2 * Copyright © 2012 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the
6 * "Software"), to deal in the Software without restriction, including
7 * without limitation the rights to use, copy, modify, merge, publish,
8 * distribute, sub license, and/or sell copies of the Software, and to
9 * permit persons to whom the Software is furnished to do so, subject to
10 * the following conditions:
12 * The above copyright notice and this permission notice (including the
13 * next paragraph) shall be included in all copies or substantial portions
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
17 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
18 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
19 * IN NO EVENT SHALL PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR
20 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
21 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
22 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25 * Zhao Yakui <yakui.zhao@intel.com>
26 * Xiang Haihao <haihao.xiang@intel.com>
35 #include "intel_batchbuffer.h"
36 #include "intel_driver.h"
38 #include "i965_defines.h"
39 #include "i965_drv_video.h"
40 #include "i965_encoder.h"
44 #ifdef SURFACE_STATE_PADDED_SIZE
45 #undef SURFACE_STATE_PADDED_SIZE
48 #define SURFACE_STATE_PADDED_SIZE SURFACE_STATE_PADDED_SIZE_GEN8
49 #define SURFACE_STATE_OFFSET(index) (SURFACE_STATE_PADDED_SIZE * index)
50 #define BINDING_TABLE_OFFSET(index) (SURFACE_STATE_OFFSET(MAX_MEDIA_SURFACES_GEN6) + sizeof(unsigned int) * index)
52 #define VME_INTRA_SHADER 0
53 #define VME_INTER_SHADER 1
54 #define VME_BINTER_SHADER 2
56 #define CURBE_ALLOCATION_SIZE 37 /* in 256-bit */
57 #define CURBE_TOTAL_DATA_LENGTH (4 * 32) /* in byte, it should be less than or equal to CURBE_ALLOCATION_SIZE * 32 */
58 #define CURBE_URB_ENTRY_LENGTH 4 /* in 256-bit, it should be less than or equal to CURBE_TOTAL_DATA_LENGTH / 32 */
60 #define VME_MSG_LENGTH 32
62 static const uint32_t gen8_vme_intra_frame[][4] = {
63 #include "shaders/vme/intra_frame_gen8.g8b"
66 static const uint32_t gen8_vme_inter_frame[][4] = {
67 #include "shaders/vme/inter_frame_gen8.g8b"
70 static const uint32_t gen8_vme_inter_bframe[][4] = {
71 #include "shaders/vme/inter_bframe_gen8.g8b"
74 static struct i965_kernel gen8_vme_kernels[] = {
77 VME_INTRA_SHADER, /*index*/
79 sizeof(gen8_vme_intra_frame),
86 sizeof(gen8_vme_inter_frame),
92 gen8_vme_inter_bframe,
93 sizeof(gen8_vme_inter_bframe),
98 static const uint32_t gen8_vme_mpeg2_intra_frame[][4] = {
99 #include "shaders/vme/intra_frame_gen8.g8b"
102 static const uint32_t gen8_vme_mpeg2_inter_frame[][4] = {
103 #include "shaders/vme/mpeg2_inter_gen8.g8b"
106 static struct i965_kernel gen8_vme_mpeg2_kernels[] = {
109 VME_INTRA_SHADER, /*index*/
110 gen8_vme_mpeg2_intra_frame,
111 sizeof(gen8_vme_mpeg2_intra_frame),
117 gen8_vme_mpeg2_inter_frame,
118 sizeof(gen8_vme_mpeg2_inter_frame),
123 /* only used for VME source surface state */
125 gen8_vme_source_surface_state(VADriverContextP ctx,
127 struct object_surface *obj_surface,
128 struct intel_encoder_context *encoder_context)
130 struct gen6_vme_context *vme_context = encoder_context->vme_context;
132 vme_context->vme_surface2_setup(ctx,
133 &vme_context->gpe_context,
135 BINDING_TABLE_OFFSET(index),
136 SURFACE_STATE_OFFSET(index));
140 gen8_vme_media_source_surface_state(VADriverContextP ctx,
142 struct object_surface *obj_surface,
143 struct intel_encoder_context *encoder_context)
145 struct gen6_vme_context *vme_context = encoder_context->vme_context;
147 vme_context->vme_media_rw_surface_setup(ctx,
148 &vme_context->gpe_context,
150 BINDING_TABLE_OFFSET(index),
151 SURFACE_STATE_OFFSET(index));
155 gen8_vme_media_chroma_source_surface_state(VADriverContextP ctx,
157 struct object_surface *obj_surface,
158 struct intel_encoder_context *encoder_context)
160 struct gen6_vme_context *vme_context = encoder_context->vme_context;
162 vme_context->vme_media_chroma_surface_setup(ctx,
163 &vme_context->gpe_context,
165 BINDING_TABLE_OFFSET(index),
166 SURFACE_STATE_OFFSET(index));
170 gen8_vme_output_buffer_setup(VADriverContextP ctx,
171 struct encode_state *encode_state,
173 struct intel_encoder_context *encoder_context)
176 struct i965_driver_data *i965 = i965_driver_data(ctx);
177 struct gen6_vme_context *vme_context = encoder_context->vme_context;
178 VAEncSequenceParameterBufferH264 *pSequenceParameter = (VAEncSequenceParameterBufferH264 *)encode_state->seq_param_ext->buffer;
179 VAEncSliceParameterBufferH264 *pSliceParameter = (VAEncSliceParameterBufferH264 *)encode_state->slice_params_ext[0]->buffer;
180 int is_intra = pSliceParameter->slice_type == SLICE_TYPE_I;
181 int width_in_mbs = pSequenceParameter->picture_width_in_mbs;
182 int height_in_mbs = pSequenceParameter->picture_height_in_mbs;
184 vme_context->vme_output.num_blocks = width_in_mbs * height_in_mbs;
185 vme_context->vme_output.pitch = 16; /* in bytes, always 16 */
188 vme_context->vme_output.size_block = INTRA_VME_OUTPUT_IN_BYTES * 2;
190 vme_context->vme_output.size_block = INTRA_VME_OUTPUT_IN_BYTES * 24;
192 * Inter MV . 32-byte Intra search + 16 IME info + 128 IME MV + 32 IME Ref
193 * + 16 FBR Info + 128 FBR MV + 32 FBR Ref.
194 * 16 * (2 + 2 * (1 + 8 + 2))= 16 * 24.
197 vme_context->vme_output.bo = dri_bo_alloc(i965->intel.bufmgr,
199 vme_context->vme_output.num_blocks * vme_context->vme_output.size_block,
201 assert(vme_context->vme_output.bo);
202 vme_context->vme_buffer_suface_setup(ctx,
203 &vme_context->gpe_context,
204 &vme_context->vme_output,
205 BINDING_TABLE_OFFSET(index),
206 SURFACE_STATE_OFFSET(index));
210 gen8_vme_output_vme_batchbuffer_setup(VADriverContextP ctx,
211 struct encode_state *encode_state,
213 struct intel_encoder_context *encoder_context)
216 struct i965_driver_data *i965 = i965_driver_data(ctx);
217 struct gen6_vme_context *vme_context = encoder_context->vme_context;
218 VAEncSequenceParameterBufferH264 *pSequenceParameter = (VAEncSequenceParameterBufferH264 *)encode_state->seq_param_ext->buffer;
219 int width_in_mbs = pSequenceParameter->picture_width_in_mbs;
220 int height_in_mbs = pSequenceParameter->picture_height_in_mbs;
222 vme_context->vme_batchbuffer.num_blocks = width_in_mbs * height_in_mbs + 1;
223 vme_context->vme_batchbuffer.size_block = 64; /* 4 OWORDs */
224 vme_context->vme_batchbuffer.pitch = 16;
225 vme_context->vme_batchbuffer.bo = dri_bo_alloc(i965->intel.bufmgr,
227 vme_context->vme_batchbuffer.num_blocks * vme_context->vme_batchbuffer.size_block,
230 vme_context->vme_buffer_suface_setup(ctx,
231 &vme_context->gpe_context,
232 &vme_context->vme_batchbuffer,
233 BINDING_TABLE_OFFSET(index),
234 SURFACE_STATE_OFFSET(index));
239 gen8_vme_surface_setup(VADriverContextP ctx,
240 struct encode_state *encode_state,
242 struct intel_encoder_context *encoder_context)
244 struct object_surface *obj_surface;
246 /*Setup surfaces state*/
247 /* current picture for encoding */
248 obj_surface = encode_state->input_yuv_object;
249 gen8_vme_source_surface_state(ctx, 0, obj_surface, encoder_context);
250 gen8_vme_media_source_surface_state(ctx, 4, obj_surface, encoder_context);
251 gen8_vme_media_chroma_source_surface_state(ctx, 6, obj_surface, encoder_context);
254 VAEncSliceParameterBufferH264 *slice_param = (VAEncSliceParameterBufferH264 *)encode_state->slice_params_ext[0]->buffer;
257 slice_type = intel_avc_enc_slice_type_fixup(slice_param->slice_type);
258 assert(slice_type != SLICE_TYPE_I && slice_type != SLICE_TYPE_SI);
260 intel_avc_vme_reference_state(ctx, encode_state, encoder_context, 0, 1, gen8_vme_source_surface_state);
262 if (slice_type == SLICE_TYPE_B)
263 intel_avc_vme_reference_state(ctx, encode_state, encoder_context, 1, 2, gen8_vme_source_surface_state);
267 gen8_vme_output_buffer_setup(ctx, encode_state, 3, encoder_context);
268 gen8_vme_output_vme_batchbuffer_setup(ctx, encode_state, 5, encoder_context);
270 return VA_STATUS_SUCCESS;
273 static VAStatus gen8_vme_interface_setup(VADriverContextP ctx,
274 struct encode_state *encode_state,
275 struct intel_encoder_context *encoder_context)
277 struct gen6_vme_context *vme_context = encoder_context->vme_context;
278 struct gen8_interface_descriptor_data *desc;
281 unsigned char *desc_ptr;
283 bo = vme_context->gpe_context.dynamic_state.bo;
286 desc_ptr = (unsigned char *)bo->virtual + vme_context->gpe_context.idrt_offset;
288 desc = (struct gen8_interface_descriptor_data *)desc_ptr;
290 for (i = 0; i < vme_context->vme_kernel_sum; i++) {
291 struct i965_kernel *kernel;
292 kernel = &vme_context->gpe_context.kernels[i];
293 assert(sizeof(*desc) == 32);
294 /*Setup the descritor table*/
295 memset(desc, 0, sizeof(*desc));
296 desc->desc0.kernel_start_pointer = kernel->kernel_offset >> 6;
297 desc->desc3.sampler_count = 0; /* FIXME: */
298 desc->desc3.sampler_state_pointer = 0;
299 desc->desc4.binding_table_entry_count = 1; /* FIXME: */
300 desc->desc4.binding_table_pointer = (BINDING_TABLE_OFFSET(0) >> 5);
301 desc->desc5.constant_urb_entry_read_offset = 0;
302 desc->desc5.constant_urb_entry_read_length = CURBE_URB_ENTRY_LENGTH;
310 return VA_STATUS_SUCCESS;
313 static VAStatus gen8_vme_constant_setup(VADriverContextP ctx,
314 struct encode_state *encode_state,
315 struct intel_encoder_context *encoder_context)
317 struct gen6_vme_context *vme_context = encoder_context->vme_context;
318 unsigned char *constant_buffer;
319 unsigned int *vme_state_message;
322 vme_state_message = (unsigned int *)vme_context->vme_state_message;
324 if (encoder_context->codec == CODEC_H264) {
325 if (vme_context->h264_level >= 30) {
328 if (vme_context->h264_level >= 31)
331 } else if (encoder_context->codec == CODEC_MPEG2) {
335 vme_state_message[31] = mv_num;
337 dri_bo_map(vme_context->gpe_context.dynamic_state.bo, 1);
338 assert(vme_context->gpe_context.dynamic_state.bo->virtual);
339 constant_buffer = (unsigned char *)vme_context->gpe_context.dynamic_state.bo->virtual +
340 vme_context->gpe_context.curbe_offset;
342 /* VME MV/Mb cost table is passed by using const buffer */
343 /* Now it uses the fixed search path. So it is constructed directly
346 memcpy(constant_buffer, (char *)vme_context->vme_state_message, 128);
348 dri_bo_unmap(vme_context->gpe_context.dynamic_state.bo);
350 return VA_STATUS_SUCCESS;
353 #define MB_SCOREBOARD_A (1 << 0)
354 #define MB_SCOREBOARD_B (1 << 1)
355 #define MB_SCOREBOARD_C (1 << 2)
357 /* check whether the mb of (x_index, y_index) is out of bound */
358 static inline int loop_in_bounds(int x_index, int y_index, int first_mb, int num_mb, int mb_width, int mb_height)
361 if (x_index < 0 || x_index >= mb_width)
363 if (y_index < 0 || y_index >= mb_height)
366 mb_index = y_index * mb_width + x_index;
367 if (mb_index < first_mb || mb_index > (first_mb + num_mb))
373 gen8wa_vme_walker_fill_vme_batchbuffer(VADriverContextP ctx,
374 struct encode_state *encode_state,
375 int mb_width, int mb_height,
377 int transform_8x8_mode_flag,
378 struct intel_encoder_context *encoder_context)
380 struct gen6_vme_context *vme_context = encoder_context->vme_context;
383 unsigned int *command_ptr;
385 #define USE_SCOREBOARD (1 << 21)
387 dri_bo_map(vme_context->vme_batchbuffer.bo, 1);
388 command_ptr = vme_context->vme_batchbuffer.bo->virtual;
390 for (s = 0; s < encode_state->num_slice_params_ext; s++) {
391 VAEncSliceParameterBufferH264 *pSliceParameter = (VAEncSliceParameterBufferH264 *)encode_state->slice_params_ext[s]->buffer;
392 int first_mb = pSliceParameter->macroblock_address;
393 int num_mb = pSliceParameter->num_macroblocks;
394 unsigned int mb_intra_ub, score_dep;
395 int x_outer, y_outer, x_inner, y_inner;
398 x_outer = first_mb % mb_width;
399 y_outer = first_mb / mb_width;
402 for (; x_outer < (mb_width -2 ) && !loop_in_bounds(x_outer, y_outer, first_mb, num_mb, mb_width, mb_height); ) {
405 for (; !loop_in_bounds(x_inner, y_inner, first_mb, num_mb, mb_width, mb_height);) {
409 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_AE;
410 score_dep |= MB_SCOREBOARD_A;
412 if (y_inner != mb_row) {
413 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_B;
414 score_dep |= MB_SCOREBOARD_B;
416 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_D;
417 if (x_inner != (mb_width -1)) {
418 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_C;
419 score_dep |= MB_SCOREBOARD_C;
423 *command_ptr++ = (CMD_MEDIA_OBJECT | (8 - 2));
424 *command_ptr++ = kernel;
425 *command_ptr++ = USE_SCOREBOARD;
428 /* the (X, Y) term of scoreboard */
429 *command_ptr++ = ((y_inner << 16) | x_inner);
430 *command_ptr++ = score_dep;
432 *command_ptr++ = (mb_width << 16 | y_inner << 8 | x_inner);
433 *command_ptr++ = ((1 << 18) | (1 << 16) | transform_8x8_mode_flag | (mb_intra_ub << 8));
434 *command_ptr++ = CMD_MEDIA_STATE_FLUSH;
443 xtemp_outer = mb_width - 2;
446 x_outer = xtemp_outer;
447 y_outer = first_mb / mb_width;
448 for (;!loop_in_bounds(x_outer, y_outer, first_mb, num_mb, mb_width, mb_height); ) {
451 for (; !loop_in_bounds(x_inner, y_inner, first_mb, num_mb, mb_width, mb_height);) {
455 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_AE;
456 score_dep |= MB_SCOREBOARD_A;
458 if (y_inner != mb_row) {
459 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_B;
460 score_dep |= MB_SCOREBOARD_B;
462 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_D;
464 if (x_inner != (mb_width -1)) {
465 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_C;
466 score_dep |= MB_SCOREBOARD_C;
470 *command_ptr++ = (CMD_MEDIA_OBJECT | (8 - 2));
471 *command_ptr++ = kernel;
472 *command_ptr++ = USE_SCOREBOARD;
475 /* the (X, Y) term of scoreboard */
476 *command_ptr++ = ((y_inner << 16) | x_inner);
477 *command_ptr++ = score_dep;
479 *command_ptr++ = (mb_width << 16 | y_inner << 8 | x_inner);
480 *command_ptr++ = ((1 << 18) | (1 << 16) | transform_8x8_mode_flag | (mb_intra_ub << 8));
482 *command_ptr++ = CMD_MEDIA_STATE_FLUSH;
488 if (x_outer >= mb_width) {
490 x_outer = xtemp_outer;
495 *command_ptr++ = MI_BATCH_BUFFER_END;
498 dri_bo_unmap(vme_context->vme_batchbuffer.bo);
502 gen8_vme_fill_vme_batchbuffer(VADriverContextP ctx,
503 struct encode_state *encode_state,
504 int mb_width, int mb_height,
506 int transform_8x8_mode_flag,
507 struct intel_encoder_context *encoder_context)
509 struct gen6_vme_context *vme_context = encoder_context->vme_context;
510 int mb_x = 0, mb_y = 0;
512 unsigned int *command_ptr;
514 dri_bo_map(vme_context->vme_batchbuffer.bo, 1);
515 command_ptr = vme_context->vme_batchbuffer.bo->virtual;
517 for (s = 0; s < encode_state->num_slice_params_ext; s++) {
518 VAEncSliceParameterBufferH264 *pSliceParameter = (VAEncSliceParameterBufferH264 *)encode_state->slice_params_ext[s]->buffer;
519 int slice_mb_begin = pSliceParameter->macroblock_address;
520 int slice_mb_number = pSliceParameter->num_macroblocks;
521 unsigned int mb_intra_ub;
522 int slice_mb_x = pSliceParameter->macroblock_address % mb_width;
523 for (i = 0; i < slice_mb_number; ) {
524 int mb_count = i + slice_mb_begin;
525 mb_x = mb_count % mb_width;
526 mb_y = mb_count / mb_width;
529 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_AE;
532 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_B;
534 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_D;
535 if (mb_x != (mb_width -1))
536 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_C;
540 mb_intra_ub &= ~(INTRA_PRED_AVAIL_FLAG_AE);
541 mb_intra_ub &= ~(INTRA_PRED_AVAIL_FLAG_BCD_MASK);
542 if ((i == (mb_width - 1)) && slice_mb_x) {
543 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_C;
547 if ((i == mb_width) && slice_mb_x) {
548 mb_intra_ub &= ~(INTRA_PRED_AVAIL_FLAG_D);
550 *command_ptr++ = (CMD_MEDIA_OBJECT | (8 - 2));
551 *command_ptr++ = kernel;
558 *command_ptr++ = (mb_width << 16 | mb_y << 8 | mb_x);
559 *command_ptr++ = ( (1 << 16) | transform_8x8_mode_flag | (mb_intra_ub << 8));
561 *command_ptr++ = CMD_MEDIA_STATE_FLUSH;
567 *command_ptr++ = MI_BATCH_BUFFER_END;
570 dri_bo_unmap(vme_context->vme_batchbuffer.bo);
573 static void gen8_vme_media_init(VADriverContextP ctx, struct intel_encoder_context *encoder_context)
575 struct gen6_vme_context *vme_context = encoder_context->vme_context;
577 gen8_gpe_context_init(ctx, &vme_context->gpe_context);
579 /* VME output buffer */
580 dri_bo_unreference(vme_context->vme_output.bo);
581 vme_context->vme_output.bo = NULL;
583 dri_bo_unreference(vme_context->vme_batchbuffer.bo);
584 vme_context->vme_batchbuffer.bo = NULL;
587 dri_bo_unreference(vme_context->vme_state.bo);
588 vme_context->vme_state.bo = NULL;
591 static void gen8_vme_pipeline_programing(VADriverContextP ctx,
592 struct encode_state *encode_state,
593 struct intel_encoder_context *encoder_context)
595 struct gen6_vme_context *vme_context = encoder_context->vme_context;
596 struct intel_batchbuffer *batch = encoder_context->base.batch;
597 VAEncPictureParameterBufferH264 *pPicParameter = (VAEncPictureParameterBufferH264 *)encode_state->pic_param_ext->buffer;
598 VAEncSliceParameterBufferH264 *pSliceParameter = (VAEncSliceParameterBufferH264 *)encode_state->slice_params_ext[0]->buffer;
599 VAEncSequenceParameterBufferH264 *pSequenceParameter = (VAEncSequenceParameterBufferH264 *)encode_state->seq_param_ext->buffer;
600 int width_in_mbs = pSequenceParameter->picture_width_in_mbs;
601 int height_in_mbs = pSequenceParameter->picture_height_in_mbs;
603 bool allow_hwscore = true;
606 for (s = 0; s < encode_state->num_slice_params_ext; s++) {
607 pSliceParameter = (VAEncSliceParameterBufferH264 *)encode_state->slice_params_ext[s]->buffer;
608 if ((pSliceParameter->macroblock_address % width_in_mbs)) {
609 allow_hwscore = false;
614 if ((pSliceParameter->slice_type == SLICE_TYPE_I) ||
615 (pSliceParameter->slice_type == SLICE_TYPE_I)) {
616 kernel_shader = VME_INTRA_SHADER;
617 } else if ((pSliceParameter->slice_type == SLICE_TYPE_P) ||
618 (pSliceParameter->slice_type == SLICE_TYPE_SP)) {
619 kernel_shader = VME_INTER_SHADER;
621 kernel_shader = VME_BINTER_SHADER;
623 kernel_shader = VME_INTER_SHADER;
626 gen8wa_vme_walker_fill_vme_batchbuffer(ctx,
628 width_in_mbs, height_in_mbs,
630 pPicParameter->pic_fields.bits.transform_8x8_mode_flag,
633 gen8_vme_fill_vme_batchbuffer(ctx,
635 width_in_mbs, height_in_mbs,
637 pPicParameter->pic_fields.bits.transform_8x8_mode_flag,
640 intel_batchbuffer_start_atomic(batch, 0x1000);
641 gen8_gpe_pipeline_setup(ctx, &vme_context->gpe_context, batch);
642 BEGIN_BATCH(batch, 3);
643 OUT_BATCH(batch, MI_BATCH_BUFFER_START | (1 << 8) | (1 << 0));
645 vme_context->vme_batchbuffer.bo,
646 I915_GEM_DOMAIN_COMMAND, 0,
649 ADVANCE_BATCH(batch);
651 intel_batchbuffer_end_atomic(batch);
654 static VAStatus gen8_vme_prepare(VADriverContextP ctx,
655 struct encode_state *encode_state,
656 struct intel_encoder_context *encoder_context)
658 VAStatus vaStatus = VA_STATUS_SUCCESS;
659 VAEncSliceParameterBufferH264 *pSliceParameter = (VAEncSliceParameterBufferH264 *)encode_state->slice_params_ext[0]->buffer;
660 int is_intra = pSliceParameter->slice_type == SLICE_TYPE_I;
661 VAEncSequenceParameterBufferH264 *pSequenceParameter = (VAEncSequenceParameterBufferH264 *)encode_state->seq_param_ext->buffer;
662 struct gen6_vme_context *vme_context = encoder_context->vme_context;
664 if (!vme_context->h264_level ||
665 (vme_context->h264_level != pSequenceParameter->level_idc)) {
666 vme_context->h264_level = pSequenceParameter->level_idc;
669 intel_vme_update_mbmv_cost(ctx, encode_state, encoder_context);
671 /*Setup all the memory object*/
672 gen8_vme_surface_setup(ctx, encode_state, is_intra, encoder_context);
673 gen8_vme_interface_setup(ctx, encode_state, encoder_context);
674 //gen8_vme_vme_state_setup(ctx, encode_state, is_intra, encoder_context);
675 gen8_vme_constant_setup(ctx, encode_state, encoder_context);
677 /*Programing media pipeline*/
678 gen8_vme_pipeline_programing(ctx, encode_state, encoder_context);
683 static VAStatus gen8_vme_run(VADriverContextP ctx,
684 struct encode_state *encode_state,
685 struct intel_encoder_context *encoder_context)
687 struct intel_batchbuffer *batch = encoder_context->base.batch;
689 intel_batchbuffer_flush(batch);
691 return VA_STATUS_SUCCESS;
694 static VAStatus gen8_vme_stop(VADriverContextP ctx,
695 struct encode_state *encode_state,
696 struct intel_encoder_context *encoder_context)
698 return VA_STATUS_SUCCESS;
702 gen8_vme_pipeline(VADriverContextP ctx,
704 struct encode_state *encode_state,
705 struct intel_encoder_context *encoder_context)
707 gen8_vme_media_init(ctx, encoder_context);
708 gen8_vme_prepare(ctx, encode_state, encoder_context);
709 gen8_vme_run(ctx, encode_state, encoder_context);
710 gen8_vme_stop(ctx, encode_state, encoder_context);
712 return VA_STATUS_SUCCESS;
716 gen8_vme_mpeg2_output_buffer_setup(VADriverContextP ctx,
717 struct encode_state *encode_state,
720 struct intel_encoder_context *encoder_context)
723 struct i965_driver_data *i965 = i965_driver_data(ctx);
724 struct gen6_vme_context *vme_context = encoder_context->vme_context;
725 VAEncSequenceParameterBufferMPEG2 *seq_param = (VAEncSequenceParameterBufferMPEG2 *)encode_state->seq_param_ext->buffer;
726 int width_in_mbs = ALIGN(seq_param->picture_width, 16) / 16;
727 int height_in_mbs = ALIGN(seq_param->picture_height, 16) / 16;
729 vme_context->vme_output.num_blocks = width_in_mbs * height_in_mbs;
730 vme_context->vme_output.pitch = 16; /* in bytes, always 16 */
733 vme_context->vme_output.size_block = INTRA_VME_OUTPUT_IN_BYTES * 2;
735 vme_context->vme_output.size_block = INTRA_VME_OUTPUT_IN_BYTES * 24;
737 * Inter MV . 32-byte Intra search + 16 IME info + 128 IME MV + 32 IME Ref
738 * + 16 FBR Info + 128 FBR MV + 32 FBR Ref.
739 * 16 * (2 + 2 * (1 + 8 + 2))= 16 * 24.
742 vme_context->vme_output.bo = dri_bo_alloc(i965->intel.bufmgr,
744 vme_context->vme_output.num_blocks * vme_context->vme_output.size_block,
746 assert(vme_context->vme_output.bo);
747 vme_context->vme_buffer_suface_setup(ctx,
748 &vme_context->gpe_context,
749 &vme_context->vme_output,
750 BINDING_TABLE_OFFSET(index),
751 SURFACE_STATE_OFFSET(index));
755 gen8_vme_mpeg2_output_vme_batchbuffer_setup(VADriverContextP ctx,
756 struct encode_state *encode_state,
758 struct intel_encoder_context *encoder_context)
761 struct i965_driver_data *i965 = i965_driver_data(ctx);
762 struct gen6_vme_context *vme_context = encoder_context->vme_context;
763 VAEncSequenceParameterBufferMPEG2 *seq_param = (VAEncSequenceParameterBufferMPEG2 *)encode_state->seq_param_ext->buffer;
764 int width_in_mbs = ALIGN(seq_param->picture_width, 16) / 16;
765 int height_in_mbs = ALIGN(seq_param->picture_height, 16) / 16;
767 vme_context->vme_batchbuffer.num_blocks = width_in_mbs * height_in_mbs + 1;
768 vme_context->vme_batchbuffer.size_block = 64; /* 4 OWORDs */
769 vme_context->vme_batchbuffer.pitch = 16;
770 vme_context->vme_batchbuffer.bo = dri_bo_alloc(i965->intel.bufmgr,
772 vme_context->vme_batchbuffer.num_blocks * vme_context->vme_batchbuffer.size_block,
774 vme_context->vme_buffer_suface_setup(ctx,
775 &vme_context->gpe_context,
776 &vme_context->vme_batchbuffer,
777 BINDING_TABLE_OFFSET(index),
778 SURFACE_STATE_OFFSET(index));
782 gen8_vme_mpeg2_surface_setup(VADriverContextP ctx,
783 struct encode_state *encode_state,
785 struct intel_encoder_context *encoder_context)
787 struct object_surface *obj_surface;
789 /*Setup surfaces state*/
790 /* current picture for encoding */
791 obj_surface = encode_state->input_yuv_object;
792 gen8_vme_source_surface_state(ctx, 0, obj_surface, encoder_context);
793 gen8_vme_media_source_surface_state(ctx, 4, obj_surface, encoder_context);
794 gen8_vme_media_chroma_source_surface_state(ctx, 6, obj_surface, encoder_context);
798 obj_surface = encode_state->reference_objects[0];
800 if (obj_surface->bo != NULL)
801 gen8_vme_source_surface_state(ctx, 1, obj_surface, encoder_context);
804 obj_surface = encode_state->reference_objects[1];
806 if (obj_surface && obj_surface->bo != NULL)
807 gen8_vme_source_surface_state(ctx, 2, obj_surface, encoder_context);
811 gen8_vme_mpeg2_output_buffer_setup(ctx, encode_state, 3, is_intra, encoder_context);
812 gen8_vme_mpeg2_output_vme_batchbuffer_setup(ctx, encode_state, 5, encoder_context);
814 return VA_STATUS_SUCCESS;
818 gen8wa_vme_mpeg2_walker_fill_vme_batchbuffer(VADriverContextP ctx,
819 struct encode_state *encode_state,
820 int mb_width, int mb_height,
822 struct intel_encoder_context *encoder_context)
824 struct gen6_vme_context *vme_context = encoder_context->vme_context;
825 unsigned int *command_ptr;
827 #define MPEG2_SCOREBOARD (1 << 21)
829 dri_bo_map(vme_context->vme_batchbuffer.bo, 1);
830 command_ptr = vme_context->vme_batchbuffer.bo->virtual;
833 unsigned int mb_intra_ub, score_dep;
834 int x_outer, y_outer, x_inner, y_inner;
837 int num_mb = mb_width * mb_height;
843 for (; x_outer < (mb_width -2 ) && !loop_in_bounds(x_outer, y_outer, first_mb, num_mb, mb_width, mb_height); ) {
846 for (; !loop_in_bounds(x_inner, y_inner, first_mb, num_mb, mb_width, mb_height);) {
850 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_AE;
851 score_dep |= MB_SCOREBOARD_A;
854 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_B;
855 score_dep |= MB_SCOREBOARD_B;
858 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_D;
860 if (x_inner != (mb_width -1)) {
861 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_C;
862 score_dep |= MB_SCOREBOARD_C;
866 *command_ptr++ = (CMD_MEDIA_OBJECT | (8 - 2));
867 *command_ptr++ = kernel;
868 *command_ptr++ = MPEG2_SCOREBOARD;
871 /* the (X, Y) term of scoreboard */
872 *command_ptr++ = ((y_inner << 16) | x_inner);
873 *command_ptr++ = score_dep;
875 *command_ptr++ = (mb_width << 16 | y_inner << 8 | x_inner);
876 *command_ptr++ = ((1 << 18) | (1 << 16) | (mb_intra_ub << 8));
877 *command_ptr++ = CMD_MEDIA_STATE_FLUSH;
886 xtemp_outer = mb_width - 2;
889 x_outer = xtemp_outer;
891 for (;!loop_in_bounds(x_outer, y_outer, first_mb, num_mb, mb_width, mb_height); ) {
894 for (; !loop_in_bounds(x_inner, y_inner, first_mb, num_mb, mb_width, mb_height);) {
898 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_AE;
899 score_dep |= MB_SCOREBOARD_A;
902 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_B;
903 score_dep |= MB_SCOREBOARD_B;
906 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_D;
908 if (x_inner != (mb_width -1)) {
909 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_C;
910 score_dep |= MB_SCOREBOARD_C;
914 *command_ptr++ = (CMD_MEDIA_OBJECT | (8 - 2));
915 *command_ptr++ = kernel;
916 *command_ptr++ = MPEG2_SCOREBOARD;
919 /* the (X, Y) term of scoreboard */
920 *command_ptr++ = ((y_inner << 16) | x_inner);
921 *command_ptr++ = score_dep;
923 *command_ptr++ = (mb_width << 16 | y_inner << 8 | x_inner);
924 *command_ptr++ = ((1 << 18) | (1 << 16) | (mb_intra_ub << 8));
926 *command_ptr++ = CMD_MEDIA_STATE_FLUSH;
932 if (x_outer >= mb_width) {
934 x_outer = xtemp_outer;
939 *command_ptr++ = MI_BATCH_BUFFER_END;
942 dri_bo_unmap(vme_context->vme_batchbuffer.bo);
947 gen8_vme_mpeg2_fill_vme_batchbuffer(VADriverContextP ctx,
948 struct encode_state *encode_state,
949 int mb_width, int mb_height,
951 int transform_8x8_mode_flag,
952 struct intel_encoder_context *encoder_context)
954 struct gen6_vme_context *vme_context = encoder_context->vme_context;
955 int mb_x = 0, mb_y = 0;
957 unsigned int *command_ptr;
960 dri_bo_map(vme_context->vme_batchbuffer.bo, 1);
961 command_ptr = vme_context->vme_batchbuffer.bo->virtual;
963 for (s = 0; s < encode_state->num_slice_params_ext; s++) {
964 VAEncSliceParameterBufferMPEG2 *slice_param = (VAEncSliceParameterBufferMPEG2 *)encode_state->slice_params_ext[s]->buffer;
966 for (j = 0; j < encode_state->slice_params_ext[s]->num_elements; j++) {
967 int slice_mb_begin = slice_param->macroblock_address;
968 int slice_mb_number = slice_param->num_macroblocks;
969 unsigned int mb_intra_ub;
971 for (i = 0; i < slice_mb_number;) {
972 int mb_count = i + slice_mb_begin;
974 mb_x = mb_count % mb_width;
975 mb_y = mb_count / mb_width;
979 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_AE;
983 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_B;
986 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_D;
988 if (mb_x != (mb_width -1))
989 mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_C;
992 *command_ptr++ = (CMD_MEDIA_OBJECT | (8 - 2));
993 *command_ptr++ = kernel;
1000 *command_ptr++ = (mb_width << 16 | mb_y << 8 | mb_x);
1001 *command_ptr++ = ( (1 << 16) | transform_8x8_mode_flag | (mb_intra_ub << 8));
1003 *command_ptr++ = CMD_MEDIA_STATE_FLUSH;
1012 *command_ptr++ = MI_BATCH_BUFFER_END;
1015 dri_bo_unmap(vme_context->vme_batchbuffer.bo);
1019 gen8_vme_mpeg2_pipeline_programing(VADriverContextP ctx,
1020 struct encode_state *encode_state,
1022 struct intel_encoder_context *encoder_context)
1024 struct gen6_vme_context *vme_context = encoder_context->vme_context;
1025 struct intel_batchbuffer *batch = encoder_context->base.batch;
1026 VAEncSequenceParameterBufferMPEG2 *seq_param = (VAEncSequenceParameterBufferMPEG2 *)encode_state->seq_param_ext->buffer;
1027 int width_in_mbs = ALIGN(seq_param->picture_width, 16) / 16;
1028 int height_in_mbs = ALIGN(seq_param->picture_height, 16) / 16;
1029 bool allow_hwscore = true;
1032 VAEncPictureParameterBufferMPEG2 *pic_param = NULL;
1034 for (s = 0; s < encode_state->num_slice_params_ext; s++) {
1036 VAEncSliceParameterBufferMPEG2 *slice_param = (VAEncSliceParameterBufferMPEG2 *)encode_state->slice_params_ext[s]->buffer;
1038 for (j = 0; j < encode_state->slice_params_ext[s]->num_elements; j++) {
1039 if (slice_param->macroblock_address % width_in_mbs) {
1040 allow_hwscore = false;
1046 pic_param = (VAEncPictureParameterBufferMPEG2 *)encode_state->pic_param_ext->buffer;
1047 if (pic_param->picture_type == VAEncPictureTypeIntra) {
1048 allow_hwscore = false;
1049 kernel_shader = VME_INTRA_SHADER;
1051 kernel_shader = VME_INTER_SHADER;
1055 gen8wa_vme_mpeg2_walker_fill_vme_batchbuffer(ctx,
1057 width_in_mbs, height_in_mbs,
1061 gen8_vme_mpeg2_fill_vme_batchbuffer(ctx,
1063 width_in_mbs, height_in_mbs,
1064 is_intra ? VME_INTRA_SHADER : VME_INTER_SHADER,
1068 intel_batchbuffer_start_atomic(batch, 0x1000);
1069 gen8_gpe_pipeline_setup(ctx, &vme_context->gpe_context, batch);
1070 BEGIN_BATCH(batch, 4);
1071 OUT_BATCH(batch, MI_BATCH_BUFFER_START | (1 << 8) | (1 << 0));
1073 vme_context->vme_batchbuffer.bo,
1074 I915_GEM_DOMAIN_COMMAND, 0,
1076 OUT_BATCH(batch, 0);
1077 OUT_BATCH(batch, 0);
1078 ADVANCE_BATCH(batch);
1080 intel_batchbuffer_end_atomic(batch);
1084 gen8_vme_mpeg2_prepare(VADriverContextP ctx,
1085 struct encode_state *encode_state,
1086 struct intel_encoder_context *encoder_context)
1088 VAStatus vaStatus = VA_STATUS_SUCCESS;
1089 VAEncSliceParameterBufferMPEG2 *slice_param = (VAEncSliceParameterBufferMPEG2 *)encode_state->slice_params_ext[0]->buffer;
1090 VAEncSequenceParameterBufferMPEG2 *seq_param = (VAEncSequenceParameterBufferMPEG2 *)encode_state->seq_param_ext->buffer;
1091 struct gen6_vme_context *vme_context = encoder_context->vme_context;
1093 if ((!vme_context->mpeg2_level) ||
1094 (vme_context->mpeg2_level != (seq_param->sequence_extension.bits.profile_and_level_indication & MPEG2_LEVEL_MASK))) {
1095 vme_context->mpeg2_level = seq_param->sequence_extension.bits.profile_and_level_indication & MPEG2_LEVEL_MASK;
1099 /*Setup all the memory object*/
1100 gen8_vme_mpeg2_surface_setup(ctx, encode_state, slice_param->is_intra_slice, encoder_context);
1101 gen8_vme_interface_setup(ctx, encode_state, encoder_context);
1102 //gen8_vme_vme_state_setup(ctx, encode_state, slice_param->is_intra_slice, encoder_context);
1103 intel_vme_mpeg2_state_setup(ctx, encode_state, encoder_context);
1104 gen8_vme_constant_setup(ctx, encode_state, encoder_context);
1106 /*Programing media pipeline*/
1107 gen8_vme_mpeg2_pipeline_programing(ctx, encode_state, slice_param->is_intra_slice, encoder_context);
1113 gen8_vme_mpeg2_pipeline(VADriverContextP ctx,
1115 struct encode_state *encode_state,
1116 struct intel_encoder_context *encoder_context)
1118 gen8_vme_media_init(ctx, encoder_context);
1119 gen8_vme_mpeg2_prepare(ctx, encode_state, encoder_context);
1120 gen8_vme_run(ctx, encode_state, encoder_context);
1121 gen8_vme_stop(ctx, encode_state, encoder_context);
1123 return VA_STATUS_SUCCESS;
1127 gen8_vme_context_destroy(void *context)
1129 struct gen6_vme_context *vme_context = context;
1131 gen8_gpe_context_destroy(&vme_context->gpe_context);
1133 dri_bo_unreference(vme_context->vme_output.bo);
1134 vme_context->vme_output.bo = NULL;
1136 dri_bo_unreference(vme_context->vme_state.bo);
1137 vme_context->vme_state.bo = NULL;
1139 dri_bo_unreference(vme_context->vme_batchbuffer.bo);
1140 vme_context->vme_batchbuffer.bo = NULL;
1142 if (vme_context->vme_state_message) {
1143 free(vme_context->vme_state_message);
1144 vme_context->vme_state_message = NULL;
1150 Bool gen8_vme_context_init(VADriverContextP ctx, struct intel_encoder_context *encoder_context)
1152 struct gen6_vme_context *vme_context = calloc(1, sizeof(struct gen6_vme_context));
1153 struct i965_kernel *vme_kernel_list = NULL;
1154 int i965_kernel_num;
1156 switch (encoder_context->codec) {
1158 vme_kernel_list = gen8_vme_kernels;
1159 encoder_context->vme_pipeline = gen8_vme_pipeline;
1160 i965_kernel_num = sizeof(gen8_vme_kernels) / sizeof(struct i965_kernel);
1164 vme_kernel_list = gen8_vme_mpeg2_kernels;
1165 encoder_context->vme_pipeline = gen8_vme_mpeg2_pipeline;
1166 i965_kernel_num = sizeof(gen8_vme_mpeg2_kernels) / sizeof(struct i965_kernel);
1171 /* never get here */
1176 vme_context->vme_kernel_sum = i965_kernel_num;
1177 vme_context->gpe_context.surface_state_binding_table.length = (SURFACE_STATE_PADDED_SIZE + sizeof(unsigned int)) * MAX_MEDIA_SURFACES_GEN6;
1179 vme_context->gpe_context.idrt_size = sizeof(struct gen8_interface_descriptor_data) * MAX_INTERFACE_DESC_GEN6;
1180 vme_context->gpe_context.curbe_size = CURBE_TOTAL_DATA_LENGTH;
1181 vme_context->gpe_context.sampler_size = 0;
1184 vme_context->gpe_context.vfe_state.max_num_threads = 60 - 1;
1185 vme_context->gpe_context.vfe_state.num_urb_entries = 64;
1186 vme_context->gpe_context.vfe_state.gpgpu_mode = 0;
1187 vme_context->gpe_context.vfe_state.urb_entry_size = 16;
1188 vme_context->gpe_context.vfe_state.curbe_allocation_size = CURBE_ALLOCATION_SIZE - 1;
1190 gen7_vme_scoreboard_init(ctx, vme_context);
1192 gen8_gpe_load_kernels(ctx,
1193 &vme_context->gpe_context,
1196 vme_context->vme_surface2_setup = gen8_gpe_surface2_setup;
1197 vme_context->vme_media_rw_surface_setup = gen8_gpe_media_rw_surface_setup;
1198 vme_context->vme_buffer_suface_setup = gen8_gpe_buffer_suface_setup;
1199 vme_context->vme_media_chroma_surface_setup = gen8_gpe_media_chroma_surface_setup;
1201 encoder_context->vme_context = vme_context;
1202 encoder_context->vme_context_destroy = gen8_vme_context_destroy;
1204 vme_context->vme_state_message = malloc(VME_MSG_LENGTH * sizeof(int));