Browse code

vaapi: Add VP9 hwaccell support

Signed-off-by: Timo Rothenpieler <timo@rothenpieler.org>

Timo Rothenpieler authored on 2015/12/20 04:31:00
Showing 7 changed files
... ...
@@ -46,6 +46,7 @@ version <next>:
46 46
 - mips64r6 option has been removed
47 47
 - DXVA2-accelerated VP9 decoding
48 48
 - SOFAlizer: virtual binaural acoustics filter
49
+- VAAPI VP9 hwaccel
49 50
 
50 51
 
51 52
 version 2.8:
... ...
@@ -2577,6 +2577,8 @@ vp9_d3d11va_hwaccel_deps="d3d11va DXVA_PicParams_VP9"
2577 2577
 vp9_d3d11va_hwaccel_select="vp9_decoder"
2578 2578
 vp9_dxva2_hwaccel_deps="dxva2 DXVA_PicParams_VP9"
2579 2579
 vp9_dxva2_hwaccel_select="vp9_decoder"
2580
+vp9_vaapi_hwaccel_deps="vaapi VADecPictureParameterBufferVP9"
2581
+vp9_vaapi_hwaccel_select="vp9_decoder"
2580 2582
 wmv3_crystalhd_decoder_select="crystalhd"
2581 2583
 wmv3_d3d11va_hwaccel_select="vc1_d3d11va_hwaccel"
2582 2584
 wmv3_dxva2_hwaccel_select="vc1_dxva2_hwaccel"
... ...
@@ -5319,6 +5321,7 @@ check_type "windows.h d3d11.h" "ID3D11VideoContext"
5319 5319
 check_type "d3d9.h dxva2api.h" DXVA2_ConfigPictureDecode -D_WIN32_WINNT=0x0602
5320 5320
 
5321 5321
 check_type "va/va.h" "VAPictureParameterBufferHEVC"
5322
+check_type "va/va.h" "VADecPictureParameterBufferVP9"
5322 5323
 
5323 5324
 check_type "vdpau/vdpau.h" "VdpPictureInfoHEVC"
5324 5325
 
... ...
@@ -751,6 +751,7 @@ OBJS-$(CONFIG_VC1_VAAPI_HWACCEL)          += vaapi_vc1.o
751 751
 OBJS-$(CONFIG_VC1_VDPAU_HWACCEL)          += vdpau_vc1.o
752 752
 OBJS-$(CONFIG_VP9_D3D11VA_HWACCEL)        += dxva2_vp9.o
753 753
 OBJS-$(CONFIG_VP9_DXVA2_HWACCEL)          += dxva2_vp9.o
754
+OBJS-$(CONFIG_VP9_VAAPI_HWACCEL)          += vaapi_vp9.o
754 755
 
755 756
 # libavformat dependencies
756 757
 OBJS-$(CONFIG_ADTS_MUXER)              += mpeg4audio.o
... ...
@@ -112,6 +112,7 @@ void avcodec_register_all(void)
112 112
     REGISTER_HWACCEL(VC1_QSV,           vc1_qsv);
113 113
     REGISTER_HWACCEL(VP9_D3D11VA,       vp9_d3d11va);
114 114
     REGISTER_HWACCEL(VP9_DXVA2,         vp9_dxva2);
115
+    REGISTER_HWACCEL(VP9_VAAPI,         vp9_vaapi);
115 116
     REGISTER_HWACCEL(WMV3_D3D11VA,      wmv3_d3d11va);
116 117
     REGISTER_HWACCEL(WMV3_DXVA2,        wmv3_dxva2);
117 118
     REGISTER_HWACCEL(WMV3_VAAPI,        wmv3_vaapi);
118 119
new file mode 100644
... ...
@@ -0,0 +1,168 @@
0
+/*
1
+ * VP9 HW decode acceleration through VA API
2
+ *
3
+ * Copyright (C) 2015 Timo Rothenpieler <timo@rothenpieler.org>
4
+ *
5
+ * This file is part of FFmpeg.
6
+ *
7
+ * FFmpeg is free software; you can redistribute it and/or
8
+ * modify it under the terms of the GNU Lesser General Public
9
+ * License as published by the Free Software Foundation; either
10
+ * version 2.1 of the License, or (at your option) any later version.
11
+ *
12
+ * FFmpeg is distributed in the hope that it will be useful,
13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15
+ * Lesser General Public License for more details.
16
+ *
17
+ * You should have received a copy of the GNU Lesser General Public
18
+ * License along with FFmpeg; if not, write to the Free Software
19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20
+ */
21
+
22
+#include "libavutil/pixdesc.h"
23
+#include "vaapi_internal.h"
24
+#include "vp9.h"
25
+
26
+static void fill_picture_parameters(AVCodecContext                 *avctx,
27
+                                    const VP9SharedContext         *h,
28
+                                    VADecPictureParameterBufferVP9 *pp)
29
+{
30
+    const AVPixFmtDescriptor *pixdesc = av_pix_fmt_desc_get(avctx->sw_pix_fmt);
31
+    int i;
32
+
33
+    pp->frame_width = avctx->width;
34
+    pp->frame_height = avctx->height;
35
+
36
+    pp->frame_header_length_in_bytes = h->h.uncompressed_header_size;
37
+    pp->first_partition_size = h->h.compressed_header_size;
38
+
39
+    pp->profile = h->h.profile;
40
+
41
+    pp->filter_level = h->h.filter.level;
42
+    pp->sharpness_level = h->h.filter.sharpness;
43
+    pp->log2_tile_rows = h->h.tiling.log2_tile_rows;
44
+    pp->log2_tile_columns = h->h.tiling.log2_tile_cols;
45
+
46
+    pp->pic_fields.bits.subsampling_x = pixdesc->log2_chroma_w;
47
+    pp->pic_fields.bits.subsampling_y = pixdesc->log2_chroma_h;
48
+    pp->pic_fields.bits.frame_type = !h->h.keyframe;
49
+    pp->pic_fields.bits.show_frame = !h->h.invisible;
50
+    pp->pic_fields.bits.error_resilient_mode = h->h.errorres;
51
+    pp->pic_fields.bits.intra_only = h->h.intraonly;
52
+    pp->pic_fields.bits.allow_high_precision_mv = h->h.keyframe ? 0 : h->h.highprecisionmvs;
53
+    pp->pic_fields.bits.mcomp_filter_type = h->h.filtermode ^ (h->h.filtermode <= 1);
54
+    pp->pic_fields.bits.frame_parallel_decoding_mode = h->h.parallelmode;
55
+    pp->pic_fields.bits.reset_frame_context = h->h.resetctx;
56
+    pp->pic_fields.bits.refresh_frame_context = h->h.refreshctx;
57
+    pp->pic_fields.bits.frame_context_idx = h->h.framectxid;
58
+
59
+    pp->pic_fields.bits.segmentation_enabled = h->h.segmentation.enabled;
60
+    pp->pic_fields.bits.segmentation_temporal_update = h->h.segmentation.temporal;
61
+    pp->pic_fields.bits.segmentation_update_map = h->h.segmentation.update_map;
62
+
63
+    pp->pic_fields.bits.last_ref_frame = h->h.refidx[0];
64
+    pp->pic_fields.bits.last_ref_frame_sign_bias = h->h.signbias[0];
65
+    pp->pic_fields.bits.golden_ref_frame = h->h.refidx[1];
66
+    pp->pic_fields.bits.golden_ref_frame_sign_bias = h->h.signbias[1];
67
+    pp->pic_fields.bits.alt_ref_frame = h->h.refidx[2];
68
+    pp->pic_fields.bits.alt_ref_frame_sign_bias = h->h.signbias[2];
69
+    pp->pic_fields.bits.lossless_flag = h->h.lossless;
70
+
71
+    for (i = 0; i < 7; i++)
72
+        pp->mb_segment_tree_probs[i] = h->h.segmentation.prob[i];
73
+
74
+    if (h->h.segmentation.temporal) {
75
+        for (i = 0; i < 3; i++)
76
+            pp->segment_pred_probs[i] = h->h.segmentation.pred_prob[i];
77
+    } else {
78
+        memset(pp->segment_pred_probs, 255, sizeof(pp->segment_pred_probs));
79
+    }
80
+
81
+    for (i = 0; i < 8; i++) {
82
+        if (h->refs[i].f->buf[0]) {
83
+            pp->reference_frames[i] = ff_vaapi_get_surface_id(h->refs[i].f);
84
+        } else {
85
+            pp->reference_frames[i] = VA_INVALID_ID;
86
+        }
87
+    }
88
+}
89
+
90
+static int vaapi_vp9_start_frame(AVCodecContext          *avctx,
91
+                                 av_unused const uint8_t *buffer,
92
+                                 av_unused uint32_t       size)
93
+{
94
+    const VP9SharedContext *h = avctx->priv_data;
95
+    FFVAContext * const vactx = ff_vaapi_get_context(avctx);
96
+    VADecPictureParameterBufferVP9 *pic_param;
97
+
98
+    vactx->slice_param_size = sizeof(VASliceParameterBufferVP9);
99
+
100
+    pic_param = ff_vaapi_alloc_pic_param(vactx, sizeof(VADecPictureParameterBufferVP9));
101
+    if (!pic_param)
102
+        return -1;
103
+    fill_picture_parameters(avctx, h, pic_param);
104
+
105
+    return 0;
106
+}
107
+
108
+static int vaapi_vp9_end_frame(AVCodecContext *avctx)
109
+{
110
+    FFVAContext * const vactx = ff_vaapi_get_context(avctx);
111
+    const VP9SharedContext *h = avctx->priv_data;
112
+    int ret;
113
+
114
+    ret = ff_vaapi_commit_slices(vactx);
115
+    if (ret < 0)
116
+        goto finish;
117
+
118
+    ret = ff_vaapi_render_picture(vactx, ff_vaapi_get_surface_id(h->frames[CUR_FRAME].tf.f));
119
+    if (ret < 0)
120
+        goto finish;
121
+
122
+finish:
123
+    ff_vaapi_common_end_frame(avctx);
124
+    return ret;
125
+}
126
+
127
+static int vaapi_vp9_decode_slice(AVCodecContext *avctx,
128
+                                  const uint8_t  *buffer,
129
+                                  uint32_t        size)
130
+{
131
+    FFVAContext * const vactx = ff_vaapi_get_context(avctx);
132
+    const VP9SharedContext *h = avctx->priv_data;
133
+    VASliceParameterBufferVP9 *slice_param;
134
+    int i;
135
+
136
+    slice_param = (VASliceParameterBufferVP9*)ff_vaapi_alloc_slice(vactx, buffer, size);
137
+    if (!slice_param)
138
+        return -1;
139
+
140
+    for (i = 0; i < 8; i++) {
141
+        slice_param->seg_param[i].segment_flags.fields.segment_reference_enabled = h->h.segmentation.feat[i].ref_enabled;
142
+        slice_param->seg_param[i].segment_flags.fields.segment_reference = h->h.segmentation.feat[i].ref_val;
143
+        slice_param->seg_param[i].segment_flags.fields.segment_reference_skipped = h->h.segmentation.feat[i].skip_enabled;
144
+
145
+        memcpy(slice_param->seg_param[i].filter_level, h->h.segmentation.feat[i].lflvl, sizeof(slice_param->seg_param[i].filter_level));
146
+
147
+        slice_param->seg_param[i].luma_dc_quant_scale = h->h.segmentation.feat[i].qmul[0][0];
148
+        slice_param->seg_param[i].luma_ac_quant_scale = h->h.segmentation.feat[i].qmul[0][1];
149
+        slice_param->seg_param[i].chroma_dc_quant_scale = h->h.segmentation.feat[i].qmul[1][0];
150
+        slice_param->seg_param[i].chroma_ac_quant_scale = h->h.segmentation.feat[i].qmul[1][1];
151
+    }
152
+
153
+    return 0;
154
+}
155
+
156
+AVHWAccel ff_vp9_vaapi_hwaccel = {
157
+    .name                 = "vp9_vaapi",
158
+    .type                 = AVMEDIA_TYPE_VIDEO,
159
+    .id                   = AV_CODEC_ID_VP9,
160
+    .pix_fmt              = AV_PIX_FMT_VAAPI,
161
+    .start_frame          = vaapi_vp9_start_frame,
162
+    .end_frame            = vaapi_vp9_end_frame,
163
+    .decode_slice         = vaapi_vp9_decode_slice,
164
+    .init                 = ff_vaapi_context_init,
165
+    .uninit               = ff_vaapi_context_fini,
166
+    .priv_data_size       = sizeof(FFVAContext),
167
+};
... ...
@@ -29,7 +29,7 @@
29 29
 #include "libavutil/version.h"
30 30
 
31 31
 #define LIBAVCODEC_VERSION_MAJOR  57
32
-#define LIBAVCODEC_VERSION_MINOR  19
32
+#define LIBAVCODEC_VERSION_MINOR  20
33 33
 #define LIBAVCODEC_VERSION_MICRO 100
34 34
 
35 35
 #define LIBAVCODEC_VERSION_INT  AV_VERSION_INT(LIBAVCODEC_VERSION_MAJOR, \
... ...
@@ -240,7 +240,7 @@ fail:
240 240
 
241 241
 static int update_size(AVCodecContext *ctx, int w, int h)
242 242
 {
243
-#define HWACCEL_MAX (CONFIG_VP9_DXVA2_HWACCEL + CONFIG_VP9_D3D11VA_HWACCEL)
243
+#define HWACCEL_MAX (CONFIG_VP9_DXVA2_HWACCEL + CONFIG_VP9_D3D11VA_HWACCEL + CONFIG_VP9_VAAPI_HWACCEL)
244 244
     enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmtp = pix_fmts;
245 245
     VP9Context *s = ctx->priv_data;
246 246
     uint8_t *p;
... ...
@@ -261,6 +261,9 @@ static int update_size(AVCodecContext *ctx, int w, int h)
261 261
 #if CONFIG_VP9_D3D11VA_HWACCEL
262 262
         *fmtp++ = AV_PIX_FMT_D3D11VA_VLD;
263 263
 #endif
264
+#if CONFIG_VP9_VAAPI_HWACCEL
265
+        *fmtp++ = AV_PIX_FMT_VAAPI;
266
+#endif
264 267
     }
265 268
 
266 269
     *fmtp++ = s->pix_fmt;