aacenc: use generational cache instead of resetting.

Approximately 11% faster transcoding from mp3 with
default settings.
Signed-off-by: Reimar Döffinger <Reimar.Doeffinger@gmx.de>

libavcodec/aacenc.c

@@ -78,11 +78,10 @@ static void put_audio_specific_config(AVCodecContext *avctx)
 
 void ff_quantize_band_cost_cache_init(struct AACEncContext *s)
 {
-    int sf, g;
-    for (sf = 0; sf < 256; sf++) {
-        for (g = 0; g < 128; g++) {
-            s->quantize_band_cost_cache[sf][g].bits = -1;
-        }
+    ++s->quantize_band_cost_cache_generation;
+    if (s->quantize_band_cost_cache_generation == 0) {
+        memset(s->quantize_band_cost_cache, 0, sizeof(s->quantize_band_cost_cache));
+        s->quantize_band_cost_cache_generation = 1;
     }
 }
 

libavcodec/aacenc.h

@@ -84,10 +84,10 @@ extern AACCoefficientsEncoder ff_aac_coders[];
 typedef struct AACQuantizeBandCostCacheEntry {
     float rd;
     float energy;
-    int bits; ///< -1 means uninitialized entry
+    int bits;
     char cb;
     char rtz;
-    char padding[2]; ///< Keeps the entry size a multiple of 32 bits
+    uint16_t generation;
 } AACQuantizeBandCostCacheEntry;
 
 /**
@@ -126,6 +126,7 @@ typedef struct AACEncContext {
     DECLARE_ALIGNED(16, int,   qcoefs)[96];      ///< quantized coefficients
     DECLARE_ALIGNED(32, float, scoefs)[1024];    ///< scaled coefficients
 
+    uint16_t quantize_band_cost_cache_generation;
     AACQuantizeBandCostCacheEntry quantize_band_cost_cache[256][128]; ///< memoization area for quantize_band_cost
 
     struct {

libavcodec/aacenc_quantization_misc.h

@@ -36,11 +36,12 @@ static inline float quantize_band_cost_cached(struct AACEncContext *s, int w, in
     AACQuantizeBandCostCacheEntry *entry;
     av_assert1(scale_idx >= 0 && scale_idx < 256);
     entry = &s->quantize_band_cost_cache[scale_idx][w*16+g];
-    if (entry->bits < 0 || entry->cb != cb || entry->rtz != rtz) {
+    if (entry->generation != s->quantize_band_cost_cache_generation || entry->cb != cb || entry->rtz != rtz) {
         entry->rd = quantize_band_cost(s, in, scaled, size, scale_idx,
                                        cb, lambda, uplim, &entry->bits, &entry->energy, rtz);
         entry->cb = cb;
         entry->rtz = rtz;
+        entry->generation = s->quantize_band_cost_cache_generation;
     }
     if (bits)
         *bits = entry->bits;