Merge remote-tracking branch 'qatar/master'

* qatar/master: (26 commits)
  avconv: deprecate the -deinterlace option
  doc: Fix the name of the new function
  aacenc: make sure to encode enough frames to cover all input samples.
  aacenc: only use the number of input samples provided by the user.
  wmadec: Verify bitstream size makes sense before calling init_get_bits.
  kmvc: Log into a context at a log level constant.
  mpeg12: Pad framerate tab to 16 entries.
  kgv1dec: Increase offsets array size so it is large enough.
  kmvc: Check palsize.
  nsvdec: Propagate errors
  nsvdec: Be more careful with av_malloc().
  nsvdec: Fix use of uninitialized streams.
  movenc: cosmetics: Get rid of camelCase identifiers
  swscale: more generic check for planar destination formats with alpha
  doc: Document mov/mp4 fragmentation options
  build: Use order-only prerequisites for creating FATE reference file dirs.
  x86 dsputil: provide SSE2/SSSE3 versions of bswap_buf
  rtsp: Remove some unused variables from ff_rtsp_connect().
  avutil: make intfloat api public
  avformat_write_header(): detail error message
  ...

Conflicts:
	doc/APIchanges
	doc/ffmpeg.texi
	doc/muxers.texi
	ffmpeg.c
	libavcodec/kmvc.c
	libavcodec/x86/Makefile
	libavcodec/x86/dsputil_yasm.asm
	libavcodec/x86/pngdsp-init.c
	libavformat/movenc.c
	libavformat/movenc.h
	libavformat/mpegtsenc.c
	libavformat/nsvdec.c
	libavformat/utils.c
	libavutil/avutil.h
	libswscale/swscale.c

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

10 files changed, 347 insertions, 122 deletions

diff --git a/libavcodec/aacenc.c b/libavcodec/aacenc.c
index 2ff6f9cc04..51c2fa8662 100644
--- a/libavcodec/aacenc.c
+++ b/libavcodec/aacenc.c
@@ -223,8 +223,9 @@ WINDOW_FUNC(eight_short)
     const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
     const float *in = audio + 448;
     float *out = sce->ret;
+    int w;
 
-    for (int w = 0; w < 8; w++) {
+    for (w = 0; w < 8; w++) {
         dsp->vector_fmul        (out, in, w ? pwindow : swindow, 128);
         out += 128;
         in  += 128;
@@ -476,7 +477,7 @@ static void put_bitstream_info(AVCodecContext *avctx, AACEncContext *s,
  * Channels are reordered from Libav's default order to AAC order.
  */
 static void deinterleave_input_samples(AACEncContext *s,
-                                       const float *samples)
+                                       const float *samples, int nb_samples)
 {
     int ch, i;
     const int sinc = s->channels;
@@ -490,10 +491,12 @@ static void deinterleave_input_samples(AACEncContext *s,
         memcpy(&s->planar_samples[ch][1024], &s->planar_samples[ch][2048], 1024 * sizeof(s->planar_samples[0][0]));
 
         /* deinterleave */
-        for (i = 2048; i < 3072; i++) {
+        for (i = 2048; i < 2048 + nb_samples; i++) {
             s->planar_samples[ch][i] = *sptr;
             sptr += sinc;
         }
+        memset(&s->planar_samples[ch][i], 0,
+               (3072 - i) * sizeof(s->planar_samples[0][0]));
     }
 }
 
@@ -507,14 +510,12 @@ static int aac_encode_frame(AVCodecContext *avctx,
     int chan_el_counter[4];
     FFPsyWindowInfo windows[AAC_MAX_CHANNELS];
 
-    if (s->last_frame)
+    if (s->last_frame == 2)
         return 0;
 
-    if (data) {
-        deinterleave_input_samples(s, data);
-        if (s->psypp)
-            ff_psy_preprocess(s->psypp, s->planar_samples, s->channels);
-    }
+    deinterleave_input_samples(s, data, data ? avctx->frame_size : 0);
+    if (s->psypp)
+        ff_psy_preprocess(s->psypp, s->planar_samples, s->channels);
 
     if (!avctx->frame_number)
         return 0;
@@ -645,7 +646,7 @@ static int aac_encode_frame(AVCodecContext *avctx,
     }
 
     if (!data)
-        s->last_frame = 1;
+        s->last_frame++;
 
     return put_bits_count(&s->pb)>>3;
 }
@@ -686,11 +687,12 @@ static av_cold int dsp_init(AVCodecContext *avctx, AACEncContext *s)
 
 static av_cold int alloc_buffers(AVCodecContext *avctx, AACEncContext *s)
 {
+    int ch;
     FF_ALLOCZ_OR_GOTO(avctx, s->buffer.samples, 3 * 1024 * s->channels * sizeof(s->buffer.samples[0]), alloc_fail);
     FF_ALLOCZ_OR_GOTO(avctx, s->cpe, sizeof(ChannelElement) * s->chan_map[0], alloc_fail);
     FF_ALLOCZ_OR_GOTO(avctx, avctx->extradata, 5 + FF_INPUT_BUFFER_PADDING_SIZE, alloc_fail);
 
-    for(int ch = 0; ch < s->channels; ch++)
+    for(ch = 0; ch < s->channels; ch++)
         s->planar_samples[ch] = s->buffer.samples + 3 * 1024 * ch;
 
     return 0;
diff --git a/libavcodec/kmvc.c b/libavcodec/kmvc.c
index 9c98badbde..6d5af5d657 100644
--- a/libavcodec/kmvc.c
+++ b/libavcodec/kmvc.c
@@ -33,6 +33,7 @@
 #define KMVC_KEYFRAME 0x80
 #define KMVC_PALETTE  0x40
 #define KMVC_METHOD   0x0F
+#define MAX_PALSIZE   256
 
 /*
  * Decoder context
@@ -43,7 +44,7 @@ typedef struct KmvcContext {
 
     int setpal;
     int palsize;
-    uint32_t pal[256];
+    uint32_t pal[MAX_PALSIZE];
     uint8_t *cur, *prev;
     uint8_t *frm0, *frm1;
     GetByteContext g;
@@ -376,14 +377,15 @@ static av_cold int decode_init(AVCodecContext * avctx)
     }
 
     if (avctx->extradata_size < 12) {
-        av_log(NULL, 0, "Extradata missing, decoding may not work properly...\n");
+        av_log(avctx, AV_LOG_WARNING,
+               "Extradata missing, decoding may not work properly...\n");
         c->palsize = 127;
     } else {
         c->palsize = AV_RL16(avctx->extradata + 10);
-        if (c->palsize > 255U) {
+        if (c->palsize >= (unsigned)MAX_PALSIZE) {
             c->palsize = 127;
-            av_log(NULL, AV_LOG_ERROR, "palsize too big\n");
-            return -1;
+            av_log(avctx, AV_LOG_ERROR, "KMVC palette too large\n");
+            return AVERROR_INVALIDDATA;
         }
     }
 
diff --git a/libavcodec/mpc7.c b/libavcodec/mpc7.c
index 0bde329e08..c60a621c65 100644
--- a/libavcodec/mpc7.c
+++ b/libavcodec/mpc7.c
@@ -53,7 +53,7 @@ static av_cold int mpc7_decode_init(AVCodecContext * avctx)
     int i, j;
     MPCContext *c = avctx->priv_data;
     GetBitContext gb;
-    uint8_t buf[16];
+    LOCAL_ALIGNED_16(uint8_t, buf, [16]);
     static int vlc_initialized = 0;
 
     static VLC_TYPE scfi_table[1 << MPC7_SCFI_BITS][2];
diff --git a/libavcodec/mpeg12data.c b/libavcodec/mpeg12data.c
index c40883692b..309ec4efd6 100644
--- a/libavcodec/mpeg12data.c
+++ b/libavcodec/mpeg12data.c
@@ -305,7 +305,7 @@ const uint8_t ff_mpeg12_mbMotionVectorTable[17][2] = {
 { 0xc, 10 },
 };
 
-const AVRational avpriv_frame_rate_tab[] = {
+const AVRational avpriv_frame_rate_tab[16] = {
     {    0,    0},
     {24000, 1001},
     {   24,    1},
diff --git a/libavcodec/wmadec.c b/libavcodec/wmadec.c
index 0b2e49981d..7d7cc7f7bf 100644
--- a/libavcodec/wmadec.c
+++ b/libavcodec/wmadec.c
@@ -891,6 +891,8 @@ static int wma_decode_superframe(AVCodecContext *avctx, void *data,
 
         /* read each frame starting from bit_offset */
         pos = bit_offset + 4 + 4 + s->byte_offset_bits + 3;
+        if (pos >= MAX_CODED_SUPERFRAME_SIZE * 8)
+            return AVERROR_INVALIDDATA;
         init_get_bits(&s->gb, buf + (pos >> 3), (MAX_CODED_SUPERFRAME_SIZE - (pos >> 3))*8);
         len = pos & 7;
         if (len > 0)
diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile
index 5ba3942933..282bc916bd 100644
--- a/libavcodec/x86/Makefile
+++ b/libavcodec/x86/Makefile
@@ -48,6 +48,7 @@ MMX-OBJS-$(CONFIG_GPL)                 += x86/idct_mmx.o
 MMX-OBJS-$(CONFIG_LPC)                 += x86/lpc_mmx.o
 YASM-OBJS-$(CONFIG_PRORES_LGPL_DECODER)     += x86/proresdsp.o
 MMX-OBJS-$(CONFIG_PRORES_LGPL_DECODER)      += x86/proresdsp-init.o
+YASM-OBJS-$(CONFIG_PNG_DECODER)        += x86/pngdsp.o
 MMX-OBJS-$(CONFIG_PNG_DECODER)         += x86/pngdsp-init.o
 YASM-OBJS-$(CONFIG_PRORES_DECODER)     += x86/proresdsp.o
 MMX-OBJS-$(CONFIG_PRORES_DECODER)      += x86/proresdsp-init.o
diff --git a/libavcodec/x86/dsputil_mmx.c b/libavcodec/x86/dsputil_mmx.c
index ad29029528..4402995da5 100644
--- a/libavcodec/x86/dsputil_mmx.c
+++ b/libavcodec/x86/dsputil_mmx.c
@@ -2392,6 +2392,9 @@ void ff_apply_window_int16_ssse3     (int16_t *output, const int16_t *input,
 void ff_apply_window_int16_ssse3_atom(int16_t *output, const int16_t *input,
                                       const int16_t *window, unsigned int len);
 
+void ff_bswap32_buf_ssse3(uint32_t *dst, const uint32_t *src, int w);
+void ff_bswap32_buf_sse2(uint32_t *dst, const uint32_t *src, int w);
+
 void ff_add_hfyu_median_prediction_mmx2(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
 int  ff_add_hfyu_left_prediction_ssse3(uint8_t *dst, const uint8_t *src, int w, int left);
 int  ff_add_hfyu_left_prediction_sse4(uint8_t *dst, const uint8_t *src, int w, int left);
@@ -2880,6 +2883,7 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
                     c->apply_window_int16 = ff_apply_window_int16_sse2;
                 }
             }
+            c->bswap_buf = ff_bswap32_buf_sse2;
 #endif
         }
         if (mm_flags & AV_CPU_FLAG_SSSE3) {
@@ -2892,6 +2896,7 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
             if (!(mm_flags & (AV_CPU_FLAG_SSE42|AV_CPU_FLAG_3DNOW))) { // cachesplit
                 c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_ssse3;
             }
+            c->bswap_buf = ff_bswap32_buf_ssse3;
 #endif
         }
 
diff --git a/libavcodec/x86/dsputil_yasm.asm b/libavcodec/x86/dsputil_yasm.asm
index 6c52ac1ffb..ad9ec2c339 100644
--- a/libavcodec/x86/dsputil_yasm.asm
+++ b/libavcodec/x86/dsputil_yasm.asm
@@ -30,6 +30,7 @@ pb_zzzz3333zzzzbbbb: db -1,-1,-1,-1,3,3,3,3,-1,-1,-1,-1,11,11,11,11
 pb_zz11zz55zz99zzdd: db -1,-1,1,1,-1,-1,5,5,-1,-1,9,9,-1,-1,13,13
 pb_revwords: db 14, 15, 12, 13, 10, 11, 8, 9, 6, 7, 4, 5, 2, 3, 0, 1
 pd_16384: times 4 dd 16384
+pb_bswap32: db 3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12
 
 SECTION_TEXT
 
@@ -1180,3 +1181,126 @@ BUTTERFLIES_FLOAT_INTERLEAVE
 INIT_YMM avx
 BUTTERFLIES_FLOAT_INTERLEAVE
 %endif
+
+INIT_XMM sse2
+; %1 = aligned/unaligned
+%macro BSWAP_LOOPS_SSE2  1
+    mov      r3, r2
+    sar      r2, 3
+    jz       .left4_%1
+.loop8_%1:
+    mov%1    m0, [r1 +  0]
+    mov%1    m1, [r1 + 16]
+    pshuflw  m0, m0, 10110001b
+    pshuflw  m1, m1, 10110001b
+    pshufhw  m0, m0, 10110001b
+    pshufhw  m1, m1, 10110001b
+    mova     m2, m0
+    mova     m3, m1
+    psllw    m0, 8
+    psllw    m1, 8
+    psrlw    m2, 8
+    psrlw    m3, 8
+    por      m2, m0
+    por      m3, m1
+    mova     [r0 +  0], m2
+    mova     [r0 + 16], m3
+    add      r1, 32
+    add      r0, 32
+    dec      r2
+    jnz      .loop8_%1
+.left4_%1:
+    mov      r2, r3
+    and      r3, 4
+    jz       .left
+    mov%1    m0, [r1]
+    pshuflw  m0, m0, 10110001b
+    pshufhw  m0, m0, 10110001b
+    mova     m2, m0
+    psllw    m0, 8
+    psrlw    m2, 8
+    por      m2, m0
+    mova     [r0], m2
+    add      r1, 16
+    add      r0, 16
+%endmacro
+
+; void bswap_buf(uint32_t *dst, const uint32_t *src, int w);
+cglobal bswap32_buf, 3,4,5
+    mov      r3, r1
+    and      r3, 15
+    jz       .start_align
+    BSWAP_LOOPS_SSE2  u
+    jmp      .left
+.start_align:
+    BSWAP_LOOPS_SSE2  a
+.left:
+    and      r2, 3
+    jz       .end
+.loop2:
+    mov      r3d, [r1]
+    bswap    r3d
+    mov      [r0], r3d
+    add      r1, 4
+    add      r0, 4
+    dec      r2
+    jnz      .loop2
+.end
+    RET
+
+; %1 = aligned/unaligned
+%macro BSWAP_LOOPS_SSSE3  1
+    mov      r3, r2
+    sar      r2, 3
+    jz       .left4_%1
+.loop8_%1:
+    mov%1    m0, [r1 +  0]
+    mov%1    m1, [r1 + 16]
+    pshufb   m0, m2
+    pshufb   m1, m2
+    mova     [r0 +  0], m0
+    mova     [r0 + 16], m1
+    add      r0, 32
+    add      r1, 32
+    dec      r2
+    jnz      .loop8_%1
+.left4_%1:
+    mov      r2, r3
+    and      r3, 4
+    jz       .left2
+    mov%1    m0, [r1]
+    pshufb   m0, m2
+    mova     [r0], m0
+    add      r1, 16
+    add      r0, 16
+%endmacro
+
+INIT_XMM ssse3
+; void bswap_buf(uint32_t *dst, const uint32_t *src, int w);
+cglobal bswap32_buf, 3,4,3
+    mov      r3, r1
+    mova     m2, [pb_bswap32]
+    and      r3, 15
+    jz       .start_align
+    BSWAP_LOOPS_SSSE3  u
+    jmp      .left2
+.start_align:
+    BSWAP_LOOPS_SSSE3  a
+.left2:
+    mov      r3, r2
+    and      r2, 2
+    jz       .left1
+    movq     m0, [r1]
+    pshufb   m0, m2
+    movq     [r0], m0
+    add      r1, 8
+    add      r0, 8
+.left1:
+    and      r3, 1
+    jz       .end
+    mov      r2d, [r1]
+    bswap    r2d
+    mov      [r0], r2d
+.end:
+    RET
+
diff --git a/libavcodec/x86/pngdsp-init.c b/libavcodec/x86/pngdsp-init.c
index 089147c9e4..f122b242fb 100644
--- a/libavcodec/x86/pngdsp-init.c
+++ b/libavcodec/x86/pngdsp-init.c
@@ -19,118 +19,33 @@
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
+#include "libavutil/common.h"
 #include "libavutil/cpu.h"
-#include "libavutil/x86_cpu.h"
 #include "libavcodec/pngdsp.h"
-#include "dsputil_mmx.h"
 
-#define PAETH(cpu, abs3)\
-static void add_png_paeth_prediction_##cpu(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp)\
-{\
-    x86_reg i, end;\
-    if(bpp>4) add_png_paeth_prediction_##cpu(dst+bpp/2, src+bpp/2, top+bpp/2, w-bpp/2, -bpp);\
-    if(bpp<0) bpp=-bpp;\
-    i= -bpp;\
-    end = w-3;\
-    __asm__ volatile(\
-        "pxor      %%mm7, %%mm7 \n"\
-        "movd    (%1,%0), %%mm0 \n"\
-        "movd    (%2,%0), %%mm1 \n"\
-        "punpcklbw %%mm7, %%mm0 \n"\
-        "punpcklbw %%mm7, %%mm1 \n"\
-        "add       %4, %0 \n"\
-        "1: \n"\
-        "movq      %%mm1, %%mm2 \n"\
-        "movd    (%2,%0), %%mm1 \n"\
-        "movq      %%mm2, %%mm3 \n"\
-        "punpcklbw %%mm7, %%mm1 \n"\
-        "movq      %%mm2, %%mm4 \n"\
-        "psubw     %%mm1, %%mm3 \n"\
-        "psubw     %%mm0, %%mm4 \n"\
-        "movq      %%mm3, %%mm5 \n"\
-        "paddw     %%mm4, %%mm5 \n"\
-        abs3\
-        "movq      %%mm4, %%mm6 \n"\
-        "pminsw    %%mm5, %%mm6 \n"\
-        "pcmpgtw   %%mm6, %%mm3 \n"\
-        "pcmpgtw   %%mm5, %%mm4 \n"\
-        "movq      %%mm4, %%mm6 \n"\
-        "pand      %%mm3, %%mm4 \n"\
-        "pandn     %%mm3, %%mm6 \n"\
-        "pandn     %%mm0, %%mm3 \n"\
-        "movd    (%3,%0), %%mm0 \n"\
-        "pand      %%mm1, %%mm6 \n"\
-        "pand      %%mm4, %%mm2 \n"\
-        "punpcklbw %%mm7, %%mm0 \n"\
-        "paddw     %%mm6, %%mm0 \n"\
-        "paddw     %%mm2, %%mm3 \n"\
-        "paddw     %%mm3, %%mm0 \n"\
-        "pand      %6   , %%mm0 \n"\
-        "movq      %%mm0, %%mm3 \n"\
-        "packuswb  %%mm3, %%mm3 \n"\
-        "movd      %%mm3, (%1,%0) \n"\
-        "add       %4, %0 \n"\
-        "cmp       %5, %0 \n"\
-        "jle 1b \n"\
-        :"+r"(i)\
-        :"r"(dst), "r"(top), "r"(src), "r"((x86_reg)bpp), "g"(end),\
-         "m"(ff_pw_255)\
-        :"memory"\
-    );\
-}
-
-#define ABS3_MMX2\
-        "psubw     %%mm5, %%mm7 \n"\
-        "pmaxsw    %%mm7, %%mm5 \n"\
-        "pxor      %%mm6, %%mm6 \n"\
-        "pxor      %%mm7, %%mm7 \n"\
-        "psubw     %%mm3, %%mm6 \n"\
-        "psubw     %%mm4, %%mm7 \n"\
-        "pmaxsw    %%mm6, %%mm3 \n"\
-        "pmaxsw    %%mm7, %%mm4 \n"\
-        "pxor      %%mm7, %%mm7 \n"
-
-#define ABS3_SSSE3\
-        "pabsw     %%mm3, %%mm3 \n"\
-        "pabsw     %%mm4, %%mm4 \n"\
-        "pabsw     %%mm5, %%mm5 \n"
-
-PAETH(mmx2, ABS3_MMX2)
-#if HAVE_SSSE3
-PAETH(ssse3, ABS3_SSSE3)
-#endif
-
-static void add_bytes_l2_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w)
-{
-    x86_reg i=0;
-    __asm__ volatile(
-        "jmp 2f                         \n\t"
-        "1:                             \n\t"
-        "movq   (%2, %0), %%mm0         \n\t"
-        "movq  8(%2, %0), %%mm1         \n\t"
-        "paddb  (%3, %0), %%mm0         \n\t"
-        "paddb 8(%3, %0), %%mm1         \n\t"
-        "movq %%mm0,  (%1, %0)          \n\t"
-        "movq %%mm1, 8(%1, %0)          \n\t"
-        "add $16, %0                    \n\t"
-        "2:                             \n\t"
-        "cmp %4, %0                     \n\t"
-        " js 1b                         \n\t"
-        : "+r" (i)
-        : "r"(dst), "r"(src1), "r"(src2), "r"((x86_reg) w - 15)
-    );
-    for (; i < w; i++)
-        dst[i] = src1[i] + src2[i];
-}
+void ff_add_png_paeth_prediction_mmx2 (uint8_t *dst, uint8_t *src,
+                                       uint8_t *top, int w, int bpp);
+void ff_add_png_paeth_prediction_ssse3(uint8_t *dst, uint8_t *src,
+                                       uint8_t *top, int w, int bpp);
+void ff_add_bytes_l2_mmx (uint8_t *dst, uint8_t *src1,
+                          uint8_t *src2, int w);
+void ff_add_bytes_l2_sse2(uint8_t *dst, uint8_t *src1,
+                          uint8_t *src2, int w);
 
 void ff_pngdsp_init_x86(PNGDSPContext *dsp)
 {
+#if HAVE_YASM
     int flags = av_get_cpu_flags();
 
+#if ARCH_X86_32
     if (flags & AV_CPU_FLAG_MMX)
-        dsp->add_bytes_l2         = add_bytes_l2_mmx;
+        dsp->add_bytes_l2         = ff_add_bytes_l2_mmx;
+#endif
     if (flags & AV_CPU_FLAG_MMX2)
-        dsp->add_paeth_prediction = add_png_paeth_prediction_mmx2;
-    if (HAVE_SSSE3 && flags & AV_CPU_FLAG_SSSE3)
-        dsp->add_paeth_prediction = add_png_paeth_prediction_ssse3;
+        dsp->add_paeth_prediction = ff_add_png_paeth_prediction_mmx2;
+    if (flags & AV_CPU_FLAG_SSE2)
+        dsp->add_bytes_l2         = ff_add_bytes_l2_sse2;
+    if (flags & AV_CPU_FLAG_SSSE3)
+        dsp->add_paeth_prediction = ff_add_png_paeth_prediction_ssse3;
+#endif
 }
diff --git a/libavcodec/x86/pngdsp.asm b/libavcodec/x86/pngdsp.asm
new file mode 100644
index 0000000000..8c4cb19da0
--- /dev/null
+++ b/libavcodec/x86/pngdsp.asm
@@ -0,0 +1,174 @@
+;******************************************************************************
+;* x86 optimizations for PNG decoding
+;*
+;* Copyright (c) 2008 Loren Merritt <lorenm@u.washington.edu>
+;* Copyright (c) 2012 Ronald S. Bultje <rsbultje@gmail.com>
+;*
+;* This file is part of Libav.
+;*
+;* Libav is free software; you can redistribute it and/or
+;* modify it under the terms of the GNU Lesser General Public
+;* License as published by the Free Software Foundation; either
+;* version 2.1 of the License, or (at your option) any later version.
+;*
+;* Libav is distributed in the hope that it will be useful,
+;* but WITHOUT ANY WARRANTY; without even the implied warranty of
+;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+;* Lesser General Public License for more details.
+;*
+;* You should have received a copy of the GNU Lesser General Public
+;* License along with Libav; if not, write to the Free Software
+;* 51, Inc., Foundation Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+;******************************************************************************
+
+%include "x86inc.asm"
+%include "x86util.asm"
+
+SECTION_RODATA
+
+cextern pw_255
+
+section .text align=16
+
+; %1 = nr. of xmm registers used
+%macro ADD_BYTES_FN 1
+cglobal add_bytes_l2, 4, 6, %1, dst, src1, src2, wa, w, i
+%if ARCH_X86_64
+    movsxd             waq, wad
+%endif
+    xor                 iq, iq
+
+    ; vector loop
+    mov                 wq, waq
+    and                waq, ~(mmsize*2-1)
+    jmp .end_v
+.loop_v:
+    mova                m0, [src1q+iq]
+    mova                m1, [src1q+iq+mmsize]
+    paddb               m0, [src2q+iq]
+    paddb               m1, [src2q+iq+mmsize]
+    mova  [dstq+iq       ], m0
+    mova  [dstq+iq+mmsize], m1
+    add                 iq, mmsize*2
+.end_v:
+    cmp                 iq, waq
+    jl .loop_v
+
+%if mmsize == 16
+    ; vector loop
+    mov                 wq, waq
+    and                waq, ~7
+    jmp .end_l
+.loop_l:
+    movq               mm0, [src1q+iq]
+    paddb              mm0, [src2q+iq]
+    movq  [dstq+iq       ], mm0
+    add                 iq, 8
+.end_l:
+    cmp                 iq, waq
+    jl .loop_l
+%endif
+
+    ; scalar loop for leftover
+    jmp .end_s
+.loop_s:
+    mov                wab, [src1q+iq]
+    add                wab, [src2q+iq]
+    mov          [dstq+iq], wab
+    inc                 iq
+.end_s:
+    cmp                 iq, wq
+    jl .loop_s
+    REP_RET
+%endmacro
+
+%if ARCH_X86_32
+INIT_MMX mmx
+ADD_BYTES_FN 0
+%endif
+
+INIT_XMM sse2
+ADD_BYTES_FN 2
+
+%macro ADD_PAETH_PRED_FN 1
+cglobal add_png_paeth_prediction, 5, 7, %1, dst, src, top, w, bpp, end, cntr
+%if ARCH_X86_64
+    movsxd            bppq, bppd
+    movsxd              wq, wd
+%endif
+    lea               endq, [dstq+wq-(mmsize/2-1)]
+    sub               topq, dstq
+    sub               srcq, dstq
+    sub               dstq, bppq
+    pxor                m7, m7
+
+    PUSH              dstq
+    lea              cntrq, [bppq-1]
+    shr              cntrq, 2 + mmsize/16
+.bpp_loop:
+    lea               dstq, [dstq+cntrq*(mmsize/2)]
+    movh                m0, [dstq]
+    movh                m1, [topq+dstq]
+    punpcklbw           m0, m7
+    punpcklbw           m1, m7
+    add               dstq, bppq
+.loop:
+    mova                m2, m1
+    movh                m1, [topq+dstq]
+    mova                m3, m2
+    punpcklbw           m1, m7
+    mova                m4, m2
+    psubw               m3, m1
+    psubw               m4, m0
+    mova                m5, m3
+    paddw               m5, m4
+%if cpuflag(ssse3)
+    pabsw               m3, m3
+    pabsw               m4, m4
+    pabsw               m5, m5
+%else ; !cpuflag(ssse3)
+    psubw               m7, m5
+    pmaxsw              m5, m7
+    pxor                m6, m6
+    pxor                m7, m7
+    psubw               m6, m3
+    psubw               m7, m4
+    pmaxsw              m3, m6
+    pmaxsw              m4, m7
+    pxor                m7, m7
+%endif ; cpuflag(ssse3)
+    mova                m6, m4
+    pminsw              m6, m5
+    pcmpgtw             m3, m6
+    pcmpgtw             m4, m5
+    mova                m6, m4
+    pand                m4, m3
+    pandn               m6, m3
+    pandn               m3, m0
+    movh                m0, [srcq+dstq]
+    pand                m6, m1
+    pand                m2, m4
+    punpcklbw           m0, m7
+    paddw               m0, m6
+    paddw               m3, m2
+    paddw               m0, m3
+    pand                m0, [pw_255]
+    mova                m3, m0
+    packuswb            m3, m3
+    movh            [dstq], m3
+    add               dstq, bppq
+    cmp               dstq, endq
+    jle .loop
+
+    mov               dstq, [rsp]
+    dec              cntrq
+    jge .bpp_loop
+    POP               dstq
+    RET
+%endmacro
+
+INIT_MMX mmx2
+ADD_PAETH_PRED_FN 0
+
+INIT_MMX ssse3
+ADD_PAETH_PRED_FN 0