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
63 /* UVD decoder representation */
65 struct pipe_video_codec base
;
69 unsigned stream_handle
;
71 unsigned frame_number
;
73 struct pipe_screen
*screen
;
74 struct radeon_winsys
* ws
;
75 struct radeon_winsys_cs
* cs
;
79 struct rvid_buffer msg_fb_it_buffers
[NUM_BUFFERS
];
85 struct rvid_buffer bs_buffers
[NUM_BUFFERS
];
89 struct rvid_buffer dpb
;
91 struct rvid_buffer ctx
;
94 /* flush IB to the hardware */
95 static void flush(struct ruvd_decoder
*dec
)
97 dec
->ws
->cs_flush(dec
->cs
, RADEON_FLUSH_ASYNC
, NULL
);
100 /* add a new set register command to the IB */
101 static void set_reg(struct ruvd_decoder
*dec
, unsigned reg
, uint32_t val
)
103 radeon_emit(dec
->cs
, RUVD_PKT0(reg
>> 2, 0));
104 radeon_emit(dec
->cs
, val
);
107 /* send a command to the VCPU through the GPCOM registers */
108 static void send_cmd(struct ruvd_decoder
*dec
, unsigned cmd
,
109 struct pb_buffer
* buf
, uint32_t off
,
110 enum radeon_bo_usage usage
, enum radeon_bo_domain domain
)
114 reloc_idx
= dec
->ws
->cs_add_buffer(dec
->cs
, buf
, usage
, domain
,
116 if (!dec
->use_legacy
) {
118 addr
= dec
->ws
->buffer_get_virtual_address(buf
);
120 set_reg(dec
, RUVD_GPCOM_VCPU_DATA0
, addr
);
121 set_reg(dec
, RUVD_GPCOM_VCPU_DATA1
, addr
>> 32);
123 set_reg(dec
, RUVD_GPCOM_VCPU_DATA0
, off
);
124 set_reg(dec
, RUVD_GPCOM_VCPU_DATA1
, reloc_idx
* 4);
126 set_reg(dec
, RUVD_GPCOM_VCPU_CMD
, cmd
<< 1);
129 /* do the codec needs an IT buffer ?*/
130 static bool have_it(struct ruvd_decoder
*dec
)
132 return dec
->stream_type
== RUVD_CODEC_H264_PERF
||
133 dec
->stream_type
== RUVD_CODEC_H265
;
136 /* map the next available message/feedback/itscaling buffer */
137 static void map_msg_fb_it_buf(struct ruvd_decoder
*dec
)
139 struct rvid_buffer
* buf
;
142 /* grab the current message/feedback buffer */
143 buf
= &dec
->msg_fb_it_buffers
[dec
->cur_buffer
];
145 /* and map it for CPU access */
146 ptr
= dec
->ws
->buffer_map(buf
->res
->buf
, dec
->cs
, PIPE_TRANSFER_WRITE
);
148 /* calc buffer offsets */
149 dec
->msg
= (struct ruvd_msg
*)ptr
;
150 dec
->fb
= (uint32_t *)(ptr
+ FB_BUFFER_OFFSET
);
152 dec
->it
= (uint8_t *)(ptr
+ FB_BUFFER_OFFSET
+ dec
->fb_size
);
155 /* unmap and send a message command to the VCPU */
156 static void send_msg_buf(struct ruvd_decoder
*dec
)
158 struct rvid_buffer
* buf
;
160 /* ignore the request if message/feedback buffer isn't mapped */
161 if (!dec
->msg
|| !dec
->fb
)
164 /* grab the current message buffer */
165 buf
= &dec
->msg_fb_it_buffers
[dec
->cur_buffer
];
167 /* unmap the buffer */
168 dec
->ws
->buffer_unmap(buf
->res
->buf
);
173 /* and send it to the hardware */
174 send_cmd(dec
, RUVD_CMD_MSG_BUFFER
, buf
->res
->buf
, 0,
175 RADEON_USAGE_READ
, RADEON_DOMAIN_GTT
);
178 /* cycle to the next set of buffers */
179 static void next_buffer(struct ruvd_decoder
*dec
)
182 dec
->cur_buffer
%= NUM_BUFFERS
;
185 /* convert the profile into something UVD understands */
186 static uint32_t profile2stream_type(struct ruvd_decoder
*dec
, unsigned family
)
188 switch (u_reduce_video_profile(dec
->base
.profile
)) {
189 case PIPE_VIDEO_FORMAT_MPEG4_AVC
:
190 return (family
>= CHIP_TONGA
) ?
191 RUVD_CODEC_H264_PERF
: RUVD_CODEC_H264
;
193 case PIPE_VIDEO_FORMAT_VC1
:
194 return RUVD_CODEC_VC1
;
196 case PIPE_VIDEO_FORMAT_MPEG12
:
197 return RUVD_CODEC_MPEG2
;
199 case PIPE_VIDEO_FORMAT_MPEG4
:
200 return RUVD_CODEC_MPEG4
;
202 case PIPE_VIDEO_FORMAT_HEVC
:
203 return RUVD_CODEC_H265
;
211 static unsigned calc_ctx_size_h264_perf(struct ruvd_decoder
*dec
)
213 unsigned width_in_mb
, height_in_mb
, ctx_size
;
214 unsigned width
= align(dec
->base
.width
, VL_MACROBLOCK_WIDTH
);
215 unsigned height
= align(dec
->base
.height
, VL_MACROBLOCK_HEIGHT
);
217 unsigned max_references
= dec
->base
.max_references
+ 1;
219 // picture width & height in 16 pixel units
220 width_in_mb
= width
/ VL_MACROBLOCK_WIDTH
;
221 height_in_mb
= align(height
/ VL_MACROBLOCK_HEIGHT
, 2);
223 if (!dec
->use_legacy
) {
224 unsigned fs_in_mb
= width_in_mb
* height_in_mb
;
225 unsigned num_dpb_buffer
;
226 switch(dec
->base
.level
) {
228 num_dpb_buffer
= 8100 / fs_in_mb
;
231 num_dpb_buffer
= 18000 / fs_in_mb
;
234 num_dpb_buffer
= 20480 / fs_in_mb
;
237 num_dpb_buffer
= 32768 / fs_in_mb
;
240 num_dpb_buffer
= 34816 / fs_in_mb
;
243 num_dpb_buffer
= 110400 / fs_in_mb
;
246 num_dpb_buffer
= 184320 / fs_in_mb
;
249 num_dpb_buffer
= 184320 / fs_in_mb
;
253 max_references
= MAX2(MIN2(NUM_H264_REFS
, num_dpb_buffer
), max_references
);
254 ctx_size
= max_references
* align(width_in_mb
* height_in_mb
* 192, 256);
256 // the firmware seems to always assume a minimum of ref frames
257 max_references
= MAX2(NUM_H264_REFS
, max_references
);
258 // macroblock context buffer
259 ctx_size
= align(width_in_mb
* height_in_mb
* max_references
* 192, 256);
265 static unsigned calc_ctx_size_h265_main(struct ruvd_decoder
*dec
)
267 unsigned width
= align(dec
->base
.width
, VL_MACROBLOCK_WIDTH
);
268 unsigned height
= align(dec
->base
.height
, VL_MACROBLOCK_HEIGHT
);
270 unsigned max_references
= dec
->base
.max_references
+ 1;
272 if (dec
->base
.width
* dec
->base
.height
>= 4096*2000)
273 max_references
= MAX2(max_references
, 8);
275 max_references
= MAX2(max_references
, 17);
277 width
= align (width
, 16);
278 height
= align (height
, 16);
279 return ((width
+ 255) / 16) * ((height
+ 255) / 16) * 16 * max_references
+ 52 * 1024;
282 static unsigned calc_ctx_size_h265_main10(struct ruvd_decoder
*dec
, struct pipe_h265_picture_desc
*pic
)
284 unsigned block_size
, log2_ctb_size
, width_in_ctb
, height_in_ctb
, num_16x16_block_per_ctb
;
285 unsigned context_buffer_size_per_ctb_row
, cm_buffer_size
, max_mb_address
, db_left_tile_pxl_size
;
286 unsigned db_left_tile_ctx_size
= 4096 / 16 * (32 + 16 * 4);
288 unsigned width
= align(dec
->base
.width
, VL_MACROBLOCK_WIDTH
);
289 unsigned height
= align(dec
->base
.height
, VL_MACROBLOCK_HEIGHT
);
290 unsigned coeff_10bit
= (pic
->pps
->sps
->bit_depth_luma_minus8
|| pic
->pps
->sps
->bit_depth_chroma_minus8
) ? 2 : 1;
292 unsigned max_references
= dec
->base
.max_references
+ 1;
294 if (dec
->base
.width
* dec
->base
.height
>= 4096*2000)
295 max_references
= MAX2(max_references
, 8);
297 max_references
= MAX2(max_references
, 17);
299 block_size
= (1 << (pic
->pps
->sps
->log2_min_luma_coding_block_size_minus3
+ 3));
300 log2_ctb_size
= block_size
+ pic
->pps
->sps
->log2_diff_max_min_luma_coding_block_size
;
302 width_in_ctb
= (width
+ ((1 << log2_ctb_size
) - 1)) >> log2_ctb_size
;
303 height_in_ctb
= (height
+ ((1 << log2_ctb_size
) - 1)) >> log2_ctb_size
;
305 num_16x16_block_per_ctb
= ((1 << log2_ctb_size
) >> 4) * ((1 << log2_ctb_size
) >> 4);
306 context_buffer_size_per_ctb_row
= align(width_in_ctb
* num_16x16_block_per_ctb
* 16, 256);
307 max_mb_address
= (unsigned) ceil(height
* 8 / 2048.0);
309 cm_buffer_size
= max_references
* context_buffer_size_per_ctb_row
* height_in_ctb
;
310 db_left_tile_pxl_size
= coeff_10bit
* (max_mb_address
* 2 * 2048 + 1024);
312 return cm_buffer_size
+ db_left_tile_ctx_size
+ db_left_tile_pxl_size
;
315 /* calculate size of reference picture buffer */
316 static unsigned calc_dpb_size(struct ruvd_decoder
*dec
)
318 unsigned width_in_mb
, height_in_mb
, image_size
, dpb_size
;
320 // always align them to MB size for dpb calculation
321 unsigned width
= align(dec
->base
.width
, VL_MACROBLOCK_WIDTH
);
322 unsigned height
= align(dec
->base
.height
, VL_MACROBLOCK_HEIGHT
);
324 // always one more for currently decoded picture
325 unsigned max_references
= dec
->base
.max_references
+ 1;
327 // aligned size of a single frame
328 image_size
= width
* height
;
329 image_size
+= image_size
/ 2;
330 image_size
= align(image_size
, 1024);
332 // picture width & height in 16 pixel units
333 width_in_mb
= width
/ VL_MACROBLOCK_WIDTH
;
334 height_in_mb
= align(height
/ VL_MACROBLOCK_HEIGHT
, 2);
336 switch (u_reduce_video_profile(dec
->base
.profile
)) {
337 case PIPE_VIDEO_FORMAT_MPEG4_AVC
: {
338 if (!dec
->use_legacy
) {
339 unsigned fs_in_mb
= width_in_mb
* height_in_mb
;
340 unsigned alignment
= 64, num_dpb_buffer
;
342 if (dec
->stream_type
== RUVD_CODEC_H264_PERF
)
344 switch(dec
->base
.level
) {
346 num_dpb_buffer
= 8100 / fs_in_mb
;
349 num_dpb_buffer
= 18000 / fs_in_mb
;
352 num_dpb_buffer
= 20480 / fs_in_mb
;
355 num_dpb_buffer
= 32768 / fs_in_mb
;
358 num_dpb_buffer
= 34816 / fs_in_mb
;
361 num_dpb_buffer
= 110400 / fs_in_mb
;
364 num_dpb_buffer
= 184320 / fs_in_mb
;
367 num_dpb_buffer
= 184320 / fs_in_mb
;
371 max_references
= MAX2(MIN2(NUM_H264_REFS
, num_dpb_buffer
), max_references
);
372 dpb_size
= image_size
* max_references
;
373 if ((dec
->stream_type
!= RUVD_CODEC_H264_PERF
) ||
374 (((struct r600_common_screen
*)dec
->screen
)->family
< CHIP_POLARIS10
)) {
375 dpb_size
+= max_references
* align(width_in_mb
* height_in_mb
* 192, alignment
);
376 dpb_size
+= align(width_in_mb
* height_in_mb
* 32, alignment
);
379 // the firmware seems to allways assume a minimum of ref frames
380 max_references
= MAX2(NUM_H264_REFS
, max_references
);
381 // reference picture buffer
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 // macroblock context buffer
386 dpb_size
+= width_in_mb
* height_in_mb
* max_references
* 192;
388 dpb_size
+= width_in_mb
* height_in_mb
* 32;
394 case PIPE_VIDEO_FORMAT_HEVC
:
395 if (dec
->base
.width
* dec
->base
.height
>= 4096*2000)
396 max_references
= MAX2(max_references
, 8);
398 max_references
= MAX2(max_references
, 17);
400 width
= align (width
, 16);
401 height
= align (height
, 16);
402 if (dec
->base
.profile
== PIPE_VIDEO_PROFILE_HEVC_MAIN_10
)
403 dpb_size
= align((width
* height
* 9) / 4, 256) * max_references
;
405 dpb_size
= align((width
* height
* 3) / 2, 256) * max_references
;
408 case PIPE_VIDEO_FORMAT_VC1
:
409 // the firmware seems to allways assume a minimum of ref frames
410 max_references
= MAX2(NUM_VC1_REFS
, max_references
);
412 // reference picture buffer
413 dpb_size
= image_size
* max_references
;
416 dpb_size
+= width_in_mb
* height_in_mb
* 128;
419 dpb_size
+= width_in_mb
* 64;
422 dpb_size
+= width_in_mb
* 128;
425 dpb_size
+= align(MAX2(width_in_mb
, height_in_mb
) * 7 * 16, 64);
428 case PIPE_VIDEO_FORMAT_MPEG12
:
429 // reference picture buffer, must be big enough for all frames
430 dpb_size
= image_size
* NUM_MPEG2_REFS
;
433 case PIPE_VIDEO_FORMAT_MPEG4
:
434 // reference picture buffer
435 dpb_size
= image_size
* max_references
;
438 dpb_size
+= width_in_mb
* height_in_mb
* 64;
441 dpb_size
+= align(width_in_mb
* height_in_mb
* 32, 64);
443 dpb_size
= MAX2(dpb_size
, 30 * 1024 * 1024);
447 // something is missing here
450 // at least use a sane default value
451 dpb_size
= 32 * 1024 * 1024;
457 /* free associated data in the video buffer callback */
458 static void ruvd_destroy_associated_data(void *data
)
460 /* NOOP, since we only use an intptr */
463 /* get h264 specific message bits */
464 static struct ruvd_h264
get_h264_msg(struct ruvd_decoder
*dec
, struct pipe_h264_picture_desc
*pic
)
466 struct ruvd_h264 result
;
468 memset(&result
, 0, sizeof(result
));
469 switch (pic
->base
.profile
) {
470 case PIPE_VIDEO_PROFILE_MPEG4_AVC_BASELINE
:
471 result
.profile
= RUVD_H264_PROFILE_BASELINE
;
474 case PIPE_VIDEO_PROFILE_MPEG4_AVC_MAIN
:
475 result
.profile
= RUVD_H264_PROFILE_MAIN
;
478 case PIPE_VIDEO_PROFILE_MPEG4_AVC_HIGH
:
479 result
.profile
= RUVD_H264_PROFILE_HIGH
;
487 result
.level
= dec
->base
.level
;
489 result
.sps_info_flags
= 0;
490 result
.sps_info_flags
|= pic
->pps
->sps
->direct_8x8_inference_flag
<< 0;
491 result
.sps_info_flags
|= pic
->pps
->sps
->mb_adaptive_frame_field_flag
<< 1;
492 result
.sps_info_flags
|= pic
->pps
->sps
->frame_mbs_only_flag
<< 2;
493 result
.sps_info_flags
|= pic
->pps
->sps
->delta_pic_order_always_zero_flag
<< 3;
495 result
.bit_depth_luma_minus8
= pic
->pps
->sps
->bit_depth_luma_minus8
;
496 result
.bit_depth_chroma_minus8
= pic
->pps
->sps
->bit_depth_chroma_minus8
;
497 result
.log2_max_frame_num_minus4
= pic
->pps
->sps
->log2_max_frame_num_minus4
;
498 result
.pic_order_cnt_type
= pic
->pps
->sps
->pic_order_cnt_type
;
499 result
.log2_max_pic_order_cnt_lsb_minus4
= pic
->pps
->sps
->log2_max_pic_order_cnt_lsb_minus4
;
501 switch (dec
->base
.chroma_format
) {
502 case PIPE_VIDEO_CHROMA_FORMAT_NONE
:
505 case PIPE_VIDEO_CHROMA_FORMAT_400
:
506 result
.chroma_format
= 0;
508 case PIPE_VIDEO_CHROMA_FORMAT_420
:
509 result
.chroma_format
= 1;
511 case PIPE_VIDEO_CHROMA_FORMAT_422
:
512 result
.chroma_format
= 2;
514 case PIPE_VIDEO_CHROMA_FORMAT_444
:
515 result
.chroma_format
= 3;
519 result
.pps_info_flags
= 0;
520 result
.pps_info_flags
|= pic
->pps
->transform_8x8_mode_flag
<< 0;
521 result
.pps_info_flags
|= pic
->pps
->redundant_pic_cnt_present_flag
<< 1;
522 result
.pps_info_flags
|= pic
->pps
->constrained_intra_pred_flag
<< 2;
523 result
.pps_info_flags
|= pic
->pps
->deblocking_filter_control_present_flag
<< 3;
524 result
.pps_info_flags
|= pic
->pps
->weighted_bipred_idc
<< 4;
525 result
.pps_info_flags
|= pic
->pps
->weighted_pred_flag
<< 6;
526 result
.pps_info_flags
|= pic
->pps
->bottom_field_pic_order_in_frame_present_flag
<< 7;
527 result
.pps_info_flags
|= pic
->pps
->entropy_coding_mode_flag
<< 8;
529 result
.num_slice_groups_minus1
= pic
->pps
->num_slice_groups_minus1
;
530 result
.slice_group_map_type
= pic
->pps
->slice_group_map_type
;
531 result
.slice_group_change_rate_minus1
= pic
->pps
->slice_group_change_rate_minus1
;
532 result
.pic_init_qp_minus26
= pic
->pps
->pic_init_qp_minus26
;
533 result
.chroma_qp_index_offset
= pic
->pps
->chroma_qp_index_offset
;
534 result
.second_chroma_qp_index_offset
= pic
->pps
->second_chroma_qp_index_offset
;
536 memcpy(result
.scaling_list_4x4
, pic
->pps
->ScalingList4x4
, 6*16);
537 memcpy(result
.scaling_list_8x8
, pic
->pps
->ScalingList8x8
, 2*64);
539 if (dec
->stream_type
== RUVD_CODEC_H264_PERF
) {
540 memcpy(dec
->it
, result
.scaling_list_4x4
, 6*16);
541 memcpy((dec
->it
+ 96), result
.scaling_list_8x8
, 2*64);
544 result
.num_ref_frames
= pic
->num_ref_frames
;
546 result
.num_ref_idx_l0_active_minus1
= pic
->num_ref_idx_l0_active_minus1
;
547 result
.num_ref_idx_l1_active_minus1
= pic
->num_ref_idx_l1_active_minus1
;
549 result
.frame_num
= pic
->frame_num
;
550 memcpy(result
.frame_num_list
, pic
->frame_num_list
, 4*16);
551 result
.curr_field_order_cnt_list
[0] = pic
->field_order_cnt
[0];
552 result
.curr_field_order_cnt_list
[1] = pic
->field_order_cnt
[1];
553 memcpy(result
.field_order_cnt_list
, pic
->field_order_cnt_list
, 4*16*2);
555 result
.decoded_pic_idx
= pic
->frame_num
;
560 /* get h265 specific message bits */
561 static struct ruvd_h265
get_h265_msg(struct ruvd_decoder
*dec
, struct pipe_video_buffer
*target
,
562 struct pipe_h265_picture_desc
*pic
)
564 struct ruvd_h265 result
;
567 memset(&result
, 0, sizeof(result
));
569 result
.sps_info_flags
= 0;
570 result
.sps_info_flags
|= pic
->pps
->sps
->scaling_list_enabled_flag
<< 0;
571 result
.sps_info_flags
|= pic
->pps
->sps
->amp_enabled_flag
<< 1;
572 result
.sps_info_flags
|= pic
->pps
->sps
->sample_adaptive_offset_enabled_flag
<< 2;
573 result
.sps_info_flags
|= pic
->pps
->sps
->pcm_enabled_flag
<< 3;
574 result
.sps_info_flags
|= pic
->pps
->sps
->pcm_loop_filter_disabled_flag
<< 4;
575 result
.sps_info_flags
|= pic
->pps
->sps
->long_term_ref_pics_present_flag
<< 5;
576 result
.sps_info_flags
|= pic
->pps
->sps
->sps_temporal_mvp_enabled_flag
<< 6;
577 result
.sps_info_flags
|= pic
->pps
->sps
->strong_intra_smoothing_enabled_flag
<< 7;
578 result
.sps_info_flags
|= pic
->pps
->sps
->separate_colour_plane_flag
<< 8;
579 if (((struct r600_common_screen
*)dec
->screen
)->family
== CHIP_CARRIZO
)
580 result
.sps_info_flags
|= 1 << 9;
581 if (pic
->UseRefPicList
== true)
582 result
.sps_info_flags
|= 1 << 10;
584 result
.chroma_format
= pic
->pps
->sps
->chroma_format_idc
;
585 result
.bit_depth_luma_minus8
= pic
->pps
->sps
->bit_depth_luma_minus8
;
586 result
.bit_depth_chroma_minus8
= pic
->pps
->sps
->bit_depth_chroma_minus8
;
587 result
.log2_max_pic_order_cnt_lsb_minus4
= pic
->pps
->sps
->log2_max_pic_order_cnt_lsb_minus4
;
588 result
.sps_max_dec_pic_buffering_minus1
= pic
->pps
->sps
->sps_max_dec_pic_buffering_minus1
;
589 result
.log2_min_luma_coding_block_size_minus3
= pic
->pps
->sps
->log2_min_luma_coding_block_size_minus3
;
590 result
.log2_diff_max_min_luma_coding_block_size
= pic
->pps
->sps
->log2_diff_max_min_luma_coding_block_size
;
591 result
.log2_min_transform_block_size_minus2
= pic
->pps
->sps
->log2_min_transform_block_size_minus2
;
592 result
.log2_diff_max_min_transform_block_size
= pic
->pps
->sps
->log2_diff_max_min_transform_block_size
;
593 result
.max_transform_hierarchy_depth_inter
= pic
->pps
->sps
->max_transform_hierarchy_depth_inter
;
594 result
.max_transform_hierarchy_depth_intra
= pic
->pps
->sps
->max_transform_hierarchy_depth_intra
;
595 result
.pcm_sample_bit_depth_luma_minus1
= pic
->pps
->sps
->pcm_sample_bit_depth_luma_minus1
;
596 result
.pcm_sample_bit_depth_chroma_minus1
= pic
->pps
->sps
->pcm_sample_bit_depth_chroma_minus1
;
597 result
.log2_min_pcm_luma_coding_block_size_minus3
= pic
->pps
->sps
->log2_min_pcm_luma_coding_block_size_minus3
;
598 result
.log2_diff_max_min_pcm_luma_coding_block_size
= pic
->pps
->sps
->log2_diff_max_min_pcm_luma_coding_block_size
;
599 result
.num_short_term_ref_pic_sets
= pic
->pps
->sps
->num_short_term_ref_pic_sets
;
601 result
.pps_info_flags
= 0;
602 result
.pps_info_flags
|= pic
->pps
->dependent_slice_segments_enabled_flag
<< 0;
603 result
.pps_info_flags
|= pic
->pps
->output_flag_present_flag
<< 1;
604 result
.pps_info_flags
|= pic
->pps
->sign_data_hiding_enabled_flag
<< 2;
605 result
.pps_info_flags
|= pic
->pps
->cabac_init_present_flag
<< 3;
606 result
.pps_info_flags
|= pic
->pps
->constrained_intra_pred_flag
<< 4;
607 result
.pps_info_flags
|= pic
->pps
->transform_skip_enabled_flag
<< 5;
608 result
.pps_info_flags
|= pic
->pps
->cu_qp_delta_enabled_flag
<< 6;
609 result
.pps_info_flags
|= pic
->pps
->pps_slice_chroma_qp_offsets_present_flag
<< 7;
610 result
.pps_info_flags
|= pic
->pps
->weighted_pred_flag
<< 8;
611 result
.pps_info_flags
|= pic
->pps
->weighted_bipred_flag
<< 9;
612 result
.pps_info_flags
|= pic
->pps
->transquant_bypass_enabled_flag
<< 10;
613 result
.pps_info_flags
|= pic
->pps
->tiles_enabled_flag
<< 11;
614 result
.pps_info_flags
|= pic
->pps
->entropy_coding_sync_enabled_flag
<< 12;
615 result
.pps_info_flags
|= pic
->pps
->uniform_spacing_flag
<< 13;
616 result
.pps_info_flags
|= pic
->pps
->loop_filter_across_tiles_enabled_flag
<< 14;
617 result
.pps_info_flags
|= pic
->pps
->pps_loop_filter_across_slices_enabled_flag
<< 15;
618 result
.pps_info_flags
|= pic
->pps
->deblocking_filter_override_enabled_flag
<< 16;
619 result
.pps_info_flags
|= pic
->pps
->pps_deblocking_filter_disabled_flag
<< 17;
620 result
.pps_info_flags
|= pic
->pps
->lists_modification_present_flag
<< 18;
621 result
.pps_info_flags
|= pic
->pps
->slice_segment_header_extension_present_flag
<< 19;
622 //result.pps_info_flags |= pic->pps->deblocking_filter_control_present_flag; ???
624 result
.num_extra_slice_header_bits
= pic
->pps
->num_extra_slice_header_bits
;
625 result
.num_long_term_ref_pic_sps
= pic
->pps
->sps
->num_long_term_ref_pics_sps
;
626 result
.num_ref_idx_l0_default_active_minus1
= pic
->pps
->num_ref_idx_l0_default_active_minus1
;
627 result
.num_ref_idx_l1_default_active_minus1
= pic
->pps
->num_ref_idx_l1_default_active_minus1
;
628 result
.pps_cb_qp_offset
= pic
->pps
->pps_cb_qp_offset
;
629 result
.pps_cr_qp_offset
= pic
->pps
->pps_cr_qp_offset
;
630 result
.pps_beta_offset_div2
= pic
->pps
->pps_beta_offset_div2
;
631 result
.pps_tc_offset_div2
= pic
->pps
->pps_tc_offset_div2
;
632 result
.diff_cu_qp_delta_depth
= pic
->pps
->diff_cu_qp_delta_depth
;
633 result
.num_tile_columns_minus1
= pic
->pps
->num_tile_columns_minus1
;
634 result
.num_tile_rows_minus1
= pic
->pps
->num_tile_rows_minus1
;
635 result
.log2_parallel_merge_level_minus2
= pic
->pps
->log2_parallel_merge_level_minus2
;
636 result
.init_qp_minus26
= pic
->pps
->init_qp_minus26
;
638 for (i
= 0; i
< 19; ++i
)
639 result
.column_width_minus1
[i
] = pic
->pps
->column_width_minus1
[i
];
641 for (i
= 0; i
< 21; ++i
)
642 result
.row_height_minus1
[i
] = pic
->pps
->row_height_minus1
[i
];
644 result
.num_delta_pocs_ref_rps_idx
= pic
->NumDeltaPocsOfRefRpsIdx
;
645 result
.curr_idx
= pic
->CurrPicOrderCntVal
;
646 result
.curr_poc
= pic
->CurrPicOrderCntVal
;
648 vl_video_buffer_set_associated_data(target
, &dec
->base
,
649 (void *)(uintptr_t)pic
->CurrPicOrderCntVal
,
650 &ruvd_destroy_associated_data
);
652 for (i
= 0; i
< 16; ++i
) {
653 struct pipe_video_buffer
*ref
= pic
->ref
[i
];
654 uintptr_t ref_pic
= 0;
656 result
.poc_list
[i
] = pic
->PicOrderCntVal
[i
];
659 ref_pic
= (uintptr_t)vl_video_buffer_get_associated_data(ref
, &dec
->base
);
662 result
.ref_pic_list
[i
] = ref_pic
;
665 for (i
= 0; i
< 8; ++i
) {
666 result
.ref_pic_set_st_curr_before
[i
] = 0xFF;
667 result
.ref_pic_set_st_curr_after
[i
] = 0xFF;
668 result
.ref_pic_set_lt_curr
[i
] = 0xFF;
671 for (i
= 0; i
< pic
->NumPocStCurrBefore
; ++i
)
672 result
.ref_pic_set_st_curr_before
[i
] = pic
->RefPicSetStCurrBefore
[i
];
674 for (i
= 0; i
< pic
->NumPocStCurrAfter
; ++i
)
675 result
.ref_pic_set_st_curr_after
[i
] = pic
->RefPicSetStCurrAfter
[i
];
677 for (i
= 0; i
< pic
->NumPocLtCurr
; ++i
)
678 result
.ref_pic_set_lt_curr
[i
] = pic
->RefPicSetLtCurr
[i
];
680 for (i
= 0; i
< 6; ++i
)
681 result
.ucScalingListDCCoefSizeID2
[i
] = pic
->pps
->sps
->ScalingListDCCoeff16x16
[i
];
683 for (i
= 0; i
< 2; ++i
)
684 result
.ucScalingListDCCoefSizeID3
[i
] = pic
->pps
->sps
->ScalingListDCCoeff32x32
[i
];
686 memcpy(dec
->it
, pic
->pps
->sps
->ScalingList4x4
, 6 * 16);
687 memcpy(dec
->it
+ 96, pic
->pps
->sps
->ScalingList8x8
, 6 * 64);
688 memcpy(dec
->it
+ 480, pic
->pps
->sps
->ScalingList16x16
, 6 * 64);
689 memcpy(dec
->it
+ 864, pic
->pps
->sps
->ScalingList32x32
, 2 * 64);
691 for (i
= 0 ; i
< 2 ; i
++) {
692 for (int j
= 0 ; j
< 15 ; j
++)
693 result
.direct_reflist
[i
][j
] = pic
->RefPicList
[i
][j
];
696 if ((pic
->base
.profile
== PIPE_VIDEO_PROFILE_HEVC_MAIN_10
) &&
697 (target
->buffer_format
== PIPE_FORMAT_NV12
)) {
698 result
.p010_mode
= 0;
699 result
.luma_10to8
= 5;
700 result
.chroma_10to8
= 5;
701 result
.sclr_luma10to8
= 4;
702 result
.sclr_chroma10to8
= 4;
712 NumShortTermPictureSliceHeaderBits;
713 NumLongTermPictureSliceHeaderBits;
721 /* get vc1 specific message bits */
722 static struct ruvd_vc1
get_vc1_msg(struct pipe_vc1_picture_desc
*pic
)
724 struct ruvd_vc1 result
;
726 memset(&result
, 0, sizeof(result
));
728 switch(pic
->base
.profile
) {
729 case PIPE_VIDEO_PROFILE_VC1_SIMPLE
:
730 result
.profile
= RUVD_VC1_PROFILE_SIMPLE
;
734 case PIPE_VIDEO_PROFILE_VC1_MAIN
:
735 result
.profile
= RUVD_VC1_PROFILE_MAIN
;
739 case PIPE_VIDEO_PROFILE_VC1_ADVANCED
:
740 result
.profile
= RUVD_VC1_PROFILE_ADVANCED
;
748 /* fields common for all profiles */
749 result
.sps_info_flags
|= pic
->postprocflag
<< 7;
750 result
.sps_info_flags
|= pic
->pulldown
<< 6;
751 result
.sps_info_flags
|= pic
->interlace
<< 5;
752 result
.sps_info_flags
|= pic
->tfcntrflag
<< 4;
753 result
.sps_info_flags
|= pic
->finterpflag
<< 3;
754 result
.sps_info_flags
|= pic
->psf
<< 1;
756 result
.pps_info_flags
|= pic
->range_mapy_flag
<< 31;
757 result
.pps_info_flags
|= pic
->range_mapy
<< 28;
758 result
.pps_info_flags
|= pic
->range_mapuv_flag
<< 27;
759 result
.pps_info_flags
|= pic
->range_mapuv
<< 24;
760 result
.pps_info_flags
|= pic
->multires
<< 21;
761 result
.pps_info_flags
|= pic
->maxbframes
<< 16;
762 result
.pps_info_flags
|= pic
->overlap
<< 11;
763 result
.pps_info_flags
|= pic
->quantizer
<< 9;
764 result
.pps_info_flags
|= pic
->panscan_flag
<< 7;
765 result
.pps_info_flags
|= pic
->refdist_flag
<< 6;
766 result
.pps_info_flags
|= pic
->vstransform
<< 0;
768 /* some fields only apply to main/advanced profile */
769 if (pic
->base
.profile
!= PIPE_VIDEO_PROFILE_VC1_SIMPLE
) {
770 result
.pps_info_flags
|= pic
->syncmarker
<< 20;
771 result
.pps_info_flags
|= pic
->rangered
<< 19;
772 result
.pps_info_flags
|= pic
->loopfilter
<< 5;
773 result
.pps_info_flags
|= pic
->fastuvmc
<< 4;
774 result
.pps_info_flags
|= pic
->extended_mv
<< 3;
775 result
.pps_info_flags
|= pic
->extended_dmv
<< 8;
776 result
.pps_info_flags
|= pic
->dquant
<< 1;
779 result
.chroma_format
= 1;
782 //(((unsigned int)(pPicParams->advance.reserved1)) << SPS_INFO_VC1_RESERVED_SHIFT)
785 uint8_t frame_coding_mode
786 uint8_t deblockEnable
793 /* extract the frame number from a referenced video buffer */
794 static uint32_t get_ref_pic_idx(struct ruvd_decoder
*dec
, struct pipe_video_buffer
*ref
)
796 uint32_t min
= MAX2(dec
->frame_number
, NUM_MPEG2_REFS
) - NUM_MPEG2_REFS
;
797 uint32_t max
= MAX2(dec
->frame_number
, 1) - 1;
800 /* seems to be the most sane fallback */
804 /* get the frame number from the associated data */
805 frame
= (uintptr_t)vl_video_buffer_get_associated_data(ref
, &dec
->base
);
807 /* limit the frame number to a valid range */
808 return MAX2(MIN2(frame
, max
), min
);
811 /* get mpeg2 specific msg bits */
812 static struct ruvd_mpeg2
get_mpeg2_msg(struct ruvd_decoder
*dec
,
813 struct pipe_mpeg12_picture_desc
*pic
)
815 const int *zscan
= pic
->alternate_scan
? vl_zscan_alternate
: vl_zscan_normal
;
816 struct ruvd_mpeg2 result
;
819 memset(&result
, 0, sizeof(result
));
820 result
.decoded_pic_idx
= dec
->frame_number
;
821 for (i
= 0; i
< 2; ++i
)
822 result
.ref_pic_idx
[i
] = get_ref_pic_idx(dec
, pic
->ref
[i
]);
824 result
.load_intra_quantiser_matrix
= 1;
825 result
.load_nonintra_quantiser_matrix
= 1;
827 for (i
= 0; i
< 64; ++i
) {
828 result
.intra_quantiser_matrix
[i
] = pic
->intra_matrix
[zscan
[i
]];
829 result
.nonintra_quantiser_matrix
[i
] = pic
->non_intra_matrix
[zscan
[i
]];
832 result
.profile_and_level_indication
= 0;
833 result
.chroma_format
= 0x1;
835 result
.picture_coding_type
= pic
->picture_coding_type
;
836 result
.f_code
[0][0] = pic
->f_code
[0][0] + 1;
837 result
.f_code
[0][1] = pic
->f_code
[0][1] + 1;
838 result
.f_code
[1][0] = pic
->f_code
[1][0] + 1;
839 result
.f_code
[1][1] = pic
->f_code
[1][1] + 1;
840 result
.intra_dc_precision
= pic
->intra_dc_precision
;
841 result
.pic_structure
= pic
->picture_structure
;
842 result
.top_field_first
= pic
->top_field_first
;
843 result
.frame_pred_frame_dct
= pic
->frame_pred_frame_dct
;
844 result
.concealment_motion_vectors
= pic
->concealment_motion_vectors
;
845 result
.q_scale_type
= pic
->q_scale_type
;
846 result
.intra_vlc_format
= pic
->intra_vlc_format
;
847 result
.alternate_scan
= pic
->alternate_scan
;
852 /* get mpeg4 specific msg bits */
853 static struct ruvd_mpeg4
get_mpeg4_msg(struct ruvd_decoder
*dec
,
854 struct pipe_mpeg4_picture_desc
*pic
)
856 struct ruvd_mpeg4 result
;
859 memset(&result
, 0, sizeof(result
));
860 result
.decoded_pic_idx
= dec
->frame_number
;
861 for (i
= 0; i
< 2; ++i
)
862 result
.ref_pic_idx
[i
] = get_ref_pic_idx(dec
, pic
->ref
[i
]);
864 result
.variant_type
= 0;
865 result
.profile_and_level_indication
= 0xF0; // ASP Level0
867 result
.video_object_layer_verid
= 0x5; // advanced simple
868 result
.video_object_layer_shape
= 0x0; // rectangular
870 result
.video_object_layer_width
= dec
->base
.width
;
871 result
.video_object_layer_height
= dec
->base
.height
;
873 result
.vop_time_increment_resolution
= pic
->vop_time_increment_resolution
;
875 result
.flags
|= pic
->short_video_header
<< 0;
876 //result.flags |= obmc_disable << 1;
877 result
.flags
|= pic
->interlaced
<< 2;
878 result
.flags
|= 1 << 3; // load_intra_quant_mat
879 result
.flags
|= 1 << 4; // load_nonintra_quant_mat
880 result
.flags
|= pic
->quarter_sample
<< 5;
881 result
.flags
|= 1 << 6; // complexity_estimation_disable
882 result
.flags
|= pic
->resync_marker_disable
<< 7;
883 //result.flags |= data_partitioned << 8;
884 //result.flags |= reversible_vlc << 9;
885 result
.flags
|= 0 << 10; // newpred_enable
886 result
.flags
|= 0 << 11; // reduced_resolution_vop_enable
887 //result.flags |= scalability << 12;
888 //result.flags |= is_object_layer_identifier << 13;
889 //result.flags |= fixed_vop_rate << 14;
890 //result.flags |= newpred_segment_type << 15;
892 result
.quant_type
= pic
->quant_type
;
894 for (i
= 0; i
< 64; ++i
) {
895 result
.intra_quant_mat
[i
] = pic
->intra_matrix
[vl_zscan_normal
[i
]];
896 result
.nonintra_quant_mat
[i
] = pic
->non_intra_matrix
[vl_zscan_normal
[i
]];
902 uint8_t vop_coding_type
903 uint8_t vop_fcode_forward
904 uint8_t vop_fcode_backward
905 uint8_t rounding_control
906 uint8_t alternate_vertical_scan_flag
907 uint8_t top_field_first
914 * destroy this video decoder
916 static void ruvd_destroy(struct pipe_video_codec
*decoder
)
918 struct ruvd_decoder
*dec
= (struct ruvd_decoder
*)decoder
;
923 map_msg_fb_it_buf(dec
);
924 memset(dec
->msg
, 0, sizeof(*dec
->msg
));
925 dec
->msg
->size
= sizeof(*dec
->msg
);
926 dec
->msg
->msg_type
= RUVD_MSG_DESTROY
;
927 dec
->msg
->stream_handle
= dec
->stream_handle
;
932 dec
->ws
->cs_destroy(dec
->cs
);
934 for (i
= 0; i
< NUM_BUFFERS
; ++i
) {
935 rvid_destroy_buffer(&dec
->msg_fb_it_buffers
[i
]);
936 rvid_destroy_buffer(&dec
->bs_buffers
[i
]);
939 rvid_destroy_buffer(&dec
->dpb
);
940 rvid_destroy_buffer(&dec
->ctx
);
946 * start decoding of a new frame
948 static void ruvd_begin_frame(struct pipe_video_codec
*decoder
,
949 struct pipe_video_buffer
*target
,
950 struct pipe_picture_desc
*picture
)
952 struct ruvd_decoder
*dec
= (struct ruvd_decoder
*)decoder
;
957 frame
= ++dec
->frame_number
;
958 vl_video_buffer_set_associated_data(target
, decoder
, (void *)frame
,
959 &ruvd_destroy_associated_data
);
962 dec
->bs_ptr
= dec
->ws
->buffer_map(
963 dec
->bs_buffers
[dec
->cur_buffer
].res
->buf
,
964 dec
->cs
, PIPE_TRANSFER_WRITE
);
968 * decode a macroblock
970 static void ruvd_decode_macroblock(struct pipe_video_codec
*decoder
,
971 struct pipe_video_buffer
*target
,
972 struct pipe_picture_desc
*picture
,
973 const struct pipe_macroblock
*macroblocks
,
974 unsigned num_macroblocks
)
976 /* not supported (yet) */
983 static void ruvd_decode_bitstream(struct pipe_video_codec
*decoder
,
984 struct pipe_video_buffer
*target
,
985 struct pipe_picture_desc
*picture
,
986 unsigned num_buffers
,
987 const void * const *buffers
,
988 const unsigned *sizes
)
990 struct ruvd_decoder
*dec
= (struct ruvd_decoder
*)decoder
;
998 for (i
= 0; i
< num_buffers
; ++i
) {
999 struct rvid_buffer
*buf
= &dec
->bs_buffers
[dec
->cur_buffer
];
1000 unsigned new_size
= dec
->bs_size
+ sizes
[i
];
1002 if (new_size
> buf
->res
->buf
->size
) {
1003 dec
->ws
->buffer_unmap(buf
->res
->buf
);
1004 if (!rvid_resize_buffer(dec
->screen
, dec
->cs
, buf
, new_size
)) {
1005 RVID_ERR("Can't resize bitstream buffer!");
1009 dec
->bs_ptr
= dec
->ws
->buffer_map(buf
->res
->buf
, dec
->cs
,
1010 PIPE_TRANSFER_WRITE
);
1014 dec
->bs_ptr
+= dec
->bs_size
;
1017 memcpy(dec
->bs_ptr
, buffers
[i
], sizes
[i
]);
1018 dec
->bs_size
+= sizes
[i
];
1019 dec
->bs_ptr
+= sizes
[i
];
1024 * end decoding of the current frame
1026 static void ruvd_end_frame(struct pipe_video_codec
*decoder
,
1027 struct pipe_video_buffer
*target
,
1028 struct pipe_picture_desc
*picture
)
1030 struct ruvd_decoder
*dec
= (struct ruvd_decoder
*)decoder
;
1031 struct pb_buffer
*dt
;
1032 struct rvid_buffer
*msg_fb_it_buf
, *bs_buf
;
1040 msg_fb_it_buf
= &dec
->msg_fb_it_buffers
[dec
->cur_buffer
];
1041 bs_buf
= &dec
->bs_buffers
[dec
->cur_buffer
];
1043 bs_size
= align(dec
->bs_size
, 128);
1044 memset(dec
->bs_ptr
, 0, bs_size
- dec
->bs_size
);
1045 dec
->ws
->buffer_unmap(bs_buf
->res
->buf
);
1047 map_msg_fb_it_buf(dec
);
1048 dec
->msg
->size
= sizeof(*dec
->msg
);
1049 dec
->msg
->msg_type
= RUVD_MSG_DECODE
;
1050 dec
->msg
->stream_handle
= dec
->stream_handle
;
1051 dec
->msg
->status_report_feedback_number
= dec
->frame_number
;
1053 dec
->msg
->body
.decode
.stream_type
= dec
->stream_type
;
1054 dec
->msg
->body
.decode
.decode_flags
= 0x1;
1055 dec
->msg
->body
.decode
.width_in_samples
= dec
->base
.width
;
1056 dec
->msg
->body
.decode
.height_in_samples
= dec
->base
.height
;
1058 if ((picture
->profile
== PIPE_VIDEO_PROFILE_VC1_SIMPLE
) ||
1059 (picture
->profile
== PIPE_VIDEO_PROFILE_VC1_MAIN
)) {
1060 dec
->msg
->body
.decode
.width_in_samples
= align(dec
->msg
->body
.decode
.width_in_samples
, 16) / 16;
1061 dec
->msg
->body
.decode
.height_in_samples
= align(dec
->msg
->body
.decode
.height_in_samples
, 16) / 16;
1064 dec
->msg
->body
.decode
.dpb_size
= dec
->dpb
.res
->buf
->size
;
1065 dec
->msg
->body
.decode
.bsd_size
= bs_size
;
1066 dec
->msg
->body
.decode
.db_pitch
= align(dec
->base
.width
, 16);
1068 if (dec
->stream_type
== RUVD_CODEC_H264_PERF
&&
1069 ((struct r600_common_screen
*)dec
->screen
)->family
>= CHIP_POLARIS10
)
1070 dec
->msg
->body
.decode
.dpb_reserved
= dec
->ctx
.res
->buf
->size
;
1072 dt
= dec
->set_dtb(dec
->msg
, (struct vl_video_buffer
*)target
);
1073 if (((struct r600_common_screen
*)dec
->screen
)->family
>= CHIP_STONEY
)
1074 dec
->msg
->body
.decode
.dt_wa_chroma_top_offset
= dec
->msg
->body
.decode
.dt_pitch
/ 2;
1076 switch (u_reduce_video_profile(picture
->profile
)) {
1077 case PIPE_VIDEO_FORMAT_MPEG4_AVC
:
1078 dec
->msg
->body
.decode
.codec
.h264
= get_h264_msg(dec
, (struct pipe_h264_picture_desc
*)picture
);
1081 case PIPE_VIDEO_FORMAT_HEVC
:
1082 dec
->msg
->body
.decode
.codec
.h265
= get_h265_msg(dec
, target
, (struct pipe_h265_picture_desc
*)picture
);
1083 if (dec
->ctx
.res
== NULL
) {
1085 if (dec
->base
.profile
== PIPE_VIDEO_PROFILE_HEVC_MAIN_10
)
1086 ctx_size
= calc_ctx_size_h265_main10(dec
, (struct pipe_h265_picture_desc
*)picture
);
1088 ctx_size
= calc_ctx_size_h265_main(dec
);
1089 if (!rvid_create_buffer(dec
->screen
, &dec
->ctx
, ctx_size
, PIPE_USAGE_DEFAULT
)) {
1090 RVID_ERR("Can't allocated context buffer.\n");
1092 rvid_clear_buffer(decoder
->context
, &dec
->ctx
);
1096 dec
->msg
->body
.decode
.dpb_reserved
= dec
->ctx
.res
->buf
->size
;
1099 case PIPE_VIDEO_FORMAT_VC1
:
1100 dec
->msg
->body
.decode
.codec
.vc1
= get_vc1_msg((struct pipe_vc1_picture_desc
*)picture
);
1103 case PIPE_VIDEO_FORMAT_MPEG12
:
1104 dec
->msg
->body
.decode
.codec
.mpeg2
= get_mpeg2_msg(dec
, (struct pipe_mpeg12_picture_desc
*)picture
);
1107 case PIPE_VIDEO_FORMAT_MPEG4
:
1108 dec
->msg
->body
.decode
.codec
.mpeg4
= get_mpeg4_msg(dec
, (struct pipe_mpeg4_picture_desc
*)picture
);
1116 dec
->msg
->body
.decode
.db_surf_tile_config
= dec
->msg
->body
.decode
.dt_surf_tile_config
;
1117 dec
->msg
->body
.decode
.extension_support
= 0x1;
1119 /* set at least the feedback buffer size */
1120 dec
->fb
[0] = dec
->fb_size
;
1124 send_cmd(dec
, RUVD_CMD_DPB_BUFFER
, dec
->dpb
.res
->buf
, 0,
1125 RADEON_USAGE_READWRITE
, RADEON_DOMAIN_VRAM
);
1126 if ((u_reduce_video_profile(picture
->profile
) == PIPE_VIDEO_FORMAT_HEVC
) ||
1127 (dec
->stream_type
== RUVD_CODEC_H264_PERF
&&
1128 ((struct r600_common_screen
*)dec
->screen
)->family
>= CHIP_POLARIS10
)) {
1129 send_cmd(dec
, RUVD_CMD_CONTEXT_BUFFER
, dec
->ctx
.res
->buf
, 0,
1130 RADEON_USAGE_READWRITE
, RADEON_DOMAIN_VRAM
);
1132 send_cmd(dec
, RUVD_CMD_BITSTREAM_BUFFER
, bs_buf
->res
->buf
,
1133 0, RADEON_USAGE_READ
, RADEON_DOMAIN_GTT
);
1134 send_cmd(dec
, RUVD_CMD_DECODING_TARGET_BUFFER
, dt
, 0,
1135 RADEON_USAGE_WRITE
, RADEON_DOMAIN_VRAM
);
1136 send_cmd(dec
, RUVD_CMD_FEEDBACK_BUFFER
, msg_fb_it_buf
->res
->buf
,
1137 FB_BUFFER_OFFSET
, RADEON_USAGE_WRITE
, RADEON_DOMAIN_GTT
);
1139 send_cmd(dec
, RUVD_CMD_ITSCALING_TABLE_BUFFER
, msg_fb_it_buf
->res
->buf
,
1140 FB_BUFFER_OFFSET
+ dec
->fb_size
, RADEON_USAGE_READ
, RADEON_DOMAIN_GTT
);
1141 set_reg(dec
, RUVD_ENGINE_CNTL
, 1);
1148 * flush any outstanding command buffers to the hardware
1150 static void ruvd_flush(struct pipe_video_codec
*decoder
)
1155 * create and UVD decoder
1157 struct pipe_video_codec
*ruvd_create_decoder(struct pipe_context
*context
,
1158 const struct pipe_video_codec
*templ
,
1159 ruvd_set_dtb set_dtb
)
1161 struct radeon_winsys
* ws
= ((struct r600_common_context
*)context
)->ws
;
1162 struct r600_common_context
*rctx
= (struct r600_common_context
*)context
;
1164 unsigned width
= templ
->width
, height
= templ
->height
;
1165 unsigned bs_buf_size
;
1166 struct radeon_info info
;
1167 struct ruvd_decoder
*dec
;
1170 ws
->query_info(ws
, &info
);
1172 switch(u_reduce_video_profile(templ
->profile
)) {
1173 case PIPE_VIDEO_FORMAT_MPEG12
:
1174 if (templ
->entrypoint
> PIPE_VIDEO_ENTRYPOINT_BITSTREAM
|| info
.family
< CHIP_PALM
)
1175 return vl_create_mpeg12_decoder(context
, templ
);
1178 case PIPE_VIDEO_FORMAT_MPEG4
:
1179 width
= align(width
, VL_MACROBLOCK_WIDTH
);
1180 height
= align(height
, VL_MACROBLOCK_HEIGHT
);
1182 case PIPE_VIDEO_FORMAT_MPEG4_AVC
:
1183 width
= align(width
, VL_MACROBLOCK_WIDTH
);
1184 height
= align(height
, VL_MACROBLOCK_HEIGHT
);
1192 dec
= CALLOC_STRUCT(ruvd_decoder
);
1197 if (info
.drm_major
< 3)
1198 dec
->use_legacy
= true;
1201 dec
->base
.context
= context
;
1202 dec
->base
.width
= width
;
1203 dec
->base
.height
= height
;
1205 dec
->base
.destroy
= ruvd_destroy
;
1206 dec
->base
.begin_frame
= ruvd_begin_frame
;
1207 dec
->base
.decode_macroblock
= ruvd_decode_macroblock
;
1208 dec
->base
.decode_bitstream
= ruvd_decode_bitstream
;
1209 dec
->base
.end_frame
= ruvd_end_frame
;
1210 dec
->base
.flush
= ruvd_flush
;
1212 dec
->stream_type
= profile2stream_type(dec
, info
.family
);
1213 dec
->set_dtb
= set_dtb
;
1214 dec
->stream_handle
= rvid_alloc_stream_handle();
1215 dec
->screen
= context
->screen
;
1217 dec
->cs
= ws
->cs_create(rctx
->ctx
, RING_UVD
, NULL
, NULL
);
1219 RVID_ERR("Can't get command submission context.\n");
1223 dec
->fb_size
= (info
.family
== CHIP_TONGA
) ? FB_BUFFER_SIZE_TONGA
:
1225 bs_buf_size
= width
* height
* (512 / (16 * 16));
1226 for (i
= 0; i
< NUM_BUFFERS
; ++i
) {
1227 unsigned msg_fb_it_size
= FB_BUFFER_OFFSET
+ dec
->fb_size
;
1228 STATIC_ASSERT(sizeof(struct ruvd_msg
) <= FB_BUFFER_OFFSET
);
1230 msg_fb_it_size
+= IT_SCALING_TABLE_SIZE
;
1231 if (!rvid_create_buffer(dec
->screen
, &dec
->msg_fb_it_buffers
[i
],
1232 msg_fb_it_size
, PIPE_USAGE_STAGING
)) {
1233 RVID_ERR("Can't allocated message buffers.\n");
1237 if (!rvid_create_buffer(dec
->screen
, &dec
->bs_buffers
[i
],
1238 bs_buf_size
, PIPE_USAGE_STAGING
)) {
1239 RVID_ERR("Can't allocated bitstream buffers.\n");
1243 rvid_clear_buffer(context
, &dec
->msg_fb_it_buffers
[i
]);
1244 rvid_clear_buffer(context
, &dec
->bs_buffers
[i
]);
1247 dpb_size
= calc_dpb_size(dec
);
1249 if (!rvid_create_buffer(dec
->screen
, &dec
->dpb
, dpb_size
, PIPE_USAGE_DEFAULT
)) {
1250 RVID_ERR("Can't allocated dpb.\n");
1254 rvid_clear_buffer(context
, &dec
->dpb
);
1256 if (dec
->stream_type
== RUVD_CODEC_H264_PERF
&& info
.family
>= CHIP_POLARIS10
) {
1257 unsigned ctx_size
= calc_ctx_size_h264_perf(dec
);
1258 if (!rvid_create_buffer(dec
->screen
, &dec
->ctx
, ctx_size
, PIPE_USAGE_DEFAULT
)) {
1259 RVID_ERR("Can't allocated context buffer.\n");
1262 rvid_clear_buffer(context
, &dec
->ctx
);
1265 map_msg_fb_it_buf(dec
);
1266 dec
->msg
->size
= sizeof(*dec
->msg
);
1267 dec
->msg
->msg_type
= RUVD_MSG_CREATE
;
1268 dec
->msg
->stream_handle
= dec
->stream_handle
;
1269 dec
->msg
->body
.create
.stream_type
= dec
->stream_type
;
1270 dec
->msg
->body
.create
.width_in_samples
= dec
->base
.width
;
1271 dec
->msg
->body
.create
.height_in_samples
= dec
->base
.height
;
1272 dec
->msg
->body
.create
.dpb_size
= dpb_size
;
1280 if (dec
->cs
) dec
->ws
->cs_destroy(dec
->cs
);
1282 for (i
= 0; i
< NUM_BUFFERS
; ++i
) {
1283 rvid_destroy_buffer(&dec
->msg_fb_it_buffers
[i
]);
1284 rvid_destroy_buffer(&dec
->bs_buffers
[i
]);
1287 rvid_destroy_buffer(&dec
->dpb
);
1288 rvid_destroy_buffer(&dec
->ctx
);
1295 /* calculate top/bottom offset */
1296 static unsigned texture_offset(struct radeon_surf
*surface
, unsigned layer
)
1298 return surface
->level
[0].offset
+
1299 layer
* surface
->level
[0].slice_size
;
1302 /* hw encode the aspect of macro tiles */
1303 static unsigned macro_tile_aspect(unsigned macro_tile_aspect
)
1305 switch (macro_tile_aspect
) {
1307 case 1: macro_tile_aspect
= 0; break;
1308 case 2: macro_tile_aspect
= 1; break;
1309 case 4: macro_tile_aspect
= 2; break;
1310 case 8: macro_tile_aspect
= 3; break;
1312 return macro_tile_aspect
;
1315 /* hw encode the bank width and height */
1316 static unsigned bank_wh(unsigned bankwh
)
1320 case 1: bankwh
= 0; break;
1321 case 2: bankwh
= 1; break;
1322 case 4: bankwh
= 2; break;
1323 case 8: bankwh
= 3; break;
1329 * fill decoding target field from the luma and chroma surfaces
1331 void ruvd_set_dt_surfaces(struct ruvd_msg
*msg
, struct radeon_surf
*luma
,
1332 struct radeon_surf
*chroma
)
1334 msg
->body
.decode
.dt_pitch
= luma
->level
[0].pitch_bytes
;
1335 switch (luma
->level
[0].mode
) {
1336 case RADEON_SURF_MODE_LINEAR_ALIGNED
:
1337 msg
->body
.decode
.dt_tiling_mode
= RUVD_TILE_LINEAR
;
1338 msg
->body
.decode
.dt_array_mode
= RUVD_ARRAY_MODE_LINEAR
;
1340 case RADEON_SURF_MODE_1D
:
1341 msg
->body
.decode
.dt_tiling_mode
= RUVD_TILE_8X8
;
1342 msg
->body
.decode
.dt_array_mode
= RUVD_ARRAY_MODE_1D_THIN
;
1344 case RADEON_SURF_MODE_2D
:
1345 msg
->body
.decode
.dt_tiling_mode
= RUVD_TILE_8X8
;
1346 msg
->body
.decode
.dt_array_mode
= RUVD_ARRAY_MODE_2D_THIN
;
1353 msg
->body
.decode
.dt_luma_top_offset
= texture_offset(luma
, 0);
1354 msg
->body
.decode
.dt_chroma_top_offset
= texture_offset(chroma
, 0);
1355 if (msg
->body
.decode
.dt_field_mode
) {
1356 msg
->body
.decode
.dt_luma_bottom_offset
= texture_offset(luma
, 1);
1357 msg
->body
.decode
.dt_chroma_bottom_offset
= texture_offset(chroma
, 1);
1359 msg
->body
.decode
.dt_luma_bottom_offset
= msg
->body
.decode
.dt_luma_top_offset
;
1360 msg
->body
.decode
.dt_chroma_bottom_offset
= msg
->body
.decode
.dt_chroma_top_offset
;
1363 assert(luma
->bankw
== chroma
->bankw
);
1364 assert(luma
->bankh
== chroma
->bankh
);
1365 assert(luma
->mtilea
== chroma
->mtilea
);
1367 msg
->body
.decode
.dt_surf_tile_config
|= RUVD_BANK_WIDTH(bank_wh(luma
->bankw
));
1368 msg
->body
.decode
.dt_surf_tile_config
|= RUVD_BANK_HEIGHT(bank_wh(luma
->bankh
));
1369 msg
->body
.decode
.dt_surf_tile_config
|= RUVD_MACRO_TILE_ASPECT_RATIO(macro_tile_aspect(luma
->mtilea
));