lavc: Edge emulation with dst/src linesize
authorRonald S. Bultje <rbultje@gmail.com>
Mon, 14 Oct 2013 01:13:10 +0000 (03:13 +0200)
committerLuca Barbato <lu_zero@gentoo.org>
Fri, 15 Nov 2013 09:16:27 +0000 (10:16 +0100)
Allow supporting files for which the image stride is smaller than
the maximum block size + number of subpel mc taps, e.g. a 64x64 VP9
file or a 16x16 VP8 file with -fflags +emu_edge.

16 files changed:
libavcodec/cavs.c
libavcodec/h264.c
libavcodec/hevc.c
libavcodec/mpegvideo_enc.c
libavcodec/mpegvideo_motion.c
libavcodec/rv34.c
libavcodec/svq3.c
libavcodec/vc1dec.c
libavcodec/videodsp.h
libavcodec/videodsp_template.c
libavcodec/vp3.c
libavcodec/vp56.c
libavcodec/vp8.c
libavcodec/wmv2.c
libavcodec/x86/videodsp.asm
libavcodec/x86/videodsp_init.c

index 19b9929..5a81089 100644 (file)
@@ -408,7 +408,8 @@ static inline void mc_dir_part(AVSContext *h, AVFrame *pic, int chroma_height,
         full_mx + 16 /* FIXME */ > pic_width + extra_width ||
         full_my + 16 /* FIXME */ > pic_height + extra_height) {
         h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
-                                 src_y - 2 - 2 * h->l_stride, h->l_stride,
+                                 src_y - 2 - 2 * h->l_stride,
+                                 h->l_stride, h->l_stride,
                                  16 + 5, 16 + 5 /* FIXME */,
                                  full_mx - 2, full_my - 2,
                                  pic_width, pic_height);
@@ -421,7 +422,7 @@ static inline void mc_dir_part(AVSContext *h, AVFrame *pic, int chroma_height,
 
     if (emu) {
         h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cb,
-                                 h->c_stride,
+                                 h->c_stride, h->c_stride,
                                  9, 9 /* FIXME */,
                                  mx >> 3, my >> 3,
                                  pic_width >> 1, pic_height >> 1);
@@ -431,7 +432,7 @@ static inline void mc_dir_part(AVSContext *h, AVFrame *pic, int chroma_height,
 
     if (emu) {
         h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cr,
-                                 h->c_stride,
+                                 h->c_stride, h->c_stride,
                                  9, 9 /* FIXME */,
                                  mx >> 3, my >> 3,
                                  pic_width >> 1, pic_height >> 1);
index 1444fc6..6f6727b 100644 (file)
@@ -899,7 +899,7 @@ static av_always_inline void mc_dir_part(H264Context *h, Picture *pic,
         full_my + 16 /*FIXME*/ > pic_height + extra_height) {
         h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
                                  src_y - (2 << pixel_shift) - 2 * h->mb_linesize,
-                                 h->mb_linesize,
+                                 h->mb_linesize, h->mb_linesize,
                                  16 + 5, 16 + 5 /*FIXME*/, full_mx - 2,
                                  full_my - 2, pic_width, pic_height);
         src_y = h->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
@@ -918,7 +918,7 @@ static av_always_inline void mc_dir_part(H264Context *h, Picture *pic,
         if (emu) {
             h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
                                      src_cb - (2 << pixel_shift) - 2 * h->mb_linesize,
-                                     h->mb_linesize,
+                                     h->mb_linesize, h->mb_linesize,
                                      16 + 5, 16 + 5 /*FIXME*/,
                                      full_mx - 2, full_my - 2,
                                      pic_width, pic_height);
@@ -932,7 +932,7 @@ static av_always_inline void mc_dir_part(H264Context *h, Picture *pic,
         if (emu) {
             h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
                                      src_cr - (2 << pixel_shift) - 2 * h->mb_linesize,
-                                     h->mb_linesize,
+                                     h->mb_linesize, h->mb_linesize,
                                      16 + 5, 16 + 5 /*FIXME*/,
                                      full_mx - 2, full_my - 2,
                                      pic_width, pic_height);
@@ -957,7 +957,8 @@ static av_always_inline void mc_dir_part(H264Context *h, Picture *pic,
              (my >> ysh) * h->mb_uvlinesize;
 
     if (emu) {
-        h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cb, h->mb_uvlinesize,
+        h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cb,
+                                 h->mb_uvlinesize, h->mb_uvlinesize,
                                  9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
                                  pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
         src_cb = h->edge_emu_buffer;
@@ -967,7 +968,8 @@ static av_always_inline void mc_dir_part(H264Context *h, Picture *pic,
               mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
 
     if (emu) {
-        h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cr, h->mb_uvlinesize,
+        h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cr,
+                                 h->mb_uvlinesize, h->mb_uvlinesize,
                                  9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
                                  pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
         src_cr = h->edge_emu_buffer;
index e01db16..f00667c 100644 (file)
@@ -1481,7 +1481,8 @@ static void luma_mc(HEVCContext *s, int16_t *dst, ptrdiff_t dststride,
         y_off >= pic_height - block_h - ff_hevc_qpel_extra_after[my]) {
         int offset = extra_top * srcstride + (extra_left << s->sps->pixel_shift);
 
-        s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src - offset, srcstride,
+        s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src - offset,
+                                 srcstride, srcstride,
                                  block_w + ff_hevc_qpel_extra[mx],
                                  block_h + ff_hevc_qpel_extra[my],
                                  x_off - extra_left, y_off - extra_top,
@@ -1532,7 +1533,8 @@ static void chroma_mc(HEVCContext *s, int16_t *dst1, int16_t *dst2,
         int offset1 = EPEL_EXTRA_BEFORE * (src1stride + (1 << s->sps->pixel_shift));
         int offset2 = EPEL_EXTRA_BEFORE * (src2stride + (1 << s->sps->pixel_shift));
 
-        s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src1 - offset1, src1stride,
+        s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src1 - offset1,
+                                 src1stride, src1stride,
                                  block_w + EPEL_EXTRA, block_h + EPEL_EXTRA,
                                  x_off - EPEL_EXTRA_BEFORE,
                                  y_off - EPEL_EXTRA_BEFORE,
@@ -1542,7 +1544,8 @@ static void chroma_mc(HEVCContext *s, int16_t *dst1, int16_t *dst2,
         s->hevcdsp.put_hevc_epel[!!my][!!mx](dst1, dststride, src1, src1stride,
                                              block_w, block_h, mx, my, lc->mc_buffer);
 
-        s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src2 - offset2, src2stride,
+        s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src2 - offset2,
+                                 src2stride, src2stride,
                                  block_w + EPEL_EXTRA, block_h + EPEL_EXTRA,
                                  x_off - EPEL_EXTRA_BEFORE,
                                  y_off - EPEL_EXTRA_BEFORE,
index 1acd0ff..b7dda07 100644 (file)
@@ -1699,15 +1699,19 @@ static av_always_inline void encode_mb_internal(MpegEncContext *s,
 
     if (mb_x * 16 + 16 > s->width || mb_y * 16 + 16 > s->height) {
         uint8_t *ebuf = s->edge_emu_buffer + 32;
-        s->vdsp.emulated_edge_mc(ebuf, ptr_y, wrap_y, 16, 16, mb_x * 16,
-                                 mb_y * 16, s->width, s->height);
+        s->vdsp.emulated_edge_mc(ebuf, ptr_y,
+                                 wrap_y, wrap_y,
+                                 16, 16, mb_x * 16, mb_y * 16,
+                                 s->width, s->height);
         ptr_y = ebuf;
-        s->vdsp.emulated_edge_mc(ebuf + 18 * wrap_y, ptr_cb, wrap_c, 8,
-                                 mb_block_height, mb_x * 8, mb_y * 8,
+        s->vdsp.emulated_edge_mc(ebuf + 18 * wrap_y, ptr_cb,
+                                 wrap_c, wrap_c,
+                                 8, mb_block_height, mb_x * 8, mb_y * 8,
                                  s->width >> 1, s->height >> 1);
         ptr_cb = ebuf + 18 * wrap_y;
-        s->vdsp.emulated_edge_mc(ebuf + 18 * wrap_y + 8, ptr_cr, wrap_c, 8,
-                                 mb_block_height, mb_x * 8, mb_y * 8,
+        s->vdsp.emulated_edge_mc(ebuf + 18 * wrap_y + 8, ptr_cr,
+                                 wrap_c, wrap_c,
+                                 8, mb_block_height, mb_x * 8, mb_y * 8,
                                  s->width >> 1, s->height >> 1);
         ptr_cr = ebuf + 18 * wrap_y + 8;
     }
index df7ddf7..fafbaa9 100644 (file)
@@ -63,7 +63,7 @@ static void gmc1_motion(MpegEncContext *s,
         if ((unsigned)src_x >= FFMAX(s->h_edge_pos - 17, 0) ||
             (unsigned)src_y >= FFMAX(s->v_edge_pos - 17, 0)) {
             s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
-                                     linesize,
+                                     linesize, linesize,
                                      17, 17,
                                      src_x, src_y,
                                      s->h_edge_pos, s->v_edge_pos);
@@ -109,7 +109,7 @@ static void gmc1_motion(MpegEncContext *s,
         if ((unsigned)src_x >= FFMAX((s->h_edge_pos >> 1) - 9, 0) ||
             (unsigned)src_y >= FFMAX((s->v_edge_pos >> 1) - 9, 0)) {
             s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
-                                     uvlinesize,
+                                     uvlinesize, uvlinesize,
                                      9, 9,
                                      src_x, src_y,
                                      s->h_edge_pos >> 1, s->v_edge_pos >> 1);
@@ -123,7 +123,7 @@ static void gmc1_motion(MpegEncContext *s,
     ptr = ref_picture[2] + offset;
     if (emu) {
         s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
-                                 uvlinesize,
+                                 uvlinesize, uvlinesize,
                                  9, 9,
                                  src_x, src_y,
                                  s->h_edge_pos >> 1, s->v_edge_pos >> 1);
@@ -216,7 +216,7 @@ static inline int hpel_motion(MpegEncContext *s,
         if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 1) - 8, 0) ||
             (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y & 1) - 8, 0)) {
             s->vdsp.emulated_edge_mc(s->edge_emu_buffer, src,
-                                     s->linesize,
+                                     s->linesize, s->linesize,
                                      9, 9,
                                      src_x, src_y, s->h_edge_pos,
                                      s->v_edge_pos);
@@ -321,7 +321,7 @@ void mpeg_motion_internal(MpegEncContext *s,
             return;
         }
         s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y,
-                                 s->linesize,
+                                 s->linesize, s->linesize,
                                  17, 17 + field_based,
                                  src_x, src_y << field_based,
                                  s->h_edge_pos, s->v_edge_pos);
@@ -329,12 +329,12 @@ void mpeg_motion_internal(MpegEncContext *s,
         if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
             uint8_t *uvbuf = s->edge_emu_buffer + 18 * s->linesize;
             s->vdsp.emulated_edge_mc(uvbuf, ptr_cb,
-                                     s->uvlinesize,
+                                     s->uvlinesize, s->uvlinesize,
                                      9, 9 + field_based,
                                      uvsrc_x, uvsrc_y << field_based,
                                      s->h_edge_pos >> 1, s->v_edge_pos >> 1);
             s->vdsp.emulated_edge_mc(uvbuf + 16, ptr_cr,
-                                     s->uvlinesize,
+                                     s->uvlinesize, s->uvlinesize,
                                      9, 9 + field_based,
                                      uvsrc_x, uvsrc_y << field_based,
                                      s->h_edge_pos >> 1, s->v_edge_pos >> 1);
@@ -542,7 +542,7 @@ static inline void qpel_motion(MpegEncContext *s,
     if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 16, 0) ||
         (unsigned)src_y > FFMAX(v_edge_pos - (motion_y & 3) - h, 0)) {
         s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y,
-                                 s->linesize,
+                                 s->linesize, s->linesize,
                                  17, 17 + field_based,
                                  src_x, src_y << field_based,
                                  s->h_edge_pos, s->v_edge_pos);
@@ -550,12 +550,12 @@ static inline void qpel_motion(MpegEncContext *s,
         if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
             uint8_t *uvbuf = s->edge_emu_buffer + 18 * s->linesize;
             s->vdsp.emulated_edge_mc(uvbuf, ptr_cb,
-                                     s->uvlinesize,
+                                     s->uvlinesize, s->uvlinesize,
                                      9, 9 + field_based,
                                      uvsrc_x, uvsrc_y << field_based,
                                      s->h_edge_pos >> 1, s->v_edge_pos >> 1);
             s->vdsp.emulated_edge_mc(uvbuf + 16, ptr_cr,
-                                     s->uvlinesize,
+                                     s->uvlinesize, s->uvlinesize,
                                      9, 9 + field_based,
                                      uvsrc_x, uvsrc_y << field_based,
                                      s->h_edge_pos >> 1, s->v_edge_pos >> 1);
@@ -625,8 +625,8 @@ static void chroma_4mv_motion(MpegEncContext *s,
     if (s->flags & CODEC_FLAG_EMU_EDGE) {
         if ((unsigned)src_x > FFMAX((s->h_edge_pos >> 1) - (dxy & 1) - 8, 0) ||
             (unsigned)src_y > FFMAX((s->v_edge_pos >> 1) - (dxy >> 1) - 8, 0)) {
-            s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
-                                     ptr, s->uvlinesize,
+            s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
+                                     s->uvlinesize, s->uvlinesize,
                                      9, 9, src_x, src_y,
                                      s->h_edge_pos >> 1, s->v_edge_pos >> 1);
             ptr = s->edge_emu_buffer;
@@ -637,8 +637,8 @@ static void chroma_4mv_motion(MpegEncContext *s,
 
     ptr = ref_picture[2] + offset;
     if (emu) {
-        s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
-                                 ptr, s->uvlinesize,
+        s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
+                                 s->uvlinesize, s->uvlinesize,
                                  9, 9, src_x, src_y,
                                  s->h_edge_pos >> 1, s->v_edge_pos >> 1);
         ptr = s->edge_emu_buffer;
@@ -786,7 +786,7 @@ static inline void apply_8x8(MpegEncContext *s,
                 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 8, 0) ||
                     (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y & 3) - 8, 0)) {
                     s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
-                                             s->linesize,
+                                             s->linesize, s->linesize,
                                              9, 9,
                                              src_x, src_y,
                                              s->h_edge_pos,
index 8038f8f..c02ea74 100644 (file)
@@ -723,12 +723,18 @@ static inline void rv34_mc(RV34DecContext *r, const int block_type,
         uint8_t *uvbuf = s->edge_emu_buffer + 22 * s->linesize;
 
         srcY -= 2 + 2*s->linesize;
-        s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, (width<<3)+6, (height<<3)+6,
+        s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY,
+                                 s->linesize, s->linesize,
+                                 (width << 3) + 6, (height << 3) + 6,
                             src_x - 2, src_y - 2, s->h_edge_pos, s->v_edge_pos);
         srcY = s->edge_emu_buffer + 2 + 2*s->linesize;
-        s->vdsp.emulated_edge_mc(uvbuf     , srcU, s->uvlinesize, (width<<2)+1, (height<<2)+1,
+        s->vdsp.emulated_edge_mc(uvbuf, srcU,
+                                 s->uvlinesize,s->uvlinesize,
+                                 (width << 2) + 1, (height << 2) + 1,
                             uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
-        s->vdsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, (width<<2)+1, (height<<2)+1,
+        s->vdsp.emulated_edge_mc(uvbuf + 16, srcV,
+                                 s->uvlinesize, s->uvlinesize,
+                                 (width << 2) + 1, (height << 2) + 1,
                             uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
         srcU = uvbuf;
         srcV = uvbuf + 16;
index 775e071..cdff5af 100644 (file)
@@ -312,7 +312,8 @@ static inline void svq3_mc_dir_part(SVQ3Context *s,
     src  = pic->f.data[0] + mx + my * h->linesize;
 
     if (emu) {
-        h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src, h->linesize,
+        h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src,
+                                 h->linesize, h->linesize,
                                  width + 1, height + 1,
                                  mx, my, s->h_edge_pos, s->v_edge_pos);
         src = h->edge_emu_buffer;
@@ -338,7 +339,8 @@ static inline void svq3_mc_dir_part(SVQ3Context *s,
             src  = pic->f.data[i] + mx + my * h->uvlinesize;
 
             if (emu) {
-                h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src, h->uvlinesize,
+                h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src,
+                                         h->uvlinesize, h->uvlinesize,
                                          width + 1, height + 1,
                                          mx, my, (s->h_edge_pos >> 1),
                                          s->v_edge_pos >> 1);
index 231f425..0f22ebe 100644 (file)
@@ -452,14 +452,19 @@ static void vc1_mc_1mv(VC1Context *v, int dir)
         uint8_t *uvbuf = s->edge_emu_buffer + 19 * s->linesize;
 
         srcY -= s->mspel * (1 + s->linesize);
-        s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize,
+        s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY,
+                                 s->linesize, s->linesize,
                                  17 + s->mspel * 2, 17 + s->mspel * 2,
                                  src_x - s->mspel, src_y - s->mspel,
                                  s->h_edge_pos, v_edge_pos);
         srcY = s->edge_emu_buffer;
-        s->vdsp.emulated_edge_mc(uvbuf     , srcU, s->uvlinesize, 8 + 1, 8 + 1,
+        s->vdsp.emulated_edge_mc(uvbuf, srcU,
+                                 s->uvlinesize, s->uvlinesize,
+                                 8 + 1, 8 + 1,
                                  uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
-        s->vdsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, 8 + 1, 8 + 1,
+        s->vdsp.emulated_edge_mc(uvbuf + 16, srcV,
+                                 s->uvlinesize, s->uvlinesize,
+                                 8 + 1, 8 + 1,
                                  uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
         srcU = uvbuf;
         srcV = uvbuf + 16;
@@ -693,7 +698,8 @@ static void vc1_mc_4mv_luma(VC1Context *v, int n, int dir, int avg)
         || (unsigned)(src_y - (s->mspel << fieldmv)) > v_edge_pos - (my & 3) - ((8 + s->mspel * 2) << fieldmv)) {
         srcY -= s->mspel * (1 + (s->linesize << fieldmv));
         /* check emulate edge stride and offset */
-        s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize,
+        s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY,
+                                 s->linesize, s->linesize,
                                  9 + s->mspel * 2, (9 + s->mspel * 2) << fieldmv,
                                  src_x - s->mspel, src_y - (s->mspel << fieldmv),
                                  s->h_edge_pos, v_edge_pos);
@@ -908,10 +914,12 @@ static void vc1_mc_4mv_chroma(VC1Context *v, int dir)
         || s->h_edge_pos < 18 || v_edge_pos < 18
         || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 9
         || (unsigned)uvsrc_y > (v_edge_pos    >> 1) - 9) {
-        s->vdsp.emulated_edge_mc(s->edge_emu_buffer     , srcU, s->uvlinesize,
+        s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcU,
+                                 s->uvlinesize, s->uvlinesize,
                                  8 + 1, 8 + 1, uvsrc_x, uvsrc_y,
                                  s->h_edge_pos >> 1, v_edge_pos >> 1);
-        s->vdsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV, s->uvlinesize,
+        s->vdsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV,
+                                 s->uvlinesize, s->uvlinesize,
                                  8 + 1, 8 + 1, uvsrc_x, uvsrc_y,
                                  s->h_edge_pos >> 1, v_edge_pos >> 1);
         srcU = s->edge_emu_buffer;
@@ -1024,10 +1032,12 @@ static void vc1_mc_4mv_chroma4(VC1Context *v, int dir, int dir2, int avg)
             || s->h_edge_pos < 10 || v_edge_pos < (5 << fieldmv)
             || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 5
             || (unsigned)uvsrc_y > v_edge_pos - (5 << fieldmv)) {
-            s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcU, s->uvlinesize,
+            s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcU,
+                                     s->uvlinesize, s->uvlinesize,
                                      5, (5 << fieldmv), uvsrc_x, uvsrc_y,
                                      s->h_edge_pos >> 1, v_edge_pos);
-            s->vdsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV, s->uvlinesize,
+            s->vdsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV,
+                                     s->uvlinesize, s->uvlinesize,
                                      5, (5 << fieldmv), uvsrc_x, uvsrc_y,
                                      s->h_edge_pos >> 1, v_edge_pos);
             srcU = s->edge_emu_buffer;
@@ -1955,14 +1965,19 @@ static void vc1_interp_mc(VC1Context *v)
         uint8_t *uvbuf = s->edge_emu_buffer + 19 * s->linesize;
 
         srcY -= s->mspel * (1 + s->linesize);
-        s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize,
+        s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY,
+                                 s->linesize, s->linesize,
                                  17 + s->mspel * 2, 17 + s->mspel * 2,
                                  src_x - s->mspel, src_y - s->mspel,
                                  s->h_edge_pos, v_edge_pos);
         srcY = s->edge_emu_buffer;
-        s->vdsp.emulated_edge_mc(uvbuf     , srcU, s->uvlinesize, 8 + 1, 8 + 1,
+        s->vdsp.emulated_edge_mc(uvbuf, srcU,
+                                 s->uvlinesize, s->uvlinesize,
+                                 8 + 1, 8 + 1,
                                  uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
-        s->vdsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, 8 + 1, 8 + 1,
+        s->vdsp.emulated_edge_mc(uvbuf + 16, srcV,
+                                 s->uvlinesize, s->uvlinesize,
+                                 8 + 1, 8 + 1,
                                  uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
         srcU = uvbuf;
         srcV = uvbuf + 16;
index bd2f763..2211c5d 100644 (file)
@@ -36,8 +36,10 @@ typedef struct VideoDSPContext {
      *
      * @param buf destination buffer
      * @param src source buffer
-     * @param linesize number of bytes between 2 vertically adjacent samples
-     *                 in both the source and destination buffers
+     * @param buf_linesize number of bytes between 2 vertically adjacent
+     *                     samples in the destination buffer
+     * @param src_linesize number of bytes between 2 vertically adjacent
+     *                     samples in both the source buffer
      * @param block_w width of block
      * @param block_h height of block
      * @param src_x x coordinate of the top left sample of the block in the
@@ -48,7 +50,9 @@ typedef struct VideoDSPContext {
      * @param h height of the source buffer
      */
     void (*emulated_edge_mc)(uint8_t *buf, const uint8_t *src,
-                             ptrdiff_t linesize, int block_w, int block_h,
+                             ptrdiff_t buf_linesize,
+                             ptrdiff_t src_linesize,
+                             int block_w, int block_h,
                              int src_x, int src_y, int w, int h);
 
     /**
index 5d5a32f..98313ac 100644 (file)
@@ -22,7 +22,8 @@
 #include "bit_depth_template.c"
 
 static void FUNC(ff_emulated_edge_mc)(uint8_t *buf, const uint8_t *src,
-                                      ptrdiff_t linesize,
+                                      ptrdiff_t buf_linesize,
+                                      ptrdiff_t src_linesize,
                                       int block_w, int block_h,
                                       int src_x, int src_y, int w, int h)
 {
@@ -30,10 +31,10 @@ static void FUNC(ff_emulated_edge_mc)(uint8_t *buf, const uint8_t *src,
     int start_y, start_x, end_y, end_x;
 
     if (src_y >= h) {
-        src  += (h - 1 - src_y) * linesize;
+        src  += (h - 1 - src_y) * src_linesize;
         src_y = h - 1;
     } else if (src_y <= -block_h) {
-        src  += (1 - block_h - src_y) * linesize;
+        src  += (1 - block_h - src_y) * src_linesize;
         src_y = 1 - block_h;
     }
     if (src_x >= w) {
@@ -52,30 +53,30 @@ static void FUNC(ff_emulated_edge_mc)(uint8_t *buf, const uint8_t *src,
     assert(start_x < end_x && block_w);
 
     w    = end_x - start_x;
-    src += start_y * linesize + start_x * sizeof(pixel);
+    src += start_y * src_linesize + start_x * sizeof(pixel);
     buf += start_x * sizeof(pixel);
 
     // top
     for (y = 0; y < start_y; y++) {
         memcpy(buf, src, w * sizeof(pixel));
-        buf += linesize;
+        buf += buf_linesize;
     }
 
     // copy existing part
     for (; y < end_y; y++) {
         memcpy(buf, src, w * sizeof(pixel));
-        src += linesize;
-        buf += linesize;
+        src += src_linesize;
+        buf += buf_linesize;
     }
 
     // bottom
-    src -= linesize;
+    src -= src_linesize;
     for (; y < block_h; y++) {
         memcpy(buf, src, w * sizeof(pixel));
-        buf += linesize;
+        buf += buf_linesize;
     }
 
-    buf -= block_h * linesize + start_x * sizeof(pixel);
+    buf -= block_h * buf_linesize + start_x * sizeof(pixel);
     while (block_h--) {
         pixel *bufp = (pixel *) buf;
 
@@ -88,6 +89,6 @@ static void FUNC(ff_emulated_edge_mc)(uint8_t *buf, const uint8_t *src,
         for (x = end_x; x < block_w; x++) {
             bufp[x] = bufp[end_x - 1];
         }
-        buf += linesize;
+        buf += buf_linesize;
     }
 }
index d596816..adcecbc 100644 (file)
@@ -1543,7 +1543,11 @@ static void render_slice(Vp3DecodeContext *s, int slice)
                             uint8_t *temp= s->edge_emu_buffer;
                             if(stride<0) temp -= 8*stride;
 
-                            s->vdsp.emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
+                            s->vdsp.emulated_edge_mc(temp, motion_source,
+                                                     stride, stride,
+                                                     9, 9, src_x, src_y,
+                                                     plane_width,
+                                                     plane_height);
                             motion_source= temp;
                         }
                     }
index 53a0271..7fbf9a0 100644 (file)
@@ -341,7 +341,8 @@ static void vp56_mc(VP56Context *s, int b, int plane, uint8_t *src,
         y<0 || y+12>=s->plane_height[plane]) {
         s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
                             src + s->block_offset[b] + (dy-2)*stride + (dx-2),
-                            stride, 12, 12, x, y,
+                            stride, stride,
+                            12, 12, x, y,
                             s->plane_width[plane],
                             s->plane_height[plane]);
         src_block = s->edge_emu_buffer;
index b1b4f2d..2660395 100644 (file)
@@ -1196,7 +1196,9 @@ void vp8_mc_luma(VP8Context *s, VP8ThreadData *td, uint8_t *dst,
         src += y_off * linesize + x_off;
         if (x_off < mx_idx || x_off >= width  - block_w - subpel_idx[2][mx] ||
             y_off < my_idx || y_off >= height - block_h - subpel_idx[2][my]) {
-            s->vdsp.emulated_edge_mc(td->edge_emu_buffer, src - my_idx * linesize - mx_idx, linesize,
+            s->vdsp.emulated_edge_mc(td->edge_emu_buffer,
+                                     src - my_idx * linesize - mx_idx,
+                                     linesize, linesize,
                                      block_w + subpel_idx[1][mx], block_h + subpel_idx[1][my],
                                      x_off - mx_idx, y_off - my_idx, width, height);
             src = td->edge_emu_buffer + mx_idx + linesize * my_idx;
@@ -1246,13 +1248,17 @@ void vp8_mc_chroma(VP8Context *s, VP8ThreadData *td, uint8_t *dst1, uint8_t *dst
         ff_thread_await_progress(ref, (3 + y_off + block_h + subpel_idx[2][my]) >> 3, 0);
         if (x_off < mx_idx || x_off >= width  - block_w - subpel_idx[2][mx] ||
             y_off < my_idx || y_off >= height - block_h - subpel_idx[2][my]) {
-            s->vdsp.emulated_edge_mc(td->edge_emu_buffer, src1 - my_idx * linesize - mx_idx, linesize,
+            s->vdsp.emulated_edge_mc(td->edge_emu_buffer,
+                                     src1 - my_idx * linesize - mx_idx,
+                                     linesize, linesize,
                                      block_w + subpel_idx[1][mx], block_h + subpel_idx[1][my],
                                      x_off - mx_idx, y_off - my_idx, width, height);
             src1 = td->edge_emu_buffer + mx_idx + linesize * my_idx;
             mc_func[my_idx][mx_idx](dst1, linesize, src1, linesize, block_h, mx, my);
 
-            s->vdsp.emulated_edge_mc(td->edge_emu_buffer, src2 - my_idx * linesize - mx_idx, linesize,
+            s->vdsp.emulated_edge_mc(td->edge_emu_buffer,
+                                     src2 - my_idx * linesize - mx_idx,
+                                     linesize, linesize,
                                      block_w + subpel_idx[1][mx], block_h + subpel_idx[1][my],
                                      x_off - mx_idx, y_off - my_idx, width, height);
             src2 = td->edge_emu_buffer + mx_idx + linesize * my_idx;
index f4bc907..b330776 100644 (file)
@@ -120,8 +120,12 @@ void ff_mspel_motion(MpegEncContext *s,
     if(s->flags&CODEC_FLAG_EMU_EDGE){
         if(src_x<1 || src_y<1 || src_x + 17  >= s->h_edge_pos
                               || src_y + h+1 >= v_edge_pos){
-            s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr - 1 - s->linesize, s->linesize, 19, 19,
-                             src_x-1, src_y-1, s->h_edge_pos, s->v_edge_pos);
+            s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
+                                     ptr - 1 - s->linesize,
+                                     s->linesize, s->linesize,
+                                     19, 19,
+                                     src_x - 1, src_y - 1,
+                                     s->h_edge_pos, s->v_edge_pos);
             ptr= s->edge_emu_buffer + 1 + s->linesize;
             emu=1;
         }
@@ -161,16 +165,22 @@ void ff_mspel_motion(MpegEncContext *s,
     offset = (src_y * uvlinesize) + src_x;
     ptr = ref_picture[1] + offset;
     if(emu){
-        s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9,
-                         src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1);
+        s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
+                                 s->uvlinesize, s->uvlinesize,
+                                 9, 9,
+                                 src_x, src_y,
+                                 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
         ptr= s->edge_emu_buffer;
     }
     pix_op[1][dxy](dest_cb, ptr, uvlinesize, h >> 1);
 
     ptr = ref_picture[2] + offset;
     if(emu){
-        s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9,
-                         src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1);
+        s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
+                                 s->uvlinesize, s->uvlinesize,
+                                 9, 9,
+                                 src_x, src_y,
+                                 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
         ptr= s->edge_emu_buffer;
     }
     pix_op[1][dxy](dest_cr, ptr, uvlinesize, h >> 1);
dissimilarity index 84%
index 34f007f..59f1937 100644 (file)
-;******************************************************************************
-;* Core video DSP functions
-;* 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
-;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
-;******************************************************************************
-
-%include "libavutil/x86/x86util.asm"
-
-SECTION .text
-
-; void ff_emu_edge_core(uint8_t *buf, const uint8_t *src, x86_reg linesize,
-;                       x86_reg start_y, x86_reg end_y, x86_reg block_h,
-;                       x86_reg start_x, x86_reg end_x, x86_reg block_w);
-;
-; The actual function itself is below. It basically wraps a very simple
-; w = end_x - start_x
-; if (w) {
-;   if (w > 22) {
-;     jump to the slow loop functions
-;   } else {
-;     jump to the fast loop functions
-;   }
-; }
-;
-; ... and then the same for left/right extend also. See below for loop
-; function implementations. Fast are fixed-width, slow is variable-width
-
-%macro EMU_EDGE_FUNC 0
-%if ARCH_X86_64
-%define w_reg r7
-cglobal emu_edge_core, 6, 9, 1
-    mov         r8, r5          ; save block_h
-%else
-%define w_reg r6
-cglobal emu_edge_core, 2, 7, 0
-    mov         r4, r4m         ; end_y
-    mov         r5, r5m         ; block_h
-%endif
-
-    ; start with vertical extend (top/bottom) and body pixel copy
-    mov      w_reg, r7m
-    sub      w_reg, r6m         ; w = start_x - end_x
-    sub         r5, r4
-%if ARCH_X86_64
-    sub         r4, r3
-%else
-    sub         r4, dword r3m
-%endif
-    cmp      w_reg, 22
-    jg .slow_v_extend_loop
-%if ARCH_X86_32
-    mov         r2, r2m         ; linesize
-%endif
-    sal      w_reg, 7           ; w * 128
-%ifdef PIC
-    lea        rax, [.emuedge_v_extend_1 - (.emuedge_v_extend_2 - .emuedge_v_extend_1)]
-    add      w_reg, rax
-%else
-    lea      w_reg, [.emuedge_v_extend_1 - (.emuedge_v_extend_2 - .emuedge_v_extend_1)+w_reg]
-%endif
-    call     w_reg              ; fast top extend, body copy and bottom extend
-.v_extend_end:
-
-    ; horizontal extend (left/right)
-    mov      w_reg, r6m         ; start_x
-    sub         r0, w_reg
-%if ARCH_X86_64
-    mov         r3, r0          ; backup of buf+block_h*linesize
-    mov         r5, r8
-%else
-    mov        r0m, r0          ; backup of buf+block_h*linesize
-    mov         r5, r5m
-%endif
-    test     w_reg, w_reg
-    jz .right_extend
-    cmp      w_reg, 22
-    jg .slow_left_extend_loop
-    mov         r1, w_reg
-    dec      w_reg
-    ; FIXME we can do a if size == 1 here if that makes any speed difference, test me
-    sar      w_reg, 1
-    sal      w_reg, 6
-    ; r0=buf+block_h*linesize,r7(64)/r6(32)=start_x offset for funcs
-    ; r6(rax)/r3(ebx)=val,r2=linesize,r1=start_x,r5=block_h
-%ifdef PIC
-    lea        rax, [.emuedge_extend_left_2]
-    add      w_reg, rax
-%else
-    lea      w_reg, [.emuedge_extend_left_2+w_reg]
-%endif
-    call     w_reg
-
-    ; now r3(64)/r0(32)=buf,r2=linesize,r8/r5=block_h,r6/r3=val, r7/r6=end_x, r1=block_w
-.right_extend:
-%if ARCH_X86_32
-    mov         r0, r0m
-    mov         r5, r5m
-%endif
-    mov      w_reg, r7m         ; end_x
-    mov         r1, r8m         ; block_w
-    mov         r4, r1
-    sub         r1, w_reg
-    jz .h_extend_end            ; if (end_x == block_w) goto h_extend_end
-    cmp         r1, 22
-    jg .slow_right_extend_loop
-    dec         r1
-    ; FIXME we can do a if size == 1 here if that makes any speed difference, test me
-    sar         r1, 1
-    sal         r1, 6
-%ifdef PIC
-    lea        rax, [.emuedge_extend_right_2]
-    add         r1, rax
-%else
-    lea         r1, [.emuedge_extend_right_2+r1]
-%endif
-    call        r1
-.h_extend_end:
-    RET
-
-%if ARCH_X86_64
-%define vall  al
-%define valh  ah
-%define valw  ax
-%define valw2 r7w
-%define valw3 r3w
-%if WIN64
-%define valw4 r7w
-%else ; unix64
-%define valw4 r3w
-%endif
-%define vald eax
-%else
-%define vall  bl
-%define valh  bh
-%define valw  bx
-%define valw2 r6w
-%define valw3 valw2
-%define valw4 valw3
-%define vald ebx
-%define stack_offset 0x14
-%endif
-
-%endmacro
-
-; macro to read/write a horizontal number of pixels (%2) to/from registers
-; on x86-64, - fills xmm0-15 for consecutive sets of 16 pixels
-;            - if (%2 & 15 == 8) fills the last 8 bytes into rax
-;            - else if (%2 & 8)  fills 8 bytes into mm0
-;            - if (%2 & 7 == 4)  fills the last 4 bytes into rax
-;            - else if (%2 & 4)  fills 4 bytes into mm0-1
-;            - if (%2 & 3 == 3)  fills 2 bytes into r7/r3, and 1 into eax
-;              (note that we're using r3 for body/bottom because it's a shorter
-;               opcode, and then the loop fits in 128 bytes)
-;            - else              fills remaining bytes into rax
-; on x86-32, - fills mm0-7 for consecutive sets of 8 pixels
-;            - if (%2 & 7 == 4)  fills 4 bytes into ebx
-;            - else if (%2 & 4)  fills 4 bytes into mm0-7
-;            - if (%2 & 3 == 3)  fills 2 bytes into r6, and 1 into ebx
-;            - else              fills remaining bytes into ebx
-; writing data out is in the same way
-%macro READ_NUM_BYTES 2
-%assign %%src_off 0 ; offset in source buffer
-%assign %%smidx   0 ; mmx register idx
-%assign %%sxidx   0 ; xmm register idx
-
-%if cpuflag(sse)
-%rep %2/16
-    movups xmm %+ %%sxidx, [r1+%%src_off]
-%assign %%src_off %%src_off+16
-%assign %%sxidx   %%sxidx+1
-%endrep ; %2/16
-%endif
-
-%if ARCH_X86_64
-%if (%2-%%src_off) == 8
-    mov           rax, [r1+%%src_off]
-%assign %%src_off %%src_off+8
-%endif ; (%2-%%src_off) == 8
-%endif ; x86-64
-
-%rep (%2-%%src_off)/8
-    movq    mm %+ %%smidx, [r1+%%src_off]
-%assign %%src_off %%src_off+8
-%assign %%smidx   %%smidx+1
-%endrep ; (%2-%%dst_off)/8
-
-%if (%2-%%src_off) == 4
-    mov          vald, [r1+%%src_off]
-%elif (%2-%%src_off) & 4
-    movd    mm %+ %%smidx, [r1+%%src_off]
-%assign %%src_off %%src_off+4
-%endif ; (%2-%%src_off) ==/& 4
-
-%if (%2-%%src_off) == 1
-    mov          vall, [r1+%%src_off]
-%elif (%2-%%src_off) == 2
-    mov          valw, [r1+%%src_off]
-%elif (%2-%%src_off) == 3
-%ifidn %1, top
-    mov         valw2, [r1+%%src_off]
-%elifidn %1, body
-    mov         valw3, [r1+%%src_off]
-%elifidn %1, bottom
-    mov         valw4, [r1+%%src_off]
-%endif ; %1 ==/!= top
-    mov          vall, [r1+%%src_off+2]
-%endif ; (%2-%%src_off) == 1/2/3
-%endmacro ; READ_NUM_BYTES
-
-%macro WRITE_NUM_BYTES 2
-%assign %%dst_off 0 ; offset in destination buffer
-%assign %%dmidx   0 ; mmx register idx
-%assign %%dxidx   0 ; xmm register idx
-
-%if cpuflag(sse)
-%rep %2/16
-    movups [r0+%%dst_off], xmm %+ %%dxidx
-%assign %%dst_off %%dst_off+16
-%assign %%dxidx   %%dxidx+1
-%endrep ; %2/16
-%endif
-
-%if ARCH_X86_64
-%if (%2-%%dst_off) == 8
-    mov    [r0+%%dst_off], rax
-%assign %%dst_off %%dst_off+8
-%endif ; (%2-%%dst_off) == 8
-%endif ; x86-64
-
-%rep (%2-%%dst_off)/8
-    movq   [r0+%%dst_off], mm %+ %%dmidx
-%assign %%dst_off %%dst_off+8
-%assign %%dmidx   %%dmidx+1
-%endrep ; (%2-%%dst_off)/8
-
-%if (%2-%%dst_off) == 4
-    mov    [r0+%%dst_off], vald
-%elif (%2-%%dst_off) & 4
-    movd   [r0+%%dst_off], mm %+ %%dmidx
-%assign %%dst_off %%dst_off+4
-%endif ; (%2-%%dst_off) ==/& 4
-
-%if (%2-%%dst_off) == 1
-    mov    [r0+%%dst_off], vall
-%elif (%2-%%dst_off) == 2
-    mov    [r0+%%dst_off], valw
-%elif (%2-%%dst_off) == 3
-%ifidn %1, top
-    mov    [r0+%%dst_off], valw2
-%elifidn %1, body
-    mov    [r0+%%dst_off], valw3
-%elifidn %1, bottom
-    mov    [r0+%%dst_off], valw4
-%endif ; %1 ==/!= top
-    mov  [r0+%%dst_off+2], vall
-%endif ; (%2-%%dst_off) == 1/2/3
-%endmacro ; WRITE_NUM_BYTES
-
-; vertical top/bottom extend and body copy fast loops
-; these are function pointers to set-width line copy functions, i.e.
-; they read a fixed number of pixels into set registers, and write
-; those out into the destination buffer
-; r0=buf,r1=src,r2=linesize,r3(64)/r3m(32)=start_x,r4=end_y,r5=block_h
-; r6(eax/64)/r3(ebx/32)=val_reg
-%macro VERTICAL_EXTEND 0
-%assign %%n 1
-%rep 22
-ALIGN 128
-.emuedge_v_extend_ %+ %%n:
-    ; extend pixels above body
-%if ARCH_X86_64
-    test           r3 , r3                   ; if (!start_y)
-    jz .emuedge_copy_body_ %+ %%n %+ _loop   ;   goto body
-%else ; ARCH_X86_32
-    cmp      dword r3m, 0
-    je .emuedge_copy_body_ %+ %%n %+ _loop
-%endif ; ARCH_X86_64/32
-    READ_NUM_BYTES  top,    %%n              ; read bytes
-.emuedge_extend_top_ %+ %%n %+ _loop:        ; do {
-    WRITE_NUM_BYTES top,    %%n              ;   write bytes
-    add            r0 , r2                   ;   dst += linesize
-%if ARCH_X86_64
-    dec            r3d
-%else ; ARCH_X86_32
-    dec      dword r3m
-%endif ; ARCH_X86_64/32
-    jnz .emuedge_extend_top_ %+ %%n %+ _loop ; } while (--start_y)
-
-    ; copy body pixels
-.emuedge_copy_body_ %+ %%n %+ _loop:         ; do {
-    READ_NUM_BYTES  body,   %%n              ;   read bytes
-    WRITE_NUM_BYTES body,   %%n              ;   write bytes
-    add            r0 , r2                   ;   dst += linesize
-    add            r1 , r2                   ;   src += linesize
-    dec            r4d
-    jnz .emuedge_copy_body_ %+ %%n %+ _loop  ; } while (--end_y)
-
-    ; copy bottom pixels
-    test           r5 , r5                   ; if (!block_h)
-    jz .emuedge_v_extend_end_ %+ %%n         ;   goto end
-    sub            r1 , r2                   ; src -= linesize
-    READ_NUM_BYTES  bottom, %%n              ; read bytes
-.emuedge_extend_bottom_ %+ %%n %+ _loop:     ; do {
-    WRITE_NUM_BYTES bottom, %%n              ;   write bytes
-    add            r0 , r2                   ;   dst += linesize
-    dec            r5d
-    jnz .emuedge_extend_bottom_ %+ %%n %+ _loop ; } while (--block_h)
-
-.emuedge_v_extend_end_ %+ %%n:
-%if ARCH_X86_64
-    ret
-%else ; ARCH_X86_32
-    rep ret
-%endif ; ARCH_X86_64/32
-%assign %%n %%n+1
-%endrep
-%endmacro VERTICAL_EXTEND
-
-; left/right (horizontal) fast extend functions
-; these are essentially identical to the vertical extend ones above,
-; just left/right separated because number of pixels to extend is
-; obviously not the same on both sides.
-; for reading, pixels are placed in eax (x86-64) or ebx (x86-64) in the
-; lowest two bytes of the register (so val*0x0101), and are splatted
-; into each byte of mm0 as well if n_pixels >= 8
-
-%macro READ_V_PIXEL 2
-    mov        vall, %2
-    mov        valh, vall
-%if %1 >= 8
-    movd        mm0, vald
-%if cpuflag(mmxext)
-    pshufw      mm0, mm0, 0
-%else ; mmx
-    punpcklwd   mm0, mm0
-    punpckldq   mm0, mm0
-%endif ; sse
-%endif ; %1 >= 8
-%endmacro
-
-%macro WRITE_V_PIXEL 2
-%assign %%dst_off 0
-%rep %1/8
-    movq [%2+%%dst_off], mm0
-%assign %%dst_off %%dst_off+8
-%endrep
-%if %1 & 4
-%if %1 >= 8
-    movd [%2+%%dst_off], mm0
-%else ; %1 < 8
-    mov  [%2+%%dst_off]  , valw
-    mov  [%2+%%dst_off+2], valw
-%endif ; %1 >=/< 8
-%assign %%dst_off %%dst_off+4
-%endif ; %1 & 4
-%if %1&2
-    mov  [%2+%%dst_off], valw
-%endif ; %1 & 2
-%endmacro
-
-; r0=buf+block_h*linesize, r1=start_x, r2=linesize, r5=block_h, r6/r3=val
-%macro LEFT_EXTEND 0
-%assign %%n 2
-%rep 11
-ALIGN 64
-.emuedge_extend_left_ %+ %%n:          ; do {
-    sub         r0, r2                 ;   dst -= linesize
-    READ_V_PIXEL  %%n, [r0+r1]         ;   read pixels
-    WRITE_V_PIXEL %%n, r0              ;   write pixels
-    dec         r5
-    jnz .emuedge_extend_left_ %+ %%n   ; } while (--block_h)
-%if ARCH_X86_64
-    ret
-%else ; ARCH_X86_32
-    rep ret
-%endif ; ARCH_X86_64/32
-%assign %%n %%n+2
-%endrep
-%endmacro ; LEFT_EXTEND
-
-; r3/r0=buf+block_h*linesize, r2=linesize, r8/r5=block_h, r0/r6=end_x, r6/r3=val
-%macro RIGHT_EXTEND 0
-%assign %%n 2
-%rep 11
-ALIGN 64
-.emuedge_extend_right_ %+ %%n:          ; do {
-%if ARCH_X86_64
-    sub        r3, r2                   ;   dst -= linesize
-    READ_V_PIXEL  %%n, [r3+w_reg-1]     ;   read pixels
-    WRITE_V_PIXEL %%n, r3+r4-%%n        ;   write pixels
-    dec       r8
-%else ; ARCH_X86_32
-    sub        r0, r2                   ;   dst -= linesize
-    READ_V_PIXEL  %%n, [r0+w_reg-1]     ;   read pixels
-    WRITE_V_PIXEL %%n, r0+r4-%%n        ;   write pixels
-    dec     r5
-%endif ; ARCH_X86_64/32
-    jnz .emuedge_extend_right_ %+ %%n   ; } while (--block_h)
-%if ARCH_X86_64
-    ret
-%else ; ARCH_X86_32
-    rep ret
-%endif ; ARCH_X86_64/32
-%assign %%n %%n+2
-%endrep
-
-%if ARCH_X86_32
-%define stack_offset 0x10
-%endif
-%endmacro ; RIGHT_EXTEND
-
-; below follow the "slow" copy/extend functions, these act on a non-fixed
-; width specified in a register, and run a loop to copy the full amount
-; of bytes. They are optimized for copying of large amounts of pixels per
-; line, so they unconditionally splat data into mm registers to copy 8
-; bytes per loop iteration. It could be considered to use xmm for x86-64
-; also, but I haven't optimized this as much (i.e. FIXME)
-%macro V_COPY_NPX 4-5
-%if %0 == 4
-    test     w_reg, %4
-    jz .%1_skip_%4_px
-%else ; %0 == 5
-.%1_%4_px_loop:
-%endif
-    %3          %2, [r1+cnt_reg]
-    %3 [r0+cnt_reg], %2
-    add    cnt_reg, %4
-%if %0 == 5
-    sub      w_reg, %4
-    test     w_reg, %5
-    jnz .%1_%4_px_loop
-%endif
-.%1_skip_%4_px:
-%endmacro
-
-%macro V_COPY_ROW 2
-%ifidn %1, bottom
-    sub         r1, linesize
-%endif
-.%1_copy_loop:
-    xor    cnt_reg, cnt_reg
-%if notcpuflag(sse)
-%define linesize r2m
-    V_COPY_NPX %1,  mm0, movq,    8, 0xFFFFFFF8
-%else ; sse
-    V_COPY_NPX %1, xmm0, movups, 16, 0xFFFFFFF0
-%if ARCH_X86_64
-%define linesize r2
-    V_COPY_NPX %1, rax , mov,     8
-%else ; ARCH_X86_32
-%define linesize r2m
-    V_COPY_NPX %1,  mm0, movq,    8
-%endif ; ARCH_X86_64/32
-%endif ; sse
-    V_COPY_NPX %1, vald, mov,     4
-    V_COPY_NPX %1, valw, mov,     2
-    V_COPY_NPX %1, vall, mov,     1
-    mov      w_reg, cnt_reg
-%ifidn %1, body
-    add         r1, linesize
-%endif
-    add         r0, linesize
-    dec         %2
-    jnz .%1_copy_loop
-%endmacro
-
-%macro SLOW_V_EXTEND 0
-.slow_v_extend_loop:
-; r0=buf,r1=src,r2(64)/r2m(32)=linesize,r3(64)/r3m(32)=start_x,r4=end_y,r5=block_h
-; r8(64)/r3(later-64)/r2(32)=cnt_reg,r6(64)/r3(32)=val_reg,r7(64)/r6(32)=w=end_x-start_x
-%if ARCH_X86_64
-    push        r8              ; save old value of block_h
-    test        r3, r3
-%define cnt_reg r8
-    jz .do_body_copy            ; if (!start_y) goto do_body_copy
-    V_COPY_ROW top, r3
-%else
-    cmp  dword r3m, 0
-%define cnt_reg r2
-    je .do_body_copy            ; if (!start_y) goto do_body_copy
-    V_COPY_ROW top, dword r3m
-%endif
-
-.do_body_copy:
-    V_COPY_ROW body, r4
-
-%if ARCH_X86_64
-    pop         r8              ; restore old value of block_h
-%define cnt_reg r3
-%endif
-    test        r5, r5
-%if ARCH_X86_64
-    jz .v_extend_end
-%else
-    jz .skip_bottom_extend
-%endif
-    V_COPY_ROW bottom, r5
-%if ARCH_X86_32
-.skip_bottom_extend:
-    mov         r2, r2m
-%endif
-    jmp .v_extend_end
-%endmacro
-
-%macro SLOW_LEFT_EXTEND 0
-.slow_left_extend_loop:
-; r0=buf+block_h*linesize,r2=linesize,r6(64)/r3(32)=val,r5=block_h,r4=cntr,r7/r6=start_x
-    mov         r4, 8
-    sub         r0, linesize
-    READ_V_PIXEL 8, [r0+w_reg]
-.left_extend_8px_loop:
-    movq [r0+r4-8], mm0
-    add         r4, 8
-    cmp         r4, w_reg
-    jle .left_extend_8px_loop
-    sub         r4, 8
-    cmp         r4, w_reg
-    jge .left_extend_loop_end
-.left_extend_2px_loop:
-    mov    [r0+r4], valw
-    add         r4, 2
-    cmp         r4, w_reg
-    jl .left_extend_2px_loop
-.left_extend_loop_end:
-    dec         r5
-    jnz .slow_left_extend_loop
-%if ARCH_X86_32
-    mov         r2, r2m
-%endif
-    jmp .right_extend
-%endmacro
-
-%macro SLOW_RIGHT_EXTEND 0
-.slow_right_extend_loop:
-; r3(64)/r0(32)=buf+block_h*linesize,r2=linesize,r4=block_w,r8(64)/r5(32)=block_h,
-; r7(64)/r6(32)=end_x,r6/r3=val,r1=cntr
-%if ARCH_X86_64
-%define buf_reg r3
-%define bh_reg r8
-%else
-%define buf_reg r0
-%define bh_reg r5
-%endif
-    lea         r1, [r4-8]
-    sub    buf_reg, linesize
-    READ_V_PIXEL 8, [buf_reg+w_reg-1]
-.right_extend_8px_loop:
-    movq [buf_reg+r1], mm0
-    sub         r1, 8
-    cmp         r1, w_reg
-    jge .right_extend_8px_loop
-    add         r1, 8
-    cmp         r1, w_reg
-    je .right_extend_loop_end
-.right_extend_2px_loop:
-    sub         r1, 2
-    mov [buf_reg+r1], valw
-    cmp         r1, w_reg
-    jg .right_extend_2px_loop
-.right_extend_loop_end:
-    dec         bh_reg
-    jnz .slow_right_extend_loop
-    jmp .h_extend_end
-%endmacro
-
-%macro emu_edge 1
-INIT_XMM %1
-EMU_EDGE_FUNC
-VERTICAL_EXTEND
-LEFT_EXTEND
-RIGHT_EXTEND
-SLOW_V_EXTEND
-SLOW_LEFT_EXTEND
-SLOW_RIGHT_EXTEND
-%endmacro
-
-emu_edge sse
-%if ARCH_X86_32
-emu_edge mmx
-%endif
-
-%macro PREFETCH_FN 1
-cglobal prefetch, 3, 3, 0, buf, stride, h
-.loop:
-    %1      [bufq]
-    add      bufq, strideq
-    dec        hd
-    jg .loop
-    REP_RET
-%endmacro
-
-INIT_MMX mmxext
-PREFETCH_FN prefetcht0
-%if ARCH_X86_32
-INIT_MMX 3dnow
-PREFETCH_FN prefetch
-%endif
+;******************************************************************************
+;* Core video DSP functions
+;* 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
+;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+;******************************************************************************
+
+%include "libavutil/x86/x86util.asm"
+
+SECTION .text
+
+; slow vertical extension loop function. Works with variable-width, and
+; does per-line reading/writing of source data
+
+%macro V_COPY_ROW 2 ; type (top/body/bottom), h
+.%1_y_loop:                                     ; do {
+    mov              wq, r7mp                   ;   initialize w (r7mp = wmp)
+.%1_x_loop:                                     ;   do {
+    movu             m0, [srcq+wq]              ;     m0 = read($mmsize)
+    movu      [dstq+wq], m0                     ;     write(m0, $mmsize)
+    add              wq, mmsize                 ;     w -= $mmsize
+    cmp              wq, -mmsize                ;   } while (w > $mmsize);
+    jl .%1_x_loop
+    movu             m0, [srcq-mmsize]          ;     m0 = read($mmsize)
+    movu  [dstq-mmsize], m0                     ;     write(m0, $mmsize)
+%ifidn %1, body                                 ;   if ($type == body) {
+    add            srcq, src_strideq            ;     src += src_stride
+%endif                                          ;   }
+    add            dstq, dst_strideq            ;   dst += dst_stride
+    dec              %2                         ; } while (--$h);
+    jnz .%1_y_loop
+%endmacro
+
+%macro vvar_fn 0
+; .----. <- zero
+; |    |    <- top is copied from first line in body of source
+; |----| <- start_y
+; |    |    <- body is copied verbatim (line-by-line) from source
+; |----| <- end_y
+; |    |    <- bottom is copied from last line in body of source
+; '----' <- bh
+%if ARCH_X86_64
+cglobal emu_edge_vvar, 7, 8, 1, dst, src, dst_stride, src_stride, \
+                                start_y, end_y, bh, w
+%else ; x86-32
+cglobal emu_edge_vvar, 1, 6, 1, dst, src, start_y, end_y, bh, w
+%define src_strideq r3mp
+%define dst_strideq r2mp
+    mov            srcq, r1mp
+    mov        start_yq, r4mp
+    mov          end_yq, r5mp
+    mov             bhq, r6mp
+%endif
+    sub             bhq, end_yq                 ; bh    -= end_q
+    sub          end_yq, start_yq               ; end_q -= start_q
+    add            srcq, r7mp                   ; (r7mp = wmp)
+    add            dstq, r7mp                   ; (r7mp = wmp)
+    neg            r7mp                         ; (r7mp = wmp)
+    test       start_yq, start_yq               ; if (start_q) {
+    jz .body
+    V_COPY_ROW      top, start_yq               ;   v_copy_row(top, start_yq)
+.body:                                          ; }
+    V_COPY_ROW     body, end_yq                 ; v_copy_row(body, end_yq)
+    test            bhq, bhq                    ; if (bh) {
+    jz .end
+    sub            srcq, src_strideq            ;   src -= src_stride
+    V_COPY_ROW   bottom, bhq                    ;   v_copy_row(bottom, bh)
+.end:                                           ; }
+    RET
+%endmacro
+
+%if ARCH_X86_32
+INIT_MMX mmx
+vvar_fn
+%endif
+
+INIT_XMM sse
+vvar_fn
+
+%macro hvar_fn 0
+cglobal emu_edge_hvar, 5, 6, 1, dst, dst_stride, start_x, n_words, h, w
+    lea            dstq, [dstq+n_wordsq*2]
+    neg        n_wordsq
+    lea        start_xq, [start_xq+n_wordsq*2]
+.y_loop:                                        ; do {
+    ; FIXME also write a ssse3 version using pshufb
+    movzx            wd, byte [dstq+start_xq]   ;   w = read(1)
+    imul             wd, 0x01010101             ;   w *= 0x01010101
+    movd             m0, wd
+    mov              wq, n_wordsq               ;   initialize w
+%if cpuflag(sse)
+    shufps           m0, m0, q0000              ;   splat
+%else ; mmx
+    punpckldq        m0, m0                     ;   splat
+%endif ; mmx/sse
+.x_loop:                                        ;   do {
+    movu    [dstq+wq*2], m0                     ;     write($reg, $mmsize)
+    add              wq, mmsize/2               ;     w -= $mmsize/2
+    cmp              wq, -mmsize/2              ;   } while (w > $mmsize/2)
+    jl .x_loop
+    movu  [dstq-mmsize], m0                     ;   write($reg, $mmsize)
+    add            dstq, dst_strideq            ;   dst += dst_stride
+    dec              hq                         ; } while (h--)
+    jnz .y_loop
+    RET
+%endmacro
+
+%if ARCH_X86_32
+INIT_MMX mmx
+hvar_fn
+%endif
+
+INIT_XMM sse
+hvar_fn
+
+; macro to read/write a horizontal number of pixels (%2) to/from registers
+; on sse, - fills xmm0-15 for consecutive sets of 16 pixels
+;         - if (%2 & 8)  fills 8 bytes into xmm$next
+;         - if (%2 & 4)  fills 4 bytes into xmm$next
+;         - if (%2 & 3)  fills 1, 2 or 4 bytes in eax
+; on mmx, - fills mm0-7 for consecutive sets of 8 pixels
+;         - if (%2 & 4)  fills 4 bytes into mm$next
+;         - if (%2 & 3)  fills 1, 2 or 4 bytes in eax
+; writing data out is in the same way
+%macro READ_NUM_BYTES 2
+%assign %%off 0 ; offset in source buffer
+%assign %%idx 0 ; mmx/xmm register index
+
+%rep %2/mmsize
+    movu     m %+ %%idx, [srcq+%%off]
+%assign %%off %%off+mmsize
+%assign %%idx %%idx+1
+%endrep ; %2/mmsize
+
+%if mmsize == 16
+%if (%2-%%off) >= 8
+%if %2 > 16 && (%2-%%off) > 8
+    movu     m %+ %%idx, [srcq+%2-16]
+%assign %%off %2
+%else
+    movq     m %+ %%idx, [srcq+%%off]
+%assign %%off %%off+8
+%endif
+%assign %%idx %%idx+1
+%endif ; (%2-%%off) >= 8
+%endif
+
+%if (%2-%%off) >= 4
+%if %2 > 8 && (%2-%%off) > 4
+    movq     m %+ %%idx, [srcq+%2-8]
+%assign %%off %2
+%else
+    movd     m %+ %%idx, [srcq+%%off]
+%assign %%off %%off+4
+%endif
+%assign %%idx %%idx+1
+%endif ; (%2-%%off) >= 4
+
+%if (%2-%%off) >= 1
+%if %2 >= 4
+    movd     m %+ %%idx, [srcq+%2-4]
+%elif (%2-%%off) == 1
+    mov            valb, [srcq+%2-1]
+%elif (%2-%%off) == 2
+    mov            valw, [srcq+%2-2]
+%elifidn %1, body
+    mov            vald, [srcq+%2-3]
+%else
+    movd     m %+ %%idx, [srcq+%2-3]
+%endif
+%endif ; (%2-%%off) >= 1
+%endmacro ; READ_NUM_BYTES
+
+%macro WRITE_NUM_BYTES 2
+%assign %%off 0 ; offset in destination buffer
+%assign %%idx 0 ; mmx/xmm register index
+
+%rep %2/mmsize
+    movu   [dstq+%%off], m %+ %%idx
+%assign %%off %%off+mmsize
+%assign %%idx %%idx+1
+%endrep ; %2/mmsize
+
+%if mmsize == 16
+%if (%2-%%off) >= 8
+%if %2 > 16 && (%2-%%off) > 8
+    movu   [dstq+%2-16], m %+ %%idx
+%assign %%off %2
+%else
+    movq   [dstq+%%off], m %+ %%idx
+%assign %%off %%off+8
+%endif
+%assign %%idx %%idx+1
+%endif ; (%2-%%off) >= 8
+%endif
+
+%if (%2-%%off) >= 4
+%if %2 > 8 && (%2-%%off) > 4
+    movq    [dstq+%2-8], m %+ %%idx
+%assign %%off %2
+%else
+    movd   [dstq+%%off], m %+ %%idx
+%assign %%off %%off+4
+%endif
+%assign %%idx %%idx+1
+%endif ; (%2-%%off) >= 4
+
+%if (%2-%%off) >= 1
+%if %2 >= 4
+    movd    [dstq+%2-4], m %+ %%idx
+%elif (%2-%%off) == 1
+    mov     [dstq+%2-1], valb
+%elif (%2-%%off) == 2
+    mov     [dstq+%2-2], valw
+%elifidn %1, body
+    mov     [dstq+%2-3], valw
+    shr            vald, 16
+    mov     [dstq+%2-1], valb
+%else
+    movd           vald, m %+ %%idx
+    mov     [dstq+%2-3], valw
+    shr            vald, 16
+    mov     [dstq+%2-1], valb
+%endif
+%endif ; (%2-%%off) >= 1
+%endmacro ; WRITE_NUM_BYTES
+
+; vertical top/bottom extend and body copy fast loops
+; these are function pointers to set-width line copy functions, i.e.
+; they read a fixed number of pixels into set registers, and write
+; those out into the destination buffer
+%macro VERTICAL_EXTEND 2
+%assign %%n %1
+%rep 1+%2-%1
+%if %%n <= 3
+%if ARCH_X86_64
+cglobal emu_edge_vfix %+ %%n, 6, 8, 0, dst, src, dst_stride, src_stride, \
+                                       start_y, end_y, val, bh
+    mov             bhq, r6mp                   ; r6mp = bhmp
+%else ; x86-32
+cglobal emu_edge_vfix %+ %%n, 0, 6, 0, val, dst, src, start_y, end_y, bh
+    mov            dstq, r0mp
+    mov            srcq, r1mp
+    mov        start_yq, r4mp
+    mov          end_yq, r5mp
+    mov             bhq, r6mp
+%define dst_strideq r2mp
+%define src_strideq r3mp
+%endif ; x86-64/32
+%else
+%if ARCH_X86_64
+cglobal emu_edge_vfix %+ %%n, 7, 7, 1, dst, src, dst_stride, src_stride, \
+                                       start_y, end_y, bh
+%else ; x86-32
+cglobal emu_edge_vfix %+ %%n, 1, 5, 1, dst, src, start_y, end_y, bh
+    mov            srcq, r1mp
+    mov        start_yq, r4mp
+    mov          end_yq, r5mp
+    mov             bhq, r6mp
+%define dst_strideq r2mp
+%define src_strideq r3mp
+%endif ; x86-64/32
+%endif
+    ; FIXME move this to c wrapper?
+    sub             bhq, end_yq                 ; bh    -= end_y
+    sub          end_yq, start_yq               ; end_y -= start_y
+
+    ; extend pixels above body
+    test       start_yq, start_yq               ; if (start_y) {
+    jz .body_loop
+    READ_NUM_BYTES  top, %%n                    ;   $variable_regs = read($n)
+.top_loop:                                      ;   do {
+    WRITE_NUM_BYTES top, %%n                    ;     write($variable_regs, $n)
+    add            dstq, dst_strideq            ;     dst += linesize
+    dec        start_yq                         ;   } while (--start_y)
+    jnz .top_loop                               ; }
+
+    ; copy body pixels
+.body_loop:                                     ; do {
+    READ_NUM_BYTES  body, %%n                   ;   $variable_regs = read($n)
+    WRITE_NUM_BYTES body, %%n                   ;   write($variable_regs, $n)
+    add            dstq, dst_strideq            ;   dst += dst_stride
+    add            srcq, src_strideq            ;   src += src_stride
+    dec          end_yq                         ; } while (--end_y)
+    jnz .body_loop
+
+    ; copy bottom pixels
+    test            bhq, bhq                    ; if (block_h) {
+    jz .end
+    sub            srcq, src_strideq            ;   src -= linesize
+    READ_NUM_BYTES  bottom, %%n                 ;   $variable_regs = read($n)
+.bottom_loop:                                   ;   do {
+    WRITE_NUM_BYTES bottom, %%n                 ;     write($variable_regs, $n)
+    add            dstq, dst_strideq            ;     dst += linesize
+    dec             bhq                         ;   } while (--bh)
+    jnz .bottom_loop                            ; }
+
+.end:
+    RET
+%assign %%n %%n+1
+%endrep ; 1+%2-%1
+%endmacro ; VERTICAL_EXTEND
+
+INIT_MMX mmx
+VERTICAL_EXTEND 1, 15
+%if ARCH_X86_32
+VERTICAL_EXTEND 16, 22
+%endif
+
+INIT_XMM sse
+VERTICAL_EXTEND 16, 22
+
+; left/right (horizontal) fast extend functions
+; these are essentially identical to the vertical extend ones above,
+; just left/right separated because number of pixels to extend is
+; obviously not the same on both sides.
+
+%macro READ_V_PIXEL 2
+%if %1 == 2
+    movzx          valw, byte %2
+    imul           valw, 0x0101
+%else
+    movzx          vald, byte %2
+    imul           vald, 0x01010101
+%if %1 >= 8
+    movd             m0, vald
+%if mmsize == 16
+    shufps           m0, m0, q0000
+%else
+    punpckldq        m0, m0
+%endif
+%endif ; %1 >= 8
+%endif
+%endmacro ; READ_V_PIXEL
+
+%macro WRITE_V_PIXEL 2
+%assign %%off 0
+%rep %1/mmsize
+    movu     [%2+%%off], m0
+%assign %%off %%off+mmsize
+%endrep ; %1/mmsize
+
+%if mmsize == 16
+%if %1-%%off >= 8
+%if %1 > 16 && %1-%%off > 8
+    movu     [%2+%1-16], m0
+%assign %%off %1
+%else
+    movq     [%2+%%off], m0
+%assign %%off %%off+8
+%endif
+%endif ; %1-%%off >= 8
+%endif
+
+%if %1-%%off >= 4
+%if %1 > 8 %% %1-%%off > 4
+    movq      [%2+%1-8], m0
+%assign %%off %1
+%elif %1 >= 8 && %1-%%off >= 4
+    movd     [%2+%%off], m0
+%assign %%off %%off+4
+%else
+    mov      [%2+%%off], vald
+%assign %%off %%off+4
+%endif
+%endif ; %1-%%off >= 4
+
+%if %1-%%off >= 2
+%if %1 >= 8
+    movd      [%2+%1-4], m0
+%else
+    mov      [%2+%%off], valw
+%endif
+%endif ; (%1-%%off)/2
+%endmacro ; WRITE_V_PIXEL
+
+%macro H_EXTEND 2
+%assign %%n %1
+%rep 1+(%2-%1)/2
+cglobal emu_edge_hfix %+ %%n, 4, 5, 1, dst, dst_stride, start_x, bh, val
+.loop_y:                                        ; do {
+    READ_V_PIXEL    %%n, [dstq+start_xq]        ;   $variable_regs = read($n)
+    WRITE_V_PIXEL   %%n, dstq                   ;   write($variable_regs, $n)
+    add            dstq, dst_strideq            ;   dst += dst_stride
+    dec             bhq                         ; } while (--bh)
+    jnz .loop_y
+    RET
+%assign %%n %%n+2
+%endrep ; 1+(%2-%1)/2
+%endmacro ; H_EXTEND
+
+INIT_MMX mmx
+H_EXTEND 2, 14
+%if ARCH_X86_32
+H_EXTEND 16, 22
+%endif
+
+INIT_XMM sse
+H_EXTEND 16, 22
+
+%macro PREFETCH_FN 1
+cglobal prefetch, 3, 3, 0, buf, stride, h
+.loop:
+    %1      [bufq]
+    add      bufq, strideq
+    dec        hd
+    jg .loop
+    REP_RET
+%endmacro
+
+INIT_MMX mmxext
+PREFETCH_FN prefetcht0
+%if ARCH_X86_32
+INIT_MMX 3dnow
+PREFETCH_FN prefetch
+%endif
index f7fd267..79d9801 100644 (file)
 #include "libavcodec/videodsp.h"
 
 #if HAVE_YASM
-typedef void emu_edge_core_func(uint8_t *buf, const uint8_t *src,
-                                x86_reg linesize, x86_reg start_y,
-                                x86_reg end_y, x86_reg block_h,
-                                x86_reg start_x, x86_reg end_x,
-                                x86_reg block_w);
-extern emu_edge_core_func ff_emu_edge_core_mmx;
-extern emu_edge_core_func ff_emu_edge_core_sse;
-
-static av_always_inline void emulated_edge_mc(uint8_t *buf, const uint8_t *src,
-                                              ptrdiff_t linesize,
-                                              int block_w, int block_h,
-                                              int src_x, int src_y,
-                                              int w, int h,
-                                              emu_edge_core_func *core_fn)
+typedef void emu_edge_vfix_func(uint8_t *dst, const uint8_t *src,
+                                x86_reg dst_stride, x86_reg src_stride,
+                                x86_reg start_y, x86_reg end_y, x86_reg bh);
+typedef void emu_edge_vvar_func(uint8_t *dst, const uint8_t *src,
+                                x86_reg dst_stride, x86_reg src_stride,
+                                x86_reg start_y, x86_reg end_y, x86_reg bh,
+                                x86_reg w);
+
+extern emu_edge_vfix_func ff_emu_edge_vfix1_mmx;
+extern emu_edge_vfix_func ff_emu_edge_vfix2_mmx;
+extern emu_edge_vfix_func ff_emu_edge_vfix3_mmx;
+extern emu_edge_vfix_func ff_emu_edge_vfix4_mmx;
+extern emu_edge_vfix_func ff_emu_edge_vfix5_mmx;
+extern emu_edge_vfix_func ff_emu_edge_vfix6_mmx;
+extern emu_edge_vfix_func ff_emu_edge_vfix7_mmx;
+extern emu_edge_vfix_func ff_emu_edge_vfix8_mmx;
+extern emu_edge_vfix_func ff_emu_edge_vfix9_mmx;
+extern emu_edge_vfix_func ff_emu_edge_vfix10_mmx;
+extern emu_edge_vfix_func ff_emu_edge_vfix11_mmx;
+extern emu_edge_vfix_func ff_emu_edge_vfix12_mmx;
+extern emu_edge_vfix_func ff_emu_edge_vfix13_mmx;
+extern emu_edge_vfix_func ff_emu_edge_vfix14_mmx;
+extern emu_edge_vfix_func ff_emu_edge_vfix15_mmx;
+extern emu_edge_vfix_func ff_emu_edge_vfix16_mmx;
+extern emu_edge_vfix_func ff_emu_edge_vfix17_mmx;
+extern emu_edge_vfix_func ff_emu_edge_vfix18_mmx;
+extern emu_edge_vfix_func ff_emu_edge_vfix19_mmx;
+extern emu_edge_vfix_func ff_emu_edge_vfix20_mmx;
+extern emu_edge_vfix_func ff_emu_edge_vfix21_mmx;
+extern emu_edge_vfix_func ff_emu_edge_vfix22_mmx;
+#if ARCH_X86_32
+static emu_edge_vfix_func *vfixtbl_mmx[22] = {
+    &ff_emu_edge_vfix1_mmx,  &ff_emu_edge_vfix2_mmx,  &ff_emu_edge_vfix3_mmx,
+    &ff_emu_edge_vfix4_mmx,  &ff_emu_edge_vfix5_mmx,  &ff_emu_edge_vfix6_mmx,
+    &ff_emu_edge_vfix7_mmx,  &ff_emu_edge_vfix8_mmx,  &ff_emu_edge_vfix9_mmx,
+    &ff_emu_edge_vfix10_mmx, &ff_emu_edge_vfix11_mmx, &ff_emu_edge_vfix12_mmx,
+    &ff_emu_edge_vfix13_mmx, &ff_emu_edge_vfix14_mmx, &ff_emu_edge_vfix15_mmx,
+    &ff_emu_edge_vfix16_mmx, &ff_emu_edge_vfix17_mmx, &ff_emu_edge_vfix18_mmx,
+    &ff_emu_edge_vfix19_mmx, &ff_emu_edge_vfix20_mmx, &ff_emu_edge_vfix21_mmx,
+    &ff_emu_edge_vfix22_mmx
+};
+#endif
+extern emu_edge_vvar_func ff_emu_edge_vvar_mmx;
+extern emu_edge_vfix_func ff_emu_edge_vfix16_sse;
+extern emu_edge_vfix_func ff_emu_edge_vfix17_sse;
+extern emu_edge_vfix_func ff_emu_edge_vfix18_sse;
+extern emu_edge_vfix_func ff_emu_edge_vfix19_sse;
+extern emu_edge_vfix_func ff_emu_edge_vfix20_sse;
+extern emu_edge_vfix_func ff_emu_edge_vfix21_sse;
+extern emu_edge_vfix_func ff_emu_edge_vfix22_sse;
+static emu_edge_vfix_func *vfixtbl_sse[22] = {
+    ff_emu_edge_vfix1_mmx,  ff_emu_edge_vfix2_mmx,  ff_emu_edge_vfix3_mmx,
+    ff_emu_edge_vfix4_mmx,  ff_emu_edge_vfix5_mmx,  ff_emu_edge_vfix6_mmx,
+    ff_emu_edge_vfix7_mmx,  ff_emu_edge_vfix8_mmx,  ff_emu_edge_vfix9_mmx,
+    ff_emu_edge_vfix10_mmx, ff_emu_edge_vfix11_mmx, ff_emu_edge_vfix12_mmx,
+    ff_emu_edge_vfix13_mmx, ff_emu_edge_vfix14_mmx, ff_emu_edge_vfix15_mmx,
+    ff_emu_edge_vfix16_sse, ff_emu_edge_vfix17_sse, ff_emu_edge_vfix18_sse,
+    ff_emu_edge_vfix19_sse, ff_emu_edge_vfix20_sse, ff_emu_edge_vfix21_sse,
+    ff_emu_edge_vfix22_sse
+};
+extern emu_edge_vvar_func ff_emu_edge_vvar_sse;
+
+typedef void emu_edge_hfix_func(uint8_t *dst, x86_reg dst_stride,
+                                x86_reg start_x, x86_reg bh);
+typedef void emu_edge_hvar_func(uint8_t *dst, x86_reg dst_stride,
+                                x86_reg start_x, x86_reg n_words, x86_reg bh);
+
+extern emu_edge_hfix_func ff_emu_edge_hfix2_mmx;
+extern emu_edge_hfix_func ff_emu_edge_hfix4_mmx;
+extern emu_edge_hfix_func ff_emu_edge_hfix6_mmx;
+extern emu_edge_hfix_func ff_emu_edge_hfix8_mmx;
+extern emu_edge_hfix_func ff_emu_edge_hfix10_mmx;
+extern emu_edge_hfix_func ff_emu_edge_hfix12_mmx;
+extern emu_edge_hfix_func ff_emu_edge_hfix14_mmx;
+extern emu_edge_hfix_func ff_emu_edge_hfix16_mmx;
+extern emu_edge_hfix_func ff_emu_edge_hfix18_mmx;
+extern emu_edge_hfix_func ff_emu_edge_hfix20_mmx;
+extern emu_edge_hfix_func ff_emu_edge_hfix22_mmx;
+#if ARCH_X86_32
+static emu_edge_hfix_func *hfixtbl_mmx[11] = {
+    ff_emu_edge_hfix2_mmx,  ff_emu_edge_hfix4_mmx,  ff_emu_edge_hfix6_mmx,
+    ff_emu_edge_hfix8_mmx,  ff_emu_edge_hfix10_mmx, ff_emu_edge_hfix12_mmx,
+    ff_emu_edge_hfix14_mmx, ff_emu_edge_hfix16_mmx, ff_emu_edge_hfix18_mmx,
+    ff_emu_edge_hfix20_mmx, ff_emu_edge_hfix22_mmx
+};
+#endif
+extern emu_edge_hvar_func ff_emu_edge_hvar_mmx;
+extern emu_edge_hfix_func ff_emu_edge_hfix16_sse;
+extern emu_edge_hfix_func ff_emu_edge_hfix18_sse;
+extern emu_edge_hfix_func ff_emu_edge_hfix20_sse;
+extern emu_edge_hfix_func ff_emu_edge_hfix22_sse;
+static emu_edge_hfix_func *hfixtbl_sse[11] = {
+    ff_emu_edge_hfix2_mmx,  ff_emu_edge_hfix4_mmx,  ff_emu_edge_hfix6_mmx,
+    ff_emu_edge_hfix8_mmx,  ff_emu_edge_hfix10_mmx, ff_emu_edge_hfix12_mmx,
+    ff_emu_edge_hfix14_mmx, ff_emu_edge_hfix16_sse, ff_emu_edge_hfix18_sse,
+    ff_emu_edge_hfix20_sse, ff_emu_edge_hfix22_sse
+};
+extern emu_edge_hvar_func ff_emu_edge_hvar_sse;
+
+static av_always_inline void emulated_edge_mc(uint8_t *dst, const uint8_t *src,
+                                              ptrdiff_t dst_stride,
+                                              ptrdiff_t src_stride,
+                                              x86_reg block_w, x86_reg block_h,
+                                              x86_reg src_x, x86_reg src_y,
+                                              x86_reg w, x86_reg h,
+                                              emu_edge_vfix_func **vfix_tbl,
+                                              emu_edge_vvar_func *v_extend_var,
+                                              emu_edge_hfix_func **hfix_tbl,
+                                              emu_edge_hvar_func *h_extend_var)
 {
-    int start_y, start_x, end_y, end_x, src_y_add = 0;
+    x86_reg start_y, start_x, end_y, end_x, src_y_add = 0, p;
+
+    if (!w || !h)
+         return;
 
     if (src_y >= h) {
-        src_y_add = h - 1 - src_y;
-        src_y     = h - 1;
+        src  -= src_y * src_stride;
+        src_y = src_y_add = h - 1;
     } else if (src_y <= -block_h) {
-        src_y_add = 1 - block_h - src_y;
-        src_y     = 1 - block_h;
+        src  -= src_y*src_stride;
+        src_y = src_y_add = 1 - block_h;
     }
     if (src_x >= w) {
         src   += w - 1 - src_x;
@@ -68,30 +166,62 @@ static av_always_inline void emulated_edge_mc(uint8_t *buf, const uint8_t *src,
     assert(start_y < end_y && block_h > 0);
 
     // fill in the to-be-copied part plus all above/below
-    src += (src_y_add + start_y) * linesize + start_x;
-    buf += start_x;
-    core_fn(buf, src, linesize, start_y, end_y,
-            block_h, start_x, end_x, block_w);
+    src += (src_y_add + start_y) * src_stride + start_x;
+    w = end_x - start_x;
+    if (w <= 22) {
+        vfix_tbl[w - 1](dst + start_x, src,
+                        dst_stride, src_stride,
+                        start_y, end_y, block_h);
+    } else {
+        v_extend_var(dst + start_x, src, dst_stride, src_stride,
+                     start_y, end_y, block_h, w);
+    }
+
+    // fill left
+    if (start_x) {
+        if (start_x <= 22) {
+            hfix_tbl[(start_x - 1) >> 1](dst, dst_stride, start_x, block_h);
+        } else {
+            h_extend_var(dst, dst_stride,
+                         start_x, (start_x + 1) >> 1, block_h);
+        }
+    }
+
+    // fill right
+    p = block_w - end_x;
+    if (p) {
+        if (p <= 22) {
+            hfix_tbl[(p - 1) >> 1](dst + end_x - (p & 1), dst_stride,
+                                   -!(p & 1), block_h);
+        } else {
+            h_extend_var(dst + end_x - (p & 1), dst_stride,
+                         -!(p & 1), (p + 1) >> 1, block_h);
+        }
+    }
 }
 
 #if ARCH_X86_32
 static av_noinline void emulated_edge_mc_mmx(uint8_t *buf, const uint8_t *src,
-                                             ptrdiff_t linesize,
+                                             ptrdiff_t buf_stride,
+                                             ptrdiff_t src_stride,
                                              int block_w, int block_h,
                                              int src_x, int src_y, int w, int h)
 {
-    emulated_edge_mc(buf, src, linesize, block_w, block_h, src_x, src_y,
-                     w, h, &ff_emu_edge_core_mmx);
+    emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h,
+                     src_x, src_y, w, h, vfixtbl_mmx, &ff_emu_edge_vvar_mmx,
+                     hfixtbl_mmx, &ff_emu_edge_hvar_mmx);
 }
 #endif
 
-static av_noinline void emulated_edge_mc_sse(uint8_t *buf, const uint8_t *src,
-                                             ptrdiff_t linesize,
+static av_noinline void emulated_edge_mc_sse(uint8_t * buf,const uint8_t *src,
+                                             ptrdiff_t buf_stride,
+                                             ptrdiff_t src_stride,
                                              int block_w, int block_h,
                                              int src_x, int src_y, int w, int h)
 {
-    emulated_edge_mc(buf, src, linesize, block_w, block_h, src_x, src_y,
-                     w, h, &ff_emu_edge_core_sse);
+    emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h,
+                     src_x, src_y, w, h, vfixtbl_sse, &ff_emu_edge_vvar_sse,
+                     hfixtbl_sse, &ff_emu_edge_hvar_sse);
 }
 #endif /* HAVE_YASM */