diff options
author | Lasse Collin <lasse.collin@tukaani.org> | 2024-02-12 17:09:10 +0200 |
---|---|---|
committer | Lasse Collin <lasse.collin@tukaani.org> | 2024-02-14 18:31:16 +0200 |
commit | 88276f9f2cb4871c7eb86952d93d07c1cf6caa66 (patch) | |
tree | b1fc176acae77892173a113f3cced7f211e84856 | |
parent | liblzma: LZ decoder: Add unlikely(). (diff) | |
download | xz-88276f9f2cb4871c7eb86952d93d07c1cf6caa66.tar.xz |
liblzma: Optimize literal_subcoder() macro slightly.
-rw-r--r-- | src/liblzma/lzma/lzma_common.h | 22 | ||||
-rw-r--r-- | src/liblzma/lzma/lzma_decoder.c | 12 | ||||
-rw-r--r-- | src/liblzma/lzma/lzma_encoder.c | 6 | ||||
-rw-r--r-- | src/liblzma/lzma/lzma_encoder_optimum_normal.c | 2 | ||||
-rw-r--r-- | src/liblzma/lzma/lzma_encoder_private.h | 4 |
5 files changed, 24 insertions, 22 deletions
diff --git a/src/liblzma/lzma/lzma_common.h b/src/liblzma/lzma/lzma_common.h index d46b8502..c3c587f0 100644 --- a/src/liblzma/lzma/lzma_common.h +++ b/src/liblzma/lzma/lzma_common.h @@ -125,31 +125,33 @@ typedef enum { /// /// Match byte is used when the previous LZMA symbol was something else than /// a literal (that is, it was some kind of match). -#define LITERAL_CODER_SIZE 0x300 +#define LITERAL_CODER_SIZE UINT32_C(0x300) /// Maximum number of literal coders #define LITERAL_CODERS_MAX (1 << LZMA_LCLP_MAX) +/// Calculates the literal_mask that literal_subcoder() needs. +#define literal_mask_calc(lc, lp) \ + ((UINT32_C(0x100) << (lp)) - (UINT32_C(0x100) >> (lc))) + /// Locate the literal coder for the next literal byte. The choice depends on /// - the lowest literal_pos_bits bits of the position of the current /// byte; and /// - the highest literal_context_bits bits of the previous byte. -#define literal_subcoder(probs, lc, lp_mask, pos, prev_byte) \ - ((probs)[(((pos) & (lp_mask)) << (lc)) \ - + ((uint32_t)(prev_byte) >> (8U - (lc)))]) +#define literal_subcoder(probs, lc, literal_mask, pos, prev_byte) \ + ((probs) + UINT32_C(3) * \ + (((((pos) << 8) + (prev_byte)) & (literal_mask)) << (lc))) static inline void -literal_init(probability (*probs)[LITERAL_CODER_SIZE], - uint32_t lc, uint32_t lp) +literal_init(probability *probs, uint32_t lc, uint32_t lp) { assert(lc + lp <= LZMA_LCLP_MAX); - const uint32_t coders = 1U << (lc + lp); + const size_t coders = LITERAL_CODER_SIZE << (lc + lp); - for (uint32_t i = 0; i < coders; ++i) - for (uint32_t j = 0; j < LITERAL_CODER_SIZE; ++j) - bit_reset(probs[i][j]); + for (size_t i = 0; i < coders; ++i) + bit_reset(probs[i]); return; } diff --git a/src/liblzma/lzma/lzma_decoder.c b/src/liblzma/lzma/lzma_decoder.c index 81149006..66d2818d 100644 --- a/src/liblzma/lzma/lzma_decoder.c +++ b/src/liblzma/lzma/lzma_decoder.c @@ -109,7 +109,7 @@ typedef struct { /////////////////// /// Literals; see comments in lzma_common.h. - probability literal[LITERAL_CODERS_MAX][LITERAL_CODER_SIZE]; + probability literal[LITERAL_CODERS_MAX * LITERAL_CODER_SIZE]; /// If 1, it's a match. Otherwise it's a single 8-bit literal. probability is_match[STATES][POS_STATES_MAX]; @@ -168,7 +168,7 @@ typedef struct { uint32_t pos_mask; // (1U << pb) - 1 uint32_t literal_context_bits; - uint32_t literal_pos_mask; + uint32_t literal_mask; /// Uncompressed size as bytes, or LZMA_VLI_UNKNOWN if end of /// payload marker is expected. @@ -280,7 +280,7 @@ lzma_decode(void *coder_ptr, lzma_dict *restrict dictptr, uint32_t offset = coder->offset; uint32_t len = coder->len; - const uint32_t literal_pos_mask = coder->literal_pos_mask; + const uint32_t literal_mask = coder->literal_mask; const uint32_t literal_context_bits = coder->literal_context_bits; // Temporary variables @@ -359,7 +359,7 @@ lzma_decode(void *coder_ptr, lzma_dict *restrict dictptr, // Get the correct probability array from lp and // lc params. probs = literal_subcoder(coder->literal, - literal_context_bits, literal_pos_mask, + literal_context_bits, literal_mask, dict.pos, dict_get0(&dict)); if (is_literal_state(state)) { @@ -684,7 +684,7 @@ slow: rc_update_0(coder->is_match[state][pos_state]); probs = literal_subcoder(coder->literal, - literal_context_bits, literal_pos_mask, + literal_context_bits, literal_mask, dict.pos, dict_get0(&dict)); symbol = 1; @@ -1034,7 +1034,7 @@ lzma_decoder_reset(void *coder_ptr, const void *opt) literal_init(coder->literal, options->lc, options->lp); coder->literal_context_bits = options->lc; - coder->literal_pos_mask = (1U << options->lp) - 1; + coder->literal_mask = literal_mask_calc(options->lc, options->lp); // State coder->state = STATE_LIT_LIT; diff --git a/src/liblzma/lzma/lzma_encoder.c b/src/liblzma/lzma/lzma_encoder.c index 89d4f4e5..543ca321 100644 --- a/src/liblzma/lzma/lzma_encoder.c +++ b/src/liblzma/lzma/lzma_encoder.c @@ -48,7 +48,7 @@ literal(lzma_lzma1_encoder *coder, lzma_mf *mf, uint32_t position) const uint8_t cur_byte = mf->buffer[ mf->read_pos - mf->read_ahead]; probability *subcoder = literal_subcoder(coder->literal, - coder->literal_context_bits, coder->literal_pos_mask, + coder->literal_context_bits, coder->literal_mask, position, mf->buffer[mf->read_pos - mf->read_ahead - 1]); if (is_literal_state(coder->state)) { @@ -282,7 +282,7 @@ encode_init(lzma_lzma1_encoder *coder, lzma_mf *mf) mf_skip(mf, 1); mf->read_ahead = 0; rc_bit(&coder->rc, &coder->is_match[0][0], 0); - rc_bittree(&coder->rc, coder->literal[0], 8, mf->buffer[0]); + rc_bittree(&coder->rc, coder->literal + 0, 8, mf->buffer[0]); ++coder->uncomp_size; } @@ -534,7 +534,7 @@ lzma_lzma_encoder_reset(lzma_lzma1_encoder *coder, coder->pos_mask = (1U << options->pb) - 1; coder->literal_context_bits = options->lc; - coder->literal_pos_mask = (1U << options->lp) - 1; + coder->literal_mask = literal_mask_calc(options->lc, options->lp); // Range coder rc_reset(&coder->rc); diff --git a/src/liblzma/lzma/lzma_encoder_optimum_normal.c b/src/liblzma/lzma/lzma_encoder_optimum_normal.c index 6b384fe0..a6c0398f 100644 --- a/src/liblzma/lzma/lzma_encoder_optimum_normal.c +++ b/src/liblzma/lzma/lzma_encoder_optimum_normal.c @@ -23,7 +23,7 @@ get_literal_price(const lzma_lzma1_encoder *const coder, const uint32_t pos, uint32_t match_byte, uint32_t symbol) { const probability *const subcoder = literal_subcoder(coder->literal, - coder->literal_context_bits, coder->literal_pos_mask, + coder->literal_context_bits, coder->literal_mask, pos, prev_byte); uint32_t price = 0; diff --git a/src/liblzma/lzma/lzma_encoder_private.h b/src/liblzma/lzma/lzma_encoder_private.h index 6c79b0a0..eeea5e9c 100644 --- a/src/liblzma/lzma/lzma_encoder_private.h +++ b/src/liblzma/lzma/lzma_encoder_private.h @@ -115,10 +115,10 @@ struct lzma_lzma1_encoder_s { uint32_t pos_mask; ///< (1 << pos_bits) - 1 uint32_t literal_context_bits; - uint32_t literal_pos_mask; + uint32_t literal_mask; // These are the same as in lzma_decoder.c. See comments there. - probability literal[LITERAL_CODERS_MAX][LITERAL_CODER_SIZE]; + probability literal[LITERAL_CODERS_MAX * LITERAL_CODER_SIZE]; probability is_match[STATES][POS_STATES_MAX]; probability is_rep[STATES]; probability is_rep0[STATES]; |