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
, domain
,
118 if (!dec
->use_legacy
) {
120 addr
= dec
->ws
->buffer_get_virtual_address(buf
);
122 set_reg(dec
, RUVD_GPCOM_VCPU_DATA0
, addr
);
123 set_reg(dec
, RUVD_GPCOM_VCPU_DATA1
, addr
>> 32);
125 set_reg(dec
, RUVD_GPCOM_VCPU_DATA0
, off
);
126 set_reg(dec
, RUVD_GPCOM_VCPU_DATA1
, reloc_idx
* 4);
128 set_reg(dec
, RUVD_GPCOM_VCPU_CMD
, cmd
<< 1);
131 /* do the codec needs an IT buffer ?*/
132 static bool have_it(struct ruvd_decoder
*dec
)
134 return dec
->stream_type
== RUVD_CODEC_H264_PERF
||
135 dec
->stream_type
== RUVD_CODEC_H265
;
138 /* map the next available message/feedback/itscaling buffer */
139 static void map_msg_fb_it_buf(struct ruvd_decoder
*dec
)
141 struct rvid_buffer
* buf
;
144 /* grab the current message/feedback buffer */
145 buf
= &dec
->msg_fb_it_buffers
[dec
->cur_buffer
];
147 /* and map it for CPU access */
148 ptr
= dec
->ws
->buffer_map(buf
->res
->buf
, dec
->cs
, PIPE_TRANSFER_WRITE
);
150 /* calc buffer offsets */
151 dec
->msg
= (struct ruvd_msg
*)ptr
;
152 dec
->fb
= (uint32_t *)(ptr
+ FB_BUFFER_OFFSET
);
154 dec
->it
= (uint8_t *)(ptr
+ FB_BUFFER_OFFSET
+ dec
->fb_size
);
157 /* unmap and send a message command to the VCPU */
158 static void send_msg_buf(struct ruvd_decoder
*dec
)
160 struct rvid_buffer
* buf
;
162 /* ignore the request if message/feedback buffer isn't mapped */
163 if (!dec
->msg
|| !dec
->fb
)
166 /* grab the current message buffer */
167 buf
= &dec
->msg_fb_it_buffers
[dec
->cur_buffer
];
169 /* unmap the buffer */
170 dec
->ws
->buffer_unmap(buf
->res
->buf
);
176 if (dec
->sessionctx
.res
)
177 send_cmd(dec
, RUVD_CMD_SESSION_CONTEXT_BUFFER
,
178 dec
->sessionctx
.res
->buf
, 0, RADEON_USAGE_READWRITE
,
181 /* and send it to the hardware */
182 send_cmd(dec
, RUVD_CMD_MSG_BUFFER
, buf
->res
->buf
, 0,
183 RADEON_USAGE_READ
, RADEON_DOMAIN_GTT
);
186 /* cycle to the next set of buffers */
187 static void next_buffer(struct ruvd_decoder
*dec
)
190 dec
->cur_buffer
%= NUM_BUFFERS
;
193 /* convert the profile into something UVD understands */
194 static uint32_t profile2stream_type(struct ruvd_decoder
*dec
, unsigned family
)
196 switch (u_reduce_video_profile(dec
->base
.profile
)) {
197 case PIPE_VIDEO_FORMAT_MPEG4_AVC
:
198 return (family
>= CHIP_TONGA
) ?
199 RUVD_CODEC_H264_PERF
: RUVD_CODEC_H264
;
201 case PIPE_VIDEO_FORMAT_VC1
:
202 return RUVD_CODEC_VC1
;
204 case PIPE_VIDEO_FORMAT_MPEG12
:
205 return RUVD_CODEC_MPEG2
;
207 case PIPE_VIDEO_FORMAT_MPEG4
:
208 return RUVD_CODEC_MPEG4
;
210 case PIPE_VIDEO_FORMAT_HEVC
:
211 return RUVD_CODEC_H265
;
219 static unsigned calc_ctx_size_h264_perf(struct ruvd_decoder
*dec
)
221 unsigned width_in_mb
, height_in_mb
, ctx_size
;
222 unsigned width
= align(dec
->base
.width
, VL_MACROBLOCK_WIDTH
);
223 unsigned height
= align(dec
->base
.height
, VL_MACROBLOCK_HEIGHT
);
225 unsigned max_references
= dec
->base
.max_references
+ 1;
227 // picture width & height in 16 pixel units
228 width_in_mb
= width
/ VL_MACROBLOCK_WIDTH
;
229 height_in_mb
= align(height
/ VL_MACROBLOCK_HEIGHT
, 2);
231 if (!dec
->use_legacy
) {
232 unsigned fs_in_mb
= width_in_mb
* height_in_mb
;
233 unsigned num_dpb_buffer
;
234 switch(dec
->base
.level
) {
236 num_dpb_buffer
= 8100 / fs_in_mb
;
239 num_dpb_buffer
= 18000 / fs_in_mb
;
242 num_dpb_buffer
= 20480 / fs_in_mb
;
245 num_dpb_buffer
= 32768 / fs_in_mb
;
248 num_dpb_buffer
= 34816 / fs_in_mb
;
251 num_dpb_buffer
= 110400 / fs_in_mb
;
254 num_dpb_buffer
= 184320 / fs_in_mb
;
257 num_dpb_buffer
= 184320 / fs_in_mb
;
261 max_references
= MAX2(MIN2(NUM_H264_REFS
, num_dpb_buffer
), max_references
);
262 ctx_size
= max_references
* align(width_in_mb
* height_in_mb
* 192, 256);
264 // the firmware seems to always assume a minimum of ref frames
265 max_references
= MAX2(NUM_H264_REFS
, max_references
);
266 // macroblock context buffer
267 ctx_size
= align(width_in_mb
* height_in_mb
* max_references
* 192, 256);
273 static unsigned calc_ctx_size_h265_main(struct ruvd_decoder
*dec
)
275 unsigned width
= align(dec
->base
.width
, VL_MACROBLOCK_WIDTH
);
276 unsigned height
= align(dec
->base
.height
, VL_MACROBLOCK_HEIGHT
);
278 unsigned max_references
= dec
->base
.max_references
+ 1;
280 if (dec
->base
.width
* dec
->base
.height
>= 4096*2000)
281 max_references
= MAX2(max_references
, 8);
283 max_references
= MAX2(max_references
, 17);
285 width
= align (width
, 16);
286 height
= align (height
, 16);
287 return ((width
+ 255) / 16) * ((height
+ 255) / 16) * 16 * max_references
+ 52 * 1024;
290 static unsigned calc_ctx_size_h265_main10(struct ruvd_decoder
*dec
, struct pipe_h265_picture_desc
*pic
)
292 unsigned block_size
, log2_ctb_size
, width_in_ctb
, height_in_ctb
, num_16x16_block_per_ctb
;
293 unsigned context_buffer_size_per_ctb_row
, cm_buffer_size
, max_mb_address
, db_left_tile_pxl_size
;
294 unsigned db_left_tile_ctx_size
= 4096 / 16 * (32 + 16 * 4);
296 unsigned width
= align(dec
->base
.width
, VL_MACROBLOCK_WIDTH
);
297 unsigned height
= align(dec
->base
.height
, VL_MACROBLOCK_HEIGHT
);
298 unsigned coeff_10bit
= (pic
->pps
->sps
->bit_depth_luma_minus8
|| pic
->pps
->sps
->bit_depth_chroma_minus8
) ? 2 : 1;
300 unsigned max_references
= dec
->base
.max_references
+ 1;
302 if (dec
->base
.width
* dec
->base
.height
>= 4096*2000)
303 max_references
= MAX2(max_references
, 8);
305 max_references
= MAX2(max_references
, 17);
307 block_size
= (1 << (pic
->pps
->sps
->log2_min_luma_coding_block_size_minus3
+ 3));
308 log2_ctb_size
= block_size
+ pic
->pps
->sps
->log2_diff_max_min_luma_coding_block_size
;
310 width_in_ctb
= (width
+ ((1 << log2_ctb_size
) - 1)) >> log2_ctb_size
;
311 height_in_ctb
= (height
+ ((1 << log2_ctb_size
) - 1)) >> log2_ctb_size
;
313 num_16x16_block_per_ctb
= ((1 << log2_ctb_size
) >> 4) * ((1 << log2_ctb_size
) >> 4);
314 context_buffer_size_per_ctb_row
= align(width_in_ctb
* num_16x16_block_per_ctb
* 16, 256);
315 max_mb_address
= (unsigned) ceil(height
* 8 / 2048.0);
317 cm_buffer_size
= max_references
* context_buffer_size_per_ctb_row
* height_in_ctb
;
318 db_left_tile_pxl_size
= coeff_10bit
* (max_mb_address
* 2 * 2048 + 1024);
320 return cm_buffer_size
+ db_left_tile_ctx_size
+ db_left_tile_pxl_size
;
323 /* calculate size of reference picture buffer */
324 static unsigned calc_dpb_size(struct ruvd_decoder
*dec
)
326 unsigned width_in_mb
, height_in_mb
, image_size
, dpb_size
;
328 // always align them to MB size for dpb calculation
329 unsigned width
= align(dec
->base
.width
, VL_MACROBLOCK_WIDTH
);
330 unsigned height
= align(dec
->base
.height
, VL_MACROBLOCK_HEIGHT
);
332 // always one more for currently decoded picture
333 unsigned max_references
= dec
->base
.max_references
+ 1;
335 // aligned size of a single frame
336 image_size
= width
* height
;
337 image_size
+= image_size
/ 2;
338 image_size
= align(image_size
, 1024);
340 // picture width & height in 16 pixel units
341 width_in_mb
= width
/ VL_MACROBLOCK_WIDTH
;
342 height_in_mb
= align(height
/ VL_MACROBLOCK_HEIGHT
, 2);
344 switch (u_reduce_video_profile(dec
->base
.profile
)) {
345 case PIPE_VIDEO_FORMAT_MPEG4_AVC
: {
346 if (!dec
->use_legacy
) {
347 unsigned fs_in_mb
= width_in_mb
* height_in_mb
;
348 unsigned alignment
= 64, num_dpb_buffer
;
350 if (dec
->stream_type
== RUVD_CODEC_H264_PERF
)
352 switch(dec
->base
.level
) {
354 num_dpb_buffer
= 8100 / fs_in_mb
;
357 num_dpb_buffer
= 18000 / fs_in_mb
;
360 num_dpb_buffer
= 20480 / fs_in_mb
;
363 num_dpb_buffer
= 32768 / fs_in_mb
;
366 num_dpb_buffer
= 34816 / fs_in_mb
;
369 num_dpb_buffer
= 110400 / fs_in_mb
;
372 num_dpb_buffer
= 184320 / fs_in_mb
;
375 num_dpb_buffer
= 184320 / fs_in_mb
;
379 max_references
= MAX2(MIN2(NUM_H264_REFS
, num_dpb_buffer
), max_references
);
380 dpb_size
= image_size
* max_references
;
381 if ((dec
->stream_type
!= RUVD_CODEC_H264_PERF
) ||
382 (((struct r600_common_screen
*)dec
->screen
)->family
< CHIP_POLARIS10
)) {
383 dpb_size
+= max_references
* align(width_in_mb
* height_in_mb
* 192, alignment
);
384 dpb_size
+= align(width_in_mb
* height_in_mb
* 32, alignment
);
387 // the firmware seems to allways assume a minimum of ref frames
388 max_references
= MAX2(NUM_H264_REFS
, max_references
);
389 // reference picture buffer
390 dpb_size
= image_size
* max_references
;
391 if ((dec
->stream_type
!= RUVD_CODEC_H264_PERF
) ||
392 (((struct r600_common_screen
*)dec
->screen
)->family
< CHIP_POLARIS10
)) {
393 // macroblock context buffer
394 dpb_size
+= width_in_mb
* height_in_mb
* max_references
* 192;
396 dpb_size
+= width_in_mb
* height_in_mb
* 32;
402 case PIPE_VIDEO_FORMAT_HEVC
:
403 if (dec
->base
.width
* dec
->base
.height
>= 4096*2000)
404 max_references
= MAX2(max_references
, 8);
406 max_references
= MAX2(max_references
, 17);
408 width
= align (width
, 16);
409 height
= align (height
, 16);
410 if (dec
->base
.profile
== PIPE_VIDEO_PROFILE_HEVC_MAIN_10
)
411 dpb_size
= align((width
* height
* 9) / 4, 256) * max_references
;
413 dpb_size
= align((width
* height
* 3) / 2, 256) * max_references
;
416 case PIPE_VIDEO_FORMAT_VC1
:
417 // the firmware seems to allways assume a minimum of ref frames
418 max_references
= MAX2(NUM_VC1_REFS
, max_references
);
420 // reference picture buffer
421 dpb_size
= image_size
* max_references
;
424 dpb_size
+= width_in_mb
* height_in_mb
* 128;
427 dpb_size
+= width_in_mb
* 64;
430 dpb_size
+= width_in_mb
* 128;
433 dpb_size
+= align(MAX2(width_in_mb
, height_in_mb
) * 7 * 16, 64);
436 case PIPE_VIDEO_FORMAT_MPEG12
:
437 // reference picture buffer, must be big enough for all frames
438 dpb_size
= image_size
* NUM_MPEG2_REFS
;
441 case PIPE_VIDEO_FORMAT_MPEG4
:
442 // reference picture buffer
443 dpb_size
= image_size
* max_references
;
446 dpb_size
+= width_in_mb
* height_in_mb
* 64;
449 dpb_size
+= align(width_in_mb
* height_in_mb
* 32, 64);
451 dpb_size
= MAX2(dpb_size
, 30 * 1024 * 1024);
455 // something is missing here
458 // at least use a sane default value
459 dpb_size
= 32 * 1024 * 1024;
465 /* free associated data in the video buffer callback */
466 static void ruvd_destroy_associated_data(void *data
)
468 /* NOOP, since we only use an intptr */
471 /* get h264 specific message bits */
472 static struct ruvd_h264
get_h264_msg(struct ruvd_decoder
*dec
, struct pipe_h264_picture_desc
*pic
)
474 struct ruvd_h264 result
;
476 memset(&result
, 0, sizeof(result
));
477 switch (pic
->base
.profile
) {
478 case PIPE_VIDEO_PROFILE_MPEG4_AVC_BASELINE
:
479 result
.profile
= RUVD_H264_PROFILE_BASELINE
;
482 case PIPE_VIDEO_PROFILE_MPEG4_AVC_MAIN
:
483 result
.profile
= RUVD_H264_PROFILE_MAIN
;
486 case PIPE_VIDEO_PROFILE_MPEG4_AVC_HIGH
:
487 result
.profile
= RUVD_H264_PROFILE_HIGH
;
495 result
.level
= dec
->base
.level
;
497 result
.sps_info_flags
= 0;
498 result
.sps_info_flags
|= pic
->pps
->sps
->direct_8x8_inference_flag
<< 0;
499 result
.sps_info_flags
|= pic
->pps
->sps
->mb_adaptive_frame_field_flag
<< 1;
500 result
.sps_info_flags
|= pic
->pps
->sps
->frame_mbs_only_flag
<< 2;
501 result
.sps_info_flags
|= pic
->pps
->sps
->delta_pic_order_always_zero_flag
<< 3;
503 result
.bit_depth_luma_minus8
= pic
->pps
->sps
->bit_depth_luma_minus8
;
504 result
.bit_depth_chroma_minus8
= pic
->pps
->sps
->bit_depth_chroma_minus8
;
505 result
.log2_max_frame_num_minus4
= pic
->pps
->sps
->log2_max_frame_num_minus4
;
506 result
.pic_order_cnt_type
= pic
->pps
->sps
->pic_order_cnt_type
;
507 result
.log2_max_pic_order_cnt_lsb_minus4
= pic
->pps
->sps
->log2_max_pic_order_cnt_lsb_minus4
;
509 switch (dec
->base
.chroma_format
) {
510 case PIPE_VIDEO_CHROMA_FORMAT_NONE
:
513 case PIPE_VIDEO_CHROMA_FORMAT_400
:
514 result
.chroma_format
= 0;
516 case PIPE_VIDEO_CHROMA_FORMAT_420
:
517 result
.chroma_format
= 1;
519 case PIPE_VIDEO_CHROMA_FORMAT_422
:
520 result
.chroma_format
= 2;
522 case PIPE_VIDEO_CHROMA_FORMAT_444
:
523 result
.chroma_format
= 3;
527 result
.pps_info_flags
= 0;
528 result
.pps_info_flags
|= pic
->pps
->transform_8x8_mode_flag
<< 0;
529 result
.pps_info_flags
|= pic
->pps
->redundant_pic_cnt_present_flag
<< 1;
530 result
.pps_info_flags
|= pic
->pps
->constrained_intra_pred_flag
<< 2;
531 result
.pps_info_flags
|= pic
->pps
->deblocking_filter_control_present_flag
<< 3;
532 result
.pps_info_flags
|= pic
->pps
->weighted_bipred_idc
<< 4;
533 result
.pps_info_flags
|= pic
->pps
->weighted_pred_flag
<< 6;
534 result
.pps_info_flags
|= pic
->pps
->bottom_field_pic_order_in_frame_present_flag
<< 7;
535 result
.pps_info_flags
|= pic
->pps
->entropy_coding_mode_flag
<< 8;
537 result
.num_slice_groups_minus1
= pic
->pps
->num_slice_groups_minus1
;
538 result
.slice_group_map_type
= pic
->pps
->slice_group_map_type
;
539 result
.slice_group_change_rate_minus1
= pic
->pps
->slice_group_change_rate_minus1
;
540 result
.pic_init_qp_minus26
= pic
->pps
->pic_init_qp_minus26
;
541 result
.chroma_qp_index_offset
= pic
->pps
->chroma_qp_index_offset
;
542 result
.second_chroma_qp_index_offset
= pic
->pps
->second_chroma_qp_index_offset
;
544 memcpy(result
.scaling_list_4x4
, pic
->pps
->ScalingList4x4
, 6*16);
545 memcpy(result
.scaling_list_8x8
, pic
->pps
->ScalingList8x8
, 2*64);
547 if (dec
->stream_type
== RUVD_CODEC_H264_PERF
) {
548 memcpy(dec
->it
, result
.scaling_list_4x4
, 6*16);
549 memcpy((dec
->it
+ 96), result
.scaling_list_8x8
, 2*64);
552 result
.num_ref_frames
= pic
->num_ref_frames
;
554 result
.num_ref_idx_l0_active_minus1
= pic
->num_ref_idx_l0_active_minus1
;
555 result
.num_ref_idx_l1_active_minus1
= pic
->num_ref_idx_l1_active_minus1
;
557 result
.frame_num
= pic
->frame_num
;
558 memcpy(result
.frame_num_list
, pic
->frame_num_list
, 4*16);
559 result
.curr_field_order_cnt_list
[0] = pic
->field_order_cnt
[0];
560 result
.curr_field_order_cnt_list
[1] = pic
->field_order_cnt
[1];
561 memcpy(result
.field_order_cnt_list
, pic
->field_order_cnt_list
, 4*16*2);
563 result
.decoded_pic_idx
= pic
->frame_num
;
568 /* get h265 specific message bits */
569 static struct ruvd_h265
get_h265_msg(struct ruvd_decoder
*dec
, struct pipe_video_buffer
*target
,
570 struct pipe_h265_picture_desc
*pic
)
572 struct ruvd_h265 result
;
575 memset(&result
, 0, sizeof(result
));
577 result
.sps_info_flags
= 0;
578 result
.sps_info_flags
|= pic
->pps
->sps
->scaling_list_enabled_flag
<< 0;
579 result
.sps_info_flags
|= pic
->pps
->sps
->amp_enabled_flag
<< 1;
580 result
.sps_info_flags
|= pic
->pps
->sps
->sample_adaptive_offset_enabled_flag
<< 2;
581 result
.sps_info_flags
|= pic
->pps
->sps
->pcm_enabled_flag
<< 3;
582 result
.sps_info_flags
|= pic
->pps
->sps
->pcm_loop_filter_disabled_flag
<< 4;
583 result
.sps_info_flags
|= pic
->pps
->sps
->long_term_ref_pics_present_flag
<< 5;
584 result
.sps_info_flags
|= pic
->pps
->sps
->sps_temporal_mvp_enabled_flag
<< 6;
585 result
.sps_info_flags
|= pic
->pps
->sps
->strong_intra_smoothing_enabled_flag
<< 7;
586 result
.sps_info_flags
|= pic
->pps
->sps
->separate_colour_plane_flag
<< 8;
587 if (((struct r600_common_screen
*)dec
->screen
)->family
== CHIP_CARRIZO
)
588 result
.sps_info_flags
|= 1 << 9;
589 if (pic
->UseRefPicList
== true)
590 result
.sps_info_flags
|= 1 << 10;
592 result
.chroma_format
= pic
->pps
->sps
->chroma_format_idc
;
593 result
.bit_depth_luma_minus8
= pic
->pps
->sps
->bit_depth_luma_minus8
;
594 result
.bit_depth_chroma_minus8
= pic
->pps
->sps
->bit_depth_chroma_minus8
;
595 result
.log2_max_pic_order_cnt_lsb_minus4
= pic
->pps
->sps
->log2_max_pic_order_cnt_lsb_minus4
;
596 result
.sps_max_dec_pic_buffering_minus1
= pic
->pps
->sps
->sps_max_dec_pic_buffering_minus1
;
597 result
.log2_min_luma_coding_block_size_minus3
= pic
->pps
->sps
->log2_min_luma_coding_block_size_minus3
;
598 result
.log2_diff_max_min_luma_coding_block_size
= pic
->pps
->sps
->log2_diff_max_min_luma_coding_block_size
;
599 result
.log2_min_transform_block_size_minus2
= pic
->pps
->sps
->log2_min_transform_block_size_minus2
;
600 result
.log2_diff_max_min_transform_block_size
= pic
->pps
->sps
->log2_diff_max_min_transform_block_size
;
601 result
.max_transform_hierarchy_depth_inter
= pic
->pps
->sps
->max_transform_hierarchy_depth_inter
;
602 result
.max_transform_hierarchy_depth_intra
= pic
->pps
->sps
->max_transform_hierarchy_depth_intra
;
603 result
.pcm_sample_bit_depth_luma_minus1
= pic
->pps
->sps
->pcm_sample_bit_depth_luma_minus1
;
604 result
.pcm_sample_bit_depth_chroma_minus1
= pic
->pps
->sps
->pcm_sample_bit_depth_chroma_minus1
;
605 result
.log2_min_pcm_luma_coding_block_size_minus3
= pic
->pps
->sps
->log2_min_pcm_luma_coding_block_size_minus3
;
606 result
.log2_diff_max_min_pcm_luma_coding_block_size
= pic
->pps
->sps
->log2_diff_max_min_pcm_luma_coding_block_size
;
607 result
.num_short_term_ref_pic_sets
= pic
->pps
->sps
->num_short_term_ref_pic_sets
;
609 result
.pps_info_flags
= 0;
610 result
.pps_info_flags
|= pic
->pps
->dependent_slice_segments_enabled_flag
<< 0;
611 result
.pps_info_flags
|= pic
->pps
->output_flag_present_flag
<< 1;
612 result
.pps_info_flags
|= pic
->pps
->sign_data_hiding_enabled_flag
<< 2;
613 result
.pps_info_flags
|= pic
->pps
->cabac_init_present_flag
<< 3;
614 result
.pps_info_flags
|= pic
->pps
->constrained_intra_pred_flag
<< 4;
615 result
.pps_info_flags
|= pic
->pps
->transform_skip_enabled_flag
<< 5;
616 result
.pps_info_flags
|= pic
->pps
->cu_qp_delta_enabled_flag
<< 6;
617 result
.pps_info_flags
|= pic
->pps
->pps_slice_chroma_qp_offsets_present_flag
<< 7;
618 result
.pps_info_flags
|= pic
->pps
->weighted_pred_flag
<< 8;
619 result
.pps_info_flags
|= pic
->pps
->weighted_bipred_flag
<< 9;
620 result
.pps_info_flags
|= pic
->pps
->transquant_bypass_enabled_flag
<< 10;
621 result
.pps_info_flags
|= pic
->pps
->tiles_enabled_flag
<< 11;
622 result
.pps_info_flags
|= pic
->pps
->entropy_coding_sync_enabled_flag
<< 12;
623 result
.pps_info_flags
|= pic
->pps
->uniform_spacing_flag
<< 13;
624 result
.pps_info_flags
|= pic
->pps
->loop_filter_across_tiles_enabled_flag
<< 14;
625 result
.pps_info_flags
|= pic
->pps
->pps_loop_filter_across_slices_enabled_flag
<< 15;
626 result
.pps_info_flags
|= pic
->pps
->deblocking_filter_override_enabled_flag
<< 16;
627 result
.pps_info_flags
|= pic
->pps
->pps_deblocking_filter_disabled_flag
<< 17;
628 result
.pps_info_flags
|= pic
->pps
->lists_modification_present_flag
<< 18;
629 result
.pps_info_flags
|= pic
->pps
->slice_segment_header_extension_present_flag
<< 19;
630 //result.pps_info_flags |= pic->pps->deblocking_filter_control_present_flag; ???
632 result
.num_extra_slice_header_bits
= pic
->pps
->num_extra_slice_header_bits
;
633 result
.num_long_term_ref_pic_sps
= pic
->pps
->sps
->num_long_term_ref_pics_sps
;
634 result
.num_ref_idx_l0_default_active_minus1
= pic
->pps
->num_ref_idx_l0_default_active_minus1
;
635 result
.num_ref_idx_l1_default_active_minus1
= pic
->pps
->num_ref_idx_l1_default_active_minus1
;
636 result
.pps_cb_qp_offset
= pic
->pps
->pps_cb_qp_offset
;
637 result
.pps_cr_qp_offset
= pic
->pps
->pps_cr_qp_offset
;
638 result
.pps_beta_offset_div2
= pic
->pps
->pps_beta_offset_div2
;
639 result
.pps_tc_offset_div2
= pic
->pps
->pps_tc_offset_div2
;
640 result
.diff_cu_qp_delta_depth
= pic
->pps
->diff_cu_qp_delta_depth
;
641 result
.num_tile_columns_minus1
= pic
->pps
->num_tile_columns_minus1
;
642 result
.num_tile_rows_minus1
= pic
->pps
->num_tile_rows_minus1
;
643 result
.log2_parallel_merge_level_minus2
= pic
->pps
->log2_parallel_merge_level_minus2
;
644 result
.init_qp_minus26
= pic
->pps
->init_qp_minus26
;
646 for (i
= 0; i
< 19; ++i
)
647 result
.column_width_minus1
[i
] = pic
->pps
->column_width_minus1
[i
];
649 for (i
= 0; i
< 21; ++i
)
650 result
.row_height_minus1
[i
] = pic
->pps
->row_height_minus1
[i
];
652 result
.num_delta_pocs_ref_rps_idx
= pic
->NumDeltaPocsOfRefRpsIdx
;
653 result
.curr_idx
= pic
->CurrPicOrderCntVal
;
654 result
.curr_poc
= pic
->CurrPicOrderCntVal
;
656 vl_video_buffer_set_associated_data(target
, &dec
->base
,
657 (void *)(uintptr_t)pic
->CurrPicOrderCntVal
,
658 &ruvd_destroy_associated_data
);
660 for (i
= 0; i
< 16; ++i
) {
661 struct pipe_video_buffer
*ref
= pic
->ref
[i
];
662 uintptr_t ref_pic
= 0;
664 result
.poc_list
[i
] = pic
->PicOrderCntVal
[i
];
667 ref_pic
= (uintptr_t)vl_video_buffer_get_associated_data(ref
, &dec
->base
);
670 result
.ref_pic_list
[i
] = ref_pic
;
673 for (i
= 0; i
< 8; ++i
) {
674 result
.ref_pic_set_st_curr_before
[i
] = 0xFF;
675 result
.ref_pic_set_st_curr_after
[i
] = 0xFF;
676 result
.ref_pic_set_lt_curr
[i
] = 0xFF;
679 for (i
= 0; i
< pic
->NumPocStCurrBefore
; ++i
)
680 result
.ref_pic_set_st_curr_before
[i
] = pic
->RefPicSetStCurrBefore
[i
];
682 for (i
= 0; i
< pic
->NumPocStCurrAfter
; ++i
)
683 result
.ref_pic_set_st_curr_after
[i
] = pic
->RefPicSetStCurrAfter
[i
];
685 for (i
= 0; i
< pic
->NumPocLtCurr
; ++i
)
686 result
.ref_pic_set_lt_curr
[i
] = pic
->RefPicSetLtCurr
[i
];
688 for (i
= 0; i
< 6; ++i
)
689 result
.ucScalingListDCCoefSizeID2
[i
] = pic
->pps
->sps
->ScalingListDCCoeff16x16
[i
];
691 for (i
= 0; i
< 2; ++i
)
692 result
.ucScalingListDCCoefSizeID3
[i
] = pic
->pps
->sps
->ScalingListDCCoeff32x32
[i
];
694 memcpy(dec
->it
, pic
->pps
->sps
->ScalingList4x4
, 6 * 16);
695 memcpy(dec
->it
+ 96, pic
->pps
->sps
->ScalingList8x8
, 6 * 64);
696 memcpy(dec
->it
+ 480, pic
->pps
->sps
->ScalingList16x16
, 6 * 64);
697 memcpy(dec
->it
+ 864, pic
->pps
->sps
->ScalingList32x32
, 2 * 64);
699 for (i
= 0 ; i
< 2 ; i
++) {
700 for (int j
= 0 ; j
< 15 ; j
++)
701 result
.direct_reflist
[i
][j
] = pic
->RefPicList
[i
][j
];
704 if ((pic
->base
.profile
== PIPE_VIDEO_PROFILE_HEVC_MAIN_10
) &&
705 (target
->buffer_format
== PIPE_FORMAT_NV12
)) {
706 result
.p010_mode
= 0;
707 result
.luma_10to8
= 5;
708 result
.chroma_10to8
= 5;
709 result
.sclr_luma10to8
= 4;
710 result
.sclr_chroma10to8
= 4;
720 NumShortTermPictureSliceHeaderBits;
721 NumLongTermPictureSliceHeaderBits;
729 /* get vc1 specific message bits */
730 static struct ruvd_vc1
get_vc1_msg(struct pipe_vc1_picture_desc
*pic
)
732 struct ruvd_vc1 result
;
734 memset(&result
, 0, sizeof(result
));
736 switch(pic
->base
.profile
) {
737 case PIPE_VIDEO_PROFILE_VC1_SIMPLE
:
738 result
.profile
= RUVD_VC1_PROFILE_SIMPLE
;
742 case PIPE_VIDEO_PROFILE_VC1_MAIN
:
743 result
.profile
= RUVD_VC1_PROFILE_MAIN
;
747 case PIPE_VIDEO_PROFILE_VC1_ADVANCED
:
748 result
.profile
= RUVD_VC1_PROFILE_ADVANCED
;
756 /* fields common for all profiles */
757 result
.sps_info_flags
|= pic
->postprocflag
<< 7;
758 result
.sps_info_flags
|= pic
->pulldown
<< 6;
759 result
.sps_info_flags
|= pic
->interlace
<< 5;
760 result
.sps_info_flags
|= pic
->tfcntrflag
<< 4;
761 result
.sps_info_flags
|= pic
->finterpflag
<< 3;
762 result
.sps_info_flags
|= pic
->psf
<< 1;
764 result
.pps_info_flags
|= pic
->range_mapy_flag
<< 31;
765 result
.pps_info_flags
|= pic
->range_mapy
<< 28;
766 result
.pps_info_flags
|= pic
->range_mapuv_flag
<< 27;
767 result
.pps_info_flags
|= pic
->range_mapuv
<< 24;
768 result
.pps_info_flags
|= pic
->multires
<< 21;
769 result
.pps_info_flags
|= pic
->maxbframes
<< 16;
770 result
.pps_info_flags
|= pic
->overlap
<< 11;
771 result
.pps_info_flags
|= pic
->quantizer
<< 9;
772 result
.pps_info_flags
|= pic
->panscan_flag
<< 7;
773 result
.pps_info_flags
|= pic
->refdist_flag
<< 6;
774 result
.pps_info_flags
|= pic
->vstransform
<< 0;
776 /* some fields only apply to main/advanced profile */
777 if (pic
->base
.profile
!= PIPE_VIDEO_PROFILE_VC1_SIMPLE
) {
778 result
.pps_info_flags
|= pic
->syncmarker
<< 20;
779 result
.pps_info_flags
|= pic
->rangered
<< 19;
780 result
.pps_info_flags
|= pic
->loopfilter
<< 5;
781 result
.pps_info_flags
|= pic
->fastuvmc
<< 4;
782 result
.pps_info_flags
|= pic
->extended_mv
<< 3;
783 result
.pps_info_flags
|= pic
->extended_dmv
<< 8;
784 result
.pps_info_flags
|= pic
->dquant
<< 1;
787 result
.chroma_format
= 1;
790 //(((unsigned int)(pPicParams->advance.reserved1)) << SPS_INFO_VC1_RESERVED_SHIFT)
793 uint8_t frame_coding_mode
794 uint8_t deblockEnable
801 /* extract the frame number from a referenced video buffer */
802 static uint32_t get_ref_pic_idx(struct ruvd_decoder
*dec
, struct pipe_video_buffer
*ref
)
804 uint32_t min
= MAX2(dec
->frame_number
, NUM_MPEG2_REFS
) - NUM_MPEG2_REFS
;
805 uint32_t max
= MAX2(dec
->frame_number
, 1) - 1;
808 /* seems to be the most sane fallback */
812 /* get the frame number from the associated data */
813 frame
= (uintptr_t)vl_video_buffer_get_associated_data(ref
, &dec
->base
);
815 /* limit the frame number to a valid range */
816 return MAX2(MIN2(frame
, max
), min
);
819 /* get mpeg2 specific msg bits */
820 static struct ruvd_mpeg2
get_mpeg2_msg(struct ruvd_decoder
*dec
,
821 struct pipe_mpeg12_picture_desc
*pic
)
823 const int *zscan
= pic
->alternate_scan
? vl_zscan_alternate
: vl_zscan_normal
;
824 struct ruvd_mpeg2 result
;
827 memset(&result
, 0, sizeof(result
));
828 result
.decoded_pic_idx
= dec
->frame_number
;
829 for (i
= 0; i
< 2; ++i
)
830 result
.ref_pic_idx
[i
] = get_ref_pic_idx(dec
, pic
->ref
[i
]);
832 result
.load_intra_quantiser_matrix
= 1;
833 result
.load_nonintra_quantiser_matrix
= 1;
835 for (i
= 0; i
< 64; ++i
) {
836 result
.intra_quantiser_matrix
[i
] = pic
->intra_matrix
[zscan
[i
]];
837 result
.nonintra_quantiser_matrix
[i
] = pic
->non_intra_matrix
[zscan
[i
]];
840 result
.profile_and_level_indication
= 0;
841 result
.chroma_format
= 0x1;
843 result
.picture_coding_type
= pic
->picture_coding_type
;
844 result
.f_code
[0][0] = pic
->f_code
[0][0] + 1;
845 result
.f_code
[0][1] = pic
->f_code
[0][1] + 1;
846 result
.f_code
[1][0] = pic
->f_code
[1][0] + 1;
847 result
.f_code
[1][1] = pic
->f_code
[1][1] + 1;
848 result
.intra_dc_precision
= pic
->intra_dc_precision
;
849 result
.pic_structure
= pic
->picture_structure
;
850 result
.top_field_first
= pic
->top_field_first
;
851 result
.frame_pred_frame_dct
= pic
->frame_pred_frame_dct
;
852 result
.concealment_motion_vectors
= pic
->concealment_motion_vectors
;
853 result
.q_scale_type
= pic
->q_scale_type
;
854 result
.intra_vlc_format
= pic
->intra_vlc_format
;
855 result
.alternate_scan
= pic
->alternate_scan
;
860 /* get mpeg4 specific msg bits */
861 static struct ruvd_mpeg4
get_mpeg4_msg(struct ruvd_decoder
*dec
,
862 struct pipe_mpeg4_picture_desc
*pic
)
864 struct ruvd_mpeg4 result
;
867 memset(&result
, 0, sizeof(result
));
868 result
.decoded_pic_idx
= dec
->frame_number
;
869 for (i
= 0; i
< 2; ++i
)
870 result
.ref_pic_idx
[i
] = get_ref_pic_idx(dec
, pic
->ref
[i
]);
872 result
.variant_type
= 0;
873 result
.profile_and_level_indication
= 0xF0; // ASP Level0
875 result
.video_object_layer_verid
= 0x5; // advanced simple
876 result
.video_object_layer_shape
= 0x0; // rectangular
878 result
.video_object_layer_width
= dec
->base
.width
;
879 result
.video_object_layer_height
= dec
->base
.height
;
881 result
.vop_time_increment_resolution
= pic
->vop_time_increment_resolution
;
883 result
.flags
|= pic
->short_video_header
<< 0;
884 //result.flags |= obmc_disable << 1;
885 result
.flags
|= pic
->interlaced
<< 2;
886 result
.flags
|= 1 << 3; // load_intra_quant_mat
887 result
.flags
|= 1 << 4; // load_nonintra_quant_mat
888 result
.flags
|= pic
->quarter_sample
<< 5;
889 result
.flags
|= 1 << 6; // complexity_estimation_disable
890 result
.flags
|= pic
->resync_marker_disable
<< 7;
891 //result.flags |= data_partitioned << 8;
892 //result.flags |= reversible_vlc << 9;
893 result
.flags
|= 0 << 10; // newpred_enable
894 result
.flags
|= 0 << 11; // reduced_resolution_vop_enable
895 //result.flags |= scalability << 12;
896 //result.flags |= is_object_layer_identifier << 13;
897 //result.flags |= fixed_vop_rate << 14;
898 //result.flags |= newpred_segment_type << 15;
900 result
.quant_type
= pic
->quant_type
;
902 for (i
= 0; i
< 64; ++i
) {
903 result
.intra_quant_mat
[i
] = pic
->intra_matrix
[vl_zscan_normal
[i
]];
904 result
.nonintra_quant_mat
[i
] = pic
->non_intra_matrix
[vl_zscan_normal
[i
]];
910 uint8_t vop_coding_type
911 uint8_t vop_fcode_forward
912 uint8_t vop_fcode_backward
913 uint8_t rounding_control
914 uint8_t alternate_vertical_scan_flag
915 uint8_t top_field_first
922 * destroy this video decoder
924 static void ruvd_destroy(struct pipe_video_codec
*decoder
)
926 struct ruvd_decoder
*dec
= (struct ruvd_decoder
*)decoder
;
931 map_msg_fb_it_buf(dec
);
932 memset(dec
->msg
, 0, sizeof(*dec
->msg
));
933 dec
->msg
->size
= sizeof(*dec
->msg
);
934 dec
->msg
->msg_type
= RUVD_MSG_DESTROY
;
935 dec
->msg
->stream_handle
= dec
->stream_handle
;
940 dec
->ws
->cs_destroy(dec
->cs
);
942 for (i
= 0; i
< NUM_BUFFERS
; ++i
) {
943 rvid_destroy_buffer(&dec
->msg_fb_it_buffers
[i
]);
944 rvid_destroy_buffer(&dec
->bs_buffers
[i
]);
947 rvid_destroy_buffer(&dec
->dpb
);
948 rvid_destroy_buffer(&dec
->ctx
);
949 rvid_destroy_buffer(&dec
->sessionctx
);
955 * start decoding of a new frame
957 static void ruvd_begin_frame(struct pipe_video_codec
*decoder
,
958 struct pipe_video_buffer
*target
,
959 struct pipe_picture_desc
*picture
)
961 struct ruvd_decoder
*dec
= (struct ruvd_decoder
*)decoder
;
966 frame
= ++dec
->frame_number
;
967 vl_video_buffer_set_associated_data(target
, decoder
, (void *)frame
,
968 &ruvd_destroy_associated_data
);
971 dec
->bs_ptr
= dec
->ws
->buffer_map(
972 dec
->bs_buffers
[dec
->cur_buffer
].res
->buf
,
973 dec
->cs
, PIPE_TRANSFER_WRITE
);
977 * decode a macroblock
979 static void ruvd_decode_macroblock(struct pipe_video_codec
*decoder
,
980 struct pipe_video_buffer
*target
,
981 struct pipe_picture_desc
*picture
,
982 const struct pipe_macroblock
*macroblocks
,
983 unsigned num_macroblocks
)
985 /* not supported (yet) */
992 static void ruvd_decode_bitstream(struct pipe_video_codec
*decoder
,
993 struct pipe_video_buffer
*target
,
994 struct pipe_picture_desc
*picture
,
995 unsigned num_buffers
,
996 const void * const *buffers
,
997 const unsigned *sizes
)
999 struct ruvd_decoder
*dec
= (struct ruvd_decoder
*)decoder
;
1007 for (i
= 0; i
< num_buffers
; ++i
) {
1008 struct rvid_buffer
*buf
= &dec
->bs_buffers
[dec
->cur_buffer
];
1009 unsigned new_size
= dec
->bs_size
+ sizes
[i
];
1011 if (new_size
> buf
->res
->buf
->size
) {
1012 dec
->ws
->buffer_unmap(buf
->res
->buf
);
1013 if (!rvid_resize_buffer(dec
->screen
, dec
->cs
, buf
, new_size
)) {
1014 RVID_ERR("Can't resize bitstream buffer!");
1018 dec
->bs_ptr
= dec
->ws
->buffer_map(buf
->res
->buf
, dec
->cs
,
1019 PIPE_TRANSFER_WRITE
);
1023 dec
->bs_ptr
+= dec
->bs_size
;
1026 memcpy(dec
->bs_ptr
, buffers
[i
], sizes
[i
]);
1027 dec
->bs_size
+= sizes
[i
];
1028 dec
->bs_ptr
+= sizes
[i
];
1033 * end decoding of the current frame
1035 static void ruvd_end_frame(struct pipe_video_codec
*decoder
,
1036 struct pipe_video_buffer
*target
,
1037 struct pipe_picture_desc
*picture
)
1039 struct ruvd_decoder
*dec
= (struct ruvd_decoder
*)decoder
;
1040 struct pb_buffer
*dt
;
1041 struct rvid_buffer
*msg_fb_it_buf
, *bs_buf
;
1049 msg_fb_it_buf
= &dec
->msg_fb_it_buffers
[dec
->cur_buffer
];
1050 bs_buf
= &dec
->bs_buffers
[dec
->cur_buffer
];
1052 bs_size
= align(dec
->bs_size
, 128);
1053 memset(dec
->bs_ptr
, 0, bs_size
- dec
->bs_size
);
1054 dec
->ws
->buffer_unmap(bs_buf
->res
->buf
);
1056 map_msg_fb_it_buf(dec
);
1057 dec
->msg
->size
= sizeof(*dec
->msg
);
1058 dec
->msg
->msg_type
= RUVD_MSG_DECODE
;
1059 dec
->msg
->stream_handle
= dec
->stream_handle
;
1060 dec
->msg
->status_report_feedback_number
= dec
->frame_number
;
1062 dec
->msg
->body
.decode
.stream_type
= dec
->stream_type
;
1063 dec
->msg
->body
.decode
.decode_flags
= 0x1;
1064 dec
->msg
->body
.decode
.width_in_samples
= dec
->base
.width
;
1065 dec
->msg
->body
.decode
.height_in_samples
= dec
->base
.height
;
1067 if ((picture
->profile
== PIPE_VIDEO_PROFILE_VC1_SIMPLE
) ||
1068 (picture
->profile
== PIPE_VIDEO_PROFILE_VC1_MAIN
)) {
1069 dec
->msg
->body
.decode
.width_in_samples
= align(dec
->msg
->body
.decode
.width_in_samples
, 16) / 16;
1070 dec
->msg
->body
.decode
.height_in_samples
= align(dec
->msg
->body
.decode
.height_in_samples
, 16) / 16;
1073 dec
->msg
->body
.decode
.dpb_size
= dec
->dpb
.res
->buf
->size
;
1074 dec
->msg
->body
.decode
.bsd_size
= bs_size
;
1075 dec
->msg
->body
.decode
.db_pitch
= align(dec
->base
.width
, 16);
1077 if (dec
->stream_type
== RUVD_CODEC_H264_PERF
&&
1078 ((struct r600_common_screen
*)dec
->screen
)->family
>= CHIP_POLARIS10
)
1079 dec
->msg
->body
.decode
.dpb_reserved
= dec
->ctx
.res
->buf
->size
;
1081 dt
= dec
->set_dtb(dec
->msg
, (struct vl_video_buffer
*)target
);
1082 if (((struct r600_common_screen
*)dec
->screen
)->family
>= CHIP_STONEY
)
1083 dec
->msg
->body
.decode
.dt_wa_chroma_top_offset
= dec
->msg
->body
.decode
.dt_pitch
/ 2;
1085 switch (u_reduce_video_profile(picture
->profile
)) {
1086 case PIPE_VIDEO_FORMAT_MPEG4_AVC
:
1087 dec
->msg
->body
.decode
.codec
.h264
= get_h264_msg(dec
, (struct pipe_h264_picture_desc
*)picture
);
1090 case PIPE_VIDEO_FORMAT_HEVC
:
1091 dec
->msg
->body
.decode
.codec
.h265
= get_h265_msg(dec
, target
, (struct pipe_h265_picture_desc
*)picture
);
1092 if (dec
->ctx
.res
== NULL
) {
1094 if (dec
->base
.profile
== PIPE_VIDEO_PROFILE_HEVC_MAIN_10
)
1095 ctx_size
= calc_ctx_size_h265_main10(dec
, (struct pipe_h265_picture_desc
*)picture
);
1097 ctx_size
= calc_ctx_size_h265_main(dec
);
1098 if (!rvid_create_buffer(dec
->screen
, &dec
->ctx
, ctx_size
, PIPE_USAGE_DEFAULT
)) {
1099 RVID_ERR("Can't allocated context buffer.\n");
1101 rvid_clear_buffer(decoder
->context
, &dec
->ctx
);
1105 dec
->msg
->body
.decode
.dpb_reserved
= dec
->ctx
.res
->buf
->size
;
1108 case PIPE_VIDEO_FORMAT_VC1
:
1109 dec
->msg
->body
.decode
.codec
.vc1
= get_vc1_msg((struct pipe_vc1_picture_desc
*)picture
);
1112 case PIPE_VIDEO_FORMAT_MPEG12
:
1113 dec
->msg
->body
.decode
.codec
.mpeg2
= get_mpeg2_msg(dec
, (struct pipe_mpeg12_picture_desc
*)picture
);
1116 case PIPE_VIDEO_FORMAT_MPEG4
:
1117 dec
->msg
->body
.decode
.codec
.mpeg4
= get_mpeg4_msg(dec
, (struct pipe_mpeg4_picture_desc
*)picture
);
1125 dec
->msg
->body
.decode
.db_surf_tile_config
= dec
->msg
->body
.decode
.dt_surf_tile_config
;
1126 dec
->msg
->body
.decode
.extension_support
= 0x1;
1128 /* set at least the feedback buffer size */
1129 dec
->fb
[0] = dec
->fb_size
;
1133 send_cmd(dec
, RUVD_CMD_DPB_BUFFER
, dec
->dpb
.res
->buf
, 0,
1134 RADEON_USAGE_READWRITE
, RADEON_DOMAIN_VRAM
);
1136 send_cmd(dec
, RUVD_CMD_CONTEXT_BUFFER
, dec
->ctx
.res
->buf
, 0,
1137 RADEON_USAGE_READWRITE
, RADEON_DOMAIN_VRAM
);
1138 send_cmd(dec
, RUVD_CMD_BITSTREAM_BUFFER
, bs_buf
->res
->buf
,
1139 0, RADEON_USAGE_READ
, RADEON_DOMAIN_GTT
);
1140 send_cmd(dec
, RUVD_CMD_DECODING_TARGET_BUFFER
, dt
, 0,
1141 RADEON_USAGE_WRITE
, RADEON_DOMAIN_VRAM
);
1142 send_cmd(dec
, RUVD_CMD_FEEDBACK_BUFFER
, msg_fb_it_buf
->res
->buf
,
1143 FB_BUFFER_OFFSET
, RADEON_USAGE_WRITE
, RADEON_DOMAIN_GTT
);
1145 send_cmd(dec
, RUVD_CMD_ITSCALING_TABLE_BUFFER
, msg_fb_it_buf
->res
->buf
,
1146 FB_BUFFER_OFFSET
+ dec
->fb_size
, RADEON_USAGE_READ
, RADEON_DOMAIN_GTT
);
1147 set_reg(dec
, RUVD_ENGINE_CNTL
, 1);
1149 flush(dec
, RADEON_FLUSH_ASYNC
);
1154 * flush any outstanding command buffers to the hardware
1156 static void ruvd_flush(struct pipe_video_codec
*decoder
)
1161 * create and UVD decoder
1163 struct pipe_video_codec
*ruvd_create_decoder(struct pipe_context
*context
,
1164 const struct pipe_video_codec
*templ
,
1165 ruvd_set_dtb set_dtb
)
1167 struct radeon_winsys
* ws
= ((struct r600_common_context
*)context
)->ws
;
1168 struct r600_common_context
*rctx
= (struct r600_common_context
*)context
;
1170 unsigned width
= templ
->width
, height
= templ
->height
;
1171 unsigned bs_buf_size
;
1172 struct radeon_info info
;
1173 struct ruvd_decoder
*dec
;
1176 ws
->query_info(ws
, &info
);
1178 switch(u_reduce_video_profile(templ
->profile
)) {
1179 case PIPE_VIDEO_FORMAT_MPEG12
:
1180 if (templ
->entrypoint
> PIPE_VIDEO_ENTRYPOINT_BITSTREAM
|| info
.family
< CHIP_PALM
)
1181 return vl_create_mpeg12_decoder(context
, templ
);
1184 case PIPE_VIDEO_FORMAT_MPEG4
:
1185 width
= align(width
, VL_MACROBLOCK_WIDTH
);
1186 height
= align(height
, VL_MACROBLOCK_HEIGHT
);
1188 case PIPE_VIDEO_FORMAT_MPEG4_AVC
:
1189 width
= align(width
, VL_MACROBLOCK_WIDTH
);
1190 height
= align(height
, VL_MACROBLOCK_HEIGHT
);
1198 dec
= CALLOC_STRUCT(ruvd_decoder
);
1203 if (info
.drm_major
< 3)
1204 dec
->use_legacy
= true;
1207 dec
->base
.context
= context
;
1208 dec
->base
.width
= width
;
1209 dec
->base
.height
= height
;
1211 dec
->base
.destroy
= ruvd_destroy
;
1212 dec
->base
.begin_frame
= ruvd_begin_frame
;
1213 dec
->base
.decode_macroblock
= ruvd_decode_macroblock
;
1214 dec
->base
.decode_bitstream
= ruvd_decode_bitstream
;
1215 dec
->base
.end_frame
= ruvd_end_frame
;
1216 dec
->base
.flush
= ruvd_flush
;
1218 dec
->stream_type
= profile2stream_type(dec
, info
.family
);
1219 dec
->set_dtb
= set_dtb
;
1220 dec
->stream_handle
= rvid_alloc_stream_handle();
1221 dec
->screen
= context
->screen
;
1223 dec
->cs
= ws
->cs_create(rctx
->ctx
, RING_UVD
, NULL
, NULL
);
1225 RVID_ERR("Can't get command submission context.\n");
1229 dec
->fb_size
= (info
.family
== CHIP_TONGA
) ? FB_BUFFER_SIZE_TONGA
:
1231 bs_buf_size
= width
* height
* (512 / (16 * 16));
1232 for (i
= 0; i
< NUM_BUFFERS
; ++i
) {
1233 unsigned msg_fb_it_size
= FB_BUFFER_OFFSET
+ dec
->fb_size
;
1234 STATIC_ASSERT(sizeof(struct ruvd_msg
) <= FB_BUFFER_OFFSET
);
1236 msg_fb_it_size
+= IT_SCALING_TABLE_SIZE
;
1237 if (!rvid_create_buffer(dec
->screen
, &dec
->msg_fb_it_buffers
[i
],
1238 msg_fb_it_size
, PIPE_USAGE_STAGING
)) {
1239 RVID_ERR("Can't allocated message buffers.\n");
1243 if (!rvid_create_buffer(dec
->screen
, &dec
->bs_buffers
[i
],
1244 bs_buf_size
, PIPE_USAGE_STAGING
)) {
1245 RVID_ERR("Can't allocated bitstream buffers.\n");
1249 rvid_clear_buffer(context
, &dec
->msg_fb_it_buffers
[i
]);
1250 rvid_clear_buffer(context
, &dec
->bs_buffers
[i
]);
1253 dpb_size
= calc_dpb_size(dec
);
1255 if (!rvid_create_buffer(dec
->screen
, &dec
->dpb
, dpb_size
, PIPE_USAGE_DEFAULT
)) {
1256 RVID_ERR("Can't allocated dpb.\n");
1260 rvid_clear_buffer(context
, &dec
->dpb
);
1262 if (dec
->stream_type
== RUVD_CODEC_H264_PERF
&& info
.family
>= CHIP_POLARIS10
) {
1263 unsigned ctx_size
= calc_ctx_size_h264_perf(dec
);
1264 if (!rvid_create_buffer(dec
->screen
, &dec
->ctx
, ctx_size
, PIPE_USAGE_DEFAULT
)) {
1265 RVID_ERR("Can't allocated context buffer.\n");
1268 rvid_clear_buffer(context
, &dec
->ctx
);
1271 if (info
.family
>= CHIP_POLARIS10
&& info
.drm_minor
>= 3) {
1272 if (!rvid_create_buffer(dec
->screen
, &dec
->sessionctx
,
1273 UVD_SESSION_CONTEXT_SIZE
,
1274 PIPE_USAGE_DEFAULT
)) {
1275 RVID_ERR("Can't allocated session ctx.\n");
1278 rvid_clear_buffer(context
, &dec
->sessionctx
);
1281 map_msg_fb_it_buf(dec
);
1282 dec
->msg
->size
= sizeof(*dec
->msg
);
1283 dec
->msg
->msg_type
= RUVD_MSG_CREATE
;
1284 dec
->msg
->stream_handle
= dec
->stream_handle
;
1285 dec
->msg
->body
.create
.stream_type
= dec
->stream_type
;
1286 dec
->msg
->body
.create
.width_in_samples
= dec
->base
.width
;
1287 dec
->msg
->body
.create
.height_in_samples
= dec
->base
.height
;
1288 dec
->msg
->body
.create
.dpb_size
= dpb_size
;
1299 if (dec
->cs
) dec
->ws
->cs_destroy(dec
->cs
);
1301 for (i
= 0; i
< NUM_BUFFERS
; ++i
) {
1302 rvid_destroy_buffer(&dec
->msg_fb_it_buffers
[i
]);
1303 rvid_destroy_buffer(&dec
->bs_buffers
[i
]);
1306 rvid_destroy_buffer(&dec
->dpb
);
1307 rvid_destroy_buffer(&dec
->ctx
);
1308 rvid_destroy_buffer(&dec
->sessionctx
);
1315 /* calculate top/bottom offset */
1316 static unsigned texture_offset(struct radeon_surf
*surface
, unsigned layer
)
1318 return surface
->level
[0].offset
+
1319 layer
* surface
->level
[0].slice_size
;
1322 /* hw encode the aspect of macro tiles */
1323 static unsigned macro_tile_aspect(unsigned macro_tile_aspect
)
1325 switch (macro_tile_aspect
) {
1327 case 1: macro_tile_aspect
= 0; break;
1328 case 2: macro_tile_aspect
= 1; break;
1329 case 4: macro_tile_aspect
= 2; break;
1330 case 8: macro_tile_aspect
= 3; break;
1332 return macro_tile_aspect
;
1335 /* hw encode the bank width and height */
1336 static unsigned bank_wh(unsigned bankwh
)
1340 case 1: bankwh
= 0; break;
1341 case 2: bankwh
= 1; break;
1342 case 4: bankwh
= 2; break;
1343 case 8: bankwh
= 3; break;
1349 * fill decoding target field from the luma and chroma surfaces
1351 void ruvd_set_dt_surfaces(struct ruvd_msg
*msg
, struct radeon_surf
*luma
,
1352 struct radeon_surf
*chroma
)
1354 msg
->body
.decode
.dt_pitch
= luma
->level
[0].pitch_bytes
;
1355 switch (luma
->level
[0].mode
) {
1356 case RADEON_SURF_MODE_LINEAR_ALIGNED
:
1357 msg
->body
.decode
.dt_tiling_mode
= RUVD_TILE_LINEAR
;
1358 msg
->body
.decode
.dt_array_mode
= RUVD_ARRAY_MODE_LINEAR
;
1360 case RADEON_SURF_MODE_1D
:
1361 msg
->body
.decode
.dt_tiling_mode
= RUVD_TILE_8X8
;
1362 msg
->body
.decode
.dt_array_mode
= RUVD_ARRAY_MODE_1D_THIN
;
1364 case RADEON_SURF_MODE_2D
:
1365 msg
->body
.decode
.dt_tiling_mode
= RUVD_TILE_8X8
;
1366 msg
->body
.decode
.dt_array_mode
= RUVD_ARRAY_MODE_2D_THIN
;
1373 msg
->body
.decode
.dt_luma_top_offset
= texture_offset(luma
, 0);
1374 msg
->body
.decode
.dt_chroma_top_offset
= texture_offset(chroma
, 0);
1375 if (msg
->body
.decode
.dt_field_mode
) {
1376 msg
->body
.decode
.dt_luma_bottom_offset
= texture_offset(luma
, 1);
1377 msg
->body
.decode
.dt_chroma_bottom_offset
= texture_offset(chroma
, 1);
1379 msg
->body
.decode
.dt_luma_bottom_offset
= msg
->body
.decode
.dt_luma_top_offset
;
1380 msg
->body
.decode
.dt_chroma_bottom_offset
= msg
->body
.decode
.dt_chroma_top_offset
;
1383 assert(luma
->bankw
== chroma
->bankw
);
1384 assert(luma
->bankh
== chroma
->bankh
);
1385 assert(luma
->mtilea
== chroma
->mtilea
);
1387 msg
->body
.decode
.dt_surf_tile_config
|= RUVD_BANK_WIDTH(bank_wh(luma
->bankw
));
1388 msg
->body
.decode
.dt_surf_tile_config
|= RUVD_BANK_HEIGHT(bank_wh(luma
->bankh
));
1389 msg
->body
.decode
.dt_surf_tile_config
|= RUVD_MACRO_TILE_ASPECT_RATIO(macro_tile_aspect(luma
->mtilea
));