#include <stdlib.h>
#include <string.h>
#include <assert.h>
+#include <math.h>
#include "intel_batchbuffer.h"
#include "i965_defines.h"
#include "i965_encoder_utils.h"
#include "gen6_mfc.h"
#include "gen6_vme.h"
+#include "intel_media.h"
+
+#define SURFACE_STATE_PADDED_SIZE MAX(SURFACE_STATE_PADDED_SIZE_GEN6, SURFACE_STATE_PADDED_SIZE_GEN7)
+#define SURFACE_STATE_OFFSET(index) (SURFACE_STATE_PADDED_SIZE * index)
+#define BINDING_TABLE_OFFSET(index) (SURFACE_STATE_OFFSET(MAX_MEDIA_SURFACES_GEN6) + sizeof(unsigned int) * index)
+
+static const uint32_t gen6_mfc_batchbuffer_avc_intra[][4] = {
+#include "shaders/utils/mfc_batchbuffer_avc_intra.g6b"
+};
+
+static const uint32_t gen6_mfc_batchbuffer_avc_inter[][4] = {
+#include "shaders/utils/mfc_batchbuffer_avc_inter.g6b"
+};
+
+static struct i965_kernel gen6_mfc_kernels[] = {
+ {
+ "MFC AVC INTRA BATCHBUFFER ",
+ MFC_BATCHBUFFER_AVC_INTRA,
+ gen6_mfc_batchbuffer_avc_intra,
+ sizeof(gen6_mfc_batchbuffer_avc_intra),
+ NULL
+ },
+
+ {
+ "MFC AVC INTER BATCHBUFFER ",
+ MFC_BATCHBUFFER_AVC_INTER,
+ gen6_mfc_batchbuffer_avc_inter,
+ sizeof(gen6_mfc_batchbuffer_avc_inter),
+ NULL
+ },
+};
static void
-gen6_mfc_pipe_mode_select(VADriverContextP ctx, struct intel_encoder_context *encoder_context)
+gen6_mfc_pipe_mode_select(VADriverContextP ctx,
+ int standard_select,
+ struct intel_encoder_context *encoder_context)
{
struct intel_batchbuffer *batch = encoder_context->base.batch;
+ struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
+
+ assert(standard_select == MFX_FORMAT_AVC);
BEGIN_BCS_BATCH(batch, 4);
OUT_BCS_BATCH(batch, MFX_PIPE_MODE_SELECT | (4 - 2));
OUT_BCS_BATCH(batch,
(1 << 10) | /* disable Stream-Out , advanced QP/bitrate control need enable it*/
- (1 << 9) | /* Post Deblocking Output */
- (0 << 8) | /* Pre Deblocking Output */
+ ((!!mfc_context->post_deblocking_output.bo) << 9) | /* Post Deblocking Output */
+ ((!!mfc_context->pre_deblocking_output.bo) << 8) | /* Pre Deblocking Output */
(0 << 7) | /* disable TLB prefectch */
(0 << 5) | /* not in stitch mode */
(1 << 4) | /* encoding mode */
ADVANCE_BCS_BATCH(batch);
}
-static void
+void
gen6_mfc_pipe_buf_addr_state(VADriverContextP ctx, struct intel_encoder_context *encoder_context)
{
struct intel_batchbuffer *batch = encoder_context->base.batch;
OUT_BCS_BATCH(batch, MFX_PIPE_BUF_ADDR_STATE | (24 - 2));
- OUT_BCS_BATCH(batch, 0); /* pre output addr */
+ if (mfc_context->pre_deblocking_output.bo)
+ OUT_BCS_RELOC(batch, mfc_context->pre_deblocking_output.bo,
+ I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
+ 0);
+ else
+ OUT_BCS_BATCH(batch, 0); /* pre output addr */
- OUT_BCS_RELOC(batch, mfc_context->post_deblocking_output.bo,
- I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
- 0); /* post output addr */
+ if (mfc_context->post_deblocking_output.bo)
+ OUT_BCS_RELOC(batch, mfc_context->post_deblocking_output.bo,
+ I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
+ 0); /* post output addr */
+ else
+ OUT_BCS_BATCH(batch, 0);
OUT_BCS_RELOC(batch, mfc_context->uncompressed_picture_source.bo,
I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
ADVANCE_BCS_BATCH(batch);
}
-static void
+void
gen6_mfc_bsp_buf_base_addr_state(VADriverContextP ctx, struct intel_encoder_context *encoder_context)
{
struct intel_batchbuffer *batch = encoder_context->base.batch;
(pPicParameter->pic_fields.bits.entropy_coding_mode_flag << 7) | /*0:CAVLC encoding mode,1:CABAC*/
(0 << 6) | /*Only valid for VLD decoding mode*/
(0 << 5) | /*Constrained Intra Predition Flag, from PPS*/
- (pSequenceParameter->direct_8x8_inference_flag << 4) | /*Direct 8x8 inference flag*/
+ (pSequenceParameter->seq_fields.bits.direct_8x8_inference_flag << 4) | /*Direct 8x8 inference flag*/
(pPicParameter->pic_fields.bits.transform_8x8_mode_flag << 3) | /*8x8 or 4x4 IDCT Transform Mode Flag*/
(1 << 2) | /*Frame MB only flag*/
(0 << 1) | /*MBAFF mode is in active*/
ADVANCE_BCS_BATCH(batch);
}
-static void gen6_mfc_avc_directmode_state(VADriverContextP ctx, struct intel_encoder_context *encoder_context)
+static void
+gen6_mfc_avc_directmode_state(VADriverContextP ctx, struct intel_encoder_context *encoder_context)
{
struct intel_batchbuffer *batch = encoder_context->base.batch;
struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
ADVANCE_BCS_BATCH(batch);
}
-static void gen6_mfc_avc_slice_state(VADriverContextP ctx,
- int slice_type,
- struct encode_state *encode_state,
- struct intel_encoder_context *encoder_context,
- int rate_control_enable,
- int qp)
+static void
+gen6_mfc_avc_slice_state(VADriverContextP ctx,
+ VAEncPictureParameterBufferH264 *pic_param,
+ VAEncSliceParameterBufferH264 *slice_param,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context,
+ int rate_control_enable,
+ int qp,
+ struct intel_batchbuffer *batch)
{
- struct intel_batchbuffer *batch = encoder_context->base.batch;
struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
- VAEncSliceParameterBufferH264 *pSliceParameter = (VAEncSliceParameterBufferH264 *)encode_state->slice_params_ext[0]->buffer; /* TODO: multi slices support */
- int bit_rate_control_target;
- if ( slice_type == SLICE_TYPE_I )
- bit_rate_control_target = 0;
- else
- bit_rate_control_target = 1;
- int maxQpN = mfc_context->bit_rate_control_context[bit_rate_control_target].MaxQpNegModifier;
- int maxQpP = mfc_context->bit_rate_control_context[bit_rate_control_target].MaxQpPosModifier;
- unsigned char correct[6];
+ int width_in_mbs = (mfc_context->surface_state.width + 15) / 16;
+ int height_in_mbs = (mfc_context->surface_state.height + 15) / 16;
+ int beginmb = slice_param->macroblock_address;
+ int endmb = beginmb + slice_param->num_macroblocks;
+ int beginx = beginmb % width_in_mbs;
+ int beginy = beginmb / width_in_mbs;
+ int nextx = endmb % width_in_mbs;
+ int nexty = endmb / width_in_mbs;
+ int slice_type = intel_avc_enc_slice_type_fixup(slice_param->slice_type);
+ int last_slice = (endmb == (width_in_mbs * height_in_mbs));
+ int maxQpN, maxQpP;
+ unsigned char correct[6], grow, shrink;
int i;
+ int weighted_pred_idc = 0;
+ unsigned int luma_log2_weight_denom = slice_param->luma_log2_weight_denom;
+ unsigned int chroma_log2_weight_denom = slice_param->chroma_log2_weight_denom;
+ int num_ref_l0 = 0, num_ref_l1 = 0;
+
+ if (batch == NULL)
+ batch = encoder_context->base.batch;
+
+ if (slice_type == SLICE_TYPE_I) {
+ luma_log2_weight_denom = 0;
+ chroma_log2_weight_denom = 0;
+ } else if (slice_type == SLICE_TYPE_P) {
+ weighted_pred_idc = pic_param->pic_fields.bits.weighted_pred_flag;
+ num_ref_l0 = pic_param->num_ref_idx_l0_active_minus1 + 1;
+
+ if (slice_param->num_ref_idx_active_override_flag)
+ num_ref_l0 = slice_param->num_ref_idx_l0_active_minus1 + 1;
+ } else if (slice_type == SLICE_TYPE_B) {
+ weighted_pred_idc = pic_param->pic_fields.bits.weighted_bipred_idc;
+ num_ref_l0 = pic_param->num_ref_idx_l0_active_minus1 + 1;
+ num_ref_l1 = pic_param->num_ref_idx_l1_active_minus1 + 1;
+
+ if (slice_param->num_ref_idx_active_override_flag) {
+ num_ref_l0 = slice_param->num_ref_idx_l0_active_minus1 + 1;
+ num_ref_l1 = slice_param->num_ref_idx_l1_active_minus1 + 1;
+ }
+
+ if (weighted_pred_idc == 2) {
+ /* 8.4.3 - Derivation process for prediction weights (8-279) */
+ luma_log2_weight_denom = 5;
+ chroma_log2_weight_denom = 5;
+ }
+ }
+
+ maxQpN = mfc_context->bit_rate_control_context[slice_type].MaxQpNegModifier;
+ maxQpP = mfc_context->bit_rate_control_context[slice_type].MaxQpPosModifier;
for (i = 0; i < 6; i++)
- correct[i] = mfc_context->bit_rate_control_context[bit_rate_control_target].Correct[i];
- unsigned char grow = mfc_context->bit_rate_control_context[bit_rate_control_target].GrowInit +
- (mfc_context->bit_rate_control_context[bit_rate_control_target].GrowResistance << 4);
- unsigned char shrink = mfc_context->bit_rate_control_context[bit_rate_control_target].ShrinkInit +
- (mfc_context->bit_rate_control_context[bit_rate_control_target].ShrinkResistance << 4);
+ correct[i] = mfc_context->bit_rate_control_context[slice_type].Correct[i];
+
+ grow = mfc_context->bit_rate_control_context[slice_type].GrowInit +
+ (mfc_context->bit_rate_control_context[slice_type].GrowResistance << 4);
+ shrink = mfc_context->bit_rate_control_context[slice_type].ShrinkInit +
+ (mfc_context->bit_rate_control_context[slice_type].ShrinkResistance << 4);
BEGIN_BCS_BATCH(batch, 11);;
OUT_BCS_BATCH(batch, MFX_AVC_SLICE_STATE | (11 - 2) );
-
OUT_BCS_BATCH(batch, slice_type); /*Slice Type: I:P:B Slice*/
- if ( slice_type == SLICE_TYPE_I ) {
- OUT_BCS_BATCH(batch, 0); /*no reference frames and pred_weight_table*/
- } else {
- OUT_BCS_BATCH(batch, 0x00010000); /*1 reference frame*/
- }
+ OUT_BCS_BATCH(batch,
+ (num_ref_l0 << 16) |
+ (num_ref_l1 << 24) |
+ (chroma_log2_weight_denom << 8) |
+ (luma_log2_weight_denom << 0));
OUT_BCS_BATCH(batch,
- (pSliceParameter->direct_spatial_mv_pred_flag<<29) | /*Direct Prediction Type*/
- (0<<24) | /*Enable deblocking operation*/
+ (weighted_pred_idc << 30) |
+ (slice_param->direct_spatial_mv_pred_flag<<29) | /*Direct Prediction Type*/
+ (slice_param->disable_deblocking_filter_idc << 27) |
+ (slice_param->cabac_init_idc << 24) |
(qp<<16) | /*Slice Quantization Parameter*/
- 0x0202 );
- OUT_BCS_BATCH(batch, 0); /*First MB X&Y , the postion of current slice*/
- OUT_BCS_BATCH(batch, ( ((mfc_context->surface_state.height+15)/16) << 16) );
-
+ ((slice_param->slice_beta_offset_div2 & 0xf) << 8) |
+ ((slice_param->slice_alpha_c0_offset_div2 & 0xf) << 0));
+ OUT_BCS_BATCH(batch,
+ (beginy << 24) | /*First MB X&Y , the begin postion of current slice*/
+ (beginx << 16) |
+ slice_param->macroblock_address );
+ OUT_BCS_BATCH(batch, (nexty << 16) | nextx); /*Next slice first MB X&Y*/
OUT_BCS_BATCH(batch,
- (rate_control_enable<<31) | /*in CBR mode RateControlCounterEnable = enable*/
- (1<<30) | /*ResetRateControlCounter*/
- (0<<28) | /*RC Triggle Mode = Always Rate Control*/
- (4<<24) | /*RC Stable Tolerance, middle level*/
- (rate_control_enable<<23) | /*RC Panic Enable*/
- (0<<22) | /*QP mode, don't modfiy CBP*/
- (0<<21) | /*MB Type Direct Conversion Enabled*/
- (0<<20) | /*MB Type Skip Conversion Enabled*/
- (1<<19) | /*IsLastSlice*/
- (0<<18) | /*BitstreamOutputFlag Compressed BitStream Output Disable Flag 0:enable 1:disable*/
- (1<<17) | /*HeaderPresentFlag*/
- (1<<16) | /*SliceData PresentFlag*/
- (1<<15) | /*TailPresentFlag*/
- (1<<13) | /*RBSP NAL TYPE*/
- (0<<12) ); /*CabacZeroWordInsertionEnable*/
-
+ (0/*rate_control_enable*/ << 31) | /*in CBR mode RateControlCounterEnable = enable*/
+ (1 << 30) | /*ResetRateControlCounter*/
+ (0 << 28) | /*RC Triggle Mode = Always Rate Control*/
+ (4 << 24) | /*RC Stable Tolerance, middle level*/
+ (0/*rate_control_enable*/ << 23) | /*RC Panic Enable*/
+ (0 << 22) | /*QP mode, don't modfiy CBP*/
+ (0 << 21) | /*MB Type Direct Conversion Enabled*/
+ (0 << 20) | /*MB Type Skip Conversion Enabled*/
+ (last_slice << 19) | /*IsLastSlice*/
+ (0 << 18) | /*BitstreamOutputFlag Compressed BitStream Output Disable Flag 0:enable 1:disable*/
+ (1 << 17) | /*HeaderPresentFlag*/
+ (1 << 16) | /*SliceData PresentFlag*/
+ (1 << 15) | /*TailPresentFlag*/
+ (1 << 13) | /*RBSP NAL TYPE*/
+ (0 << 12) ); /*CabacZeroWordInsertionEnable*/
OUT_BCS_BATCH(batch, mfc_context->mfc_indirect_pak_bse_object.offset);
-
- OUT_BCS_BATCH(batch, (maxQpN<<24) | /*Target QP - 24 is lowest QP*/
- (maxQpP<<16) | /*Target QP + 20 is highest QP*/
- (shrink<<8) |
- (grow<<0));
- OUT_BCS_BATCH(batch, (correct[5] << 20) |
+ OUT_BCS_BATCH(batch,
+ (maxQpN << 24) | /*Target QP - 24 is lowest QP*/
+ (maxQpP << 16) | /*Target QP + 20 is highest QP*/
+ (shrink << 8) |
+ (grow << 0));
+ OUT_BCS_BATCH(batch,
+ (correct[5] << 20) |
(correct[4] << 16) |
(correct[3] << 12) |
(correct[2] << 8) |
ADVANCE_BCS_BATCH(batch);
}
+
static void gen6_mfc_avc_qm_state(VADriverContextP ctx, struct intel_encoder_context *encoder_context)
{
struct intel_batchbuffer *batch = encoder_context->base.batch;
ADVANCE_BCS_BATCH(batch);
}
-static void gen6_mfc_avc_ref_idx_state(VADriverContextP ctx, struct intel_encoder_context *encoder_context)
-{
- struct intel_batchbuffer *batch = encoder_context->base.batch;
- int i;
-
- BEGIN_BCS_BATCH(batch, 10);
- OUT_BCS_BATCH(batch, MFX_AVC_REF_IDX_STATE | 8);
- OUT_BCS_BATCH(batch, 0); //Select L0
- OUT_BCS_BATCH(batch, 0x80808020); //Only 1 reference
- for(i = 0; i < 7; i++) {
- OUT_BCS_BATCH(batch, 0x80808080);
- }
- ADVANCE_BCS_BATCH(batch);
-
- BEGIN_BCS_BATCH(batch, 10);
- OUT_BCS_BATCH(batch, MFX_AVC_REF_IDX_STATE | 8);
- OUT_BCS_BATCH(batch, 1); //Select L1
- OUT_BCS_BATCH(batch, 0x80808022); //Only 1 reference
- for(i = 0; i < 7; i++) {
- OUT_BCS_BATCH(batch, 0x80808080);
- }
- ADVANCE_BCS_BATCH(batch);
-}
-
static void
gen6_mfc_avc_insert_object(VADriverContextP ctx, struct intel_encoder_context *encoder_context,
unsigned int *insert_data, int lenght_in_dws, int data_bits_in_last_dw,
- int skip_emul_byte_count, int is_last_header, int is_end_of_slice, int emulation_flag)
+ int skip_emul_byte_count, int is_last_header, int is_end_of_slice, int emulation_flag,
+ struct intel_batchbuffer *batch)
{
- struct i965_driver_data *i965 = i965_driver_data(ctx);
- struct intel_batchbuffer *batch = encoder_context->base.batch;
+ if (batch == NULL)
+ batch = encoder_context->base.batch;
BEGIN_BCS_BATCH(batch, lenght_in_dws + 2);
- if (IS_GEN7(i965->intel.device_id))
- OUT_BCS_BATCH(batch, MFX_INSERT_OBJECT | (lenght_in_dws + 2 - 2));
- else
- OUT_BCS_BATCH(batch, MFC_AVC_INSERT_OBJECT | (lenght_in_dws + 2 - 2));
+ OUT_BCS_BATCH(batch, MFC_AVC_INSERT_OBJECT | (lenght_in_dws + 2 - 2));
OUT_BCS_BATCH(batch,
(0 << 16) | /* always start at offset 0 */
ADVANCE_BCS_BATCH(batch);
}
+void
+gen6_mfc_init(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
+{
+ struct i965_driver_data *i965 = i965_driver_data(ctx);
+ struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
+ dri_bo *bo;
+ int i;
+ int width_in_mbs = 0;
+ int height_in_mbs = 0;
+ int slice_batchbuffer_size;
+
+ if (encoder_context->codec == CODEC_H264) {
+ VAEncSequenceParameterBufferH264 *pSequenceParameter = (VAEncSequenceParameterBufferH264 *)encode_state->seq_param_ext->buffer;
+ width_in_mbs = pSequenceParameter->picture_width_in_mbs;
+ height_in_mbs = pSequenceParameter->picture_height_in_mbs;
+ } else {
+ VAEncSequenceParameterBufferMPEG2 *pSequenceParameter = (VAEncSequenceParameterBufferMPEG2 *)encode_state->seq_param_ext->buffer;
+
+ assert(encoder_context->codec == CODEC_MPEG2);
+
+ width_in_mbs = ALIGN(pSequenceParameter->picture_width, 16) / 16;
+ height_in_mbs = ALIGN(pSequenceParameter->picture_height, 16) / 16;
+ }
+
+ slice_batchbuffer_size = 64 * width_in_mbs * height_in_mbs + 4096 +
+ (SLICE_HEADER + SLICE_TAIL) * encode_state->num_slice_params_ext;
+
+ /*Encode common setup for MFC*/
+ dri_bo_unreference(mfc_context->post_deblocking_output.bo);
+ mfc_context->post_deblocking_output.bo = NULL;
+
+ dri_bo_unreference(mfc_context->pre_deblocking_output.bo);
+ mfc_context->pre_deblocking_output.bo = NULL;
+
+ dri_bo_unreference(mfc_context->uncompressed_picture_source.bo);
+ mfc_context->uncompressed_picture_source.bo = NULL;
+
+ dri_bo_unreference(mfc_context->mfc_indirect_pak_bse_object.bo);
+ mfc_context->mfc_indirect_pak_bse_object.bo = NULL;
+
+ for (i = 0; i < NUM_MFC_DMV_BUFFERS; i++){
+ if ( mfc_context->direct_mv_buffers[i].bo != NULL);
+ dri_bo_unreference(mfc_context->direct_mv_buffers[i].bo);
+ mfc_context->direct_mv_buffers[i].bo = NULL;
+ }
+
+ for (i = 0; i < MAX_MFC_REFERENCE_SURFACES; i++){
+ if (mfc_context->reference_surfaces[i].bo != NULL)
+ dri_bo_unreference(mfc_context->reference_surfaces[i].bo);
+ mfc_context->reference_surfaces[i].bo = NULL;
+ }
+
+ dri_bo_unreference(mfc_context->intra_row_store_scratch_buffer.bo);
+ bo = dri_bo_alloc(i965->intel.bufmgr,
+ "Buffer",
+ width_in_mbs * 64,
+ 64);
+ assert(bo);
+ mfc_context->intra_row_store_scratch_buffer.bo = bo;
+
+ dri_bo_unreference(mfc_context->macroblock_status_buffer.bo);
+ bo = dri_bo_alloc(i965->intel.bufmgr,
+ "Buffer",
+ width_in_mbs * height_in_mbs * 16,
+ 64);
+ assert(bo);
+ mfc_context->macroblock_status_buffer.bo = bo;
+
+ dri_bo_unreference(mfc_context->deblocking_filter_row_store_scratch_buffer.bo);
+ bo = dri_bo_alloc(i965->intel.bufmgr,
+ "Buffer",
+ 4 * width_in_mbs * 64, /* 4 * width_in_mbs * 64 */
+ 64);
+ assert(bo);
+ mfc_context->deblocking_filter_row_store_scratch_buffer.bo = bo;
+
+ dri_bo_unreference(mfc_context->bsd_mpc_row_store_scratch_buffer.bo);
+ bo = dri_bo_alloc(i965->intel.bufmgr,
+ "Buffer",
+ 128 * width_in_mbs, /* 2 * widht_in_mbs * 64 */
+ 0x1000);
+ assert(bo);
+ mfc_context->bsd_mpc_row_store_scratch_buffer.bo = bo;
+
+ dri_bo_unreference(mfc_context->mfc_batchbuffer_surface.bo);
+ mfc_context->mfc_batchbuffer_surface.bo = NULL;
+
+ dri_bo_unreference(mfc_context->aux_batchbuffer_surface.bo);
+ mfc_context->aux_batchbuffer_surface.bo = NULL;
+
+ if (mfc_context->aux_batchbuffer)
+ intel_batchbuffer_free(mfc_context->aux_batchbuffer);
+
+ mfc_context->aux_batchbuffer = intel_batchbuffer_new(&i965->intel, I915_EXEC_BSD,
+ slice_batchbuffer_size);
+ mfc_context->aux_batchbuffer_surface.bo = mfc_context->aux_batchbuffer->buffer;
+ dri_bo_reference(mfc_context->aux_batchbuffer_surface.bo);
+ mfc_context->aux_batchbuffer_surface.pitch = 16;
+ mfc_context->aux_batchbuffer_surface.num_blocks = mfc_context->aux_batchbuffer->size / 16;
+ mfc_context->aux_batchbuffer_surface.size_block = 16;
+
+ i965_gpe_context_init(ctx, &mfc_context->gpe_context);
+}
+
+static void gen6_mfc_avc_pipeline_picture_programing( VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
+{
+ struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
+
+ mfc_context->pipe_mode_select(ctx, MFX_FORMAT_AVC, encoder_context);
+ mfc_context->set_surface_state(ctx, encoder_context);
+ mfc_context->ind_obj_base_addr_state(ctx, encoder_context);
+ gen6_mfc_pipe_buf_addr_state(ctx, encoder_context);
+ gen6_mfc_bsp_buf_base_addr_state(ctx, encoder_context);
+ mfc_context->avc_img_state(ctx, encode_state, encoder_context);
+ mfc_context->avc_qm_state(ctx, encoder_context);
+ mfc_context->avc_fqm_state(ctx, encoder_context);
+ gen6_mfc_avc_directmode_state(ctx, encoder_context);
+ intel_mfc_avc_ref_idx_state(ctx, encode_state, encoder_context);
+}
+
+
+VAStatus
+gen6_mfc_run(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
+{
+ struct intel_batchbuffer *batch = encoder_context->base.batch;
+
+ intel_batchbuffer_flush(batch); //run the pipeline
+
+ return VA_STATUS_SUCCESS;
+}
+
+VAStatus
+gen6_mfc_stop(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context,
+ int *encoded_bits_size)
+{
+ VAStatus vaStatus = VA_STATUS_ERROR_UNKNOWN;
+ VAEncPictureParameterBufferH264 *pPicParameter = (VAEncPictureParameterBufferH264 *)encode_state->pic_param_ext->buffer;
+ VACodedBufferSegment *coded_buffer_segment;
+
+ vaStatus = i965_MapBuffer(ctx, pPicParameter->coded_buf, (void **)&coded_buffer_segment);
+ assert(vaStatus == VA_STATUS_SUCCESS);
+ *encoded_bits_size = coded_buffer_segment->size * 8;
+ i965_UnmapBuffer(ctx, pPicParameter->coded_buf);
+
+ return VA_STATUS_SUCCESS;
+}
+
+#if __SOFTWARE__
+
static int
gen6_mfc_avc_pak_object_intra(VADriverContextP ctx, int x, int y, int end_mb, int qp,unsigned int *msg,
struct intel_encoder_context *encoder_context,
- unsigned char target_mb_size, unsigned char max_mb_size)
+ unsigned char target_mb_size, unsigned char max_mb_size,
+ struct intel_batchbuffer *batch)
{
- struct intel_batchbuffer *batch = encoder_context->base.batch;
int len_in_dwords = 11;
+ if (batch == NULL)
+ batch = encoder_context->base.batch;
+
BEGIN_BCS_BATCH(batch, len_in_dwords);
OUT_BCS_BATCH(batch, MFC_AVC_PAK_OBJECT | (len_in_dwords - 2));
(1 << 17) | /* CbpDcV */
(msg[0] & 0xFFFF) );
- OUT_BCS_BATCH(batch, (0xFFFF<<16) | (y << 8) | x); /* Code Block Pattern for Y*/
+ OUT_BCS_BATCH(batch, (0xFFFF << 16) | (y << 8) | x); /* Code Block Pattern for Y*/
OUT_BCS_BATCH(batch, 0x000F000F); /* Code Block Pattern */
OUT_BCS_BATCH(batch, (0 << 27) | (end_mb << 26) | qp); /* Last MB */
return len_in_dwords;
}
-static int gen6_mfc_avc_pak_object_inter(VADriverContextP ctx, int x, int y, int end_mb, int qp, unsigned int offset,
- struct intel_encoder_context *encoder_context,
- unsigned char target_mb_size,unsigned char max_mb_size, int slice_type)
+static int
+gen6_mfc_avc_pak_object_inter(VADriverContextP ctx, int x, int y, int end_mb, int qp,
+ unsigned int *msg, unsigned int offset,
+ struct intel_encoder_context *encoder_context,
+ unsigned char target_mb_size,unsigned char max_mb_size, int slice_type,
+ struct intel_batchbuffer *batch)
{
- struct intel_batchbuffer *batch = encoder_context->base.batch;
+ struct gen6_vme_context *vme_context = encoder_context->vme_context;
int len_in_dwords = 11;
+ if (batch == NULL)
+ batch = encoder_context->base.batch;
+
BEGIN_BCS_BATCH(batch, len_in_dwords);
OUT_BCS_BATCH(batch, MFC_AVC_PAK_OBJECT | (len_in_dwords - 2));
- OUT_BCS_BATCH(batch, 32); /* 32 MV*/
+ OUT_BCS_BATCH(batch, msg[2]); /* 32 MV*/
OUT_BCS_BATCH(batch, offset);
- OUT_BCS_BATCH(batch,
- (1 << 24) | /* PackedMvNum, Debug*/
- (4 << 20) | /* 8 MV, SNB don't use it*/
- (1 << 19) | /* CbpDcY */
- (1 << 18) | /* CbpDcU */
- (1 << 17) | /* CbpDcV */
- (0 << 15) | /* Transform8x8Flag = 0*/
- (0 << 14) | /* Frame based*/
- (0 << 13) | /* Inter MB */
- (1 << 8) | /* MbType = P_L0_16x16 */
- (0 << 7) | /* MBZ for frame */
- (0 << 6) | /* MBZ */
- (2 << 4) | /* MBZ for inter*/
- (0 << 3) | /* MBZ */
- (0 << 2) | /* SkipMbFlag */
- (0 << 0)); /* InterMbMode */
+ OUT_BCS_BATCH(batch, msg[0]);
OUT_BCS_BATCH(batch, (0xFFFF<<16) | (y << 8) | x); /* Code Block Pattern for Y*/
OUT_BCS_BATCH(batch, 0x000F000F); /* Code Block Pattern */
/*Stuff for Inter MB*/
- OUT_BCS_BATCH(batch, 0x0);
- OUT_BCS_BATCH(batch, 0x0);
- OUT_BCS_BATCH(batch, 0x0);
+ OUT_BCS_BATCH(batch, msg[1]);
+ OUT_BCS_BATCH(batch, vme_context->ref_index_in_mb[0]);
+ OUT_BCS_BATCH(batch, vme_context->ref_index_in_mb[1]);
/*MaxSizeInWord and TargetSzieInWord*/
OUT_BCS_BATCH(batch, (max_mb_size << 24) |
return len_in_dwords;
}
-static void gen6_mfc_init(VADriverContextP ctx, struct intel_encoder_context *encoder_context)
+static void
+gen6_mfc_avc_pipeline_slice_programing(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context,
+ int slice_index,
+ struct intel_batchbuffer *slice_batch)
{
- struct i965_driver_data *i965 = i965_driver_data(ctx);
struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
- dri_bo *bo;
- int i;
+ struct gen6_vme_context *vme_context = encoder_context->vme_context;
+ VAEncSequenceParameterBufferH264 *pSequenceParameter = (VAEncSequenceParameterBufferH264 *)encode_state->seq_param_ext->buffer;
+ VAEncPictureParameterBufferH264 *pPicParameter = (VAEncPictureParameterBufferH264 *)encode_state->pic_param_ext->buffer;
+ VAEncSliceParameterBufferH264 *pSliceParameter = (VAEncSliceParameterBufferH264 *)encode_state->slice_params_ext[slice_index]->buffer;
+ unsigned int *msg = NULL, offset = 0;
+ int width_in_mbs = (mfc_context->surface_state.width + 15) / 16;
+ int height_in_mbs = (mfc_context->surface_state.height + 15) / 16;
+ int last_slice = (pSliceParameter->macroblock_address + pSliceParameter->num_macroblocks) == (width_in_mbs * height_in_mbs);
+ int i,x,y;
+ int qp = pPicParameter->pic_init_qp + pSliceParameter->slice_qp_delta;
+ unsigned int rate_control_mode = encoder_context->rate_control_mode;
+ unsigned int tail_data[] = { 0x0, 0x0 };
+ int slice_type = intel_avc_enc_slice_type_fixup(pSliceParameter->slice_type);
+ int is_intra = slice_type == SLICE_TYPE_I;
+ int qp_slice;
+
+ qp_slice = qp;
+ if (rate_control_mode == VA_RC_CBR) {
+ qp = mfc_context->bit_rate_control_context[slice_type].QpPrimeY;
+ if (encode_state->slice_header_index[slice_index] == 0) {
+ pSliceParameter->slice_qp_delta = qp - pPicParameter->pic_init_qp;
+ qp_slice = qp;
+ }
+ }
- /*Encode common setup for MFC*/
- dri_bo_unreference(mfc_context->post_deblocking_output.bo);
- mfc_context->post_deblocking_output.bo = NULL;
+ /* only support for 8-bit pixel bit-depth */
+ assert(pSequenceParameter->bit_depth_luma_minus8 == 0);
+ assert(pSequenceParameter->bit_depth_chroma_minus8 == 0);
+ assert(pPicParameter->pic_init_qp >= 0 && pPicParameter->pic_init_qp < 52);
+ assert(qp >= 0 && qp < 52);
- dri_bo_unreference(mfc_context->pre_deblocking_output.bo);
- mfc_context->pre_deblocking_output.bo = NULL;
+ gen6_mfc_avc_slice_state(ctx,
+ pPicParameter,
+ pSliceParameter,
+ encode_state, encoder_context,
+ (rate_control_mode == VA_RC_CBR), qp_slice, slice_batch);
- dri_bo_unreference(mfc_context->uncompressed_picture_source.bo);
- mfc_context->uncompressed_picture_source.bo = NULL;
+ if ( slice_index == 0)
+ intel_mfc_avc_pipeline_header_programing(ctx, encode_state, encoder_context, slice_batch);
- dri_bo_unreference(mfc_context->mfc_indirect_pak_bse_object.bo);
- mfc_context->mfc_indirect_pak_bse_object.bo = NULL;
+ intel_avc_slice_insert_packed_data(ctx, encode_state, encoder_context, slice_index, slice_batch);
- for (i = 0; i < NUM_MFC_DMV_BUFFERS; i++){
- if ( mfc_context->direct_mv_buffers[i].bo != NULL);
- dri_bo_unreference(mfc_context->direct_mv_buffers[i].bo);
- mfc_context->direct_mv_buffers[i].bo = NULL;
- }
+ dri_bo_map(vme_context->vme_output.bo , 1);
+ msg = (unsigned int *)vme_context->vme_output.bo->virtual;
- for (i = 0; i < MAX_MFC_REFERENCE_SURFACES; i++){
- if (mfc_context->reference_surfaces[i].bo != NULL)
- dri_bo_unreference(mfc_context->reference_surfaces[i].bo);
- mfc_context->reference_surfaces[i].bo = NULL;
+ if (is_intra) {
+ msg += pSliceParameter->macroblock_address * INTRA_VME_OUTPUT_IN_DWS;
+ } else {
+ msg += pSliceParameter->macroblock_address * INTER_VME_OUTPUT_IN_DWS;
+ msg += 32; /* the first 32 DWs are MVs */
+ offset = pSliceParameter->macroblock_address * INTER_VME_OUTPUT_IN_BYTES;
}
+
+ for (i = pSliceParameter->macroblock_address;
+ i < pSliceParameter->macroblock_address + pSliceParameter->num_macroblocks; i++) {
+ int last_mb = (i == (pSliceParameter->macroblock_address + pSliceParameter->num_macroblocks - 1) );
+ x = i % width_in_mbs;
+ y = i / width_in_mbs;
+
+ if (is_intra) {
+ assert(msg);
+ gen6_mfc_avc_pak_object_intra(ctx, x, y, last_mb, qp, msg, encoder_context, 0, 0, slice_batch);
+ msg += INTRA_VME_OUTPUT_IN_DWS;
+ } else {
+ if (msg[0] & INTRA_MB_FLAG_MASK) {
+ gen6_mfc_avc_pak_object_intra(ctx, x, y, last_mb, qp, msg, encoder_context, 0, 0, slice_batch);
+ } else {
+ gen6_mfc_avc_pak_object_inter(ctx, x, y, last_mb, qp, msg, offset, encoder_context, 0, 0, slice_type, slice_batch);
+ }
- dri_bo_unreference(mfc_context->intra_row_store_scratch_buffer.bo);
- bo = dri_bo_alloc(i965->intel.bufmgr,
- "Buffer",
- 128 * 64,
- 64);
- assert(bo);
- mfc_context->intra_row_store_scratch_buffer.bo = bo;
+ msg += INTER_VME_OUTPUT_IN_DWS;
+ offset += INTER_VME_OUTPUT_IN_BYTES;
+ }
+ }
+
+ dri_bo_unmap(vme_context->vme_output.bo);
- dri_bo_unreference(mfc_context->macroblock_status_buffer.bo);
- bo = dri_bo_alloc(i965->intel.bufmgr,
- "Buffer",
- 128*128*16,
- 64);
- assert(bo);
- mfc_context->macroblock_status_buffer.bo = bo;
+ if ( last_slice ) {
+ mfc_context->insert_object(ctx, encoder_context,
+ tail_data, 2, 8,
+ 2, 1, 1, 0, slice_batch);
+ } else {
+ mfc_context->insert_object(ctx, encoder_context,
+ tail_data, 1, 8,
+ 1, 1, 1, 0, slice_batch);
+ }
- dri_bo_unreference(mfc_context->deblocking_filter_row_store_scratch_buffer.bo);
- bo = dri_bo_alloc(i965->intel.bufmgr,
- "Buffer",
- 49152, /* 6 * 128 * 64 */
- 64);
- assert(bo);
- mfc_context->deblocking_filter_row_store_scratch_buffer.bo = bo;
- dri_bo_unreference(mfc_context->bsd_mpc_row_store_scratch_buffer.bo);
- bo = dri_bo_alloc(i965->intel.bufmgr,
- "Buffer",
- 12288, /* 1.5 * 128 * 64 */
- 0x1000);
- assert(bo);
- mfc_context->bsd_mpc_row_store_scratch_buffer.bo = bo;
}
-void gen6_mfc_avc_pipeline_programing(VADriverContextP ctx,
- struct encode_state *encode_state,
- struct intel_encoder_context *encoder_context)
+static dri_bo *
+gen6_mfc_avc_software_batchbuffer(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
{
- struct intel_batchbuffer *batch = encoder_context->base.batch;
+ struct i965_driver_data *i965 = i965_driver_data(ctx);
struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
- struct gen6_vme_context *vme_context = encoder_context->vme_context;
- VAEncSequenceParameterBufferH264 *pSequenceParameter = (VAEncSequenceParameterBufferH264 *)encode_state->seq_param_ext->buffer;
- VAEncPictureParameterBufferH264 *pPicParameter = (VAEncPictureParameterBufferH264 *)encode_state->pic_param_ext->buffer;
- VAEncSliceParameterBufferH264 *pSliceParameter = (VAEncSliceParameterBufferH264 *)encode_state->slice_params_ext[0]->buffer; /* FIXME: multi slices */
- VAEncH264DecRefPicMarkingBuffer *pDecRefPicMarking = NULL;
- unsigned int *msg = NULL, offset = 0;
- int emit_new_state = 1, object_len_in_bytes;
- int is_intra = pSliceParameter->slice_type == SLICE_TYPE_I;
- int width_in_mbs = (mfc_context->surface_state.width + 15) / 16;
- int height_in_mbs = (mfc_context->surface_state.height + 15) / 16;
- int x,y;
- int rate_control_mode = pSequenceParameter->rate_control_method;
- unsigned char target_mb_size = mfc_context->bit_rate_control_context[1-is_intra].TargetSizeInWord;
- unsigned char max_mb_size = mfc_context->bit_rate_control_context[1-is_intra].MaxSizeInWord;
- int qp = pPicParameter->pic_init_qp + pSliceParameter->slice_qp_delta;
- unsigned char *slice_header = NULL;
- int slice_header_length_in_bits = 0;
- unsigned int tail_data[] = { 0x0 };
-
- if (encode_state->dec_ref_pic_marking)
- pDecRefPicMarking = (VAEncH264DecRefPicMarkingBuffer *)encode_state->dec_ref_pic_marking->buffer;
+ struct intel_batchbuffer *batch;;
+ dri_bo *batch_bo;
+ int i;
- slice_header_length_in_bits = build_avc_slice_header(pSequenceParameter, pPicParameter, pSliceParameter, pDecRefPicMarking, &slice_header);
+ batch = mfc_context->aux_batchbuffer;
+ batch_bo = batch->buffer;
- if ( rate_control_mode == 0) {
- qp = mfc_context->bit_rate_control_context[1-is_intra].QpPrimeY;
+ for (i = 0; i < encode_state->num_slice_params_ext; i++) {
+ gen6_mfc_avc_pipeline_slice_programing(ctx, encode_state, encoder_context, i, batch);
}
- intel_batchbuffer_start_atomic_bcs(batch, 0x1000);
+ intel_batchbuffer_align(batch, 8);
- if (is_intra) {
- dri_bo_map(vme_context->vme_output.bo , 1);
- msg = (unsigned int *)vme_context->vme_output.bo->virtual;
- }
+ BEGIN_BCS_BATCH(batch, 2);
+ OUT_BCS_BATCH(batch, 0);
+ OUT_BCS_BATCH(batch, MI_BATCH_BUFFER_END);
+ ADVANCE_BCS_BATCH(batch);
- for (y = 0; y < height_in_mbs; y++) {
- for (x = 0; x < width_in_mbs; x++) {
- int last_mb = (y == (height_in_mbs-1)) && ( x == (width_in_mbs-1) );
-
- if (emit_new_state) {
- intel_batchbuffer_emit_mi_flush(batch);
- gen6_mfc_pipe_mode_select(ctx, encoder_context);
- gen6_mfc_surface_state(ctx, encoder_context);
- gen6_mfc_ind_obj_base_addr_state(ctx, encoder_context);
- gen6_mfc_pipe_buf_addr_state(ctx, encoder_context);
- gen6_mfc_bsp_buf_base_addr_state(ctx, encoder_context);
- gen6_mfc_avc_img_state(ctx, encode_state,encoder_context);
- gen6_mfc_avc_qm_state(ctx, encoder_context);
- gen6_mfc_avc_fqm_state(ctx, encoder_context);
- gen6_mfc_avc_directmode_state(ctx, encoder_context);
- gen6_mfc_avc_ref_idx_state(ctx, encoder_context);
- gen6_mfc_avc_slice_state(ctx, pSliceParameter->slice_type,
- encode_state, encoder_context,
- rate_control_mode == 0, pPicParameter->pic_init_qp + pSliceParameter->slice_qp_delta);
-
- if (encode_state->packed_header_data[VAEncPackedHeaderSPS]) {
- VAEncPackedHeaderParameterBuffer *param = NULL;
- unsigned int *header_data = (unsigned int *)encode_state->packed_header_data[VAEncPackedHeaderSPS]->buffer;
- unsigned int length_in_bits;
-
- assert(encode_state->packed_header_param[VAEncPackedHeaderSPS]);
- param = (VAEncPackedHeaderParameterBuffer *)encode_state->packed_header_param[VAEncPackedHeaderSPS]->buffer;
- length_in_bits = param->length_in_bits[0];
-
- gen6_mfc_avc_insert_object(ctx,
- encoder_context,
- header_data,
- ALIGN(length_in_bits, 32) >> 5,
- length_in_bits & 0x1f,
- param->skip_emulation_check_count,
- 0,
- 0,
- param->insert_emulation_bytes);
- }
+ dri_bo_reference(batch_bo);
- if (encode_state->packed_header_data[VAEncPackedHeaderPPS]) {
- VAEncPackedHeaderParameterBuffer *param = NULL;
- unsigned int *header_data = (unsigned int *)encode_state->packed_header_data[VAEncPackedHeaderPPS]->buffer;
- unsigned int length_in_bits;
-
- assert(encode_state->packed_header_param[VAEncPackedHeaderPPS]);
- param = (VAEncPackedHeaderParameterBuffer *)encode_state->packed_header_param[VAEncPackedHeaderPPS]->buffer;
- length_in_bits = param->length_in_bits[0];
-
- gen6_mfc_avc_insert_object(ctx,
- encoder_context,
- header_data,
- ALIGN(length_in_bits, 32) >> 5,
- length_in_bits & 0x1f,
- param->skip_emulation_check_count,
- 0,
- 0,
- param->insert_emulation_bytes);
- }
+ intel_batchbuffer_free(batch);
+ mfc_context->aux_batchbuffer = NULL;
- gen6_mfc_avc_insert_object(ctx, encoder_context,
- (unsigned int *)slice_header, ALIGN(slice_header_length_in_bits, 32) >> 5, slice_header_length_in_bits & 0x1f,
- 5, /* first 5 bytes are start code + nal unit type */
- 1, 0, 1);
- emit_new_state = 0;
- }
+ return batch_bo;
+}
- if (is_intra) {
- assert(msg);
- object_len_in_bytes = gen6_mfc_avc_pak_object_intra(ctx, x, y, last_mb, qp, msg, encoder_context,target_mb_size, max_mb_size);
- msg += 4;
- } else {
- object_len_in_bytes = gen6_mfc_avc_pak_object_inter(ctx, x, y, last_mb, qp, offset, encoder_context, target_mb_size, max_mb_size, pSliceParameter->slice_type);
- offset += 64;
- }
+#else
- if (intel_batchbuffer_check_free_space(batch, object_len_in_bytes) == 0) {
- assert(0);
- intel_batchbuffer_end_atomic(batch);
- intel_batchbuffer_flush(batch);
- emit_new_state = 1;
- intel_batchbuffer_start_atomic_bcs(batch, 0x1000);
- }
- }
- }
+static void
+gen6_mfc_batchbuffer_surfaces_input(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
- gen6_mfc_avc_insert_object(ctx, encoder_context,
- tail_data, sizeof(tail_data) >> 2, 32,
- sizeof(tail_data), 1, 1, 1);
+{
+ struct gen6_vme_context *vme_context = encoder_context->vme_context;
+ struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
- if (is_intra)
- dri_bo_unmap(vme_context->vme_output.bo);
+ assert(vme_context->vme_output.bo);
+ mfc_context->buffer_suface_setup(ctx,
+ &mfc_context->gpe_context,
+ &vme_context->vme_output,
+ BINDING_TABLE_OFFSET(BIND_IDX_VME_OUTPUT),
+ SURFACE_STATE_OFFSET(BIND_IDX_VME_OUTPUT));
+ assert(mfc_context->aux_batchbuffer_surface.bo);
+ mfc_context->buffer_suface_setup(ctx,
+ &mfc_context->gpe_context,
+ &mfc_context->aux_batchbuffer_surface,
+ BINDING_TABLE_OFFSET(BIND_IDX_MFC_SLICE_HEADER),
+ SURFACE_STATE_OFFSET(BIND_IDX_MFC_SLICE_HEADER));
+}
- free(slice_header);
+static void
+gen6_mfc_batchbuffer_surfaces_output(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
- intel_batchbuffer_end_atomic(batch);
+{
+ struct i965_driver_data *i965 = i965_driver_data(ctx);
+ struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
+ VAEncSequenceParameterBufferH264 *pSequenceParameter = (VAEncSequenceParameterBufferH264 *)encode_state->seq_param_ext->buffer;
+ int width_in_mbs = pSequenceParameter->picture_width_in_mbs;
+ int height_in_mbs = pSequenceParameter->picture_height_in_mbs;
+ mfc_context->mfc_batchbuffer_surface.num_blocks = width_in_mbs * height_in_mbs + encode_state->num_slice_params_ext * 8 + 1;
+ mfc_context->mfc_batchbuffer_surface.size_block = 16 * CMD_LEN_IN_OWORD; /* 3 OWORDs */
+ mfc_context->mfc_batchbuffer_surface.pitch = 16;
+ mfc_context->mfc_batchbuffer_surface.bo = dri_bo_alloc(i965->intel.bufmgr,
+ "MFC batchbuffer",
+ mfc_context->mfc_batchbuffer_surface.num_blocks * mfc_context->mfc_batchbuffer_surface.size_block,
+ 0x1000);
+ mfc_context->buffer_suface_setup(ctx,
+ &mfc_context->gpe_context,
+ &mfc_context->mfc_batchbuffer_surface,
+ BINDING_TABLE_OFFSET(BIND_IDX_MFC_BATCHBUFFER),
+ SURFACE_STATE_OFFSET(BIND_IDX_MFC_BATCHBUFFER));
}
-static void
-gen6_mfc_free_avc_surface(void **data)
+static void
+gen6_mfc_batchbuffer_surfaces_setup(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
{
- struct gen6_mfc_avc_surface_aux *avc_surface = *data;
-
- if (!avc_surface)
- return;
-
- dri_bo_unreference(avc_surface->dmv_top);
- avc_surface->dmv_top = NULL;
- dri_bo_unreference(avc_surface->dmv_bottom);
- avc_surface->dmv_bottom = NULL;
-
- free(avc_surface);
- *data = NULL;
+ gen6_mfc_batchbuffer_surfaces_input(ctx, encode_state, encoder_context);
+ gen6_mfc_batchbuffer_surfaces_output(ctx, encode_state, encoder_context);
}
-static void gen6_mfc_bit_rate_control_context_init(struct encode_state *encode_state,
- struct gen6_mfc_context *mfc_context)
+static void
+gen6_mfc_batchbuffer_idrt_setup(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
{
- VAEncSequenceParameterBufferH264 *pSequenceParameter = (VAEncSequenceParameterBufferH264 *)encode_state->seq_param_ext->buffer;
-
- int width_in_mbs = (mfc_context->surface_state.width + 15) / 16;
- int height_in_mbs = (mfc_context->surface_state.height + 15) / 16;
- float fps = pSequenceParameter->time_scale * 0.5 / pSequenceParameter->num_units_in_tick ;
- int inter_mb_size = pSequenceParameter->bits_per_second * 1.0 / (fps+4.0) / width_in_mbs / height_in_mbs;
- int intra_mb_size = inter_mb_size * 5.0;
+ struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
+ struct gen6_interface_descriptor_data *desc;
int i;
-
- mfc_context->bit_rate_control_context[0].target_mb_size = intra_mb_size;
- mfc_context->bit_rate_control_context[0].target_frame_size = intra_mb_size * width_in_mbs * height_in_mbs;
- mfc_context->bit_rate_control_context[1].target_mb_size = inter_mb_size;
- mfc_context->bit_rate_control_context[1].target_frame_size = inter_mb_size * width_in_mbs * height_in_mbs;
-
- for(i = 0 ; i < 2; i++) {
- mfc_context->bit_rate_control_context[i].QpPrimeY = 26;
- mfc_context->bit_rate_control_context[i].MaxQpNegModifier = 6;
- mfc_context->bit_rate_control_context[i].MaxQpPosModifier = 6;
- mfc_context->bit_rate_control_context[i].GrowInit = 6;
- mfc_context->bit_rate_control_context[i].GrowResistance = 4;
- mfc_context->bit_rate_control_context[i].ShrinkInit = 6;
- mfc_context->bit_rate_control_context[i].ShrinkResistance = 4;
-
- mfc_context->bit_rate_control_context[i].Correct[0] = 8;
- mfc_context->bit_rate_control_context[i].Correct[1] = 4;
- mfc_context->bit_rate_control_context[i].Correct[2] = 2;
- mfc_context->bit_rate_control_context[i].Correct[3] = 2;
- mfc_context->bit_rate_control_context[i].Correct[4] = 4;
- mfc_context->bit_rate_control_context[i].Correct[5] = 8;
+ dri_bo *bo;
+
+ bo = mfc_context->gpe_context.idrt.bo;
+ dri_bo_map(bo, 1);
+ assert(bo->virtual);
+ desc = bo->virtual;
+
+ for (i = 0; i < mfc_context->gpe_context.num_kernels; i++) {
+ struct i965_kernel *kernel;
+
+ kernel = &mfc_context->gpe_context.kernels[i];
+ assert(sizeof(*desc) == 32);
+
+ /*Setup the descritor table*/
+ memset(desc, 0, sizeof(*desc));
+ desc->desc0.kernel_start_pointer = (kernel->bo->offset >> 6);
+ desc->desc2.sampler_count = 0;
+ desc->desc2.sampler_state_pointer = 0;
+ desc->desc3.binding_table_entry_count = 2;
+ desc->desc3.binding_table_pointer = (BINDING_TABLE_OFFSET(0) >> 5);
+ desc->desc4.constant_urb_entry_read_offset = 0;
+ desc->desc4.constant_urb_entry_read_length = 4;
+
+ /*kernel start*/
+ dri_bo_emit_reloc(bo,
+ I915_GEM_DOMAIN_INSTRUCTION, 0,
+ 0,
+ i * sizeof(*desc) + offsetof(struct gen6_interface_descriptor_data, desc0),
+ kernel->bo);
+ desc++;
}
-
- mfc_context->bit_rate_control_context[0].TargetSizeInWord = (intra_mb_size + 16)/ 16;
- mfc_context->bit_rate_control_context[1].TargetSizeInWord = (inter_mb_size + 16)/ 16;
- mfc_context->bit_rate_control_context[0].MaxSizeInWord = mfc_context->bit_rate_control_context[0].TargetSizeInWord * 1.5;
- mfc_context->bit_rate_control_context[1].MaxSizeInWord = mfc_context->bit_rate_control_context[1].TargetSizeInWord * 1.5;
+ dri_bo_unmap(bo);
}
-static int gen6_mfc_bit_rate_control_context_update(struct encode_state *encode_state,
- struct gen6_mfc_context *mfc_context,
- int current_frame_size)
+static void
+gen6_mfc_batchbuffer_constant_setup(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
{
- VAEncSliceParameterBufferH264 *pSliceParameter = (VAEncSliceParameterBufferH264 *)encode_state->slice_params_ext[0]->buffer;
- int control_index = 1 - (pSliceParameter->slice_type == SLICE_TYPE_I);
- int oldQp = mfc_context->bit_rate_control_context[control_index].QpPrimeY;
-
- /*
- printf("conrol_index = %d, start_qp = %d, result = %d, target = %d\n", control_index,
- mfc_context->bit_rate_control_context[control_index].QpPrimeY, current_frame_size,
- mfc_context->bit_rate_control_context[control_index].target_frame_size );
- */
-
- if ( current_frame_size > mfc_context->bit_rate_control_context[control_index].target_frame_size * 4.0 ) {
- mfc_context->bit_rate_control_context[control_index].QpPrimeY += 4;
- } else if ( current_frame_size > mfc_context->bit_rate_control_context[control_index].target_frame_size * 2.0 ) {
- mfc_context->bit_rate_control_context[control_index].QpPrimeY += 3;
- } else if ( current_frame_size > mfc_context->bit_rate_control_context[control_index].target_frame_size * 1.50 ) {
- mfc_context->bit_rate_control_context[control_index].QpPrimeY += 2;
- } else if ( current_frame_size > mfc_context->bit_rate_control_context[control_index].target_frame_size * 1.20 ) {
- mfc_context->bit_rate_control_context[control_index].QpPrimeY ++;
- } else if (current_frame_size < mfc_context->bit_rate_control_context[control_index].target_frame_size * 0.30 ) {
- mfc_context->bit_rate_control_context[control_index].QpPrimeY -= 3;
- } else if (current_frame_size < mfc_context->bit_rate_control_context[control_index].target_frame_size * 0.50 ) {
- mfc_context->bit_rate_control_context[control_index].QpPrimeY -= 2;
- } else if (current_frame_size < mfc_context->bit_rate_control_context[control_index].target_frame_size * 0.80 ) {
- mfc_context->bit_rate_control_context[control_index].QpPrimeY --;
- }
+ struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
- if ( mfc_context->bit_rate_control_context[control_index].QpPrimeY > 51)
- mfc_context->bit_rate_control_context[control_index].QpPrimeY = 51;
- if ( mfc_context->bit_rate_control_context[control_index].QpPrimeY < 1)
- mfc_context->bit_rate_control_context[control_index].QpPrimeY = 1;
-
- if ( mfc_context->bit_rate_control_context[control_index].QpPrimeY != oldQp)
- return 0;
+ (void)mfc_context;
+}
- return 1;
+static void
+gen6_mfc_batchbuffer_emit_object_command(struct intel_batchbuffer *batch,
+ int index,
+ int head_offset,
+ int batchbuffer_offset,
+ int head_size,
+ int tail_size,
+ int number_mb_cmds,
+ int first_object,
+ int last_object,
+ int last_slice,
+ int mb_x,
+ int mb_y,
+ int width_in_mbs,
+ int qp,
+ unsigned int ref_index[2])
+{
+ BEGIN_BATCH(batch, 14);
+
+ OUT_BATCH(batch, CMD_MEDIA_OBJECT | (14 - 2));
+ OUT_BATCH(batch, index);
+ OUT_BATCH(batch, 0);
+ OUT_BATCH(batch, 0);
+ OUT_BATCH(batch, 0);
+ OUT_BATCH(batch, 0);
+
+ /*inline data */
+ OUT_BATCH(batch, head_offset);
+ OUT_BATCH(batch, batchbuffer_offset);
+ OUT_BATCH(batch,
+ head_size << 16 |
+ tail_size);
+ OUT_BATCH(batch,
+ number_mb_cmds << 16 |
+ first_object << 2 |
+ last_object << 1 |
+ last_slice);
+ OUT_BATCH(batch,
+ mb_y << 8 |
+ mb_x);
+ OUT_BATCH(batch,
+ qp << 16 |
+ width_in_mbs);
+ OUT_BATCH(batch, ref_index[0]);
+ OUT_BATCH(batch, ref_index[1]);
+
+ ADVANCE_BATCH(batch);
}
-static VAStatus gen6_mfc_avc_prepare(VADriverContextP ctx,
- struct encode_state *encode_state,
- struct intel_encoder_context *encoder_context)
+static void
+gen6_mfc_avc_batchbuffer_slice_command(VADriverContextP ctx,
+ struct intel_encoder_context *encoder_context,
+ VAEncSliceParameterBufferH264 *slice_param,
+ int head_offset,
+ unsigned short head_size,
+ unsigned short tail_size,
+ int batchbuffer_offset,
+ int qp,
+ int last_slice)
{
- struct i965_driver_data *i965 = i965_driver_data(ctx);
+ struct intel_batchbuffer *batch = encoder_context->base.batch;
struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
- struct object_surface *obj_surface;
- struct object_buffer *obj_buffer;
- struct gen6_mfc_avc_surface_aux* gen6_avc_surface;
- dri_bo *bo;
- VAEncPictureParameterBufferH264 *pPicParameter = (VAEncPictureParameterBufferH264 *)encode_state->pic_param_ext->buffer;
- VAStatus vaStatus = VA_STATUS_SUCCESS;
+ struct gen6_vme_context *vme_context = encoder_context->vme_context;
+ int width_in_mbs = (mfc_context->surface_state.width + 15) / 16;
+ int total_mbs = slice_param->num_macroblocks;
+ int number_mb_cmds = 128;
+ int starting_mb = 0;
+ int last_object = 0;
+ int first_object = 1;
int i;
+ int mb_x, mb_y;
+ int index = (slice_param->slice_type == SLICE_TYPE_I) ? MFC_BATCHBUFFER_AVC_INTRA : MFC_BATCHBUFFER_AVC_INTER;
+
+ for (i = 0; i < total_mbs / number_mb_cmds; i++) {
+ last_object = (total_mbs - starting_mb) == number_mb_cmds;
+ mb_x = (slice_param->macroblock_address + starting_mb) % width_in_mbs;
+ mb_y = (slice_param->macroblock_address + starting_mb) / width_in_mbs;
+ assert(mb_x <= 255 && mb_y <= 255);
+
+ starting_mb += number_mb_cmds;
+
+ gen6_mfc_batchbuffer_emit_object_command(batch,
+ index,
+ head_offset,
+ batchbuffer_offset,
+ head_size,
+ tail_size,
+ number_mb_cmds,
+ first_object,
+ last_object,
+ last_slice,
+ mb_x,
+ mb_y,
+ width_in_mbs,
+ qp,
+ vme_context->ref_index_in_mb);
+
+ if (first_object) {
+ head_offset += head_size;
+ batchbuffer_offset += head_size;
+ }
+
+ if (last_object) {
+ head_offset += tail_size;
+ batchbuffer_offset += tail_size;
+ }
+
+ batchbuffer_offset += number_mb_cmds * CMD_LEN_IN_OWORD;
- /*Setup all the input&output object*/
-
- /* Setup current frame and current direct mv buffer*/
- obj_surface = SURFACE(pPicParameter->CurrPic.picture_id);
- assert(obj_surface);
- i965_check_alloc_surface_bo(ctx, obj_surface, 1, VA_FOURCC('N','V','1','2'));
- if ( obj_surface->private_data == NULL) {
- gen6_avc_surface = calloc(sizeof(struct gen6_mfc_avc_surface_aux), 1);
- gen6_avc_surface->dmv_top =
- dri_bo_alloc(i965->intel.bufmgr,
- "Buffer",
- 68*8192,
- 64);
- gen6_avc_surface->dmv_bottom =
- dri_bo_alloc(i965->intel.bufmgr,
- "Buffer",
- 68*8192,
- 64);
- assert(gen6_avc_surface->dmv_top);
- assert(gen6_avc_surface->dmv_bottom);
- obj_surface->private_data = (void *)gen6_avc_surface;
- obj_surface->free_private_data = (void *)gen6_mfc_free_avc_surface;
+ first_object = 0;
}
- gen6_avc_surface = (struct gen6_mfc_avc_surface_aux*) obj_surface->private_data;
- mfc_context->direct_mv_buffers[NUM_MFC_DMV_BUFFERS - 2].bo = gen6_avc_surface->dmv_top;
- mfc_context->direct_mv_buffers[NUM_MFC_DMV_BUFFERS - 1].bo = gen6_avc_surface->dmv_bottom;
- dri_bo_reference(gen6_avc_surface->dmv_top);
- dri_bo_reference(gen6_avc_surface->dmv_bottom);
-
- mfc_context->post_deblocking_output.bo = obj_surface->bo;
- dri_bo_reference(mfc_context->post_deblocking_output.bo);
-
- mfc_context->surface_state.width = obj_surface->orig_width;
- mfc_context->surface_state.height = obj_surface->orig_height;
- mfc_context->surface_state.w_pitch = obj_surface->width;
- mfc_context->surface_state.h_pitch = obj_surface->height;
-
- /* Setup reference frames and direct mv buffers*/
- for(i = 0; i < MAX_MFC_REFERENCE_SURFACES; i++) {
- if ( pPicParameter->ReferenceFrames[i].picture_id != VA_INVALID_ID ) {
- obj_surface = SURFACE(pPicParameter->ReferenceFrames[i].picture_id);
- assert(obj_surface);
- if (obj_surface->bo != NULL) {
- mfc_context->reference_surfaces[i].bo = obj_surface->bo;
- dri_bo_reference(obj_surface->bo);
- }
- /* Check DMV buffer */
- if ( obj_surface->private_data == NULL) {
-
- gen6_avc_surface = calloc(sizeof(struct gen6_mfc_avc_surface_aux), 1);
- gen6_avc_surface->dmv_top =
- dri_bo_alloc(i965->intel.bufmgr,
- "Buffer",
- 68*8192,
- 64);
- gen6_avc_surface->dmv_bottom =
- dri_bo_alloc(i965->intel.bufmgr,
- "Buffer",
- 68*8192,
- 64);
- assert(gen6_avc_surface->dmv_top);
- assert(gen6_avc_surface->dmv_bottom);
- obj_surface->private_data = gen6_avc_surface;
- obj_surface->free_private_data = gen6_mfc_free_avc_surface;
- }
-
- gen6_avc_surface = (struct gen6_mfc_avc_surface_aux*) obj_surface->private_data;
- /* Setup DMV buffer */
- mfc_context->direct_mv_buffers[i*2].bo = gen6_avc_surface->dmv_top;
- mfc_context->direct_mv_buffers[i*2+1].bo = gen6_avc_surface->dmv_bottom;
- dri_bo_reference(gen6_avc_surface->dmv_top);
- dri_bo_reference(gen6_avc_surface->dmv_bottom);
- } else {
- break;
+
+ if (!last_object) {
+ last_object = 1;
+ number_mb_cmds = total_mbs % number_mb_cmds;
+ mb_x = (slice_param->macroblock_address + starting_mb) % width_in_mbs;
+ mb_y = (slice_param->macroblock_address + starting_mb) / width_in_mbs;
+ assert(mb_x <= 255 && mb_y <= 255);
+ starting_mb += number_mb_cmds;
+
+ gen6_mfc_batchbuffer_emit_object_command(batch,
+ index,
+ head_offset,
+ batchbuffer_offset,
+ head_size,
+ tail_size,
+ number_mb_cmds,
+ first_object,
+ last_object,
+ last_slice,
+ mb_x,
+ mb_y,
+ width_in_mbs,
+ qp,
+ vme_context->ref_index_in_mb);
+ }
+}
+
+/*
+ * return size in Owords (16bytes)
+ */
+static int
+gen6_mfc_avc_batchbuffer_slice(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context,
+ int slice_index,
+ int batchbuffer_offset)
+{
+ struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
+ struct intel_batchbuffer *slice_batch = mfc_context->aux_batchbuffer;
+ VAEncSequenceParameterBufferH264 *pSequenceParameter = (VAEncSequenceParameterBufferH264 *)encode_state->seq_param_ext->buffer;
+ VAEncPictureParameterBufferH264 *pPicParameter = (VAEncPictureParameterBufferH264 *)encode_state->pic_param_ext->buffer;
+ VAEncSliceParameterBufferH264 *pSliceParameter = (VAEncSliceParameterBufferH264 *)encode_state->slice_params_ext[slice_index]->buffer;
+ int width_in_mbs = (mfc_context->surface_state.width + 15) / 16;
+ int height_in_mbs = (mfc_context->surface_state.height + 15) / 16;
+ int last_slice = (pSliceParameter->macroblock_address + pSliceParameter->num_macroblocks) == (width_in_mbs * height_in_mbs);
+ int qp = pPicParameter->pic_init_qp + pSliceParameter->slice_qp_delta;
+ unsigned int rate_control_mode = encoder_context->rate_control_mode;
+ unsigned int tail_data[] = { 0x0, 0x0 };
+ long head_offset;
+ int old_used = intel_batchbuffer_used_size(slice_batch), used;
+ unsigned short head_size, tail_size;
+ int slice_type = intel_avc_enc_slice_type_fixup(pSliceParameter->slice_type);
+ int qp_slice;
+
+ qp_slice = qp;
+ if (rate_control_mode == VA_RC_CBR) {
+ qp = mfc_context->bit_rate_control_context[slice_type].QpPrimeY;
+ if (encode_state->slice_header_index[slice_index] == 0) {
+ pSliceParameter->slice_qp_delta = qp - pPicParameter->pic_init_qp;
+ /* Use the adjusted qp when slice_header is generated by driver */
+ qp_slice = qp;
}
}
-
- obj_surface = SURFACE(encode_state->current_render_target);
- assert(obj_surface && obj_surface->bo);
- mfc_context->uncompressed_picture_source.bo = obj_surface->bo;
- dri_bo_reference(mfc_context->uncompressed_picture_source.bo);
-
- obj_buffer = BUFFER (pPicParameter->CodedBuf); /* FIXME: fix this later */
- bo = obj_buffer->buffer_store->bo;
- assert(bo);
- mfc_context->mfc_indirect_pak_bse_object.bo = bo;
- mfc_context->mfc_indirect_pak_bse_object.offset = ALIGN(sizeof(VACodedBufferSegment), 64);
- mfc_context->mfc_indirect_pak_bse_object.end_offset = ALIGN (obj_buffer->size_element - 0x1000, 0x1000);
- dri_bo_reference(mfc_context->mfc_indirect_pak_bse_object.bo);
-
- /*Programing bit rate control */
- if ( mfc_context->bit_rate_control_context[0].MaxSizeInWord == 0 )
- gen6_mfc_bit_rate_control_context_init(encode_state, mfc_context);
-
- /*Programing bcs pipeline*/
- gen6_mfc_avc_pipeline_programing(ctx, encode_state, encoder_context); //filling the pipeline
-
- return vaStatus;
+
+ /* only support for 8-bit pixel bit-depth */
+ assert(pSequenceParameter->bit_depth_luma_minus8 == 0);
+ assert(pSequenceParameter->bit_depth_chroma_minus8 == 0);
+ assert(pPicParameter->pic_init_qp >= 0 && pPicParameter->pic_init_qp < 52);
+ assert(qp >= 0 && qp < 52);
+
+ head_offset = old_used / 16;
+ gen6_mfc_avc_slice_state(ctx,
+ pPicParameter,
+ pSliceParameter,
+ encode_state,
+ encoder_context,
+ (rate_control_mode == VA_RC_CBR),
+ qp_slice,
+ slice_batch);
+
+ if (slice_index == 0)
+ intel_mfc_avc_pipeline_header_programing(ctx, encode_state, encoder_context, slice_batch);
+
+ intel_avc_slice_insert_packed_data(ctx, encode_state, encoder_context, slice_index, slice_batch);
+
+ intel_batchbuffer_align(slice_batch, 16); /* aligned by an Oword */
+ used = intel_batchbuffer_used_size(slice_batch);
+ head_size = (used - old_used) / 16;
+ old_used = used;
+
+ /* tail */
+ if (last_slice) {
+ mfc_context->insert_object(ctx,
+ encoder_context,
+ tail_data,
+ 2,
+ 8,
+ 2,
+ 1,
+ 1,
+ 0,
+ slice_batch);
+ } else {
+ mfc_context->insert_object(ctx,
+ encoder_context,
+ tail_data,
+ 1,
+ 8,
+ 1,
+ 1,
+ 1,
+ 0,
+ slice_batch);
+ }
+
+ intel_batchbuffer_align(slice_batch, 16); /* aligned by an Oword */
+ used = intel_batchbuffer_used_size(slice_batch);
+ tail_size = (used - old_used) / 16;
+
+
+ gen6_mfc_avc_batchbuffer_slice_command(ctx,
+ encoder_context,
+ pSliceParameter,
+ head_offset,
+ head_size,
+ tail_size,
+ batchbuffer_offset,
+ qp,
+ last_slice);
+
+ return head_size + tail_size + pSliceParameter->num_macroblocks * CMD_LEN_IN_OWORD;
}
-static VAStatus gen6_mfc_run(VADriverContextP ctx,
- struct encode_state *encode_state,
- struct intel_encoder_context *encoder_context)
+static void
+gen6_mfc_avc_batchbuffer_pipeline(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
{
+ struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
struct intel_batchbuffer *batch = encoder_context->base.batch;
+ int i, size, offset = 0;
+ intel_batchbuffer_start_atomic(batch, 0x4000);
+ gen6_gpe_pipeline_setup(ctx, &mfc_context->gpe_context, batch);
- intel_batchbuffer_flush(batch); //run the pipeline
+ for ( i = 0; i < encode_state->num_slice_params_ext; i++) {
+ size = gen6_mfc_avc_batchbuffer_slice(ctx, encode_state, encoder_context, i, offset);
+ offset += size;
+ }
- return VA_STATUS_SUCCESS;
+ intel_batchbuffer_end_atomic(batch);
+ intel_batchbuffer_flush(batch);
}
-static VAStatus gen6_mfc_stop(VADriverContextP ctx,
- struct encode_state *encode_state,
- struct intel_encoder_context *encoder_context,
- int *encoded_bits_size)
+static void
+gen6_mfc_build_avc_batchbuffer(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
+{
+ gen6_mfc_batchbuffer_surfaces_setup(ctx, encode_state, encoder_context);
+ gen6_mfc_batchbuffer_idrt_setup(ctx, encode_state, encoder_context);
+ gen6_mfc_batchbuffer_constant_setup(ctx, encode_state, encoder_context);
+ gen6_mfc_avc_batchbuffer_pipeline(ctx, encode_state, encoder_context);
+}
+
+static dri_bo *
+gen6_mfc_avc_hardware_batchbuffer(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
{
struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
- unsigned int *status_mem;
- unsigned int buffer_size_bits = 0;
- int width_in_mbs = (mfc_context->surface_state.width + 15) / 16;
- int height_in_mbs = (mfc_context->surface_state.height + 15) / 16;
- int i;
- dri_bo_map(mfc_context->macroblock_status_buffer.bo, 1);
- status_mem = (unsigned int *)mfc_context->macroblock_status_buffer.bo->virtual;
- //Detecting encoder buffer size and bit rate control result
- for(i = 0; i < width_in_mbs * height_in_mbs; i++) {
- unsigned short current_mb = status_mem[1] >> 16;
- buffer_size_bits += current_mb;
- status_mem += 4;
- }
- dri_bo_unmap(mfc_context->macroblock_status_buffer.bo);
+ gen6_mfc_build_avc_batchbuffer(ctx, encode_state, encoder_context);
+ dri_bo_reference(mfc_context->mfc_batchbuffer_surface.bo);
- *encoded_bits_size = buffer_size_bits;
+ return mfc_context->mfc_batchbuffer_surface.bo;
+}
- return VA_STATUS_SUCCESS;
+#endif
+
+
+static void
+gen6_mfc_avc_pipeline_programing(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
+{
+ struct intel_batchbuffer *batch = encoder_context->base.batch;
+ dri_bo *slice_batch_bo;
+
+ if ( intel_mfc_interlace_check(ctx, encode_state, encoder_context) ) {
+ fprintf(stderr, "Current VA driver don't support interlace mode!\n");
+ assert(0);
+ return;
+ }
+
+#if __SOFTWARE__
+ slice_batch_bo = gen6_mfc_avc_software_batchbuffer(ctx, encode_state, encoder_context);
+#else
+ slice_batch_bo = gen6_mfc_avc_hardware_batchbuffer(ctx, encode_state, encoder_context);
+#endif
+
+ // begin programing
+ intel_batchbuffer_start_atomic_bcs(batch, 0x4000);
+ intel_batchbuffer_emit_mi_flush(batch);
+
+ // picture level programing
+ gen6_mfc_avc_pipeline_picture_programing(ctx, encode_state, encoder_context);
+
+ BEGIN_BCS_BATCH(batch, 2);
+ OUT_BCS_BATCH(batch, MI_BATCH_BUFFER_START | (1 << 8));
+ OUT_BCS_RELOC(batch,
+ slice_batch_bo,
+ I915_GEM_DOMAIN_COMMAND, 0,
+ 0);
+ ADVANCE_BCS_BATCH(batch);
+
+ // end programing
+ intel_batchbuffer_end_atomic(batch);
+
+ dri_bo_unreference(slice_batch_bo);
}
-static VAStatus
+VAStatus
gen6_mfc_avc_encode_picture(VADriverContextP ctx,
struct encode_state *encode_state,
struct intel_encoder_context *encoder_context)
{
- VAEncSequenceParameterBufferH264 *pSequenceParameter = (VAEncSequenceParameterBufferH264 *)encode_state->seq_param_ext->buffer;
struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
- int rate_control_mode = pSequenceParameter->rate_control_method;
- int MAX_CBR_INTERATE = 4;
+ unsigned int rate_control_mode = encoder_context->rate_control_mode;
int current_frame_bits_size;
- int i;
+ int sts;
- for(i = 0; i < MAX_CBR_INTERATE; i++) {
- gen6_mfc_init(ctx, encoder_context);
- gen6_mfc_avc_prepare(ctx, encode_state, encoder_context);
+ for (;;) {
+ gen6_mfc_init(ctx, encode_state, encoder_context);
+ intel_mfc_avc_prepare(ctx, encode_state, encoder_context);
+ /*Programing bcs pipeline*/
+ gen6_mfc_avc_pipeline_programing(ctx, encode_state, encoder_context); //filling the pipeline
gen6_mfc_run(ctx, encode_state, encoder_context);
- gen6_mfc_stop(ctx, encode_state, encoder_context, ¤t_frame_bits_size);
- if ( rate_control_mode == 0) {
- if ( gen6_mfc_bit_rate_control_context_update( encode_state, mfc_context, current_frame_bits_size) )
+ if (rate_control_mode == VA_RC_CBR /*|| rate_control_mode == VA_RC_VBR*/) {
+ gen6_mfc_stop(ctx, encode_state, encoder_context, ¤t_frame_bits_size);
+ sts = intel_mfc_brc_postpack(encode_state, mfc_context, current_frame_bits_size);
+ if (sts == BRC_NO_HRD_VIOLATION) {
+ intel_mfc_hrd_context_update(encode_state, mfc_context);
break;
+ }
+ else if (sts == BRC_OVERFLOW_WITH_MIN_QP || sts == BRC_UNDERFLOW_WITH_MAX_QP) {
+ if (!mfc_context->hrd.violation_noted) {
+ fprintf(stderr, "Unrepairable %s!\n", (sts == BRC_OVERFLOW_WITH_MIN_QP)? "overflow": "underflow");
+ mfc_context->hrd.violation_noted = 1;
+ }
+ return VA_STATUS_SUCCESS;
+ }
} else {
break;
}
return VA_STATUS_SUCCESS;
}
-static VAStatus
+VAStatus
gen6_mfc_pipeline(VADriverContextP ctx,
VAProfile profile,
struct encode_state *encode_state,
VAStatus vaStatus;
switch (profile) {
- case VAProfileH264Baseline:
+ case VAProfileH264ConstrainedBaseline:
+ case VAProfileH264Main:
+ case VAProfileH264High:
vaStatus = gen6_mfc_avc_encode_picture(ctx, encode_state, encoder_context);
break;
return vaStatus;
}
-static void
+void
gen6_mfc_context_destroy(void *context)
{
struct gen6_mfc_context *mfc_context = context;
mfc_context->reference_surfaces[i].bo = NULL;
}
+ i965_gpe_context_destroy(&mfc_context->gpe_context);
+
+ dri_bo_unreference(mfc_context->mfc_batchbuffer_surface.bo);
+ mfc_context->mfc_batchbuffer_surface.bo = NULL;
+
+ dri_bo_unreference(mfc_context->aux_batchbuffer_surface.bo);
+ mfc_context->aux_batchbuffer_surface.bo = NULL;
+
+ if (mfc_context->aux_batchbuffer)
+ intel_batchbuffer_free(mfc_context->aux_batchbuffer);
+
+ mfc_context->aux_batchbuffer = NULL;
+
free(mfc_context);
}
Bool gen6_mfc_context_init(VADriverContextP ctx, struct intel_encoder_context *encoder_context)
{
- encoder_context->mfc_context = calloc(1, sizeof(struct gen6_mfc_context));
+ struct gen6_mfc_context *mfc_context = calloc(1, sizeof(struct gen6_mfc_context));
+
+ if (!mfc_context)
+ return False;
+
+ mfc_context->gpe_context.surface_state_binding_table.length = (SURFACE_STATE_PADDED_SIZE + sizeof(unsigned int)) * MAX_MEDIA_SURFACES_GEN6;
+
+ mfc_context->gpe_context.idrt.max_entries = MAX_GPE_KERNELS;
+ mfc_context->gpe_context.idrt.entry_size = sizeof(struct gen6_interface_descriptor_data);
+
+ mfc_context->gpe_context.curbe.length = 32 * 4;
+
+ mfc_context->gpe_context.vfe_state.max_num_threads = 60 - 1;
+ mfc_context->gpe_context.vfe_state.num_urb_entries = 16;
+ mfc_context->gpe_context.vfe_state.gpgpu_mode = 0;
+ mfc_context->gpe_context.vfe_state.urb_entry_size = 59 - 1;
+ mfc_context->gpe_context.vfe_state.curbe_allocation_size = 37 - 1;
+
+ i965_gpe_load_kernels(ctx,
+ &mfc_context->gpe_context,
+ gen6_mfc_kernels,
+ NUM_MFC_KERNEL);
+
+ mfc_context->pipe_mode_select = gen6_mfc_pipe_mode_select;
+ mfc_context->set_surface_state = gen6_mfc_surface_state;
+ mfc_context->ind_obj_base_addr_state = gen6_mfc_ind_obj_base_addr_state;
+ mfc_context->avc_img_state = gen6_mfc_avc_img_state;
+ mfc_context->avc_qm_state = gen6_mfc_avc_qm_state;
+ mfc_context->avc_fqm_state = gen6_mfc_avc_fqm_state;
+ mfc_context->insert_object = gen6_mfc_avc_insert_object;
+ mfc_context->buffer_suface_setup = i965_gpe_buffer_suface_setup;
+
+ encoder_context->mfc_context = mfc_context;
encoder_context->mfc_context_destroy = gen6_mfc_context_destroy;
encoder_context->mfc_pipeline = gen6_mfc_pipeline;
+ encoder_context->mfc_brc_prepare = intel_mfc_brc_prepare;
return True;
}
projects / platform / upstream / libva-intel-driver.git / blobdiff