1 /**************************************************************************
3 * Copyright 2011 Advanced Micro Devices, Inc.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
30 * Christian König <christian.koenig@amd.com>
34 #include <sys/types.h>
40 #include "pipe/p_video_codec.h"
42 #include "util/u_memory.h"
43 #include "util/u_video.h"
45 #include "vl/vl_defines.h"
46 #include "vl/vl_mpeg12_decoder.h"
48 #include "r600_pipe_common.h"
49 #include "radeon_video.h"
50 #include "radeon_uvd.h"
54 #define NUM_MPEG2_REFS 6
55 #define NUM_H264_REFS 17
56 #define NUM_VC1_REFS 5
58 #define FB_BUFFER_OFFSET 0x1000
59 #define FB_BUFFER_SIZE 2048
60 #define FB_BUFFER_SIZE_TONGA (2048 * 64)
61 #define IT_SCALING_TABLE_SIZE 992
62 #define UVD_SESSION_CONTEXT_SIZE (128 * 1024)
64 /* UVD decoder representation */
66 struct pipe_video_codec base
;
70 unsigned stream_handle
;
72 unsigned frame_number
;
74 struct pipe_screen
*screen
;
75 struct radeon_winsys
* ws
;
76 struct radeon_winsys_cs
* cs
;
80 struct rvid_buffer msg_fb_it_buffers
[NUM_BUFFERS
];
86 struct rvid_buffer bs_buffers
[NUM_BUFFERS
];
90 struct rvid_buffer dpb
;
92 struct rvid_buffer ctx
;
93 struct rvid_buffer sessionctx
;
96 /* flush IB to the hardware */
97 static int flush(struct ruvd_decoder
*dec
, unsigned flags
)
99 return dec
->ws
->cs_flush(dec
->cs
, flags
, NULL
);
102 /* add a new set register command to the IB */
103 static void set_reg(struct ruvd_decoder
*dec
, unsigned reg
, uint32_t val
)
105 radeon_emit(dec
->cs
, RUVD_PKT0(reg
>> 2, 0));
106 radeon_emit(dec
->cs
, val
);
109 /* send a command to the VCPU through the GPCOM registers */
110 static void send_cmd(struct ruvd_decoder
*dec
, unsigned cmd
,
111 struct pb_buffer
* buf
, uint32_t off
,
112 enum radeon_bo_usage usage
, enum radeon_bo_domain domain
)
116 reloc_idx
= dec
->ws
->cs_add_buffer(dec
->cs
, buf
, usage
| RADEON_USAGE_SYNCHRONIZED
,
119 if (!dec
->use_legacy
) {
121 addr
= dec
->ws
->buffer_get_virtual_address(buf
);
123 set_reg(dec
, RUVD_GPCOM_VCPU_DATA0
, addr
);
124 set_reg(dec
, RUVD_GPCOM_VCPU_DATA1
, addr
>> 32);
126 off
+= dec
->ws
->buffer_get_reloc_offset(buf
);
127 set_reg(dec
, RUVD_GPCOM_VCPU_DATA0
, off
);
128 set_reg(dec
, RUVD_GPCOM_VCPU_DATA1
, reloc_idx
* 4);
130 set_reg(dec
, RUVD_GPCOM_VCPU_CMD
, cmd
<< 1);
133 /* do the codec needs an IT buffer ?*/
134 static bool have_it(struct ruvd_decoder
*dec
)
136 return dec
->stream_type
== RUVD_CODEC_H264_PERF
||
137 dec
->stream_type
== RUVD_CODEC_H265
;
140 /* map the next available message/feedback/itscaling buffer */
141 static void map_msg_fb_it_buf(struct ruvd_decoder
*dec
)
143 struct rvid_buffer
* buf
;
146 /* grab the current message/feedback buffer */
147 buf
= &dec
->msg_fb_it_buffers
[dec
->cur_buffer
];
149 /* and map it for CPU access */
150 ptr
= dec
->ws
->buffer_map(buf
->res
->buf
, dec
->cs
, PIPE_TRANSFER_WRITE
);
152 /* calc buffer offsets */
153 dec
->msg
= (struct ruvd_msg
*)ptr
;
154 dec
->fb
= (uint32_t *)(ptr
+ FB_BUFFER_OFFSET
);
156 dec
->it
= (uint8_t *)(ptr
+ FB_BUFFER_OFFSET
+ dec
->fb_size
);
159 /* unmap and send a message command to the VCPU */
160 static void send_msg_buf(struct ruvd_decoder
*dec
)
162 struct rvid_buffer
* buf
;
164 /* ignore the request if message/feedback buffer isn't mapped */
165 if (!dec
->msg
|| !dec
->fb
)
168 /* grab the current message buffer */
169 buf
= &dec
->msg_fb_it_buffers
[dec
->cur_buffer
];
171 /* unmap the buffer */
172 dec
->ws
->buffer_unmap(buf
->res
->buf
);
178 if (dec
->sessionctx
.res
)
179 send_cmd(dec
, RUVD_CMD_SESSION_CONTEXT_BUFFER
,
180 dec
->sessionctx
.res
->buf
, 0, RADEON_USAGE_READWRITE
,
183 /* and send it to the hardware */
184 send_cmd(dec
, RUVD_CMD_MSG_BUFFER
, buf
->res
->buf
, 0,
185 RADEON_USAGE_READ
, RADEON_DOMAIN_GTT
);
188 /* cycle to the next set of buffers */
189 static void next_buffer(struct ruvd_decoder
*dec
)
192 dec
->cur_buffer
%= NUM_BUFFERS
;
195 /* convert the profile into something UVD understands */
196 static uint32_t profile2stream_type(struct ruvd_decoder
*dec
, unsigned family
)
198 switch (u_reduce_video_profile(dec
->base
.profile
)) {
199 case PIPE_VIDEO_FORMAT_MPEG4_AVC
:
200 return (family
>= CHIP_TONGA
) ?
201 RUVD_CODEC_H264_PERF
: RUVD_CODEC_H264
;
203 case PIPE_VIDEO_FORMAT_VC1
:
204 return RUVD_CODEC_VC1
;
206 case PIPE_VIDEO_FORMAT_MPEG12
:
207 return RUVD_CODEC_MPEG2
;
209 case PIPE_VIDEO_FORMAT_MPEG4
:
210 return RUVD_CODEC_MPEG4
;
212 case PIPE_VIDEO_FORMAT_HEVC
:
213 return RUVD_CODEC_H265
;
221 static unsigned calc_ctx_size_h264_perf(struct ruvd_decoder
*dec
)
223 unsigned width_in_mb
, height_in_mb
, ctx_size
;
224 unsigned width
= align(dec
->base
.width
, VL_MACROBLOCK_WIDTH
);
225 unsigned height
= align(dec
->base
.height
, VL_MACROBLOCK_HEIGHT
);
227 unsigned max_references
= dec
->base
.max_references
+ 1;
229 // picture width & height in 16 pixel units
230 width_in_mb
= width
/ VL_MACROBLOCK_WIDTH
;
231 height_in_mb
= align(height
/ VL_MACROBLOCK_HEIGHT
, 2);
233 if (!dec
->use_legacy
) {
234 unsigned fs_in_mb
= width_in_mb
* height_in_mb
;
235 unsigned num_dpb_buffer
;
236 switch(dec
->base
.level
) {
238 num_dpb_buffer
= 8100 / fs_in_mb
;
241 num_dpb_buffer
= 18000 / fs_in_mb
;
244 num_dpb_buffer
= 20480 / fs_in_mb
;
247 num_dpb_buffer
= 32768 / fs_in_mb
;
250 num_dpb_buffer
= 34816 / fs_in_mb
;
253 num_dpb_buffer
= 110400 / fs_in_mb
;
256 num_dpb_buffer
= 184320 / fs_in_mb
;
259 num_dpb_buffer
= 184320 / fs_in_mb
;
263 max_references
= MAX2(MIN2(NUM_H264_REFS
, num_dpb_buffer
), max_references
);
264 ctx_size
= max_references
* align(width_in_mb
* height_in_mb
* 192, 256);
266 // the firmware seems to always assume a minimum of ref frames
267 max_references
= MAX2(NUM_H264_REFS
, max_references
);
268 // macroblock context buffer
269 ctx_size
= align(width_in_mb
* height_in_mb
* max_references
* 192, 256);
275 static unsigned calc_ctx_size_h265_main(struct ruvd_decoder
*dec
)
277 unsigned width
= align(dec
->base
.width
, VL_MACROBLOCK_WIDTH
);
278 unsigned height
= align(dec
->base
.height
, VL_MACROBLOCK_HEIGHT
);
280 unsigned max_references
= dec
->base
.max_references
+ 1;
282 if (dec
->base
.width
* dec
->base
.height
>= 4096*2000)
283 max_references
= MAX2(max_references
, 8);
285 max_references
= MAX2(max_references
, 17);
287 width
= align (width
, 16);
288 height
= align (height
, 16);
289 return ((width
+ 255) / 16) * ((height
+ 255) / 16) * 16 * max_references
+ 52 * 1024;
292 static unsigned calc_ctx_size_h265_main10(struct ruvd_decoder
*dec
, struct pipe_h265_picture_desc
*pic
)
294 unsigned block_size
, log2_ctb_size
, width_in_ctb
, height_in_ctb
, num_16x16_block_per_ctb
;
295 unsigned context_buffer_size_per_ctb_row
, cm_buffer_size
, max_mb_address
, db_left_tile_pxl_size
;
296 unsigned db_left_tile_ctx_size
= 4096 / 16 * (32 + 16 * 4);
298 unsigned width
= align(dec
->base
.width
, VL_MACROBLOCK_WIDTH
);
299 unsigned height
= align(dec
->base
.height
, VL_MACROBLOCK_HEIGHT
);
300 unsigned coeff_10bit
= (pic
->pps
->sps
->bit_depth_luma_minus8
|| pic
->pps
->sps
->bit_depth_chroma_minus8
) ? 2 : 1;
302 unsigned max_references
= dec
->base
.max_references
+ 1;
304 if (dec
->base
.width
* dec
->base
.height
>= 4096*2000)
305 max_references
= MAX2(max_references
, 8);
307 max_references
= MAX2(max_references
, 17);
309 block_size
= (1 << (pic
->pps
->sps
->log2_min_luma_coding_block_size_minus3
+ 3));
310 log2_ctb_size
= block_size
+ pic
->pps
->sps
->log2_diff_max_min_luma_coding_block_size
;
312 width_in_ctb
= (width
+ ((1 << log2_ctb_size
) - 1)) >> log2_ctb_size
;
313 height_in_ctb
= (height
+ ((1 << log2_ctb_size
) - 1)) >> log2_ctb_size
;
315 num_16x16_block_per_ctb
= ((1 << log2_ctb_size
) >> 4) * ((1 << log2_ctb_size
) >> 4);
316 context_buffer_size_per_ctb_row
= align(width_in_ctb
* num_16x16_block_per_ctb
* 16, 256);
317 max_mb_address
= (unsigned) ceil(height
* 8 / 2048.0);
319 cm_buffer_size
= max_references
* context_buffer_size_per_ctb_row
* height_in_ctb
;
320 db_left_tile_pxl_size
= coeff_10bit
* (max_mb_address
* 2 * 2048 + 1024);
322 return cm_buffer_size
+ db_left_tile_ctx_size
+ db_left_tile_pxl_size
;
325 /* calculate size of reference picture buffer */
326 static unsigned calc_dpb_size(struct ruvd_decoder
*dec
)
328 unsigned width_in_mb
, height_in_mb
, image_size
, dpb_size
;
330 // always align them to MB size for dpb calculation
331 unsigned width
= align(dec
->base
.width
, VL_MACROBLOCK_WIDTH
);
332 unsigned height
= align(dec
->base
.height
, VL_MACROBLOCK_HEIGHT
);
334 // always one more for currently decoded picture
335 unsigned max_references
= dec
->base
.max_references
+ 1;
337 // aligned size of a single frame
338 image_size
= width
* height
;
339 image_size
+= image_size
/ 2;
340 image_size
= align(image_size
, 1024);
342 // picture width & height in 16 pixel units
343 width_in_mb
= width
/ VL_MACROBLOCK_WIDTH
;
344 height_in_mb
= align(height
/ VL_MACROBLOCK_HEIGHT
, 2);
346 switch (u_reduce_video_profile(dec
->base
.profile
)) {
347 case PIPE_VIDEO_FORMAT_MPEG4_AVC
: {
348 if (!dec
->use_legacy
) {
349 unsigned fs_in_mb
= width_in_mb
* height_in_mb
;
350 unsigned alignment
= 64, num_dpb_buffer
;
352 if (dec
->stream_type
== RUVD_CODEC_H264_PERF
)
354 switch(dec
->base
.level
) {
356 num_dpb_buffer
= 8100 / fs_in_mb
;
359 num_dpb_buffer
= 18000 / fs_in_mb
;
362 num_dpb_buffer
= 20480 / fs_in_mb
;
365 num_dpb_buffer
= 32768 / fs_in_mb
;
368 num_dpb_buffer
= 34816 / fs_in_mb
;
371 num_dpb_buffer
= 110400 / fs_in_mb
;
374 num_dpb_buffer
= 184320 / fs_in_mb
;
377 num_dpb_buffer
= 184320 / fs_in_mb
;
381 max_references
= MAX2(MIN2(NUM_H264_REFS
, num_dpb_buffer
), max_references
);
382 dpb_size
= image_size
* max_references
;
383 if ((dec
->stream_type
!= RUVD_CODEC_H264_PERF
) ||
384 (((struct r600_common_screen
*)dec
->screen
)->family
< CHIP_POLARIS10
)) {
385 dpb_size
+= max_references
* align(width_in_mb
* height_in_mb
* 192, alignment
);
386 dpb_size
+= align(width_in_mb
* height_in_mb
* 32, alignment
);
389 // the firmware seems to allways assume a minimum of ref frames
390 max_references
= MAX2(NUM_H264_REFS
, max_references
);
391 // reference picture buffer
392 dpb_size
= image_size
* max_references
;
393 if ((dec
->stream_type
!= RUVD_CODEC_H264_PERF
) ||
394 (((struct r600_common_screen
*)dec
->screen
)->family
< CHIP_POLARIS10
)) {
395 // macroblock context buffer
396 dpb_size
+= width_in_mb
* height_in_mb
* max_references
* 192;
398 dpb_size
+= width_in_mb
* height_in_mb
* 32;
404 case PIPE_VIDEO_FORMAT_HEVC
:
405 if (dec
->base
.width
* dec
->base
.height
>= 4096*2000)
406 max_references
= MAX2(max_references
, 8);
408 max_references
= MAX2(max_references
, 17);
410 width
= align (width
, 16);
411 height
= align (height
, 16);
412 if (dec
->base
.profile
== PIPE_VIDEO_PROFILE_HEVC_MAIN_10
)
413 dpb_size
= align((width
* height
* 9) / 4, 256) * max_references
;
415 dpb_size
= align((width
* height
* 3) / 2, 256) * max_references
;
418 case PIPE_VIDEO_FORMAT_VC1
:
419 // the firmware seems to allways assume a minimum of ref frames
420 max_references
= MAX2(NUM_VC1_REFS
, max_references
);
422 // reference picture buffer
423 dpb_size
= image_size
* max_references
;
426 dpb_size
+= width_in_mb
* height_in_mb
* 128;
429 dpb_size
+= width_in_mb
* 64;
432 dpb_size
+= width_in_mb
* 128;
435 dpb_size
+= align(MAX2(width_in_mb
, height_in_mb
) * 7 * 16, 64);
438 case PIPE_VIDEO_FORMAT_MPEG12
:
439 // reference picture buffer, must be big enough for all frames
440 dpb_size
= image_size
* NUM_MPEG2_REFS
;
443 case PIPE_VIDEO_FORMAT_MPEG4
:
444 // reference picture buffer
445 dpb_size
= image_size
* max_references
;
448 dpb_size
+= width_in_mb
* height_in_mb
* 64;
451 dpb_size
+= align(width_in_mb
* height_in_mb
* 32, 64);
453 dpb_size
= MAX2(dpb_size
, 30 * 1024 * 1024);
457 // something is missing here
460 // at least use a sane default value
461 dpb_size
= 32 * 1024 * 1024;
467 /* free associated data in the video buffer callback */
468 static void ruvd_destroy_associated_data(void *data
)
470 /* NOOP, since we only use an intptr */
473 /* get h264 specific message bits */
474 static struct ruvd_h264
get_h264_msg(struct ruvd_decoder
*dec
, struct pipe_h264_picture_desc
*pic
)
476 struct ruvd_h264 result
;
478 memset(&result
, 0, sizeof(result
));
479 switch (pic
->base
.profile
) {
480 case PIPE_VIDEO_PROFILE_MPEG4_AVC_BASELINE
:
481 case PIPE_VIDEO_PROFILE_MPEG4_AVC_CONSTRAINED_BASELINE
:
482 result
.profile
= RUVD_H264_PROFILE_BASELINE
;
485 case PIPE_VIDEO_PROFILE_MPEG4_AVC_MAIN
:
486 result
.profile
= RUVD_H264_PROFILE_MAIN
;
489 case PIPE_VIDEO_PROFILE_MPEG4_AVC_HIGH
:
490 result
.profile
= RUVD_H264_PROFILE_HIGH
;
498 result
.level
= dec
->base
.level
;
500 result
.sps_info_flags
= 0;
501 result
.sps_info_flags
|= pic
->pps
->sps
->direct_8x8_inference_flag
<< 0;
502 result
.sps_info_flags
|= pic
->pps
->sps
->mb_adaptive_frame_field_flag
<< 1;
503 result
.sps_info_flags
|= pic
->pps
->sps
->frame_mbs_only_flag
<< 2;
504 result
.sps_info_flags
|= pic
->pps
->sps
->delta_pic_order_always_zero_flag
<< 3;
506 result
.bit_depth_luma_minus8
= pic
->pps
->sps
->bit_depth_luma_minus8
;
507 result
.bit_depth_chroma_minus8
= pic
->pps
->sps
->bit_depth_chroma_minus8
;
508 result
.log2_max_frame_num_minus4
= pic
->pps
->sps
->log2_max_frame_num_minus4
;
509 result
.pic_order_cnt_type
= pic
->pps
->sps
->pic_order_cnt_type
;
510 result
.log2_max_pic_order_cnt_lsb_minus4
= pic
->pps
->sps
->log2_max_pic_order_cnt_lsb_minus4
;
512 switch (dec
->base
.chroma_format
) {
513 case PIPE_VIDEO_CHROMA_FORMAT_NONE
:
516 case PIPE_VIDEO_CHROMA_FORMAT_400
:
517 result
.chroma_format
= 0;
519 case PIPE_VIDEO_CHROMA_FORMAT_420
:
520 result
.chroma_format
= 1;
522 case PIPE_VIDEO_CHROMA_FORMAT_422
:
523 result
.chroma_format
= 2;
525 case PIPE_VIDEO_CHROMA_FORMAT_444
:
526 result
.chroma_format
= 3;
530 result
.pps_info_flags
= 0;
531 result
.pps_info_flags
|= pic
->pps
->transform_8x8_mode_flag
<< 0;
532 result
.pps_info_flags
|= pic
->pps
->redundant_pic_cnt_present_flag
<< 1;
533 result
.pps_info_flags
|= pic
->pps
->constrained_intra_pred_flag
<< 2;
534 result
.pps_info_flags
|= pic
->pps
->deblocking_filter_control_present_flag
<< 3;
535 result
.pps_info_flags
|= pic
->pps
->weighted_bipred_idc
<< 4;
536 result
.pps_info_flags
|= pic
->pps
->weighted_pred_flag
<< 6;
537 result
.pps_info_flags
|= pic
->pps
->bottom_field_pic_order_in_frame_present_flag
<< 7;
538 result
.pps_info_flags
|= pic
->pps
->entropy_coding_mode_flag
<< 8;
540 result
.num_slice_groups_minus1
= pic
->pps
->num_slice_groups_minus1
;
541 result
.slice_group_map_type
= pic
->pps
->slice_group_map_type
;
542 result
.slice_group_change_rate_minus1
= pic
->pps
->slice_group_change_rate_minus1
;
543 result
.pic_init_qp_minus26
= pic
->pps
->pic_init_qp_minus26
;
544 result
.chroma_qp_index_offset
= pic
->pps
->chroma_qp_index_offset
;
545 result
.second_chroma_qp_index_offset
= pic
->pps
->second_chroma_qp_index_offset
;
547 memcpy(result
.scaling_list_4x4
, pic
->pps
->ScalingList4x4
, 6*16);
548 memcpy(result
.scaling_list_8x8
, pic
->pps
->ScalingList8x8
, 2*64);
550 if (dec
->stream_type
== RUVD_CODEC_H264_PERF
) {
551 memcpy(dec
->it
, result
.scaling_list_4x4
, 6*16);
552 memcpy((dec
->it
+ 96), result
.scaling_list_8x8
, 2*64);
555 result
.num_ref_frames
= pic
->num_ref_frames
;
557 result
.num_ref_idx_l0_active_minus1
= pic
->num_ref_idx_l0_active_minus1
;
558 result
.num_ref_idx_l1_active_minus1
= pic
->num_ref_idx_l1_active_minus1
;
560 result
.frame_num
= pic
->frame_num
;
561 memcpy(result
.frame_num_list
, pic
->frame_num_list
, 4*16);
562 result
.curr_field_order_cnt_list
[0] = pic
->field_order_cnt
[0];
563 result
.curr_field_order_cnt_list
[1] = pic
->field_order_cnt
[1];
564 memcpy(result
.field_order_cnt_list
, pic
->field_order_cnt_list
, 4*16*2);
566 result
.decoded_pic_idx
= pic
->frame_num
;
571 /* get h265 specific message bits */
572 static struct ruvd_h265
get_h265_msg(struct ruvd_decoder
*dec
, struct pipe_video_buffer
*target
,
573 struct pipe_h265_picture_desc
*pic
)
575 struct ruvd_h265 result
;
578 memset(&result
, 0, sizeof(result
));
580 result
.sps_info_flags
= 0;
581 result
.sps_info_flags
|= pic
->pps
->sps
->scaling_list_enabled_flag
<< 0;
582 result
.sps_info_flags
|= pic
->pps
->sps
->amp_enabled_flag
<< 1;
583 result
.sps_info_flags
|= pic
->pps
->sps
->sample_adaptive_offset_enabled_flag
<< 2;
584 result
.sps_info_flags
|= pic
->pps
->sps
->pcm_enabled_flag
<< 3;
585 result
.sps_info_flags
|= pic
->pps
->sps
->pcm_loop_filter_disabled_flag
<< 4;
586 result
.sps_info_flags
|= pic
->pps
->sps
->long_term_ref_pics_present_flag
<< 5;
587 result
.sps_info_flags
|= pic
->pps
->sps
->sps_temporal_mvp_enabled_flag
<< 6;
588 result
.sps_info_flags
|= pic
->pps
->sps
->strong_intra_smoothing_enabled_flag
<< 7;
589 result
.sps_info_flags
|= pic
->pps
->sps
->separate_colour_plane_flag
<< 8;
590 if (((struct r600_common_screen
*)dec
->screen
)->family
== CHIP_CARRIZO
)
591 result
.sps_info_flags
|= 1 << 9;
592 if (pic
->UseRefPicList
== true)
593 result
.sps_info_flags
|= 1 << 10;
595 result
.chroma_format
= pic
->pps
->sps
->chroma_format_idc
;
596 result
.bit_depth_luma_minus8
= pic
->pps
->sps
->bit_depth_luma_minus8
;
597 result
.bit_depth_chroma_minus8
= pic
->pps
->sps
->bit_depth_chroma_minus8
;
598 result
.log2_max_pic_order_cnt_lsb_minus4
= pic
->pps
->sps
->log2_max_pic_order_cnt_lsb_minus4
;
599 result
.sps_max_dec_pic_buffering_minus1
= pic
->pps
->sps
->sps_max_dec_pic_buffering_minus1
;
600 result
.log2_min_luma_coding_block_size_minus3
= pic
->pps
->sps
->log2_min_luma_coding_block_size_minus3
;
601 result
.log2_diff_max_min_luma_coding_block_size
= pic
->pps
->sps
->log2_diff_max_min_luma_coding_block_size
;
602 result
.log2_min_transform_block_size_minus2
= pic
->pps
->sps
->log2_min_transform_block_size_minus2
;
603 result
.log2_diff_max_min_transform_block_size
= pic
->pps
->sps
->log2_diff_max_min_transform_block_size
;
604 result
.max_transform_hierarchy_depth_inter
= pic
->pps
->sps
->max_transform_hierarchy_depth_inter
;
605 result
.max_transform_hierarchy_depth_intra
= pic
->pps
->sps
->max_transform_hierarchy_depth_intra
;
606 result
.pcm_sample_bit_depth_luma_minus1
= pic
->pps
->sps
->pcm_sample_bit_depth_luma_minus1
;
607 result
.pcm_sample_bit_depth_chroma_minus1
= pic
->pps
->sps
->pcm_sample_bit_depth_chroma_minus1
;
608 result
.log2_min_pcm_luma_coding_block_size_minus3
= pic
->pps
->sps
->log2_min_pcm_luma_coding_block_size_minus3
;
609 result
.log2_diff_max_min_pcm_luma_coding_block_size
= pic
->pps
->sps
->log2_diff_max_min_pcm_luma_coding_block_size
;
610 result
.num_short_term_ref_pic_sets
= pic
->pps
->sps
->num_short_term_ref_pic_sets
;
612 result
.pps_info_flags
= 0;
613 result
.pps_info_flags
|= pic
->pps
->dependent_slice_segments_enabled_flag
<< 0;
614 result
.pps_info_flags
|= pic
->pps
->output_flag_present_flag
<< 1;
615 result
.pps_info_flags
|= pic
->pps
->sign_data_hiding_enabled_flag
<< 2;
616 result
.pps_info_flags
|= pic
->pps
->cabac_init_present_flag
<< 3;
617 result
.pps_info_flags
|= pic
->pps
->constrained_intra_pred_flag
<< 4;
618 result
.pps_info_flags
|= pic
->pps
->transform_skip_enabled_flag
<< 5;
619 result
.pps_info_flags
|= pic
->pps
->cu_qp_delta_enabled_flag
<< 6;
620 result
.pps_info_flags
|= pic
->pps
->pps_slice_chroma_qp_offsets_present_flag
<< 7;
621 result
.pps_info_flags
|= pic
->pps
->weighted_pred_flag
<< 8;
622 result
.pps_info_flags
|= pic
->pps
->weighted_bipred_flag
<< 9;
623 result
.pps_info_flags
|= pic
->pps
->transquant_bypass_enabled_flag
<< 10;
624 result
.pps_info_flags
|= pic
->pps
->tiles_enabled_flag
<< 11;
625 result
.pps_info_flags
|= pic
->pps
->entropy_coding_sync_enabled_flag
<< 12;
626 result
.pps_info_flags
|= pic
->pps
->uniform_spacing_flag
<< 13;
627 result
.pps_info_flags
|= pic
->pps
->loop_filter_across_tiles_enabled_flag
<< 14;
628 result
.pps_info_flags
|= pic
->pps
->pps_loop_filter_across_slices_enabled_flag
<< 15;
629 result
.pps_info_flags
|= pic
->pps
->deblocking_filter_override_enabled_flag
<< 16;
630 result
.pps_info_flags
|= pic
->pps
->pps_deblocking_filter_disabled_flag
<< 17;
631 result
.pps_info_flags
|= pic
->pps
->lists_modification_present_flag
<< 18;
632 result
.pps_info_flags
|= pic
->pps
->slice_segment_header_extension_present_flag
<< 19;
633 //result.pps_info_flags |= pic->pps->deblocking_filter_control_present_flag; ???
635 result
.num_extra_slice_header_bits
= pic
->pps
->num_extra_slice_header_bits
;
636 result
.num_long_term_ref_pic_sps
= pic
->pps
->sps
->num_long_term_ref_pics_sps
;
637 result
.num_ref_idx_l0_default_active_minus1
= pic
->pps
->num_ref_idx_l0_default_active_minus1
;
638 result
.num_ref_idx_l1_default_active_minus1
= pic
->pps
->num_ref_idx_l1_default_active_minus1
;
639 result
.pps_cb_qp_offset
= pic
->pps
->pps_cb_qp_offset
;
640 result
.pps_cr_qp_offset
= pic
->pps
->pps_cr_qp_offset
;
641 result
.pps_beta_offset_div2
= pic
->pps
->pps_beta_offset_div2
;
642 result
.pps_tc_offset_div2
= pic
->pps
->pps_tc_offset_div2
;
643 result
.diff_cu_qp_delta_depth
= pic
->pps
->diff_cu_qp_delta_depth
;
644 result
.num_tile_columns_minus1
= pic
->pps
->num_tile_columns_minus1
;
645 result
.num_tile_rows_minus1
= pic
->pps
->num_tile_rows_minus1
;
646 result
.log2_parallel_merge_level_minus2
= pic
->pps
->log2_parallel_merge_level_minus2
;
647 result
.init_qp_minus26
= pic
->pps
->init_qp_minus26
;
649 for (i
= 0; i
< 19; ++i
)
650 result
.column_width_minus1
[i
] = pic
->pps
->column_width_minus1
[i
];
652 for (i
= 0; i
< 21; ++i
)
653 result
.row_height_minus1
[i
] = pic
->pps
->row_height_minus1
[i
];
655 result
.num_delta_pocs_ref_rps_idx
= pic
->NumDeltaPocsOfRefRpsIdx
;
656 result
.curr_idx
= pic
->CurrPicOrderCntVal
;
657 result
.curr_poc
= pic
->CurrPicOrderCntVal
;
659 vl_video_buffer_set_associated_data(target
, &dec
->base
,
660 (void *)(uintptr_t)pic
->CurrPicOrderCntVal
,
661 &ruvd_destroy_associated_data
);
663 for (i
= 0; i
< 16; ++i
) {
664 struct pipe_video_buffer
*ref
= pic
->ref
[i
];
665 uintptr_t ref_pic
= 0;
667 result
.poc_list
[i
] = pic
->PicOrderCntVal
[i
];
670 ref_pic
= (uintptr_t)vl_video_buffer_get_associated_data(ref
, &dec
->base
);
673 result
.ref_pic_list
[i
] = ref_pic
;
676 for (i
= 0; i
< 8; ++i
) {
677 result
.ref_pic_set_st_curr_before
[i
] = 0xFF;
678 result
.ref_pic_set_st_curr_after
[i
] = 0xFF;
679 result
.ref_pic_set_lt_curr
[i
] = 0xFF;
682 for (i
= 0; i
< pic
->NumPocStCurrBefore
; ++i
)
683 result
.ref_pic_set_st_curr_before
[i
] = pic
->RefPicSetStCurrBefore
[i
];
685 for (i
= 0; i
< pic
->NumPocStCurrAfter
; ++i
)
686 result
.ref_pic_set_st_curr_after
[i
] = pic
->RefPicSetStCurrAfter
[i
];
688 for (i
= 0; i
< pic
->NumPocLtCurr
; ++i
)
689 result
.ref_pic_set_lt_curr
[i
] = pic
->RefPicSetLtCurr
[i
];
691 for (i
= 0; i
< 6; ++i
)
692 result
.ucScalingListDCCoefSizeID2
[i
] = pic
->pps
->sps
->ScalingListDCCoeff16x16
[i
];
694 for (i
= 0; i
< 2; ++i
)
695 result
.ucScalingListDCCoefSizeID3
[i
] = pic
->pps
->sps
->ScalingListDCCoeff32x32
[i
];
697 memcpy(dec
->it
, pic
->pps
->sps
->ScalingList4x4
, 6 * 16);
698 memcpy(dec
->it
+ 96, pic
->pps
->sps
->ScalingList8x8
, 6 * 64);
699 memcpy(dec
->it
+ 480, pic
->pps
->sps
->ScalingList16x16
, 6 * 64);
700 memcpy(dec
->it
+ 864, pic
->pps
->sps
->ScalingList32x32
, 2 * 64);
702 for (i
= 0 ; i
< 2 ; i
++) {
703 for (int j
= 0 ; j
< 15 ; j
++)
704 result
.direct_reflist
[i
][j
] = pic
->RefPicList
[i
][j
];
707 if (pic
->base
.profile
== PIPE_VIDEO_PROFILE_HEVC_MAIN_10
) {
708 if (target
->buffer_format
== PIPE_FORMAT_P016
) {
709 result
.p010_mode
= 1;
712 result
.luma_10to8
= 5;
713 result
.chroma_10to8
= 5;
714 result
.sclr_luma10to8
= 4;
715 result
.sclr_chroma10to8
= 4;
726 NumShortTermPictureSliceHeaderBits;
727 NumLongTermPictureSliceHeaderBits;
735 /* get vc1 specific message bits */
736 static struct ruvd_vc1
get_vc1_msg(struct pipe_vc1_picture_desc
*pic
)
738 struct ruvd_vc1 result
;
740 memset(&result
, 0, sizeof(result
));
742 switch(pic
->base
.profile
) {
743 case PIPE_VIDEO_PROFILE_VC1_SIMPLE
:
744 result
.profile
= RUVD_VC1_PROFILE_SIMPLE
;
748 case PIPE_VIDEO_PROFILE_VC1_MAIN
:
749 result
.profile
= RUVD_VC1_PROFILE_MAIN
;
753 case PIPE_VIDEO_PROFILE_VC1_ADVANCED
:
754 result
.profile
= RUVD_VC1_PROFILE_ADVANCED
;
762 /* fields common for all profiles */
763 result
.sps_info_flags
|= pic
->postprocflag
<< 7;
764 result
.sps_info_flags
|= pic
->pulldown
<< 6;
765 result
.sps_info_flags
|= pic
->interlace
<< 5;
766 result
.sps_info_flags
|= pic
->tfcntrflag
<< 4;
767 result
.sps_info_flags
|= pic
->finterpflag
<< 3;
768 result
.sps_info_flags
|= pic
->psf
<< 1;
770 result
.pps_info_flags
|= pic
->range_mapy_flag
<< 31;
771 result
.pps_info_flags
|= pic
->range_mapy
<< 28;
772 result
.pps_info_flags
|= pic
->range_mapuv_flag
<< 27;
773 result
.pps_info_flags
|= pic
->range_mapuv
<< 24;
774 result
.pps_info_flags
|= pic
->multires
<< 21;
775 result
.pps_info_flags
|= pic
->maxbframes
<< 16;
776 result
.pps_info_flags
|= pic
->overlap
<< 11;
777 result
.pps_info_flags
|= pic
->quantizer
<< 9;
778 result
.pps_info_flags
|= pic
->panscan_flag
<< 7;
779 result
.pps_info_flags
|= pic
->refdist_flag
<< 6;
780 result
.pps_info_flags
|= pic
->vstransform
<< 0;
782 /* some fields only apply to main/advanced profile */
783 if (pic
->base
.profile
!= PIPE_VIDEO_PROFILE_VC1_SIMPLE
) {
784 result
.pps_info_flags
|= pic
->syncmarker
<< 20;
785 result
.pps_info_flags
|= pic
->rangered
<< 19;
786 result
.pps_info_flags
|= pic
->loopfilter
<< 5;
787 result
.pps_info_flags
|= pic
->fastuvmc
<< 4;
788 result
.pps_info_flags
|= pic
->extended_mv
<< 3;
789 result
.pps_info_flags
|= pic
->extended_dmv
<< 8;
790 result
.pps_info_flags
|= pic
->dquant
<< 1;
793 result
.chroma_format
= 1;
796 //(((unsigned int)(pPicParams->advance.reserved1)) << SPS_INFO_VC1_RESERVED_SHIFT)
799 uint8_t frame_coding_mode
800 uint8_t deblockEnable
807 /* extract the frame number from a referenced video buffer */
808 static uint32_t get_ref_pic_idx(struct ruvd_decoder
*dec
, struct pipe_video_buffer
*ref
)
810 uint32_t min
= MAX2(dec
->frame_number
, NUM_MPEG2_REFS
) - NUM_MPEG2_REFS
;
811 uint32_t max
= MAX2(dec
->frame_number
, 1) - 1;
814 /* seems to be the most sane fallback */
818 /* get the frame number from the associated data */
819 frame
= (uintptr_t)vl_video_buffer_get_associated_data(ref
, &dec
->base
);
821 /* limit the frame number to a valid range */
822 return MAX2(MIN2(frame
, max
), min
);
825 /* get mpeg2 specific msg bits */
826 static struct ruvd_mpeg2
get_mpeg2_msg(struct ruvd_decoder
*dec
,
827 struct pipe_mpeg12_picture_desc
*pic
)
829 const int *zscan
= pic
->alternate_scan
? vl_zscan_alternate
: vl_zscan_normal
;
830 struct ruvd_mpeg2 result
;
833 memset(&result
, 0, sizeof(result
));
834 result
.decoded_pic_idx
= dec
->frame_number
;
835 for (i
= 0; i
< 2; ++i
)
836 result
.ref_pic_idx
[i
] = get_ref_pic_idx(dec
, pic
->ref
[i
]);
838 result
.load_intra_quantiser_matrix
= 1;
839 result
.load_nonintra_quantiser_matrix
= 1;
841 for (i
= 0; i
< 64; ++i
) {
842 result
.intra_quantiser_matrix
[i
] = pic
->intra_matrix
[zscan
[i
]];
843 result
.nonintra_quantiser_matrix
[i
] = pic
->non_intra_matrix
[zscan
[i
]];
846 result
.profile_and_level_indication
= 0;
847 result
.chroma_format
= 0x1;
849 result
.picture_coding_type
= pic
->picture_coding_type
;
850 result
.f_code
[0][0] = pic
->f_code
[0][0] + 1;
851 result
.f_code
[0][1] = pic
->f_code
[0][1] + 1;
852 result
.f_code
[1][0] = pic
->f_code
[1][0] + 1;
853 result
.f_code
[1][1] = pic
->f_code
[1][1] + 1;
854 result
.intra_dc_precision
= pic
->intra_dc_precision
;
855 result
.pic_structure
= pic
->picture_structure
;
856 result
.top_field_first
= pic
->top_field_first
;
857 result
.frame_pred_frame_dct
= pic
->frame_pred_frame_dct
;
858 result
.concealment_motion_vectors
= pic
->concealment_motion_vectors
;
859 result
.q_scale_type
= pic
->q_scale_type
;
860 result
.intra_vlc_format
= pic
->intra_vlc_format
;
861 result
.alternate_scan
= pic
->alternate_scan
;
866 /* get mpeg4 specific msg bits */
867 static struct ruvd_mpeg4
get_mpeg4_msg(struct ruvd_decoder
*dec
,
868 struct pipe_mpeg4_picture_desc
*pic
)
870 struct ruvd_mpeg4 result
;
873 memset(&result
, 0, sizeof(result
));
874 result
.decoded_pic_idx
= dec
->frame_number
;
875 for (i
= 0; i
< 2; ++i
)
876 result
.ref_pic_idx
[i
] = get_ref_pic_idx(dec
, pic
->ref
[i
]);
878 result
.variant_type
= 0;
879 result
.profile_and_level_indication
= 0xF0; // ASP Level0
881 result
.video_object_layer_verid
= 0x5; // advanced simple
882 result
.video_object_layer_shape
= 0x0; // rectangular
884 result
.video_object_layer_width
= dec
->base
.width
;
885 result
.video_object_layer_height
= dec
->base
.height
;
887 result
.vop_time_increment_resolution
= pic
->vop_time_increment_resolution
;
889 result
.flags
|= pic
->short_video_header
<< 0;
890 //result.flags |= obmc_disable << 1;
891 result
.flags
|= pic
->interlaced
<< 2;
892 result
.flags
|= 1 << 3; // load_intra_quant_mat
893 result
.flags
|= 1 << 4; // load_nonintra_quant_mat
894 result
.flags
|= pic
->quarter_sample
<< 5;
895 result
.flags
|= 1 << 6; // complexity_estimation_disable
896 result
.flags
|= pic
->resync_marker_disable
<< 7;
897 //result.flags |= data_partitioned << 8;
898 //result.flags |= reversible_vlc << 9;
899 result
.flags
|= 0 << 10; // newpred_enable
900 result
.flags
|= 0 << 11; // reduced_resolution_vop_enable
901 //result.flags |= scalability << 12;
902 //result.flags |= is_object_layer_identifier << 13;
903 //result.flags |= fixed_vop_rate << 14;
904 //result.flags |= newpred_segment_type << 15;
906 result
.quant_type
= pic
->quant_type
;
908 for (i
= 0; i
< 64; ++i
) {
909 result
.intra_quant_mat
[i
] = pic
->intra_matrix
[vl_zscan_normal
[i
]];
910 result
.nonintra_quant_mat
[i
] = pic
->non_intra_matrix
[vl_zscan_normal
[i
]];
916 uint8_t vop_coding_type
917 uint8_t vop_fcode_forward
918 uint8_t vop_fcode_backward
919 uint8_t rounding_control
920 uint8_t alternate_vertical_scan_flag
921 uint8_t top_field_first
928 * destroy this video decoder
930 static void ruvd_destroy(struct pipe_video_codec
*decoder
)
932 struct ruvd_decoder
*dec
= (struct ruvd_decoder
*)decoder
;
937 map_msg_fb_it_buf(dec
);
938 memset(dec
->msg
, 0, sizeof(*dec
->msg
));
939 dec
->msg
->size
= sizeof(*dec
->msg
);
940 dec
->msg
->msg_type
= RUVD_MSG_DESTROY
;
941 dec
->msg
->stream_handle
= dec
->stream_handle
;
946 dec
->ws
->cs_destroy(dec
->cs
);
948 for (i
= 0; i
< NUM_BUFFERS
; ++i
) {
949 rvid_destroy_buffer(&dec
->msg_fb_it_buffers
[i
]);
950 rvid_destroy_buffer(&dec
->bs_buffers
[i
]);
953 rvid_destroy_buffer(&dec
->dpb
);
954 rvid_destroy_buffer(&dec
->ctx
);
955 rvid_destroy_buffer(&dec
->sessionctx
);
961 * start decoding of a new frame
963 static void ruvd_begin_frame(struct pipe_video_codec
*decoder
,
964 struct pipe_video_buffer
*target
,
965 struct pipe_picture_desc
*picture
)
967 struct ruvd_decoder
*dec
= (struct ruvd_decoder
*)decoder
;
972 frame
= ++dec
->frame_number
;
973 vl_video_buffer_set_associated_data(target
, decoder
, (void *)frame
,
974 &ruvd_destroy_associated_data
);
977 dec
->bs_ptr
= dec
->ws
->buffer_map(
978 dec
->bs_buffers
[dec
->cur_buffer
].res
->buf
,
979 dec
->cs
, PIPE_TRANSFER_WRITE
);
983 * decode a macroblock
985 static void ruvd_decode_macroblock(struct pipe_video_codec
*decoder
,
986 struct pipe_video_buffer
*target
,
987 struct pipe_picture_desc
*picture
,
988 const struct pipe_macroblock
*macroblocks
,
989 unsigned num_macroblocks
)
991 /* not supported (yet) */
998 static void ruvd_decode_bitstream(struct pipe_video_codec
*decoder
,
999 struct pipe_video_buffer
*target
,
1000 struct pipe_picture_desc
*picture
,
1001 unsigned num_buffers
,
1002 const void * const *buffers
,
1003 const unsigned *sizes
)
1005 struct ruvd_decoder
*dec
= (struct ruvd_decoder
*)decoder
;
1013 for (i
= 0; i
< num_buffers
; ++i
) {
1014 struct rvid_buffer
*buf
= &dec
->bs_buffers
[dec
->cur_buffer
];
1015 unsigned new_size
= dec
->bs_size
+ sizes
[i
];
1017 if (new_size
> buf
->res
->buf
->size
) {
1018 dec
->ws
->buffer_unmap(buf
->res
->buf
);
1019 if (!rvid_resize_buffer(dec
->screen
, dec
->cs
, buf
, new_size
)) {
1020 RVID_ERR("Can't resize bitstream buffer!");
1024 dec
->bs_ptr
= dec
->ws
->buffer_map(buf
->res
->buf
, dec
->cs
,
1025 PIPE_TRANSFER_WRITE
);
1029 dec
->bs_ptr
+= dec
->bs_size
;
1032 memcpy(dec
->bs_ptr
, buffers
[i
], sizes
[i
]);
1033 dec
->bs_size
+= sizes
[i
];
1034 dec
->bs_ptr
+= sizes
[i
];
1039 * end decoding of the current frame
1041 static void ruvd_end_frame(struct pipe_video_codec
*decoder
,
1042 struct pipe_video_buffer
*target
,
1043 struct pipe_picture_desc
*picture
)
1045 struct ruvd_decoder
*dec
= (struct ruvd_decoder
*)decoder
;
1046 struct pb_buffer
*dt
;
1047 struct rvid_buffer
*msg_fb_it_buf
, *bs_buf
;
1055 msg_fb_it_buf
= &dec
->msg_fb_it_buffers
[dec
->cur_buffer
];
1056 bs_buf
= &dec
->bs_buffers
[dec
->cur_buffer
];
1058 bs_size
= align(dec
->bs_size
, 128);
1059 memset(dec
->bs_ptr
, 0, bs_size
- dec
->bs_size
);
1060 dec
->ws
->buffer_unmap(bs_buf
->res
->buf
);
1062 map_msg_fb_it_buf(dec
);
1063 dec
->msg
->size
= sizeof(*dec
->msg
);
1064 dec
->msg
->msg_type
= RUVD_MSG_DECODE
;
1065 dec
->msg
->stream_handle
= dec
->stream_handle
;
1066 dec
->msg
->status_report_feedback_number
= dec
->frame_number
;
1068 dec
->msg
->body
.decode
.stream_type
= dec
->stream_type
;
1069 dec
->msg
->body
.decode
.decode_flags
= 0x1;
1070 dec
->msg
->body
.decode
.width_in_samples
= dec
->base
.width
;
1071 dec
->msg
->body
.decode
.height_in_samples
= dec
->base
.height
;
1073 if ((picture
->profile
== PIPE_VIDEO_PROFILE_VC1_SIMPLE
) ||
1074 (picture
->profile
== PIPE_VIDEO_PROFILE_VC1_MAIN
)) {
1075 dec
->msg
->body
.decode
.width_in_samples
= align(dec
->msg
->body
.decode
.width_in_samples
, 16) / 16;
1076 dec
->msg
->body
.decode
.height_in_samples
= align(dec
->msg
->body
.decode
.height_in_samples
, 16) / 16;
1079 dec
->msg
->body
.decode
.dpb_size
= dec
->dpb
.res
->buf
->size
;
1080 dec
->msg
->body
.decode
.bsd_size
= bs_size
;
1081 dec
->msg
->body
.decode
.db_pitch
= align(dec
->base
.width
, 16);
1083 if (dec
->stream_type
== RUVD_CODEC_H264_PERF
&&
1084 ((struct r600_common_screen
*)dec
->screen
)->family
>= CHIP_POLARIS10
)
1085 dec
->msg
->body
.decode
.dpb_reserved
= dec
->ctx
.res
->buf
->size
;
1087 dt
= dec
->set_dtb(dec
->msg
, (struct vl_video_buffer
*)target
);
1088 if (((struct r600_common_screen
*)dec
->screen
)->family
>= CHIP_STONEY
)
1089 dec
->msg
->body
.decode
.dt_wa_chroma_top_offset
= dec
->msg
->body
.decode
.dt_pitch
/ 2;
1091 switch (u_reduce_video_profile(picture
->profile
)) {
1092 case PIPE_VIDEO_FORMAT_MPEG4_AVC
:
1093 dec
->msg
->body
.decode
.codec
.h264
= get_h264_msg(dec
, (struct pipe_h264_picture_desc
*)picture
);
1096 case PIPE_VIDEO_FORMAT_HEVC
:
1097 dec
->msg
->body
.decode
.codec
.h265
= get_h265_msg(dec
, target
, (struct pipe_h265_picture_desc
*)picture
);
1098 if (dec
->ctx
.res
== NULL
) {
1100 if (dec
->base
.profile
== PIPE_VIDEO_PROFILE_HEVC_MAIN_10
)
1101 ctx_size
= calc_ctx_size_h265_main10(dec
, (struct pipe_h265_picture_desc
*)picture
);
1103 ctx_size
= calc_ctx_size_h265_main(dec
);
1104 if (!rvid_create_buffer(dec
->screen
, &dec
->ctx
, ctx_size
, PIPE_USAGE_DEFAULT
)) {
1105 RVID_ERR("Can't allocated context buffer.\n");
1107 rvid_clear_buffer(decoder
->context
, &dec
->ctx
);
1111 dec
->msg
->body
.decode
.dpb_reserved
= dec
->ctx
.res
->buf
->size
;
1114 case PIPE_VIDEO_FORMAT_VC1
:
1115 dec
->msg
->body
.decode
.codec
.vc1
= get_vc1_msg((struct pipe_vc1_picture_desc
*)picture
);
1118 case PIPE_VIDEO_FORMAT_MPEG12
:
1119 dec
->msg
->body
.decode
.codec
.mpeg2
= get_mpeg2_msg(dec
, (struct pipe_mpeg12_picture_desc
*)picture
);
1122 case PIPE_VIDEO_FORMAT_MPEG4
:
1123 dec
->msg
->body
.decode
.codec
.mpeg4
= get_mpeg4_msg(dec
, (struct pipe_mpeg4_picture_desc
*)picture
);
1131 dec
->msg
->body
.decode
.db_surf_tile_config
= dec
->msg
->body
.decode
.dt_surf_tile_config
;
1132 dec
->msg
->body
.decode
.extension_support
= 0x1;
1134 /* set at least the feedback buffer size */
1135 dec
->fb
[0] = dec
->fb_size
;
1139 send_cmd(dec
, RUVD_CMD_DPB_BUFFER
, dec
->dpb
.res
->buf
, 0,
1140 RADEON_USAGE_READWRITE
, RADEON_DOMAIN_VRAM
);
1142 send_cmd(dec
, RUVD_CMD_CONTEXT_BUFFER
, dec
->ctx
.res
->buf
, 0,
1143 RADEON_USAGE_READWRITE
, RADEON_DOMAIN_VRAM
);
1144 send_cmd(dec
, RUVD_CMD_BITSTREAM_BUFFER
, bs_buf
->res
->buf
,
1145 0, RADEON_USAGE_READ
, RADEON_DOMAIN_GTT
);
1146 send_cmd(dec
, RUVD_CMD_DECODING_TARGET_BUFFER
, dt
, 0,
1147 RADEON_USAGE_WRITE
, RADEON_DOMAIN_VRAM
);
1148 send_cmd(dec
, RUVD_CMD_FEEDBACK_BUFFER
, msg_fb_it_buf
->res
->buf
,
1149 FB_BUFFER_OFFSET
, RADEON_USAGE_WRITE
, RADEON_DOMAIN_GTT
);
1151 send_cmd(dec
, RUVD_CMD_ITSCALING_TABLE_BUFFER
, msg_fb_it_buf
->res
->buf
,
1152 FB_BUFFER_OFFSET
+ dec
->fb_size
, RADEON_USAGE_READ
, RADEON_DOMAIN_GTT
);
1153 set_reg(dec
, RUVD_ENGINE_CNTL
, 1);
1155 flush(dec
, RADEON_FLUSH_ASYNC
);
1160 * flush any outstanding command buffers to the hardware
1162 static void ruvd_flush(struct pipe_video_codec
*decoder
)
1167 * create and UVD decoder
1169 struct pipe_video_codec
*ruvd_create_decoder(struct pipe_context
*context
,
1170 const struct pipe_video_codec
*templ
,
1171 ruvd_set_dtb set_dtb
)
1173 struct radeon_winsys
* ws
= ((struct r600_common_context
*)context
)->ws
;
1174 struct r600_common_context
*rctx
= (struct r600_common_context
*)context
;
1176 unsigned width
= templ
->width
, height
= templ
->height
;
1177 unsigned bs_buf_size
;
1178 struct radeon_info info
;
1179 struct ruvd_decoder
*dec
;
1182 ws
->query_info(ws
, &info
);
1184 switch(u_reduce_video_profile(templ
->profile
)) {
1185 case PIPE_VIDEO_FORMAT_MPEG12
:
1186 if (templ
->entrypoint
> PIPE_VIDEO_ENTRYPOINT_BITSTREAM
|| info
.family
< CHIP_PALM
)
1187 return vl_create_mpeg12_decoder(context
, templ
);
1190 case PIPE_VIDEO_FORMAT_MPEG4
:
1191 width
= align(width
, VL_MACROBLOCK_WIDTH
);
1192 height
= align(height
, VL_MACROBLOCK_HEIGHT
);
1194 case PIPE_VIDEO_FORMAT_MPEG4_AVC
:
1195 width
= align(width
, VL_MACROBLOCK_WIDTH
);
1196 height
= align(height
, VL_MACROBLOCK_HEIGHT
);
1204 dec
= CALLOC_STRUCT(ruvd_decoder
);
1209 if (info
.drm_major
< 3)
1210 dec
->use_legacy
= true;
1213 dec
->base
.context
= context
;
1214 dec
->base
.width
= width
;
1215 dec
->base
.height
= height
;
1217 dec
->base
.destroy
= ruvd_destroy
;
1218 dec
->base
.begin_frame
= ruvd_begin_frame
;
1219 dec
->base
.decode_macroblock
= ruvd_decode_macroblock
;
1220 dec
->base
.decode_bitstream
= ruvd_decode_bitstream
;
1221 dec
->base
.end_frame
= ruvd_end_frame
;
1222 dec
->base
.flush
= ruvd_flush
;
1224 dec
->stream_type
= profile2stream_type(dec
, info
.family
);
1225 dec
->set_dtb
= set_dtb
;
1226 dec
->stream_handle
= rvid_alloc_stream_handle();
1227 dec
->screen
= context
->screen
;
1229 dec
->cs
= ws
->cs_create(rctx
->ctx
, RING_UVD
, NULL
, NULL
);
1231 RVID_ERR("Can't get command submission context.\n");
1235 dec
->fb_size
= (info
.family
== CHIP_TONGA
) ? FB_BUFFER_SIZE_TONGA
:
1237 bs_buf_size
= width
* height
* (512 / (16 * 16));
1238 for (i
= 0; i
< NUM_BUFFERS
; ++i
) {
1239 unsigned msg_fb_it_size
= FB_BUFFER_OFFSET
+ dec
->fb_size
;
1240 STATIC_ASSERT(sizeof(struct ruvd_msg
) <= FB_BUFFER_OFFSET
);
1242 msg_fb_it_size
+= IT_SCALING_TABLE_SIZE
;
1243 if (!rvid_create_buffer(dec
->screen
, &dec
->msg_fb_it_buffers
[i
],
1244 msg_fb_it_size
, PIPE_USAGE_STAGING
)) {
1245 RVID_ERR("Can't allocated message buffers.\n");
1249 if (!rvid_create_buffer(dec
->screen
, &dec
->bs_buffers
[i
],
1250 bs_buf_size
, PIPE_USAGE_STAGING
)) {
1251 RVID_ERR("Can't allocated bitstream buffers.\n");
1255 rvid_clear_buffer(context
, &dec
->msg_fb_it_buffers
[i
]);
1256 rvid_clear_buffer(context
, &dec
->bs_buffers
[i
]);
1259 dpb_size
= calc_dpb_size(dec
);
1261 if (!rvid_create_buffer(dec
->screen
, &dec
->dpb
, dpb_size
, PIPE_USAGE_DEFAULT
)) {
1262 RVID_ERR("Can't allocated dpb.\n");
1266 rvid_clear_buffer(context
, &dec
->dpb
);
1268 if (dec
->stream_type
== RUVD_CODEC_H264_PERF
&& info
.family
>= CHIP_POLARIS10
) {
1269 unsigned ctx_size
= calc_ctx_size_h264_perf(dec
);
1270 if (!rvid_create_buffer(dec
->screen
, &dec
->ctx
, ctx_size
, PIPE_USAGE_DEFAULT
)) {
1271 RVID_ERR("Can't allocated context buffer.\n");
1274 rvid_clear_buffer(context
, &dec
->ctx
);
1277 if (info
.family
>= CHIP_POLARIS10
&& info
.drm_minor
>= 3) {
1278 if (!rvid_create_buffer(dec
->screen
, &dec
->sessionctx
,
1279 UVD_SESSION_CONTEXT_SIZE
,
1280 PIPE_USAGE_DEFAULT
)) {
1281 RVID_ERR("Can't allocated session ctx.\n");
1284 rvid_clear_buffer(context
, &dec
->sessionctx
);
1287 map_msg_fb_it_buf(dec
);
1288 dec
->msg
->size
= sizeof(*dec
->msg
);
1289 dec
->msg
->msg_type
= RUVD_MSG_CREATE
;
1290 dec
->msg
->stream_handle
= dec
->stream_handle
;
1291 dec
->msg
->body
.create
.stream_type
= dec
->stream_type
;
1292 dec
->msg
->body
.create
.width_in_samples
= dec
->base
.width
;
1293 dec
->msg
->body
.create
.height_in_samples
= dec
->base
.height
;
1294 dec
->msg
->body
.create
.dpb_size
= dpb_size
;
1305 if (dec
->cs
) dec
->ws
->cs_destroy(dec
->cs
);
1307 for (i
= 0; i
< NUM_BUFFERS
; ++i
) {
1308 rvid_destroy_buffer(&dec
->msg_fb_it_buffers
[i
]);
1309 rvid_destroy_buffer(&dec
->bs_buffers
[i
]);
1312 rvid_destroy_buffer(&dec
->dpb
);
1313 rvid_destroy_buffer(&dec
->ctx
);
1314 rvid_destroy_buffer(&dec
->sessionctx
);
1321 /* calculate top/bottom offset */
1322 static unsigned texture_offset(struct radeon_surf
*surface
, unsigned layer
)
1324 return surface
->u
.legacy
.level
[0].offset
+
1325 layer
* surface
->u
.legacy
.level
[0].slice_size
;
1328 /* hw encode the aspect of macro tiles */
1329 static unsigned macro_tile_aspect(unsigned macro_tile_aspect
)
1331 switch (macro_tile_aspect
) {
1333 case 1: macro_tile_aspect
= 0; break;
1334 case 2: macro_tile_aspect
= 1; break;
1335 case 4: macro_tile_aspect
= 2; break;
1336 case 8: macro_tile_aspect
= 3; break;
1338 return macro_tile_aspect
;
1341 /* hw encode the bank width and height */
1342 static unsigned bank_wh(unsigned bankwh
)
1346 case 1: bankwh
= 0; break;
1347 case 2: bankwh
= 1; break;
1348 case 4: bankwh
= 2; break;
1349 case 8: bankwh
= 3; break;
1355 * fill decoding target field from the luma and chroma surfaces
1357 void ruvd_set_dt_surfaces(struct ruvd_msg
*msg
, struct radeon_surf
*luma
,
1358 struct radeon_surf
*chroma
)
1360 msg
->body
.decode
.dt_pitch
= luma
->u
.legacy
.level
[0].nblk_x
;
1361 switch (luma
->u
.legacy
.level
[0].mode
) {
1362 case RADEON_SURF_MODE_LINEAR_ALIGNED
:
1363 msg
->body
.decode
.dt_tiling_mode
= RUVD_TILE_LINEAR
;
1364 msg
->body
.decode
.dt_array_mode
= RUVD_ARRAY_MODE_LINEAR
;
1366 case RADEON_SURF_MODE_1D
:
1367 msg
->body
.decode
.dt_tiling_mode
= RUVD_TILE_8X8
;
1368 msg
->body
.decode
.dt_array_mode
= RUVD_ARRAY_MODE_1D_THIN
;
1370 case RADEON_SURF_MODE_2D
:
1371 msg
->body
.decode
.dt_tiling_mode
= RUVD_TILE_8X8
;
1372 msg
->body
.decode
.dt_array_mode
= RUVD_ARRAY_MODE_2D_THIN
;
1379 msg
->body
.decode
.dt_luma_top_offset
= texture_offset(luma
, 0);
1380 msg
->body
.decode
.dt_chroma_top_offset
= texture_offset(chroma
, 0);
1381 if (msg
->body
.decode
.dt_field_mode
) {
1382 msg
->body
.decode
.dt_luma_bottom_offset
= texture_offset(luma
, 1);
1383 msg
->body
.decode
.dt_chroma_bottom_offset
= texture_offset(chroma
, 1);
1385 msg
->body
.decode
.dt_luma_bottom_offset
= msg
->body
.decode
.dt_luma_top_offset
;
1386 msg
->body
.decode
.dt_chroma_bottom_offset
= msg
->body
.decode
.dt_chroma_top_offset
;
1389 assert(luma
->u
.legacy
.bankw
== chroma
->u
.legacy
.bankw
);
1390 assert(luma
->u
.legacy
.bankh
== chroma
->u
.legacy
.bankh
);
1391 assert(luma
->u
.legacy
.mtilea
== chroma
->u
.legacy
.mtilea
);
1393 msg
->body
.decode
.dt_surf_tile_config
|= RUVD_BANK_WIDTH(bank_wh(luma
->u
.legacy
.bankw
));
1394 msg
->body
.decode
.dt_surf_tile_config
|= RUVD_BANK_HEIGHT(bank_wh(luma
->u
.legacy
.bankh
));
1395 msg
->body
.decode
.dt_surf_tile_config
|= RUVD_MACRO_TILE_ASPECT_RATIO(macro_tile_aspect(luma
->u
.legacy
.mtilea
));