Merge commit '3c650efb81aaa3b395ba4606ee68a47ee4efb57b'
authorMichael Niedermayer <michaelni@gmx.at>
Mon, 7 Jul 2014 13:54:17 +0000 (15:54 +0200)
committerMichael Niedermayer <michaelni@gmx.at>
Mon, 7 Jul 2014 14:17:27 +0000 (16:17 +0200)
* commit '3c650efb81aaa3b395ba4606ee68a47ee4efb57b':
  dsputil: Move draw_edges() to mpegvideoencdsp

Conflicts:
libavcodec/mpegvideo_enc.c
libavcodec/x86/Makefile
libavcodec/x86/dsputil_init.c
libavcodec/x86/dsputil_mmx.c
libavcodec/x86/dsputil_x86.h

Merged-by: Michael Niedermayer <michaelni@gmx.at>
16 files changed:
libavcodec/diracdec.c
libavcodec/dsputil.c
libavcodec/dsputil.h
libavcodec/mpegvideo.h
libavcodec/mpegvideo_enc.c
libavcodec/mpegvideoencdsp.c
libavcodec/mpegvideoencdsp.h
libavcodec/snow.c
libavcodec/snow.h
libavcodec/snowenc.c
libavcodec/utils.c
libavcodec/x86/Makefile
libavcodec/x86/dsputil_init.c
libavcodec/x86/dsputil_mmx.c [deleted file]
libavcodec/x86/dsputil_x86.h
libavcodec/x86/mpegvideoencdsp_init.c

index d7cca7e..d1c3758 100644 (file)
@@ -34,6 +34,8 @@
 #include "golomb.h"
 #include "dirac_arith.h"
 #include "mpeg12data.h"
+#include "libavcodec/mpegvideo.h"
+#include "mpegvideoencdsp.h"
 #include "dirac_dwt.h"
 #include "dirac.h"
 #include "diracdsp.h"
@@ -137,6 +139,7 @@ typedef struct Plane {
 typedef struct DiracContext {
     AVCodecContext *avctx;
     DSPContext dsp;
+    MpegvideoEncDSPContext mpvencdsp;
     DiracDSPContext diracdsp;
     GetBitContext gb;
     dirac_source_params source;
@@ -424,6 +427,7 @@ static av_cold int dirac_decode_init(AVCodecContext *avctx)
 
     ff_dsputil_init(&s->dsp, avctx);
     ff_diracdsp_init(&s->diracdsp);
+    ff_mpegvideoencdsp_init(&s->mpvencdsp, avctx);
 
     for (i = 0; i < MAX_FRAMES; i++) {
         s->all_frames[i].avframe = av_frame_alloc();
@@ -1556,7 +1560,7 @@ static void interpolate_refplane(DiracContext *s, DiracFrame *ref, int plane, in
     int i, edge = EDGE_WIDTH/2;
 
     ref->hpel[plane][0] = ref->avframe->data[plane];
-    s->dsp.draw_edges(ref->hpel[plane][0], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM); /* EDGE_TOP | EDGE_BOTTOM values just copied to make it build, this needs to be ensured */
+    s->mpvencdsp.draw_edges(ref->hpel[plane][0], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM); /* EDGE_TOP | EDGE_BOTTOM values just copied to make it build, this needs to be ensured */
 
     /* no need for hpel if we only have fpel vectors */
     if (!s->mv_precision)
@@ -1573,9 +1577,9 @@ static void interpolate_refplane(DiracContext *s, DiracFrame *ref, int plane, in
         s->diracdsp.dirac_hpel_filter(ref->hpel[plane][1], ref->hpel[plane][2],
                                       ref->hpel[plane][3], ref->hpel[plane][0],
                                       ref->avframe->linesize[plane], width, height);
-        s->dsp.draw_edges(ref->hpel[plane][1], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM);
-        s->dsp.draw_edges(ref->hpel[plane][2], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM);
-        s->dsp.draw_edges(ref->hpel[plane][3], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM);
+        s->mpvencdsp.draw_edges(ref->hpel[plane][1], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM);
+        s->mpvencdsp.draw_edges(ref->hpel[plane][2], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM);
+        s->mpvencdsp.draw_edges(ref->hpel[plane][3], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM);
     }
     ref->interpolated[plane] = 1;
 }
index 640a4bf..5d7fbb1 100644 (file)
@@ -937,34 +937,6 @@ WRAPPER8_16_SQ(quant_psnr8x8_c, quant_psnr16_c)
 WRAPPER8_16_SQ(rd8x8_c, rd16_c)
 WRAPPER8_16_SQ(bit8x8_c, bit16_c)
 
-/* draw the edges of width 'w' of an image of size width, height */
-// FIXME: Check that this is OK for MPEG-4 interlaced.
-static void draw_edges_8_c(uint8_t *buf, int wrap, int width, int height,
-                           int w, int h, int sides)
-{
-    uint8_t *ptr = buf, *last_line;
-    int i;
-
-    /* left and right */
-    for (i = 0; i < height; i++) {
-        memset(ptr - w, ptr[0], w);
-        memset(ptr + width, ptr[width - 1], w);
-        ptr += wrap;
-    }
-
-    /* top and bottom + corners */
-    buf -= w;
-    last_line = buf + (height - 1) * wrap;
-    if (sides & EDGE_TOP)
-        for (i = 0; i < h; i++)
-            // top
-            memcpy(buf - (i + 1) * wrap, buf, width + w + w);
-    if (sides & EDGE_BOTTOM)
-        for (i = 0; i < h; i++)
-            // bottom
-            memcpy(last_line + (i + 1) * wrap, last_line, width + w + w);
-}
-
 /* init static data */
 av_cold void ff_dsputil_static_init(void)
 {
@@ -1067,8 +1039,6 @@ av_cold void ff_dsputil_init(DSPContext *c, AVCodecContext *avctx)
     ff_dsputil_init_dwt(c);
 #endif
 
-    c->draw_edges = draw_edges_8_c;
-
     switch (avctx->bits_per_raw_sample) {
     case 9:
     case 10:
index 8dafbbd..0d7dd14 100644 (file)
@@ -99,12 +99,6 @@ typedef struct DSPContext {
     /* (I)DCT */
     void (*fdct)(int16_t *block /* align 16 */);
     void (*fdct248)(int16_t *block /* align 16 */);
-
-    void (*draw_edges)(uint8_t *buf, int wrap, int width, int height,
-                       int w, int h, int sides);
-#define EDGE_WIDTH 16
-#define EDGE_TOP    1
-#define EDGE_BOTTOM 2
 } DSPContext;
 
 void ff_dsputil_static_init(void);
index 795a3bd..cf0c47b 100644 (file)
@@ -81,6 +81,8 @@ enum OutputFormat {
 
 #define INPLACE_OFFSET 16
 
+#define EDGE_WIDTH 16
+
 /* Start codes. */
 #define SEQ_END_CODE            0x000001b7
 #define SEQ_START_CODE          0x000001b3
index 0d630df..d39f88e 100644 (file)
@@ -1145,11 +1145,11 @@ static int load_input_picture(MpegEncContext *s, const AVFrame *pic_arg)
                         }
                     }
                     if ((s->width & 15) || (s->height & (vpad-1))) {
-                        s->dsp.draw_edges(dst, dst_stride,
-                                          w, h,
-                                          16>>h_shift,
-                                          vpad>>v_shift,
-                                          EDGE_BOTTOM);
+                        s->mpvencdsp.draw_edges(dst, dst_stride,
+                                                w, h,
+                                                16>>h_shift,
+                                                vpad>>v_shift,
+                                                EDGE_BOTTOM);
                     }
                 }
             }
@@ -1529,18 +1529,25 @@ static void frame_end(MpegEncContext *s)
         const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(s->avctx->pix_fmt);
         int hshift = desc->log2_chroma_w;
         int vshift = desc->log2_chroma_h;
-        s->dsp.draw_edges(s->current_picture.f->data[0], s->current_picture.f->linesize[0],
-                          s->h_edge_pos, s->v_edge_pos,
-                          EDGE_WIDTH, EDGE_WIDTH,
-                          EDGE_TOP | EDGE_BOTTOM);
-        s->dsp.draw_edges(s->current_picture.f->data[1], s->current_picture.f->linesize[1],
-                          s->h_edge_pos >> hshift, s->v_edge_pos >> vshift,
-                          EDGE_WIDTH >> hshift, EDGE_WIDTH >> vshift,
-                          EDGE_TOP | EDGE_BOTTOM);
-        s->dsp.draw_edges(s->current_picture.f->data[2], s->current_picture.f->linesize[2],
-                          s->h_edge_pos >> hshift, s->v_edge_pos >> vshift,
-                          EDGE_WIDTH >> hshift, EDGE_WIDTH >> vshift,
-                          EDGE_TOP | EDGE_BOTTOM);
+        s->mpvencdsp.draw_edges(s->current_picture.f->data[0],
+                                s->current_picture.f->linesize[0],
+                                s->h_edge_pos, s->v_edge_pos,
+                                EDGE_WIDTH, EDGE_WIDTH,
+                                EDGE_TOP | EDGE_BOTTOM);
+        s->mpvencdsp.draw_edges(s->current_picture.f->data[1],
+                                s->current_picture.f->linesize[1],
+                                s->h_edge_pos >> hshift,
+                                s->v_edge_pos >> vshift,
+                                EDGE_WIDTH >> hshift,
+                                EDGE_WIDTH >> vshift,
+                                EDGE_TOP | EDGE_BOTTOM);
+        s->mpvencdsp.draw_edges(s->current_picture.f->data[2],
+                                s->current_picture.f->linesize[2],
+                                s->h_edge_pos >> hshift,
+                                s->v_edge_pos >> vshift,
+                                EDGE_WIDTH >> hshift,
+                                EDGE_WIDTH >> vshift,
+                                EDGE_TOP | EDGE_BOTTOM);
     }
 
     emms_c();
index bde4345..10ad369 100644 (file)
@@ -18,6 +18,7 @@
 
 #include <assert.h>
 #include <stdint.h>
+#include <string.h>
 
 #include "config.h"
 #include "libavutil/avassert.h"
@@ -125,6 +126,34 @@ static int pix_norm1_c(uint8_t *pix, int line_size)
     return s;
 }
 
+/* draw the edges of width 'w' of an image of size width, height */
+// FIXME: Check that this is OK for MPEG-4 interlaced.
+static void draw_edges_8_c(uint8_t *buf, int wrap, int width, int height,
+                           int w, int h, int sides)
+{
+    uint8_t *ptr = buf, *last_line;
+    int i;
+
+    /* left and right */
+    for (i = 0; i < height; i++) {
+        memset(ptr - w, ptr[0], w);
+        memset(ptr + width, ptr[width - 1], w);
+        ptr += wrap;
+    }
+
+    /* top and bottom + corners */
+    buf -= w;
+    last_line = buf + (height - 1) * wrap;
+    if (sides & EDGE_TOP)
+        for (i = 0; i < h; i++)
+            // top
+            memcpy(buf - (i + 1) * wrap, buf, width + w + w);
+    if (sides & EDGE_BOTTOM)
+        for (i = 0; i < h; i++)
+            // bottom
+            memcpy(last_line + (i + 1) * wrap, last_line, width + w + w);
+}
+
 av_cold void ff_mpegvideoencdsp_init(MpegvideoEncDSPContext *c,
                                      AVCodecContext *avctx)
 {
@@ -139,6 +168,8 @@ av_cold void ff_mpegvideoencdsp_init(MpegvideoEncDSPContext *c,
     c->pix_sum   = pix_sum_c;
     c->pix_norm1 = pix_norm1_c;
 
+    c->draw_edges = draw_edges_8_c;
+
     if (ARCH_ARM)
         ff_mpegvideoencdsp_init_arm(c, avctx);
     if (ARCH_PPC)
index 81e3fe6..e12f4c6 100644 (file)
@@ -26,6 +26,9 @@
 #define BASIS_SHIFT 16
 #define RECON_SHIFT 6
 
+#define EDGE_TOP    1
+#define EDGE_BOTTOM 2
+
 typedef struct MpegvideoEncDSPContext {
     int (*try_8x8basis)(int16_t rem[64], int16_t weight[64],
                         int16_t basis[64], int scale);
@@ -36,6 +39,9 @@ typedef struct MpegvideoEncDSPContext {
 
     void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src,
                       int src_wrap, int width, int height);
+
+    void (*draw_edges)(uint8_t *buf, int wrap, int width, int height,
+                       int w, int h, int sides);
 } MpegvideoEncDSPContext;
 
 void ff_mpegvideoencdsp_init(MpegvideoEncDSPContext *c,
index e1ed297..711d1a4 100644 (file)
@@ -433,6 +433,7 @@ av_cold int ff_snow_common_init(AVCodecContext *avctx){
     ff_videodsp_init(&s->vdsp, 8);
     ff_dwt_init(&s->dwt);
     ff_h264qpel_init(&s->h264qpel, 8);
+    ff_mpegvideoencdsp_init(&s->mpvencdsp, avctx);
 
 #define mcf(dx,dy)\
     s->qdsp.put_qpel_pixels_tab       [0][dy+dx/4]=\
@@ -642,16 +643,16 @@ int ff_snow_frame_start(SnowContext *s){
    int h= s->avctx->height;
 
     if (s->current_picture->data[0] && !(s->avctx->flags&CODEC_FLAG_EMU_EDGE)) {
-        s->dsp.draw_edges(s->current_picture->data[0],
-                          s->current_picture->linesize[0], w   , h   ,
-                          EDGE_WIDTH  , EDGE_WIDTH  , EDGE_TOP | EDGE_BOTTOM);
+        s->mpvencdsp.draw_edges(s->current_picture->data[0],
+                                s->current_picture->linesize[0], w   , h   ,
+                                EDGE_WIDTH  , EDGE_WIDTH  , EDGE_TOP | EDGE_BOTTOM);
         if (s->current_picture->data[2]) {
-            s->dsp.draw_edges(s->current_picture->data[1],
-                            s->current_picture->linesize[1], w>>s->chroma_h_shift, h>>s->chroma_v_shift,
-                            EDGE_WIDTH>>s->chroma_h_shift, EDGE_WIDTH>>s->chroma_v_shift, EDGE_TOP | EDGE_BOTTOM);
-            s->dsp.draw_edges(s->current_picture->data[2],
-                            s->current_picture->linesize[2], w>>s->chroma_h_shift, h>>s->chroma_v_shift,
-                            EDGE_WIDTH>>s->chroma_h_shift, EDGE_WIDTH>>s->chroma_v_shift, EDGE_TOP | EDGE_BOTTOM);
+            s->mpvencdsp.draw_edges(s->current_picture->data[1],
+                                    s->current_picture->linesize[1], w>>s->chroma_h_shift, h>>s->chroma_v_shift,
+                                    EDGE_WIDTH>>s->chroma_h_shift, EDGE_WIDTH>>s->chroma_v_shift, EDGE_TOP | EDGE_BOTTOM);
+            s->mpvencdsp.draw_edges(s->current_picture->data[2],
+                                    s->current_picture->linesize[2], w>>s->chroma_h_shift, h>>s->chroma_v_shift,
+                                    EDGE_WIDTH>>s->chroma_h_shift, EDGE_WIDTH>>s->chroma_v_shift, EDGE_TOP | EDGE_BOTTOM);
         }
     }
 
index a9c8518..2cda5b3 100644 (file)
@@ -115,6 +115,7 @@ typedef struct SnowContext{
     QpelDSPContext qdsp;
     VideoDSPContext vdsp;
     H264QpelContext h264qpel;
+    MpegvideoEncDSPContext mpvencdsp;
     SnowDWTContext dwt;
     AVFrame *new_picture;
     AVFrame *input_picture;              ///< new_picture with the internal linesizes
index a39cd84..cb83821 100644 (file)
@@ -1568,10 +1568,10 @@ static int encode_frame(AVCodecContext *avctx, AVPacket *pkt,
             memcpy(&s->input_picture->data[i][y * s->input_picture->linesize[i]],
                    &pict->data[i][y * pict->linesize[i]],
                    width>>hshift);
-        s->dsp.draw_edges(s->input_picture->data[i], s->input_picture->linesize[i],
-                            width >> hshift, height >> vshift,
-                            EDGE_WIDTH >> hshift, EDGE_WIDTH >> vshift,
-                            EDGE_TOP | EDGE_BOTTOM);
+        s->mpvencdsp.draw_edges(s->input_picture->data[i], s->input_picture->linesize[i],
+                                width >> hshift, height >> vshift,
+                                EDGE_WIDTH >> hshift, EDGE_WIDTH >> vshift,
+                                EDGE_TOP | EDGE_BOTTOM);
 
     }
     emms_c();
index 65f8c36..9fa8e16 100644 (file)
@@ -43,6 +43,7 @@
 #include "avcodec.h"
 #include "dsputil.h"
 #include "libavutil/opt.h"
+#include "mpegvideo.h"
 #include "thread.h"
 #include "frame_thread_encoder.h"
 #include "internal.h"
index ac336c7..adcf627 100644 (file)
@@ -58,7 +58,6 @@ OBJS-$(CONFIG_VP8_DECODER)             += x86/vp8dsp_init.o
 OBJS-$(CONFIG_VP9_DECODER)             += x86/vp9dsp_init.o
 OBJS-$(CONFIG_WEBP_DECODER)            += x86/vp8dsp_init.o
 
-MMX-OBJS-$(CONFIG_DSPUTIL)             += x86/dsputil_mmx.o
 MMX-OBJS-$(CONFIG_DIRAC_DECODER)       += x86/dirac_dwt.o
 MMX-OBJS-$(CONFIG_ENCODERS)            += x86/fdct.o
 MMX-OBJS-$(CONFIG_IDCTDSP)             += x86/idctdsp_mmx.o             \
index c1b110b..85c637b 100644 (file)
 #include "libavcodec/dsputil.h"
 #include "dsputil_x86.h"
 
-static av_cold void dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx,
-                                     int cpu_flags, unsigned high_bit_depth)
-{
-#if HAVE_MMX_INLINE
-    if (!high_bit_depth) {
-        c->draw_edges   = ff_draw_edges_mmx;
-    }
-
-#endif /* HAVE_MMX_INLINE */
-}
-
 av_cold void ff_dsputil_init_x86(DSPContext *c, AVCodecContext *avctx,
                                  unsigned high_bit_depth)
 {
-    int cpu_flags = av_get_cpu_flags();
-
-    if (X86_MMX(cpu_flags))
-        dsputil_init_mmx(c, avctx, cpu_flags, high_bit_depth);
-
     if (CONFIG_ENCODERS)
         ff_dsputilenc_init_mmx(c, avctx, high_bit_depth);
 }
diff --git a/libavcodec/x86/dsputil_mmx.c b/libavcodec/x86/dsputil_mmx.c
deleted file mode 100644 (file)
index dca971e..0000000
+++ /dev/null
@@ -1,150 +0,0 @@
-/*
- * MMX optimized DSP utils
- * Copyright (c) 2000, 2001 Fabrice Bellard
- * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
- *
- * This file is part of FFmpeg.
- *
- * FFmpeg 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.
- *
- * FFmpeg 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 FFmpeg; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
- *
- * MMX optimization by Nick Kurshev <nickols_k@mail.ru>
- */
-
-#include "config.h"
-#include "libavutil/avassert.h"
-#include "libavutil/cpu.h"
-#include "libavutil/x86/asm.h"
-#include "libavcodec/pixels.h"
-#include "libavcodec/videodsp.h"
-#include "dsputil_x86.h"
-#include "inline_asm.h"
-
-#if HAVE_INLINE_ASM
-
-/* Draw the edges of width 'w' of an image of size width, height
- * this MMX version can only handle w == 8 || w == 16. */
-void ff_draw_edges_mmx(uint8_t *buf, int wrap, int width, int height,
-                       int w, int h, int sides)
-{
-    uint8_t *ptr, *last_line;
-    int i;
-
-    last_line = buf + (height - 1) * wrap;
-    /* left and right */
-    ptr = buf;
-    if (w == 8) {
-        __asm__ volatile (
-            "1:                             \n\t"
-            "movd            (%0), %%mm0    \n\t"
-            "punpcklbw      %%mm0, %%mm0    \n\t"
-            "punpcklwd      %%mm0, %%mm0    \n\t"
-            "punpckldq      %%mm0, %%mm0    \n\t"
-            "movq           %%mm0, -8(%0)   \n\t"
-            "movq      -8(%0, %2), %%mm1    \n\t"
-            "punpckhbw      %%mm1, %%mm1    \n\t"
-            "punpckhwd      %%mm1, %%mm1    \n\t"
-            "punpckhdq      %%mm1, %%mm1    \n\t"
-            "movq           %%mm1, (%0, %2) \n\t"
-            "add               %1, %0       \n\t"
-            "cmp               %3, %0       \n\t"
-            "jb                1b           \n\t"
-            : "+r" (ptr)
-            : "r" ((x86_reg) wrap), "r" ((x86_reg) width),
-              "r" (ptr + wrap * height));
-    } else if (w == 16) {
-        __asm__ volatile (
-            "1:                                 \n\t"
-            "movd            (%0), %%mm0        \n\t"
-            "punpcklbw      %%mm0, %%mm0        \n\t"
-            "punpcklwd      %%mm0, %%mm0        \n\t"
-            "punpckldq      %%mm0, %%mm0        \n\t"
-            "movq           %%mm0, -8(%0)       \n\t"
-            "movq           %%mm0, -16(%0)      \n\t"
-            "movq      -8(%0, %2), %%mm1        \n\t"
-            "punpckhbw      %%mm1, %%mm1        \n\t"
-            "punpckhwd      %%mm1, %%mm1        \n\t"
-            "punpckhdq      %%mm1, %%mm1        \n\t"
-            "movq           %%mm1,  (%0, %2)    \n\t"
-            "movq           %%mm1, 8(%0, %2)    \n\t"
-            "add               %1, %0           \n\t"
-            "cmp               %3, %0           \n\t"
-            "jb                1b               \n\t"
-            : "+r"(ptr)
-            : "r"((x86_reg)wrap), "r"((x86_reg)width), "r"(ptr + wrap * height)
-            );
-    } else {
-        av_assert1(w == 4);
-        __asm__ volatile (
-            "1:                             \n\t"
-            "movd            (%0), %%mm0    \n\t"
-            "punpcklbw      %%mm0, %%mm0    \n\t"
-            "punpcklwd      %%mm0, %%mm0    \n\t"
-            "movd           %%mm0, -4(%0)   \n\t"
-            "movd      -4(%0, %2), %%mm1    \n\t"
-            "punpcklbw      %%mm1, %%mm1    \n\t"
-            "punpckhwd      %%mm1, %%mm1    \n\t"
-            "punpckhdq      %%mm1, %%mm1    \n\t"
-            "movd           %%mm1, (%0, %2) \n\t"
-            "add               %1, %0       \n\t"
-            "cmp               %3, %0       \n\t"
-            "jb                1b           \n\t"
-            : "+r" (ptr)
-            : "r" ((x86_reg) wrap), "r" ((x86_reg) width),
-              "r" (ptr + wrap * height));
-    }
-
-    /* top and bottom (and hopefully also the corners) */
-    if (sides & EDGE_TOP) {
-        for (i = 0; i < h; i += 4) {
-            ptr = buf - (i + 1) * wrap - w;
-            __asm__ volatile (
-                "1:                             \n\t"
-                "movq (%1, %0), %%mm0           \n\t"
-                "movq    %%mm0, (%0)            \n\t"
-                "movq    %%mm0, (%0, %2)        \n\t"
-                "movq    %%mm0, (%0, %2, 2)     \n\t"
-                "movq    %%mm0, (%0, %3)        \n\t"
-                "add        $8, %0              \n\t"
-                "cmp        %4, %0              \n\t"
-                "jb         1b                  \n\t"
-                : "+r" (ptr)
-                : "r" ((x86_reg) buf - (x86_reg) ptr - w),
-                  "r" ((x86_reg) - wrap), "r" ((x86_reg) - wrap * 3),
-                  "r" (ptr + width + 2 * w));
-        }
-    }
-
-    if (sides & EDGE_BOTTOM) {
-        for (i = 0; i < h; i += 4) {
-            ptr = last_line + (i + 1) * wrap - w;
-            __asm__ volatile (
-                "1:                             \n\t"
-                "movq (%1, %0), %%mm0           \n\t"
-                "movq    %%mm0, (%0)            \n\t"
-                "movq    %%mm0, (%0, %2)        \n\t"
-                "movq    %%mm0, (%0, %2, 2)     \n\t"
-                "movq    %%mm0, (%0, %3)        \n\t"
-                "add        $8, %0              \n\t"
-                "cmp        %4, %0              \n\t"
-                "jb         1b                  \n\t"
-                : "+r" (ptr)
-                : "r" ((x86_reg) last_line - (x86_reg) ptr - w),
-                  "r" ((x86_reg) wrap), "r" ((x86_reg) wrap * 3),
-                  "r" (ptr + width + 2 * w));
-        }
-    }
-}
-
-#endif /* HAVE_INLINE_ASM */
index c94d004..3e6f364 100644 (file)
@@ -31,9 +31,6 @@ void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx,
                             unsigned high_bit_depth);
 void ff_dsputil_init_pix_mmx(DSPContext *c, AVCodecContext *avctx);
 
-void ff_draw_edges_mmx(uint8_t *buf, int wrap, int width, int height,
-                       int w, int h, int sides);
-
 
 void ff_mmx_idct(int16_t *block);
 void ff_mmxext_idct(int16_t *block);
index 1684189..d91b902 100644 (file)
@@ -17,6 +17,7 @@
  */
 
 #include "libavutil/attributes.h"
+#include "libavutil/avassert.h"
 #include "libavutil/cpu.h"
 #include "libavutil/x86/cpu.h"
 #include "libavcodec/avcodec.h"
@@ -96,6 +97,120 @@ int ff_pix_norm1_sse2(uint8_t *pix, int line_size);
 #undef PHADDD
 #endif /* HAVE_SSSE3_INLINE */
 
+/* Draw the edges of width 'w' of an image of size width, height
+ * this MMX version can only handle w == 8 || w == 16. */
+static void draw_edges_mmx(uint8_t *buf, int wrap, int width, int height,
+                           int w, int h, int sides)
+{
+    uint8_t *ptr, *last_line;
+    int i;
+
+    last_line = buf + (height - 1) * wrap;
+    /* left and right */
+    ptr = buf;
+    if (w == 8) {
+        __asm__ volatile (
+            "1:                             \n\t"
+            "movd            (%0), %%mm0    \n\t"
+            "punpcklbw      %%mm0, %%mm0    \n\t"
+            "punpcklwd      %%mm0, %%mm0    \n\t"
+            "punpckldq      %%mm0, %%mm0    \n\t"
+            "movq           %%mm0, -8(%0)   \n\t"
+            "movq      -8(%0, %2), %%mm1    \n\t"
+            "punpckhbw      %%mm1, %%mm1    \n\t"
+            "punpckhwd      %%mm1, %%mm1    \n\t"
+            "punpckhdq      %%mm1, %%mm1    \n\t"
+            "movq           %%mm1, (%0, %2) \n\t"
+            "add               %1, %0       \n\t"
+            "cmp               %3, %0       \n\t"
+            "jb                1b           \n\t"
+            : "+r" (ptr)
+            : "r" ((x86_reg) wrap), "r" ((x86_reg) width),
+              "r" (ptr + wrap * height));
+    } else if (w == 16) {
+        __asm__ volatile (
+            "1:                                 \n\t"
+            "movd            (%0), %%mm0        \n\t"
+            "punpcklbw      %%mm0, %%mm0        \n\t"
+            "punpcklwd      %%mm0, %%mm0        \n\t"
+            "punpckldq      %%mm0, %%mm0        \n\t"
+            "movq           %%mm0, -8(%0)       \n\t"
+            "movq           %%mm0, -16(%0)      \n\t"
+            "movq      -8(%0, %2), %%mm1        \n\t"
+            "punpckhbw      %%mm1, %%mm1        \n\t"
+            "punpckhwd      %%mm1, %%mm1        \n\t"
+            "punpckhdq      %%mm1, %%mm1        \n\t"
+            "movq           %%mm1,  (%0, %2)    \n\t"
+            "movq           %%mm1, 8(%0, %2)    \n\t"
+            "add               %1, %0           \n\t"
+            "cmp               %3, %0           \n\t"
+            "jb                1b               \n\t"
+            : "+r"(ptr)
+            : "r"((x86_reg)wrap), "r"((x86_reg)width), "r"(ptr + wrap * height)
+            );
+    } else {
+        av_assert1(w == 4);
+        __asm__ volatile (
+            "1:                             \n\t"
+            "movd            (%0), %%mm0    \n\t"
+            "punpcklbw      %%mm0, %%mm0    \n\t"
+            "punpcklwd      %%mm0, %%mm0    \n\t"
+            "movd           %%mm0, -4(%0)   \n\t"
+            "movd      -4(%0, %2), %%mm1    \n\t"
+            "punpcklbw      %%mm1, %%mm1    \n\t"
+            "punpckhwd      %%mm1, %%mm1    \n\t"
+            "punpckhdq      %%mm1, %%mm1    \n\t"
+            "movd           %%mm1, (%0, %2) \n\t"
+            "add               %1, %0       \n\t"
+            "cmp               %3, %0       \n\t"
+            "jb                1b           \n\t"
+            : "+r" (ptr)
+            : "r" ((x86_reg) wrap), "r" ((x86_reg) width),
+              "r" (ptr + wrap * height));
+    }
+
+    /* top and bottom (and hopefully also the corners) */
+    if (sides & EDGE_TOP) {
+        for (i = 0; i < h; i += 4) {
+            ptr = buf - (i + 1) * wrap - w;
+            __asm__ volatile (
+                "1:                             \n\t"
+                "movq (%1, %0), %%mm0           \n\t"
+                "movq    %%mm0, (%0)            \n\t"
+                "movq    %%mm0, (%0, %2)        \n\t"
+                "movq    %%mm0, (%0, %2, 2)     \n\t"
+                "movq    %%mm0, (%0, %3)        \n\t"
+                "add        $8, %0              \n\t"
+                "cmp        %4, %0              \n\t"
+                "jb         1b                  \n\t"
+                : "+r" (ptr)
+                : "r" ((x86_reg) buf - (x86_reg) ptr - w),
+                  "r" ((x86_reg) - wrap), "r" ((x86_reg) - wrap * 3),
+                  "r" (ptr + width + 2 * w));
+        }
+    }
+
+    if (sides & EDGE_BOTTOM) {
+        for (i = 0; i < h; i += 4) {
+            ptr = last_line + (i + 1) * wrap - w;
+            __asm__ volatile (
+                "1:                             \n\t"
+                "movq (%1, %0), %%mm0           \n\t"
+                "movq    %%mm0, (%0)            \n\t"
+                "movq    %%mm0, (%0, %2)        \n\t"
+                "movq    %%mm0, (%0, %2, 2)     \n\t"
+                "movq    %%mm0, (%0, %3)        \n\t"
+                "add        $8, %0              \n\t"
+                "cmp        %4, %0              \n\t"
+                "jb         1b                  \n\t"
+                : "+r" (ptr)
+                : "r" ((x86_reg) last_line - (x86_reg) ptr - w),
+                  "r" ((x86_reg) wrap), "r" ((x86_reg) wrap * 3),
+                  "r" (ptr + width + 2 * w));
+        }
+    }
+}
+
 #endif /* HAVE_INLINE_ASM */
 
 av_cold void ff_mpegvideoencdsp_init_x86(MpegvideoEncDSPContext *c,
@@ -124,6 +239,10 @@ av_cold void ff_mpegvideoencdsp_init_x86(MpegvideoEncDSPContext *c,
             c->try_8x8basis = try_8x8basis_mmx;
         }
         c->add_8x8basis = add_8x8basis_mmx;
+
+        if (avctx->bits_per_raw_sample <= 8) {
+            c->draw_edges = draw_edges_mmx;
+        }
     }
 
     if (INLINE_AMD3DNOW(cpu_flags)) {