From fbaf73a33d505a32431ede6599f5b742ade045e1 Mon Sep 17 00:00:00 2001 From: Michael Niedermayer Date: Tue, 27 May 2014 01:36:23 +0200 Subject: [PATCH] avcodec/lagarithrac: lag_get_rac: use normal division maybe 0.3% faster and simpler code --- libavcodec/lagarithrac.c | 3 --- libavcodec/lagarithrac.h | 15 +++------------ 2 files changed, 3 insertions(+), 15 deletions(-) diff --git a/libavcodec/lagarithrac.c b/libavcodec/lagarithrac.c index c24064607c..37ac2cf570 100644 --- a/libavcodec/lagarithrac.c +++ b/libavcodec/lagarithrac.c @@ -53,7 +53,4 @@ void ff_lag_rac_init(lag_rac *l, GetBitContext *gb, int length) j++; l->range_hash[i] = j; } - - /* Add conversion factor to hash_shift so we don't have to in lag_get_rac. */ - l->hash_shift += 23; } diff --git a/libavcodec/lagarithrac.h b/libavcodec/lagarithrac.h index 7d60d56c2f..dfdfea0db3 100644 --- a/libavcodec/lagarithrac.h +++ b/libavcodec/lagarithrac.h @@ -72,9 +72,8 @@ static inline void lag_rac_refill(lag_rac *l) */ static inline uint8_t lag_get_rac(lag_rac *l) { - unsigned range_scaled, low_scaled, div; + unsigned range_scaled, low_scaled; int val; - uint8_t shift; lag_rac_refill(l); @@ -85,16 +84,8 @@ static inline uint8_t lag_get_rac(lag_rac *l) if (l->low < range_scaled * l->prob[1]) { val = 0; } else { - /* FIXME __builtin_clz is ~20% faster here, but not allowed in generic code. */ - shift = 30 - av_log2(range_scaled); - div = ((range_scaled << shift) + (1 << 23) - 1) >> 23; - /* low>>24 ensures that any cases too big for exact FASTDIV are - * under- rather than over-estimated - */ - low_scaled = FASTDIV(l->low - (l->low >> 24), div); - shift -= l->hash_shift; - low_scaled = (low_scaled >> (32 - shift)); - /* low_scaled is now a lower bound of low/range_scaled */ + low_scaled = l->low / (range_scaled<<(l->hash_shift)); + val = l->range_hash[low_scaled]; while (l->low >= range_scaled * l->prob[val + 1]) val++;