Merge remote-tracking branch 'qatar/master'

* qatar/master:
  qdm2: Use floating point synthesis filter.
  h264: correct border check.
  h264: fix loopfilter with threading at slice boundaries.
  Fix ff_mpa_synth_filter_fixed() prototype
  Rename costablegen.c ---> cos_tablegen.c.
  Collapse tableprint.c into tableprint.h.
  Simplify trig table rules
  Remove potentially unstable filenames from comments in generated files.
  Ignore generated tables and generated table generator programs.
  Simplify CLEANFILES make variable by using wildcards.
  Remove silly insults from avformat_version() Doxygen documentation.
  mpegaudiodsp: fix x86 and ppc makefiles
  configure: Adjust AVX assembler check.
  mpegaudio: remove unused version of SAME_HEADER_MASK
  mpegaudio: remove useless #undef at end of file
  asfdec: add missing #include for av_bswap32()
  mpegaudio: merge two #if CONFIG_FLOAT blocks
  mpegaudio: move some struct definitions from mpegaudio.h
  Move some mpegaudio functions to new mpegaudiodsp subsystem

Conflicts:
	libavcodec/h264.c
	libavcodec/x86/Makefile

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

.gitignore

@@ -16,6 +16,9 @@ ffmpeg
 ffplay
 ffprobe
 ffserver
+libavcodec/*_tablegen
+libavcodec/*_tables.c
+libavcodec/*_tables.h
 libavcodec/libavcodec*
 libavcore/libavcore*
 libavdevice/libavdevice*

configure

@@ -956,6 +956,7 @@ CONFIG_LIST="
     mdct
     memalign_hack
     mlib
+    mpegaudiodsp
     network
     nonfree
     pic
@@ -1244,6 +1245,7 @@ symver_if_any="symver_asm_label symver_gnu_asm"
 dct_select="rdft"
 mdct_select="fft"
 rdft_select="fft"
+mpegaudiodsp_select="dct"
 
 # decoders / encoders / hardware accelerators
 aac_decoder_select="mdct sinewin"
@@ -1297,11 +1299,16 @@ ljpeg_encoder_select="aandct"
 loco_decoder_select="golomb"
 mjpeg_encoder_select="aandct"
 mlp_decoder_select="mlp_parser"
-mp1float_decoder_select="dct"
-mp2float_decoder_select="dct"
-mp3adufloat_decoder_select="dct"
-mp3float_decoder_select="dct"
-mp3on4float_decoder_select="dct"
+mp1_decoder_select="mpegaudiodsp"
+mp2_decoder_select="mpegaudiodsp"
+mp3adu_decoder_select="mpegaudiodsp"
+mp3_decoder_select="mpegaudiodsp"
+mp3on4_decoder_select="mpegaudiodsp"
+mp1float_decoder_select="mpegaudiodsp"
+mp2float_decoder_select="mpegaudiodsp"
+mp3adufloat_decoder_select="mpegaudiodsp"
+mp3float_decoder_select="mpegaudiodsp"
+mp3on4float_decoder_select="mpegaudiodsp"
 mpeg1video_encoder_select="aandct"
 mpeg2video_encoder_select="aandct"
 mpeg4_decoder_select="h263_decoder mpeg4video_parser"
@@ -1331,7 +1338,7 @@ nellymoser_encoder_select="mdct sinewin"
 png_decoder_select="zlib"
 png_encoder_select="zlib"
 qcelp_decoder_select="lsp"
-qdm2_decoder_select="mdct rdft"
+qdm2_decoder_select="mdct rdft mpegaudiodsp"
 ra_144_encoder_select="lpc"
 rv10_decoder_select="h263_decoder"
 rv10_encoder_select="h263_encoder"
@@ -2753,7 +2760,7 @@ EOF
 
         check_yasm "pextrd [eax], xmm0, 1" && enable yasm ||
             die "yasm not found, use --disable-yasm for a crippled build"
-        check_yasm "vpaddw xmm0, xmm0, xmm0" || disable avx
+        check_yasm "vextractf128 xmm0, ymm0, 0" || disable avx
     fi
 
     case "$cpu" in

libavcodec/Makefile

@@ -41,6 +41,9 @@ OBJS-$(CONFIG_HUFFMAN)                 += huffman.o
 OBJS-$(CONFIG_LPC)                     += lpc.o
 OBJS-$(CONFIG_LSP)                     += lsp.o
 OBJS-$(CONFIG_MDCT)                    += mdct_fixed.o mdct_float.o
+OBJS-$(CONFIG_MPEGAUDIODSP)            += mpegaudiodsp.o                \
+                                          mpegaudiodsp_fixed.o          \
+                                          mpegaudiodsp_float.o
 RDFT-OBJS-$(CONFIG_HARDCODED_TABLES)   += sin_tables.o
 OBJS-$(CONFIG_RDFT)                    += rdft.o $(RDFT-OBJS-yes)
 OBJS-$(CONFIG_SINEWIN)                 += sinewin.o
@@ -676,24 +679,23 @@ TESTPROGS = cabac dct eval fft fft-fixed h264 iirfilter rangecoder snow
 TESTPROGS-$(HAVE_MMX) += motion
 TESTOBJS = dctref.o
 
-HOSTPROGS = costablegen
+HOSTPROGS = aac_tablegen aacps_tablegen cbrt_tablegen cos_tablegen      \
+            dv_tablegen motionpixels_tablegen mpegaudio_tablegen        \
+            pcm_tablegen qdm2_tablegen sinewin_tablegen
 
 DIRS = alpha arm bfin mlib ppc ps2 sh4 sparc x86
 
-CLEANFILES = sin_tables.c cos_tables.c *_tables.h *_tablegen$(HOSTEXESUF)
+CLEANFILES = *_tables.c *_tables.h *_tablegen$(HOSTEXESUF)
 
 include $(SUBDIR)../subdir.mak
 
 $(SUBDIR)dct-test$(EXESUF): $(SUBDIR)dctref.o
 
-$(SUBDIR)cos_tables.c: $(SUBDIR)costablegen$(HOSTEXESUF)
-	$(M)./$< > $@
-
-$(SUBDIR)cos_fixed_tables.c: $(SUBDIR)costablegen$(HOSTEXESUF)
-	$(M)./$< cos fixed > $@
+TRIG_TABLES  = cos cos_fixed sin
+TRIG_TABLES := $(TRIG_TABLES:%=$(SUBDIR)%_tables.c)
 
-$(SUBDIR)sin_tables.c: $(SUBDIR)costablegen$(HOSTEXESUF)
-	$(M)./$< sin > $@
+$(TRIG_TABLES): $(SUBDIR)%_tables.c: $(SUBDIR)cos_tablegen$(HOSTEXESUF)
+	$(M)./$< $* > $@
 
 ifdef CONFIG_SMALL
 $(SUBDIR)%_tablegen$(HOSTEXESUF): HOSTCFLAGS += -DCONFIG_SMALL=1
@@ -701,9 +703,6 @@ else
 $(SUBDIR)%_tablegen$(HOSTEXESUF): HOSTCFLAGS += -DCONFIG_SMALL=0
 endif
 
-$(SUBDIR)%_tablegen$(HOSTEXESUF): $(SUBDIR)%_tablegen.c $(SUBDIR)%_tablegen.h $(SUBDIR)tableprint.c
-	$(HOSTCC) $(HOSTCFLAGS) $(HOSTLDFLAGS) -o $@ $(filter %.c,$^) $(HOSTLIBS)
-
 GEN_HEADERS = cbrt_tables.h aacps_tables.h aac_tables.h dv_tables.h     \
               sinewin_tables.h mpegaudio_tables.h motionpixels_tables.h \
               pcm_tables.h qdm2_tables.h

libavcodec/costablegen.c → libavcodec/cos_tablegen.c

@@ -51,10 +51,10 @@ int main(int argc, char *argv[])
 {
     int i, j;
     int do_sin = argc > 1 && !strcmp(argv[1], "sin");
-    int fixed  = argc > 2 && !strcmp(argv[2], "fixed");
+    int fixed  = argc > 1 &&  strstr(argv[1], "fixed");
     double (*func)(double) = do_sin ? sin : cos;
 
-    printf("/* This file was generated by libavcodec/costablegen */\n");
+    printf("/* This file was automatically generated. */\n");
     printf("#define CONFIG_FFT_FLOAT %d\n", !fixed);
     printf("#include \"libavcodec/%s\"\n", do_sin ? "rdft.h" : "fft.h");
     for (i = 4; i <= BITS; i++) {

libavcodec/h264.c

@@ -1002,7 +1002,7 @@ static inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
                                   int linesize, int uvlinesize,
                                   int xchg, int simple, int pixel_shift){
     MpegEncContext * const s = &h->s;
-    int deblock_left;
+    int deblock_topleft;
     int deblock_top;
     int top_idx = 1;
     uint8_t *top_border_m1;
@@ -1018,11 +1018,11 @@ static inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
     }
 
     if(h->deblocking_filter == 2) {
-        deblock_left = h->left_type[0];
-        deblock_top  = h->top_type;
+        deblock_topleft = h->slice_table[h->mb_xy - 1 - s->mb_stride] == h->slice_num;
+        deblock_top     = h->top_type;
     } else {
-        deblock_left = (s->mb_x > 0);
-        deblock_top =  (s->mb_y > !!MB_FIELD);
+        deblock_topleft = (s->mb_x > 0);
+        deblock_top     = (s->mb_y > !!MB_FIELD);
     }
 
     src_y  -=   linesize + 1 + pixel_shift;
@@ -1045,7 +1045,7 @@ if (xchg) AV_SWAP64(b,a);\
 else      AV_COPY64(b,a);
 
     if(deblock_top){
-        if(deblock_left){
+        if(deblock_topleft){
             XCHG(top_border_m1 + (8 << pixel_shift), src_y - (7 << pixel_shift), 1);
         }
         XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
@@ -1056,7 +1056,7 @@ else      AV_COPY64(b,a);
     }
     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
         if(deblock_top){
-            if(deblock_left){
+            if(deblock_topleft){
                 XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
                 XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
             }
@@ -2561,18 +2561,16 @@ static int fill_filter_caches(H264Context *h, int mb_type){
     return 0;
 }
 
-static void loop_filter(H264Context *h){
+static void loop_filter(H264Context *h, int start_x, int end_x){
     MpegEncContext * const s = &h->s;
     uint8_t  *dest_y, *dest_cb, *dest_cr;
     int linesize, uvlinesize, mb_x, mb_y;
     const int end_mb_y= s->mb_y + FRAME_MBAFF;
     const int old_slice_type= h->slice_type;
-    const int end_mb_x  = s->mb_x;
     const int pixel_shift = h->pixel_shift;
 
     if(h->deblocking_filter) {
-        int start_x= s->resync_mb_y == s->mb_y ? s->resync_mb_x : 0;
-        for(mb_x= start_x; mb_x<end_mb_x; mb_x++){
+        for(mb_x= start_x; mb_x<end_x; mb_x++){
             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
                 int mb_xy, mb_type;
                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
@@ -2617,7 +2615,7 @@ static void loop_filter(H264Context *h){
         }
     }
     h->slice_type= old_slice_type;
-    s->mb_x= end_mb_x;
+    s->mb_x= end_x;
     s->mb_y= end_mb_y - FRAME_MBAFF;
     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
@@ -2672,6 +2670,7 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
     H264Context *h = *(void**)arg;
     MpegEncContext * const s = &h->s;
     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
+    int lf_x_start = s->mb_x;
 
     s->mb_skip_run= -1;
 
@@ -2710,6 +2709,7 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
 
             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
+                if (s->mb_x >= lf_x_start) loop_filter(h, lf_x_start, s->mb_x + 1);
                 return 0;
             }
             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
@@ -2719,8 +2719,8 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
             }
 
             if( ++s->mb_x >= s->mb_width ) {
-                loop_filter(h);
-                s->mb_x = 0;
+                loop_filter(h, lf_x_start, s->mb_x);
+                s->mb_x = lf_x_start = 0;
                 decode_finish_row(h);
                 ++s->mb_y;
                 if(FIELD_OR_MBAFF_PICTURE) {
@@ -2731,10 +2731,9 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
             }
 
             if( eos || s->mb_y >= s->mb_height ) {
-                if(s->mb_x)
-                    loop_filter(h);
                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
+                if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
                 return 0;
             }
         }
@@ -2756,13 +2755,12 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
             if(ret<0){
                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
-
                 return -1;
             }
 
             if(++s->mb_x >= s->mb_width){
-                loop_filter(h);
-                s->mb_x=0;
+                loop_filter(h, lf_x_start, s->mb_x);
+                s->mb_x = lf_x_start = 0;
                 decode_finish_row(h);
                 ++s->mb_y;
                 if(FIELD_OR_MBAFF_PICTURE) {
@@ -2789,9 +2787,8 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
-                    if(s->mb_x)
-                        loop_filter(h);
                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
+                    if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
 
                     return 0;
                 }else{

libavcodec/mpc.c

@@ -29,6 +29,7 @@
 #include "avcodec.h"
 #include "get_bits.h"
 #include "dsputil.h"
+#include "mpegaudiodsp.h"
 #include "mpegaudio.h"
 
 #include "mpc.h"
@@ -51,7 +52,8 @@ static void mpc_synth(MPCContext *c, int16_t *out, int channels)
     for(ch = 0;  ch < channels; ch++){
         samples_ptr = samples + ch;
         for(i = 0; i < SAMPLES_PER_BAND; i++) {
-            ff_mpa_synth_filter_fixed(c->synth_buf[ch], &(c->synth_buf_offset[ch]),
+            ff_mpa_synth_filter_fixed(&c->mpadsp,
+                                c->synth_buf[ch], &(c->synth_buf_offset[ch]),
                                 ff_mpa_synth_window_fixed, &dither_state,
                                 samples_ptr, channels,
                                 c->sb_samples[ch][i]);

libavcodec/mpc.h

@@ -52,6 +52,7 @@ typedef struct {
 
 typedef struct {
     DSPContext dsp;
+    MPADSPContext mpadsp;
     GetBitContext gb;
     int IS, MSS, gapless;
     int lastframelen;

libavcodec/mpc7.c

@@ -29,7 +29,7 @@
 #include "avcodec.h"
 #include "get_bits.h"
 #include "dsputil.h"
-#include "mpegaudio.h"
+#include "mpegaudiodsp.h"
 #include "libavutil/audioconvert.h"
 
 #include "mpc.h"
@@ -68,6 +68,7 @@ static av_cold int mpc7_decode_init(AVCodecContext * avctx)
     memset(c->oldDSCF, 0, sizeof(c->oldDSCF));
     av_lfg_init(&c->rnd, 0xDEADBEEF);
     dsputil_init(&c->dsp, avctx);
+    ff_mpadsp_init(&c->mpadsp);
     c->dsp.bswap_buf((uint32_t*)buf, (const uint32_t*)avctx->extradata, 4);
     ff_mpc_init();
     init_get_bits(&gb, buf, 128);

libavcodec/mpc8.c

@@ -29,7 +29,7 @@
 #include "avcodec.h"
 #include "get_bits.h"
 #include "dsputil.h"
-#include "mpegaudio.h"
+#include "mpegaudiodsp.h"
 #include "libavutil/audioconvert.h"
 
 #include "mpc.h"
@@ -120,6 +120,7 @@ static av_cold int mpc8_decode_init(AVCodecContext * avctx)
     memset(c->oldDSCF, 0, sizeof(c->oldDSCF));
     av_lfg_init(&c->rnd, 0xDEADBEEF);
     dsputil_init(&c->dsp, avctx);
+    ff_mpadsp_init(&c->mpadsp);
 
     ff_mpc_init();
 

libavcodec/mpegaudio.h

@@ -31,9 +31,6 @@
 #endif
 
 #include "avcodec.h"
-#include "get_bits.h"
-#include "dsputil.h"
-#include "dct.h"
 
 /* max frame size, in samples */
 #define MPA_FRAME_SIZE 1152
@@ -50,10 +47,6 @@
 #define MPA_DUAL    2
 #define MPA_MONO    3
 
-/* header + layer + bitrate + freq + lsf/mpeg25 */
-#define SAME_HEADER_MASK \
-   (0xffe00000 | (3 << 17) | (0xf << 12) | (3 << 10) | (3 << 19))
-
 #define MP3_MASK 0xFFFE0CCF
 
 #ifndef FRAC_BITS
@@ -65,47 +58,20 @@
 
 #define FIX(a)   ((int)((a) * FRAC_ONE))
 
-#if CONFIG_FLOAT
-typedef float OUT_INT;
-#else
-typedef int16_t OUT_INT;
-#define OUT_SHIFT (WFRAC_BITS + FRAC_BITS - 15)
-#endif
-
 #if CONFIG_FLOAT
 #   define INTFLOAT float
 typedef float MPA_INT;
+typedef float OUT_INT;
 #elif FRAC_BITS <= 15
 #   define INTFLOAT int
 typedef int16_t MPA_INT;
+typedef int16_t OUT_INT;
 #else
 #   define INTFLOAT int
 typedef int32_t MPA_INT;
+typedef int16_t OUT_INT;
 #endif
 
-#define BACKSTEP_SIZE 512
-#define EXTRABYTES 24
-
-/* layer 3 "granule" */
-typedef struct GranuleDef {
-    uint8_t scfsi;
-    int part2_3_length;
-    int big_values;
-    int global_gain;
-    int scalefac_compress;
-    uint8_t block_type;
-    uint8_t switch_point;
-    int table_select[3];
-    int subblock_gain[3];
-    uint8_t scalefac_scale;
-    uint8_t count1table_select;
-    int region_size[3]; /* number of huffman codes in each region */
-    int preflag;
-    int short_start, long_end; /* long/short band indexes */
-    uint8_t scale_factors[40];
-    INTFLOAT sb_hybrid[SBLIMIT * 18]; /* 576 samples */
-} GranuleDef;
-
 #define MPA_DECODE_HEADER \
     int frame_size; \
     int error_protection; \
@@ -122,58 +88,8 @@ typedef struct MPADecodeHeader {
   MPA_DECODE_HEADER
 } MPADecodeHeader;
 
-typedef struct MPADecodeContext {
-    MPA_DECODE_HEADER
-    uint8_t last_buf[2*BACKSTEP_SIZE + EXTRABYTES];
-    int last_buf_size;
-    /* next header (used in free format parsing) */
-    uint32_t free_format_next_header;
-    GetBitContext gb;
-    GetBitContext in_gb;
-    DECLARE_ALIGNED(16, MPA_INT, synth_buf)[MPA_MAX_CHANNELS][512 * 2];
-    int synth_buf_offset[MPA_MAX_CHANNELS];
-    DECLARE_ALIGNED(16, INTFLOAT, sb_samples)[MPA_MAX_CHANNELS][36][SBLIMIT];
-    INTFLOAT mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
-    GranuleDef granules[2][2]; /* Used in Layer 3 */
-#ifdef DEBUG
-    int frame_count;
-#endif
-    int adu_mode; ///< 0 for standard mp3, 1 for adu formatted mp3
-    int dither_state;
-    int error_recognition;
-    AVCodecContext* avctx;
-#if CONFIG_FLOAT
-    DCTContext dct;
-#endif
-    void (*apply_window_mp3)(MPA_INT *synth_buf, MPA_INT *window,
-                             int *dither_state, OUT_INT *samples, int incr);
-} MPADecodeContext;
-
-/* layer 3 huffman tables */
-typedef struct HuffTable {
-    int xsize;
-    const uint8_t *bits;
-    const uint16_t *codes;
-} HuffTable;
-
 int ff_mpa_l2_select_table(int bitrate, int nb_channels, int freq, int lsf);
 int ff_mpa_decode_header(AVCodecContext *avctx, uint32_t head, int *sample_rate, int *channels, int *frame_size, int *bitrate);
-extern MPA_INT ff_mpa_synth_window_fixed[];
-void ff_mpa_synth_init_fixed(MPA_INT *window);
-void ff_mpa_synth_filter_fixed(MPA_INT *synth_buf_ptr, int *synth_buf_offset,
-                         MPA_INT *window, int *dither_state,
-                         OUT_INT *samples, int incr,
-                         INTFLOAT sb_samples[SBLIMIT]);
-
-void ff_mpa_synth_init_float(MPA_INT *window);
-void ff_mpa_synth_filter_float(MPADecodeContext *s,
-                         MPA_INT *synth_buf_ptr, int *synth_buf_offset,
-                         MPA_INT *window, int *dither_state,
-                         OUT_INT *samples, int incr,
-                         INTFLOAT sb_samples[SBLIMIT]);
-
-void ff_mpegaudiodec_init_mmx(MPADecodeContext *s);
-void ff_mpegaudiodec_init_altivec(MPADecodeContext *s);
 
 /* fast header check for resync */
 static inline int ff_mpa_check_header(uint32_t header){