Merge remote-tracking branch 'qatar/master'

* qatar/master:
  dnxhddec: avoid a branch in 10-bit decode_dct_block()
  H.264: Add optimizations to predict x86 assembly.
  riff: Add mpgv MPEG-2 fourcc
  add Flash Screen Video 2 decoder

Conflicts:
	configure
	doc/general.texi
	libavcodec/Makefile
	libavcodec/allcodecs.c
	libavcodec/version.h

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

8 files changed, 736 insertions, 729 deletions

diff --git a/libavcodec/Makefile b/libavcodec/Makefile
index 8b3d35ab38..9706939aec 100644
--- a/libavcodec/Makefile
+++ b/libavcodec/Makefile
@@ -153,6 +153,7 @@ OBJS-$(CONFIG_FLAC_ENCODER)            += flacenc.o flacdata.o flac.o vorbis_dat
 OBJS-$(CONFIG_FLASHSV_DECODER)         += flashsv.o
 OBJS-$(CONFIG_FLASHSV_ENCODER)         += flashsvenc.o
 OBJS-$(CONFIG_FLASHSV2_ENCODER)        += flashsv2enc.o
+OBJS-$(CONFIG_FLASHSV2_DECODER)        += flashsv.o
 OBJS-$(CONFIG_FLIC_DECODER)            += flicvideo.o
 OBJS-$(CONFIG_FOURXM_DECODER)          += 4xm.o
 OBJS-$(CONFIG_FRAPS_DECODER)           += fraps.o
diff --git a/libavcodec/allcodecs.c b/libavcodec/allcodecs.c
index e6305cf7d2..5480c7e78b 100644
--- a/libavcodec/allcodecs.c
+++ b/libavcodec/allcodecs.c
@@ -109,7 +109,7 @@ void avcodec_register_all(void)
     REGISTER_ENCDEC  (FFV1, ffv1);
     REGISTER_ENCDEC  (FFVHUFF, ffvhuff);
     REGISTER_ENCDEC  (FLASHSV, flashsv);
-    REGISTER_ENCODER (FLASHSV2, flashsv2);
+    REGISTER_ENCDEC (FLASHSV2, flashsv2);
     REGISTER_DECODER (FLIC, flic);
     REGISTER_ENCDEC  (FLV, flv);
     REGISTER_DECODER (FOURXM, fourxm);
diff --git a/libavcodec/dnxhddec.c b/libavcodec/dnxhddec.c
index a7ad620fd8..bfc4de645c 100644
--- a/libavcodec/dnxhddec.c
+++ b/libavcodec/dnxhddec.c
@@ -246,7 +246,7 @@ static av_always_inline void dnxhd_decode_dct_block(DNXHDContext *ctx,
         //av_log(ctx->avctx, AV_LOG_DEBUG, "j %d\n", j);
         //av_log(ctx->avctx, AV_LOG_DEBUG, "level %d, weight %d\n", level, weight_matrix[i]);
         level = (2*level+1) * qscale * weight_matrix[i];
-        if (weight_matrix[i] != level_bias)
+        if (level_bias < 32 || weight_matrix[i] != level_bias)
             level += level_bias;
         level >>= level_shift;
 
diff --git a/libavcodec/flashsv.c b/libavcodec/flashsv.c
index 96045e5eac..6b649bb874 100644
--- a/libavcodec/flashsv.c
+++ b/libavcodec/flashsv.c
@@ -25,6 +25,8 @@
  * Flash Screen Video decoder
  * @author Alex Beregszaszi
  * @author Benjamin Larsson
+ * @author Daniel Verkamp
+ * @author Konstantin Shishkov
  *
  * A description of the bitstream format for Flash Screen Video version 1/2
  * is part of the SWF File Format Specification (version 10), which can be
@@ -35,9 +37,17 @@
 #include <stdlib.h>
 #include <zlib.h>
 
+#include "libavutil/intreadwrite.h"
 #include "avcodec.h"
+#include "bytestream.h"
 #include "get_bits.h"
 
+typedef struct BlockInfo {
+    uint8_t *pos;
+    int      size;
+    int      unp_size;
+} BlockInfo;
+
 typedef struct FlashSVContext {
     AVCodecContext *avctx;
     AVFrame         frame;
@@ -46,9 +56,50 @@ typedef struct FlashSVContext {
     uint8_t        *tmpblock;
     int             block_size;
     z_stream        zstream;
+    int             ver;
+    const uint32_t *pal;
+    int             is_keyframe;
+    uint8_t        *keyframedata;
+    uint8_t        *keyframe;
+    BlockInfo      *blocks;
+    uint8_t        *deflate_block;
+    int             deflate_block_size;
+    int             color_depth;
+    int             zlibprime_curr, zlibprime_prev;
+    int             diff_start, diff_height;
 } FlashSVContext;
 
 
+static int decode_hybrid(const uint8_t *sptr, uint8_t *dptr, int dx, int dy,
+                         int h, int w, int stride, const uint32_t *pal)
+{
+    int x, y;
+    const uint8_t *orig_src = sptr;
+
+    for (y = dx+h; y > dx; y--) {
+        uint8_t *dst = dptr + (y * stride) + dy * 3;
+        for (x = 0; x < w; x++) {
+            if (*sptr & 0x80) {
+                /* 15-bit color */
+                unsigned c = AV_RB16(sptr) & ~0x8000;
+                unsigned b =  c        & 0x1F;
+                unsigned g = (c >>  5) & 0x1F;
+                unsigned r =  c >> 10;
+                /* 000aaabb -> aaabbaaa  */
+                *dst++ = (b << 3) | (b >> 2);
+                *dst++ = (g << 3) | (g >> 2);
+                *dst++ = (r << 3) | (r >> 2);
+                sptr += 2;
+            } else {
+                /* palette index */
+                uint32_t c = pal[*sptr++];
+                bytestream_put_le24(&dst, c);
+            }
+        }
+    }
+    return sptr - orig_src;
+}
+
 static av_cold int flashsv_decode_init(AVCodecContext *avctx)
 {
     FlashSVContext *s = avctx->priv_data;
@@ -71,9 +122,42 @@ static av_cold int flashsv_decode_init(AVCodecContext *avctx)
 }
 
 
+static void flashsv2_prime(FlashSVContext *s, uint8_t *src,
+                           int size, int unp_size)
+{
+    z_stream zs;
+
+    zs.zalloc = NULL;
+    zs.zfree  = NULL;
+    zs.opaque = NULL;
+
+    s->zstream.next_in   = src;
+    s->zstream.avail_in  = size;
+    s->zstream.next_out  = s->tmpblock;
+    s->zstream.avail_out = s->block_size * 3;
+    inflate(&s->zstream, Z_SYNC_FLUSH);
+
+    deflateInit(&zs, 0);
+    zs.next_in   = s->tmpblock;
+    zs.avail_in  = s->block_size * 3 - s->zstream.avail_out;
+    zs.next_out  = s->deflate_block;
+    zs.avail_out = s->deflate_block_size;
+    deflate(&zs, Z_SYNC_FLUSH);
+    deflateEnd(&zs);
+
+    inflateReset(&s->zstream);
+
+    s->zstream.next_in   = s->deflate_block;
+    s->zstream.avail_in  = s->deflate_block_size - zs.avail_out;
+    s->zstream.next_out  = s->tmpblock;
+    s->zstream.avail_out = s->block_size * 3;
+    inflate(&s->zstream, Z_SYNC_FLUSH);
+}
+
 static int flashsv_decode_block(AVCodecContext *avctx, AVPacket *avpkt,
                                 GetBitContext *gb, int block_size,
-                                int width, int height, int x_pos, int y_pos)
+                                int width, int height, int x_pos, int y_pos,
+                                int blk_idx)
 {
     struct FlashSVContext *s = avctx->priv_data;
     uint8_t *line = s->tmpblock;
@@ -82,6 +166,10 @@ static int flashsv_decode_block(AVCodecContext *avctx, AVPacket *avpkt,
     if (ret != Z_OK) {
         //return -1;
     }
+    if (s->zlibprime_curr || s->zlibprime_prev) {
+        flashsv2_prime(s, s->blocks[blk_idx].pos, s->blocks[blk_idx].size,
+                       s->blocks[blk_idx].unp_size);
+    }
     s->zstream.next_in   = avpkt->data + get_bits_count(gb) / 8;
     s->zstream.avail_in  = block_size;
     s->zstream.next_out  = s->tmpblock;
@@ -96,19 +184,48 @@ static int flashsv_decode_block(AVCodecContext *avctx, AVPacket *avpkt,
     if (ret != Z_OK && ret != Z_STREAM_END) {
         //return -1;
     }
-    /* Flash Screen Video stores the image upside down, so copy
-     * lines to destination in reverse order. */
-    for (k = 1; k <= height; k++) {
-        memcpy(s->frame.data[0] + x_pos * 3 +
-               (s->image_height - y_pos - k) * s->frame.linesize[0],
-               line, width * 3);
-        /* advance source pointer to next line */
-        line += width * 3;
+
+    if (s->is_keyframe) {
+        s->blocks[blk_idx].pos      = s->keyframedata + (get_bits_count(gb) / 8);
+        s->blocks[blk_idx].size     = block_size;
+        s->blocks[blk_idx].unp_size = s->block_size * 3 - s->zstream.avail_out;
+    }
+    if (!s->color_depth) {
+        /* Flash Screen Video stores the image upside down, so copy
+         * lines to destination in reverse order. */
+        for (k = 1; k <= s->diff_height; k++) {
+            memcpy(s->frame.data[0] + x_pos * 3 +
+                   (s->image_height - y_pos - s->diff_start - k) * s->frame.linesize[0],
+                   line, width * 3);
+            /* advance source pointer to next line */
+            line += width * 3;
+        }
+    } else {
+        /* hybrid 15-bit/palette mode */
+        decode_hybrid(s->tmpblock, s->frame.data[0],
+                      s->image_height - (y_pos + 1 + s->diff_start + s->diff_height),
+                      x_pos, s->diff_height, width,
+                      s->frame.linesize[0], s->pal);
     }
     skip_bits_long(gb, 8 * block_size); /* skip the consumed bits */
     return 0;
 }
 
+static int calc_deflate_block_size(int tmpblock_size)
+{
+    z_stream zstream;
+    int size;
+
+    zstream.zalloc = Z_NULL;
+    zstream.zfree  = Z_NULL;
+    zstream.opaque = Z_NULL;
+    if (deflateInit(&zstream, 0) != Z_OK)
+        return -1;
+    size = deflateBound(&zstream, tmpblock_size);
+    deflateEnd(&zstream);
+
+    return size;
+}
 
 static int flashsv_decode_frame(AVCodecContext *avctx, void *data,
                                 int *data_size, AVPacket *avpkt)
@@ -132,6 +249,18 @@ static int flashsv_decode_frame(AVCodecContext *avctx, void *data,
     s->block_height = 16 * (get_bits(&gb,  4) + 1);
     s->image_height =       get_bits(&gb, 12);
 
+    if (s->ver == 2) {
+        skip_bits(&gb, 6);
+        if (get_bits1(&gb)) {
+            av_log_missing_feature(avctx, "iframe", 1);
+            return AVERROR_PATCHWELCOME;
+        }
+        if (get_bits1(&gb)) {
+            av_log_missing_feature(avctx, "custom palette", 1);
+            return AVERROR_PATCHWELCOME;
+        }
+    }
+
     /* calculate number of blocks and size of border (partial) blocks */
     h_blocks = s->image_width  / s->block_width;
     h_part   = s->image_width  % s->block_width;
@@ -141,11 +270,25 @@ static int flashsv_decode_frame(AVCodecContext *avctx, void *data,
     /* the block size could change between frames, make sure the buffer
      * is large enough, if not, get a larger one */
     if (s->block_size < s->block_width * s->block_height) {
-        av_free(s->tmpblock);
-        if ((s->tmpblock = av_malloc(3 * s->block_width * s->block_height)) == NULL) {
+        int tmpblock_size = 3 * s->block_width * s->block_height;
+
+        s->tmpblock = av_realloc(s->tmpblock, tmpblock_size);
+        if (!s->tmpblock) {
             av_log(avctx, AV_LOG_ERROR, "Can't allocate decompression buffer.\n");
             return AVERROR(ENOMEM);
         }
+        if (s->ver == 2) {
+            s->deflate_block_size = calc_deflate_block_size(tmpblock_size);
+            if (s->deflate_block_size <= 0) {
+                av_log(avctx, AV_LOG_ERROR, "Can't determine deflate buffer size.\n");
+                return -1;
+            }
+            s->deflate_block = av_realloc(s->deflate_block, s->deflate_block_size);
+            if (!s->deflate_block) {
+                av_log(avctx, AV_LOG_ERROR, "Can't allocate deflate buffer.\n");
+                return AVERROR(ENOMEM);
+            }
+        }
     }
     s->block_size = s->block_width * s->block_height;
 
@@ -164,6 +307,16 @@ static int flashsv_decode_frame(AVCodecContext *avctx, void *data,
         return AVERROR_INVALIDDATA;
     }
 
+    /* we care for keyframes only in Screen Video v2 */
+    s->is_keyframe = (avpkt->flags & AV_PKT_FLAG_KEY) && (s->ver == 2);
+    if (s->is_keyframe) {
+        s->keyframedata = av_realloc(s->keyframedata, avpkt->size);
+        memcpy(s->keyframedata, avpkt->data, avpkt->size);
+        s->blocks = av_realloc(s->blocks,
+                               (v_blocks + !!v_part) * (h_blocks + !!h_part)
+                               * sizeof(s->blocks[0]));
+    }
+
     av_dlog(avctx, "image: %dx%d block: %dx%d num: %dx%d part: %dx%d\n",
             s->image_width, s->image_height, s->block_width, s->block_height,
             h_blocks, v_blocks, h_part, v_part);
@@ -187,25 +340,90 @@ static int flashsv_decode_frame(AVCodecContext *avctx, void *data,
         for (i = 0; i < h_blocks + (h_part ? 1 : 0); i++) {
             int x_pos = i * s->block_width; // horizontal position in frame
             int cur_blk_width = (i < h_blocks) ? s->block_width : h_part;
+            int has_diff = 0;
 
             /* get the size of the compressed zlib chunk */
             int size = get_bits(&gb, 16);
+
+            s->color_depth    = 0;
+            s->zlibprime_curr = 0;
+            s->zlibprime_prev = 0;
+            s->diff_start     = 0;
+            s->diff_height    = cur_blk_height;
+
             if (8 * size > get_bits_left(&gb)) {
                 avctx->release_buffer(avctx, &s->frame);
                 s->frame.data[0] = NULL;
                 return AVERROR_INVALIDDATA;
             }
 
+            if (s->ver == 2 && size) {
+                skip_bits(&gb, 3);
+                s->color_depth    = get_bits(&gb, 2);
+                has_diff          = get_bits1(&gb);
+                s->zlibprime_curr = get_bits1(&gb);
+                s->zlibprime_prev = get_bits1(&gb);
+
+                if (s->color_depth != 0 && s->color_depth != 2) {
+                    av_log(avctx, AV_LOG_ERROR,
+                           "%dx%d invalid color depth %d\n", i, j, s->color_depth);
+                    return -1;
+                }
+
+                if (has_diff) {
+                    s->diff_start  = get_bits(&gb, 8);
+                    s->diff_height = get_bits(&gb, 8);
+                    av_log(avctx, AV_LOG_DEBUG,
+                           "%dx%d diff start %d height %d\n",
+                           i, j, s->diff_start, s->diff_height);
+                    size -= 2;
+                }
+
+                if (s->zlibprime_prev)
+                    av_log(avctx, AV_LOG_DEBUG, "%dx%d zlibprime_prev\n", i, j);
+
+                if (s->zlibprime_curr) {
+                    int col = get_bits(&gb, 8);
+                    int row = get_bits(&gb, 8);
+                    av_log(avctx, AV_LOG_DEBUG, "%dx%d zlibprime_curr %dx%d\n", i, j, col, row);
+                    size -= 2;
+                    av_log_missing_feature(avctx, "zlibprime_curr", 1);
+                    return AVERROR_PATCHWELCOME;
+                }
+                size--; // account for flags byte
+            }
+
+            if (has_diff) {
+                int k;
+                int off = (s->image_height - y_pos - 1) * s->frame.linesize[0];
+
+                for (k = 0; k < cur_blk_height; k++)
+                    memcpy(s->frame.data[0] + off - k*s->frame.linesize[0] + x_pos*3,
+                           s->keyframe + off - k*s->frame.linesize[0] + x_pos*3,
+                           cur_blk_width * 3);
+            }
+
             /* skip unchanged blocks, which have size 0 */
             if (size) {
                 if (flashsv_decode_block(avctx, avpkt, &gb, size,
                                          cur_blk_width, cur_blk_height,
-                                         x_pos, y_pos))
+                                         x_pos, y_pos,
+                                         i + j * (h_blocks + !!h_part)))
                     av_log(avctx, AV_LOG_ERROR,
                            "error in decompression of block %dx%d\n", i, j);
             }
         }
     }
+    if (s->is_keyframe && s->ver == 2) {
+        if (!s->keyframe) {
+            s->keyframe = av_malloc(s->frame.linesize[0] * avctx->height);
+            if (!s->keyframe) {
+                av_log(avctx, AV_LOG_ERROR, "Cannot allocate image data\n");
+                return AVERROR(ENOMEM);
+            }
+        }
+        memcpy(s->keyframe, s->frame.data[0], s->frame.linesize[0] * avctx->height);
+    }
 
     *data_size = sizeof(AVFrame);
     *(AVFrame*)data = s->frame;
@@ -234,6 +452,7 @@ static av_cold int flashsv_decode_end(AVCodecContext *avctx)
 }
 
 
+#if CONFIG_FLASHSV_DECODER
 AVCodec ff_flashsv_decoder = {
     .name           = "flashsv",
     .type           = AVMEDIA_TYPE_VIDEO,
@@ -246,3 +465,67 @@ AVCodec ff_flashsv_decoder = {
     .pix_fmts       = (const enum PixelFormat[]){PIX_FMT_BGR24, PIX_FMT_NONE},
     .long_name      = NULL_IF_CONFIG_SMALL("Flash Screen Video v1"),
 };
+#endif /* CONFIG_FLASHSV_DECODER */
+
+#if CONFIG_FLASHSV2_DECODER
+static const uint32_t ff_flashsv2_default_palette[128] = {
+    0x000000, 0x333333, 0x666666, 0x999999, 0xCCCCCC, 0xFFFFFF,
+    0x330000, 0x660000, 0x990000, 0xCC0000, 0xFF0000, 0x003300,
+    0x006600, 0x009900, 0x00CC00, 0x00FF00, 0x000033, 0x000066,
+    0x000099, 0x0000CC, 0x0000FF, 0x333300, 0x666600, 0x999900,
+    0xCCCC00, 0xFFFF00, 0x003333, 0x006666, 0x009999, 0x00CCCC,
+    0x00FFFF, 0x330033, 0x660066, 0x990099, 0xCC00CC, 0xFF00FF,
+    0xFFFF33, 0xFFFF66, 0xFFFF99, 0xFFFFCC, 0xFF33FF, 0xFF66FF,
+    0xFF99FF, 0xFFCCFF, 0x33FFFF, 0x66FFFF, 0x99FFFF, 0xCCFFFF,
+    0xCCCC33, 0xCCCC66, 0xCCCC99, 0xCCCCFF, 0xCC33CC, 0xCC66CC,
+    0xCC99CC, 0xCCFFCC, 0x33CCCC, 0x66CCCC, 0x99CCCC, 0xFFCCCC,
+    0x999933, 0x999966, 0x9999CC, 0x9999FF, 0x993399, 0x996699,
+    0x99CC99, 0x99FF99, 0x339999, 0x669999, 0xCC9999, 0xFF9999,
+    0x666633, 0x666699, 0x6666CC, 0x6666FF, 0x663366, 0x669966,
+    0x66CC66, 0x66FF66, 0x336666, 0x996666, 0xCC6666, 0xFF6666,
+    0x333366, 0x333399, 0x3333CC, 0x3333FF, 0x336633, 0x339933,
+    0x33CC33, 0x33FF33, 0x663333, 0x993333, 0xCC3333, 0xFF3333,
+    0x003366, 0x336600, 0x660033, 0x006633, 0x330066, 0x663300,
+    0x336699, 0x669933, 0x993366, 0x339966, 0x663399, 0x996633,
+    0x6699CC, 0x99CC66, 0xCC6699, 0x66CC99, 0x9966CC, 0xCC9966,
+    0x99CCFF, 0xCCFF99, 0xFF99CC, 0x99FFCC, 0xCC99FF, 0xFFCC99,
+    0x111111, 0x222222, 0x444444, 0x555555, 0xAAAAAA, 0xBBBBBB,
+    0xDDDDDD, 0xEEEEEE
+};
+
+static av_cold int flashsv2_decode_init(AVCodecContext *avctx)
+{
+    FlashSVContext *s = avctx->priv_data;
+    flashsv_decode_init(avctx);
+    s->pal = ff_flashsv2_default_palette;
+    s->ver = 2;
+
+    return 0;
+}
+
+static av_cold int flashsv2_decode_end(AVCodecContext *avctx)
+{
+    FlashSVContext *s = avctx->priv_data;
+
+    av_freep(&s->keyframedata);
+    av_freep(&s->blocks);
+    av_freep(&s->keyframe);
+    av_freep(&s->deflate_block);
+    flashsv_decode_end(avctx);
+
+    return 0;
+}
+
+AVCodec ff_flashsv2_decoder = {
+    .name           = "flashsv2",
+    .type           = AVMEDIA_TYPE_VIDEO,
+    .id             = CODEC_ID_FLASHSV2,
+    .priv_data_size = sizeof(FlashSVContext),
+    .init           = flashsv2_decode_init,
+    .close          = flashsv2_decode_end,
+    .decode         = flashsv_decode_frame,
+    .capabilities   = CODEC_CAP_DR1,
+    .pix_fmts       = (const enum PixelFormat[]){PIX_FMT_BGR24, PIX_FMT_NONE},
+    .long_name      = NULL_IF_CONFIG_SMALL("Flash Screen Video v2"),
+};
+#endif /* CONFIG_FLASHSV2_DECODER */
diff --git a/libavcodec/version.h b/libavcodec/version.h
index 475e748ca7..1883768a29 100644
--- a/libavcodec/version.h
+++ b/libavcodec/version.h
@@ -21,7 +21,7 @@
 #define AVCODEC_VERSION_H
 
 #define LIBAVCODEC_VERSION_MAJOR 53
-#define LIBAVCODEC_VERSION_MINOR  8
+#define LIBAVCODEC_VERSION_MINOR  9
 #define LIBAVCODEC_VERSION_MICRO  0
 
 #define LIBAVCODEC_VERSION_INT  AV_VERSION_INT(LIBAVCODEC_VERSION_MAJOR, \
diff --git a/libavcodec/x86/h264_intrapred.asm b/libavcodec/x86/h264_intrapred.asm
index a9da6a19db..a178a335ed 100644
--- a/libavcodec/x86/h264_intrapred.asm
+++ b/libavcodec/x86/h264_intrapred.asm
@@ -2611,12 +2611,11 @@ cglobal pred4x4_down_left_mmxext, 3,3
     punpckldq m1, [r1]
     movq      m2, m1
     movq      m3, m1
-    movq      m4, m1
     psllq     m1, 8
     pxor      m2, m1
     psrlq     m2, 8
-    pxor      m3, m2
-    PRED4x4_LOWPASS m0, m1, m3, m4, m5
+    pxor      m2, m3
+    PRED4x4_LOWPASS m0, m1, m2, m3, m4
     lea       r1, [r0+r2*2]
     psrlq     m0, 8
     movd      [r0+r2*1], m0
diff --git a/libavcodec/x86/h264_intrapred_10bit.asm b/libavcodec/x86/h264_intrapred_10bit.asm
index 24a7bfa875..e14e31a38c 100644
--- a/libavcodec/x86/h264_intrapred_10bit.asm
+++ b/libavcodec/x86/h264_intrapred_10bit.asm
@@ -27,8 +27,6 @@
 
 SECTION_RODATA
 
-SECTION .text
-
 cextern pw_16
 cextern pw_8
 cextern pw_4
@@ -42,6 +40,8 @@ pw_512:       times 8 dw 512
 pd_17:        times 4 dd 17
 pd_16:        times 4 dd 16
 
+SECTION .text
+
 ; dest, left, right, src
 ; output: %1 = (t[n-1] + t[n]*2 + t[n+1] + 2) >> 2
 %macro PRED4x4_LOWPASS 4
@@ -64,13 +64,11 @@ cglobal pred4x4_down_right_10_%1, 3,3
     movq      m3, [r0]
     punpckhdq m1, m2
     PALIGNR   m3, m1, 10, m1
-    mova      m1, m3
     movhps    m4, [r1+r2*1-8]
-    PALIGNR   m3, m4, 14, m4
-    mova      m2, m3
+    PALIGNR   m0, m3, m4, 14, m4
     movhps    m4, [r1+r2*2-8]
-    PALIGNR   m3, m4, 14, m4
-    PRED4x4_LOWPASS m0, m3, m1, m2
+    PALIGNR   m2, m0, m4, 14, m4
+    PRED4x4_LOWPASS m0, m2, m3, m0
     movq      [r1+r2*2], m0
     psrldq    m0, 2
     movq      [r1+r2*1], m0
@@ -104,22 +102,20 @@ cglobal pred4x4_vertical_right_10_%1, 3,3,6
     pavgw   m5, m0
     movhps  m1, [r0+r2*1-8]
     PALIGNR m0, m1, 14, m1      ; ....t3t2t1t0ltl0
-    mova    m1, m0
     movhps  m2, [r0+r2*2-8]
-    PALIGNR m0, m2, 14, m2      ; ..t3t2t1t0ltl0l1
-    mova    m2, m0
+    PALIGNR m1, m0, m2, 14, m2  ; ..t3t2t1t0ltl0l1
     movhps  m3, [r1+r2*1-8]
-    PALIGNR m0, m3, 14, m3      ; t3t2t1t0ltl0l1l2
-    PRED4x4_LOWPASS m3, m1, m0, m2
-    pslldq  m1, m3, 12
-    psrldq  m3, 4
+    PALIGNR m2, m1, m3, 14, m3  ; t3t2t1t0ltl0l1l2
+    PRED4x4_LOWPASS m1, m0, m2, m1
+    pslldq  m0, m1, 12
+    psrldq  m1, 4
     movq    [r0+r2*1], m5
-    movq    [r0+r2*2], m3
-    PALIGNR m5, m1, 14, m2
-    pslldq  m1, 2
+    movq    [r0+r2*2], m1
+    PALIGNR m5, m0, 14, m2
+    pslldq  m0, 2
     movq    [r1+r2*1], m5
-    PALIGNR m3, m1, 14, m1
-    movq    [r1+r2*2], m3
+    PALIGNR m1, m0, 14, m0
+    movq    [r1+r2*2], m1
     RET
 %endmacro
 
@@ -152,9 +148,9 @@ cglobal pred4x4_horizontal_down_10_%1, 3,3
     punpckhdq  m1, m2          ; l0 l1 l2 l3
     punpckhqdq m1, m0          ; t2 t1 t0 lt l0 l1 l2 l3
     psrldq     m0, m1, 4       ; .. .. t2 t1 t0 lt l0 l1
-    psrldq     m2, m1, 2       ; .. t2 t1 t0 lt l0 l1 l2
-    pavgw      m5, m1, m2
-    PRED4x4_LOWPASS m3, m1, m0, m2
+    psrldq     m3, m1, 2       ; .. t2 t1 t0 lt l0 l1 l2
+    pavgw      m5, m1, m3
+    PRED4x4_LOWPASS m3, m1, m0, m3
     punpcklwd  m5, m3
     psrldq     m3, 8
     PALIGNR    m3, m5, 12, m4
@@ -220,17 +216,15 @@ cglobal pred4x4_dc_10_mmxext, 3,3
 ;-----------------------------------------------------------------------------
 ; void pred4x4_down_left(pixel *src, const pixel *topright, int stride)
 ;-----------------------------------------------------------------------------
-;TODO: more AVX here
 %macro PRED4x4_DL 1
 cglobal pred4x4_down_left_10_%1, 3,3
     sub        r0, r2
-    movq       m1, [r0]
-    movhps     m1, [r1]
-    pslldq     m5, m1, 2
-    pxor       m2, m5, m1
-    psrldq     m2, 2
-    pxor       m3, m1, m2
-    PRED4x4_LOWPASS m0, m5, m3, m1
+    movq       m0, [r0]
+    movhps     m0, [r1]
+    psrldq     m2, m0, 2
+    pslldq     m3, m0, 2
+    pshufhw    m2, m2, 10100100b
+    PRED4x4_LOWPASS m0, m3, m2, m0
     lea        r1, [r0+r2*2]
     movhps     [r1+r2*2], m0
     psrldq     m0, 2
@@ -257,10 +251,10 @@ cglobal pred4x4_vertical_left_10_%1, 3,3
     sub        r0, r2
     movu       m1, [r0]
     movhps     m1, [r1]
-    psrldq     m3, m1, 2
+    psrldq     m0, m1, 2
     psrldq     m2, m1, 4
-    pavgw      m4, m3, m1
-    PRED4x4_LOWPASS m0, m1, m2, m3
+    pavgw      m4, m0, m1
+    PRED4x4_LOWPASS m0, m1, m2, m0
     lea        r1, [r0+r2*2]
     movq       [r0+r2*1], m4
     movq       [r0+r2*2], m0
@@ -298,13 +292,13 @@ cglobal pred4x4_horizontal_up_10_mmxext, 3,3
     pavgw     m2, m0
 
     pshufw    m5, m0, 11111110b
-    PRED4x4_LOWPASS m3, m0, m5, m1
+    PRED4x4_LOWPASS m1, m0, m5, m1
     movq      m6, m2
-    punpcklwd m6, m3
+    punpcklwd m6, m1
     movq      [r0+r2*1], m6
     psrlq     m2, 16
-    psrlq     m3, 16
-    punpcklwd m2, m3
+    psrlq     m1, 16
+    punpcklwd m2, m1
     movq      [r0+r2*2], m2
     psrlq     m2, 32
     movd      [r1+r2*1], m2
@@ -333,7 +327,7 @@ cglobal pred8x8_vertical_10_sse2, 2,2
 ;-----------------------------------------------------------------------------
 INIT_XMM
 cglobal pred8x8_horizontal_10_sse2, 2,3
-    mov          r2, 4
+    mov         r2d, 4
 .loop:
     movq         m0, [r0+r1*0-8]
     movq         m1, [r0+r1*1-8]
@@ -344,7 +338,7 @@ cglobal pred8x8_horizontal_10_sse2, 2,3
     mova  [r0+r1*0], m0
     mova  [r0+r1*1], m1
     lea          r0, [r0+r1*2]
-    dec          r2
+    dec          r2d
     jg .loop
     REP_RET
 
@@ -362,53 +356,53 @@ cglobal pred8x8_horizontal_10_sse2, 2,3
 %endmacro
 
 %macro PRED8x8_DC 2
-cglobal pred8x8_dc_10_%1, 2,4
-%ifdef ARCH_X86_64
-%define t0 r10
-%else
-%define t0 r0m
-%endif
+cglobal pred8x8_dc_10_%1, 2,6
     sub         r0, r1
     pxor        m4, m4
     movq        m0, [r0+0]
     movq        m1, [r0+8]
-    HADDW       m0, m2
-    mov         t0, r0
-    HADDW       m1, m2
+%if mmsize==16
+    punpcklwd   m0, m1
+    movhlps     m1, m0
+    paddw       m0, m1
+%else
+    pshufw      m2, m0, 00001110b
+    pshufw      m3, m1, 00001110b
+    paddw       m0, m2
+    paddw       m1, m3
+    punpcklwd   m0, m1
+%endif
+    %2          m2, m0, 00001110b
+    paddw       m0, m2
 
+    lea         r5, [r1*3]
+    lea         r4, [r0+r1*4]
     movzx      r2d, word [r0+r1*1-2]
     movzx      r3d, word [r0+r1*2-2]
-    lea         r0, [r0+r1*2]
     add        r2d, r3d
-    movzx      r3d, word [r0+r1*1-2]
+    movzx      r3d, word [r0+r5*1-2]
     add        r2d, r3d
-    movzx      r3d, word [r0+r1*2-2]
+    movzx      r3d, word [r4-2]
     add        r2d, r3d
-    lea         r0, [r0+r1*2]
     movd        m2, r2d            ; s2
 
-    movzx      r2d, word [r0+r1*1-2]
-    movzx      r3d, word [r0+r1*2-2]
-    lea         r0, [r0+r1*2]
+    movzx      r2d, word [r4+r1*1-2]
+    movzx      r3d, word [r4+r1*2-2]
     add        r2d, r3d
-    movzx      r3d, word [r0+r1*1-2]
+    movzx      r3d, word [r4+r5*1-2]
     add        r2d, r3d
-    movzx      r3d, word [r0+r1*2-2]
+    movzx      r3d, word [r4+r1*4-2]
     add        r2d, r3d
     movd        m3, r2d            ; s3
 
-    punpcklwd   m0, m1
-    mov         r0, t0
     punpcklwd   m2, m3
     punpckldq   m0, m2            ; s0, s1, s2, s3
     %2          m3, m0, 11110110b ; s2, s1, s3, s3
-    lea         r2, [r1+r1*2]
     %2          m0, m0, 01110100b ; s0, s1, s3, s1
     paddw       m0, m3
-    lea         r3, [r0+r1*4]
     psrlw       m0, 2
     pavgw       m0, m4            ; s0+s2, s1, s3, s1+s3
-%ifidn %1, sse2
+%if mmsize==16
     punpcklwd   m0, m0
     pshufd      m3, m0, 11111010b
     punpckldq   m0, m0
@@ -421,12 +415,12 @@ cglobal pred8x8_dc_10_%1, 2,4
 %endif
     MOV8   r0+r1*1, m1, m2
     MOV8   r0+r1*2, m1, m2
-    MOV8   r0+r2*1, m1, m2
+    MOV8   r0+r5*1, m1, m2
     MOV8   r0+r1*4, m1, m2
-    MOV8   r3+r1*1, m3, m4
-    MOV8   r3+r1*2, m3, m4
-    MOV8   r3+r2*1, m3, m4
-    MOV8   r3+r1*4, m3, m4
+    MOV8   r4+r1*1, m3, m4
+    MOV8   r4+r1*2, m3, m4
+    MOV8   r4+r5*1, m3, m4
+    MOV8   r4+r1*4, m3, m4
     RET
 %endmacro
 
@@ -438,39 +432,29 @@ PRED8x8_DC sse2  , pshuflw
 ;-----------------------------------------------------------------------------
 ; void pred8x8_top_dc(pixel *src, int stride)
 ;-----------------------------------------------------------------------------
-%macro PRED8x8_TOP_DC 2
-cglobal pred8x8_top_dc_10_%1, 2,4
+INIT_XMM
+cglobal pred8x8_top_dc_10_sse2, 2,4
     sub         r0, r1
-    movq        m0, [r0+0]
-    movq        m1, [r0+8]
-    HADDW       m0, m2
-    HADDW       m1, m3
-    lea         r2, [r1+r1*2]
-    paddw       m0, [pw_2]
-    paddw       m1, [pw_2]
+    mova        m0, [r0]
+    pshuflw     m1, m0, 0x4e
+    pshufhw     m1, m1, 0x4e
+    paddw       m0, m1
+    pshuflw     m1, m0, 0xb1
+    pshufhw     m1, m1, 0xb1
+    paddw       m0, m1
+    lea         r2, [r1*3]
     lea         r3, [r0+r1*4]
+    paddw       m0, [pw_2]
     psrlw       m0, 2
-    psrlw       m1, 2
-    %2          m0, m0, 0
-    %2          m1, m1, 0
-%ifidn %1, sse2
-    punpcklqdq  m0, m1
-%endif
-    MOV8   r0+r1*1, m0, m1
-    MOV8   r0+r1*2, m0, m1
-    MOV8   r0+r2*1, m0, m1
-    MOV8   r0+r1*4, m0, m1
-    MOV8   r3+r1*1, m0, m1
-    MOV8   r3+r1*2, m0, m1
-    MOV8   r3+r2*1, m0, m1
-    MOV8   r3+r1*4, m0, m1
+    mova [r0+r1*1], m0
+    mova [r0+r1*2], m0
+    mova [r0+r2*1], m0
+    mova [r0+r1*4], m0
+    mova [r3+r1*1], m0
+    mova [r3+r1*2], m0
+    mova [r3+r2*1], m0
+    mova [r3+r1*4], m0
     RET
-%endmacro
-
-INIT_MMX
-PRED8x8_TOP_DC mmxext, pshufw
-INIT_XMM
-PRED8x8_TOP_DC sse2  , pshuflw
 
 ;-----------------------------------------------------------------------------
 ; void pred8x8_plane(pixel *src, int stride)
@@ -478,7 +462,7 @@ PRED8x8_TOP_DC sse2  , pshuflw
 INIT_XMM
 cglobal pred8x8_plane_10_sse2, 2,7,7
     sub       r0, r1
-    lea       r2, [r1+r1*2]
+    lea       r2, [r1*3]
     lea       r3, [r0+r1*4]
     mova      m2, [r0]
     pmaddwd   m2, [pw_m32101234]
@@ -500,7 +484,7 @@ cglobal pred8x8_plane_10_sse2, 2,7,7
     movzx    r5d, word [r3+r2*1-2] ; src[6*stride-1]
     movzx    r6d, word [r0+r1*1-2] ; src[0*stride-1]
     sub      r5d, r6d
-    lea      r5d, [r5+r5*2]
+    lea      r5d, [r5*3]
     add      r4d, r5d
     movzx    r6d, word [r3+r1*4-2] ; src[7*stride-1]
     movzx    r5d, word [r0+r1*0-2] ; src[ -stride-1]
@@ -540,8 +524,8 @@ cglobal pred8x8_plane_10_sse2, 2,7,7
 ;-----------------------------------------------------------------------------
 %macro PRED8x8L_128_DC 1
 cglobal pred8x8l_128_dc_10_%1, 4,4
-    mova      m0, [pw_512]
-    lea       r1, [r3+r3*2]
+    mova      m0, [pw_512] ; (1<<(BIT_DEPTH-1))
+    lea       r1, [r3*3]
     lea       r2, [r0+r3*4]
     MOV8 r0+r3*0, m0, m0
     MOV8 r0+r3*1, m0, m0
@@ -565,37 +549,17 @@ PRED8x8L_128_DC sse2
 %macro PRED8x8L_TOP_DC 1
 cglobal pred8x8l_top_dc_10_%1, 4,4,6
     sub         r0, r3
-    pxor        m7, m7
-    mova        m0, [r0-16]
-    mova        m3, [r0]
-    mova        m1, [r0+16]
-    mova        m2, m3
-    mova        m4, m3
-    PALIGNR     m2, m0, 14, m0
-    PALIGNR     m1, m4,  2, m4
-    test        r1, r1 ; top_left
-    jz .fix_lt_2
-    test        r2, r2 ; top_right
-    jz .fix_tr_1
-    jmp .body
-.fix_lt_2:
-    mova        m5, m3
-    pxor        m5, m2
-    pslldq      m5, 14
-    psrldq      m5, 14
-    pxor        m2, m5
-    test        r2, r2 ; top_right
-    jnz .body
-.fix_tr_1:
-    mova        m5, m3
-    pxor        m5, m1
-    psrldq      m5, 14
-    pslldq      m5, 14
-    pxor        m1, m5
-.body
-    lea         r1, [r3+r3*2]
+    mova        m0, [r0]
+    shr        r1d, 14
+    shr        r2d, 13
+    neg         r1
+    pslldq      m1, m0, 2
+    psrldq      m2, m0, 2
+    pinsrw      m1, [r0+r1], 0
+    pinsrw      m2, [r0+r2+14], 7
+    lea         r1, [r3*3]
     lea         r2, [r0+r3*4]
-    PRED4x4_LOWPASS m0, m2, m1, m3
+    PRED4x4_LOWPASS m0, m2, m1, m0
     HADDW       m0, m1
     paddw       m0, [pw_4]
     psrlw       m0, 3
@@ -612,110 +576,70 @@ cglobal pred8x8l_top_dc_10_%1, 4,4,6
 %endmacro
 
 INIT_XMM
-%define PALIGNR PALIGNR_MMX
 PRED8x8L_TOP_DC sse2
-%define PALIGNR PALIGNR_SSSE3
-PRED8x8L_TOP_DC ssse3
+%ifdef HAVE_AVX
+INIT_AVX
+PRED8x8L_TOP_DC avx
+%endif
 
 ;-----------------------------------------------------------------------------
 ;void pred8x8l_dc(pixel *src, int has_topleft, int has_topright, int stride)
 ;-----------------------------------------------------------------------------
 ;TODO: see if scalar is faster
 %macro PRED8x8L_DC 1
-cglobal pred8x8l_dc_10_%1, 4,5,8
+cglobal pred8x8l_dc_10_%1, 4,6,6
     sub         r0, r3
-    lea         r4, [r0+r3*2]
-    mova        m0, [r0+r3*1-16]
-    punpckhwd   m0, [r0+r3*0-16]
-    mova        m1, [r4+r3*1-16]
-    punpckhwd   m1, [r0+r3*2-16]
-    mov         r4, r0
+    lea         r4, [r0+r3*4]
+    lea         r5, [r3*3]
+    mova        m0, [r0+r3*2-16]
+    punpckhwd   m0, [r0+r3*1-16]
+    mova        m1, [r4+r3*0-16]
+    punpckhwd   m1, [r0+r5*1-16]
     punpckhdq   m1, m0
-    lea         r0, [r0+r3*4]
-    mova        m2, [r0+r3*1-16]
-    punpckhwd   m2, [r0+r3*0-16]
-    lea         r0, [r0+r3*2]
-    mova        m3, [r0+r3*1-16]
-    punpckhwd   m3, [r0+r3*0-16]
+    mova        m2, [r4+r3*2-16]
+    punpckhwd   m2, [r4+r3*1-16]
+    mova        m3, [r4+r3*4-16]
+    punpckhwd   m3, [r4+r5*1-16]
     punpckhdq   m3, m2
     punpckhqdq  m3, m1
-    lea         r0, [r0+r3*2]
-    mova        m0, [r0+r3*0-16]
-    mova        m1, [r4]
-    mov         r0, r4
-    mova        m4, m3
-    mova        m2, m3
-    PALIGNR     m4, m0, 14, m0
-    PALIGNR     m1, m2,  2, m2
-    test        r1, r1
-    jnz .do_left
-.fix_lt_1:
-    mova        m5, m3
-    pxor        m5, m4
-    psrldq      m5, 14
-    pslldq      m5, 12
-    pxor        m1, m5
-    jmp .do_left
-.fix_lt_2:
-    mova        m5, m3
-    pxor        m5, m2
-    pslldq      m5, 14
-    psrldq      m5, 14
-    pxor        m2, m5
-    test        r2, r2
-    jnz .body
-.fix_tr_1:
-    mova        m5, m3
-    pxor        m5, m1
-    psrldq      m5, 14
-    pslldq      m5, 14
-    pxor        m1, m5
-    jmp .body
-.do_left:
-    mova        m0, m4
-    PRED4x4_LOWPASS m2, m1, m4, m3
-    mova        m4, m0
-    mova        m7, m2
-    PRED4x4_LOWPASS m1, m3, m0, m4
-    pslldq      m1, 14
-    PALIGNR     m7, m1, 14, m3
-    mova        m0, [r0-16]
-    mova        m3, [r0]
-    mova        m1, [r0+16]
-    mova        m2, m3
-    mova        m4, m3
-    PALIGNR     m2, m0, 14, m0
-    PALIGNR     m1, m4,  2, m4
-    test        r1, r1
-    jz .fix_lt_2
-    test        r2, r2
-    jz .fix_tr_1
-.body
-    lea         r1, [r3+r3*2]
-    PRED4x4_LOWPASS m6, m2, m1, m3
-    HADDW       m7, m0
-    HADDW       m6, m0
-    lea         r2, [r0+r3*4]
-    paddw       m7, [pw_8]
-    paddw       m7, m6
-    psrlw       m7, 4
-    SPLATW      m7, m7
-    mova [r0+r3*1], m7
-    mova [r0+r3*2], m7
-    mova [r0+r1*1], m7
-    mova [r0+r3*4], m7
-    mova [r2+r3*1], m7
-    mova [r2+r3*2], m7
-    mova [r2+r1*1], m7
-    mova [r2+r3*4], m7
+    mova        m0, [r0]
+    shr        r1d, 14
+    shr        r2d, 13
+    neg         r1
+    pslldq      m1, m0, 2
+    psrldq      m2, m0, 2
+    pinsrw      m1, [r0+r1], 0
+    pinsrw      m2, [r0+r2+14], 7
+    not         r1
+    and         r1, r3
+    pslldq      m4, m3, 2
+    psrldq      m5, m3, 2
+    pshuflw     m4, m4, 11100101b
+    pinsrw      m5, [r0+r1-2], 7
+    PRED4x4_LOWPASS m3, m4, m5, m3
+    PRED4x4_LOWPASS m0, m2, m1, m0
+    paddw       m0, m3
+    HADDW       m0, m1
+    paddw       m0, [pw_8]
+    psrlw       m0, 4
+    SPLATW      m0, m0
+    mova [r0+r3*1], m0
+    mova [r0+r3*2], m0
+    mova [r0+r5*1], m0
+    mova [r0+r3*4], m0
+    mova [r4+r3*1], m0
+    mova [r4+r3*2], m0
+    mova [r4+r5*1], m0
+    mova [r4+r3*4], m0
     RET
 %endmacro
 
 INIT_XMM
-%define PALIGNR PALIGNR_MMX
 PRED8x8L_DC sse2
-%define PALIGNR PALIGNR_SSSE3
-PRED8x8L_DC ssse3
+%ifdef HAVE_AVX
+INIT_AVX
+PRED8x8L_DC avx
+%endif
 
 ;-----------------------------------------------------------------------------
 ; void pred8x8l_vertical(pixel *src, int has_topleft, int has_topright, int stride)
@@ -723,36 +647,17 @@ PRED8x8L_DC ssse3
 %macro PRED8x8L_VERTICAL 1
 cglobal pred8x8l_vertical_10_%1, 4,4,6
     sub         r0, r3
-    mova        m0, [r0-16]
-    mova        m3, [r0]
-    mova        m1, [r0+16]
-    mova        m2, m3
-    mova        m4, m3
-    PALIGNR     m2, m0, 14, m0
-    PALIGNR     m1, m4,  2, m4
-    test        r1, r1 ; top_left
-    jz .fix_lt_2
-    test        r2, r2 ; top_right
-    jz .fix_tr_1
-    jmp .body
-.fix_lt_2:
-    mova        m5, m3
-    pxor        m5, m2
-    pslldq      m5, 14
-    psrldq      m5, 14
-    pxor        m2, m5
-    test        r2, r2 ; top_right
-    jnz .body
-.fix_tr_1:
-    mova        m5, m3
-    pxor        m5, m1
-    psrldq      m5, 14
-    pslldq      m5, 14
-    pxor        m1, m5
-.body
-    lea         r1, [r3+r3*2]
+    mova        m0, [r0]
+    shr        r1d, 14
+    shr        r2d, 13
+    neg         r1
+    pslldq      m1, m0, 2
+    psrldq      m2, m0, 2
+    pinsrw      m1, [r0+r1], 0
+    pinsrw      m2, [r0+r2+14], 7
+    lea         r1, [r3*3]
     lea         r2, [r0+r3*4]
-    PRED4x4_LOWPASS m0, m2, m1, m3
+    PRED4x4_LOWPASS m0, m2, m1, m0
     mova [r0+r3*1], m0
     mova [r0+r3*2], m0
     mova [r0+r1*1], m0
@@ -765,70 +670,56 @@ cglobal pred8x8l_vertical_10_%1, 4,4,6
 %endmacro
 
 INIT_XMM
-%define PALIGNR PALIGNR_MMX
 PRED8x8L_VERTICAL sse2
-%define PALIGNR PALIGNR_SSSE3
-PRED8x8L_VERTICAL ssse3
+%ifdef HAVE_AVX
+INIT_AVX
+PRED8x8L_VERTICAL avx
+%endif
 
 ;-----------------------------------------------------------------------------
 ; void pred8x8l_horizontal(uint8_t *src, int has_topleft, int has_topright, int stride)
 ;-----------------------------------------------------------------------------
 %macro PRED8x8L_HORIZONTAL 1
-cglobal pred8x8l_horizontal_10_%1, 4,4,8
-    sub         r0, r3
-    lea         r2, [r0+r3*2]
-    mova        m0, [r0+r3*1-16]
-    test        r1, r1
-    lea         r1, [r0+r3]
-    cmovnz      r1, r0
-    punpckhwd   m0, [r1+r3*0-16]
-    mova        m1, [r2+r3*1-16]
-    punpckhwd   m1, [r0+r3*2-16]
-    mov         r2, r0
+cglobal pred8x8l_horizontal_10_%1, 4,4,5
+    mova        m0, [r0-16]
+    shr        r1d, 14
+    dec         r1
+    and         r1, r3
+    sub         r1, r3
+    punpckhwd   m0, [r0+r1-16]
+    mova        m1, [r0+r3*2-16]
+    punpckhwd   m1, [r0+r3*1-16]
+    lea         r2, [r0+r3*4]
+    lea         r1, [r3*3]
     punpckhdq   m1, m0
-    lea         r0, [r0+r3*4]
-    mova        m2, [r0+r3*1-16]
-    punpckhwd   m2, [r0+r3*0-16]
-    lea         r0, [r0+r3*2]
-    mova        m3, [r0+r3*1-16]
-    punpckhwd   m3, [r0+r3*0-16]
+    mova        m2, [r2+r3*0-16]
+    punpckhwd   m2, [r0+r1-16]
+    mova        m3, [r2+r3*2-16]
+    punpckhwd   m3, [r2+r3*1-16]
     punpckhdq   m3, m2
     punpckhqdq  m3, m1
-    lea         r0, [r0+r3*2]
-    mova        m0, [r0+r3*0-16]
-    mova        m1, [r1+r3*0-16]
-    mov         r0, r2
-    mova        m4, m3
-    mova        m2, m3
-    PALIGNR     m4, m0, 14, m0
-    PALIGNR     m1, m2,  2, m2
-    mova        m0, m4
-    PRED4x4_LOWPASS m2, m1, m4, m3
-    mova        m4, m0
-    mova        m7, m2
-    PRED4x4_LOWPASS m1, m3, m0, m4
-    pslldq      m1, 14
-    PALIGNR     m7, m1, 14, m3
-    lea         r1, [r3+r3*2]
-    punpckhwd   m3, m7, m7
-    punpcklwd   m7, m7
+    PALIGNR     m4, m3, [r2+r1-16], 14, m0
+    pslldq      m0, m4, 2
+    pshuflw     m0, m0, 11100101b
+    PRED4x4_LOWPASS m4, m3, m0, m4
+    punpckhwd   m3, m4, m4
+    punpcklwd   m4, m4
     pshufd      m0, m3, 0xff
     pshufd      m1, m3, 0xaa
-    lea         r2, [r0+r3*4]
     pshufd      m2, m3, 0x55
     pshufd      m3, m3, 0x00
-    pshufd      m4, m7, 0xff
-    pshufd      m5, m7, 0xaa
-    pshufd      m6, m7, 0x55
-    pshufd      m7, m7, 0x00
-    mova [r0+r3*1], m0
-    mova [r0+r3*2], m1
-    mova [r0+r1*1], m2
-    mova [r0+r3*4], m3
-    mova [r2+r3*1], m4
-    mova [r2+r3*2], m5
-    mova [r2+r1*1], m6
-    mova [r2+r3*4], m7
+    mova [r0+r3*0], m0
+    mova [r0+r3*1], m1
+    mova [r0+r3*2], m2
+    mova [r0+r1*1], m3
+    pshufd      m0, m4, 0xff
+    pshufd      m1, m4, 0xaa
+    pshufd      m2, m4, 0x55
+    pshufd      m3, m4, 0x00
+    mova [r2+r3*0], m0
+    mova [r2+r3*1], m1
+    mova [r2+r3*2], m2
+    mova [r2+r1*1], m3
     RET
 %endmacro
 
@@ -837,116 +728,68 @@ INIT_XMM
 PRED8x8L_HORIZONTAL sse2
 %define PALIGNR PALIGNR_SSSE3
 PRED8x8L_HORIZONTAL ssse3
+%ifdef HAVE_AVX
+INIT_AVX
+PRED8x8L_HORIZONTAL avx
+%endif
 
 ;-----------------------------------------------------------------------------
 ;void pred8x8l_down_left(pixel *src, int has_topleft, int has_topright, int stride)
 ;-----------------------------------------------------------------------------
 %macro PRED8x8L_DOWN_LEFT 1
-cglobal pred8x8l_down_left_10_%1, 4,4,8
+cglobal pred8x8l_down_left_10_%1, 4,4,7
     sub         r0, r3
-    mova        m0, [r0-16]
     mova        m3, [r0]
+    shr        r1d, 14
+    neg         r1
+    shr        r2d, 13
+    pslldq      m1, m3, 2
+    psrldq      m2, m3, 2
+    pinsrw      m1, [r0+r1], 0
+    pinsrw      m2, [r0+r2+14], 7
+    PRED4x4_LOWPASS m6, m2, m1, m3
+    jz .fix_tr ; flags from shr r2d
     mova        m1, [r0+16]
-    mova        m2, m3
-    mova        m4, m3
-    PALIGNR     m2, m0, 14, m0
-    PALIGNR     m1, m4,  2, m4
-    test        r1, r1
-    jz .fix_lt_2
-    test        r2, r2
-    jz .fix_tr_1
-    jmp .do_top
-.fix_lt_2:
-    mova        m5, m3
-    pxor        m5, m2
-    pslldq      m5, 14
-    psrldq      m5, 14
-    pxor        m2, m5
-    test        r2, r2
-    jnz .do_top
-.fix_tr_1:
-    mova        m5, m3
-    pxor        m5, m1
-    psrldq      m5, 14
-    pslldq      m5, 14
-    pxor        m1, m5
-    jmp .do_top
-.fix_tr_2:
-    punpckhwd   m3, m3
-    pshufd      m1, m3, 0xFF
-    jmp .do_topright
-.do_top:
-    PRED4x4_LOWPASS m4, m2, m1, m3
-    mova        m7, m4
-    test        r2, r2
-    jz .fix_tr_2
-    mova        m0, [r0+16]
-    mova        m5, m0
-    mova        m2, m0
-    mova        m4, m0
-    psrldq      m5, 14
-    PALIGNR     m2, m3, 14, m3
-    PALIGNR     m5, m4,  2, m4
-    PRED4x4_LOWPASS m1, m2, m5, m0
+    psrldq      m5, m1, 2
+    PALIGNR     m2, m1, m3, 14, m3
+    pshufhw     m5, m5, 10100100b
+    PRED4x4_LOWPASS m1, m2, m5, m1
 .do_topright:
-    lea         r1, [r3+r3*2]
-    mova        m6, m1
-    psrldq      m1, 14
-    mova        m4, m1
+    lea         r1, [r3*3]
+    psrldq      m5, m1, 14
     lea         r2, [r0+r3*4]
-    mova        m2, m6
-    PALIGNR     m2, m7,  2, m0
-    mova        m3, m6
-    PALIGNR     m3, m7, 14, m0
-    PALIGNR     m4, m6,  2, m0
-    mova        m5, m7
-    mova        m1, m7
-    mova        m7, m6
-    pslldq      m1, 2
-    PRED4x4_LOWPASS m0, m1, m2, m5
-    PRED4x4_LOWPASS m1, m3, m4, m7
+    PALIGNR     m2, m1, m6,  2, m0
+    PALIGNR     m3, m1, m6, 14, m0
+    PALIGNR     m5, m1,  2, m0
+    pslldq      m4, m6, 2
+    PRED4x4_LOWPASS m6, m4, m2, m6
+    PRED4x4_LOWPASS m1, m3, m5, m1
     mova [r2+r3*4], m1
-    mova        m2, m0
-    pslldq      m1, 2
-    psrldq      m2, 14
-    pslldq      m0, 2
-    por         m1, m2
+    PALIGNR     m1, m6, 14, m2
+    pslldq      m6, 2
     mova [r2+r1*1], m1
-    mova        m2, m0
-    pslldq      m1, 2
-    psrldq      m2, 14
-    pslldq      m0, 2
-    por         m1, m2
+    PALIGNR     m1, m6, 14, m2
+    pslldq      m6, 2
     mova [r2+r3*2], m1
-    mova        m2, m0
-    pslldq      m1, 2
-    psrldq      m2, 14
-    pslldq      m0, 2
-    por         m1, m2
+    PALIGNR     m1, m6, 14, m2
+    pslldq      m6, 2
     mova [r2+r3*1], m1
-    mova        m2, m0
-    pslldq      m1, 2
-    psrldq      m2, 14
-    pslldq      m0, 2
-    por         m1, m2
+    PALIGNR     m1, m6, 14, m2
+    pslldq      m6, 2
     mova [r0+r3*4], m1
-    mova        m2, m0
-    pslldq      m1, 2
-    psrldq      m2, 14
-    pslldq      m0, 2
-    por         m1, m2
+    PALIGNR     m1, m6, 14, m2
+    pslldq      m6, 2
     mova [r0+r1*1], m1
-    mova        m2, m0
-    pslldq      m1, 2
-    psrldq      m2, 14
-    pslldq      m0, 2
-    por         m1, m2
+    PALIGNR     m1, m6, 14, m2
+    pslldq      m6, 2
     mova [r0+r3*2], m1
-    pslldq      m1, 2
-    psrldq      m0, 14
-    por         m1, m0
+    PALIGNR     m1, m6, 14, m6
     mova [r0+r3*1], m1
     RET
+.fix_tr:
+    punpckhwd   m3, m3
+    pshufd      m1, m3, 0xFF
+    jmp .do_topright
 %endmacro
 
 INIT_XMM
@@ -954,139 +797,73 @@ INIT_XMM
 PRED8x8L_DOWN_LEFT sse2
 %define PALIGNR PALIGNR_SSSE3
 PRED8x8L_DOWN_LEFT ssse3
+%ifdef HAVE_AVX
+INIT_AVX
+PRED8x8L_DOWN_LEFT avx
+%endif
 
 ;-----------------------------------------------------------------------------
-;void pred8x8l_down_right_mxext(pixel *src, int has_topleft, int has_topright, int stride)
+;void pred8x8l_down_right(pixel *src, int has_topleft, int has_topright, int stride)
 ;-----------------------------------------------------------------------------
 %macro PRED8x8L_DOWN_RIGHT 1
+; standard forbids this when has_topleft is false
+; no need to check
 cglobal pred8x8l_down_right_10_%1, 4,5,8
     sub         r0, r3
-    lea         r4, [r0+r3*2]
+    lea         r4, [r0+r3*4]
+    lea         r1, [r3*3]
     mova        m0, [r0+r3*1-16]
     punpckhwd   m0, [r0+r3*0-16]
-    mova        m1, [r4+r3*1-16]
+    mova        m1, [r0+r1*1-16]
     punpckhwd   m1, [r0+r3*2-16]
-    mov         r4, r0
     punpckhdq   m1, m0
-    lea         r0, [r0+r3*4]
-    mova        m2, [r0+r3*1-16]
-    punpckhwd   m2, [r0+r3*0-16]
-    lea         r0, [r0+r3*2]
-    mova        m3, [r0+r3*1-16]
-    punpckhwd   m3, [r0+r3*0-16]
+    mova        m2, [r4+r3*1-16]
+    punpckhwd   m2, [r4+r3*0-16]
+    mova        m3, [r4+r1*1-16]
+    punpckhwd   m3, [r4+r3*2-16]
     punpckhdq   m3, m2
     punpckhqdq  m3, m1
-    lea         r0, [r0+r3*2]
-    mova        m0, [r0+r3*0-16]
-    mova        m1, [r4]
-    mov         r0, r4
-    mova        m4, m3
-    mova        m2, m3
-    PALIGNR     m4, m0, 14, m0
-    PALIGNR     m1, m2,  2, m2
-    test        r1, r1 ; top_left
-    jz .fix_lt_1
-.do_left:
-    mova        m0, m4
-    PRED4x4_LOWPASS m2, m1, m4, m3
-    mova        m4, m0
-    mova        m7, m2
-    mova        m6, m2
-    PRED4x4_LOWPASS m1, m3, m0, m4
-    pslldq      m1, 14
-    PALIGNR     m7, m1, 14, m3
-    mova        m0, [r0-16]
+    mova        m0, [r4+r3*4-16]
+    mova        m1, [r0]
+    PALIGNR     m4, m3, m0, 14, m0
+    PALIGNR     m1, m3,  2, m2
+    pslldq      m0, m4, 2
+    pshuflw     m0, m0, 11100101b
+    PRED4x4_LOWPASS m6, m1, m4, m3
+    PRED4x4_LOWPASS m4, m3, m0, m4
     mova        m3, [r0]
-    mova        m1, [r0+16]
-    mova        m2, m3
-    mova        m4, m3
-    PALIGNR     m2, m0, 14, m0
-    PALIGNR     m1, m4,  2, m4
-    test        r1, r1 ; top_left
-    jz .fix_lt_2
-    test        r2, r2 ; top_right
-    jz .fix_tr_1
-.do_top:
-    PRED4x4_LOWPASS m4, m2, m1, m3
-    mova        m5, m4
-    jmp .body
-.fix_lt_1:
-    mova        m5, m3
-    pxor        m5, m4
-    psrldq      m5, 14
-    pslldq      m5, 12
-    pxor        m1, m5
-    jmp .do_left
-.fix_lt_2:
-    mova        m5, m3
-    pxor        m5, m2
-    pslldq      m5, 14
-    psrldq      m5, 14
-    pxor        m2, m5
-    test        r2, r2 ; top_right
-    jnz .do_top
-.fix_tr_1:
-    mova        m5, m3
-    pxor        m5, m1
-    psrldq      m5, 14
-    pslldq      m5, 14
-    pxor        m1, m5
-    jmp .do_top
-.body
-    lea         r1, [r3+r3*2]
-    mova        m1, m7
-    mova        m7, m5
-    mova        m5, m6
-    mova        m2, m7
-    lea         r2, [r0+r3*4]
-    PALIGNR     m2, m6,  2, m0
-    mova        m3, m7
-    PALIGNR     m3, m6, 14, m0
-    mova        m4, m7
-    psrldq      m4, 2
-    PRED4x4_LOWPASS m0, m1, m2, m5
-    PRED4x4_LOWPASS m1, m3, m4, m7
-    mova [r2+r3*4], m0
-    mova        m2, m1
-    psrldq      m0, 2
-    pslldq      m2, 14
-    psrldq      m1, 2
-    por         m0, m2
-    mova [r2+r1*1], m0
-    mova        m2, m1
-    psrldq      m0, 2
-    pslldq      m2, 14
-    psrldq      m1, 2
-    por         m0, m2
-    mova [r2+r3*2], m0
-    mova        m2, m1
-    psrldq      m0, 2
-    pslldq      m2, 14
-    psrldq      m1, 2
-    por         m0, m2
-    mova [r2+r3*1], m0
-    mova        m2, m1
-    psrldq      m0, 2
-    pslldq      m2, 14
-    psrldq      m1, 2
-    por         m0, m2
-    mova [r0+r3*4], m0
-    mova        m2, m1
-    psrldq      m0, 2
-    pslldq      m2, 14
-    psrldq      m1, 2
-    por         m0, m2
-    mova [r0+r1*1], m0
-    mova        m2, m1
-    psrldq      m0, 2
-    pslldq      m2, 14
-    psrldq      m1, 2
-    por         m0, m2
-    mova [r0+r3*2], m0
-    psrldq      m0, 2
-    pslldq      m1, 14
-    por         m0, m1
-    mova [r0+r3*1], m0
+    shr        r2d, 13
+    pslldq      m1, m3, 2
+    psrldq      m2, m3, 2
+    pinsrw      m1, [r0-2], 0
+    pinsrw      m2, [r0+r2+14], 7
+    PRED4x4_LOWPASS m3, m2, m1, m3
+    PALIGNR     m2, m3, m6,  2, m0
+    PALIGNR     m5, m3, m6, 14, m0
+    psrldq      m7, m3, 2
+    PRED4x4_LOWPASS m6, m4, m2, m6
+    PRED4x4_LOWPASS m3, m5, m7, m3
+    mova [r4+r3*4], m6
+    PALIGNR     m3, m6, 14, m2
+    pslldq      m6, 2
+    mova [r0+r3*1], m3
+    PALIGNR     m3, m6, 14, m2
+    pslldq      m6, 2
+    mova [r0+r3*2], m3
+    PALIGNR     m3, m6, 14, m2
+    pslldq      m6, 2
+    mova [r0+r1*1], m3
+    PALIGNR     m3, m6, 14, m2
+    pslldq      m6, 2
+    mova [r0+r3*4], m3
+    PALIGNR     m3, m6, 14, m2
+    pslldq      m6, 2
+    mova [r4+r3*1], m3
+    PALIGNR     m3, m6, 14, m2
+    pslldq      m6, 2
+    mova [r4+r3*2], m3
+    PALIGNR     m3, m6, 14, m6
+    mova [r4+r1*1], m3
     RET
 %endmacro
 
@@ -1095,114 +872,69 @@ INIT_XMM
 PRED8x8L_DOWN_RIGHT sse2
 %define PALIGNR PALIGNR_SSSE3
 PRED8x8L_DOWN_RIGHT ssse3
+%ifdef HAVE_AVX
+INIT_AVX
+PRED8x8L_DOWN_RIGHT avx
+%endif
 
 ;-----------------------------------------------------------------------------
 ; void pred8x8l_vertical_right(pixel *src, int has_topleft, int has_topright, int stride)
 ;-----------------------------------------------------------------------------
 %macro PRED8x8L_VERTICAL_RIGHT 1
-cglobal pred8x8l_vertical_right_10_%1, 4,5,8
+; likewise with 8x8l_down_right
+cglobal pred8x8l_vertical_right_10_%1, 4,5,7
     sub         r0, r3
-    lea         r4, [r0+r3*2]
+    lea         r4, [r0+r3*4]
+    lea         r1, [r3*3]
     mova        m0, [r0+r3*1-16]
     punpckhwd   m0, [r0+r3*0-16]
-    mova        m1, [r4+r3*1-16]
+    mova        m1, [r0+r1*1-16]
     punpckhwd   m1, [r0+r3*2-16]
-    mov         r4, r0
     punpckhdq   m1, m0
-    lea         r0, [r0+r3*4]
-    mova        m2, [r0+r3*1-16]
-    punpckhwd   m2, [r0+r3*0-16]
-    lea         r0, [r0+r3*2]
-    mova        m3, [r0+r3*1-16]
-    punpckhwd   m3, [r0+r3*0-16]
+    mova        m2, [r4+r3*1-16]
+    punpckhwd   m2, [r4+r3*0-16]
+    mova        m3, [r4+r1*1-16]
+    punpckhwd   m3, [r4+r3*2-16]
     punpckhdq   m3, m2
     punpckhqdq  m3, m1
-    lea         r0, [r0+r3*2]
-    mova        m0, [r0+r3*0-16]
-    mova        m1, [r4]
-    mov         r0, r4
-    mova        m4, m3
-    mova        m2, m3
-    PALIGNR     m4, m0, 14, m0
-    PALIGNR     m1, m2,  2, m2
-    test        r1, r1
-    jz .fix_lt_1
-    jmp .do_left
-.fix_lt_1:
-    mova        m5, m3
-    pxor        m5, m4
-    psrldq      m5, 14
-    pslldq      m5, 12
-    pxor        m1, m5
-    jmp .do_left
-.fix_lt_2:
-    mova        m5, m3
-    pxor        m5, m2
-    pslldq      m5, 14
-    psrldq      m5, 14
-    pxor        m2, m5
-    test        r2, r2
-    jnz .do_top
-.fix_tr_1:
-    mova        m5, m3
-    pxor        m5, m1
-    psrldq      m5, 14
-    pslldq      m5, 14
-    pxor        m1, m5
-    jmp .do_top
-.do_left:
-    mova        m0, m4
-    PRED4x4_LOWPASS m2, m1, m4, m3
-    mova        m7, m2
-    mova        m0, [r0-16]
-    mova        m3, [r0]
-    mova        m1, [r0+16]
-    mova        m2, m3
-    mova        m4, m3
-    PALIGNR     m2, m0, 14, m0
-    PALIGNR     m1, m4,  2, m4
-    test        r1, r1
-    jz .fix_lt_2
-    test        r2, r2
-    jz .fix_tr_1
-.do_top
-    PRED4x4_LOWPASS m6, m2, m1, m3
-    lea         r1, [r3+r3*2]
-    mova        m2, m6
-    mova        m3, m6
-    PALIGNR     m3, m7, 14, m0
-    PALIGNR     m6, m7, 12, m1
-    mova        m4, m3
-    pavgw       m3, m2
-    lea         r2, [r0+r3*4]
-    PRED4x4_LOWPASS m0, m6, m2, m4
-    mova [r0+r3*1], m3
+    mova        m0, [r4+r3*4-16]
+    mova        m1, [r0]
+    PALIGNR     m4, m3, m0, 14, m0
+    PALIGNR     m1, m3,  2, m2
+    PRED4x4_LOWPASS m3, m1, m4, m3
+    mova        m2, [r0]
+    shr        r2d, 13
+    pslldq      m1, m2, 2
+    psrldq      m5, m2, 2
+    pinsrw      m1, [r0-2], 0
+    pinsrw      m5, [r0+r2+14], 7
+    PRED4x4_LOWPASS m2, m5, m1, m2
+    PALIGNR     m6, m2, m3, 12, m1
+    PALIGNR     m5, m2, m3, 14, m0
+    PRED4x4_LOWPASS m0, m6, m2, m5
+    pavgw       m2, m5
     mova [r0+r3*2], m0
-    mova        m5, m0
-    mova        m6, m3
-    mova        m1, m7
-    mova        m2, m1
-    pslldq      m2, 2
-    mova        m3, m1
-    pslldq      m3, 4
-    PRED4x4_LOWPASS m0, m1, m3, m2
-    PALIGNR     m6, m0, 14, m2
-    mova [r0+r1*1], m6
-    pslldq      m0, 2
-    PALIGNR     m5, m0, 14, m1
-    mova [r0+r3*4], m5
-    pslldq      m0, 2
-    PALIGNR     m6, m0, 14, m2
-    mova [r2+r3*1], m6
-    pslldq      m0, 2
-    PALIGNR     m5, m0, 14, m1
-    mova [r2+r3*2], m5
-    pslldq      m0, 2
-    PALIGNR     m6, m0, 14, m2
-    mova [r2+r1*1], m6
-    pslldq      m0, 2
-    PALIGNR     m5, m0, 14, m1
-    mova [r2+r3*4], m5
+    mova [r0+r3*1], m2
+    pslldq      m6, m3, 4
+    pslldq      m1, m3, 2
+    PRED4x4_LOWPASS m1, m3, m6, m1
+    PALIGNR     m2, m1, 14, m4
+    mova [r0+r1*1], m2
+    pslldq      m1, 2
+    PALIGNR     m0, m1, 14, m3
+    mova [r0+r3*4], m0
+    pslldq      m1, 2
+    PALIGNR     m2, m1, 14, m4
+    mova [r4+r3*1], m2
+    pslldq      m1, 2
+    PALIGNR     m0, m1, 14, m3
+    mova [r4+r3*2], m0
+    pslldq      m1, 2
+    PALIGNR     m2, m1, 14, m4
+    mova [r4+r1*1], m2
+    pslldq      m1, 2
+    PALIGNR     m0, m1, 14, m1
+    mova [r4+r3*4], m0
     RET
 %endmacro
 
@@ -1211,84 +943,60 @@ INIT_XMM
 PRED8x8L_VERTICAL_RIGHT sse2
 %define PALIGNR PALIGNR_SSSE3
 PRED8x8L_VERTICAL_RIGHT ssse3
+%ifdef HAVE_AVX
+INIT_AVX
+PRED8x8L_VERTICAL_RIGHT avx
+%endif
 
 ;-----------------------------------------------------------------------------
 ; void pred8x8l_horizontal_up(pixel *src, int has_topleft, int has_topright, int stride)
 ;-----------------------------------------------------------------------------
 %macro PRED8x8L_HORIZONTAL_UP 1
-cglobal pred8x8l_horizontal_up_10_%1, 4,4,8
-    sub         r0, r3
-    lea         r2, [r0+r3*2]
-    mova        m0, [r0+r3*1-16]
-    test        r1, r1
-    lea         r1, [r0+r3]
-    cmovnz      r1, r0
-    punpckhwd   m0, [r1+r3*0-16]
-    mova        m1, [r2+r3*1-16]
-    punpckhwd   m1, [r0+r3*2-16]
-    mov         r2, r0
-    punpckhdq   m1, m0
-    lea         r0, [r0+r3*4]
-    mova        m2, [r0+r3*1-16]
-    punpckhwd   m2, [r0+r3*0-16]
-    lea         r0, [r0+r3*2]
-    mova        m3, [r0+r3*1-16]
-    punpckhwd   m3, [r0+r3*0-16]
-    punpckhdq   m3, m2
-    punpckhqdq  m3, m1
-    lea         r0, [r0+r3*2]
+cglobal pred8x8l_horizontal_up_10_%1, 4,4,6
     mova        m0, [r0+r3*0-16]
-    mova        m1, [r1+r3*0-16]
-    mov         r0, r2
-    mova        m4, m3
-    mova        m2, m3
-    PALIGNR     m4, m0, 14, m0
-    PALIGNR     m1, m2,  2, m2
-    mova        m0, m4
-    PRED4x4_LOWPASS m2, m1, m4, m3
-    mova        m4, m0
-    mova        m7, m2
-    PRED4x4_LOWPASS m1, m3, m0, m4
-    pslldq      m1, 14
-    PALIGNR     m7, m1, 14, m3
-    lea         r1, [r3+r3*2]
-    pshufd      m0, m7, 00011011b ; l6 l7 l4 l5 l2 l3 l0 l1
-    pslldq      m7, 14             ; l7 .. .. .. .. .. .. ..
-    mova        m2, m0
-    pslld       m0, 16
-    psrld       m2, 16
-    por         m2, m0            ; l7 l6 l5 l4 l3 l2 l1 l0
-    mova        m3, m2
-    mova        m4, m2
-    mova        m5, m2
-    psrldq      m2, 2
-    psrldq      m3, 4
+    punpckhwd   m0, [r0+r3*1-16]
+    shr        r1d, 14
+    dec         r1
+    and         r1, r3
+    sub         r1, r3
+    mova        m4, [r0+r1*1-16]
+    lea         r1, [r3*3]
     lea         r2, [r0+r3*4]
-    por         m2, m7            ; l7 l7 l6 l5 l4 l3 l2 l1
-    punpckhwd   m7, m7
-    por         m3, m7            ; l7 l7 l7 l6 l5 l4 l3 l2
-    pavgw       m4, m2
-    PRED4x4_LOWPASS m1, m3, m5, m2
-    mova        m5, m4
-    punpcklwd   m4, m1            ; p4 p3 p2 p1
-    punpckhwd   m5, m1            ; p8 p7 p6 p5
-    mova        m6, m5
-    mova        m7, m5
-    mova        m0, m5
-    PALIGNR     m5, m4, 4, m1
-    pshufd      m1, m6, 11111001b
-    PALIGNR     m6, m4, 8, m2
-    pshufd      m2, m7, 11111110b
-    PALIGNR     m7, m4, 12, m3
-    pshufd      m3, m0, 11111111b
-    mova [r0+r3*1], m4
-    mova [r0+r3*2], m5
-    mova [r0+r1*1], m6
-    mova [r0+r3*4], m7
+    mova        m1, [r0+r3*2-16]
+    punpckhwd   m1, [r0+r1*1-16]
+    punpckhdq   m0, m1
+    mova        m2, [r2+r3*0-16]
+    punpckhwd   m2, [r2+r3*1-16]
+    mova        m3, [r2+r3*2-16]
+    punpckhwd   m3, [r2+r1*1-16]
+    punpckhdq   m2, m3
+    punpckhqdq  m0, m2
+    PALIGNR     m1, m0, m4, 14, m4
+    psrldq      m2, m0, 2
+    pshufhw     m2, m2, 10100100b
+    PRED4x4_LOWPASS m0, m1, m2, m0
+    psrldq      m1, m0, 2
+    psrldq      m2, m0, 4
+    pshufhw     m1, m1, 10100100b
+    pshufhw     m2, m2, 01010100b
+    pavgw       m4, m0, m1
+    PRED4x4_LOWPASS m1, m2, m0, m1
+    punpckhwd   m5, m4, m1
+    punpcklwd   m4, m1
+    mova [r2+r3*0], m5
+    mova [r0+r3*0], m4
+    pshufd      m0, m5, 11111001b
+    pshufd      m1, m5, 11111110b
+    pshufd      m2, m5, 11111111b
     mova [r2+r3*1], m0
     mova [r2+r3*2], m1
     mova [r2+r1*1], m2
-    mova [r2+r3*4], m3
+    PALIGNR     m2, m5, m4, 4, m0
+    PALIGNR     m3, m5, m4, 8, m1
+    PALIGNR     m5, m5, m4, 12, m4
+    mova [r0+r3*1], m2
+    mova [r0+r3*2], m3
+    mova [r0+r1*1], m5
     RET
 %endmacro
 
@@ -1297,7 +1005,10 @@ INIT_XMM
 PRED8x8L_HORIZONTAL_UP sse2
 %define PALIGNR PALIGNR_SSSE3
 PRED8x8L_HORIZONTAL_UP ssse3
-
+%ifdef HAVE_AVX
+INIT_AVX
+PRED8x8L_HORIZONTAL_UP avx
+%endif
 
 
 ;-----------------------------------------------------------------------------
@@ -1315,7 +1026,7 @@ PRED8x8L_HORIZONTAL_UP ssse3
 %macro PRED16x16_VERTICAL 1
 cglobal pred16x16_vertical_10_%1, 2,3
     sub   r0, r1
-    mov   r2, 8
+    mov  r2d, 8
     mova  m0, [r0+ 0]
     mova  m1, [r0+mmsize]
 %if mmsize==8
@@ -1326,7 +1037,7 @@ cglobal pred16x16_vertical_10_%1, 2,3
     MOV16 r0+r1*1, m0, m1, m2, m3
     MOV16 r0+r1*2, m0, m1, m2, m3
     lea   r0, [r0+r1*2]
-    dec   r2
+    dec   r2d
     jg .loop
     REP_RET
 %endmacro
@@ -1341,7 +1052,7 @@ PRED16x16_VERTICAL sse2
 ;-----------------------------------------------------------------------------
 %macro PRED16x16_HORIZONTAL 1
 cglobal pred16x16_horizontal_10_%1, 2,3
-    mov    r2, 8
+    mov   r2d, 8
 .vloop:
     movd   m0, [r0+r1*0-4]
     movd   m1, [r0+r1*1-4]
@@ -1350,7 +1061,7 @@ cglobal pred16x16_horizontal_10_%1, 2,3
     MOV16  r0+r1*0, m0, m0, m0, m0
     MOV16  r0+r1*1, m1, m1, m1, m1
     lea    r0, [r0+r1*2]
-    dec    r2
+    dec    r2d
     jg .vloop
     REP_RET
 %endmacro
@@ -1364,8 +1075,8 @@ PRED16x16_HORIZONTAL sse2
 ; void pred16x16_dc(pixel *src, int stride)
 ;-----------------------------------------------------------------------------
 %macro PRED16x16_DC 1
-cglobal pred16x16_dc_10_%1, 2,7
-    mov        r4, r0
+cglobal pred16x16_dc_10_%1, 2,6
+    mov        r5, r0
     sub        r0, r1
     mova       m0, [r0+0]
     paddw      m0, [r0+mmsize]
@@ -1375,17 +1086,17 @@ cglobal pred16x16_dc_10_%1, 2,7
 %endif
     HADDW      m0, m2
 
-    sub        r0, 2
-    movzx     r3d, word [r0+r1*1]
-    movzx     r5d, word [r0+r1*2]
+    lea        r0, [r0+r1-2]
+    movzx     r3d, word [r0]
+    movzx     r4d, word [r0+r1]
 %rep 7
     lea        r0, [r0+r1*2]
-    movzx     r2d, word [r0+r1*1]
+    movzx     r2d, word [r0]
     add       r3d, r2d
-    movzx     r2d, word [r0+r1*2]
-    add       r5d, r2d
+    movzx     r2d, word [r0+r1]
+    add       r4d, r2d
 %endrep
-    lea       r3d, [r3+r5+16]
+    lea       r3d, [r3+r4+16]
 
     movd       m1, r3d
     paddw      m0, m1
@@ -1393,9 +1104,9 @@ cglobal pred16x16_dc_10_%1, 2,7
     SPLATW     m0, m0
     mov       r3d, 8
 .loop:
-    MOV16 r4+r1*0, m0, m0, m0, m0
-    MOV16 r4+r1*1, m0, m0, m0, m0
-    lea        r4, [r4+r1*2]
+    MOV16 r5+r1*0, m0, m0, m0, m0
+    MOV16 r5+r1*1, m0, m0, m0, m0
+    lea        r5, [r5+r1*2]
     dec       r3d
     jg .loop
     REP_RET
@@ -1442,29 +1153,29 @@ PRED16x16_TOP_DC sse2
 ; void pred16x16_left_dc(pixel *src, int stride)
 ;-----------------------------------------------------------------------------
 %macro PRED16x16_LEFT_DC 1
-cglobal pred16x16_left_dc_10_%1, 2,7
-    mov        r4, r0
+cglobal pred16x16_left_dc_10_%1, 2,6
+    mov        r5, r0
 
     sub        r0, 2
-    movzx     r5d, word [r0+r1*0]
-    movzx     r6d, word [r0+r1*1]
+    movzx     r3d, word [r0]
+    movzx     r4d, word [r0+r1]
 %rep 7
     lea        r0, [r0+r1*2]
-    movzx     r2d, word [r0+r1*0]
-    movzx     r3d, word [r0+r1*1]
-    add       r5d, r2d
-    add       r6d, r3d
+    movzx     r2d, word [r0]
+    add       r3d, r2d
+    movzx     r2d, word [r0+r1]
+    add       r4d, r2d
 %endrep
-    lea       r2d, [r5+r6+8]
-    shr       r2d, 4
+    lea       r3d, [r3+r4+8]
+    shr       r3d, 4
 
-    movd       m0, r2d
+    movd       m0, r3d
     SPLATW     m0, m0
     mov       r3d, 8
 .loop:
-    MOV16 r4+r1*0, m0, m0, m0, m0
-    MOV16 r4+r1*1, m0, m0, m0, m0
-    lea        r4, [r4+r1*2]
+    MOV16 r5+r1*0, m0, m0, m0, m0
+    MOV16 r5+r1*1, m0, m0, m0, m0
+    lea        r5, [r5+r1*2]
     dec       r3d
     jg .loop
     REP_RET
diff --git a/libavcodec/x86/h264_intrapred_init.c b/libavcodec/x86/h264_intrapred_init.c
index 5153c4e2bc..7220dd75c1 100644
--- a/libavcodec/x86/h264_intrapred_init.c
+++ b/libavcodec/x86/h264_intrapred_init.c
@@ -45,7 +45,6 @@ void ff_pred8x8_ ## TYPE ## _ ## DEPTH ## _ ## OPT (uint8_t *src, int stride);
 
 PRED8x8(dc, 10, mmxext)
 PRED8x8(dc, 10, sse2)
-PRED8x8(top_dc, 10, mmxext)
 PRED8x8(top_dc, 10, sse2)
 PRED8x8(plane, 10, sse2)
 PRED8x8(vertical, 10, sse2)
@@ -55,23 +54,28 @@ PRED8x8(horizontal, 10, sse2)
 void ff_pred8x8l_ ## TYPE ## _ ## DEPTH ## _ ## OPT (uint8_t *src, int has_topleft, int has_topright, int stride);
 
 PRED8x8L(dc, 10, sse2)
-PRED8x8L(dc, 10, ssse3)
+PRED8x8L(dc, 10, avx)
 PRED8x8L(128_dc, 10, mmxext)
 PRED8x8L(128_dc, 10, sse2)
 PRED8x8L(top_dc, 10, sse2)
-PRED8x8L(top_dc, 10, ssse3)
+PRED8x8L(top_dc, 10, avx)
 PRED8x8L(vertical, 10, sse2)
-PRED8x8L(vertical, 10, ssse3)
+PRED8x8L(vertical, 10, avx)
 PRED8x8L(horizontal, 10, sse2)
 PRED8x8L(horizontal, 10, ssse3)
+PRED8x8L(horizontal, 10, avx)
 PRED8x8L(down_left, 10, sse2)
 PRED8x8L(down_left, 10, ssse3)
+PRED8x8L(down_left, 10, avx)
 PRED8x8L(down_right, 10, sse2)
 PRED8x8L(down_right, 10, ssse3)
+PRED8x8L(down_right, 10, avx)
 PRED8x8L(vertical_right, 10, sse2)
 PRED8x8L(vertical_right, 10, ssse3)
+PRED8x8L(vertical_right, 10, avx)
 PRED8x8L(horizontal_up, 10, sse2)
 PRED8x8L(horizontal_up, 10, ssse3)
+PRED8x8L(horizontal_up, 10, avx)
 
 #define PRED16x16(TYPE, DEPTH, OPT)\
 void ff_pred16x16_ ## TYPE ## _ ## DEPTH ## _ ## OPT (uint8_t *src, int stride);
@@ -298,7 +302,6 @@ void ff_h264_pred_init_x86(H264PredContext *h, int codec_id, const int bit_depth
             h->pred4x4[HOR_UP_PRED         ] = ff_pred4x4_horizontal_up_10_mmxext;
 
             h->pred8x8[DC_PRED8x8          ] = ff_pred8x8_dc_10_mmxext;
-            h->pred8x8[TOP_DC_PRED8x8      ] = ff_pred8x8_top_dc_10_mmxext;
 
             h->pred8x8l[DC_128_PRED        ] = ff_pred8x8l_128_dc_10_mmxext;
 
@@ -344,18 +347,28 @@ void ff_h264_pred_init_x86(H264PredContext *h, int codec_id, const int bit_depth
             h->pred4x4[VERT_RIGHT_PRED     ] = ff_pred4x4_vertical_right_10_ssse3;
             h->pred4x4[HOR_DOWN_PRED       ] = ff_pred4x4_horizontal_down_10_ssse3;
 
-            h->pred8x8l[VERT_PRED           ] = ff_pred8x8l_vertical_10_ssse3;
             h->pred8x8l[HOR_PRED            ] = ff_pred8x8l_horizontal_10_ssse3;
-            h->pred8x8l[DC_PRED             ] = ff_pred8x8l_dc_10_ssse3;
-            h->pred8x8l[TOP_DC_PRED         ] = ff_pred8x8l_top_dc_10_ssse3;
             h->pred8x8l[DIAG_DOWN_LEFT_PRED ] = ff_pred8x8l_down_left_10_ssse3;
+            h->pred8x8l[DIAG_DOWN_RIGHT_PRED] = ff_pred8x8l_down_right_10_ssse3;
+            h->pred8x8l[VERT_RIGHT_PRED     ] = ff_pred8x8l_vertical_right_10_ssse3;
+            h->pred8x8l[HOR_UP_PRED         ] = ff_pred8x8l_horizontal_up_10_ssse3;
         }
 #if HAVE_AVX
         if (mm_flags & AV_CPU_FLAG_AVX) {
             h->pred4x4[DIAG_DOWN_LEFT_PRED ] = ff_pred4x4_down_left_10_avx;
             h->pred4x4[DIAG_DOWN_RIGHT_PRED] = ff_pred4x4_down_right_10_avx;
+            h->pred4x4[VERT_LEFT_PRED      ] = ff_pred4x4_vertical_left_10_avx;
             h->pred4x4[VERT_RIGHT_PRED     ] = ff_pred4x4_vertical_right_10_avx;
             h->pred4x4[HOR_DOWN_PRED       ] = ff_pred4x4_horizontal_down_10_avx;
+
+            h->pred8x8l[VERT_PRED           ] = ff_pred8x8l_vertical_10_avx;
+            h->pred8x8l[HOR_PRED            ] = ff_pred8x8l_horizontal_10_avx;
+            h->pred8x8l[DC_PRED             ] = ff_pred8x8l_dc_10_avx;
+            h->pred8x8l[TOP_DC_PRED         ] = ff_pred8x8l_top_dc_10_avx;
+            h->pred8x8l[DIAG_DOWN_RIGHT_PRED] = ff_pred8x8l_down_right_10_avx;
+            h->pred8x8l[DIAG_DOWN_LEFT_PRED ] = ff_pred8x8l_down_left_10_avx;
+            h->pred8x8l[VERT_RIGHT_PRED     ] = ff_pred8x8l_vertical_right_10_avx;
+            h->pred8x8l[HOR_UP_PRED         ] = ff_pred8x8l_horizontal_up_10_avx;
         }
 #endif /* HAVE_AVX */
     }