/* * 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 * Xiang Haihao * */ #include #include #include #include #include #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" #define MFC_SOFTWARE_HASWELL 1 #define B0_STEP_REV 2 #define IS_STEPPING_BPLUS(i965) ((i965->intel.revision) >= B0_STEP_REV) 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 }, }; #define INTER_MODE_MASK 0x03 #define INTER_8X8 0x03 #define SUBMB_SHAPE_MASK 0x00FF00 #define INTER_MV8 (4 << 20) #define INTER_MV32 (6 << 20) 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_bplus(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, 26); OUT_BCS_BATCH(batch, MFX_IND_OBJ_BASE_ADDR_STATE | (26 - 2)); /* the DW1-3 is for the MFX indirect bistream offset */ OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); /* the DW4-5 is the MFX upper bound */ OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); /* the DW6-10 is for 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, 0); OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0x80000000); /* must set, up to 2G */ OUT_BCS_BATCH(batch, 0); /* the DW11-15 is for MFX IT-COFF. Not used on encoder */ OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); /* the DW16-20 is for MFX indirect DBLK. Not used on encoder */ OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); /* the DW21-25 is for 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_BATCH(batch, 0); OUT_BCS_BATCH(batch, 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); 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; struct i965_driver_data *i965 = i965_driver_data(ctx); if (IS_STEPPING_BPLUS(i965)) { gen75_mfc_ind_obj_base_addr_state_bplus(ctx, encoder_context); return; } 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 << 8) | /* 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_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_bplus(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, 61); OUT_BCS_BATCH(batch, MFX_PIPE_BUF_ADDR_STATE | (61 - 2)); /* the DW1-3 is for pre_deblocking */ if (mfc_context->pre_deblocking_output.bo) OUT_BCS_RELOC(batch, mfc_context->pre_deblocking_output.bo, I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION, 0); else OUT_BCS_BATCH(batch, 0); /* pre output addr */ OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); /* the DW4-6 is for the post_deblocking */ 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_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); /* the DW7-9 is for the uncompressed_picture */ OUT_BCS_RELOC(batch, mfc_context->uncompressed_picture_source.bo, I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION, 0); /* uncompressed data */ OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); /* the DW10-12 is for the mb status */ OUT_BCS_RELOC(batch, mfc_context->macroblock_status_buffer.bo, I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION, 0); /* StreamOut data*/ OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); /* the DW13-15 is for the intra_row_store_scratch */ OUT_BCS_RELOC(batch, mfc_context->intra_row_store_scratch_buffer.bo, I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION, 0); OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); /* the DW16-18 is for the deblocking filter */ OUT_BCS_RELOC(batch, mfc_context->deblocking_filter_row_store_scratch_buffer.bo, I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION, 0); OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); /* the DW 19-50 is for 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_BATCH(batch, 0); } OUT_BCS_BATCH(batch, 0); /* The DW 52-54 is for the MB status buffer */ OUT_BCS_RELOC(batch, mfc_context->macroblock_status_buffer.bo, I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION, 0); /* Macroblock status buffer*/ OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); /* the DW 55-57 is the ILDB buffer */ OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); /* the DW 58-60 is the second ILDB buffer */ OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); ADVANCE_BCS_BATCH(batch); } 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; struct i965_driver_data *i965 = i965_driver_data(ctx); int i; if (IS_STEPPING_BPLUS(i965)) { gen75_mfc_pipe_buf_addr_state_bplus(ctx, encoder_context); return; } BEGIN_BCS_BATCH(batch, 25); OUT_BCS_BATCH(batch, MFX_PIPE_BUF_ADDR_STATE | (25 - 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*/ OUT_BCS_BATCH(batch, 0); ADVANCE_BCS_BATCH(batch); } static void gen75_mfc_avc_directmode_state_bplus(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, 71); OUT_BCS_BATCH(batch, MFX_AVC_DIRECTMODE_STATE | (71 - 2)); /* Reference frames and Current frames */ /* the DW1-32 is for the direct MV for reference */ for(i = 0; i < NUM_MFC_DMV_BUFFERS - 2; i += 2) { 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); OUT_BCS_BATCH(batch, 0); } else { OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); } } OUT_BCS_BATCH(batch, 0); /* the DW34-36 is the MV for the current reference */ OUT_BCS_RELOC(batch, mfc_context->direct_mv_buffers[NUM_MFC_DMV_BUFFERS - 2].bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 0); OUT_BCS_BATCH(batch, 0); 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_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; struct i965_driver_data *i965 = i965_driver_data(ctx); int i; if (IS_STEPPING_BPLUS(i965)) { gen75_mfc_avc_directmode_state_bplus(ctx, encoder_context); return; } 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_bplus(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, 10); OUT_BCS_BATCH(batch, MFX_BSP_BUF_BASE_ADDR_STATE | (10 - 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); /* the DW4-6 is for MPR Row Store Scratch Buffer Base Address */ OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); /* the DW7-9 is for Bitplane Read Buffer Base Address */ OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); OUT_BCS_BATCH(batch, 0); 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; struct i965_driver_data *i965 = i965_driver_data(ctx); if (IS_STEPPING_BPLUS(i965)) { gen75_mfc_bsp_buf_base_addr_state_bplus(ctx, encoder_context); return; } 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 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); } #ifdef MFC_SOFTWARE_HASWELL 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 = 12; unsigned int intra_msg; #define INTRA_MSG_FLAG (1 << 13) #define INTRA_MBTYPE_MASK (0x1F0000) if (batch == NULL) batch = encoder_context->base.batch; BEGIN_BCS_BATCH(batch, len_in_dwords); intra_msg = msg[0] & 0xC0FF; intra_msg |= INTRA_MSG_FLAG; intra_msg |= ((msg[0] & INTRA_MBTYPE_MASK) >> 8); 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 */ intra_msg); 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) ); OUT_BCS_BATCH(batch, 0); 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 = 12; unsigned int inter_msg = 0; 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)); inter_msg = 32; /* MV quantity */ if ((msg[0] & INTER_MODE_MASK) == INTER_8X8) { if (msg[1] & SUBMB_SHAPE_MASK) inter_msg = 128; } OUT_BCS_BATCH(batch, inter_msg); /* 32 MV*/ OUT_BCS_BATCH(batch, offset); inter_msg = msg[0] & (0x1F00FFFF); inter_msg |= INTER_MV8; if (((msg[0] & INTER_MODE_MASK) == INTER_8X8) && (msg[1] & SUBMB_SHAPE_MASK)) { inter_msg |= INTER_MV32; } OUT_BCS_BATCH(batch, inter_msg); 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 inter_msg = msg[1] >> 8; /*Stuff for Inter MB*/ OUT_BCS_BATCH(batch, inter_msg); 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) ); OUT_BCS_BATCH(batch, 0x0); ADVANCE_BCS_BATCH(batch); return len_in_dwords; } #define INTRA_RDO_OFFSET 4 #define INTER_RDO_OFFSET 54 #define INTER_MSG_OFFSET 52 #define INTER_MV_OFFSET 224 #define RDO_MASK 0xFFFF 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; unsigned char *msg_ptr = NULL; 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) intel_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_ptr = (unsigned char *)vme_context->vme_output.bo->virtual; if (is_intra) { msg = (unsigned int *) (msg_ptr + pSliceParameter->macroblock_address * vme_context->vme_output.size_block); } else { msg = (unsigned int *) (msg_ptr + pSliceParameter->macroblock_address * vme_context->vme_output.size_block); } 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; msg = (unsigned int *) (msg_ptr + i * vme_context->vme_output.size_block); if (is_intra) { assert(msg); gen75_mfc_avc_pak_object_intra(ctx, x, y, last_mb, qp, msg, encoder_context, 0, 0, slice_batch); } else { int inter_rdo, intra_rdo; inter_rdo = msg[INTER_RDO_OFFSET] & RDO_MASK; intra_rdo = msg[INTRA_RDO_OFFSET] & RDO_MASK; offset = i * vme_context->vme_output.size_block + INTER_MV_OFFSET; if (intra_rdo < inter_rdo) { gen75_mfc_avc_pak_object_intra(ctx, x, y, last_mb, qp, msg, encoder_context, 0, 0, slice_batch); } else { msg += INTER_MSG_OFFSET; gen75_mfc_avc_pak_object_inter(ctx, x, y, last_mb, qp, msg, offset, encoder_context, 0, 0, pSliceParameter->slice_type, slice_batch); } } } 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) intel_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 ( intel_mfc_interlace_check(ctx, encode_state, encoder_context) ) { fprintf(stderr, "Current VA driver don't support interlace mode!\n"); assert(0); return; } #ifdef MFC_SOFTWARE_HASWELL 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 = intel_mfc_brc_postpack(encode_state, mfc_context, current_frame_bits_size); if (sts == BRC_NO_HRD_VIOLATION) { intel_mfc_hrd_context_update(encode_state, mfc_context); break; } else if (sts == BRC_OVERFLOW_WITH_MIN_QP || sts == BRC_UNDERFLOW_WITH_MAX_QP) { if (!mfc_context->hrd.violation_noted) { fprintf(stderr, "Unrepairable %s!\n", (sts == BRC_OVERFLOW_WITH_MIN_QP)? "overflow": "underflow"); mfc_context->hrd.violation_noted = 1; } return VA_STATUS_SUCCESS; } } else { break; } } return VA_STATUS_SUCCESS; } static 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 = intel_mfc_brc_prepare; return True; }