Merge remote-tracking branch 'qatar/master'

* qatar/master:
  xan: Convert to bytestream2
  oggenc: add a page_duration option and deprecate the pagesize option
  x86: lavr: add SSE2/AVX dither_int_to_float()

Conflicts:
	libavcodec/xan.c
	libavformat/oggenc.c

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

5 files changed, 174 insertions, 34 deletions

diff --git a/doc/muxers.texi b/doc/muxers.texi
index fb642847e1..2f62d6351d 100644
--- a/doc/muxers.texi
+++ b/doc/muxers.texi
@@ -717,4 +717,19 @@ ffmpeg -i input.mp3 -i cover.png -c copy -metadata:s:v title="Album cover"
 -metadata:s:v comment="Cover (Front)" out.mp3
 @end example
 
+@section ogg
+
+Ogg container muxer.
+
+@table @option
+@item -page_duration @var{duration}
+Preferred page duration, in microseconds. The muxer will attempt to create
+pages that are approximately @var{duration} microseconds long. This allows the
+user to compromise between seek granularity and container overhead. The default
+is 1 second. A value of 0 will fill all segments, making pages as large as
+possible. A value of 1 will effectively use 1 packet-per-page in most
+situations, giving a small seek granularity at the cost of additional container
+overhead.
+@end table
+
 @c man end MUXERS
diff --git a/libavcodec/xan.c b/libavcodec/xan.c
index 3a559e929b..219eedd92d 100644
--- a/libavcodec/xan.c
+++ b/libavcodec/xan.c
@@ -142,44 +142,47 @@ static void xan_unpack(unsigned char *dest, int dest_len,
     int size;
     unsigned char *dest_org = dest;
     unsigned char *dest_end = dest + dest_len;
-    const unsigned char *src_end = src + src_len;
+    GetByteContext ctx;
 
-    while (dest < dest_end && src < src_end) {
-        opcode = *src++;
+    bytestream2_init(&ctx, src, src_len);
+    while (dest < dest_end && bytestream2_get_bytes_left(&ctx)) {
+        opcode = bytestream2_get_byte(&ctx);
 
         if (opcode < 0xe0) {
             int size2, back;
             if ((opcode & 0x80) == 0) {
                 size = opcode & 3;
 
-                back  = ((opcode & 0x60) << 3) + *src++ + 1;
+                back  = ((opcode & 0x60) << 3) + bytestream2_get_byte(&ctx) + 1;
                 size2 = ((opcode & 0x1c) >> 2) + 3;
             } else if ((opcode & 0x40) == 0) {
-                size = *src >> 6;
+                size = bytestream2_peek_byte(&ctx) >> 6;
 
-                back  = (bytestream_get_be16(&src) & 0x3fff) + 1;
+                back  = (bytestream2_get_be16(&ctx) & 0x3fff) + 1;
                 size2 = (opcode & 0x3f) + 4;
             } else {
                 size = opcode & 3;
 
-                back  = ((opcode & 0x10) << 12) + bytestream_get_be16(&src) + 1;
-                size2 = ((opcode & 0x0c) <<  6) + *src++ + 5;
+                back  = ((opcode & 0x10) << 12) + bytestream2_get_be16(&ctx) + 1;
+                size2 = ((opcode & 0x0c) <<  6) + bytestream2_get_byte(&ctx) + 5;
             }
 
             if (dest_end - dest < size + size2 ||
                 dest + size - dest_org < back ||
-                src_end - src < size)
+                bytestream2_get_bytes_left(&ctx) < size)
                 return;
-            memcpy(dest, src, size);  dest += size;  src += size;
+            bytestream2_get_buffer(&ctx, dest, size);
+            dest += size;
             av_memcpy_backptr(dest, back, size2);
             dest += size2;
         } else {
             int finish = opcode >= 0xfc;
             size = finish ? opcode & 3 : ((opcode & 0x1f) << 2) + 4;
 
-            if (dest_end - dest < size || src_end - src < size)
+            if (dest_end - dest < size || bytestream2_get_bytes_left(&ctx) < size)
                 return;
-            memcpy(dest, src, size);  dest += size;  src += size;
+            bytestream2_get_buffer(&ctx, dest, size);
+            dest += size;
             if (finish)
                 return;
         }
@@ -518,21 +521,22 @@ static int xan_decode_frame(AVCodecContext *avctx,
     const uint8_t *buf = avpkt->data;
     int ret, buf_size = avpkt->size;
     XanContext *s = avctx->priv_data;
-    const uint8_t *buf_end = buf + buf_size;
+    GetByteContext ctx;
     int tag = 0;
 
-    while (buf_end - buf > 8 && tag != VGA__TAG) {
+    bytestream2_init(&ctx, buf, buf_size);
+    while (bytestream2_get_bytes_left(&ctx) > 8 && tag != VGA__TAG) {
         unsigned *tmpptr;
         uint32_t new_pal;
         int size;
         int i;
-        tag  = bytestream_get_le32(&buf);
-        size = bytestream_get_be32(&buf);
+        tag  = bytestream2_get_le32(&ctx);
+        size = bytestream2_get_be32(&ctx);
         if(size < 0) {
             av_log(avctx, AV_LOG_ERROR, "Invalid tag size %d\n", size);
             return AVERROR_INVALIDDATA;
         }
-        size = FFMIN(size, buf_end - buf);
+        size = FFMIN(size, bytestream2_get_bytes_left(&ctx));
         switch (tag) {
         case PALT_TAG:
             if (size < PALETTE_SIZE)
@@ -547,13 +551,13 @@ static int xan_decode_frame(AVCodecContext *avctx,
             tmpptr += s->palettes_count * AVPALETTE_COUNT;
             for (i = 0; i < PALETTE_COUNT; i++) {
 #if RUNTIME_GAMMA
-                int r = gamma_corr(*buf++);
-                int g = gamma_corr(*buf++);
-                int b = gamma_corr(*buf++);
+                int r = gamma_corr(bytestream2_get_byteu(&ctx));
+                int g = gamma_corr(bytestream2_get_byteu(&ctx));
+                int b = gamma_corr(bytestream2_get_byteu(&ctx));
 #else
-                int r = gamma_lookup[*buf++];
-                int g = gamma_lookup[*buf++];
-                int b = gamma_lookup[*buf++];
+                int r = gamma_lookup[bytestream2_get_byteu(&ctx)];
+                int g = gamma_lookup[bytestream2_get_byteu(&ctx)];
+                int b = gamma_lookup[bytestream2_get_byteu(&ctx)];
 #endif
                 *tmpptr++ = (0xFFU << 24) | (r << 16) | (g << 8) | b;
             }
@@ -562,7 +566,7 @@ static int xan_decode_frame(AVCodecContext *avctx,
         case SHOT_TAG:
             if (size < 4)
                 return AVERROR_INVALIDDATA;
-            new_pal = bytestream_get_le32(&buf);
+            new_pal = bytestream2_get_le32(&ctx);
             if (new_pal < s->palettes_count) {
                 s->cur_palette = new_pal;
             } else
@@ -571,11 +575,11 @@ static int xan_decode_frame(AVCodecContext *avctx,
         case VGA__TAG:
             break;
         default:
-            buf += size;
+            bytestream2_skip(&ctx, size);
             break;
         }
     }
-    buf_size = buf_end - buf;
+    buf_size = bytestream2_get_bytes_left(&ctx);
 
     if (s->palettes_count <= 0) {
         av_log(s->avctx, AV_LOG_ERROR, "No palette found\n");
@@ -594,7 +598,7 @@ static int xan_decode_frame(AVCodecContext *avctx,
     memcpy(s->current_frame.data[1],
            s->palettes + s->cur_palette * AVPALETTE_COUNT, AVPALETTE_SIZE);
 
-    s->buf = buf;
+    s->buf = ctx.buffer;
     s->size = buf_size;
 
     if (xan_wc3_decode_frame(s) < 0)
diff --git a/libavformat/oggenc.c b/libavformat/oggenc.c
index 2ff6704698..31e28413e9 100644
--- a/libavformat/oggenc.c
+++ b/libavformat/oggenc.c
@@ -35,6 +35,7 @@
 #define MAX_PAGE_SIZE 65025
 
 typedef struct {
+    int64_t start_granule;
     int64_t granule;
     int stream_index;
     uint8_t flags;
@@ -68,6 +69,7 @@ typedef struct {
     const AVClass *class;
     OGGPageList *page_list;
     int pref_size; ///< preferred page size (0 => fill all segments)
+    int64_t pref_duration;      ///< preferred page duration (0 => fill all segments)
 } OGGContext;
 
 #define OFFSET(x) offsetof(OGGContext, x)
@@ -76,8 +78,10 @@ typedef struct {
 static const AVOption options[] = {
     { "oggpagesize", "Set preferred Ogg page size.",
       offsetof(OGGContext, pref_size), AV_OPT_TYPE_INT, {.i64 = 0}, 0, MAX_PAGE_SIZE, AV_OPT_FLAG_ENCODING_PARAM},
-    { "pagesize", "preferred page size in bytes",
+    { "pagesize", "preferred page size in bytes (deprecated)",
         OFFSET(pref_size), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, MAX_PAGE_SIZE, PARAM },
+    { "page_duration", "preferred page duration, in microseconds",
+        OFFSET(pref_duration), AV_OPT_TYPE_INT, { .i64 = 1000000 }, 0, INT64_MAX, PARAM },
     { NULL },
 };
 
@@ -185,6 +189,7 @@ static int ogg_buffer_page(AVFormatContext *s, OGGStreamContext *oggstream)
         return AVERROR(ENOMEM);
     l->page = oggstream->page;
 
+    oggstream->page.start_granule = oggstream->page.granule;
     oggstream->page_count++;
     ogg_reset_cur_page(oggstream);
 
@@ -223,6 +228,12 @@ static int ogg_buffer_data(AVFormatContext *s, AVStream *st,
         flush = 1;
     }
 
+    // avoid a continued page
+    if (!header && oggstream->page.size > 0 &&
+        MAX_PAGE_SIZE - oggstream->page.size < size) {
+        ogg_buffer_page(s, oggstream);
+    }
+
     for (i = 0; i < total_segments; ) {
         OGGPage *page = &oggstream->page;
 
@@ -245,9 +256,19 @@ static int ogg_buffer_data(AVFormatContext *s, AVStream *st,
         if (i == total_segments)
             page->granule = granule;
 
-        if (!header && (page->segments_count == 255 ||
-            (ogg->pref_size > 0 && page->size >= ogg->pref_size))) {
-           ogg_buffer_page(s, oggstream);
+        if (!header) {
+            AVStream *st = s->streams[page->stream_index];
+
+            int64_t start = av_rescale_q(page->start_granule, st->time_base,
+                                         AV_TIME_BASE_Q);
+            int64_t next  = av_rescale_q(page->granule, st->time_base,
+                                         AV_TIME_BASE_Q);
+
+            if (page->segments_count == 255 ||
+                (ogg->pref_size     > 0 && page->size   >= ogg->pref_size) ||
+                (ogg->pref_duration > 0 && next - start >= ogg->pref_duration)) {
+                ogg_buffer_page(s, oggstream);
+            }
         }
     }
 
@@ -380,9 +401,13 @@ static int ogg_build_opus_headers(AVCodecContext *avctx,
 
 static int ogg_write_header(AVFormatContext *s)
 {
+    OGGContext *ogg = s->priv_data;
     OGGStreamContext *oggstream;
     int i, j;
 
+    if (ogg->pref_size)
+        av_log(s, AV_LOG_WARNING, "The pagesize option is deprecated\n");
+
     for (i = 0; i < s->nb_streams; i++) {
         AVStream *st = s->streams[i];
         unsigned serial_num = i;
@@ -502,6 +527,9 @@ static int ogg_write_header(AVFormatContext *s)
         }
         ogg_buffer_page(s, oggstream);
     }
+
+    oggstream->page.start_granule = AV_NOPTS_VALUE;
+
     return 0;
 }
 
@@ -551,6 +579,9 @@ static int ogg_write_packet(AVFormatContext *s, AVPacket *pkt)
     else
         granule = pkt->pts + pkt->duration;
 
+    if (oggstream->page.start_granule == AV_NOPTS_VALUE)
+        oggstream->page.start_granule = pkt->pts;
+
     ret = ogg_buffer_data(s, st, pkt->data, pkt->size, granule, 0);
     if (ret < 0)
         return ret;
@@ -566,9 +597,13 @@ static int ogg_write_trailer(AVFormatContext *s)
 {
     int i;
 
-    /* flush current page */
-    for (i = 0; i < s->nb_streams; i++)
-        ogg_buffer_page(s, s->streams[i]->priv_data);
+    /* flush current page if needed */
+    for (i = 0; i < s->nb_streams; i++) {
+        OGGStreamContext *oggstream = s->streams[i]->priv_data;
+
+        if (oggstream->page.size > 0)
+            ogg_buffer_page(s, oggstream);
+    }
 
     ogg_write_pages(s, 1);
 
diff --git a/libavresample/x86/dither.asm b/libavresample/x86/dither.asm
index bfdc4e6907..757f2800bc 100644
--- a/libavresample/x86/dither.asm
+++ b/libavresample/x86/dither.asm
@@ -23,6 +23,9 @@
 
 SECTION_RODATA 32
 
+; 1.0f / (2.0f * INT32_MAX)
+pf_dither_scale: times 8 dd 2.32830643762e-10
+
 pf_s16_scale: times 4 dd 32753.0
 
 SECTION_TEXT
@@ -51,3 +54,64 @@ cglobal quantize, 4,4,3, dst, src, dither, len
     add         lenq, mmsize
     jl .loop
     REP_RET
+
+;------------------------------------------------------------------------------
+; void ff_dither_int_to_float_rectangular(float *dst, int *src, int len)
+;------------------------------------------------------------------------------
+
+%macro DITHER_INT_TO_FLOAT_RECTANGULAR 0
+cglobal dither_int_to_float_rectangular, 3,3,3, dst, src, len
+    lea         lenq, [4*lend]
+    add         srcq, lenq
+    add         dstq, lenq
+    neg         lenq
+    mova          m0, [pf_dither_scale]
+.loop:
+    cvtdq2ps      m1, [srcq+lenq]
+    cvtdq2ps      m2, [srcq+lenq+mmsize]
+    mulps         m1, m1, m0
+    mulps         m2, m2, m0
+    mova  [dstq+lenq], m1
+    mova  [dstq+lenq+mmsize], m2
+    add         lenq, 2*mmsize
+    jl .loop
+    REP_RET
+%endmacro
+
+INIT_XMM sse2
+DITHER_INT_TO_FLOAT_RECTANGULAR
+INIT_YMM avx
+DITHER_INT_TO_FLOAT_RECTANGULAR
+
+;------------------------------------------------------------------------------
+; void ff_dither_int_to_float_triangular(float *dst, int *src0, int len)
+;------------------------------------------------------------------------------
+
+%macro DITHER_INT_TO_FLOAT_TRIANGULAR 0
+cglobal dither_int_to_float_triangular, 3,4,5, dst, src0, len, src1
+    lea         lenq, [4*lend]
+    lea        src1q, [src0q+2*lenq]
+    add        src0q, lenq
+    add         dstq, lenq
+    neg         lenq
+    mova          m0, [pf_dither_scale]
+.loop:
+    cvtdq2ps      m1, [src0q+lenq]
+    cvtdq2ps      m2, [src0q+lenq+mmsize]
+    cvtdq2ps      m3, [src1q+lenq]
+    cvtdq2ps      m4, [src1q+lenq+mmsize]
+    addps         m1, m1, m3
+    addps         m2, m2, m4
+    mulps         m1, m1, m0
+    mulps         m2, m2, m0
+    mova  [dstq+lenq], m1
+    mova  [dstq+lenq+mmsize], m2
+    add         lenq, 2*mmsize
+    jl .loop
+    REP_RET
+%endmacro
+
+INIT_XMM sse2
+DITHER_INT_TO_FLOAT_TRIANGULAR
+INIT_YMM avx
+DITHER_INT_TO_FLOAT_TRIANGULAR
diff --git a/libavresample/x86/dither_init.c b/libavresample/x86/dither_init.c
index d5d92902fd..6532887c09 100644
--- a/libavresample/x86/dither_init.c
+++ b/libavresample/x86/dither_init.c
@@ -26,6 +26,12 @@
 extern void ff_quantize_sse2(int16_t *dst, const float *src, float *dither,
                              int len);
 
+extern void ff_dither_int_to_float_rectangular_sse2(float *dst, int *src, int len);
+extern void ff_dither_int_to_float_rectangular_avx(float *dst, int *src, int len);
+
+extern void ff_dither_int_to_float_triangular_sse2(float *dst, int *src0, int len);
+extern void ff_dither_int_to_float_triangular_avx(float *dst, int *src0, int len);
+
 av_cold void ff_dither_init_x86(DitherDSPContext *ddsp,
                                 enum AVResampleDitherMethod method)
 {
@@ -36,4 +42,20 @@ av_cold void ff_dither_init_x86(DitherDSPContext *ddsp,
         ddsp->ptr_align     = 16;
         ddsp->samples_align = 8;
     }
+
+    if (method == AV_RESAMPLE_DITHER_RECTANGULAR) {
+        if (EXTERNAL_SSE2(mm_flags)) {
+            ddsp->dither_int_to_float = ff_dither_int_to_float_rectangular_sse2;
+        }
+        if (EXTERNAL_AVX(mm_flags)) {
+            ddsp->dither_int_to_float = ff_dither_int_to_float_rectangular_avx;
+        }
+    } else {
+        if (EXTERNAL_SSE2(mm_flags)) {
+            ddsp->dither_int_to_float = ff_dither_int_to_float_triangular_sse2;
+        }
+        if (EXTERNAL_AVX(mm_flags)) {
+            ddsp->dither_int_to_float = ff_dither_int_to_float_triangular_avx;
+        }
+    }
 }