.memory || !lists.freelist || !lists.chains0 || !lists.chains1) error = 83; /*alloc fail*/ |
| |
| if(!error) { |
| for(i = 0; i != lists.memsize; ++i) lists.freelist[i] = &lists.memory[i]; |
| |
| bpmnode_create(&lists, leaves[0].weight, 1, 0); |
| bpmnode_create(&lists, leaves[1].weight, 2, 0); |
| |
| for(i = 0; i != lists.listsize; ++i) { |
| lists.chains0[i] = &lists.memory[0]; |
| lists.chains1[i] = &lists.memory[1]; |
| } |
| |
| /*each boundaryPM call adds one chain to the last list, and we need 2 * numpresent - 2 chains.*/ |
| for(i = 2; i != 2 * numpresent - 2; ++i) boundaryPM(&lists, leaves, numpresent, (int)maxbitlen - 1, (int)i); |
| |
| for(node = lists.chains1[maxbitlen - 1]; node; node = node->tail) { |
| for(i = 0; i != node->index; ++i) ++lengths[leaves[i].index]; |
| } |
| } |
| |
| lodepng_free(lists.memory); |
| lodepng_free(lists.freelist); |
| lodepng_free(lists.chains0); |
| lodepng_free(lists.chains1); |
| } |
| |
| lodepng_free(leaves); |
| return error; |
| } |
| |
| /*Create the Huffman tree given the symbol frequencies*/ |
| static unsigned HuffmanTree_makeFromFrequencies(HuffmanTree* tree, const unsigned* frequencies, |
| size_t mincodes, size_t numcodes, unsigned maxbitlen) { |
| unsigned error = 0; |
| while(!frequencies[numcodes - 1] && numcodes > mincodes) --numcodes; /*trim zeroes*/ |
| tree->maxbitlen = maxbitlen; |
| tree->numcodes = (unsigned)numcodes; /*number of symbols*/ |
| tree->lengths = (unsigned*)lodepng_realloc(tree->lengths, numcodes * sizeof(unsigned)); |
| if(!tree->lengths) return 83; /*alloc fail*/ |
| /*initialize all lengths to 0*/ |
| memset(tree->lengths, 0, numcodes * sizeof(unsigned)); |
| |
| error = lodepng_huffman_code_lengths(tree->lengths, frequencies, numcodes, maxbitlen); |
| if(!error) error = HuffmanTree_makeFromLengths2(tree); |
| return error; |
| } |
| |
| static unsigned HuffmanTree_getCode(const HuffmanTree* tree, unsigned index) { |
| return tree->tree1d[index]; |
| } |
| |
| static unsigned HuffmanTree_getLength(const HuffmanTree* tree, unsigned index) { |
| return tree->lengths[index]; |
| } |
| #endif /*LODEPNG_COMPILE_ENCODER*/ |
| |
| /*get the literal and length code tree of a deflated block with fixed tree, as per the deflate specification*/ |
| static unsigned generateFixedLitLenTree(HuffmanTree* tree) { |
| unsigned i, error = 0; |
| unsigned* bitlen = (unsigned*)lodepng_malloc(NUM_DEFLATE_CODE_SYMBOLS * sizeof(unsigned)); |
| if(!bitlen) return 83; /*alloc fail*/ |
| |
| /*288 possible codes: 0-255=literals, 256=endcode, 257-285=lengthcodes, 286-287=unused*/ |
| for(i = 0; i <= 143; ++i) bitlen[i] = 8; |
| for(i = 144; i <= 255; ++i) bitlen[i] = 9; |
| for(i = 256; i <= 279; ++i) bitlen[i] = 7; |
| for(i = 280; i <= 287; ++i) bitlen[i] = 8; |
| |
| error = HuffmanTree_makeFromLengths(tree, bitlen, NUM_DEFLATE_CODE_SYMBOLS, 15); |
| |
| lodepng_free(bitlen); |
| return error; |
| } |
| |
| /*get the distance code tree of a deflated block with fixed tree, as specified in the deflate specification*/ |
| static unsigned generateFixedDistanceTree(HuffmanTree* tree) { |
| unsigned i, error = 0; |
| unsigned* bitlen = (unsigned*)lodepng_malloc(NUM_DISTANCE_SYMBOLS * sizeof(unsigned)); |
| if(!bitlen) return 83; /*alloc fail*/ |
| |
| /*there are 32 distance codes, but 30-31 are unused*/ |
| for(i = 0; i != NUM_DISTANCE_SYMBOLS; ++i) bitlen[i] = 5; |
| error = HuffmanTree_makeFromLengths(tree, bitlen, NUM_DISTANCE_SYMBOLS, 15); |
| |
| lodepng_free(bitlen); |
| return error; |
| } |
| |
| #ifdef LODEPNG_COMPILE_DECODER |
| |
| /* |
| returns the code, or (unsigned)(-1) if error happened |
| inbitlength is the length of the complete buffer, in bits (so its byte length times 8) |
| */ |
| static unsigned huffmanDecodeSymbol(const unsigned char* in, size_t* bp, |
| const HuffmanTree* codetree, size_t inbitlength) { |
| unsigned treepos = 0, ct; |
| for(;;) { |
| if(*bp >= inbitlength) return (unsigned)(-1); /*error: end of input memory reached without endcode*/ |
| /* |
| decode the symbol from the tree. The "readBitFromStream" code is inlined in |
| the expression below because this is the biggest bottleneck while decoding |
| */ |
| ct = codetree->tree2d[(treepos << 1) + READBIT(*bp, in)]; |
| ++(*bp); |
| if(ct < codetree->numcodes) return ct; /*the symbol is decoded, return it*/ |
| else treepos = ct - codetree->numcodes; /*symbol not yet decoded, instead move tree position*/ |
| |
| if(treepos >= codetree->numcodes) return (unsigned)(-1); /*error: it appeared outside the codetree*/ |
| } |
| } |
| #endif /*LODEPNG_COMPILE_DECODER*/ |
| |
| #ifdef LODEPNG_COMPILE_DECODER |
| |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| /* / Inflator (Decompressor) / */ |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| |
| /*get the tree of a deflated block with fixed tree, as specified in the deflate specification*/ |
| static void getTreeInflateFixed(HuffmanTree* tree_ll, HuffmanTree* tree_d) { |
| /*TODO: check for out of memory errors*/ |
| generateFixedLitLenTree(tree_ll); |
| generateFixedDistanceTree(tree_d); |
| } |
| |
| /*get the tree of a deflated block with dynamic tree, the tree itself is also Huffman compressed with a known tree*/ |
| static unsigned getTreeInflateDynamic(HuffmanTree* tree_ll, HuffmanTree* tree_d, |
| const unsigned char* in, size_t* bp, size_t inlength) { |
| /*make sure that length values that aren't filled in will be 0, or a wrong tree will be generated*/ |
| unsigned error = 0; |
| unsigned n, HLIT, HDIST, HCLEN, i; |
| size_t inbitlength = inlength * 8; |
| |
| /*see comments in deflateDynamic for explanation of the context and these variables, it is analogous*/ |
| unsigned* bitlen_ll = 0; /*lit,len code lengths*/ |
| unsigned* bitlen_d = 0; /*dist code lengths*/ |
| /*code length code lengths ("clcl"), the bit lengths of the huffman tree used to compress bitlen_ll and bitlen_d*/ |
| unsigned* bitlen_cl = 0; |
| HuffmanTree tree_cl; /*the code tree for code length codes (the huffman tree for compressed huffman trees)*/ |
| |
| if((*bp) + 14 > (inlength << 3)) return 49; /*error: the bit pointer is or will go past the memory*/ |
| |
| /*number of literal/length codes + 257. Unlike the spec, the value 257 is added to it here already*/ |
| HLIT = readBitsFromStream(bp, in, 5) + 257; |
| /*number of distance codes. Unlike the spec, the value 1 is added to it here already*/ |
| HDIST = readBitsFromStream(bp, in, 5) + 1; |
| /*number of code length codes. Unlike the spec, the value 4 is added to it here already*/ |
| HCLEN = readBitsFromStream(bp, in, 4) + 4; |
| |
| if((*bp) + HCLEN * 3 > (inlength << 3)) return 50; /*error: the bit pointer is or will go past the memory*/ |
| |
| HuffmanTree_init(&tree_cl); |
| |
| while(!error) { |
| /*read the code length codes out of 3 * (amount of code length codes) bits*/ |
| |
| bitlen_cl = (unsigned*)lodepng_malloc(NUM_CODE_LENGTH_CODES * sizeof(unsigned)); |
| if(!bitlen_cl) ERROR_BREAK(83 /*alloc fail*/); |
| |
| for(i = 0; i != NUM_CODE_LENGTH_CODES; ++i) { |
| if(i < HCLEN) bitlen_cl[CLCL_ORDER[i]] = readBitsFromStream(bp, in, 3); |
| else bitlen_cl[CLCL_ORDER[i]] = 0; /*if not, it must stay 0*/ |
| } |
| |
| error = HuffmanTree_makeFromLengths(&tree_cl, bitlen_cl, NUM_CODE_LENGTH_CODES, 7); |
| if(error) break; |
| |
| /*now we can use this tree to read the lengths for the tree that this function will return*/ |
| bitlen_ll = (unsigned*)lodepng_malloc(NUM_DEFLATE_CODE_SYMBOLS * sizeof(unsigned)); |
| bitlen_d = (unsigned*)lodepng_malloc(NUM_DISTANCE_SYMBOLS * sizeof(unsigned)); |
| if(!bitlen_ll || !bitlen_d) ERROR_BREAK(83 /*alloc fail*/); |
| for(i = 0; i != NUM_DEFLATE_CODE_SYMBOLS; ++i) bitlen_ll[i] = 0; |
| for(i = 0; i != NUM_DISTANCE_SYMBOLS; ++i) bitlen_d[i] = 0; |
| |
| /*i is the current symbol we're reading in the part that contains the code lengths of lit/len and dist codes*/ |
| i = 0; |
| while(i < HLIT + HDIST) { |
| unsigned code = huffmanDecodeSymbol(in, bp, &tree_cl, inbitlength); |
| if(code <= 15) /*a length code*/ { |
| if(i < HLIT) bitlen_ll[i] = code; |
| else bitlen_d[i - HLIT] = code; |
| ++i; |
| } else if(code == 16) /*repeat previous*/ { |
| unsigned replength = 3; /*read in the 2 bits that indicate repeat length (3-6)*/ |
| unsigned value; /*set value to the previous code*/ |
| |
| if(i == 0) ERROR_BREAK(54); /*can't repeat previous if i is 0*/ |
| |
| if((*bp + 2) > inbitlength) ERROR_BREAK(50); /*error, bit pointer jumps past memory*/ |
| replength += readBitsFromStream(bp, in, 2); |
| |
| if(i < HLIT + 1) value = bitlen_ll[i - 1]; |
| else value = bitlen_d[i - HLIT - 1]; |
| /*repeat this value in the next lengths*/ |
| for(n = 0; n < replength; ++n) { |
| if(i >= HLIT + HDIST) ERROR_BREAK(13); /*error: i is larger than the amount of codes*/ |
| if(i < HLIT) bitlen_ll[i] = value; |
| else bitlen_d[i - HLIT] = value; |
| ++i; |
| } |
| } else if(code == 17) /*repeat "0" 3-10 times*/ { |
| unsigned replength = 3; /*read in the bits that indicate repeat length*/ |
| if((*bp + 3) > inbitlength) ERROR_BREAK(50); /*error, bit pointer jumps past memory*/ |
| replength += readBitsFromStream(bp, in, 3); |
| |
| /*repeat this value in the next lengths*/ |
| for(n = 0; n < replength; ++n) { |
| if(i >= HLIT + HDIST) ERROR_BREAK(14); /*error: i is larger than the amount of codes*/ |
| |
| if(i < HLIT) bitlen_ll[i] = 0; |
| else bitlen_d[i - HLIT] = 0; |
| ++i; |
| } |
| } else if(code == 18) /*repeat "0" 11-138 times*/ { |
| unsigned replength = 11; /*read in the bits that indicate repeat length*/ |
| if((*bp + 7) > inbitlength) ERROR_BREAK(50); /*error, bit pointer jumps past memory*/ |
| replength += readBitsFromStream(bp, in, 7); |
| |
| /*repeat this value in the next lengths*/ |
| for(n = 0; n < replength; ++n) { |
| if(i >= HLIT + HDIST) ERROR_BREAK(15); /*error: i is larger than the amount of codes*/ |
| |
| if(i < HLIT) bitlen_ll[i] = 0; |
| else bitlen_d[i - HLIT] = 0; |
| ++i; |
| } |
| } else /*if(code == (unsigned)(-1))*/ /*huffmanDecodeSymbol returns (unsigned)(-1) in case of error*/ { |
| if(code == (unsigned)(-1)) { |
| /*return error code 10 or 11 depending on the situation that happened in huffmanDecodeSymbol |
| (10=no endcode, 11=wrong jump outside of tree)*/ |
| error = (*bp) > inbitlength ? 10 : 11; |
| } |
| else error = 16; /*unexisting code, this can never happen*/ |
| break; |
| } |
| } |
| if(error) break; |
| |
| if(bitlen_ll[256] == 0) ERROR_BREAK(64); /*the length of the end code 256 must be larger than 0*/ |
| |
| /*now we've finally got HLIT and HDIST, so generate the code trees, and the function is done*/ |
| error = HuffmanTree_makeFromLengths(tree_ll, bitlen_ll, NUM_DEFLATE_CODE_SYMBOLS, 15); |
| if(error) break; |
| error = HuffmanTree_makeFromLengths(tree_d, bitlen_d, NUM_DISTANCE_SYMBOLS, 15); |
| |
| break; /*end of error-while*/ |
| } |
| |
| lodepng_free(bitlen_cl); |
| lodepng_free(bitlen_ll); |
| lodepng_free(bitlen_d); |
| HuffmanTree_cleanup(&tree_cl); |
| |
| return error; |
| } |
| |
| /*inflate a block with dynamic of fixed Huffman tree*/ |
| static unsigned inflateHuffmanBlock(ucvector* out, const unsigned char* in, size_t* bp, |
| size_t* pos, size_t inlength, unsigned btype) { |
| unsigned error = 0; |
| HuffmanTree tree_ll; /*the huffman tree for literal and length codes*/ |
| HuffmanTree tree_d; /*the huffman tree for distance codes*/ |
| size_t inbitlength = inlength * 8; |
| |
| HuffmanTree_init(&tree_ll); |
| HuffmanTree_init(&tree_d); |
| |
| if(btype == 1) getTreeInflateFixed(&tree_ll, &tree_d); |
| else if(btype == 2) error = getTreeInflateDynamic(&tree_ll, &tree_d, in, bp, inlength); |
| |
| while(!error) /*decode all symbols until end reached, breaks at end code*/ { |
| /*code_ll is literal, length or end code*/ |
| unsigned code_ll = huffmanDecodeSymbol(in, bp, &tree_ll, inbitlength); |
| if(code_ll <= 255) /*literal symbol*/ { |
| /*ucvector_push_back would do the same, but for some reason the two lines below run 10% faster*/ |
| if(!ucvector_resize(out, (*pos) + 1)) ERROR_BREAK(83 /*alloc fail*/); |
| out->data[*pos] = (unsigned char)code_ll; |
| ++(*pos); |
| } else if(code_ll >= FIRST_LENGTH_CODE_INDEX && code_ll <= LAST_LENGTH_CODE_INDEX) /*length code*/ { |
| unsigned code_d, distance; |
| unsigned numextrabits_l, numextrabits_d; /*extra bits for length and distance*/ |
| size_t start, forward, backward, length; |
| |
| /*part 1: get length base*/ |
| length = LENGTHBASE[code_ll - FIRST_LENGTH_CODE_INDEX]; |
| |
| /*part 2: get extra bits and add the value of that to length*/ |
| numextrabits_l = LENGTHEXTRA[code_ll - FIRST_LENGTH_CODE_INDEX]; |
| if((*bp + numextrabits_l) > inbitlength) ERROR_BREAK(51); /*error, bit pointer will jump past memory*/ |
| length += readBitsFromStream(bp, in, numextrabits_l); |
| |
| /*part 3: get distance code*/ |
| code_d = huffmanDecodeSymbol(in, bp, &tree_d, inbitlength); |
| if(code_d > 29) { |
| if(code_d == (unsigned)(-1)) /*huffmanDecodeSymbol returns (unsigned)(-1) in case of error*/ { |
| /*return error code 10 or 11 depending on the situation that happened in huffmanDecodeSymbol |
| (10=no endcode, 11=wrong jump outside of tree)*/ |
| error = (*bp) > inlength * 8 ? 10 : 11; |
| } |
| else error = 18; /*error: invalid distance code (30-31 are never used)*/ |
| break; |
| } |
| distance = DISTANCEBASE[code_d]; |
| |
| /*part 4: get extra bits from distance*/ |
| numextrabits_d = DISTANCEEXTRA[code_d]; |
| if((*bp + numextrabits_d) > inbitlength) ERROR_BREAK(51); /*error, bit pointer will jump past memory*/ |
| distance += readBitsFromStream(bp, in, numextrabits_d); |
| |
| /*part 5: fill in all the out[n] values based on the length and dist*/ |
| start = (*pos); |
| if(distance > start) ERROR_BREAK(52); /*too long backward distance*/ |
| backward = start - distance; |
| |
| if(!ucvector_resize(out, (*pos) + length)) ERROR_BREAK(83 /*alloc fail*/); |
| if (distance < length) { |
| for(forward = 0; forward < length; ++forward) { |
| out->data[(*pos)++] = out->data[backward++]; |
| } |
| } else { |
| memcpy(out->data + *pos, out->data + backward, length); |
| *pos += length; |
| } |
| } else if(code_ll == 256) { |
| break; /*end code, break the loop*/ |
| } else /*if(code == (unsigned)(-1))*/ /*huffmanDecodeSymbol returns (unsigned)(-1) in case of error*/ { |
| /*return error code 10 or 11 depending on the situation that happened in huffmanDecodeSymbol |
| (10=no endcode, 11=wrong jump outside of tree)*/ |
| error = ((*bp) > inlength * 8) ? 10 : 11; |
| break; |
| } |
| } |
| |
| HuffmanTree_cleanup(&tree_ll); |
| HuffmanTree_cleanup(&tree_d); |
| |
| return error; |
| } |
| |
| static unsigned inflateNoCompression(ucvector* out, const unsigned char* in, size_t* bp, size_t* pos, size_t inlength) { |
| size_t p; |
| unsigned LEN, NLEN, n, error = 0; |
| |
| /*go to first boundary of byte*/ |
| while(((*bp) & 0x7) != 0) ++(*bp); |
| p = (*bp) / 8; /*byte position*/ |
| |
| /*read LEN (2 bytes) and NLEN (2 bytes)*/ |
| if(p + 4 >= inlength) return 52; /*error, bit pointer will jump past memory*/ |
| LEN = in[p] + 256u * in[p + 1]; p += 2; |
| NLEN = in[p] + 256u * in[p + 1]; p += 2; |
| |
| /*check if 16-bit NLEN is really the one's complement of LEN*/ |
| if(LEN + NLEN != 65535) return 21; /*error: NLEN is not one's complement of LEN*/ |
| |
| if(!ucvector_resize(out, (*pos) + LEN)) return 83; /*alloc fail*/ |
| |
| /*read the literal data: LEN bytes are now stored in the out buffer*/ |
| if(p + LEN > inlength) return 23; /*error: reading outside of in buffer*/ |
| for(n = 0; n < LEN; ++n) out->data[(*pos)++] = in[p++]; |
| |
| (*bp) = p * 8; |
| |
| return error; |
| } |
| |
| static unsigned lodepng_inflatev(ucvector* out, |
| const unsigned char* in, size_t insize, |
| const LodePNGDecompressSettings* settings) { |
| /*bit pointer in the "in" data, current byte is bp >> 3, current bit is bp & 0x7 (from lsb to msb of the byte)*/ |
| size_t bp = 0; |
| unsigned BFINAL = 0; |
| size_t pos = 0; /*byte position in the out buffer*/ |
| unsigned error = 0; |
| |
| (void)settings; |
| |
| while(!BFINAL) { |
| unsigned BTYPE; |
| if(bp + 2 >= insize * 8) return 52; /*error, bit pointer will jump past memory*/ |
| BFINAL = readBitFromStream(&bp, in); |
| BTYPE = 1u * readBitFromStream(&bp, in); |
| BTYPE += 2u * readBitFromStream(&bp, in); |
| |
| if(BTYPE == 3) return 20; /*error: invalid BTYPE*/ |
| else if(BTYPE == 0) error = inflateNoCompression(out, in, &bp, &pos, insize); /*no compression*/ |
| else error = inflateHuffmanBlock(out, in, &bp, &pos, insize, BTYPE); /*compression, BTYPE 01 or 10*/ |
| |
| if(error) return error; |
| } |
| |
| return error; |
| } |
| |
| unsigned lodepng_inflate(unsigned char** out, size_t* outsize, |
| const unsigned char* in, size_t insize, |
| const LodePNGDecompressSettings* settings) { |
| unsigned error; |
| ucvector v; |
| ucvector_init_buffer(&v, *out, *outsize); |
| error = lodepng_inflatev(&v, in, insize, settings); |
| *out = v.data; |
| *outsize = v.size; |
| return error; |
| } |
| |
| static unsigned inflate(unsigned char** out, size_t* outsize, |
| const unsigned char* in, size_t insize, |
| const LodePNGDecompressSettings* settings) { |
| if(settings->custom_inflate) { |
| return settings->custom_inflate(out, outsize, in, insize, settings); |
| } else { |
| return lodepng_inflate(out, outsize, in, insize, settings); |
| } |
| } |
| |
| #endif /*LODEPNG_COMPILE_DECODER*/ |
| |
| #ifdef LODEPNG_COMPILE_ENCODER |
| |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| /* / Deflator (Compressor) / */ |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| |
| static const size_t MAX_SUPPORTED_DEFLATE_LENGTH = 258; |
| |
| /*bitlen is the size in bits of the code*/ |
| static void addHuffmanSymbol(size_t* bp, ucvector* compressed, unsigned code, unsigned bitlen) { |
| addBitsToStreamReversed(bp, compressed, code, bitlen); |
| } |
| |
| /*search the index in the array, that has the largest value smaller than or equal to the given value, |
| given array must be sorted (if no value is smaller, it returns the size of the given array)*/ |
| static size_t searchCodeIndex(const unsigned* array, size_t array_size, size_t value) { |
| /*binary search (only small gain over linear). TODO: use CPU log2 instruction for getting symbols instead*/ |
| size_t left = 1; |
| size_t right = array_size - 1; |
| |
| while(left <= right) { |
| size_t mid = (left + right) >> 1; |
| if (array[mid] >= value) right = mid - 1; |
| else left = mid + 1; |
| } |
| if(left >= array_size || array[left] > value) left--; |
| return left; |
| } |
| |
| static void addLengthDistance(uivector* values, size_t length, size_t distance) { |
| /*values in encoded vector are those used by deflate: |
| 0-255: literal bytes |
| 256: end |
| 257-285: length/distance pair (length code, followed by extra length bits, distance code, extra distance bits) |
| 286-287: invalid*/ |
| |
| unsigned length_code = (unsigned)searchCodeIndex(LENGTHBASE, 29, length); |
| unsigned extra_length = (unsigned)(length - LENGTHBASE[length_code]); |
| unsigned dist_code = (unsigned)searchCodeIndex(DISTANCEBASE, 30, distance); |
| unsigned extra_distance = (unsigned)(distance - DISTANCEBASE[dist_code]); |
| |
| uivector_push_back(values, length_code + FIRST_LENGTH_CODE_INDEX); |
| uivector_push_back(values, extra_length); |
| uivector_push_back(values, dist_code); |
| uivector_push_back(values, extra_distance); |
| } |
| |
| /*3 bytes of data get encoded into two bytes. The hash cannot use more than 3 |
| bytes as input because 3 is the minimum match length for deflate*/ |
| static const unsigned HASH_NUM_VALUES = 65536; |
| static const unsigned HASH_BIT_MASK = 65535; /*HASH_NUM_VALUES - 1, but C90 does not like that as initializer*/ |
| |
| typedef struct Hash { |
| int* head; /*hash value to head circular pos - can be outdated if went around window*/ |
| /*circular pos to prev circular pos*/ |
| unsigned short* chain; |
| int* val; /*circular pos to hash value*/ |
| |
| /*TODO: do this not only for zeros but for any repeated byte. However for PNG |
| it's always going to be the zeros that dominate, so not important for PNG*/ |
| int* headz; /*similar to head, but for chainz*/ |
| unsigned short* chainz; /*those with same amount of zeros*/ |
| unsigned short* zeros; /*length of zeros streak, used as a second hash chain*/ |
| } Hash; |
| |
| static unsigned hash_init(Hash* hash, unsigned windowsize) { |
| unsigned i; |
| hash->head = (int*)lodepng_malloc(sizeof(int) * HASH_NUM_VALUES); |
| hash->val = (int*)lodepng_malloc(sizeof(int) * windowsize); |
| hash->chain = (unsigned short*)lodepng_malloc(sizeof(unsigned short) * windowsize); |
| |
| hash->zeros = (unsigned short*)lodepng_malloc(sizeof(unsigned short) * windowsize); |
| hash->headz = (int*)lodepng_malloc(sizeof(int) * (MAX_SUPPORTED_DEFLATE_LENGTH + 1)); |
| hash->chainz = (unsigned short*)lodepng_malloc(sizeof(unsigned short) * windowsize); |
| |
| if(!hash->head || !hash->chain || !hash->val || !hash->headz|| !hash->chainz || !hash->zeros) { |
| return 83; /*alloc fail*/ |
| } |
| |
| /*initialize hash table*/ |
| for(i = 0; i != HASH_NUM_VALUES; ++i) hash->head[i] = -1; |
| for(i = 0; i != windowsize; ++i) hash->val[i] = -1; |
| for(i = 0; i != windowsize; ++i) hash->chain[i] = i; /*same value as index indicates uninitialized*/ |
| |
| for(i = 0; i <= MAX_SUPPORTED_DEFLATE_LENGTH; ++i) hash->headz[i] = -1; |
| for(i = 0; i != windowsize; ++i) hash->chainz[i] = i; /*same value as index indicates uninitialized*/ |
| |
| return 0; |
| } |
| |
| static void hash_cleanup(Hash* hash) { |
| lodepng_free(hash->head); |
| lodepng_free(hash->val); |
| lodepng_free(hash->chain); |
| |
| lodepng_free(hash->zeros); |
| lodepng_free(hash->headz); |
| lodepng_free(hash->chainz); |
| } |
| |
| |
| |
| static unsigned getHash(const unsigned char* data, size_t size, size_t pos) { |
| unsigned result = 0; |
| if(pos + 2 < size) { |
| /*A simple shift and xor hash is used. Since the data of PNGs is dominated |
| by zeroes due to the filters, a better hash does not have a significant |
| effect on speed in traversing the chain, and causes more time spend on |
| calculating the hash.*/ |
| result ^= (unsigned)(data[pos + 0] << 0u); |
| result ^= (unsigned)(data[pos + 1] << 4u); |
| result ^= (unsigned)(data[pos + 2] << 8u); |
| } else { |
| size_t amount, i; |
| if(pos >= size) return 0; |
| amount = size - pos; |
| for(i = 0; i != amount; ++i) result ^= (unsigned)(data[pos + i] << (i * 8u)); |
| } |
| return result & HASH_BIT_MASK; |
| } |
| |
| static unsigned countZeros(const unsigned char* data, size_t size, size_t pos) { |
| const unsigned char* start = data + pos; |
| const unsigned char* end = start + MAX_SUPPORTED_DEFLATE_LENGTH; |
| if(end > data + size) end = data + size; |
| data = start; |
| while(data != end && *data == 0) ++data; |
| /*subtracting two addresses returned as 32-bit number (max value is MAX_SUPPORTED_DEFLATE_LENGTH)*/ |
| return (unsigned)(data - start); |
| } |
| |
| /*wpos = pos & (windowsize - 1)*/ |
| static void updateHashChain(Hash* hash, size_t wpos, unsigned hashval, unsigned short numzeros) { |
| hash->val[wpos] = (int)hashval; |
| if(hash->head[hashval] != -1) hash->chain[wpos] = hash->head[hashval]; |
| hash->head[hashval] = (int)wpos; |
| |
| hash->zeros[wpos] = numzeros; |
| if(hash->headz[numzeros] != -1) hash->chainz[wpos] = hash->headz[numzeros]; |
| hash->headz[numzeros] = (int)wpos; |
| } |
| |
| /* |
| LZ77-encode the data. Return value is error code. The input are raw bytes, the output |
| is in the form of unsigned integers with codes representing for example literal bytes, or |
| length/distance pairs. |
| It uses a hash table technique to let it encode faster. When doing LZ77 encoding, a |
| sliding window (of windowsize) is used, and all past bytes in that window can be used as |
| the "dictionary". A brute force search through all possible distances would be slow, and |
| this hash technique is one out of several ways to speed this up. |
| */ |
| static unsigned encodeLZ77(uivector* out, Hash* hash, |
| const unsigned char* in, size_t inpos, size_t insize, unsigned windowsize, |
| unsigned minmatch, unsigned nicematch, unsigned lazymatching) { |
| size_t pos; |
| unsigned i, error = 0; |
| /*for large window lengths, assume the user wants no compression loss. Otherwise, max hash chain length speedup.*/ |
| unsigned maxchainlength = windowsize >= 8192 ? windowsize : windowsize / 8; |
| unsigned maxlazymatch = windowsize >= 8192 ? MAX_SUPPORTED_DEFLATE_LENGTH : 64; |
| |
| unsigned usezeros = 1; /*not sure if setting it to false for windowsize < 8192 is better or worse*/ |
| unsigned numzeros = 0; |
| |
| unsigned offset; /*the offset represents the distance in LZ77 terminology*/ |
| unsigned length; |
| unsigned lazy = 0; |
| unsigned lazylength = 0, lazyoffset = 0; |
| unsigned hashval; |
| unsigned current_offset, current_length; |
| unsigned prev_offset; |
| const unsigned char *lastptr, *foreptr, *backptr; |
| unsigned hashpos; |
| |
| if(windowsize == 0 || windowsize > 32768) return 60; /*error: windowsize smaller/larger than allowed*/ |
| if((windowsize & (windowsize - 1)) != 0) return 90; /*error: must be power of two*/ |
| |
| if(nicematch > MAX_SUPPORTED_DEFLATE_LENGTH) nicematch = MAX_SUPPORTED_DEFLATE_LENGTH; |
| |
| for(pos = inpos; pos < insize; ++pos) { |
| size_t wpos = pos & (windowsize - 1); /*position for in 'circular' hash buffers*/ |
| unsigned chainlength = 0; |
| |
| hashval = getHash(in, insize, pos); |
| |
| if(usezeros && hashval == 0) { |
| if(numzeros == 0) numzeros = countZeros(in, insize, pos); |
| else if(pos + numzeros > insize || in[pos + numzeros - 1] != 0) --numzeros; |
| } else { |
| numzeros = 0; |
| } |
| |
| updateHashChain(hash, wpos, hashval, numzeros); |
| |
| /*the length and offset found for the current position*/ |
| length = 0; |
| offset = 0; |
| |
| hashpos = hash->chain[wpos]; |
| |
| lastptr = &in[insize < pos + MAX_SUPPORTED_DEFLATE_LENGTH ? insize : pos + MAX_SUPPORTED_DEFLATE_LENGTH]; |
| |
| /*search for the longest string*/ |
| prev_offset = 0; |
| for(;;) { |
| if(chainlength++ >= maxchainlength) break; |
| current_offset = (unsigned)(hashpos <= wpos ? wpos - hashpos : wpos - hashpos + windowsize); |
| |
| if(current_offset < prev_offset) break; /*stop when went completely around the circular buffer*/ |
| prev_offset = current_offset; |
| if(current_offset > 0) { |
| /*test the next characters*/ |
| foreptr = &in[pos]; |
| backptr = &in[pos - current_offset]; |
| |
| /*common case in PNGs is lots of zeros. Quickly skip over them as a speedup*/ |
| if(numzeros >= 3) { |
| unsigned skip = hash->zeros[hashpos]; |
| if(skip > numzeros) skip = numzeros; |
| backptr += skip; |
| foreptr += skip; |
| } |
| |
| while(foreptr != lastptr && *backptr == *foreptr) /*maximum supported length by deflate is max length*/ { |
| ++backptr; |
| ++foreptr; |
| } |
| current_length = (unsigned)(foreptr - &in[pos]); |
| |
| if(current_length > length) { |
| length = current_length; /*the longest length*/ |
| offset = current_offset; /*the offset that is related to this longest length*/ |
| /*jump out once a length of max length is found (speed gain). This also jumps |
| out if length is MAX_SUPPORTED_DEFLATE_LENGTH*/ |
| if(current_length >= nicematch) break; |
| } |
| } |
| |
| if(hashpos == hash->chain[hashpos]) break; |
| |
| if(numzeros >= 3 && length > numzeros) { |
| hashpos = hash->chainz[hashpos]; |
| if(hash->zeros[hashpos] != numzeros) break; |
| } else { |
| hashpos = hash->chain[hashpos]; |
| /*outdated hash value, happens if particular value was not encountered in whole last window*/ |
| if(hash->val[hashpos] != (int)hashval) break; |
| } |
| } |
| |
| if(lazymatching) { |
| if(!lazy && length >= 3 && length <= maxlazymatch && length < MAX_SUPPORTED_DEFLATE_LENGTH) { |
| lazy = 1; |
| lazylength = length; |
| lazyoffset = offset; |
| continue; /*try the next byte*/ |
| } |
| if(lazy) { |
| lazy = 0; |
| if(pos == 0) ERROR_BREAK(81); |
| if(length > lazylength + 1) { |
| /*push the previous character as literal*/ |
| if(!uivector_push_back(out, in[pos - 1])) ERROR_BREAK(83 /*alloc fail*/); |
| } else { |
| length = lazylength; |
| offset = lazyoffset; |
| hash->head[hashval] = -1; /*the same hashchain update will be done, this ensures no wrong alteration*/ |
| hash->headz[numzeros] = -1; /*idem*/ |
| --pos; |
| } |
| } |
| } |
| if(length >= 3 && offset > windowsize) ERROR_BREAK(86 /*too big (or overflown negative) offset*/); |
| |
| /*encode it as length/distance pair or literal value*/ |
| if(length < 3) /*only lengths of 3 or higher are supported as length/distance pair*/ { |
| if(!uivector_push_back(out, in[pos])) ERROR_BREAK(83 /*alloc fail*/); |
| } else if(length < minmatch || (length == 3 && offset > 4096)) { |
| /*compensate for the fact that longer offsets have more extra bits, a |
| length of only 3 may be not worth it then*/ |
| if(!uivector_push_back(out, in[pos])) ERROR_BREAK(83 /*alloc fail*/); |
| } else { |
| addLengthDistance(out, length, offset); |
| for(i = 1; i < length; ++i) { |
| ++pos; |
| wpos = pos & (windowsize - 1); |
| hashval = getHash(in, insize, pos); |
| if(usezeros && hashval == 0) { |
| if(numzeros == 0) numzeros = countZeros(in, insize, pos); |
| else if(pos + numzeros > insize || in[pos + numzeros - 1] != 0) --numzeros; |
| } else { |
| numzeros = 0; |
| } |
| updateHashChain(hash, wpos, hashval, numzeros); |
| } |
| } |
| } /*end of the loop through each character of input*/ |
| |
| return error; |
| } |
| |
| /* /////////////////////////////////////////////////////////////////////////// */ |
| |
| static unsigned deflateNoCompression(ucvector* out, const unsigned char* data, size_t datasize) { |
| /*non compressed deflate block data: 1 bit BFINAL,2 bits BTYPE,(5 bits): it jumps to start of next byte, |
| 2 bytes LEN, 2 bytes NLEN, LEN bytes literal DATA*/ |
| |
| size_t i, j, numdeflateblocks = (datasize + 65534) / 65535; |
| unsigned datapos = 0; |
| for(i = 0; i != numdeflateblocks; ++i) { |
| unsigned BFINAL, BTYPE, LEN, NLEN; |
| unsigned char firstbyte; |
| |
| BFINAL = (i == numdeflateblocks - 1); |
| BTYPE = 0; |
| |
| firstbyte = (unsigned char)(BFINAL + ((BTYPE & 1) << 1) + ((BTYPE & 2) << 1)); |
| ucvector_push_back(out, firstbyte); |
| |
| LEN = 65535; |
| if(datasize - datapos < 65535) LEN = (unsigned)datasize - datapos; |
| NLEN = 65535 - LEN; |
| |
| ucvector_push_back(out, (unsigned char)(LEN & 255)); |
| ucvector_push_back(out, (unsigned char)(LEN >> 8)); |
| ucvector_push_back(out, (unsigned char)(NLEN & 255)); |
| ucvector_push_back(out, (unsigned char)(NLEN >> 8)); |
| |
| /*Decompressed data*/ |
| for(j = 0; j < 65535 && datapos < datasize; ++j) { |
| ucvector_push_back(out, data[datapos++]); |
| } |
| } |
| |
| return 0; |
| } |
| |
| /* |
| write the lz77-encoded data, which has lit, len and dist codes, to compressed stream using huffman trees. |
| tree_ll: the tree for lit and len codes. |
| tree_d: the tree for distance codes. |
| */ |
| static void writeLZ77data(size_t* bp, ucvector* out, const uivector* lz77_encoded, |
| const HuffmanTree* tree_ll, const HuffmanTree* tree_d) { |
| size_t i = 0; |
| for(i = 0; i != lz77_encoded->size; ++i) { |
| unsigned val = lz77_encoded->data[i]; |
| addHuffmanSymbol(bp, out, HuffmanTree_getCode(tree_ll, val), HuffmanTree_getLength(tree_ll, val)); |
| if(val > 256) /*for a length code, 3 more things have to be added*/ { |
| unsigned length_index = val - FIRST_LENGTH_CODE_INDEX; |
| unsigned n_length_extra_bits = LENGTHEXTRA[length_index]; |
| unsigned length_extra_bits = lz77_encoded->data[++i]; |
| |
| unsigned distance_code = lz77_encoded->data[++i]; |
| |
| unsigned distance_index = distance_code; |
| unsigned n_distance_extra_bits = DISTANCEEXTRA[distance_index]; |
| unsigned distance_extra_bits = lz77_encoded->data[++i]; |
| |
| addBitsToStream(bp, out, length_extra_bits, n_length_extra_bits); |
| addHuffmanSymbol(bp, out, HuffmanTree_getCode(tree_d, distance_code), |
| HuffmanTree_getLength(tree_d, distance_code)); |
| addBitsToStream(bp, out, distance_extra_bits, n_distance_extra_bits); |
| } |
| } |
| } |
| |
| /*Deflate for a block of type "dynamic", that is, with freely, optimally, created huffman trees*/ |
| static unsigned deflateDynamic(ucvector* out, size_t* bp, Hash* hash, |
| const unsigned char* data, size_t datapos, size_t dataend, |
| const LodePNGCompressSettings* settings, unsigned final) { |
| unsigned error = 0; |
| |
| /* |
| A block is compressed as follows: The PNG data is lz77 encoded, resulting in |
| literal bytes and length/distance pairs. This is then huffman compressed with |
| two huffman trees. One huffman tree is used for the lit and len values ("ll"), |
| another huffman tree is used for the dist values ("d"). These two trees are |
| stored using their code lengths, and to compress even more these code lengths |
| are also run-length encoded and huffman compressed. This gives a huffman tree |
| of code lengths "cl". The code lenghts used to describe this third tree are |
| the code length code lengths ("clcl"). |
| */ |
| |
| /*The lz77 encoded data, represented with integers since there will also be length and distance codes in it*/ |
| uivector lz77_encoded; |
| HuffmanTree tree_ll; /*tree for lit,len values*/ |
| HuffmanTree tree_d; /*tree for distance codes*/ |
| HuffmanTree tree_cl; /*tree for encoding the code lengths representing tree_ll and tree_d*/ |
| uivector frequencies_ll; /*frequency of lit,len codes*/ |
| uivector frequencies_d; /*frequency of dist codes*/ |
| uivector frequencies_cl; /*frequency of code length codes*/ |
| uivector bitlen_lld; /*lit,len,dist code lenghts (int bits), literally (without repeat codes).*/ |
| uivector bitlen_lld_e; /*bitlen_lld encoded with repeat codes (this is a rudemtary run length compression)*/ |
| /*bitlen_cl is the code length code lengths ("clcl"). The bit lengths of codes to represent tree_cl |
| (these are written as is in the file, it would be crazy to compress these using yet another huffman |
| tree that needs to be represented by yet another set of code lengths)*/ |
| uivector bitlen_cl; |
| size_t datasize = dataend - datapos; |
| |
| /* |
| Due to the huffman compression of huffman tree representations ("two levels"), there are some anologies: |
| bitlen_lld is to tree_cl what data is to tree_ll and tree_d. |
| bitlen_lld_e is to bitlen_lld what lz77_encoded is to data. |
| bitlen_cl is to bitlen_lld_e what bitlen_lld is to lz77_encoded. |
| */ |
| |
| unsigned BFINAL = final; |
| size_t numcodes_ll, numcodes_d, i; |
| unsigned HLIT, HDIST, HCLEN; |
| |
| uivector_init(&lz77_encoded); |
| HuffmanTree_init(&tree_ll); |
| HuffmanTree_init(&tree_d); |
| HuffmanTree_init(&tree_cl); |
| uivector_init(&frequencies_ll); |
| uivector_init(&frequencies_d); |
| uivector_init(&frequencies_cl); |
| uivector_init(&bitlen_lld); |
| uivector_init(&bitlen_lld_e); |
| uivector_init(&bitlen_cl); |
| |
| /*This while loop never loops due to a break at the end, it is here to |
| allow breaking out of it to the cleanup phase on error conditions.*/ |
| while(!error) { |
| if(settings->use_lz77) { |
| error = encodeLZ77(&lz77_encoded, hash, data, datapos, dataend, settings->windowsize, |
| settings->minmatch, settings->nicematch, settings->lazymatching); |
| if(error) break; |
| } else { |
| if(!uivector_resize(&lz77_encoded, datasize)) ERROR_BREAK(83 /*alloc fail*/); |
| for(i = datapos; i < dataend; ++i) lz77_encoded.data[i - datapos] = data[i]; /*no LZ77, but still will be Huffman compressed*/ |
| } |
| |
| if(!uivector_resizev(&frequencies_ll, 286, 0)) ERROR_BREAK(83 /*alloc fail*/); |
| if(!uivector_resizev(&frequencies_d, 30, 0)) ERROR_BREAK(83 /*alloc fail*/); |
| |
| /*Count the frequencies of lit, len and dist codes*/ |
| for(i = 0; i != lz77_encoded.size; ++i) { |
| unsigned symbol = lz77_encoded.data[i]; |
| ++frequencies_ll.data[symbol]; |
| if(symbol > 256) { |
| unsigned dist = lz77_encoded.data[i + 2]; |
| ++frequencies_d.data[dist]; |
| i += 3; |
| } |
| } |
| frequencies_ll.data[256] = 1; /*there will be exactly 1 end code, at the end of the block*/ |
| |
| /*Make both huffman trees, one for the lit and len codes, one for the dist codes*/ |
| error = HuffmanTree_makeFromFrequencies(&tree_ll, frequencies_ll.data, 257, frequencies_ll.size, 15); |
| if(error) break; |
| /*2, not 1, is chosen for mincodes: some buggy PNG decoders require at least 2 symbols in the dist tree*/ |
| error = HuffmanTree_makeFromFrequencies(&tree_d, frequencies_d.data, 2, frequencies_d.size, 15); |
| if(error) break; |
| |
| numcodes_ll = tree_ll.numcodes; if(numcodes_ll > 286) numcodes_ll = 286; |
| numcodes_d = tree_d.numcodes; if(numcodes_d > 30) numcodes_d = 30; |
| /*store the code lengths of both generated trees in bitlen_lld*/ |
| for(i = 0; i != numcodes_ll; ++i) uivector_push_back(&bitlen_lld, HuffmanTree_getLength(&tree_ll, (unsigned)i)); |
| for(i = 0; i != numcodes_d; ++i) uivector_push_back(&bitlen_lld, HuffmanTree_getLength(&tree_d, (unsigned)i)); |
| |
| /*run-length compress bitlen_ldd into bitlen_lld_e by using repeat codes 16 (copy length 3-6 times), |
| 17 (3-10 zeroes), 18 (11-138 zeroes)*/ |
| for(i = 0; i != (unsigned)bitlen_lld.size; ++i) { |
| unsigned j = 0; /*amount of repititions*/ |
| while(i + j + 1 < (unsigned)bitlen_lld.size && bitlen_lld.data[i + j + 1] == bitlen_lld.data[i]) ++j; |
| |
| if(bitlen_lld.data[i] == 0 && j >= 2) /*repeat code for zeroes*/ { |
| ++j; /*include the first zero*/ |
| if(j <= 10) /*repeat code 17 supports max 10 zeroes*/ { |
| uivector_push_back(&bitlen_lld_e, 17); |
| uivector_push_back(&bitlen_lld_e, j - 3); |
| } else /*repeat code 18 supports max 138 zeroes*/ { |
| if(j > 138) j = 138; |
| uivector_push_back(&bitlen_lld_e, 18); |
| uivector_push_back(&bitlen_lld_e, j - 11); |
| } |
| i += (j - 1); |
| } else if(j >= 3) /*repeat code for value other than zero*/ { |
| size_t k; |
| unsigned num = j / 6, rest = j % 6; |
| uivector_push_back(&bitlen_lld_e, bitlen_lld.data[i]); |
| for(k = 0; k < num; ++k) { |
| uivector_push_back(&bitlen_lld_e, 16); |
| uivector_push_back(&bitlen_lld_e, 6 - 3); |
| } |
| if(rest >= 3) { |
| uivector_push_back(&bitlen_lld_e, 16); |
| uivector_push_back(&bitlen_lld_e, rest - 3); |
| } |
| else j -= rest; |
| i += j; |
| } else /*too short to benefit from repeat code*/ { |
| uivector_push_back(&bitlen_lld_e, bitlen_lld.data[i]); |
| } |
| } |
| |
| /*generate tree_cl, the huffmantree of huffmantrees*/ |
| |
| if(!uivector_resizev(&frequencies_cl, NUM_CODE_LENGTH_CODES, 0)) ERROR_BREAK(83 /*alloc fail*/); |
| for(i = 0; i != bitlen_lld_e.size; ++i) { |
| ++frequencies_cl.data[bitlen_lld_e.data[i]]; |
| /*after a repeat code come the bits that specify the number of repetitions, |
| those don't need to be in the frequencies_cl calculation*/ |
| if(bitlen_lld_e.data[i] >= 16) ++i; |
| } |
| |
| error = HuffmanTree_makeFromFrequencies(&tree_cl, frequencies_cl.data, |
| frequencies_cl.size, frequencies_cl.size, 7); |
| if(error) break; |
| |
| if(!uivector_resize(&bitlen_cl, tree_cl.numcodes)) ERROR_BREAK(83 /*alloc fail*/); |
| for(i = 0; i != tree_cl.numcodes; ++i) { |
| /*lenghts of code length tree is in the order as specified by deflate*/ |
| bitlen_cl.data[i] = HuffmanTree_getLength(&tree_cl, CLCL_ORDER[i]); |
| } |
| while(bitlen_cl.data[bitlen_cl.size - 1] == 0 && bitlen_cl.size > 4) { |
| /*remove zeros at the end, but minimum size must be 4*/ |
| if(!uivector_resize(&bitlen_cl, bitlen_cl.size - 1)) ERROR_BREAK(83 /*alloc fail*/); |
| } |
| if(error) break; |
| |
| /* |
| Write everything into the output |
| |
| After the BFINAL and BTYPE, the dynamic block consists out of the following: |
| - 5 bits HLIT, 5 bits HDIST, 4 bits HCLEN |
| - (HCLEN+4)*3 bits code lengths of code length alphabet |
| - HLIT + 257 code lenghts of lit/length alphabet (encoded using the code length |
| alphabet, + possible repetition codes 16, 17, 18) |
| - HDIST + 1 code lengths of distance alphabet (encoded using the code length |
| alphabet, + possible repetition codes 16, 17, 18) |
| - compressed data |
| - 256 (end code) |
| */ |
| |
| /*Write block type*/ |
| addBitToStream(bp, out, BFINAL); |
| addBitToStream(bp, out, 0); /*first bit of BTYPE "dynamic"*/ |
| addBitToStream(bp, out, 1); /*second bit of BTYPE "dynamic"*/ |
| |
| /*write the HLIT, HDIST and HCLEN values*/ |
| HLIT = (unsigned)(numcodes_ll - 257); |
| HDIST = (unsigned)(numcodes_d - 1); |
| HCLEN = (unsigned)bitlen_cl.size - 4; |
| /*trim zeroes for HCLEN. HLIT and HDIST were already trimmed at tree creation*/ |
| while(!bitlen_cl.data[HCLEN + 4 - 1] && HCLEN > 0) --HCLEN; |
| addBitsToStream(bp, out, HLIT, 5); |
| addBitsToStream(bp, out, HDIST, 5); |
| addBitsToStream(bp, out, HCLEN, 4); |
| |
| /*write the code lenghts of the code length alphabet*/ |
| for(i = 0; i != HCLEN + 4; ++i) addBitsToStream(bp, out, bitlen_cl.data[i], 3); |
| |
| /*write the lenghts of the lit/len AND the dist alphabet*/ |
| for(i = 0; i != bitlen_lld_e.size; ++i) { |
| addHuffmanSymbol(bp, out, HuffmanTree_getCode(&tree_cl, bitlen_lld_e.data[i]), |
| HuffmanTree_getLength(&tree_cl, bitlen_lld_e.data[i])); |
| /*extra bits of repeat codes*/ |
| if(bitlen_lld_e.data[i] == 16) addBitsToStream(bp, out, bitlen_lld_e.data[++i], 2); |
| else if(bitlen_lld_e.data[i] == 17) addBitsToStream(bp, out, bitlen_lld_e.data[++i], 3); |
| else if(bitlen_lld_e.data[i] == 18) addBitsToStream(bp, out, bitlen_lld_e.data[++i], 7); |
| } |
| |
| /*write the compressed data symbols*/ |
| writeLZ77data(bp, out, &lz77_encoded, &tree_ll, &tree_d); |
| /*error: the length of the end code 256 must be larger than 0*/ |
| if(HuffmanTree_getLength(&tree_ll, 256) == 0) ERROR_BREAK(64); |
| |
| /*write the end code*/ |
| addHuffmanSymbol(bp, out, HuffmanTree_getCode(&tree_ll, 256), HuffmanTree_getLength(&tree_ll, 256)); |
| |
| break; /*end of error-while*/ |
| } |
| |
| /*cleanup*/ |
| uivector_cleanup(&lz77_encoded); |
| HuffmanTree_cleanup(&tree_ll); |
| HuffmanTree_cleanup(&tree_d); |
| HuffmanTree_cleanup(&tree_cl); |
| uivector_cleanup(&frequencies_ll); |
| uivector_cleanup(&frequencies_d); |
| uivector_cleanup(&frequencies_cl); |
| uivector_cleanup(&bitlen_lld_e); |
| uivector_cleanup(&bitlen_lld); |
| uivector_cleanup(&bitlen_cl); |
| |
| return error; |
| } |
| |
| static unsigned deflateFixed(ucvector* out, size_t* bp, Hash* hash, |
| const unsigned char* data, |
| size_t datapos, size_t dataend, |
| const LodePNGCompressSettings* settings, unsigned final) { |
| HuffmanTree tree_ll; /*tree for literal values and length codes*/ |
| HuffmanTree tree_d; /*tree for distance codes*/ |
| |
| unsigned BFINAL = final; |
| unsigned error = 0; |
| size_t i; |
| |
| HuffmanTree_init(&tree_ll); |
| HuffmanTree_init(&tree_d); |
| |
| generateFixedLitLenTree(&tree_ll); |
| generateFixedDistanceTree(&tree_d); |
| |
| addBitToStream(bp, out, BFINAL); |
| addBitToStream(bp, out, 1); /*first bit of BTYPE*/ |
| addBitToStream(bp, out, 0); /*second bit of BTYPE*/ |
| |
| if(settings->use_lz77) /*LZ77 encoded*/ { |
| uivector lz77_encoded; |
| uivector_init(&lz77_encoded); |
| error = encodeLZ77(&lz77_encoded, hash, data, datapos, dataend, settings->windowsize, |
| settings->minmatch, settings->nicematch, settings->lazymatching); |
| if(!error) writeLZ77data(bp, out, &lz77_encoded, &tree_ll, &tree_d); |
| uivector_cleanup(&lz77_encoded); |
| } else /*no LZ77, but still will be Huffman compressed*/ { |
| for(i = datapos; i < dataend; ++i) { |
| addHuffmanSymbol(bp, out, HuffmanTree_getCode(&tree_ll, data[i]), HuffmanTree_getLength(&tree_ll, data[i])); |
| } |
| } |
| /*add END code*/ |
| if(!error) addHuffmanSymbol(bp, out, HuffmanTree_getCode(&tree_ll, 256), HuffmanTree_getLength(&tree_ll, 256)); |
| |
| /*cleanup*/ |
| HuffmanTree_cleanup(&tree_ll); |
| HuffmanTree_cleanup(&tree_d); |
| |
| return error; |
| } |
| |
| static unsigned lodepng_deflatev(ucvector* out, const unsigned char* in, size_t insize, |
| const LodePNGCompressSettings* settings) { |
| unsigned error = 0; |
| size_t i, blocksize, numdeflateblocks; |
| size_t bp = 0; /*the bit pointer*/ |
| Hash hash; |
| |
| if(settings->btype > 2) return 61; |
| else if(settings->btype == 0) return deflateNoCompression(out, in, insize); |
| else if(settings->btype == 1) blocksize = insize; |
| else /*if(settings->btype == 2)*/ { |
| /*on PNGs, deflate blocks of 65-262k seem to give most dense encoding*/ |
| blocksize = insize / 8 + 8; |
| if(blocksize < 65536) blocksize = 65536; |
| if(blocksize > 262144) blocksize = 262144; |
| } |
| |
| numdeflateblocks = (insize + blocksize - 1) / blocksize; |
| if(numdeflateblocks == 0) numdeflateblocks = 1; |
| |
| error = hash_init(&hash, settings->windowsize); |
| if(error) return error; |
| |
| for(i = 0; i != numdeflateblocks && !error; ++i) { |
| unsigned final = (i == numdeflateblocks - 1); |
| size_t start = i * blocksize; |
| size_t end = start + blocksize; |
| if(end > insize) end = insize; |
| |
| if(settings->btype == 1) error = deflateFixed(out, &bp, &hash, in, start, end, settings, final); |
| else if(settings->btype == 2) error = deflateDynamic(out, &bp, &hash, in, start, end, settings, final); |
| } |
| |
| hash_cleanup(&hash); |
| |
| return error; |
| } |
| |
| unsigned lodepng_deflate(unsigned char** out, size_t* outsize, |
| const unsigned char* in, size_t insize, |
| const LodePNGCompressSettings* settings) { |
| unsigned error; |
| ucvector v; |
| ucvector_init_buffer(&v, *out, *outsize); |
| error = lodepng_deflatev(&v, in, insize, settings); |
| *out = v.data; |
| *outsize = v.size; |
| return error; |
| } |
| |
| static unsigned deflate(unsigned char** out, size_t* outsize, |
| const unsigned char* in, size_t insize, |
| const LodePNGCompressSettings* settings) { |
| if(settings->custom_deflate) { |
| return settings->custom_deflate(out, outsize, in, insize, settings); |
| } else { |
| return lodepng_deflate(out, outsize, in, insize, settings); |
| } |
| } |
| |
| #endif /*LODEPNG_COMPILE_DECODER*/ |
| |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| /* / Adler32 */ |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| |
| static unsigned update_adler32(unsigned adler, const unsigned char* data, unsigned len) { |
| unsigned s1 = adler & 0xffff; |
| unsigned s2 = (adler >> 16) & 0xffff; |
| |
| while(len > 0) { |
| /*at least 5552 sums can be done before the sums overflow, saving a lot of module divisions*/ |
| unsigned amount = len > 5552 ? 5552 : len; |
| len -= amount; |
| while(amount > 0) { |
| s1 += (*data++); |
| s2 += s1; |
| --amount; |
| } |
| s1 %= 65521; |
| s2 %= 65521; |
| } |
| |
| return (s2 << 16) | s1; |
| } |
| |
| /*Return the adler32 of the bytes data[0..len-1]*/ |
| static unsigned adler32(const unsigned char* data, unsigned len) { |
| return update_adler32(1L, data, len); |
| } |
| |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| /* / Zlib / */ |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| |
| #ifdef LODEPNG_COMPILE_DECODER |
| |
| unsigned lodepng_zlib_decompress(unsigned char** out, size_t* outsize, const unsigned char* in, |
| size_t insize, const LodePNGDecompressSettings* settings) { |
| unsigned error = 0; |
| unsigned CM, CINFO, FDICT; |
| |
| if(insize < 2) return 53; /*error, size of zlib data too small*/ |
| /*read information from zlib header*/ |
| if((in[0] * 256 + in[1]) % 31 != 0) { |
| /*error: 256 * in[0] + in[1] must be a multiple of 31, the FCHECK value is supposed to be made that way*/ |
| return 24; |
| } |
| |
| CM = in[0] & 15; |
| CINFO = (in[0] >> 4) & 15; |
| /*FCHECK = in[1] & 31;*/ /*FCHECK is already tested above*/ |
| FDICT = (in[1] >> 5) & 1; |
| /*FLEVEL = (in[1] >> 6) & 3;*/ /*FLEVEL is not used here*/ |
| |
| if(CM != 8 || CINFO > 7) { |
| /*error: only compression method 8: inflate with sliding window of 32k is supported by the PNG spec*/ |
| return 25; |
| } |
| if(FDICT != 0) { |
| /*error: the specification of PNG says about the zlib stream: |
| "The additional flags shall not specify a preset dictionary."*/ |
| return 26; |
| } |
| |
| error = inflate(out, outsize, in + 2, insize - 2, settings); |
| if(error) return error; |
| |
| if(!settings->ignore_adler32) { |
| unsigned ADLER32 = lodepng_read32bitInt(&in[insize - 4]); |
| unsigned checksum = adler32(*out, (unsigned)(*outsize)); |
| if(checksum != ADLER32) return 58; /*error, adler checksum not correct, data must be corrupted*/ |
| } |
| |
| return 0; /*no error*/ |
| } |
| |
| static unsigned zlib_decompress(unsigned char** out, size_t* outsize, const unsigned char* in, |
| size_t insize, const LodePNGDecompressSettings* settings) { |
| if(settings->custom_zlib) { |
| return settings->custom_zlib(out, outsize, in, insize, settings); |
| } else { |
| return lodepng_zlib_decompress(out, outsize, in, insize, settings); |
| } |
| } |
| |
| #endif /*LODEPNG_COMPILE_DECODER*/ |
| |
| #ifdef LODEPNG_COMPILE_ENCODER |
| |
| unsigned lodepng_zlib_compress(unsigned char** out, size_t* outsize, const unsigned char* in, |
| size_t insize, const LodePNGCompressSettings* settings) { |
| /*initially, *out must be NULL and outsize 0, if you just give some random *out |
| that's pointing to a non allocated buffer, this'll crash*/ |
| ucvector outv; |
| size_t i; |
| unsigned error; |
| unsigned char* deflatedata = 0; |
| size_t deflatesize = 0; |
| |
| /*zlib data: 1 byte CMF (CM+CINFO), 1 byte FLG, deflate data, 4 byte ADLER32 checksum of the Decompressed data*/ |
| unsigned CMF = 120; /*0b01111000: CM 8, CINFO 7. With CINFO 7, any window size up to 32768 can be used.*/ |
| unsigned FLEVEL = 0; |
| unsigned FDICT = 0; |
| unsigned CMFFLG = 256 * CMF + FDICT * 32 + FLEVEL * 64; |
| unsigned FCHECK = 31 - CMFFLG % 31; |
| CMFFLG += FCHECK; |
| |
| /*ucvector-controlled version of the output buffer, for dynamic array*/ |
| ucvector_init_buffer(&outv, *out, *outsize); |
| |
| ucvector_push_back(&outv, (unsigned char)(CMFFLG >> 8)); |
| ucvector_push_back(&outv, (unsigned char)(CMFFLG & 255)); |
| |
| error = deflate(&deflatedata, &deflatesize, in, insize, settings); |
| |
| if(!error) { |
| unsigned ADLER32 = adler32(in, (unsigned)insize); |
| for(i = 0; i != deflatesize; ++i) ucvector_push_back(&outv, deflatedata[i]); |
| lodepng_free(deflatedata); |
| lodepng_add32bitInt(&outv, ADLER32); |
| } |
| |
| *out = outv.data; |
| *outsize = outv.size; |
| |
| return error; |
| } |
| |
| /* compress using the default or custom zlib function */ |
| static unsigned zlib_compress(unsigned char** out, size_t* outsize, const unsigned char* in, |
| size_t insize, const LodePNGCompressSettings* settings) { |
| if(settings->custom_zlib) { |
| return settings->custom_zlib(out, outsize, in, insize, settings); |
| } else { |
| return lodepng_zlib_compress(out, outsize, in, insize, settings); |
| } |
| } |
| |
| #endif /*LODEPNG_COMPILE_ENCODER*/ |
| |
| #else /*no LODEPNG_COMPILE_ZLIB*/ |
| |
| #ifdef LODEPNG_COMPILE_DECODER |
| static unsigned zlib_decompress(unsigned char** out, size_t* outsize, const unsigned char* in, |
| size_t insize, const LodePNGDecompressSettings* settings) { |
| if(!settings->custom_zlib) return 87; /*no custom zlib function provided */ |
| return settings->custom_zlib(out, outsize, in, insize, settings); |
| } |
| #endif /*LODEPNG_COMPILE_DECODER*/ |
| #ifdef LODEPNG_COMPILE_ENCODER |
| static unsigned zlib_compress(unsigned char** out, size_t* outsize, const unsigned char* in, |
| size_t insize, const LodePNGCompressSettings* settings) { |
| if(!settings->custom_zlib) return 87; /*no custom zlib function provided */ |
| return settings->custom_zlib(out, outsize, in, insize, settings); |
| } |
| #endif /*LODEPNG_COMPILE_ENCODER*/ |
| |
| #endif /*LODEPNG_COMPILE_ZLIB*/ |
| |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| |
| #ifdef LODEPNG_COMPILE_ENCODER |
| |
| /*this is a good tradeoff between speed and compression ratio*/ |
| #define DEFAULT_WINDOWSIZE 2048 |
| |
| void lodepng_compress_settings_init(LodePNGCompressSettings* settings) { |
| /*compress with dynamic huffman tree (not in the mathematical sense, just not the predefined one)*/ |
| settings->btype = 2; |
| settings->use_lz77 = 1; |
| settings->windowsize = DEFAULT_WINDOWSIZE; |
| settings->minmatch = 3; |
| settings->nicematch = 128; |
| settings->lazymatching = 1; |
| |
| settings->custom_zlib = 0; |
| settings->custom_deflate = 0; |
| settings->custom_context = 0; |
| } |
| |
| const LodePNGCompressSettings lodepng_default_compress_settings = {2, 1, DEFAULT_WINDOWSIZE, 3, 128, 1, 0, 0, 0}; |
| |
| |
| #endif /*LODEPNG_COMPILE_ENCODER*/ |
| |
| #ifdef LODEPNG_COMPILE_DECODER |
| |
| void lodepng_decompress_settings_init(LodePNGDecompressSettings* settings) { |
| settings->ignore_adler32 = 0; |
| |
| settings->custom_zlib = 0; |
| settings->custom_inflate = 0; |
| settings->custom_context = 0; |
| } |
| |
| const LodePNGDecompressSettings lodepng_default_decompress_settings = {0, 0, 0, 0}; |
| |
| #endif /*LODEPNG_COMPILE_DECODER*/ |
| |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| /* // End of Zlib related code. Begin of PNG related code. // */ |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| |
| #ifdef LODEPNG_COMPILE_PNG |
| |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| /* / CRC32 / */ |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| |
| |
| #ifndef LODEPNG_NO_COMPILE_CRC |
| /* CRC polynomial: 0xedb88320 */ |
| static unsigned lodepng_crc32_table[256] = { |
| 0u, 1996959894u, 3993919788u, 2567524794u, 124634137u, 1886057615u, 3915621685u, 2657392035u, |
| 249268274u, 2044508324u, 3772115230u, 2547177864u, 162941995u, 2125561021u, 3887607047u, 2428444049u, |
| 498536548u, 1789927666u, 4089016648u, 2227061214u, 450548861u, 1843258603u, 4107580753u, 2211677639u, |
| 325883990u, 1684777152u, 4251122042u, 2321926636u, 335633487u, 1661365465u, 4195302755u, 2366115317u, |
| 997073096u, 1281953886u, 3579855332u, 2724688242u, 1006888145u, 1258607687u, 3524101629u, 2768942443u, |
| 901097722u, 1119000684u, 3686517206u, 2898065728u, 853044451u, 1172266101u, 3705015759u, 2882616665u, |
| 651767980u, 1373503546u, 3369554304u, 3218104598u, 565507253u, 1454621731u, 3485111705u, 3099436303u, |
| 671266974u, 1594198024u, 3322730930u, 2970347812u, 795835527u, 1483230225u, 3244367275u, 3060149565u, |
| 1994146192u, 31158534u, 2563907772u, 4023717930u, 1907459465u, 112637215u, 2680153253u, 3904427059u, |
| 2013776290u, 251722036u, 2517215374u, 3775830040u, 2137656763u, 141376813u, 2439277719u, 3865271297u, |
| 1802195444u, 476864866u, 2238001368u, 4066508878u, 1812370925u, 453092731u, 2181625025u, 4111451223u, |
| 1706088902u, 314042704u, 2344532202u, 4240017532u, 1658658271u, 366619977u, 2362670323u, 4224994405u, |
| 1303535960u, 984961486u, 2747007092u, 3569037538u, 1256170817u, 1037604311u, 2765210733u, 3554079995u, |
| 1131014506u, 879679996u, 2909243462u, 3663771856u, 1141124467u, 855842277u, 2852801631u, 3708648649u, |
| 1342533948u, 654459306u, 3188396048u, 3373015174u, 1466479909u, 544179635u, 3110523913u, 3462522015u, |
| 1591671054u, 702138776u, 2966460450u, 3352799412u, 1504918807u, 783551873u, 3082640443u, 3233442989u, |
| 3988292384u, 2596254646u, 62317068u, 1957810842u, 3939845945u, 2647816111u, 81470997u, 1943803523u, |
| 3814918930u, 2489596804u, 225274430u, 2053790376u, 3826175755u, 2466906013u, 167816743u, 2097651377u, |
| 4027552580u, 2265490386u, 503444072u, 1762050814u, 4150417245u, 2154129355u, 426522225u, 1852507879u, |
| 4275313526u, 2312317920u, 282753626u, 1742555852u, 4189708143u, 2394877945u, 397917763u, 1622183637u, |
| 3604390888u, 2714866558u, 953729732u, 1340076626u, 3518719985u, 2797360999u, 1068828381u, 1219638859u, |
| 3624741850u, 2936675148u, 906185462u, 1090812512u, 3747672003u, 2825379669u, 829329135u, 1181335161u, |
| 3412177804u, 3160834842u, 628085408u, 1382605366u, 3423369109u, 3138078467u, 570562233u, 1426400815u, |
| 3317316542u, 2998733608u, 733239954u, 1555261956u, 3268935591u, 3050360625u, 752459403u, 1541320221u, |
| 2607071920u, 3965973030u, 1969922972u, 40735498u, 2617837225u, 3943577151u, 1913087877u, 83908371u, |
| 2512341634u, 3803740692u, 2075208622u, 213261112u, 2463272603u, 3855990285u, 2094854071u, 198958881u, |
| 2262029012u, 4057260610u, 1759359992u, 534414190u, 2176718541u, 4139329115u, 1873836001u, 414664567u, |
| 2282248934u, 4279200368u, 1711684554u, 285281116u, 2405801727u, 4167216745u, 1634467795u, 376229701u, |
| 2685067896u, 3608007406u, 1308918612u, 956543938u, 2808555105u, 3495958263u, 1231636301u, 1047427035u, |
| 2932959818u, 3654703836u, 1088359270u, 936918000u, 2847714899u, 3736837829u, 1202900863u, 817233897u, |
| 3183342108u, 3401237130u, 1404277552u, 615818150u, 3134207493u, 3453421203u, 1423857449u, 601450431u, |
| 3009837614u, 3294710456u, 1567103746u, 711928724u, 3020668471u, 3272380065u, 1510334235u, 755167117u |
| }; |
| |
| /*Return the CRC of the bytes buf[0..len-1].*/ |
| unsigned lodepng_crc32(const unsigned char* data, size_t length) { |
| unsigned r = 0xffffffffu; |
| size_t i; |
| for(i = 0; i < length; ++i) { |
| r = lodepng_crc32_table[(r ^ data[i]) & 0xff] ^ (r >> 8); |
| } |
| return r ^ 0xffffffffu; |
| } |
| #else /* !LODEPNG_NO_COMPILE_CRC */ |
| unsigned lodepng_crc32(const unsigned char* data, size_t length); |
| #endif /* !LODEPNG_NO_COMPILE_CRC */ |
| |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| /* / Reading and writing single bits and bytes from/to stream for LodePNG / */ |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| |
| static unsigned char readBitFromReversedStream(size_t* bitpointer, const unsigned char* bitstream) { |
| unsigned char result = (unsigned char)((bitstream[(*bitpointer) >> 3] >> (7 - ((*bitpointer) & 0x7))) & 1); |
| ++(*bitpointer); |
| return result; |
| } |
| |
| static unsigned readBitsFromReversedStream(size_t* bitpointer, const unsigned char* bitstream, size_t nbits) { |
| unsigned result = 0; |
| size_t i; |
| for(i = 0 ; i < nbits; ++i) { |
| result <<= 1; |
| result |= (unsigned)readBitFromReversedStream(bitpointer, bitstream); |
| } |
| return result; |
| } |
| |
| #ifdef LODEPNG_COMPILE_DECODER |
| static void setBitOfReversedStream0(size_t* bitpointer, unsigned char* bitstream, unsigned char bit) { |
| /*the current bit in bitstream must be 0 for this to work*/ |
| if(bit) { |
| /*earlier bit of huffman code is in a lesser significant bit of an earlier byte*/ |
| bitstream[(*bitpointer) >> 3] |= (bit << (7 - ((*bitpointer) & 0x7))); |
| } |
| ++(*bitpointer); |
| } |
| #endif /*LODEPNG_COMPILE_DECODER*/ |
| |
| static void setBitOfReversedStream(size_t* bitpointer, unsigned char* bitstream, unsigned char bit) { |
| /*the current bit in bitstream may be 0 or 1 for this to work*/ |
| if(bit == 0) bitstream[(*bitpointer) >> 3] &= (unsigned char)(~(1 << (7 - ((*bitpointer) & 0x7)))); |
| else bitstream[(*bitpointer) >> 3] |= (1 << (7 - ((*bitpointer) & 0x7))); |
| ++(*bitpointer); |
| } |
| |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| /* / PNG chunks / */ |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| |
| unsigned lodepng_chunk_length(const unsigned char* chunk) { |
| return lodepng_read32bitInt(&chunk[0]); |
| } |
| |
| void lodepng_chunk_type(char type[5], const unsigned char* chunk) { |
| unsigned i; |
| for(i = 0; i != 4; ++i) type[i] = (char)chunk[4 + i]; |
| type[4] = 0; /*null termination char*/ |
| } |
| |
| unsigned char lodepng_chunk_type_equals(const unsigned char* chunk, const char* type) { |
| if(strlen(type) != 4) return 0; |
| return (chunk[4] == type[0] && chunk[5] == type[1] && chunk[6] == type[2] && chunk[7] == type[3]); |
| } |
| |
| unsigned char lodepng_chunk_ancillary(const unsigned char* chunk) { |
| return((chunk[4] & 32) != 0); |
| } |
| |
| unsigned char lodepng_chunk_private(const unsigned char* chunk) { |
| return((chunk[6] & 32) != 0); |
| } |
| |
| unsigned char lodepng_chunk_safetocopy(const unsigned char* chunk) { |
| return((chunk[7] & 32) != 0); |
| } |
| |
| unsigned char* lodepng_chunk_data(unsigned char* chunk) { |
| return &chunk[8]; |
| } |
| |
| const unsigned char* lodepng_chunk_data_const(const unsigned char* chunk) { |
| return &chunk[8]; |
| } |
| |
| unsigned lodepng_chunk_check_crc(const unsigned char* chunk) { |
| unsigned length = lodepng_chunk_length(chunk); |
| unsigned CRC = lodepng_read32bitInt(&chunk[length + 8]); |
| /*the CRC is taken of the data and the 4 chunk type letters, not the length*/ |
| unsigned checksum = lodepng_crc32(&chunk[4], length + 4); |
| if(CRC != checksum) return 1; |
| else return 0; |
| } |
| |
| void lodepng_chunk_generate_crc(unsigned char* chunk) { |
| unsigned length = lodepng_chunk_length(chunk); |
| unsigned CRC = lodepng_crc32(&chunk[4], length + 4); |
| lodepng_set32bitInt(chunk + 8 + length, CRC); |
| } |
| |
| unsigned char* lodepng_chunk_next(unsigned char* chunk) { |
| if(chunk[0] == 0x89 && chunk[1] == 0x50 && chunk[2] == 0x4e && chunk[3] == 0x47 |
| && chunk[4] == 0x0d && chunk[5] == 0x0a && chunk[6] == 0x1a && chunk[7] == 0x0a) { |
| /* Is PNG magic header at start of PNG file. Jump to first actual chunk. */ |
| return chunk + 8; |
| } else { |
| unsigned total_chunk_length = lodepng_chunk_length(chunk) + 12; |
| return chunk + total_chunk_length; |
| } |
| } |
| |
| const unsigned char* lodepng_chunk_next_const(const unsigned char* chunk) { |
| if(chunk[0] == 0x89 && chunk[1] == 0x50 && chunk[2] == 0x4e && chunk[3] == 0x47 |
| && chunk[4] == 0x0d && chunk[5] == 0x0a && chunk[6] == 0x1a && chunk[7] == 0x0a) { |
| /* Is PNG magic header at start of PNG file. Jump to first actual chunk. */ |
| return chunk + 8; |
| } else { |
| unsigned total_chunk_length = lodepng_chunk_length(chunk) + 12; |
| return chunk + total_chunk_length; |
| } |
| } |
| |
| unsigned char* lodepng_chunk_find(unsigned char* chunk, const unsigned char* end, const char type[5]) { |
| for(;;) { |
| if(chunk + 12 >= end) return 0; |
| if(lodepng_chunk_type_equals(chunk, type)) return chunk; |
| chunk = lodepng_chunk_next(chunk); |
| } |
| } |
| |
| const unsigned char* lodepng_chunk_find_const(const unsigned char* chunk, const unsigned char* end, const char type[5]) { |
| for(;;) { |
| if(chunk + 12 >= end) return 0; |
| if(lodepng_chunk_type_equals(chunk, type)) return chunk; |
| chunk = lodepng_chunk_next_const(chunk); |
| } |
| } |
| |
| unsigned lodepng_chunk_append(unsigned char** out, size_t* outlength, const unsigned char* chunk) { |
| unsigned i; |
| unsigned total_chunk_length = lodepng_chunk_length(chunk) + 12; |
| unsigned char *chunk_start, *new_buffer; |
| size_t new_length = (*outlength) + total_chunk_length; |
| if(new_length < total_chunk_length || new_length < (*outlength)) return 77; /*integer overflow happened*/ |
| |
| new_buffer = (unsigned char*)lodepng_realloc(*out, new_length); |
| if(!new_buffer) return 83; /*alloc fail*/ |
| (*out) = new_buffer; |
| (*outlength) = new_length; |
| chunk_start = &(*out)[new_length - total_chunk_length]; |
| |
| for(i = 0; i != total_chunk_length; ++i) chunk_start[i] = chunk[i]; |
| |
| return 0; |
| } |
| |
| unsigned lodepng_chunk_create(unsigned char** out, size_t* outlength, unsigned length, |
| const char* type, const unsigned char* data) { |
| unsigned i; |
| unsigned char *chunk, *new_buffer; |
| size_t new_length = (*outlength) + length + 12; |
| if(new_length < length + 12 || new_length < (*outlength)) return 77; /*integer overflow happened*/ |
| new_buffer = (unsigned char*)lodepng_realloc(*out, new_length); |
| if(!new_buffer) return 83; /*alloc fail*/ |
| (*out) = new_buffer; |
| (*outlength) = new_length; |
| chunk = &(*out)[(*outlength) - length - 12]; |
| |
| /*1: length*/ |
| lodepng_set32bitInt(chunk, (unsigned)length); |
| |
| /*2: chunk name (4 letters)*/ |
| chunk[4] = (unsigned char)type[0]; |
| chunk[5] = (unsigned char)type[1]; |
| chunk[6] = (unsigned char)type[2]; |
| chunk[7] = (unsigned char)type[3]; |
| |
| /*3: the data*/ |
| for(i = 0; i != length; ++i) chunk[8 + i] = data[i]; |
| |
| /*4: CRC (of the chunkname characters and the data)*/ |
| lodepng_chunk_generate_crc(chunk); |
| |
| return 0; |
| } |
| |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| /* / Color types and such / */ |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| |
| /*return type is a LodePNG error code*/ |
| static unsigned checkColorValidity(LodePNGColorType colortype, unsigned bd) /*bd = bitdepth*/ { |
| switch(colortype) { |
| case 0: if(!(bd == 1 || bd == 2 || bd == 4 || bd == 8 || bd == 16)) return 37; break; /*gray*/ |
| case 2: if(!( bd == 8 || bd == 16)) return 37; break; /*RGB*/ |
| case 3: if(!(bd == 1 || bd == 2 || bd == 4 || bd == 8 )) return 37; break; /*palette*/ |
| case 4: if(!( bd == 8 || bd == 16)) return 37; break; /*gray + alpha*/ |
| case 6: if(!( bd == 8 || bd == 16)) return 37; break; /*RGBA*/ |
| default: return 31; |
| } |
| return 0; /*allowed color type / bits combination*/ |
| } |
| |
| static unsigned getNumColorChannels(LodePNGColorType colortype) { |
| switch(colortype) { |
| case 0: return 1; /*gray*/ |
| case 2: return 3; /*RGB*/ |
| case 3: return 1; /*palette*/ |
| case 4: return 2; /*gray + alpha*/ |
| case 6: return 4; /*RGBA*/ |
| } |
| return 0; /*unexisting color type*/ |
| } |
| |
| static unsigned lodepng_get_bpp_lct(LodePNGColorType colortype, unsigned bitdepth) { |
| /*bits per pixel is amount of channels * bits per channel*/ |
| return getNumColorChannels(colortype) * bitdepth; |
| } |
| |
| /* ////////////////////////////////////////////////////////////////////////// */ |
| |
| void lodepng_color_mode_init(LodePNGColorMode* info) { |
| info->key_defined = 0; |
| info->key_r = info->key_g = info->key_b = 0; |
| info->colortype = LCT_RGBA; |
| info->bitdepth = 8; |
| info->palette = 0; |
| info->palettesize = 0; |
| } |
| |
| void lodepng_color_mode_cleanup(LodePNGColorMode* info) { |
| lodepng_palette_clear(info); |
| } |
| |
| unsigned lodepng_color_mode_copy(LodePNGColorMode* dest, const LodePNGColorMode* source) { |
| size_t i; |
| lodepng_color_mode_cleanup(dest); |
| *dest = *source; |
| if(source->palette) { |
| dest->palette = (unsigned char*)lodepng_malloc(1024); |
| if(!dest->palette && source->palettesize) return 83; /*alloc fail*/ |
| for(i = 0; i != source->palettesize * 4; ++i) dest->palette[i] = source->palette[i]; |
| } |
| return 0; |
| } |
| |
| LodePNGColorMode lodepng_color_mode_make(LodePNGColorType colortype, unsigned bitdepth) { |
| LodePNGColorMode result; |
| lodepng_color_mode_init(&result); |
| result.colortype = colortype; |
| result.bitdepth = bitdepth; |
| return result; |
| } |
| |
| static int lodepng_color_mode_equal(const LodePNGColorMode* a, const LodePNGColorMode* b) { |
| size_t i; |
| if(a->colortype != b->colortype) return 0; |
| if(a->bitdepth != b->bitdepth) return 0; |
| if(a->key_defined != b->key_defined) return 0; |
| if(a->key_defined) { |
| if(a->key_r != b->key_r) return 0; |
| if(a->key_g != b->key_g) return 0; |
| if(a->key_b != b->key_b) return 0; |
| } |
| if(a->palettesize != b->palettesize) return 0; |
| for(i = 0; i != a->palettesize * 4; ++i) { |
| if(a->palette[i] != b->palette[i]) return 0; |
| } |
| return 1; |
| } |
| |
| void lodepng_palette_clear(LodePNGColorMode* info) { |
| if(info->palette) lodepng_free(info->palette); |
| info->palette = 0; |
| info->palettesize = 0; |
| } |
| |
| unsigned lodepng_palette_add(LodePNGColorMode* info, |
| unsigned char r, unsigned char g, unsigned char b, unsigned char a) { |
| unsigned char* data; |
| /*the same resize technique as C++ std::vectors is used, and here it's made so that for a palette with |
| the max of 256 colors, it'll have the exact alloc size*/ |
| if(!info->palette) /*allocate palette if empty*/ { |
| /*room for 256 colors with 4 bytes each*/ |
| data = (unsigned char*)lodepng_realloc(info->palette, 1024); |
| if(!data) return 83; /*alloc fail*/ |
| else info->palette = data; |
| } |
| info->palette[4 * info->palettesize + 0] = r; |
| info->palette[4 * info->palettesize + 1] = g; |
| info->palette[4 * info->palettesize + 2] = b; |
| info->palette[4 * info->palettesize + 3] = a; |
| ++info->palettesize; |
| return 0; |
| } |
| |
| /*calculate bits per pixel out of colortype and bitdepth*/ |
| unsigned lodepng_get_bpp(const LodePNGColorMode* info) { |
| return lodepng_get_bpp_lct(info->colortype, info->bitdepth); |
| } |
| |
| unsigned lodepng_get_channels(const LodePNGColorMode* info) { |
| return getNumColorChannels(info->colortype); |
| } |
| |
| unsigned lodepng_is_greyscale_type(const LodePNGColorMode* info) { |
| return info->colortype == LCT_GREY || info->colortype == LCT_GREY_ALPHA; |
| } |
| |
| unsigned lodepng_is_alpha_type(const LodePNGColorMode* info) { |
| return (info->colortype & 4) != 0; /*4 or 6*/ |
| } |
| |
| unsigned lodepng_is_palette_type(const LodePNGColorMode* info) { |
| return info->colortype == LCT_PALETTE; |
| } |
| |
| unsigned lodepng_has_palette_alpha(const LodePNGColorMode* info) { |
| size_t i; |
| for(i = 0; i != info->palettesize; ++i) { |
| if(info->palette[i * 4 + 3] < 255) return 1; |
| } |
| return 0; |
| } |
| |
| unsigned lodepng_can_have_alpha(const LodePNGColorMode* info) { |
| return info->key_defined |
| || lodepng_is_alpha_type(info) |
| || lodepng_has_palette_alpha(info); |
| } |
| |
| size_t lodepng_get_raw_size_lct(unsigned w, unsigned h, LodePNGColorType colortype, unsigned bitdepth) { |
| size_t bpp = lodepng_get_bpp_lct(colortype, bitdepth); |
| size_t n = (size_t)w * (size_t)h; |
| return ((n / 8) * bpp) + ((n & 7) * bpp + 7) / 8; |
| } |
| |
| size_t lodepng_get_raw_size(unsigned w, unsigned h, const LodePNGColorMode* color) { |
| return lodepng_get_raw_size_lct(w, h, color->colortype, color->bitdepth); |
| } |
| |
| |
| #ifdef LODEPNG_COMPILE_PNG |
| #ifdef LODEPNG_COMPILE_DECODER |
| |
| /*in an idat chunk, each scanline is a multiple of 8 bits, unlike the lodepng output buffer, |
| and in addition has one extra byte per line: the filter byte. So this gives a larger |
| result than lodepng_get_raw_size. */ |
| static size_t lodepng_get_raw_size_idat(unsigned w, unsigned h, const LodePNGColorMode* color) { |
| size_t bpp = lodepng_get_bpp(color); |
| /* + 1 for the filter byte, and possibly plus padding bits per line */ |
| size_t line = ((size_t)(w / 8) * bpp) + 1 + ((w & 7) * bpp + 7) / 8; |
| return (size_t)h * line; |
| } |
| |
| /* Safely check if multiplying two integers will overflow (no undefined |
| behavior, compiler removing the code, etc...) and output result. */ |
| static int lodepng_mulofl(size_t a, size_t b, size_t* result) { |
| *result = a * b; /* Unsigned multiplication is well defined and safe in C90 */ |
| return (a != 0 && *result / a != b); |
| } |
| |
| /* Safely check if adding two integers will overflow (no undefined |
| behavior, compiler removing the code, etc...) and output result. */ |
| static int lodepng_addofl(size_t a, size_t b, size_t* result) { |
| *result = a + b; /* Unsigned addition is well defined and safe in C90 */ |
| return *result < a; |
| } |
|