X-Git-Url: http://review.tizen.org/git/?a=blobdiff_plain;f=src%2Fgen75_vme.c;h=585375766a73516ccdb6e95fa925a005a96616cd;hb=0e29103373dbadb30bab189722f19c3f9447a306;hp=471e8a8b36ba64cd2c153cc0bfbb4e1f7b87495c;hpb=1249f08f1b652a36a7cf37738a61c2d8174c6af6;p=profile%2Fivi%2Fvaapi-intel-driver.git diff --git a/src/gen75_vme.c b/src/gen75_vme.c index 471e8a8..5853757 100644 --- a/src/gen75_vme.c +++ b/src/gen75_vme.c @@ -28,6 +28,7 @@ #include #include +#include #include #include @@ -54,6 +55,7 @@ #define VME_INTRA_SHADER 0 #define VME_INTER_SHADER 1 +#define VME_BINTER_SHADER 3 #define VME_BATCHBUFFER 2 #define CURBE_ALLOCATION_SIZE 37 /* in 256-bit */ @@ -70,6 +72,10 @@ static const uint32_t gen75_vme_inter_frame[][4] = { #include "shaders/vme/inter_frame_haswell.g75b" }; +static const uint32_t gen75_vme_inter_bframe[][4] = { +#include "shaders/vme/inter_bframe_haswell.g75b" +}; + static const uint32_t gen75_vme_batchbuffer[][4] = { #include "shaders/vme/batchbuffer.g75b" }; @@ -96,6 +102,49 @@ static struct i965_kernel gen75_vme_kernels[] = { sizeof(gen75_vme_batchbuffer), NULL }, + { + "VME inter BFrame", + VME_BINTER_SHADER, + gen75_vme_inter_bframe, + sizeof(gen75_vme_inter_bframe), + NULL + } +}; + +static const uint32_t gen75_vme_mpeg2_intra_frame[][4] = { +#include "shaders/vme/intra_frame_haswell.g75b" +}; + +static const uint32_t gen75_vme_mpeg2_inter_frame[][4] = { +#include "shaders/vme/mpeg2_inter_frame_haswell.g75b" +}; + +static const uint32_t gen75_vme_mpeg2_batchbuffer[][4] = { +#include "shaders/vme/batchbuffer.g75b" +}; + +static struct i965_kernel gen75_vme_mpeg2_kernels[] = { + { + "VME Intra Frame", + VME_INTRA_SHADER, /*index*/ + gen75_vme_mpeg2_intra_frame, + sizeof(gen75_vme_mpeg2_intra_frame), + NULL + }, + { + "VME inter Frame", + VME_INTER_SHADER, + gen75_vme_mpeg2_inter_frame, + sizeof(gen75_vme_mpeg2_inter_frame), + NULL + }, + { + "VME BATCHBUFFER", + VME_BATCHBUFFER, + gen75_vme_mpeg2_batchbuffer, + sizeof(gen75_vme_mpeg2_batchbuffer), + NULL + }, }; /* only used for VME source surface state */ @@ -264,7 +313,7 @@ static VAStatus gen75_vme_interface_setup(VADriverContextP ctx, assert(bo->virtual); desc = bo->virtual; - for (i = 0; i < GEN6_VME_KERNEL_NUMBER; i++) { + for (i = 0; i < vme_context->vme_kernel_sum; i++) { struct i965_kernel *kernel; kernel = &vme_context->gpe_context.kernels[i]; assert(sizeof(*desc) == 32); @@ -301,11 +350,21 @@ static VAStatus gen75_vme_constant_setup(VADriverContextP ctx, int mv_num = 32; vme_state_message = (unsigned int *)vme_context->vme_state_message; - if (vme_context->h264_level >= 30) { - mv_num = 16; - if (vme_context->h264_level >= 31) - mv_num = 8; - } + + if (encoder_context->profile == VAProfileH264Baseline || + encoder_context->profile == VAProfileH264Main || + encoder_context->profile == VAProfileH264High) { + if (vme_context->h264_level >= 30) { + mv_num = 16; + + if (vme_context->h264_level >= 31) + mv_num = 8; + } + } else if (encoder_context->profile == VAProfileMPEG2Simple || + encoder_context->profile == VAProfileMPEG2Main) { + mv_num = 2; + } + vme_state_message[31] = mv_num; dri_bo_map(vme_context->gpe_context.curbe.bo, 1); @@ -318,7 +377,7 @@ static VAStatus gen75_vme_constant_setup(VADriverContextP ctx, */ memcpy(constant_buffer, (char *)vme_context->vme_state_message, 128); - dri_bo_unmap( vme_context->gpe_context.curbe.bo); + dri_bo_unmap(vme_context->gpe_context.curbe.bo); return VA_STATUS_SUCCESS; } @@ -393,7 +452,7 @@ static void gen75_vme_state_setup_fixup(VADriverContextP ctx, if (encoder_context->rate_control_mode == VA_RC_CQP) vme_state_message[0] = intra_mb_mode_cost_table[pic_param->pic_init_qp + slice_param->slice_qp_delta]; else - vme_state_message[0] = intra_mb_mode_cost_table[mfc_context->bit_rate_control_context[slice_param->slice_type].QpPrimeY]; + vme_state_message[0] = intra_mb_mode_cost_table[mfc_context->bit_rate_control_context[SLICE_TYPE_I].QpPrimeY]; } static VAStatus gen75_vme_vme_state_setup(VADriverContextP ctx, @@ -419,11 +478,23 @@ static VAStatus gen75_vme_vme_state_setup(VADriverContextP ctx, vme_state_message[i] = 0; } - gen75_vme_state_setup_fixup(ctx, encode_state, encoder_context, vme_state_message); + switch (encoder_context->profile) { + case VAProfileH264Baseline: + case VAProfileH264Main: + case VAProfileH264High: + gen75_vme_state_setup_fixup(ctx, encode_state, encoder_context, vme_state_message); + + break; + + default: + /* no fixup */ + break; + } return VA_STATUS_SUCCESS; } + static void gen75_vme_fill_vme_batchbuffer(VADriverContextP ctx, struct encode_state *encode_state, @@ -437,12 +508,6 @@ gen75_vme_fill_vme_batchbuffer(VADriverContextP ctx, int i, s; unsigned int *command_ptr; -#define INTRA_PRED_AVAIL_FLAG_AE 0x60 -#define INTRA_PRED_AVAIL_FLAG_B 0x10 -#define INTRA_PRED_AVAIL_FLAG_C 0x8 -#define INTRA_PRED_AVAIL_FLAG_D 0x4 -#define INTRA_PRED_AVAIL_FLAG_BCD_MASK 0x1C - dri_bo_map(vme_context->vme_batchbuffer.bo, 1); command_ptr = vme_context->vme_batchbuffer.bo->virtual; @@ -502,9 +567,7 @@ gen75_vme_fill_vme_batchbuffer(VADriverContextP ctx, 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); @@ -529,14 +592,42 @@ static void gen75_vme_pipeline_programing(VADriverContextP ctx, 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; + int kernel_shader; + bool allow_hwscore = true; + int s; - gen75_vme_fill_vme_batchbuffer(ctx, + for (s = 0; s < encode_state->num_slice_params_ext; s++) { + pSliceParameter = (VAEncSliceParameterBufferH264 *)encode_state->slice_params_ext[s]->buffer; + if ((pSliceParameter->macroblock_address % width_in_mbs)) { + allow_hwscore = false; + break; + } + } + if ((pSliceParameter->slice_type == SLICE_TYPE_I) || + (pSliceParameter->slice_type == SLICE_TYPE_I)) { + kernel_shader = VME_INTRA_SHADER; + } else if ((pSliceParameter->slice_type == SLICE_TYPE_P) || + (pSliceParameter->slice_type == SLICE_TYPE_SP)) { + kernel_shader = VME_INTER_SHADER; + } else { + kernel_shader = VME_BINTER_SHADER; + if (!allow_hwscore) + kernel_shader = VME_INTER_SHADER; + } + if (allow_hwscore) + gen7_vme_walker_fill_vme_batchbuffer(ctx, + encode_state, + width_in_mbs, height_in_mbs, + kernel_shader, + pPicParameter->pic_fields.bits.transform_8x8_mode_flag, + encoder_context); + else + gen75_vme_fill_vme_batchbuffer(ctx, encode_state, width_in_mbs, height_in_mbs, - is_intra ? VME_INTRA_SHADER : VME_INTER_SHADER, + kernel_shader, pPicParameter->pic_fields.bits.transform_8x8_mode_flag, encoder_context); @@ -615,6 +706,262 @@ gen75_vme_pipeline(VADriverContextP ctx, } static void +gen75_vme_mpeg2_output_buffer_setup(VADriverContextP ctx, + struct encode_state *encode_state, + int index, + int is_intra, + 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; + VAEncSequenceParameterBufferMPEG2 *seq_param = (VAEncSequenceParameterBufferMPEG2 *)encode_state->seq_param_ext->buffer; + int width_in_mbs = ALIGN(seq_param->picture_width, 16) / 16; + int height_in_mbs = ALIGN(seq_param->picture_height, 16) / 16; + + 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 * 2; + else + vme_context->vme_output.size_block = INTRA_VME_OUTPUT_IN_BYTES * 24; + /* + * Inter MV . 32-byte Intra search + 16 IME info + 128 IME MV + 32 IME Ref + * + 16 FBR Info + 128 FBR MV + 32 FBR Ref. + * 16 * (2 + 2 * (1 + 8 + 2))= 16 * 24. + */ + + 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_mpeg2_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; + VAEncSequenceParameterBufferMPEG2 *seq_param = (VAEncSequenceParameterBufferMPEG2 *)encode_state->seq_param_ext->buffer; + int width_in_mbs = ALIGN(seq_param->picture_width, 16) / 16; + int height_in_mbs = ALIGN(seq_param->picture_height, 16) / 16; + + vme_context->vme_batchbuffer.num_blocks = width_in_mbs * height_in_mbs + 1; + vme_context->vme_batchbuffer.size_block = 64; /* 4 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_mpeg2_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; + VAEncPictureParameterBufferMPEG2 *pic_param = (VAEncPictureParameterBufferMPEG2 *)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); + gen75_vme_media_chroma_source_surface_state(ctx, 6, obj_surface, encoder_context); + + if (!is_intra) { + /* reference 0 */ + obj_surface = SURFACE(pic_param->forward_reference_picture); + assert(obj_surface); + if ( obj_surface->bo != NULL) + gen75_vme_source_surface_state(ctx, 1, obj_surface, encoder_context); + + /* reference 1 */ + obj_surface = SURFACE(pic_param->backward_reference_picture); + if (obj_surface && obj_surface->bo != NULL) + gen75_vme_source_surface_state(ctx, 2, obj_surface, encoder_context); + } + + /* VME output */ + gen75_vme_mpeg2_output_buffer_setup(ctx, encode_state, 3, is_intra, encoder_context); + gen75_vme_mpeg2_output_vme_batchbuffer_setup(ctx, encode_state, 5, encoder_context); + + return VA_STATUS_SUCCESS; +} + +static void +gen75_vme_mpeg2_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 mb_x = 0, mb_y = 0; + int i, s, j; + 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++) { + VAEncSliceParameterBufferMPEG2 *slice_param = (VAEncSliceParameterBufferMPEG2 *)encode_state->slice_params_ext[s]->buffer; + + for (j = 0; j < encode_state->slice_params_ext[s]->num_elements; j++) { + int slice_mb_begin = slice_param->macroblock_address; + int slice_mb_number = slice_param->num_macroblocks; + unsigned int mb_intra_ub; + int slice_mb_x = slice_param->macroblock_address % mb_width; + + 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; + mb_intra_ub = 0; + + if (mb_x != 0) { + mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_AE; + } + + if (mb_y != 0) { + mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_B; + + if (mb_x != 0) + mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_D; + + if (mb_x != (mb_width -1)) + mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_C; + } + + if (i < mb_width) { + if (i == 0) + mb_intra_ub &= ~(INTRA_PRED_AVAIL_FLAG_AE); + + mb_intra_ub &= ~(INTRA_PRED_AVAIL_FLAG_BCD_MASK); + + if ((i == (mb_width - 1)) && slice_mb_x) { + mb_intra_ub |= INTRA_PRED_AVAIL_FLAG_C; + } + } + + if ((i == mb_width) && slice_mb_x) { + mb_intra_ub &= ~(INTRA_PRED_AVAIL_FLAG_D); + } + + *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++ = ( (1 << 16) | transform_8x8_mode_flag | (mb_intra_ub << 8)); + + i += 1; + } + + slice_param++; + } + } + + *command_ptr++ = 0; + *command_ptr++ = MI_BATCH_BUFFER_END; + + dri_bo_unmap(vme_context->vme_batchbuffer.bo); +} + +static void +gen75_vme_mpeg2_pipeline_programing(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; + struct intel_batchbuffer *batch = encoder_context->base.batch; + VAEncSequenceParameterBufferMPEG2 *seq_param = (VAEncSequenceParameterBufferMPEG2 *)encode_state->seq_param_ext->buffer; + int width_in_mbs = ALIGN(seq_param->picture_width, 16) / 16; + int height_in_mbs = ALIGN(seq_param->picture_height, 16) / 16; + + gen75_vme_mpeg2_fill_vme_batchbuffer(ctx, + encode_state, + width_in_mbs, height_in_mbs, + is_intra ? VME_INTRA_SHADER : VME_INTER_SHADER, + 0, + 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_mpeg2_prepare(VADriverContextP ctx, + struct encode_state *encode_state, + struct intel_encoder_context *encoder_context) +{ + VAStatus vaStatus = VA_STATUS_SUCCESS; + VAEncSliceParameterBufferMPEG2 *slice_param = (VAEncSliceParameterBufferMPEG2 *)encode_state->slice_params_ext[0]->buffer; + + /*Setup all the memory object*/ + gen75_vme_mpeg2_surface_setup(ctx, encode_state, slice_param->is_intra_slice, encoder_context); + gen75_vme_interface_setup(ctx, encode_state, encoder_context); + gen75_vme_vme_state_setup(ctx, encode_state, slice_param->is_intra_slice, encoder_context); + gen75_vme_constant_setup(ctx, encode_state, encoder_context); + + /*Programing media pipeline*/ + gen75_vme_mpeg2_pipeline_programing(ctx, encode_state, slice_param->is_intra_slice, encoder_context); + + return vaStatus; +} + +static VAStatus +gen75_vme_mpeg2_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_mpeg2_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; @@ -641,7 +988,33 @@ gen75_vme_context_destroy(void *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)); - + struct i965_kernel *vme_kernel_list = NULL; + int i965_kernel_num; + + switch (encoder_context->profile) { + case VAProfileH264Baseline: + case VAProfileH264Main: + case VAProfileH264High: + vme_kernel_list = gen75_vme_kernels; + encoder_context->vme_pipeline = gen75_vme_pipeline; + i965_kernel_num = sizeof(gen75_vme_kernels) / sizeof(struct i965_kernel); + break; + + case VAProfileMPEG2Simple: + case VAProfileMPEG2Main: + vme_kernel_list = gen75_vme_mpeg2_kernels; + encoder_context->vme_pipeline = gen75_vme_mpeg2_pipeline; + i965_kernel_num = sizeof(gen75_vme_mpeg2_kernels) / sizeof(struct i965_kernel); + + break; + + default: + /* never get here */ + assert(0); + + break; + } + vme_context->vme_kernel_sum = i965_kernel_num; 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; @@ -655,10 +1028,12 @@ Bool gen75_vme_context_init(VADriverContextP ctx, struct intel_encoder_context * vme_context->gpe_context.vfe_state.urb_entry_size = 59 - 1; vme_context->gpe_context.vfe_state.curbe_allocation_size = CURBE_ALLOCATION_SIZE - 1; + gen7_vme_scoreboard_init(ctx, vme_context); + i965_gpe_load_kernels(ctx, &vme_context->gpe_context, - gen75_vme_kernels, - GEN6_VME_KERNEL_NUMBER); + vme_kernel_list, + i965_kernel_num); 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; @@ -666,8 +1041,8 @@ Bool gen75_vme_context_init(VADriverContextP ctx, struct intel_encoder_context * encoder_context->vme_context = vme_context; encoder_context->vme_context_destroy = gen75_vme_context_destroy; - encoder_context->vme_pipeline = gen75_vme_pipeline; vme_context->vme_state_message = malloc(VME_MSG_LENGTH * sizeof(int)); + return True; }