]> git.sesse.net Git - ffmpeg/blobdiff - libavcodec/vdpau.c
floatdsp: move butterflies_float from dsputil to avfloatdsp.
[ffmpeg] / libavcodec / vdpau.c
index 805fe79fe9ab11b28048da1f3136f66dbef6e0f3..851c7cb9f322ce032e813f081a32d35121cec2c9 100644 (file)
@@ -1,23 +1,23 @@
 /*
  * Video Decode and Presentation API for UNIX (VDPAU) is used for
- * HW decode acceleration for MPEG-1/2, H.264 and VC-1.
+ * HW decode acceleration for MPEG-1/2, MPEG-4 ASP, H.264 and VC-1.
  *
  * Copyright (c) 2008 NVIDIA
  *
- * This file is part of FFmpeg.
+ * This file is part of Libav.
  *
- * FFmpeg is free software; you can redistribute it and/or
+ * 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.
  *
- * FFmpeg is distributed in the hope that it will be useful,
+ * 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 FFmpeg; if not, write to the Free Software
+ * License along with Libav; if not, write to the Free Software
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
 #include "vdpau_internal.h"
 
 /**
- * \addtogroup VDPAU_Decoding
+ * @addtogroup VDPAU_Decoding
  *
  * @{
  */
 
+int ff_vdpau_common_start_frame(AVCodecContext *avctx,
+                                av_unused const uint8_t *buffer,
+                                av_unused uint32_t size)
+{
+    AVVDPAUContext *hwctx = avctx->hwaccel_context;
+
+    hwctx->bitstream_buffers_used = 0;
+    return 0;
+}
+
+int ff_vdpau_common_end_frame(AVCodecContext *avctx)
+{
+    MpegEncContext * const s = avctx->priv_data;
+    AVVDPAUContext *hwctx = avctx->hwaccel_context;
+
+    if (hwctx->bitstream_buffers_used) {
+        VdpVideoSurface surf = ff_vdpau_get_surface_id(s->current_picture_ptr);
+
+        hwctx->render(hwctx->decoder, surf, (void *)&hwctx->info,
+                      hwctx->bitstream_buffers_used, hwctx->bitstream_buffers);
+
+        ff_draw_horiz_band(s, 0, s->avctx->height);
+        hwctx->bitstream_buffers_used = 0;
+    }
+    return 0;
+}
+
+int ff_vdpau_add_buffer(AVCodecContext *avctx,
+                        const uint8_t *buf, uint32_t size)
+{
+    AVVDPAUContext *hwctx = avctx->hwaccel_context;
+    VdpBitstreamBuffer *buffers = hwctx->bitstream_buffers;
+
+    buffers = av_fast_realloc(buffers, &hwctx->bitstream_buffers_allocated,
+                              (hwctx->bitstream_buffers_used + 1) * sizeof(*buffers));
+    if (!buffers)
+        return AVERROR(ENOMEM);
+
+    hwctx->bitstream_buffers = buffers;
+    buffers += hwctx->bitstream_buffers_used++;
+
+    buffers->struct_version  = VDP_BITSTREAM_BUFFER_VERSION;
+    buffers->bitstream       = buf;
+    buffers->bitstream_bytes = size;
+    return 0;
+}
+
+/* Obsolete non-hwaccel VDPAU support below... */
+
 void ff_vdpau_h264_set_reference_frames(MpegEncContext *s)
 {
     H264Context *h = s->avctx->priv_data;
@@ -46,7 +95,7 @@ void ff_vdpau_h264_set_reference_frames(MpegEncContext *s)
     Picture *pic;
     int i, list, pic_frame_idx;
 
-    render = (struct vdpau_render_state *)s->current_picture_ptr->data[0];
+    render = (struct vdpau_render_state *)s->current_picture_ptr->f.data[0];
     assert(render);
 
     rf = &render->info.h264.referenceFrames[0];
@@ -54,15 +103,15 @@ void ff_vdpau_h264_set_reference_frames(MpegEncContext *s)
 
     for (list = 0; list < 2; ++list) {
         Picture **lp = list ? h->long_ref : h->short_ref;
-        int ls = list ? h->long_ref_count : h->short_ref_count;
+        int ls = list ? 16 : h->short_ref_count;
 
         for (i = 0; i < ls; ++i) {
             pic = lp[i];
-            if (!pic || !pic->reference)
+            if (!pic || !pic->f.reference)
                 continue;
             pic_frame_idx = pic->long_ref ? pic->pic_id : pic->frame_num;
 
-            render_ref = (struct vdpau_render_state *)pic->data[0];
+            render_ref = (struct vdpau_render_state *)pic->f.data[0];
             assert(render_ref);
 
             rf2 = &render->info.h264.referenceFrames[0];
@@ -76,8 +125,8 @@ void ff_vdpau_h264_set_reference_frames(MpegEncContext *s)
                 ++rf2;
             }
             if (rf2 != rf) {
-                rf2->top_is_reference    |= (pic->reference & PICT_TOP_FIELD)    ? VDP_TRUE : VDP_FALSE;
-                rf2->bottom_is_reference |= (pic->reference & PICT_BOTTOM_FIELD) ? VDP_TRUE : VDP_FALSE;
+                rf2->top_is_reference    |= (pic->f.reference & PICT_TOP_FIELD)    ? VDP_TRUE : VDP_FALSE;
+                rf2->bottom_is_reference |= (pic->f.reference & PICT_BOTTOM_FIELD) ? VDP_TRUE : VDP_FALSE;
                 continue;
             }
 
@@ -86,8 +135,8 @@ void ff_vdpau_h264_set_reference_frames(MpegEncContext *s)
 
             rf->surface             = render_ref->surface;
             rf->is_long_term        = pic->long_ref;
-            rf->top_is_reference    = (pic->reference & PICT_TOP_FIELD)    ? VDP_TRUE : VDP_FALSE;
-            rf->bottom_is_reference = (pic->reference & PICT_BOTTOM_FIELD) ? VDP_TRUE : VDP_FALSE;
+            rf->top_is_reference    = (pic->f.reference & PICT_TOP_FIELD)    ? VDP_TRUE : VDP_FALSE;
+            rf->bottom_is_reference = (pic->f.reference & PICT_BOTTOM_FIELD) ? VDP_TRUE : VDP_FALSE;
             rf->field_order_cnt[0]  = pic->field_poc[0];
             rf->field_order_cnt[1]  = pic->field_poc[1];
             rf->frame_idx           = pic_frame_idx;
@@ -112,7 +161,7 @@ void ff_vdpau_add_data_chunk(MpegEncContext *s,
 {
     struct vdpau_render_state *render;
 
-    render = (struct vdpau_render_state *)s->current_picture_ptr->data[0];
+    render = (struct vdpau_render_state *)s->current_picture_ptr->f.data[0];
     assert(render);
 
     render->bitstream_buffers= av_fast_realloc(
@@ -127,19 +176,15 @@ void ff_vdpau_add_data_chunk(MpegEncContext *s,
     render->bitstream_buffers_used++;
 }
 
-void ff_vdpau_h264_picture_complete(MpegEncContext *s)
+void ff_vdpau_h264_picture_start(MpegEncContext *s)
 {
     H264Context *h = s->avctx->priv_data;
     struct vdpau_render_state *render;
     int i;
 
-    render = (struct vdpau_render_state *)s->current_picture_ptr->data[0];
+    render = (struct vdpau_render_state *)s->current_picture_ptr->f.data[0];
     assert(render);
 
-    render->info.h264.slice_count = h->slice_num;
-    if (render->info.h264.slice_count < 1)
-        return;
-
     for (i = 0; i < 2; ++i) {
         int foc = s->current_picture_ptr->field_poc[i];
         if (foc == INT_MAX)
@@ -147,8 +192,22 @@ void ff_vdpau_h264_picture_complete(MpegEncContext *s)
         render->info.h264.field_order_cnt[i] = foc;
     }
 
-    render->info.h264.is_reference                           = s->current_picture_ptr->reference ? VDP_TRUE : VDP_FALSE;
-    render->info.h264.frame_num                              = h->frame_num;
+    render->info.h264.frame_num = h->frame_num;
+}
+
+void ff_vdpau_h264_picture_complete(MpegEncContext *s)
+{
+    H264Context *h = s->avctx->priv_data;
+    struct vdpau_render_state *render;
+
+    render = (struct vdpau_render_state *)s->current_picture_ptr->f.data[0];
+    assert(render);
+
+    render->info.h264.slice_count = h->slice_num;
+    if (render->info.h264.slice_count < 1)
+        return;
+
+    render->info.h264.is_reference                           = (s->current_picture_ptr->f.reference & 3) ? VDP_TRUE : VDP_FALSE;
     render->info.h264.field_pic_flag                         = s->picture_structure != PICT_FRAME;
     render->info.h264.bottom_field_flag                      = s->picture_structure == PICT_BOTTOM_FIELD;
     render->info.h264.num_ref_frames                         = h->sps.ref_frame_count;
@@ -165,7 +224,7 @@ void ff_vdpau_h264_picture_complete(MpegEncContext *s)
     render->info.h264.num_ref_idx_l1_active_minus1           = h->pps.ref_count[1] - 1;
     render->info.h264.log2_max_frame_num_minus4              = h->sps.log2_max_frame_num - 4;
     render->info.h264.pic_order_cnt_type                     = h->sps.poc_type;
-    render->info.h264.log2_max_pic_order_cnt_lsb_minus4      = h->sps.log2_max_poc_lsb - 4;
+    render->info.h264.log2_max_pic_order_cnt_lsb_minus4      = h->sps.poc_type ? 0 : h->sps.log2_max_poc_lsb - 4;
     render->info.h264.delta_pic_order_always_zero_flag       = h->sps.delta_pic_order_always_zero_flag;
     render->info.h264.direct_8x8_inference_flag              = h->sps.direct_8x8_inference_flag;
     render->info.h264.entropy_coding_mode_flag               = h->pps.cabac;
@@ -173,7 +232,8 @@ void ff_vdpau_h264_picture_complete(MpegEncContext *s)
     render->info.h264.deblocking_filter_control_present_flag = h->pps.deblocking_filter_parameters_present;
     render->info.h264.redundant_pic_cnt_present_flag         = h->pps.redundant_pic_cnt_present;
     memcpy(render->info.h264.scaling_lists_4x4, h->pps.scaling_matrix4, sizeof(render->info.h264.scaling_lists_4x4));
-    memcpy(render->info.h264.scaling_lists_8x8, h->pps.scaling_matrix8, sizeof(render->info.h264.scaling_lists_8x8));
+    memcpy(render->info.h264.scaling_lists_8x8[0], h->pps.scaling_matrix8[0], sizeof(render->info.h264.scaling_lists_8x8[0]));
+    memcpy(render->info.h264.scaling_lists_8x8[1], h->pps.scaling_matrix8[3], sizeof(render->info.h264.scaling_lists_8x8[0]));
 
     ff_draw_horiz_band(s, 0, s->avctx->height);
     render->bitstream_buffers_used = 0;
@@ -187,7 +247,7 @@ void ff_vdpau_mpeg_picture_complete(MpegEncContext *s, const uint8_t *buf,
 
     if (!s->current_picture_ptr) return;
 
-    render = (struct vdpau_render_state *)s->current_picture_ptr->data[0];
+    render = (struct vdpau_render_state *)s->current_picture_ptr->f.data[0];
     assert(render);
 
     /* fill VdpPictureInfoMPEG1Or2 struct */
@@ -215,13 +275,13 @@ void ff_vdpau_mpeg_picture_complete(MpegEncContext *s, const uint8_t *buf,
     render->info.mpeg.backward_reference         = VDP_INVALID_HANDLE;
 
     switch(s->pict_type){
-    case  FF_B_TYPE:
-        next = (struct vdpau_render_state *)s->next_picture.data[0];
+    case  AV_PICTURE_TYPE_B:
+        next = (struct vdpau_render_state *)s->next_picture.f.data[0];
         assert(next);
         render->info.mpeg.backward_reference     = next->surface;
         // no return here, going to set forward prediction
-    case  FF_P_TYPE:
-        last = (struct vdpau_render_state *)s->last_picture.data[0];
+    case  AV_PICTURE_TYPE_P:
+        last = (struct vdpau_render_state *)s->last_picture.f.data[0];
         if (!last) // FIXME: Does this test make sense?
             last = render; // predict second field from the first
         render->info.mpeg.forward_reference      = last->surface;
@@ -242,7 +302,7 @@ void ff_vdpau_vc1_decode_picture(MpegEncContext *s, const uint8_t *buf,
     VC1Context *v = s->avctx->priv_data;
     struct vdpau_render_state *render, *last, *next;
 
-    render = (struct vdpau_render_state *)s->current_picture.data[0];
+    render = (struct vdpau_render_state *)s->current_picture.f.data[0];
     assert(render);
 
     /*  fill LvPictureInfoVC1 struct */
@@ -270,7 +330,7 @@ void ff_vdpau_vc1_decode_picture(MpegEncContext *s, const uint8_t *buf,
     /* Specific to simple/main profile only */
     render->info.vc1.multires           = v->multires;
     render->info.vc1.syncmarker         = v->s.resync_marker;
-    render->info.vc1.rangered           = v->rangered;
+    render->info.vc1.rangered           = v->rangered | (v->rangeredfrm << 1);
     render->info.vc1.maxbframes         = v->s.max_b_frames;
 
     render->info.vc1.deblockEnable      = v->postprocflag & 1;
@@ -285,13 +345,13 @@ void ff_vdpau_vc1_decode_picture(MpegEncContext *s, const uint8_t *buf,
         render->info.vc1.picture_type = s->pict_type - 1 + s->pict_type / 3;
 
     switch(s->pict_type){
-    case  FF_B_TYPE:
-        next = (struct vdpau_render_state *)s->next_picture.data[0];
+    case  AV_PICTURE_TYPE_B:
+        next = (struct vdpau_render_state *)s->next_picture.f.data[0];
         assert(next);
         render->info.vc1.backward_reference = next->surface;
         // no break here, going to set forward prediction
-    case  FF_P_TYPE:
-        last = (struct vdpau_render_state *)s->last_picture.data[0];
+    case  AV_PICTURE_TYPE_P:
+        last = (struct vdpau_render_state *)s->last_picture.f.data[0];
         if (!last) // FIXME: Does this test make sense?
             last = render; // predict second field from the first
         render->info.vc1.forward_reference = last->surface;
@@ -305,4 +365,58 @@ void ff_vdpau_vc1_decode_picture(MpegEncContext *s, const uint8_t *buf,
     render->bitstream_buffers_used        = 0;
 }
 
+void ff_vdpau_mpeg4_decode_picture(MpegEncContext *s, const uint8_t *buf,
+                                   int buf_size)
+{
+    struct vdpau_render_state *render, *last, *next;
+    int i;
+
+    if (!s->current_picture_ptr) return;
+
+    render = (struct vdpau_render_state *)s->current_picture_ptr->f.data[0];
+    assert(render);
+
+    /* fill VdpPictureInfoMPEG4Part2 struct */
+    render->info.mpeg4.trd[0]                            = s->pp_time;
+    render->info.mpeg4.trb[0]                            = s->pb_time;
+    render->info.mpeg4.trd[1]                            = s->pp_field_time >> 1;
+    render->info.mpeg4.trb[1]                            = s->pb_field_time >> 1;
+    render->info.mpeg4.vop_time_increment_resolution     = s->avctx->time_base.den;
+    render->info.mpeg4.vop_coding_type                   = 0;
+    render->info.mpeg4.vop_fcode_forward                 = s->f_code;
+    render->info.mpeg4.vop_fcode_backward                = s->b_code;
+    render->info.mpeg4.resync_marker_disable             = !s->resync_marker;
+    render->info.mpeg4.interlaced                        = !s->progressive_sequence;
+    render->info.mpeg4.quant_type                        = s->mpeg_quant;
+    render->info.mpeg4.quarter_sample                    = s->quarter_sample;
+    render->info.mpeg4.short_video_header                = s->avctx->codec->id == AV_CODEC_ID_H263;
+    render->info.mpeg4.rounding_control                  = s->no_rounding;
+    render->info.mpeg4.alternate_vertical_scan_flag      = s->alternate_scan;
+    render->info.mpeg4.top_field_first                   = s->top_field_first;
+    for (i = 0; i < 64; ++i) {
+        render->info.mpeg4.intra_quantizer_matrix[i]     = s->intra_matrix[i];
+        render->info.mpeg4.non_intra_quantizer_matrix[i] = s->inter_matrix[i];
+    }
+    render->info.mpeg4.forward_reference                 = VDP_INVALID_HANDLE;
+    render->info.mpeg4.backward_reference                = VDP_INVALID_HANDLE;
+
+    switch (s->pict_type) {
+    case AV_PICTURE_TYPE_B:
+        next = (struct vdpau_render_state *)s->next_picture.f.data[0];
+        assert(next);
+        render->info.mpeg4.backward_reference     = next->surface;
+        render->info.mpeg4.vop_coding_type        = 2;
+        // no break here, going to set forward prediction
+    case AV_PICTURE_TYPE_P:
+        last = (struct vdpau_render_state *)s->last_picture.f.data[0];
+        assert(last);
+        render->info.mpeg4.forward_reference      = last->surface;
+    }
+
+    ff_vdpau_add_data_chunk(s, buf, buf_size);
+
+    ff_draw_horiz_band(s, 0, s->avctx->height);
+    render->bitstream_buffers_used = 0;
+}
+
 /* @}*/