// File: lzham_symbol_codec.h // See Copyright Notice and license at the end of include/lzham.h #pragma once #include "lzham_prefix_coding.h" namespace lzham { class symbol_codec; const uint cSymbolCodecArithMinLen = 0x01000000U; const uint cSymbolCodecArithMaxLen = 0xFFFFFFFFU; const uint cSymbolCodecArithProbBits = 11; const uint cSymbolCodecArithProbScale = 1 << cSymbolCodecArithProbBits; const uint cSymbolCodecArithProbHalfScale = 1 << (cSymbolCodecArithProbBits - 1); const uint cSymbolCodecArithProbMoveBits = 5; typedef uint64 bit_cost_t; const uint32 cBitCostScaleShift = 24; const uint32 cBitCostScale = (1U << cBitCostScaleShift); const bit_cost_t cBitCostMax = cUINT64_MAX; inline bit_cost_t convert_to_scaled_bitcost(uint bits) { LZHAM_ASSERT(bits <= 255); uint32 scaled_bits = bits << cBitCostScaleShift; return static_cast(scaled_bits); } extern uint32 g_prob_cost[cSymbolCodecArithProbScale]; class quasi_adaptive_huffman_data_model { public: quasi_adaptive_huffman_data_model(lzham_malloc_context malloc_context = NULL, bool encoding = false, uint total_syms = 0, uint max_update_interval = 0, uint adapt_rate = 0); quasi_adaptive_huffman_data_model(const quasi_adaptive_huffman_data_model& other); ~quasi_adaptive_huffman_data_model(); bool assign(const quasi_adaptive_huffman_data_model& rhs); quasi_adaptive_huffman_data_model& operator= (const quasi_adaptive_huffman_data_model& rhs); void clear(); void set_malloc_context(lzham_malloc_context malloc_context) { m_malloc_context = malloc_context; m_initial_sym_freq.set_malloc_context(malloc_context); m_sym_freq.set_malloc_context(malloc_context); m_codes.set_malloc_context(malloc_context); m_code_sizes.set_malloc_context(malloc_context); } lzham_malloc_context get_malloc_context() const { return m_malloc_context; } bool init2(lzham_malloc_context context, bool encoding, uint total_syms, uint max_update_interval, uint adapt_rate, const uint16 *pInitial_sym_freq); bool reset(); inline uint get_total_syms() const { return m_total_syms; } void rescale(); void reset_update_rate(); bool update_sym(uint sym); inline bit_cost_t get_cost(uint sym) const { return convert_to_scaled_bitcost(m_code_sizes[sym]); } public: lzham_malloc_context m_malloc_context; lzham::vector m_initial_sym_freq; lzham::vector m_sym_freq; lzham::vector m_codes; lzham::vector m_code_sizes; prefix_coding::decoder_tables* m_pDecode_tables; uint m_total_syms; uint m_max_cycle; uint m_update_cycle; uint m_symbols_until_update; uint m_total_count; uint8 m_decoder_table_bits; uint16 m_max_update_interval; // def=16, typical range 12-128, controls the max interval between table updates, higher=longer max interval (faster decode/lower ratio) uint16 m_adapt_rate; // def=10, 8 or higher, scaled by 8, controls the slowing of the update update freq, higher=more rapid slowing (faster decode/lower ratio) bool m_encoding; bool update_tables(int force_update_cycle = -1, bool sym_freq_all_ones = false); friend class symbol_codec; }; class adaptive_bit_model { public: inline adaptive_bit_model() { clear(); } adaptive_bit_model(float prob0); adaptive_bit_model(const adaptive_bit_model& other); inline adaptive_bit_model& operator= (const adaptive_bit_model& rhs) { m_bit_0_prob = rhs.m_bit_0_prob; return *this; } inline void clear() { m_bit_0_prob = 1U << (cSymbolCodecArithProbBits - 1); } void set_probability_0(float prob0); inline void update(uint bit) { if (!bit) m_bit_0_prob += ((cSymbolCodecArithProbScale - m_bit_0_prob) >> cSymbolCodecArithProbMoveBits); else m_bit_0_prob -= (m_bit_0_prob >> cSymbolCodecArithProbMoveBits); LZHAM_ASSERT(m_bit_0_prob >= 1); LZHAM_ASSERT(m_bit_0_prob < cSymbolCodecArithProbScale); } inline bit_cost_t get_cost(uint bit) const { return g_prob_cost[bit ? (cSymbolCodecArithProbScale - m_bit_0_prob) : m_bit_0_prob]; } public: uint16 m_bit_0_prob; friend class symbol_codec; }; #if LZHAM_CPU_HAS_64BIT_REGISTERS #define LZHAM_SYMBOL_CODEC_USE_64_BIT_BUFFER 1 #else #define LZHAM_SYMBOL_CODEC_USE_64_BIT_BUFFER 0 #endif class symbol_codec { LZHAM_NO_COPY_OR_ASSIGNMENT_OP(symbol_codec); public: symbol_codec(lzham_malloc_context malloc_context); void reset(); // clear() is like reset(), except it also frees all memory. void clear(); // Encoding bool start_encoding(uint expected_file_size); bool encode_bits(uint bits, uint num_bits); bool encode_arith_init(); bool encode_align_to_byte(); bool encode(uint sym, quasi_adaptive_huffman_data_model& model); bool encode(uint bit, adaptive_bit_model& model, bool update_model = true); inline uint encode_get_total_bits_written() const { return m_total_bits_written; } bool stop_encoding(bool support_arith); const lzham::vector& get_encoding_buf() const { return m_output_buf; } lzham::vector& get_encoding_buf() { return m_output_buf; } // Decoding typedef void (*need_bytes_func_ptr)(size_t num_bytes_consumed, void *pPrivate_data, const uint8* &pBuf, size_t &buf_size, bool &eof_flag); bool start_decoding(const uint8* pBuf, size_t buf_size, bool eof_flag = true, need_bytes_func_ptr pNeed_bytes_func = NULL, void *pPrivate_data = NULL); inline void decode_set_input_buffer(const uint8* pBuf, size_t buf_size, const uint8* pBuf_next, bool eof_flag) { m_pDecode_buf = pBuf; m_pDecode_buf_next = pBuf_next; m_decode_buf_size = buf_size; m_pDecode_buf_end = pBuf + buf_size; m_decode_buf_eof = eof_flag; } inline uint64 decode_get_bytes_consumed() const { return m_pDecode_buf_next - m_pDecode_buf; } inline uint64 decode_get_bits_remaining() const { return ((m_pDecode_buf_end - m_pDecode_buf_next) << 3) + m_bit_count; } void start_arith_decoding(); uint decode_bits(uint num_bits); uint decode_peek_bits(uint num_bits); void decode_remove_bits(uint num_bits); void decode_align_to_byte(); int decode_remove_byte_from_bit_buf(); uint decode(quasi_adaptive_huffman_data_model& model); uint decode(adaptive_bit_model& model, bool update_model = true); uint64 stop_decoding(); uint get_total_model_updates() const { return m_total_model_updates; } public: lzham_malloc_context m_malloc_context; const uint8* m_pDecode_buf; const uint8* m_pDecode_buf_next; const uint8* m_pDecode_buf_end; size_t m_decode_buf_size; bool m_decode_buf_eof; need_bytes_func_ptr m_pDecode_need_bytes_func; void* m_pDecode_private_data; #if LZHAM_SYMBOL_CODEC_USE_64_BIT_BUFFER typedef uint64 bit_buf_t; enum { cBitBufSize = 64 }; #else typedef uint32 bit_buf_t; enum { cBitBufSize = 32 }; #endif bit_buf_t m_bit_buf; int m_bit_count; uint m_total_model_updates; lzham::vector m_output_buf; lzham::vector m_arith_output_buf; struct output_symbol { uint m_bits; enum { cArithSym = -1, cAlignToByteSym = -2, cArithInit = -3 }; int16 m_num_bits; uint16 m_arith_prob0; }; lzham::vector m_output_syms; uint m_total_bits_written; uint m_arith_base; uint m_arith_value; uint m_arith_length; uint m_arith_total_bits; quasi_adaptive_huffman_data_model* m_pSaved_huff_model; void* m_pSaved_model; uint m_saved_node_index; bool put_bits_init(uint expected_size); bool record_put_bits(uint bits, uint num_bits); void arith_propagate_carry(); bool arith_renorm_enc_interval(); void arith_start_encoding(); bool arith_stop_encoding(); bool put_bits(uint bits, uint num_bits); bool put_bits_align_to_byte(); bool flush_bits(); bool assemble_output_buf(); uint get_bits(uint num_bits); void remove_bits(uint num_bits); void decode_need_bytes(); enum { cNull, cEncoding, cDecoding } m_mode; }; // Optional macros for faster decompression. These macros implement the symbol_codec class's decode functionality. // This is hard to debug (and just plain ugly), but using these macros eliminate function calls, and they place the most important // member variables on the stack so they're hopefully put in registers (avoiding horrible load hit stores on some CPU's). // The user must define the LZHAM_DECODE_NEEDS_BYTES macro, which is invoked when the decode buffer is exhausted. #define LZHAM_SYMBOL_CODEC_DECODE_DECLARE(codec) \ uint arith_value = 0; \ uint arith_length = 0; \ symbol_codec::bit_buf_t bit_buf = 0; \ int bit_count = 0; \ const uint8* pDecode_buf_next = NULL; #define LZHAM_SYMBOL_CODEC_DECODE_BEGIN(codec) \ arith_value = codec.m_arith_value; \ arith_length = codec.m_arith_length; \ bit_buf = codec.m_bit_buf; \ bit_count = codec.m_bit_count; \ pDecode_buf_next = codec.m_pDecode_buf_next; #define LZHAM_SYMBOL_CODEC_DECODE_END(codec) \ codec.m_arith_value = arith_value; \ codec.m_arith_length = arith_length; \ codec.m_bit_buf = bit_buf; \ codec.m_bit_count = bit_count; \ codec.m_pDecode_buf_next = pDecode_buf_next; // The user must declare the LZHAM_DECODE_NEEDS_BYTES macro. #define LZHAM_SYMBOL_CODEC_DECODE_GET_BITS(codec, result, num_bits) \ { \ while (LZHAM_BUILTIN_EXPECT(bit_count < (int)(num_bits), 0)) \ { \ uint r; \ if (LZHAM_BUILTIN_EXPECT(pDecode_buf_next == codec.m_pDecode_buf_end, 0)) \ { \ if (LZHAM_BUILTIN_EXPECT(!codec.m_decode_buf_eof, 1)) \ { \ LZHAM_SYMBOL_CODEC_DECODE_END(codec) \ LZHAM_DECODE_NEEDS_BYTES \ LZHAM_SYMBOL_CODEC_DECODE_BEGIN(codec) \ } \ r = 0; \ if (LZHAM_BUILTIN_EXPECT(pDecode_buf_next < codec.m_pDecode_buf_end, 1)) r = *pDecode_buf_next++; \ } \ else \ r = *pDecode_buf_next++; \ bit_count += 8; \ bit_buf |= (static_cast(r) << (symbol_codec::cBitBufSize - bit_count)); \ } \ result = (num_bits) ? static_cast(bit_buf >> (symbol_codec::cBitBufSize - (num_bits))) : 0; \ bit_buf <<= (num_bits); \ bit_count -= (num_bits); \ } #define LZHAM_SYMBOL_CODEC_DECODE_ARITH_BIT(codec, result, model) \ { \ adaptive_bit_model *pModel; \ pModel = &model; \ while (LZHAM_BUILTIN_EXPECT(arith_length < cSymbolCodecArithMinLen, 0)) \ { \ uint c; codec.m_pSaved_model = pModel; \ LZHAM_SYMBOL_CODEC_DECODE_GET_BITS(codec, c, 8); \ pModel = static_cast(codec.m_pSaved_model); \ arith_value = (arith_value << 8) | c; \ arith_length <<= 8; \ } \ uint x = pModel->m_bit_0_prob * (arith_length >> cSymbolCodecArithProbBits); \ result = (arith_value >= x); \ if (!result) \ { \ pModel->m_bit_0_prob += ((cSymbolCodecArithProbScale - pModel->m_bit_0_prob) >> cSymbolCodecArithProbMoveBits); \ arith_length = x; \ } \ else \ { \ pModel->m_bit_0_prob -= (pModel->m_bit_0_prob >> cSymbolCodecArithProbMoveBits); \ arith_value -= x; \ arith_length -= x; \ } \ } #if LZHAM_SYMBOL_CODEC_USE_64_BIT_BUFFER #define LZHAM_SYMBOL_CODEC_DECODE_ADAPTIVE_HUFFMAN(codec, result, model) \ { \ quasi_adaptive_huffman_data_model* pModel; const prefix_coding::decoder_tables* pTables; \ pModel = &model; pTables = model.m_pDecode_tables; \ if (LZHAM_BUILTIN_EXPECT(bit_count < 24, 0)) \ { \ uint c; \ pDecode_buf_next += sizeof(uint32); \ if (LZHAM_BUILTIN_EXPECT(pDecode_buf_next >= codec.m_pDecode_buf_end, 0)) \ { \ pDecode_buf_next -= sizeof(uint32); \ while (bit_count < 24) \ { \ if (!codec.m_decode_buf_eof) \ { \ codec.m_pSaved_huff_model = pModel; \ LZHAM_SYMBOL_CODEC_DECODE_END(codec) \ LZHAM_DECODE_NEEDS_BYTES \ LZHAM_SYMBOL_CODEC_DECODE_BEGIN(codec) \ pModel = codec.m_pSaved_huff_model; pTables = pModel->m_pDecode_tables; \ } \ c = 0; if (pDecode_buf_next < codec.m_pDecode_buf_end) c = *pDecode_buf_next++; \ bit_count += 8; \ bit_buf |= (static_cast(c) << (symbol_codec::cBitBufSize - bit_count)); \ } \ } \ else \ { \ c = LZHAM_READ_BIG_ENDIAN_UINT32(pDecode_buf_next - sizeof(uint32)); \ bit_count += 32; \ bit_buf |= (static_cast(c) << (symbol_codec::cBitBufSize - bit_count)); \ } \ } \ uint k = static_cast((bit_buf >> (symbol_codec::cBitBufSize - 16)) + 1); \ uint len; \ if (LZHAM_BUILTIN_EXPECT(k <= pTables->m_table_max_code, 1)) \ { \ uint32 t = pTables->m_lookup[bit_buf >> (symbol_codec::cBitBufSize - pTables->m_table_bits)]; \ result = t & cUINT16_MAX; \ len = t >> 16; \ } \ else \ { \ len = pTables->m_decode_start_code_size; \ for ( ; ; ) \ { \ if (LZHAM_BUILTIN_EXPECT(k <= pTables->m_max_codes[len - 1], 0)) \ break; \ len++; \ } \ int val_ptr = pTables->m_val_ptrs[len - 1] + static_cast(bit_buf >> (symbol_codec::cBitBufSize - len)); \ if (((uint)val_ptr >= pModel->m_total_syms)) val_ptr = 0; \ result = pTables->m_sorted_symbol_order[val_ptr]; \ } \ bit_buf <<= len; \ bit_count -= len; \ uint freq = pModel->m_sym_freq[result]; \ freq++; \ pModel->m_sym_freq[result] = static_cast(freq); \ LZHAM_ASSERT(freq <= cUINT16_MAX); \ if (LZHAM_BUILTIN_EXPECT(--pModel->m_symbols_until_update == 0, 0)) \ { \ pModel->update_tables(); \ } \ } #else #define LZHAM_SYMBOL_CODEC_DECODE_ADAPTIVE_HUFFMAN(codec, result, model) \ { \ quasi_adaptive_huffman_data_model* pModel; const prefix_coding::decoder_tables* pTables; \ pModel = &model; pTables = model.m_pDecode_tables; \ while (LZHAM_BUILTIN_EXPECT(bit_count < (symbol_codec::cBitBufSize - 8), 1)) \ { \ uint c; \ if (LZHAM_BUILTIN_EXPECT(pDecode_buf_next == codec.m_pDecode_buf_end, 0)) \ { \ if (LZHAM_BUILTIN_EXPECT(!codec.m_decode_buf_eof, 1)) \ { \ codec.m_pSaved_huff_model = pModel; \ LZHAM_SYMBOL_CODEC_DECODE_END(codec) \ LZHAM_DECODE_NEEDS_BYTES \ LZHAM_SYMBOL_CODEC_DECODE_BEGIN(codec) \ pModel = codec.m_pSaved_huff_model; pTables = pModel->m_pDecode_tables; \ } \ c = 0; if (LZHAM_BUILTIN_EXPECT(pDecode_buf_next < codec.m_pDecode_buf_end, 1)) c = *pDecode_buf_next++; \ } \ else \ c = *pDecode_buf_next++; \ bit_count += 8; \ bit_buf |= (static_cast(c) << (symbol_codec::cBitBufSize - bit_count)); \ } \ uint k = static_cast((bit_buf >> (symbol_codec::cBitBufSize - 16)) + 1); \ uint len; \ if (LZHAM_BUILTIN_EXPECT(k <= pTables->m_table_max_code, 1)) \ { \ uint32 t = pTables->m_lookup[bit_buf >> (symbol_codec::cBitBufSize - pTables->m_table_bits)]; \ result = t & cUINT16_MAX; \ len = t >> 16; \ } \ else \ { \ len = pTables->m_decode_start_code_size; \ for ( ; ; ) \ { \ if (LZHAM_BUILTIN_EXPECT(k <= pTables->m_max_codes[len - 1], 0)) \ break; \ len++; \ } \ int val_ptr = pTables->m_val_ptrs[len - 1] + static_cast(bit_buf >> (symbol_codec::cBitBufSize - len)); \ if (LZHAM_BUILTIN_EXPECT(((uint)val_ptr >= pModel->m_total_syms), 0)) val_ptr = 0; \ result = pTables->m_sorted_symbol_order[val_ptr]; \ } \ bit_buf <<= len; \ bit_count -= len; \ uint freq = pModel->m_sym_freq[result]; \ freq++; \ pModel->m_sym_freq[result] = static_cast(freq); \ LZHAM_ASSERT(freq <= cUINT16_MAX); \ if (LZHAM_BUILTIN_EXPECT(--pModel->m_symbols_until_update == 0, 0)) \ { \ pModel->update_tables(); \ } \ } #endif #define LZHAM_SYMBOL_CODEC_DECODE_ALIGN_TO_BYTE(codec) if (bit_count & 7) { int dummy_result; LZHAM_NOTE_UNUSED(dummy_result); LZHAM_SYMBOL_CODEC_DECODE_GET_BITS(codec, dummy_result, bit_count & 7); } #define LZHAM_SYMBOL_CODEC_DECODE_REMOVE_BYTE_FROM_BIT_BUF(codec, result) \ { \ result = -1; \ if (bit_count >= 8) \ { \ result = static_cast(bit_buf >> (symbol_codec::cBitBufSize - 8)); \ bit_buf <<= 8; \ bit_count -= 8; \ } \ } #define LZHAM_SYMBOL_CODEC_DECODE_ARITH_START(codec) \ { \ for ( arith_value = 0, arith_length = 0; arith_length < 4; ++arith_length ) \ { \ uint val; LZHAM_SYMBOL_CODEC_DECODE_GET_BITS(codec, val, 8); \ arith_value = (arith_value << 8) | val; \ } \ arith_length = cSymbolCodecArithMaxLen; \ } } // namespace lzham