Merge remote-tracking branch 'newdev/master'

Conflicts:
	Changelog
	doc/APIchanges
	doc/optimization.txt
	libavformat/avio.h

Merged-by: Michael Niedermayer <michaelni@gmx.at>

6 files changed, 119 insertions, 90 deletions

diff --git a/libavcodec/aaccoder.c b/libavcodec/aaccoder.c
index 8063fb6cd4..79723c61ef 100644
--- a/libavcodec/aaccoder.c
+++ b/libavcodec/aaccoder.c
@@ -146,34 +146,34 @@ static av_always_inline float quantize_and_encode_band_cost_template(
             curidx *= range;
             curidx += quants[j] + off;
         }
-            curbits =  ff_aac_spectral_bits[cb-1][curidx];
-            vec     = &ff_aac_codebook_vectors[cb-1][curidx*dim];
-            if (BT_UNSIGNED) {
-                for (k = 0; k < dim; k++) {
-                    float t = fabsf(in[i+k]);
-                    float di;
-                    if (BT_ESC && vec[k] == 64.0f) { //FIXME: slow
-                        if (t >= CLIPPED_ESCAPE) {
-                            di = t - CLIPPED_ESCAPE;
-                            curbits += 21;
-                        } else {
-                            int c = av_clip(quant(t, Q), 0, 8191);
-                            di = t - c*cbrtf(c)*IQ;
-                            curbits += av_log2(c)*2 - 4 + 1;
-                        }
+        curbits =  ff_aac_spectral_bits[cb-1][curidx];
+        vec     = &ff_aac_codebook_vectors[cb-1][curidx*dim];
+        if (BT_UNSIGNED) {
+            for (k = 0; k < dim; k++) {
+                float t = fabsf(in[i+k]);
+                float di;
+                if (BT_ESC && vec[k] == 64.0f) { //FIXME: slow
+                    if (t >= CLIPPED_ESCAPE) {
+                        di = t - CLIPPED_ESCAPE;
+                        curbits += 21;
                     } else {
-                        di = t - vec[k]*IQ;
+                        int c = av_clip(quant(t, Q), 0, 8191);
+                        di = t - c*cbrtf(c)*IQ;
+                        curbits += av_log2(c)*2 - 4 + 1;
                     }
-                    if (vec[k] != 0.0f)
-                        curbits++;
-                    rd += di*di;
-                }
-            } else {
-                for (k = 0; k < dim; k++) {
-                    float di = in[i+k] - vec[k]*IQ;
-                    rd += di*di;
+                } else {
+                    di = t - vec[k]*IQ;
                 }
+                if (vec[k] != 0.0f)
+                    curbits++;
+                rd += di*di;
             }
+        } else {
+            for (k = 0; k < dim; k++) {
+                float di = in[i+k] - vec[k]*IQ;
+                rd += di*di;
+            }
+        }
         cost    += rd * lambda + curbits;
         resbits += curbits;
         if (cost >= uplim)
@@ -575,7 +575,7 @@ static void search_for_quantizers_anmr(AVCodecContext *avctx, AACEncContext *s,
         int qnrg = av_clip_uint8(log2f(sqrtf(qnrgf/qcnt))*4 - 31 + SCALE_ONE_POS - SCALE_DIV_512);
         q1 = qnrg + 30;
         q0 = qnrg - 30;
-    //av_log(NULL, AV_LOG_ERROR, "q0 %d, q1 %d\n", q0, q1);
+        //av_log(NULL, AV_LOG_ERROR, "q0 %d, q1 %d\n", q0, q1);
         if (q0 < q0low) {
             q1 += q0low - q0;
             q0  = q0low;
@@ -723,7 +723,7 @@ static void search_for_quantizers_twoloop(AVCodecContext *avctx,
             sce->zeroes[w*16+g] = !nz;
             if (nz)
                 minthr = FFMIN(minthr, uplim);
-            allz = FFMAX(allz, nz);
+            allz |= nz;
         }
     }
     for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
diff --git a/libavcodec/aacpsy.c b/libavcodec/aacpsy.c
index a987be0abb..d9896ed74f 100644
--- a/libavcodec/aacpsy.c
+++ b/libavcodec/aacpsy.c
@@ -39,8 +39,8 @@
  * constants for 3GPP AAC psychoacoustic model
  * @{
  */
-#define PSY_3GPP_SPREAD_HI   1.5f // spreading factor for ascending threshold spreading  (15 dB/Bark)
-#define PSY_3GPP_SPREAD_LOW  3.0f // spreading factor for descending threshold spreading (30 dB/Bark)
+#define PSY_3GPP_THR_SPREAD_HI   1.5f // spreading factor for low-to-hi threshold spreading  (15 dB/Bark)
+#define PSY_3GPP_THR_SPREAD_LOW  3.0f // spreading factor for hi-to-low threshold spreading  (30 dB/Bark)
 
 #define PSY_3GPP_RPEMIN      0.01f
 #define PSY_3GPP_RPELEV      2.0f
@@ -61,9 +61,7 @@
  */
 typedef struct AacPsyBand{
     float energy;    ///< band energy
-    float ffac;      ///< form factor
     float thr;       ///< energy threshold
-    float min_snr;   ///< minimal SNR
     float thr_quiet; ///< threshold in quiet
 }AacPsyBand;
 
@@ -88,17 +86,18 @@ typedef struct AacPsyChannel{
  * psychoacoustic model frame type-dependent coefficients
  */
 typedef struct AacPsyCoeffs{
-    float ath       [64]; ///< absolute threshold of hearing per bands
-    float barks     [64]; ///< Bark value for each spectral band in long frame
-    float spread_low[64]; ///< spreading factor for low-to-high threshold spreading in long frame
-    float spread_hi [64]; ///< spreading factor for high-to-low threshold spreading in long frame
+    float ath;           ///< absolute threshold of hearing per bands
+    float barks;         ///< Bark value for each spectral band in long frame
+    float spread_low[2]; ///< spreading factor for low-to-high threshold spreading in long frame
+    float spread_hi [2]; ///< spreading factor for high-to-low threshold spreading in long frame
+    float min_snr;       ///< minimal SNR
 }AacPsyCoeffs;
 
 /**
  * 3GPP TS26.403-inspired psychoacoustic model specific data
  */
 typedef struct AacPsyContext{
-    AacPsyCoeffs psy_coef[2];
+    AacPsyCoeffs psy_coef[2][64];
     AacPsyChannel *ch;
 }AacPsyContext;
 
@@ -243,27 +242,30 @@ static av_cold int psy_3gpp_init(FFPsyContext *ctx) {
 
     minath = ath(3410, ATH_ADD);
     for (j = 0; j < 2; j++) {
-        AacPsyCoeffs *coeffs = &pctx->psy_coef[j];
+        AacPsyCoeffs *coeffs = pctx->psy_coef[j];
+        const uint8_t *band_sizes = ctx->bands[j];
         float line_to_frequency = ctx->avctx->sample_rate / (j ? 256.f : 2048.0f);
         i = 0;
         prev = 0.0;
         for (g = 0; g < ctx->num_bands[j]; g++) {
-            i += ctx->bands[j][g];
+            i += band_sizes[g];
             bark = calc_bark((i-1) * line_to_frequency);
-            coeffs->barks[g] = (bark + prev) / 2.0;
+            coeffs[g].barks = (bark + prev) / 2.0;
             prev = bark;
         }
         for (g = 0; g < ctx->num_bands[j] - 1; g++) {
-            coeffs->spread_low[g] = pow(10.0, -(coeffs->barks[g+1] - coeffs->barks[g]) * PSY_3GPP_SPREAD_LOW);
-            coeffs->spread_hi [g] = pow(10.0, -(coeffs->barks[g+1] - coeffs->barks[g]) * PSY_3GPP_SPREAD_HI);
+            AacPsyCoeffs *coeff = &coeffs[g];
+            float bark_width = coeffs[g+1].barks - coeffs->barks;
+            coeff->spread_low[0] = pow(10.0, -bark_width * PSY_3GPP_THR_SPREAD_LOW);
+            coeff->spread_hi [0] = pow(10.0, -bark_width * PSY_3GPP_THR_SPREAD_HI);
         }
         start = 0;
         for (g = 0; g < ctx->num_bands[j]; g++) {
             minscale = ath(start * line_to_frequency, ATH_ADD);
-            for (i = 1; i < ctx->bands[j][g]; i++)
+            for (i = 1; i < band_sizes[g]; i++)
                 minscale = FFMIN(minscale, ath((start + i) * line_to_frequency, ATH_ADD));
-            coeffs->ath[g] = minscale - minath;
-            start += ctx->bands[j][g];
+            coeffs[g].ath = minscale - minath;
+            start += band_sizes[g];
         }
     }
 
@@ -393,9 +395,9 @@ static void psy_3gpp_analyze(FFPsyContext *ctx, int channel,
     AacPsyChannel *pch  = &pctx->ch[channel];
     int start = 0;
     int i, w, g;
-    const int num_bands       = ctx->num_bands[wi->num_windows == 8];
-    const uint8_t* band_sizes = ctx->bands[wi->num_windows == 8];
-    AacPsyCoeffs *coeffs     = &pctx->psy_coef[wi->num_windows == 8];
+    const int      num_bands  = ctx->num_bands[wi->num_windows == 8];
+    const uint8_t *band_sizes = ctx->bands[wi->num_windows == 8];
+    AacPsyCoeffs  *coeffs     = &pctx->psy_coef[wi->num_windows == 8];
 
     //calculate energies, initial thresholds and related values - 5.4.2 "Threshold Calculation"
     for (w = 0; w < wi->num_windows*16; w += 16) {
@@ -406,26 +408,37 @@ static void psy_3gpp_analyze(FFPsyContext *ctx, int channel,
                 band->energy += coefs[start+i] * coefs[start+i];
             band->thr     = band->energy * 0.001258925f;
             start        += band_sizes[g];
-
-            ctx->psy_bands[channel*PSY_MAX_BANDS+w+g].energy = band->energy;
         }
     }
-    //modify thresholds - spread, threshold in quiet - 5.4.3 "Spreaded Energy Calculation"
+    //modify thresholds and energies - spread, threshold in quiet, pre-echo control
     for (w = 0; w < wi->num_windows*16; w += 16) {
-        AacPsyBand *band = &pch->band[w];
+        AacPsyBand *bands = &pch->band[w];
+        //5.4.2.3 "Spreading" & 5.4.3 "Spreaded Energy Calculation"
         for (g = 1; g < num_bands; g++)
-            band[g].thr = FFMAX(band[g].thr, band[g-1].thr * coeffs->spread_hi [g]);
+            bands[g].thr = FFMAX(bands[g].thr, bands[g-1].thr * coeffs[g].spread_hi[0]);
         for (g = num_bands - 2; g >= 0; g--)
-            band[g].thr = FFMAX(band[g].thr, band[g+1].thr * coeffs->spread_low[g]);
+            bands[g].thr = FFMAX(bands[g].thr, bands[g+1].thr * coeffs[g].spread_low[0]);
+        //5.4.2.4 "Threshold in quiet"
         for (g = 0; g < num_bands; g++) {
-            band[g].thr_quiet = band[g].thr = FFMAX(band[g].thr, coeffs->ath[g]);
+            AacPsyBand *band = &bands[g];
+            band->thr_quiet = band->thr = FFMAX(band->thr, coeffs[g].ath);
+            //5.4.2.5 "Pre-echo control"
             if (!(wi->window_type[0] == LONG_STOP_SEQUENCE || (wi->window_type[1] == LONG_START_SEQUENCE && !w)))
-                band[g].thr = FFMAX(PSY_3GPP_RPEMIN*band[g].thr, FFMIN(band[g].thr,
-                                    PSY_3GPP_RPELEV*pch->prev_band[w+g].thr_quiet));
+                band->thr = FFMAX(PSY_3GPP_RPEMIN*band->thr, FFMIN(band->thr,
+                                  PSY_3GPP_RPELEV*pch->prev_band[w+g].thr_quiet));
+        }
+    }
+
+    for (w = 0; w < wi->num_windows*16; w += 16) {
+        for (g = 0; g < num_bands; g++) {
+            AacPsyBand *band     = &pch->band[w+g];
+            FFPsyBand  *psy_band = &ctx->psy_bands[channel*PSY_MAX_BANDS+w+g];
 
-            ctx->psy_bands[channel*PSY_MAX_BANDS+w+g].threshold = band[g].thr;
+            psy_band->threshold = band->thr;
+            psy_band->energy    = band->energy;
         }
     }
+
     memcpy(pch->prev_band, pch->band, sizeof(pch->band));
 }
 
@@ -553,22 +566,9 @@ static FFPsyWindowInfo psy_lame_window(FFPsyContext *ctx,
         if (pch->prev_attack == 3 || att_sum) {
             uselongblock = 0;
 
-            if (attacks[1] && attacks[0])
-                attacks[1] = 0;
-            if (attacks[2] && attacks[1])
-                attacks[2] = 0;
-            if (attacks[3] && attacks[2])
-                attacks[3] = 0;
-            if (attacks[4] && attacks[3])
-                attacks[4] = 0;
-            if (attacks[5] && attacks[4])
-                attacks[5] = 0;
-            if (attacks[6] && attacks[5])
-                attacks[6] = 0;
-            if (attacks[7] && attacks[6])
-                attacks[7] = 0;
-            if (attacks[8] && attacks[7])
-                attacks[8] = 0;
+            for (i = 1; i < AAC_NUM_BLOCKS_SHORT + 1; i++)
+                if (attacks[i] && attacks[i-1])
+                    attacks[i] = 0;
         }
     } else {
         /* We have no lookahead info, so just use same type as the previous sequence. */
diff --git a/libavcodec/dsputil.c b/libavcodec/dsputil.c
index 1f7bd4cbc0..b293642b1d 100644
--- a/libavcodec/dsputil.c
+++ b/libavcodec/dsputil.c
@@ -222,6 +222,12 @@ static void bswap_buf(uint32_t *dst, const uint32_t *src, int w){
     }
 }
 
+static void bswap16_buf(uint16_t *dst, const uint16_t *src, int len)
+{
+    while (len--)
+        *dst++ = av_bswap16(*src++);
+}
+
 static int sse4_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h)
 {
     int s, i;
@@ -4324,6 +4330,7 @@ av_cold void dsputil_init(DSPContext* c, AVCodecContext *avctx)
     c->add_hfyu_left_prediction  = add_hfyu_left_prediction_c;
     c->add_hfyu_left_prediction_bgr32 = add_hfyu_left_prediction_bgr32_c;
     c->bswap_buf= bswap_buf;
+    c->bswap16_buf = bswap16_buf;
 #if CONFIG_PNG_DECODER
     c->add_png_paeth_prediction= ff_add_png_paeth_prediction;
 #endif
diff --git a/libavcodec/dsputil.h b/libavcodec/dsputil.h
index 3e55d1338d..017219bc89 100644
--- a/libavcodec/dsputil.h
+++ b/libavcodec/dsputil.h
@@ -369,6 +369,7 @@ typedef struct DSPContext {
     /* this might write to dst[w] */
     void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
     void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
+    void (*bswap16_buf)(uint16_t *dst, const uint16_t *src, int len);
 
     void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
     void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
diff --git a/libavcodec/utils.c b/libavcodec/utils.c
index 40d689e1f6..5a39bff168 100644
--- a/libavcodec/utils.c
+++ b/libavcodec/utils.c
@@ -1078,13 +1078,13 @@ unsigned avcodec_version( void )
 
 const char *avcodec_configuration(void)
 {
-    return FFMPEG_CONFIGURATION;
+    return LIBAV_CONFIGURATION;
 }
 
 const char *avcodec_license(void)
 {
 #define LICENSE_PREFIX "libavcodec license: "
-    return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
+    return LICENSE_PREFIX LIBAV_LICENSE + sizeof(LICENSE_PREFIX) - 1;
 }
 
 void avcodec_init(void)
diff --git a/libavcodec/x86/mathops.h b/libavcodec/x86/mathops.h
index 5949dfe3d4..33d9a6c8ff 100644
--- a/libavcodec/x86/mathops.h
+++ b/libavcodec/x86/mathops.h
@@ -26,24 +26,45 @@
 #include "libavutil/common.h"
 
 #if ARCH_X86_32
-#define MULL(ra, rb, shift) \
-        ({ int rt, dummy; __asm__ (\
-            "imull %3               \n\t"\
-            "shrdl %4, %%edx, %%eax \n\t"\
-            : "=a"(rt), "=d"(dummy)\
-            : "a" ((int)(ra)), "rm" ((int)(rb)), "i"(shift));\
-         rt; })
 
-#define MULH(ra, rb) \
-    ({ int rt, dummy;\
-     __asm__ ("imull %3\n\t" : "=d"(rt), "=a"(dummy): "a" ((int)(ra)), "rm" ((int)(rb)));\
-     rt; })
+#define MULL MULL
+static av_always_inline av_const int MULL(int a, int b, unsigned shift)
+{
+    int rt, dummy;
+    __asm__ (
+        "imull %3               \n\t"
+        "shrdl %4, %%edx, %%eax \n\t"
+        :"=a"(rt), "=d"(dummy)
+        :"a"(a), "rm"(b), "ci"((uint8_t)shift)
+    );
+    return rt;
+}
 
-#define MUL64(ra, rb) \
-    ({ int64_t rt;\
-     __asm__ ("imull %2\n\t" : "=A"(rt) : "a" ((int)(ra)), "g" ((int)(rb)));\
-     rt; })
-#endif
+#define MULH MULH
+static av_always_inline av_const int MULH(int a, int b)
+{
+    int rt, dummy;
+    __asm__ (
+        "imull %3"
+        :"=d"(rt), "=a"(dummy)
+        :"a"(a), "rm"(b)
+    );
+    return rt;
+}
+
+#define MUL64 MUL64
+static av_always_inline av_const int64_t MUL64(int a, int b)
+{
+    int64_t rt;
+    __asm__ (
+        "imull %2"
+        :"=A"(rt)
+        :"a"(a), "rm"(b)
+    );
+    return rt;
+}
+
+#endif /* ARCH_X86_32 */
 
 #if HAVE_CMOV
 /* median of 3 */