Browse code

vaapi: Add VP8 decode hwaccel

Mark Thompson authored on 2017/11/19 02:55:24
Showing 5 changed files
... ...
@@ -2746,6 +2746,8 @@ vc1_vaapi_hwaccel_deps="vaapi"
2746 2746
 vc1_vaapi_hwaccel_select="vc1_decoder"
2747 2747
 vc1_vdpau_hwaccel_deps="vdpau"
2748 2748
 vc1_vdpau_hwaccel_select="vc1_decoder"
2749
+vp8_vaapi_hwaccel_deps="vaapi VAPictureParameterBufferVP8"
2750
+vp8_vaapi_hwaccel_select="vp8_decoder"
2749 2751
 vp9_d3d11va_hwaccel_deps="d3d11va DXVA_PicParams_VP9"
2750 2752
 vp9_d3d11va_hwaccel_select="vp9_decoder"
2751 2753
 vp9_d3d11va2_hwaccel_deps="d3d11va DXVA_PicParams_VP9"
... ...
@@ -5719,6 +5721,7 @@ check_type "windows.h d3d11.h" "ID3D11VideoContext"
5719 5719
 check_type "d3d9.h dxva2api.h" DXVA2_ConfigPictureDecode -D_WIN32_WINNT=0x0602
5720 5720
 
5721 5721
 check_type "va/va.h va/va_dec_hevc.h" "VAPictureParameterBufferHEVC"
5722
+check_type "va/va.h va/va_dec_vp8.h" "VAPictureParameterBufferVP8"
5722 5723
 check_struct "va/va.h" "VADecPictureParameterBufferVP9" bit_depth
5723 5724
 check_type "va/va.h va/va_vpp.h" "VAProcPipelineParameterBuffer"
5724 5725
 check_type "va/va.h va/va_enc_h264.h" "VAEncPictureParameterBufferH264"
... ...
@@ -871,6 +871,7 @@ OBJS-$(CONFIG_VC1_NVDEC_HWACCEL)          += nvdec_vc1.o
871 871
 OBJS-$(CONFIG_VC1_QSV_HWACCEL)            += qsvdec_other.o
872 872
 OBJS-$(CONFIG_VC1_VAAPI_HWACCEL)          += vaapi_vc1.o
873 873
 OBJS-$(CONFIG_VC1_VDPAU_HWACCEL)          += vdpau_vc1.o
874
+OBJS-$(CONFIG_VP8_VAAPI_HWACCEL)          += vaapi_vp8.o
874 875
 OBJS-$(CONFIG_VP9_D3D11VA_HWACCEL)        += dxva2_vp9.o
875 876
 OBJS-$(CONFIG_VP9_DXVA2_HWACCEL)          += dxva2_vp9.o
876 877
 OBJS-$(CONFIG_VP9_NVDEC_HWACCEL)          += nvdec_vp9.o
... ...
@@ -59,6 +59,7 @@ extern const AVHWAccel ff_vc1_dxva2_hwaccel;
59 59
 extern const AVHWAccel ff_vc1_nvdec_hwaccel;
60 60
 extern const AVHWAccel ff_vc1_vaapi_hwaccel;
61 61
 extern const AVHWAccel ff_vc1_vdpau_hwaccel;
62
+extern const AVHWAccel ff_vp8_vaapi_hwaccel;
62 63
 extern const AVHWAccel ff_vp9_d3d11va_hwaccel;
63 64
 extern const AVHWAccel ff_vp9_d3d11va2_hwaccel;
64 65
 extern const AVHWAccel ff_vp9_dxva2_hwaccel;
65 66
new file mode 100644
... ...
@@ -0,0 +1,237 @@
0
+/*
1
+ * This file is part of FFmpeg.
2
+ *
3
+ * FFmpeg is free software; you can redistribute it and/or
4
+ * modify it under the terms of the GNU Lesser General Public
5
+ * License as published by the Free Software Foundation; either
6
+ * version 2.1 of the License, or (at your option) any later version.
7
+ *
8
+ * FFmpeg is distributed in the hope that it will be useful,
9
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
10
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11
+ * Lesser General Public License for more details.
12
+ *
13
+ * You should have received a copy of the GNU Lesser General Public
14
+ * License along with FFmpeg; if not, write to the Free Software
15
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
16
+ */
17
+
18
+#include <va/va.h>
19
+#include <va/va_dec_vp8.h>
20
+
21
+#include "hwaccel.h"
22
+#include "vaapi_decode.h"
23
+#include "vp8.h"
24
+
25
+static VASurfaceID vaapi_vp8_surface_id(VP8Frame *vf)
26
+{
27
+    if (vf)
28
+        return ff_vaapi_get_surface_id(vf->tf.f);
29
+    else
30
+        return VA_INVALID_SURFACE;
31
+}
32
+
33
+static int vaapi_vp8_start_frame(AVCodecContext          *avctx,
34
+                                 av_unused const uint8_t *buffer,
35
+                                 av_unused uint32_t       size)
36
+{
37
+    const VP8Context *s = avctx->priv_data;
38
+    VAAPIDecodePicture *pic = s->framep[VP56_FRAME_CURRENT]->hwaccel_picture_private;
39
+    VAPictureParameterBufferVP8 pp;
40
+    VAProbabilityDataBufferVP8 prob;
41
+    VAIQMatrixBufferVP8 quant;
42
+    int err, i, j, k;
43
+
44
+    pic->output_surface = vaapi_vp8_surface_id(s->framep[VP56_FRAME_CURRENT]);
45
+
46
+    pp = (VAPictureParameterBufferVP8) {
47
+        .frame_width                     = avctx->width,
48
+        .frame_height                    = avctx->height,
49
+
50
+        .last_ref_frame                  = vaapi_vp8_surface_id(s->framep[VP56_FRAME_PREVIOUS]),
51
+        .golden_ref_frame                = vaapi_vp8_surface_id(s->framep[VP56_FRAME_GOLDEN]),
52
+        .alt_ref_frame                   = vaapi_vp8_surface_id(s->framep[VP56_FRAME_GOLDEN2]),
53
+        .out_of_loop_frame               = VA_INVALID_SURFACE,
54
+
55
+        .pic_fields.bits = {
56
+            .key_frame                   = !s->keyframe,
57
+            .version                     = s->profile,
58
+
59
+            .segmentation_enabled        = s->segmentation.enabled,
60
+            .update_mb_segmentation_map  = s->segmentation.update_map,
61
+            .update_segment_feature_data = s->segmentation.update_feature_data,
62
+
63
+            .filter_type                 = s->filter.simple,
64
+            .sharpness_level             = s->filter.sharpness,
65
+
66
+            .loop_filter_adj_enable      = s->lf_delta.enabled,
67
+            .mode_ref_lf_delta_update    = s->lf_delta.update,
68
+
69
+            .sign_bias_golden            = s->sign_bias[VP56_FRAME_GOLDEN],
70
+            .sign_bias_alternate         = s->sign_bias[VP56_FRAME_GOLDEN2],
71
+
72
+            .mb_no_coeff_skip            = s->mbskip_enabled,
73
+            .loop_filter_disable         = s->filter.level == 0,
74
+        },
75
+
76
+        .prob_skip_false                 = s->prob->mbskip,
77
+        .prob_intra                      = s->prob->intra,
78
+        .prob_last                       = s->prob->last,
79
+        .prob_gf                         = s->prob->golden,
80
+    };
81
+
82
+    for (i = 0; i < 3; i++)
83
+        pp.mb_segment_tree_probs[i] = s->prob->segmentid[i];
84
+
85
+    for (i = 0; i < 4; i++) {
86
+        if (s->segmentation.enabled) {
87
+            pp.loop_filter_level[i] = s->segmentation.filter_level[i];
88
+            if (!s->segmentation.absolute_vals)
89
+                pp.loop_filter_level[i] += s->filter.level;
90
+        } else {
91
+            pp.loop_filter_level[i] = s->filter.level;
92
+        }
93
+        pp.loop_filter_level[i] = av_clip_uintp2(pp.loop_filter_level[i], 6);
94
+    }
95
+
96
+    for (i = 0; i < 4; i++) {
97
+        pp.loop_filter_deltas_ref_frame[i] = s->lf_delta.ref[i];
98
+        pp.loop_filter_deltas_mode[i] = s->lf_delta.mode[i + 4];
99
+    }
100
+
101
+    if (s->keyframe) {
102
+        static const uint8_t keyframe_y_mode_probs[4] = {
103
+            145, 156, 163, 128
104
+        };
105
+        static const uint8_t keyframe_uv_mode_probs[3] = {
106
+            142, 114, 183
107
+        };
108
+        memcpy(pp.y_mode_probs,  keyframe_y_mode_probs,  4);
109
+        memcpy(pp.uv_mode_probs, keyframe_uv_mode_probs, 3);
110
+    } else {
111
+        for (i = 0; i < 4; i++)
112
+            pp.y_mode_probs[i] = s->prob->pred16x16[i];
113
+        for (i = 0; i < 3; i++)
114
+            pp.uv_mode_probs[i] = s->prob->pred8x8c[i];
115
+    }
116
+    for (i = 0; i < 2; i++)
117
+        for (j = 0; j < 19; j++)
118
+            pp.mv_probs[i][j] = s->prob->mvc[i][j];
119
+
120
+    pp.bool_coder_ctx.range = s->coder_state_at_header_end.range;
121
+    pp.bool_coder_ctx.value = s->coder_state_at_header_end.value;
122
+    pp.bool_coder_ctx.count = s->coder_state_at_header_end.bit_count;
123
+
124
+    err = ff_vaapi_decode_make_param_buffer(avctx, pic,
125
+                                            VAPictureParameterBufferType,
126
+                                            &pp, sizeof(pp));
127
+    if (err < 0)
128
+        goto fail;
129
+
130
+    for (i = 0; i < 4; i++) {
131
+        for (j = 0; j < 8; j++) {
132
+            static const int coeff_bands_inverse[8] = {
133
+                0, 1, 2, 3, 5, 6, 4, 15
134
+            };
135
+            int coeff_pos = coeff_bands_inverse[j];
136
+
137
+            for (k = 0; k < 3; k++) {
138
+                memcpy(prob.dct_coeff_probs[i][j][k],
139
+                       s->prob->token[i][coeff_pos][k], 11);
140
+            }
141
+        }
142
+    }
143
+
144
+    err = ff_vaapi_decode_make_param_buffer(avctx, pic,
145
+                                            VAProbabilityBufferType,
146
+                                            &prob, sizeof(prob));
147
+    if (err < 0)
148
+        goto fail;
149
+
150
+    for (i = 0; i < 4; i++) {
151
+        int base_qi = s->segmentation.base_quant[i];
152
+        if (!s->segmentation.absolute_vals)
153
+            base_qi += s->quant.yac_qi;
154
+
155
+        quant.quantization_index[i][0] = av_clip_uintp2(base_qi,                       7);
156
+        quant.quantization_index[i][1] = av_clip_uintp2(base_qi + s->quant.ydc_delta,  7);
157
+        quant.quantization_index[i][2] = av_clip_uintp2(base_qi + s->quant.y2dc_delta, 7);
158
+        quant.quantization_index[i][3] = av_clip_uintp2(base_qi + s->quant.y2ac_delta, 7);
159
+        quant.quantization_index[i][4] = av_clip_uintp2(base_qi + s->quant.uvdc_delta, 7);
160
+        quant.quantization_index[i][5] = av_clip_uintp2(base_qi + s->quant.uvac_delta, 7);
161
+    }
162
+
163
+    err = ff_vaapi_decode_make_param_buffer(avctx, pic,
164
+                                            VAIQMatrixBufferType,
165
+                                            &quant, sizeof(quant));
166
+    if (err < 0)
167
+        goto fail;
168
+
169
+    return 0;
170
+
171
+fail:
172
+    ff_vaapi_decode_cancel(avctx, pic);
173
+    return err;
174
+}
175
+
176
+static int vaapi_vp8_end_frame(AVCodecContext *avctx)
177
+{
178
+    const VP8Context *s = avctx->priv_data;
179
+    VAAPIDecodePicture *pic = s->framep[VP56_FRAME_CURRENT]->hwaccel_picture_private;
180
+
181
+    return ff_vaapi_decode_issue(avctx, pic);
182
+}
183
+
184
+static int vaapi_vp8_decode_slice(AVCodecContext *avctx,
185
+                                  const uint8_t  *buffer,
186
+                                  uint32_t        size)
187
+{
188
+    const VP8Context *s = avctx->priv_data;
189
+    VAAPIDecodePicture *pic = s->framep[VP56_FRAME_CURRENT]->hwaccel_picture_private;
190
+    VASliceParameterBufferVP8 sp;
191
+    int err, i;
192
+
193
+    unsigned int header_size = 3 + 7 * s->keyframe;
194
+    const uint8_t *data = buffer + header_size;
195
+    unsigned int data_size = size - header_size;
196
+
197
+    sp = (VASliceParameterBufferVP8) {
198
+        .slice_data_size   = data_size,
199
+        .slice_data_offset = 0,
200
+        .slice_data_flag   = VA_SLICE_DATA_FLAG_ALL,
201
+
202
+        .macroblock_offset = (8 * (s->coder_state_at_header_end.input - data) -
203
+                              s->coder_state_at_header_end.bit_count - 8),
204
+        .num_of_partitions = s->num_coeff_partitions + 1,
205
+    };
206
+
207
+    sp.partition_size[0] = s->header_partition_size - ((sp.macroblock_offset + 7) / 8);
208
+    for (i = 0; i < 8; i++)
209
+        sp.partition_size[i+1] = s->coeff_partition_size[i];
210
+
211
+    err = ff_vaapi_decode_make_slice_buffer(avctx, pic, &sp, sizeof(sp), data, data_size);
212
+    if (err)
213
+        goto fail;
214
+
215
+    return 0;
216
+
217
+fail:
218
+    ff_vaapi_decode_cancel(avctx, pic);
219
+    return err;
220
+}
221
+
222
+const AVHWAccel ff_vp8_vaapi_hwaccel = {
223
+    .name                 = "vp8_vaapi",
224
+    .type                 = AVMEDIA_TYPE_VIDEO,
225
+    .id                   = AV_CODEC_ID_VP8,
226
+    .pix_fmt              = AV_PIX_FMT_VAAPI,
227
+    .start_frame          = &vaapi_vp8_start_frame,
228
+    .end_frame            = &vaapi_vp8_end_frame,
229
+    .decode_slice         = &vaapi_vp8_decode_slice,
230
+    .frame_priv_data_size = sizeof(VAAPIDecodePicture),
231
+    .init                 = &ff_vaapi_decode_init,
232
+    .uninit               = &ff_vaapi_decode_uninit,
233
+    .frame_params         = &ff_vaapi_common_frame_params,
234
+    .priv_data_size       = sizeof(VAAPIDecodeContext),
235
+    .caps_internal        = HWACCEL_CAP_ASYNC_SAFE,
236
+};
... ...
@@ -2599,6 +2599,9 @@ int vp78_decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
2599 2599
         // avctx->pix_fmt already set in caller.
2600 2600
     } else if (!is_vp7 && s->pix_fmt == AV_PIX_FMT_NONE) {
2601 2601
         enum AVPixelFormat pix_fmts[] = {
2602
+#if CONFIG_VP8_VAAPI_HWACCEL
2603
+            AV_PIX_FMT_VAAPI,
2604
+#endif
2602 2605
             AV_PIX_FMT_YUV420P,
2603 2606
             AV_PIX_FMT_NONE,
2604 2607
         };
... ...
@@ -2944,6 +2947,9 @@ AVCodec ff_vp8_decoder = {
2944 2944
     .init_thread_copy      = ONLY_IF_THREADS_ENABLED(vp8_decode_init_thread_copy),
2945 2945
     .update_thread_context = ONLY_IF_THREADS_ENABLED(vp8_decode_update_thread_context),
2946 2946
     .hw_configs            = (const AVCodecHWConfigInternal*[]) {
2947
+#if CONFIG_VP8_VAAPI_HWACCEL
2948
+                               HWACCEL_VAAPI(vp8),
2949
+#endif
2947 2950
                                NULL
2948 2951
                            },
2949 2952
 };