--- /dev/null
+/*
+ * Copyright © 2012 Intel Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ * Zhao Yakui <yakui.zhao@intel.com>
+ * Xiang Haihao <haihao.xiang@intel.com>
+ *
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <math.h>
+#include <assert.h>
+
+#include "intel_batchbuffer.h"
+#include "i965_defines.h"
+#include "i965_structs.h"
+#include "i965_drv_video.h"
+#include "i965_encoder.h"
+#include "i965_encoder_utils.h"
+#include "gen6_mfc.h"
+#include "gen6_vme.h"
+
+static const uint32_t gen75_mfc_batchbuffer_avc_intra[][4] = {
+#include "shaders/utils/mfc_batchbuffer_avc_intra.g7b"
+};
+
+static const uint32_t gen75_mfc_batchbuffer_avc_inter[][4] = {
+#include "shaders/utils/mfc_batchbuffer_avc_inter.g7b"
+};
+
+static struct i965_kernel gen75_mfc_kernels[] = {
+ {
+ "MFC AVC INTRA BATCHBUFFER ",
+ MFC_BATCHBUFFER_AVC_INTRA,
+ gen75_mfc_batchbuffer_avc_intra,
+ sizeof(gen75_mfc_batchbuffer_avc_intra),
+ NULL
+ },
+
+ {
+ "MFC AVC INTER BATCHBUFFER ",
+ MFC_BATCHBUFFER_AVC_INTER,
+ gen75_mfc_batchbuffer_avc_inter,
+ sizeof(gen75_mfc_batchbuffer_avc_inter),
+ NULL
+ },
+};
+
+static void
+gen75_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_MPEG2 ||
+ standard_select == MFX_FORMAT_AVC);
+
+ BEGIN_BCS_BATCH(batch, 5);
+
+ OUT_BCS_BATCH(batch, MFX_PIPE_MODE_SELECT | (5 - 2));
+ OUT_BCS_BATCH(batch,
+ (MFX_LONG_MODE << 17) | /* Must be long format for encoder */
+ (MFD_MODE_VLD << 15) | /* VLD mode */
+ (1 << 10) | /* Stream-Out Enable */
+ ((!!mfc_context->post_deblocking_output.bo) << 9) | /* Post Deblocking Output */
+ ((!!mfc_context->pre_deblocking_output.bo) << 8) | /* Pre Deblocking Output */
+ (0 << 8) | /* Pre Deblocking Output */
+ (0 << 5) | /* not in stitch mode */
+ (1 << 4) | /* encoding mode */
+ (standard_select << 0)); /* standard select: avc or mpeg2 */
+ OUT_BCS_BATCH(batch,
+ (0 << 7) | /* expand NOA bus flag */
+ (0 << 6) | /* disable slice-level clock gating */
+ (0 << 5) | /* disable clock gating for NOA */
+ (0 << 4) | /* terminate if AVC motion and POC table error occurs */
+ (0 << 3) | /* terminate if AVC mbdata error occurs */
+ (0 << 2) | /* terminate if AVC CABAC/CAVLC decode error occurs */
+ (0 << 1) |
+ (0 << 0));
+ OUT_BCS_BATCH(batch, 0);
+ OUT_BCS_BATCH(batch, 0);
+
+ ADVANCE_BCS_BATCH(batch);
+}
+
+static void
+gen75_mfc_surface_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;
+
+ BEGIN_BCS_BATCH(batch, 6);
+
+ OUT_BCS_BATCH(batch, MFX_SURFACE_STATE | (6 - 2));
+ OUT_BCS_BATCH(batch, 0);
+ OUT_BCS_BATCH(batch,
+ ((mfc_context->surface_state.height - 1) << 18) |
+ ((mfc_context->surface_state.width - 1) << 4));
+ OUT_BCS_BATCH(batch,
+ (MFX_SURFACE_PLANAR_420_8 << 28) | /* 420 planar YUV surface */
+ (1 << 27) | /* must be 1 for interleave U/V, hardware requirement */
+ (0 << 22) | /* surface object control state, FIXME??? */
+ ((mfc_context->surface_state.w_pitch - 1) << 3) | /* pitch */
+ (0 << 2) | /* must be 0 for interleave U/V */
+ (1 << 1) | /* must be tiled */
+ (I965_TILEWALK_YMAJOR << 0)); /* tile walk, TILEWALK_YMAJOR */
+ OUT_BCS_BATCH(batch,
+ (0 << 16) | /* must be 0 for interleave U/V */
+ (mfc_context->surface_state.h_pitch)); /* y offset for U(cb) */
+ OUT_BCS_BATCH(batch, 0);
+
+ ADVANCE_BCS_BATCH(batch);
+}
+
+static void
+gen75_mfc_ind_obj_base_addr_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;
+ struct gen6_vme_context *vme_context = encoder_context->vme_context;
+
+ BEGIN_BCS_BATCH(batch, 11);
+
+ OUT_BCS_BATCH(batch, MFX_IND_OBJ_BASE_ADDR_STATE | (11 - 2));
+ OUT_BCS_BATCH(batch, 0);
+ OUT_BCS_BATCH(batch, 0);
+ /* MFX Indirect MV Object Base Address */
+ OUT_BCS_RELOC(batch, vme_context->vme_output.bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 0);
+ OUT_BCS_BATCH(batch, 0x80000000); /* must set, up to 2G */
+ OUT_BCS_BATCH(batch, 0);
+ OUT_BCS_BATCH(batch, 0);
+ OUT_BCS_BATCH(batch, 0);
+ OUT_BCS_BATCH(batch, 0);
+ /*MFC Indirect PAK-BSE Object Base Address for Encoder*/
+ OUT_BCS_RELOC(batch,
+ mfc_context->mfc_indirect_pak_bse_object.bo,
+ I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
+ 0);
+ OUT_BCS_RELOC(batch,
+ mfc_context->mfc_indirect_pak_bse_object.bo,
+ I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
+ mfc_context->mfc_indirect_pak_bse_object.end_offset);
+
+ ADVANCE_BCS_BATCH(batch);
+}
+
+static void
+gen75_mfc_avc_img_state(VADriverContextP ctx, struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
+{
+ struct intel_batchbuffer *batch = encoder_context->base.batch;
+ struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
+ VAEncPictureParameterBufferH264 *pPicParameter = (VAEncPictureParameterBufferH264 *)encode_state->pic_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;
+
+ BEGIN_BCS_BATCH(batch, 16);
+
+ OUT_BCS_BATCH(batch, MFX_AVC_IMG_STATE | (16 - 2));
+ OUT_BCS_BATCH(batch,
+ ((width_in_mbs * height_in_mbs) & 0xFFFF));
+ OUT_BCS_BATCH(batch,
+ ((height_in_mbs - 1) << 16) |
+ ((width_in_mbs - 1) << 0));
+ OUT_BCS_BATCH(batch,
+ (0 << 24) | /* Second Chroma QP Offset */
+ (0 << 16) | /* Chroma QP Offset */
+ (0 << 14) | /* Max-bit conformance Intra flag */
+ (0 << 13) | /* Max Macroblock size conformance Inter flag */
+ (pPicParameter->pic_fields.bits.weighted_pred_flag << 12) | /*Weighted_Pred_Flag */
+ (pPicParameter->pic_fields.bits.weighted_bipred_idc << 10) | /* Weighted_BiPred_Idc */
+ (0 << 8) | /* FIXME: Image Structure */
+ (0 << 0) ); /* Current Decoed Image Frame Store ID, reserved in Encode mode */
+ OUT_BCS_BATCH(batch,
+ (0 << 16) | /* Mininum Frame size */
+ (0 << 15) | /* Disable reading of Macroblock Status Buffer */
+ (0 << 14) | /* Load BitStream Pointer only once, 1 slic 1 frame */
+ (0 << 13) | /* CABAC 0 word insertion test enable */
+ (1 << 12) | /* MVUnpackedEnable,compliant to DXVA */
+ (1 << 10) | /* Chroma Format IDC, 4:2:0 */
+ (0 << 9) | /* FIXME: MbMvFormatFlag */
+ (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 */
+ (0 << 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 */
+ (0 << 0)); /* Field picture flag */
+ OUT_BCS_BATCH(batch, 0); /* Mainly about MB rate control and debug, just ignoring */
+ OUT_BCS_BATCH(batch, /* Inter and Intra Conformance Max size limit */
+ (0xBB8 << 16) | /* InterMbMaxSz */
+ (0xEE8) ); /* IntraMbMaxSz */
+ OUT_BCS_BATCH(batch, 0); /* Reserved */
+ OUT_BCS_BATCH(batch, 0); /* Slice QP Delta for bitrate control */
+ OUT_BCS_BATCH(batch, 0); /* Slice QP Delta for bitrate control */
+ OUT_BCS_BATCH(batch, 0x8C000000);
+ OUT_BCS_BATCH(batch, 0x00010000);
+ OUT_BCS_BATCH(batch, 0);
+ OUT_BCS_BATCH(batch, 0);
+ OUT_BCS_BATCH(batch, 0);
+ OUT_BCS_BATCH(batch, 0);
+
+ ADVANCE_BCS_BATCH(batch);
+}
+
+static void
+gen75_mfc_qm_state(VADriverContextP ctx,
+ int qm_type,
+ unsigned int *qm,
+ int qm_length,
+ struct intel_encoder_context *encoder_context)
+{
+ struct intel_batchbuffer *batch = encoder_context->base.batch;
+ unsigned int qm_buffer[16];
+
+ assert(qm_length <= 16);
+ assert(sizeof(*qm) == 4);
+ memcpy(qm_buffer, qm, qm_length * 4);
+
+ BEGIN_BCS_BATCH(batch, 18);
+ OUT_BCS_BATCH(batch, MFX_QM_STATE | (18 - 2));
+ OUT_BCS_BATCH(batch, qm_type << 0);
+ intel_batchbuffer_data(batch, qm_buffer, 16 * 4);
+ ADVANCE_BCS_BATCH(batch);
+}
+
+static void
+gen75_mfc_avc_qm_state(VADriverContextP ctx, struct intel_encoder_context *encoder_context)
+{
+ unsigned int qm[16] = {
+ 0x10101010, 0x10101010, 0x10101010, 0x10101010,
+ 0x10101010, 0x10101010, 0x10101010, 0x10101010,
+ 0x10101010, 0x10101010, 0x10101010, 0x10101010,
+ 0x10101010, 0x10101010, 0x10101010, 0x10101010
+ };
+
+ gen75_mfc_qm_state(ctx, MFX_QM_AVC_4X4_INTRA_MATRIX, qm, 12, encoder_context);
+ gen75_mfc_qm_state(ctx, MFX_QM_AVC_4X4_INTER_MATRIX, qm, 12, encoder_context);
+ gen75_mfc_qm_state(ctx, MFX_QM_AVC_8x8_INTRA_MATRIX, qm, 16, encoder_context);
+ gen75_mfc_qm_state(ctx, MFX_QM_AVC_8x8_INTER_MATRIX, qm, 16, encoder_context);
+}
+
+static void
+gen75_mfc_fqm_state(VADriverContextP ctx,
+ int fqm_type,
+ unsigned int *fqm,
+ int fqm_length,
+ struct intel_encoder_context *encoder_context)
+{
+ struct intel_batchbuffer *batch = encoder_context->base.batch;
+ unsigned int fqm_buffer[32];
+
+ assert(fqm_length <= 32);
+ assert(sizeof(*fqm) == 4);
+ memcpy(fqm_buffer, fqm, fqm_length * 4);
+
+ BEGIN_BCS_BATCH(batch, 34);
+ OUT_BCS_BATCH(batch, MFX_FQM_STATE | (34 - 2));
+ OUT_BCS_BATCH(batch, fqm_type << 0);
+ intel_batchbuffer_data(batch, fqm_buffer, 32 * 4);
+ ADVANCE_BCS_BATCH(batch);
+}
+
+static void
+gen75_mfc_avc_fqm_state(VADriverContextP ctx, struct intel_encoder_context *encoder_context)
+{
+ unsigned int qm[32] = {
+ 0x10001000, 0x10001000, 0x10001000, 0x10001000,
+ 0x10001000, 0x10001000, 0x10001000, 0x10001000,
+ 0x10001000, 0x10001000, 0x10001000, 0x10001000,
+ 0x10001000, 0x10001000, 0x10001000, 0x10001000,
+ 0x10001000, 0x10001000, 0x10001000, 0x10001000,
+ 0x10001000, 0x10001000, 0x10001000, 0x10001000,
+ 0x10001000, 0x10001000, 0x10001000, 0x10001000,
+ 0x10001000, 0x10001000, 0x10001000, 0x10001000
+ };
+
+ gen75_mfc_fqm_state(ctx, MFX_QM_AVC_4X4_INTRA_MATRIX, qm, 24, encoder_context);
+ gen75_mfc_fqm_state(ctx, MFX_QM_AVC_4X4_INTER_MATRIX, qm, 24, encoder_context);
+ gen75_mfc_fqm_state(ctx, MFX_QM_AVC_8x8_INTRA_MATRIX, qm, 32, encoder_context);
+ gen75_mfc_fqm_state(ctx, MFX_QM_AVC_8x8_INTER_MATRIX, qm, 32, encoder_context);
+}
+
+static void
+gen75_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,
+ struct intel_batchbuffer *batch)
+{
+ if (batch == NULL)
+ batch = encoder_context->base.batch;
+
+ BEGIN_BCS_BATCH(batch, lenght_in_dws + 2);
+
+ OUT_BCS_BATCH(batch, MFX_INSERT_OBJECT | (lenght_in_dws + 2 - 2));
+ OUT_BCS_BATCH(batch,
+ (0 << 16) | /* always start at offset 0 */
+ (data_bits_in_last_dw << 8) |
+ (skip_emul_byte_count << 4) |
+ (!!emulation_flag << 3) |
+ ((!!is_last_header) << 2) |
+ ((!!is_end_of_slice) << 1) |
+ (0 << 0)); /* FIXME: ??? */
+ intel_batchbuffer_data(batch, insert_data, lenght_in_dws * 4);
+
+ ADVANCE_BCS_BATCH(batch);
+}
+
+
+static void
+gen75_mfc_bit_rate_control_context_init(struct encode_state *encode_state,
+ struct gen6_mfc_context *mfc_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;
+ int i;
+
+ mfc_context->bit_rate_control_context[SLICE_TYPE_I].target_mb_size = intra_mb_size;
+ mfc_context->bit_rate_control_context[SLICE_TYPE_I].target_frame_size = intra_mb_size * width_in_mbs * height_in_mbs;
+ mfc_context->bit_rate_control_context[SLICE_TYPE_P].target_mb_size = inter_mb_size;
+ mfc_context->bit_rate_control_context[SLICE_TYPE_P].target_frame_size = inter_mb_size * width_in_mbs * height_in_mbs;
+ mfc_context->bit_rate_control_context[SLICE_TYPE_B].target_mb_size = inter_mb_size;
+ mfc_context->bit_rate_control_context[SLICE_TYPE_B].target_frame_size = inter_mb_size * width_in_mbs * height_in_mbs;
+
+ for(i = 0 ; i < 3; 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;
+ }
+
+ mfc_context->bit_rate_control_context[SLICE_TYPE_I].TargetSizeInWord = (intra_mb_size + 16)/ 16;
+ mfc_context->bit_rate_control_context[SLICE_TYPE_P].TargetSizeInWord = (inter_mb_size + 16)/ 16;
+ mfc_context->bit_rate_control_context[SLICE_TYPE_B].TargetSizeInWord = (inter_mb_size + 16)/ 16;
+
+ mfc_context->bit_rate_control_context[SLICE_TYPE_I].MaxSizeInWord = mfc_context->bit_rate_control_context[SLICE_TYPE_I].TargetSizeInWord * 1.5;
+ mfc_context->bit_rate_control_context[SLICE_TYPE_P].MaxSizeInWord = mfc_context->bit_rate_control_context[SLICE_TYPE_P].TargetSizeInWord * 1.5;
+ mfc_context->bit_rate_control_context[SLICE_TYPE_B].MaxSizeInWord = mfc_context->bit_rate_control_context[SLICE_TYPE_B].TargetSizeInWord * 1.5;
+}
+
+static void
+gen75_mfc_brc_init(struct encode_state *encode_state,
+ struct intel_encoder_context* encoder_context)
+{
+ struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
+ VAEncSequenceParameterBufferH264 *pSequenceParameter = (VAEncSequenceParameterBufferH264 *)encode_state->seq_param_ext->buffer;
+ VAEncMiscParameterBuffer* pMiscParamHRD = (VAEncMiscParameterBuffer*)encode_state->misc_param[VAEncMiscParameterTypeHRD]->buffer;
+ VAEncMiscParameterHRD* pParameterHRD = (VAEncMiscParameterHRD*)pMiscParamHRD->data;
+ double bitrate = pSequenceParameter->bits_per_second;
+ double framerate = (double)pSequenceParameter->time_scale /(2 * (double)pSequenceParameter->num_units_in_tick);
+ int inum = 1, pnum = 0, bnum = 0; /* Gop structure: number of I, P, B frames in the Gop. */
+ int intra_period = pSequenceParameter->intra_period;
+ int ip_period = pSequenceParameter->ip_period;
+ double qp1_size = 0.1 * 8 * 3 * (pSequenceParameter->picture_width_in_mbs<<4) * (pSequenceParameter->picture_height_in_mbs<<4)/2;
+ double qp51_size = 0.001 * 8 * 3 * (pSequenceParameter->picture_width_in_mbs<<4) * (pSequenceParameter->picture_height_in_mbs<<4)/2;
+ double bpf;
+
+ if (pSequenceParameter->ip_period) {
+ pnum = (intra_period + ip_period - 1)/ip_period - 1;
+ bnum = intra_period - inum - pnum;
+ }
+
+ mfc_context->brc.mode = encoder_context->rate_control_mode;
+
+ mfc_context->brc.target_frame_size[SLICE_TYPE_I] = (int)((double)((bitrate * intra_period)/framerate) /
+ (double)(inum + BRC_PWEIGHT * pnum + BRC_BWEIGHT * bnum));
+ mfc_context->brc.target_frame_size[SLICE_TYPE_P] = BRC_PWEIGHT * mfc_context->brc.target_frame_size[SLICE_TYPE_I];
+ mfc_context->brc.target_frame_size[SLICE_TYPE_B] = BRC_BWEIGHT * mfc_context->brc.target_frame_size[SLICE_TYPE_I];
+
+ mfc_context->brc.gop_nums[SLICE_TYPE_I] = inum;
+ mfc_context->brc.gop_nums[SLICE_TYPE_P] = pnum;
+ mfc_context->brc.gop_nums[SLICE_TYPE_B] = bnum;
+
+ bpf = mfc_context->brc.bits_per_frame = bitrate/framerate;
+
+ mfc_context->hrd.buffer_size = (double)pParameterHRD->buffer_size;
+ mfc_context->hrd.current_buffer_fullness =
+ (double)(pParameterHRD->initial_buffer_fullness < mfc_context->hrd.buffer_size)?
+ pParameterHRD->initial_buffer_fullness: mfc_context->hrd.buffer_size/2.;
+ mfc_context->hrd.target_buffer_fullness = (double)mfc_context->hrd.buffer_size/2.;
+ mfc_context->hrd.buffer_capacity = (double)mfc_context->hrd.buffer_size/qp1_size;
+ mfc_context->hrd.violation_noted = 0;
+
+ if ((bpf > qp51_size) && (bpf < qp1_size)) {
+ mfc_context->bit_rate_control_context[SLICE_TYPE_P].QpPrimeY = 51 - 50*(bpf - qp51_size)/(qp1_size - qp51_size);
+ }
+ else if (bpf >= qp1_size)
+ mfc_context->bit_rate_control_context[SLICE_TYPE_P].QpPrimeY = 1;
+ else if (bpf <= qp51_size)
+ mfc_context->bit_rate_control_context[SLICE_TYPE_P].QpPrimeY = 51;
+
+ mfc_context->bit_rate_control_context[SLICE_TYPE_I].QpPrimeY = mfc_context->bit_rate_control_context[SLICE_TYPE_P].QpPrimeY;
+ mfc_context->bit_rate_control_context[SLICE_TYPE_B].QpPrimeY = mfc_context->bit_rate_control_context[SLICE_TYPE_I].QpPrimeY;
+
+ BRC_CLIP(mfc_context->bit_rate_control_context[SLICE_TYPE_I].QpPrimeY, 1, 51);
+ BRC_CLIP(mfc_context->bit_rate_control_context[SLICE_TYPE_P].QpPrimeY, 1, 51);
+ BRC_CLIP(mfc_context->bit_rate_control_context[SLICE_TYPE_B].QpPrimeY, 1, 51);
+}
+
+static void
+gen75_mfc_hrd_context_init(struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
+{
+ struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
+ VAEncSequenceParameterBufferH264 *pSequenceParameter = (VAEncSequenceParameterBufferH264 *)encode_state->seq_param_ext->buffer;
+ unsigned int rate_control_mode = encoder_context->rate_control_mode;
+ int target_bit_rate = pSequenceParameter->bits_per_second;
+
+ // current we only support CBR mode.
+ if (rate_control_mode == VA_RC_CBR) {
+ mfc_context->vui_hrd.i_bit_rate_value = target_bit_rate >> 10;
+ mfc_context->vui_hrd.i_cpb_size_value = (target_bit_rate * 8) >> 10;
+ mfc_context->vui_hrd.i_initial_cpb_removal_delay = mfc_context->vui_hrd.i_cpb_size_value * 0.5 * 1024 / target_bit_rate * 90000;
+ mfc_context->vui_hrd.i_cpb_removal_delay = 2;
+ mfc_context->vui_hrd.i_frame_number = 0;
+
+ mfc_context->vui_hrd.i_initial_cpb_removal_delay_length = 24;
+ mfc_context->vui_hrd.i_cpb_removal_delay_length = 24;
+ mfc_context->vui_hrd.i_dpb_output_delay_length = 24;
+ }
+
+}
+
+static void
+gen75_mfc_free_avc_surface(void **data)
+{
+ 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;
+}
+
+static void gen75_mfc_init(VADriverContextP ctx, 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;
+
+ /*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",
+ 128 * 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",
+ 128*128*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",
+ 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;
+
+ 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);
+ 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
+gen75_mfc_pipe_buf_addr_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;
+ int i;
+
+ BEGIN_BCS_BATCH(batch, 24);
+
+ OUT_BCS_BATCH(batch, MFX_PIPE_BUF_ADDR_STATE | (24 - 2));
+
+ 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 */
+
+ 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,
+ 0); /* uncompressed data */
+ OUT_BCS_RELOC(batch, mfc_context->macroblock_status_buffer.bo,
+ I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
+ 0); /* StreamOut data*/
+ OUT_BCS_RELOC(batch, mfc_context->intra_row_store_scratch_buffer.bo,
+ I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
+ 0);
+ OUT_BCS_RELOC(batch, mfc_context->deblocking_filter_row_store_scratch_buffer.bo,
+ I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
+ 0);
+ /* 7..22 Reference pictures*/
+ for (i = 0; i < ARRAY_ELEMS(mfc_context->reference_surfaces); i++) {
+ if ( mfc_context->reference_surfaces[i].bo != NULL) {
+ OUT_BCS_RELOC(batch, mfc_context->reference_surfaces[i].bo,
+ I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
+ 0);
+ } else {
+ OUT_BCS_BATCH(batch, 0);
+ }
+ }
+ OUT_BCS_RELOC(batch, mfc_context->macroblock_status_buffer.bo,
+ I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
+ 0); /* Macroblock status buffer*/
+
+ ADVANCE_BCS_BATCH(batch);
+}
+
+static void
+gen75_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;
+
+ int i;
+
+ BEGIN_BCS_BATCH(batch, 69);
+
+ OUT_BCS_BATCH(batch, MFX_AVC_DIRECTMODE_STATE | (69 - 2));
+
+ /* Reference frames and Current frames */
+ for(i = 0; i < NUM_MFC_DMV_BUFFERS; i++) {
+ if ( mfc_context->direct_mv_buffers[i].bo != NULL) {
+ OUT_BCS_RELOC(batch, mfc_context->direct_mv_buffers[i].bo,
+ I915_GEM_DOMAIN_INSTRUCTION, 0,
+ 0);
+ } else {
+ OUT_BCS_BATCH(batch, 0);
+ }
+ }
+
+ /* POL list */
+ for(i = 0; i < 32; i++) {
+ OUT_BCS_BATCH(batch, i/2);
+ }
+ OUT_BCS_BATCH(batch, 0);
+ OUT_BCS_BATCH(batch, 0);
+
+ ADVANCE_BCS_BATCH(batch);
+}
+
+static void
+gen75_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
+gen75_mfc_bsp_buf_base_addr_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;
+
+ BEGIN_BCS_BATCH(batch, 4);
+
+ OUT_BCS_BATCH(batch, MFX_BSP_BUF_BASE_ADDR_STATE | (4 - 2));
+ OUT_BCS_RELOC(batch, mfc_context->bsd_mpc_row_store_scratch_buffer.bo,
+ I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
+ 0);
+ OUT_BCS_BATCH(batch, 0);
+ OUT_BCS_BATCH(batch, 0);
+
+ ADVANCE_BCS_BATCH(batch);
+}
+
+
+static void gen75_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);
+ gen75_mfc_pipe_buf_addr_state(ctx, encoder_context);
+ gen75_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);
+ gen75_mfc_avc_directmode_state(ctx, encoder_context);
+ gen75_mfc_avc_ref_idx_state(ctx, encoder_context);
+}
+
+
+static VAStatus gen75_mfc_avc_prepare(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;
+ 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;
+ int i, j, enable_avc_ildb = 0;
+ VAEncSliceParameterBufferH264 *slice_param;
+ VACodedBufferSegment *coded_buffer_segment;
+ unsigned char *flag = NULL;
+
+ for (j = 0; j < encode_state->num_slice_params_ext && enable_avc_ildb == 0; j++) {
+ assert(encode_state->slice_params_ext && encode_state->slice_params_ext[j]->buffer);
+ slice_param = (VAEncSliceParameterBufferH264 *)encode_state->slice_params_ext[j]->buffer;
+
+ for (i = 0; i < encode_state->slice_params_ext[j]->num_elements; i++) {
+ assert((slice_param->slice_type == SLICE_TYPE_I) ||
+ (slice_param->slice_type == SLICE_TYPE_SI) ||
+ (slice_param->slice_type == SLICE_TYPE_P) ||
+ (slice_param->slice_type == SLICE_TYPE_SP) ||
+ (slice_param->slice_type == SLICE_TYPE_B));
+
+ if (slice_param->disable_deblocking_filter_idc != 1) {
+ enable_avc_ildb = 1;
+ break;
+ }
+
+ slice_param++;
+ }
+ }
+
+ /*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'), SUBSAMPLE_YUV420);
+
+ 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 *)gen75_mfc_free_avc_surface;
+ }
+ 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);
+
+ if (enable_avc_ildb) {
+ mfc_context->post_deblocking_output.bo = obj_surface->bo;
+ dri_bo_reference(mfc_context->post_deblocking_output.bo);
+ } else {
+ mfc_context->pre_deblocking_output.bo = obj_surface->bo;
+ dri_bo_reference(mfc_context->pre_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 = gen75_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;
+ }
+ }
+
+ obj_surface = SURFACE(encoder_context->input_yuv_surface);
+ 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->coded_buf); /* 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 = I965_CODEDBUFFER_SIZE;
+ 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);
+
+ dri_bo_map(bo, 1);
+ coded_buffer_segment = (VACodedBufferSegment *)bo->virtual;
+ flag = (unsigned char *)(coded_buffer_segment + 1);
+ *flag = 0;
+ dri_bo_unmap(bo);
+
+ return vaStatus;
+}
+
+
+static VAStatus gen75_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;
+}
+
+
+static VAStatus
+gen75_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;
+}
+
+
+static int gen75_mfc_update_hrd(struct encode_state *encode_state,
+ struct gen6_mfc_context *mfc_context,
+ int frame_bits)
+{
+ double prev_bf = mfc_context->hrd.current_buffer_fullness;
+
+ mfc_context->hrd.current_buffer_fullness -= frame_bits;
+
+ if (mfc_context->hrd.buffer_size > 0 && mfc_context->hrd.current_buffer_fullness <= 0.) {
+ mfc_context->hrd.current_buffer_fullness = prev_bf;
+ return BRC_UNDERFLOW;
+ }
+
+ mfc_context->hrd.current_buffer_fullness += mfc_context->brc.bits_per_frame;
+ if (mfc_context->hrd.buffer_size > 0 && mfc_context->hrd.current_buffer_fullness > mfc_context->hrd.buffer_size) {
+ if (mfc_context->brc.mode == VA_RC_VBR)
+ mfc_context->hrd.current_buffer_fullness = mfc_context->hrd.buffer_size;
+ else {
+ mfc_context->hrd.current_buffer_fullness = prev_bf;
+ return BRC_OVERFLOW;
+ }
+ }
+ return BRC_NO_HRD_VIOLATION;
+}
+
+
+static int gen75_mfc_brc_postpack(struct encode_state *encode_state,
+ struct gen6_mfc_context *mfc_context,
+ int frame_bits)
+{
+ gen6_brc_status sts = BRC_NO_HRD_VIOLATION;
+ VAEncSliceParameterBufferH264 *pSliceParameter = (VAEncSliceParameterBufferH264 *)encode_state->slice_params_ext[0]->buffer;
+ int slicetype = pSliceParameter->slice_type;
+ int qpi = mfc_context->bit_rate_control_context[SLICE_TYPE_I].QpPrimeY;
+ int qpp = mfc_context->bit_rate_control_context[SLICE_TYPE_P].QpPrimeY;
+ int qpb = mfc_context->bit_rate_control_context[SLICE_TYPE_B].QpPrimeY;
+ int qp; // quantizer of previously encoded slice of current type
+ int qpn; // predicted quantizer for next frame of current type in integer format
+ double qpf; // predicted quantizer for next frame of current type in float format
+ double delta_qp; // QP correction
+ int target_frame_size, frame_size_next;
+ /* Notes:
+ * x - how far we are from HRD buffer borders
+ * y - how far we are from target HRD buffer fullness
+ */
+ double x, y;
+ double frame_size_alpha;
+
+ if (slicetype == SLICE_TYPE_SP)
+ slicetype = SLICE_TYPE_P;
+ else if (slicetype == SLICE_TYPE_SI)
+ slicetype = SLICE_TYPE_I;
+
+ qp = mfc_context->bit_rate_control_context[slicetype].QpPrimeY;
+
+ target_frame_size = mfc_context->brc.target_frame_size[slicetype];
+ if (mfc_context->hrd.buffer_capacity < 5)
+ frame_size_alpha = 0;
+ else
+ frame_size_alpha = (double)mfc_context->brc.gop_nums[slicetype];
+ if (frame_size_alpha > 30) frame_size_alpha = 30;
+ frame_size_next = target_frame_size + (double)(target_frame_size - frame_bits) /
+ (double)(frame_size_alpha + 1.);
+
+ /* frame_size_next: avoiding negative number and too small value */
+ if ((double)frame_size_next < (double)(target_frame_size * 0.25))
+ frame_size_next = (int)((double)target_frame_size * 0.25);
+
+ qpf = (double)qp * target_frame_size / frame_size_next;
+ qpn = (int)(qpf + 0.5);
+
+ if (qpn == qp) {
+ /* setting qpn we round qpf making mistakes: now we are trying to compensate this */
+ mfc_context->brc.qpf_rounding_accumulator += qpf - qpn;
+ if (mfc_context->brc.qpf_rounding_accumulator > 1.0) {
+ qpn++;
+ mfc_context->brc.qpf_rounding_accumulator = 0.;
+ } else if (mfc_context->brc.qpf_rounding_accumulator < -1.0) {
+ qpn--;
+ mfc_context->brc.qpf_rounding_accumulator = 0.;
+ }
+ }
+ /* making sure that QP is not changing too fast */
+ if ((qpn - qp) > BRC_QP_MAX_CHANGE) qpn = qp + BRC_QP_MAX_CHANGE;
+ else if ((qpn - qp) < -BRC_QP_MAX_CHANGE) qpn = qp - BRC_QP_MAX_CHANGE;
+ /* making sure that with QP predictions we did do not leave QPs range */
+ BRC_CLIP(qpn, 1, 51);
+
+ /* checking wthether HRD compliance is still met */
+ sts = gen75_mfc_update_hrd(encode_state, mfc_context, frame_bits);
+
+ /* calculating QP delta as some function*/
+ x = mfc_context->hrd.target_buffer_fullness - mfc_context->hrd.current_buffer_fullness;
+ if (x > 0) {
+ x /= mfc_context->hrd.target_buffer_fullness;
+ y = mfc_context->hrd.current_buffer_fullness;
+ }
+ else {
+ x /= (mfc_context->hrd.buffer_size - mfc_context->hrd.target_buffer_fullness);
+ y = mfc_context->hrd.buffer_size - mfc_context->hrd.current_buffer_fullness;
+ }
+ if (y < 0.01) y = 0.01;
+ if (x > 1) x = 1;
+ else if (x < -1) x = -1;
+
+ delta_qp = BRC_QP_MAX_CHANGE*exp(-1/y)*sin(BRC_PI_0_5 * x);
+ qpn = (int)(qpn + delta_qp + 0.5);
+
+ /* making sure that with QP predictions we did do not leave QPs range */
+ BRC_CLIP(qpn, 1, 51);
+
+ if (sts == BRC_NO_HRD_VIOLATION) { // no HRD violation
+ /* correcting QPs of slices of other types */
+ if (slicetype == SLICE_TYPE_P) {
+ if (abs(qpn + BRC_P_B_QP_DIFF - qpb) > 2)
+ mfc_context->bit_rate_control_context[SLICE_TYPE_B].QpPrimeY += (qpn + BRC_P_B_QP_DIFF - qpb) >> 1;
+ if (abs(qpn - BRC_I_P_QP_DIFF - qpi) > 2)
+ mfc_context->bit_rate_control_context[SLICE_TYPE_I].QpPrimeY += (qpn - BRC_I_P_QP_DIFF - qpi) >> 1;
+ } else if (slicetype == SLICE_TYPE_I) {
+ if (abs(qpn + BRC_I_B_QP_DIFF - qpb) > 4)
+ mfc_context->bit_rate_control_context[SLICE_TYPE_B].QpPrimeY += (qpn + BRC_I_B_QP_DIFF - qpb) >> 2;
+ if (abs(qpn + BRC_I_P_QP_DIFF - qpp) > 2)
+ mfc_context->bit_rate_control_context[SLICE_TYPE_P].QpPrimeY += (qpn + BRC_I_P_QP_DIFF - qpp) >> 2;
+ } else { // SLICE_TYPE_B
+ if (abs(qpn - BRC_P_B_QP_DIFF - qpp) > 2)
+ mfc_context->bit_rate_control_context[SLICE_TYPE_P].QpPrimeY += (qpn - BRC_P_B_QP_DIFF - qpp) >> 1;
+ if (abs(qpn - BRC_I_B_QP_DIFF - qpi) > 4)
+ mfc_context->bit_rate_control_context[SLICE_TYPE_I].QpPrimeY += (qpn - BRC_I_B_QP_DIFF - qpi) >> 2;
+ }
+ BRC_CLIP(mfc_context->bit_rate_control_context[SLICE_TYPE_I].QpPrimeY, 1, 51);
+ BRC_CLIP(mfc_context->bit_rate_control_context[SLICE_TYPE_P].QpPrimeY, 1, 51);
+ BRC_CLIP(mfc_context->bit_rate_control_context[SLICE_TYPE_B].QpPrimeY, 1, 51);
+ } else if (sts == BRC_UNDERFLOW) { // underflow
+ if (qpn <= qp) qpn = qp + 1;
+ if (qpn > 51) {
+ qpn = 51;
+ sts = BRC_UNDERFLOW_WITH_MAX_QP; //underflow with maxQP
+ }
+ } else if (sts == BRC_OVERFLOW) {
+ if (qpn >= qp) qpn = qp - 1;
+ if (qpn < 1) { // < 0 (?) overflow with minQP
+ qpn = 1;
+ sts = BRC_OVERFLOW_WITH_MIN_QP; // bit stuffing to be done
+ }
+ }
+
+ mfc_context->bit_rate_control_context[slicetype].QpPrimeY = qpn;
+
+ return sts;
+}
+
+static void
+gen75_mfc_hrd_context_update(struct encode_state *encode_state,
+ struct gen6_mfc_context *mfc_context)
+{
+ mfc_context->vui_hrd.i_frame_number++;
+}
+
+
+static int interlace_check(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
+{
+ struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
+ VAEncSliceParameterBufferH264 *pSliceParameter;
+ int i;
+ int mbCount = 0;
+ int width_in_mbs = (mfc_context->surface_state.width + 15) / 16;
+ int height_in_mbs = (mfc_context->surface_state.height + 15) / 16;
+
+ for (i = 0; i < encode_state->num_slice_params_ext; i++) {
+ pSliceParameter = (VAEncSliceParameterBufferH264 *)encode_state->slice_params_ext[i]->buffer;
+ mbCount += pSliceParameter->num_macroblocks;
+ }
+
+ if ( mbCount == ( width_in_mbs * height_in_mbs ) )
+ return 0;
+
+ return 1;
+}
+
+static void
+gen75_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 gen6_mfc_context *mfc_context = encoder_context->mfc_context;
+ 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 = slice_param->slice_type;
+ int last_slice = (endmb == (width_in_mbs * height_in_mbs));
+ int bit_rate_control_target, 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;
+
+ if (batch == NULL)
+ batch = encoder_context->base.batch;
+
+ bit_rate_control_target = slice_type;
+ if (slice_type == SLICE_TYPE_SP)
+ bit_rate_control_target = SLICE_TYPE_P;
+ else if (slice_type == SLICE_TYPE_SI)
+ bit_rate_control_target = SLICE_TYPE_I;
+
+ if (slice_type == SLICE_TYPE_P) {
+ weighted_pred_idc = pic_param->pic_fields.bits.weighted_pred_flag;
+ } else if (slice_type == SLICE_TYPE_B) {
+ weighted_pred_idc = pic_param->pic_fields.bits.weighted_bipred_idc;
+
+ 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[bit_rate_control_target].MaxQpNegModifier;
+ maxQpP = mfc_context->bit_rate_control_context[bit_rate_control_target].MaxQpPosModifier;
+
+ for (i = 0; i < 6; i++)
+ correct[i] = mfc_context->bit_rate_control_context[bit_rate_control_target].Correct[i];
+
+ grow = mfc_context->bit_rate_control_context[bit_rate_control_target].GrowInit +
+ (mfc_context->bit_rate_control_context[bit_rate_control_target].GrowResistance << 4);
+ shrink = mfc_context->bit_rate_control_context[bit_rate_control_target].ShrinkInit +
+ (mfc_context->bit_rate_control_context[bit_rate_control_target].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,
+ (1 << 16) | /*1 reference frame*/
+ (chroma_log2_weight_denom << 8) |
+ (luma_log2_weight_denom << 0));
+ }
+
+ OUT_BCS_BATCH(batch,
+ (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*/
+ ((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,
+ (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) |
+ (correct[4] << 16) |
+ (correct[3] << 12) |
+ (correct[2] << 8) |
+ (correct[1] << 4) |
+ (correct[0] << 0));
+ OUT_BCS_BATCH(batch, 0);
+
+ ADVANCE_BCS_BATCH(batch);
+}
+
+
+static void gen75_mfc_avc_pipeline_header_programing(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context,
+ struct intel_batchbuffer *slice_batch)
+{
+ struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
+ int idx = va_enc_packed_type_to_idx(VAEncPackedHeaderH264_SPS);
+
+ if (encode_state->packed_header_data[idx]) {
+ VAEncPackedHeaderParameterBuffer *param = NULL;
+ unsigned int *header_data = (unsigned int *)encode_state->packed_header_data[idx]->buffer;
+ unsigned int length_in_bits;
+
+ assert(encode_state->packed_header_param[idx]);
+ param = (VAEncPackedHeaderParameterBuffer *)encode_state->packed_header_param[idx]->buffer;
+ length_in_bits = param->bit_length;
+
+ mfc_context->insert_object(ctx,
+ encoder_context,
+ header_data,
+ ALIGN(length_in_bits, 32) >> 5,
+ length_in_bits & 0x1f,
+ 5, /* FIXME: check it */
+ 0,
+ 0,
+ !param->has_emulation_bytes,
+ slice_batch);
+ }
+
+ idx = va_enc_packed_type_to_idx(VAEncPackedHeaderH264_PPS);
+
+ if (encode_state->packed_header_data[idx]) {
+ VAEncPackedHeaderParameterBuffer *param = NULL;
+ unsigned int *header_data = (unsigned int *)encode_state->packed_header_data[idx]->buffer;
+ unsigned int length_in_bits;
+
+ assert(encode_state->packed_header_param[idx]);
+ param = (VAEncPackedHeaderParameterBuffer *)encode_state->packed_header_param[idx]->buffer;
+ length_in_bits = param->bit_length;
+
+ mfc_context->insert_object(ctx,
+ encoder_context,
+ header_data,
+ ALIGN(length_in_bits, 32) >> 5,
+ length_in_bits & 0x1f,
+ 5, /* FIXME: check it */
+ 0,
+ 0,
+ !param->has_emulation_bytes,
+ slice_batch);
+ }
+
+ idx = va_enc_packed_type_to_idx(VAEncPackedHeaderH264_SEI);
+
+ if (encode_state->packed_header_data[idx]) {
+ VAEncPackedHeaderParameterBuffer *param = NULL;
+ unsigned int *header_data = (unsigned int *)encode_state->packed_header_data[idx]->buffer;
+ unsigned int length_in_bits;
+
+ assert(encode_state->packed_header_param[idx]);
+ param = (VAEncPackedHeaderParameterBuffer *)encode_state->packed_header_param[idx]->buffer;
+ length_in_bits = param->bit_length;
+
+ mfc_context->insert_object(ctx,
+ encoder_context,
+ header_data,
+ ALIGN(length_in_bits, 32) >> 5,
+ length_in_bits & 0x1f,
+ 5, /* FIXME: check it */
+ 0,
+ 0,
+ !param->has_emulation_bytes,
+ slice_batch);
+ }
+}
+
+#if __SOFTWARE__
+
+static int
+gen75_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,
+ struct intel_batchbuffer *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));
+ OUT_BCS_BATCH(batch, 0);
+ OUT_BCS_BATCH(batch, 0);
+ OUT_BCS_BATCH(batch,
+ (0 << 24) | /* PackedMvNum, Debug*/
+ (0 << 20) | /* No motion vector */
+ (1 << 19) | /* CbpDcY */
+ (1 << 18) | /* CbpDcU */
+ (1 << 17) | /* CbpDcV */
+ (msg[0] & 0xFFFF) );
+
+ 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 */
+
+ /*Stuff for Intra MB*/
+ OUT_BCS_BATCH(batch, msg[1]); /* We using Intra16x16 no 4x4 predmode*/
+ OUT_BCS_BATCH(batch, msg[2]);
+ OUT_BCS_BATCH(batch, msg[3]&0xFC);
+
+ /*MaxSizeInWord and TargetSzieInWord*/
+ OUT_BCS_BATCH(batch, (max_mb_size << 24) |
+ (target_mb_size << 16) );
+
+ ADVANCE_BCS_BATCH(batch);
+
+ return len_in_dwords;
+}
+
+static int
+gen75_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)
+{
+ 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, msg[2]); /* 32 MV*/
+ OUT_BCS_BATCH(batch, offset);
+
+ 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 */
+#if 0
+ if ( slice_type == SLICE_TYPE_B) {
+ OUT_BCS_BATCH(batch, (0xF<<28) | (end_mb << 26) | qp); /* Last MB */
+ } else {
+ OUT_BCS_BATCH(batch, (end_mb << 26) | qp); /* Last MB */
+ }
+#else
+ OUT_BCS_BATCH(batch, (end_mb << 26) | qp); /* Last MB */
+#endif
+
+
+ /*Stuff for Inter MB*/
+ OUT_BCS_BATCH(batch, msg[1]);
+ OUT_BCS_BATCH(batch, 0x0);
+ OUT_BCS_BATCH(batch, 0x0);
+
+ /*MaxSizeInWord and TargetSzieInWord*/
+ OUT_BCS_BATCH(batch, (max_mb_size << 24) |
+ (target_mb_size << 16) );
+
+ ADVANCE_BCS_BATCH(batch);
+
+ return len_in_dwords;
+}
+
+static void
+gen75_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 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[slice_index]->buffer;
+ unsigned int *msg = NULL, offset = 0;
+ 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 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 char *slice_header = NULL;
+ int slice_header_length_in_bits = 0;
+ unsigned int tail_data[] = { 0x0, 0x0 };
+ int slice_type = pSliceParameter->slice_type;
+
+
+ if (rate_control_mode == VA_RC_CBR) {
+ qp = mfc_context->bit_rate_control_context[slice_type].QpPrimeY;
+ pSliceParameter->slice_qp_delta = qp - pPicParameter->pic_init_qp;
+ }
+
+ /* 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);
+
+ gen75_mfc_avc_slice_state(ctx,
+ pPicParameter,
+ pSliceParameter,
+ encode_state, encoder_context,
+ (rate_control_mode == VA_RC_CBR), qp, slice_batch);
+
+ if ( slice_index == 0)
+ gen75_mfc_avc_pipeline_header_programing(ctx, encode_state, encoder_context, slice_batch);
+
+ slice_header_length_in_bits = build_avc_slice_header(pSequenceParameter, pPicParameter, pSliceParameter, &slice_header);
+
+ // slice hander
+ mfc_context->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, slice_batch);
+
+ dri_bo_map(vme_context->vme_output.bo , 1);
+ msg = (unsigned int *)vme_context->vme_output.bo->virtual;
+
+ 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);
+ gen75_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) {
+ gen75_mfc_avc_pak_object_intra(ctx, x, y, last_mb, qp, msg, encoder_context, 0, 0, slice_batch);
+ } else {
+ gen75_mfc_avc_pak_object_inter(ctx, x, y, last_mb, qp, msg, offset, encoder_context, 0, 0, pSliceParameter->slice_type, slice_batch);
+ }
+
+ msg += INTER_VME_OUTPUT_IN_DWS;
+ offset += INTER_VME_OUTPUT_IN_BYTES;
+ }
+ }
+
+ dri_bo_unmap(vme_context->vme_output.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);
+ }
+
+ free(slice_header);
+
+}
+
+static dri_bo *
+gen75_mfc_avc_software_batchbuffer(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
+{
+ struct i965_driver_data *i965 = i965_driver_data(ctx);
+ struct intel_batchbuffer *batch = intel_batchbuffer_new(&i965->intel, I915_EXEC_BSD);
+ dri_bo *batch_bo = batch->buffer;
+ int i;
+
+ for (i = 0; i < encode_state->num_slice_params_ext; i++) {
+ gen75_mfc_avc_pipeline_slice_programing(ctx, encode_state, encoder_context, i, batch);
+ }
+
+ intel_batchbuffer_align(batch, 8);
+
+ BEGIN_BCS_BATCH(batch, 2);
+ OUT_BCS_BATCH(batch, 0);
+ OUT_BCS_BATCH(batch, MI_BATCH_BUFFER_END);
+ ADVANCE_BCS_BATCH(batch);
+
+ dri_bo_reference(batch_bo);
+ intel_batchbuffer_free(batch);
+
+ return batch_bo;
+}
+
+#else
+
+static void
+gen75_mfc_batchbuffer_surfaces_input(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
+
+{
+ struct gen6_vme_context *vme_context = encoder_context->vme_context;
+ struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
+
+ 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));
+}
+
+static void
+gen75_mfc_batchbuffer_surfaces_output(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;
+ 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
+gen75_mfc_batchbuffer_surfaces_setup(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
+{
+ gen75_mfc_batchbuffer_surfaces_input(ctx, encode_state, encoder_context);
+ gen75_mfc_batchbuffer_surfaces_output(ctx, encode_state, encoder_context);
+}
+
+static void
+gen75_mfc_batchbuffer_idrt_setup(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
+{
+ struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
+ struct gen6_interface_descriptor_data *desc;
+ int i;
+ 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++;
+ }
+
+ dri_bo_unmap(bo);
+}
+
+static void
+gen75_mfc_batchbuffer_constant_setup(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
+{
+ struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
+
+ (void)mfc_context;
+}
+
+static void
+gen75_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)
+{
+ BEGIN_BATCH(batch, 12);
+
+ OUT_BATCH(batch, CMD_MEDIA_OBJECT | (12 - 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);
+
+ ADVANCE_BATCH(batch);
+}
+
+static void
+gen75_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 intel_batchbuffer *batch = encoder_context->base.batch;
+ struct gen6_mfc_context *mfc_context = encoder_context->mfc_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;
+
+ gen75_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);
+
+ 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;
+
+ first_object = 0;
+ }
+
+ 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;
+
+ gen75_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);
+ }
+}
+
+/*
+ * return size in Owords (16bytes)
+ */
+static int
+gen75_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 char *slice_header = NULL;
+ int slice_header_length_in_bits = 0;
+ 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 = pSliceParameter->slice_type;
+
+ if (rate_control_mode == VA_RC_CBR) {
+ qp = mfc_context->bit_rate_control_context[slice_type].QpPrimeY;
+ pSliceParameter->slice_qp_delta = qp - pPicParameter->pic_init_qp;
+ }
+
+ /* 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;
+ gen75_mfc_avc_slice_state(ctx,
+ pPicParameter,
+ pSliceParameter,
+ encode_state,
+ encoder_context,
+ (rate_control_mode == VA_RC_CBR),
+ qp,
+ slice_batch);
+
+ if (slice_index == 0)
+ gen75_mfc_avc_pipeline_header_programing(ctx, encode_state, encoder_context, slice_batch);
+
+ slice_header_length_in_bits = build_avc_slice_header(pSequenceParameter, pPicParameter, pSliceParameter, &slice_header);
+
+ // slice hander
+ mfc_context->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,
+ slice_batch);
+ free(slice_header);
+
+ 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;
+
+
+ gen75_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 void
+gen75_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);
+
+ for ( i = 0; i < encode_state->num_slice_params_ext; i++) {
+ size = gen75_mfc_avc_batchbuffer_slice(ctx, encode_state, encoder_context, i, offset);
+ offset += size;
+ }
+
+ intel_batchbuffer_end_atomic(batch);
+ intel_batchbuffer_flush(batch);
+}
+
+static void
+gen75_mfc_build_avc_batchbuffer(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
+{
+ gen75_mfc_batchbuffer_surfaces_setup(ctx, encode_state, encoder_context);
+ gen75_mfc_batchbuffer_idrt_setup(ctx, encode_state, encoder_context);
+ gen75_mfc_batchbuffer_constant_setup(ctx, encode_state, encoder_context);
+ gen75_mfc_avc_batchbuffer_pipeline(ctx, encode_state, encoder_context);
+}
+
+static dri_bo *
+gen75_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;
+
+ gen75_mfc_build_avc_batchbuffer(ctx, encode_state, encoder_context);
+ dri_bo_reference(mfc_context->mfc_batchbuffer_surface.bo);
+
+ return mfc_context->mfc_batchbuffer_surface.bo;
+}
+
+#endif
+
+static void
+gen75_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 ( 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 = gen75_mfc_avc_software_batchbuffer(ctx, encode_state, encoder_context);
+#else
+ slice_batch_bo = gen75_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
+ gen75_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
+gen75_mfc_avc_encode_picture(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 rate_control_mode = encoder_context->rate_control_mode;
+ int current_frame_bits_size;
+ int sts;
+
+ for (;;) {
+ gen75_mfc_init(ctx, encoder_context);
+ gen75_mfc_avc_prepare(ctx, encode_state, encoder_context);
+ /*Programing bcs pipeline*/
+ gen75_mfc_avc_pipeline_programing(ctx, encode_state, encoder_context); //filling the pipeline
+ gen75_mfc_run(ctx, encode_state, encoder_context);
+ if (rate_control_mode == VA_RC_CBR /*|| rate_control_mode == VA_RC_VBR*/) {
+ gen75_mfc_stop(ctx, encode_state, encoder_context, ¤t_frame_bits_size);
+ sts = gen75_mfc_brc_postpack(encode_state, mfc_context, current_frame_bits_size);
+ if (sts == BRC_NO_HRD_VIOLATION) {
+ gen75_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 void gen75_mfc_brc_prepare(struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
+{
+ unsigned int rate_control_mode = encoder_context->rate_control_mode;
+ struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
+
+ if (rate_control_mode == VA_RC_CBR) {
+ /*Programing bit rate control */
+ if ( mfc_context->bit_rate_control_context[SLICE_TYPE_I].MaxSizeInWord == 0 ) {
+ gen75_mfc_bit_rate_control_context_init(encode_state, mfc_context);
+ gen75_mfc_brc_init(encode_state, encoder_context);
+ }
+
+ /*Programing HRD control */
+ if ( mfc_context->vui_hrd.i_cpb_size_value == 0 )
+ gen75_mfc_hrd_context_init(encode_state, encoder_context);
+ }
+}
+
+static void
+gen75_mfc_context_destroy(void *context)
+{
+ struct gen6_mfc_context *mfc_context = context;
+ int i;
+
+ 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++){
+ dri_bo_unreference(mfc_context->direct_mv_buffers[i].bo);
+ mfc_context->direct_mv_buffers[i].bo = NULL;
+ }
+
+ dri_bo_unreference(mfc_context->intra_row_store_scratch_buffer.bo);
+ mfc_context->intra_row_store_scratch_buffer.bo = NULL;
+
+ dri_bo_unreference(mfc_context->macroblock_status_buffer.bo);
+ mfc_context->macroblock_status_buffer.bo = NULL;
+
+ dri_bo_unreference(mfc_context->deblocking_filter_row_store_scratch_buffer.bo);
+ mfc_context->deblocking_filter_row_store_scratch_buffer.bo = NULL;
+
+ dri_bo_unreference(mfc_context->bsd_mpc_row_store_scratch_buffer.bo);
+ mfc_context->bsd_mpc_row_store_scratch_buffer.bo = NULL;
+
+
+ for (i = 0; i < MAX_MFC_REFERENCE_SURFACES; i++){
+ dri_bo_unreference(mfc_context->reference_surfaces[i].bo);
+ 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);
+}
+
+static VAStatus gen75_mfc_pipeline(VADriverContextP ctx,
+ VAProfile profile,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
+{
+ VAStatus vaStatus;
+
+ switch (profile) {
+ case VAProfileH264Baseline:
+ case VAProfileH264Main:
+ case VAProfileH264High:
+ vaStatus = gen75_mfc_avc_encode_picture(ctx, encode_state, encoder_context);
+ break;
+
+ /* FIXME: add for other profile */
+ default:
+ vaStatus = VA_STATUS_ERROR_UNSUPPORTED_PROFILE;
+ break;
+ }
+
+ return vaStatus;
+}
+
+Bool gen75_mfc_context_init(VADriverContextP ctx, struct intel_encoder_context *encoder_context)
+{
+ struct gen6_mfc_context *mfc_context = calloc(1, sizeof(struct gen6_mfc_context));
+
+ 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,
+ gen75_mfc_kernels,
+ NUM_MFC_KERNEL);
+
+ mfc_context->pipe_mode_select = gen75_mfc_pipe_mode_select;
+ mfc_context->set_surface_state = gen75_mfc_surface_state;
+ mfc_context->ind_obj_base_addr_state = gen75_mfc_ind_obj_base_addr_state;
+ mfc_context->avc_img_state = gen75_mfc_avc_img_state;
+ mfc_context->avc_qm_state = gen75_mfc_avc_qm_state;
+ mfc_context->avc_fqm_state = gen75_mfc_avc_fqm_state;
+ mfc_context->insert_object = gen75_mfc_avc_insert_object;
+ mfc_context->buffer_suface_setup = gen7_gpe_buffer_suface_setup;
+
+ encoder_context->mfc_context = mfc_context;
+ encoder_context->mfc_context_destroy = gen75_mfc_context_destroy;
+ encoder_context->mfc_pipeline = gen75_mfc_pipeline;
+ encoder_context->mfc_brc_prepare = gen75_mfc_brc_prepare;
+
+ return True;
+}
--- /dev/null
+/*
+ * Copyright © 2012 Intel Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ * Zhao Yakui <yakui.zhao@intel.com>
+ * Xiang Haihao <haihao.xiang@intel.com>
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <assert.h>
+
+#include "intel_batchbuffer.h"
+#include "intel_driver.h"
+
+#include "i965_defines.h"
+#include "i965_drv_video.h"
+#include "i965_encoder.h"
+#include "gen6_vme.h"
+#include "gen6_mfc.h"
+
+#define SURFACE_STATE_PADDED_SIZE_0_GEN7 ALIGN(sizeof(struct gen7_surface_state), 32)
+#define SURFACE_STATE_PADDED_SIZE_1_GEN7 ALIGN(sizeof(struct gen7_surface_state2), 32)
+#define SURFACE_STATE_PADDED_SIZE_GEN7 MAX(SURFACE_STATE_PADDED_SIZE_0_GEN7, SURFACE_STATE_PADDED_SIZE_1_GEN7)
+
+#define SURFACE_STATE_PADDED_SIZE_0_GEN6 ALIGN(sizeof(struct i965_surface_state), 32)
+#define SURFACE_STATE_PADDED_SIZE_1_GEN6 ALIGN(sizeof(struct i965_surface_state2), 32)
+#define SURFACE_STATE_PADDED_SIZE_GEN6 MAX(SURFACE_STATE_PADDED_SIZE_0_GEN6, SURFACE_STATE_PADDED_SIZE_1_GEN6)
+
+#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)
+
+#define VME_INTRA_SHADER 0
+#define VME_INTER_SHADER 1
+#define VME_BATCHBUFFER 2
+
+#define CURBE_ALLOCATION_SIZE 37 /* in 256-bit */
+#define CURBE_TOTAL_DATA_LENGTH (4 * 32) /* in byte, it should be less than or equal to CURBE_ALLOCATION_SIZE * 32 */
+#define CURBE_URB_ENTRY_LENGTH 4 /* in 256-bit, it should be less than or equal to CURBE_TOTAL_DATA_LENGTH / 32 */
+
+static const uint32_t gen75_vme_intra_frame[][4] = {
+#include "shaders/vme/intra_frame.g7b"
+};
+
+static const uint32_t gen75_vme_inter_frame[][4] = {
+#include "shaders/vme/inter_frame.g7b"
+};
+
+static const uint32_t gen75_vme_batchbuffer[][4] = {
+#include "shaders/vme/batchbuffer.g7b"
+};
+
+static struct i965_kernel gen75_vme_kernels[] = {
+ {
+ "VME Intra Frame",
+ VME_INTRA_SHADER, /*index*/
+ gen75_vme_intra_frame,
+ sizeof(gen75_vme_intra_frame),
+ NULL
+ },
+ {
+ "VME inter Frame",
+ VME_INTER_SHADER,
+ gen75_vme_inter_frame,
+ sizeof(gen75_vme_inter_frame),
+ NULL
+ },
+ {
+ "VME BATCHBUFFER",
+ VME_BATCHBUFFER,
+ gen75_vme_batchbuffer,
+ sizeof(gen75_vme_batchbuffer),
+ NULL
+ },
+};
+
+/* only used for VME source surface state */
+static void
+gen75_vme_source_surface_state(VADriverContextP ctx,
+ int index,
+ struct object_surface *obj_surface,
+ struct intel_encoder_context *encoder_context)
+{
+ struct gen6_vme_context *vme_context = encoder_context->vme_context;
+
+ vme_context->vme_surface2_setup(ctx,
+ &vme_context->gpe_context,
+ obj_surface,
+ BINDING_TABLE_OFFSET(index),
+ SURFACE_STATE_OFFSET(index));
+}
+
+static void
+gen75_vme_media_source_surface_state(VADriverContextP ctx,
+ int index,
+ struct object_surface *obj_surface,
+ struct intel_encoder_context *encoder_context)
+{
+ struct gen6_vme_context *vme_context = encoder_context->vme_context;
+
+ vme_context->vme_media_rw_surface_setup(ctx,
+ &vme_context->gpe_context,
+ obj_surface,
+ BINDING_TABLE_OFFSET(index),
+ SURFACE_STATE_OFFSET(index));
+}
+
+static void
+gen75_vme_output_buffer_setup(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ int index,
+ struct intel_encoder_context *encoder_context)
+
+{
+ struct i965_driver_data *i965 = i965_driver_data(ctx);
+ struct gen6_vme_context *vme_context = encoder_context->vme_context;
+ VAEncSequenceParameterBufferH264 *pSequenceParameter = (VAEncSequenceParameterBufferH264 *)encode_state->seq_param_ext->buffer;
+ VAEncSliceParameterBufferH264 *pSliceParameter = (VAEncSliceParameterBufferH264 *)encode_state->slice_params_ext[0]->buffer;
+ int is_intra = pSliceParameter->slice_type == SLICE_TYPE_I;
+ int width_in_mbs = pSequenceParameter->picture_width_in_mbs;
+ int height_in_mbs = pSequenceParameter->picture_height_in_mbs;
+
+ vme_context->vme_output.num_blocks = width_in_mbs * height_in_mbs;
+ vme_context->vme_output.pitch = 16; /* in bytes, always 16 */
+
+ if (is_intra)
+ vme_context->vme_output.size_block = INTRA_VME_OUTPUT_IN_BYTES;
+ else
+ vme_context->vme_output.size_block = INTER_VME_OUTPUT_IN_BYTES;
+
+ vme_context->vme_output.bo = dri_bo_alloc(i965->intel.bufmgr,
+ "VME output buffer",
+ vme_context->vme_output.num_blocks * vme_context->vme_output.size_block,
+ 0x1000);
+ assert(vme_context->vme_output.bo);
+ vme_context->vme_buffer_suface_setup(ctx,
+ &vme_context->gpe_context,
+ &vme_context->vme_output,
+ BINDING_TABLE_OFFSET(index),
+ SURFACE_STATE_OFFSET(index));
+}
+
+static void
+gen75_vme_output_vme_batchbuffer_setup(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ int index,
+ struct intel_encoder_context *encoder_context)
+
+{
+ struct i965_driver_data *i965 = i965_driver_data(ctx);
+ struct gen6_vme_context *vme_context = encoder_context->vme_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;
+
+ vme_context->vme_batchbuffer.num_blocks = width_in_mbs * height_in_mbs + 1;
+ vme_context->vme_batchbuffer.size_block = 32; /* 2 OWORDs */
+ vme_context->vme_batchbuffer.pitch = 16;
+ vme_context->vme_batchbuffer.bo = dri_bo_alloc(i965->intel.bufmgr,
+ "VME batchbuffer",
+ vme_context->vme_batchbuffer.num_blocks * vme_context->vme_batchbuffer.size_block,
+ 0x1000);
+ vme_context->vme_buffer_suface_setup(ctx,
+ &vme_context->gpe_context,
+ &vme_context->vme_batchbuffer,
+ BINDING_TABLE_OFFSET(index),
+ SURFACE_STATE_OFFSET(index));
+}
+
+static VAStatus
+gen75_vme_surface_setup(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ int is_intra,
+ struct intel_encoder_context *encoder_context)
+{
+ struct i965_driver_data *i965 = i965_driver_data(ctx);
+ struct object_surface *obj_surface;
+ VAEncPictureParameterBufferH264 *pPicParameter = (VAEncPictureParameterBufferH264 *)encode_state->pic_param_ext->buffer;
+
+ /*Setup surfaces state*/
+ /* current picture for encoding */
+ obj_surface = SURFACE(encoder_context->input_yuv_surface);
+ assert(obj_surface);
+ gen75_vme_source_surface_state(ctx, 0, obj_surface, encoder_context);
+ gen75_vme_media_source_surface_state(ctx, 4, obj_surface, encoder_context);
+
+ if (!is_intra) {
+ /* reference 0 */
+ obj_surface = SURFACE(pPicParameter->ReferenceFrames[0].picture_id);
+ assert(obj_surface);
+ if ( obj_surface->bo != NULL)
+ gen75_vme_source_surface_state(ctx, 1, obj_surface, encoder_context);
+
+ /* reference 1 */
+ obj_surface = SURFACE(pPicParameter->ReferenceFrames[1].picture_id);
+ assert(obj_surface);
+ if ( obj_surface->bo != NULL )
+ gen75_vme_source_surface_state(ctx, 2, obj_surface, encoder_context);
+ }
+
+ /* VME output */
+ gen75_vme_output_buffer_setup(ctx, encode_state, 3, encoder_context);
+ gen75_vme_output_vme_batchbuffer_setup(ctx, encode_state, 5, encoder_context);
+
+ return VA_STATUS_SUCCESS;
+}
+
+static VAStatus gen75_vme_interface_setup(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
+{
+ struct gen6_vme_context *vme_context = encoder_context->vme_context;
+ struct gen6_interface_descriptor_data *desc;
+ int i;
+ dri_bo *bo;
+
+ bo = vme_context->gpe_context.idrt.bo;
+ dri_bo_map(bo, 1);
+ assert(bo->virtual);
+ desc = bo->virtual;
+
+ for (i = 0; i < GEN6_VME_KERNEL_NUMBER; i++) {
+ struct i965_kernel *kernel;
+ kernel = &vme_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 = 1; /* FIXME: */
+ desc->desc2.sampler_state_pointer = (vme_context->vme_state.bo->offset >> 5);
+ desc->desc3.binding_table_entry_count = 1; /* FIXME: */
+ 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 = CURBE_URB_ENTRY_LENGTH;
+
+ /*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);
+ /*Sampler State(VME state pointer)*/
+ dri_bo_emit_reloc(bo,
+ I915_GEM_DOMAIN_INSTRUCTION, 0,
+ (1 << 2), //
+ i * sizeof(*desc) + offsetof(struct gen6_interface_descriptor_data, desc2),
+ vme_context->vme_state.bo);
+ desc++;
+ }
+ dri_bo_unmap(bo);
+
+ return VA_STATUS_SUCCESS;
+}
+
+static VAStatus gen75_vme_constant_setup(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
+{
+ struct gen6_vme_context *vme_context = encoder_context->vme_context;
+ // unsigned char *constant_buffer;
+
+ dri_bo_map(vme_context->gpe_context.curbe.bo, 1);
+ assert(vme_context->gpe_context.curbe.bo->virtual);
+ // constant_buffer = vme_context->curbe.bo->virtual;
+
+ /*TODO copy buffer into CURB*/
+
+ dri_bo_unmap( vme_context->gpe_context.curbe.bo);
+
+ return VA_STATUS_SUCCESS;
+}
+
+static const unsigned int intra_mb_mode_cost_table[] = {
+ 0x31110001, // for qp0
+ 0x09110001, // for qp1
+ 0x15030001, // for qp2
+ 0x0b030001, // for qp3
+ 0x0d030011, // for qp4
+ 0x17210011, // for qp5
+ 0x41210011, // for qp6
+ 0x19210011, // for qp7
+ 0x25050003, // for qp8
+ 0x1b130003, // for qp9
+ 0x1d130003, // for qp10
+ 0x27070021, // for qp11
+ 0x51310021, // for qp12
+ 0x29090021, // for qp13
+ 0x35150005, // for qp14
+ 0x2b0b0013, // for qp15
+ 0x2d0d0013, // for qp16
+ 0x37170007, // for qp17
+ 0x61410031, // for qp18
+ 0x39190009, // for qp19
+ 0x45250015, // for qp20
+ 0x3b1b000b, // for qp21
+ 0x3d1d000d, // for qp22
+ 0x47270017, // for qp23
+ 0x71510041, // for qp24 ! center for qp=0..30
+ 0x49290019, // for qp25
+ 0x55350025, // for qp26
+ 0x4b2b001b, // for qp27
+ 0x4d2d001d, // for qp28
+ 0x57370027, // for qp29
+ 0x81610051, // for qp30
+ 0x57270017, // for qp31
+ 0x81510041, // for qp32 ! center for qp=31..51
+ 0x59290019, // for qp33
+ 0x65350025, // for qp34
+ 0x5b2b001b, // for qp35
+ 0x5d2d001d, // for qp36
+ 0x67370027, // for qp37
+ 0x91610051, // for qp38
+ 0x69390029, // for qp39
+ 0x75450035, // for qp40
+ 0x6b3b002b, // for qp41
+ 0x6d3d002d, // for qp42
+ 0x77470037, // for qp43
+ 0xa1710061, // for qp44
+ 0x79490039, // for qp45
+ 0x85550045, // for qp46
+ 0x7b4b003b, // for qp47
+ 0x7d4d003d, // for qp48
+ 0x87570047, // for qp49
+ 0xb1810071, // for qp50
+ 0x89590049 // for qp51
+};
+
+static void gen75_vme_state_setup_fixup(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context,
+ unsigned int *vme_state_message)
+{
+ struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
+ VAEncPictureParameterBufferH264 *pic_param = (VAEncPictureParameterBufferH264 *)encode_state->pic_param_ext->buffer;
+ VAEncSliceParameterBufferH264 *slice_param = (VAEncSliceParameterBufferH264 *)encode_state->slice_params_ext[0]->buffer;
+
+ if (slice_param->slice_type != SLICE_TYPE_I &&
+ slice_param->slice_type != SLICE_TYPE_SI)
+ return;
+ if (encoder_context->rate_control_mode == VA_RC_CQP)
+ vme_state_message[16] = intra_mb_mode_cost_table[pic_param->pic_init_qp + slice_param->slice_qp_delta];
+ else
+ vme_state_message[16] = intra_mb_mode_cost_table[mfc_context->bit_rate_control_context[slice_param->slice_type].QpPrimeY];
+}
+
+static VAStatus gen75_vme_vme_state_setup(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ int is_intra,
+ struct intel_encoder_context *encoder_context)
+{
+ struct gen6_vme_context *vme_context = encoder_context->vme_context;
+ unsigned int *vme_state_message;
+ int i;
+
+ //building VME state message
+ dri_bo_map(vme_context->vme_state.bo, 1);
+ assert(vme_context->vme_state.bo->virtual);
+ vme_state_message = (unsigned int *)vme_context->vme_state.bo->virtual;
+
+ vme_state_message[0] = 0x01010101;
+ vme_state_message[1] = 0x10010101;
+ vme_state_message[2] = 0x0F0F0F0F;
+ vme_state_message[3] = 0x100F0F0F;
+ vme_state_message[4] = 0x01010101;
+ vme_state_message[5] = 0x00010101;
+ vme_state_message[6] = 0x01010101;
+ vme_state_message[7] = 0x10010101;
+ vme_state_message[8] = 0x0F0F0F0F;
+ vme_state_message[9] = 0x100F0F0F;
+ vme_state_message[10] = 0x01010101;
+ vme_state_message[11] = 0x00010101;
+ vme_state_message[12] = 0x00;
+ vme_state_message[13] = 0x00;
+
+ vme_state_message[14] = 0x4a4a;
+ vme_state_message[15] = 0x0;
+ vme_state_message[16] = 0x4a4a4a4a;
+ vme_state_message[17] = 0x4a4a4a4a;
+ vme_state_message[18] = 0x22120200;
+ vme_state_message[19] = 0x62524232;
+
+ for(i = 20; i < 32; i++) {
+ vme_state_message[i] = 0;
+ }
+ //vme_state_message[16] = 0x42424242; //cost function LUT set 0 for Intra
+
+ gen75_vme_state_setup_fixup(ctx, encode_state, encoder_context, vme_state_message);
+
+ dri_bo_unmap( vme_context->vme_state.bo);
+ return VA_STATUS_SUCCESS;
+}
+
+static void
+gen75_vme_fill_vme_batchbuffer(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ int mb_width, int mb_height,
+ int kernel,
+ int transform_8x8_mode_flag,
+ struct intel_encoder_context *encoder_context)
+{
+ struct gen6_vme_context *vme_context = encoder_context->vme_context;
+ int number_mb_cmds;
+ int mb_x = 0, mb_y = 0;
+ int i, s;
+ unsigned int *command_ptr;
+
+ dri_bo_map(vme_context->vme_batchbuffer.bo, 1);
+ command_ptr = vme_context->vme_batchbuffer.bo->virtual;
+
+ for (s = 0; s < encode_state->num_slice_params_ext; s++) {
+ VAEncSliceParameterBufferH264 *pSliceParameter = (VAEncSliceParameterBufferH264 *)encode_state->slice_params_ext[s]->buffer;
+ int slice_mb_begin = pSliceParameter->macroblock_address;
+ int slice_mb_number = pSliceParameter->num_macroblocks;
+
+ for (i = 0; i < slice_mb_number; ) {
+ int mb_count = i + slice_mb_begin;
+ mb_x = mb_count % mb_width;
+ mb_y = mb_count / mb_width;
+ if( i == 0 ) {
+ number_mb_cmds = mb_width; // we must mark the slice edge.
+ } else if ( (i + 128 ) <= slice_mb_number) {
+ number_mb_cmds = 128;
+ } else {
+ number_mb_cmds = slice_mb_number - i;
+ }
+
+ *command_ptr++ = (CMD_MEDIA_OBJECT | (8 - 2));
+ *command_ptr++ = kernel;
+ *command_ptr++ = 0;
+ *command_ptr++ = 0;
+ *command_ptr++ = 0;
+ *command_ptr++ = 0;
+
+ /*inline data */
+ *command_ptr++ = (mb_width << 16 | mb_y << 8 | mb_x);
+ *command_ptr++ = (number_mb_cmds << 16 | transform_8x8_mode_flag | ((i==0) << 1));
+
+ i += number_mb_cmds;
+ }
+ }
+
+ *command_ptr++ = 0;
+ *command_ptr++ = MI_BATCH_BUFFER_END;
+
+ dri_bo_unmap(vme_context->vme_batchbuffer.bo);
+}
+
+static void gen75_vme_media_init(VADriverContextP ctx, struct intel_encoder_context *encoder_context)
+{
+ struct i965_driver_data *i965 = i965_driver_data(ctx);
+ struct gen6_vme_context *vme_context = encoder_context->vme_context;
+ dri_bo *bo;
+
+ i965_gpe_context_init(ctx, &vme_context->gpe_context);
+
+ /* VME output buffer */
+ dri_bo_unreference(vme_context->vme_output.bo);
+ vme_context->vme_output.bo = NULL;
+
+ dri_bo_unreference(vme_context->vme_batchbuffer.bo);
+ vme_context->vme_batchbuffer.bo = NULL;
+
+ /* VME state */
+ dri_bo_unreference(vme_context->vme_state.bo);
+ bo = dri_bo_alloc(i965->intel.bufmgr,
+ "Buffer",
+ 1024*16, 64);
+ assert(bo);
+ vme_context->vme_state.bo = bo;
+}
+
+static void gen75_vme_pipeline_programing(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
+{
+ struct gen6_vme_context *vme_context = encoder_context->vme_context;
+ struct intel_batchbuffer *batch = encoder_context->base.batch;
+ VAEncPictureParameterBufferH264 *pPicParameter = (VAEncPictureParameterBufferH264 *)encode_state->pic_param_ext->buffer;
+ VAEncSliceParameterBufferH264 *pSliceParameter = (VAEncSliceParameterBufferH264 *)encode_state->slice_params_ext[0]->buffer;
+ VAEncSequenceParameterBufferH264 *pSequenceParameter = (VAEncSequenceParameterBufferH264 *)encode_state->seq_param_ext->buffer;
+ int is_intra = pSliceParameter->slice_type == SLICE_TYPE_I;
+ int width_in_mbs = pSequenceParameter->picture_width_in_mbs;
+ int height_in_mbs = pSequenceParameter->picture_height_in_mbs;
+
+ gen75_vme_fill_vme_batchbuffer(ctx,
+ encode_state,
+ width_in_mbs, height_in_mbs,
+ is_intra ? VME_INTRA_SHADER : VME_INTER_SHADER,
+ pPicParameter->pic_fields.bits.transform_8x8_mode_flag,
+ encoder_context);
+
+ intel_batchbuffer_start_atomic(batch, 0x1000);
+ gen6_gpe_pipeline_setup(ctx, &vme_context->gpe_context, batch);
+ BEGIN_BATCH(batch, 2);
+ OUT_BATCH(batch, MI_BATCH_BUFFER_START | (2 << 6));
+ OUT_RELOC(batch,
+ vme_context->vme_batchbuffer.bo,
+ I915_GEM_DOMAIN_COMMAND, 0,
+ 0);
+ ADVANCE_BATCH(batch);
+
+ intel_batchbuffer_end_atomic(batch);
+}
+
+static VAStatus gen75_vme_prepare(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
+{
+ VAStatus vaStatus = VA_STATUS_SUCCESS;
+ VAEncSliceParameterBufferH264 *pSliceParameter = (VAEncSliceParameterBufferH264 *)encode_state->slice_params_ext[0]->buffer;
+ int is_intra = pSliceParameter->slice_type == SLICE_TYPE_I;
+
+ /*Setup all the memory object*/
+ gen75_vme_surface_setup(ctx, encode_state, is_intra, encoder_context);
+ gen75_vme_interface_setup(ctx, encode_state, encoder_context);
+ gen75_vme_constant_setup(ctx, encode_state, encoder_context);
+ gen75_vme_vme_state_setup(ctx, encode_state, is_intra, encoder_context);
+
+ /*Programing media pipeline*/
+ gen75_vme_pipeline_programing(ctx, encode_state, encoder_context);
+
+ return vaStatus;
+}
+
+static VAStatus gen75_vme_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);
+
+ return VA_STATUS_SUCCESS;
+}
+
+static VAStatus gen75_vme_stop(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
+{
+ return VA_STATUS_SUCCESS;
+}
+
+static VAStatus
+gen75_vme_pipeline(VADriverContextP ctx,
+ VAProfile profile,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
+{
+ gen75_vme_media_init(ctx, encoder_context);
+ gen75_vme_prepare(ctx, encode_state, encoder_context);
+ gen75_vme_run(ctx, encode_state, encoder_context);
+ gen75_vme_stop(ctx, encode_state, encoder_context);
+
+ return VA_STATUS_SUCCESS;
+}
+
+static void
+gen75_vme_context_destroy(void *context)
+{
+ struct gen6_vme_context *vme_context = context;
+
+ i965_gpe_context_destroy(&vme_context->gpe_context);
+
+ dri_bo_unreference(vme_context->vme_output.bo);
+ vme_context->vme_output.bo = NULL;
+
+ dri_bo_unreference(vme_context->vme_state.bo);
+ vme_context->vme_state.bo = NULL;
+
+ dri_bo_unreference(vme_context->vme_batchbuffer.bo);
+ vme_context->vme_batchbuffer.bo = NULL;
+
+ free(vme_context);
+}
+
+Bool gen75_vme_context_init(VADriverContextP ctx, struct intel_encoder_context *encoder_context)
+{
+ struct gen6_vme_context *vme_context = calloc(1, sizeof(struct gen6_vme_context));
+
+ vme_context->gpe_context.surface_state_binding_table.length = (SURFACE_STATE_PADDED_SIZE + sizeof(unsigned int)) * MAX_MEDIA_SURFACES_GEN6;
+
+ vme_context->gpe_context.idrt.max_entries = MAX_INTERFACE_DESC_GEN6;
+ vme_context->gpe_context.idrt.entry_size = sizeof(struct gen6_interface_descriptor_data);
+
+ vme_context->gpe_context.curbe.length = CURBE_TOTAL_DATA_LENGTH;
+
+ vme_context->gpe_context.vfe_state.max_num_threads = 60 - 1;
+ vme_context->gpe_context.vfe_state.num_urb_entries = 16;
+ vme_context->gpe_context.vfe_state.gpgpu_mode = 0;
+ vme_context->gpe_context.vfe_state.urb_entry_size = 59 - 1;
+ vme_context->gpe_context.vfe_state.curbe_allocation_size = CURBE_ALLOCATION_SIZE - 1;
+
+ i965_gpe_load_kernels(ctx,
+ &vme_context->gpe_context,
+ gen75_vme_kernels,
+ GEN6_VME_KERNEL_NUMBER);
+ vme_context->vme_surface2_setup = gen7_gpe_surface2_setup;
+ vme_context->vme_media_rw_surface_setup = gen7_gpe_media_rw_surface_setup;
+ vme_context->vme_buffer_suface_setup = gen7_gpe_buffer_suface_setup;
+
+ encoder_context->vme_context = vme_context;
+ encoder_context->vme_context_destroy = gen75_vme_context_destroy;
+ encoder_context->vme_pipeline = gen75_vme_pipeline;
+
+ return True;
+}