[Libva] [PATCH 18/31] ENC: add REF frame QA caculation and MB level const data init for AVC MBenc stage
Sean V Kelley
seanvk at posteo.de
Tue Jan 10 23:38:00 UTC 2017
From: Pengfei Qu <Pengfei.Qu at intel.com>
Signed-off-by: Pengfei Qu <Pengfei.Qu at intel.com>
Reviewed-by: Sean V Kelley <seanvk at posteo.de>
---
src/gen9_avc_encoder.c | 226 +++++++++++++++++++++++++++++++++++++++++++++++++
1 file changed, 226 insertions(+)
diff --git a/src/gen9_avc_encoder.c b/src/gen9_avc_encoder.c
index a025f563..cd847a88 100755
--- a/src/gen9_avc_encoder.c
+++ b/src/gen9_avc_encoder.c
@@ -2618,3 +2618,229 @@ gen9_avc_kernel_brc_mb_update(VADriverContextP ctx,
return VA_STATUS_SUCCESS;
}
+
+/*
+mbenc kernel related function,it include intra dist kernel
+*/
+static int
+gen9_avc_get_biweight(int dist_scale_factor_ref_id0_list0, unsigned short weighted_bipredidc)
+{
+ int biweight = 32; // default value
+
+ /* based on kernel HLD*/
+ if (weighted_bipredidc != INTEL_AVC_WP_MODE_IMPLICIT)
+ {
+ biweight = 32;
+ }
+ else
+ {
+ biweight = (dist_scale_factor_ref_id0_list0 + 2) >> 2;
+
+ if (biweight != 16 && biweight != 21 &&
+ biweight != 32 && biweight != 43 && biweight != 48)
+ {
+ biweight = 32; // If # of B-pics between two refs is more than 3. VME does not support it.
+ }
+ }
+
+ return biweight;
+}
+
+static void
+gen9_avc_get_dist_scale_factor(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
+{
+ struct encoder_vme_mfc_context * vme_context = (struct encoder_vme_mfc_context *)encoder_context->vme_context;
+ struct avc_enc_state * avc_state = (struct avc_enc_state * )vme_context->private_enc_state;
+ VAEncSliceParameterBufferH264 * slice_param = avc_state->slice_param[0];
+ VAEncPictureParameterBufferH264 *pic_param = avc_state->pic_param;
+
+ int max_num_references;
+ VAPictureH264 *curr_pic;
+ VAPictureH264 *ref_pic_l0;
+ VAPictureH264 *ref_pic_l1;
+ int i = 0;
+ int tb = 0;
+ int td = 0;
+ int tx = 0;
+ int tmp = 0;
+ int poc0 = 0;
+ int poc1 = 0;
+
+ max_num_references = pic_param->num_ref_idx_l0_active_minus1 + 1;
+
+ memset(avc_state->dist_scale_factor_list0,0,32*sizeof(unsigned int));
+ curr_pic = &pic_param->CurrPic;
+ for(i = 0; i < max_num_references; i++)
+ {
+ ref_pic_l0 = &(slice_param->RefPicList0[i]);
+
+ if((ref_pic_l0->flags & VA_PICTURE_H264_INVALID) ||
+ (ref_pic_l0->picture_id == VA_INVALID_SURFACE) )
+ break;
+ ref_pic_l1 = &(slice_param->RefPicList1[0]);
+ if((ref_pic_l0->flags & VA_PICTURE_H264_INVALID) ||
+ (ref_pic_l0->picture_id == VA_INVALID_SURFACE) )
+ break;
+
+ poc0 = (curr_pic->TopFieldOrderCnt - ref_pic_l0->TopFieldOrderCnt);
+ poc1 = (ref_pic_l1->TopFieldOrderCnt - ref_pic_l0->TopFieldOrderCnt);
+ CLIP(poc0,-128,127);
+ CLIP(poc1,-128,127);
+ tb = poc0;
+ td = poc1;
+
+ if(td == 0)
+ {
+ td = 1;
+ }
+ tmp = (td/2 > 0)?(td/2):(-(td/2));
+ tx = (16384 + tmp)/td ;
+ tmp = (tb*tx+32)>>6;
+ CLIP(tmp,-1024,1023);
+ avc_state->dist_scale_factor_list0[i] = tmp;
+ }
+ return;
+}
+
+static unsigned int
+gen9_avc_get_qp_from_ref_list(VADriverContextP ctx,
+ VAEncSliceParameterBufferH264 *slice_param,
+ int list,
+ int ref_frame_idx)
+{
+ struct i965_driver_data *i965 = i965_driver_data(ctx);
+ struct object_surface *obj_surface;
+ struct gen9_surface_avc *avc_priv_surface;
+ VASurfaceID surface_id;
+
+ assert(slice_param);
+ assert(list < 2);
+
+ if(list == 0)
+ {
+ if(ref_frame_idx < slice_param->num_ref_idx_l0_active_minus1 + 1)
+ surface_id = slice_param->RefPicList0[ref_frame_idx].picture_id;
+ else
+ return 0;
+ }else
+ {
+ if(ref_frame_idx < slice_param->num_ref_idx_l1_active_minus1 + 1)
+ surface_id = slice_param->RefPicList1[ref_frame_idx].picture_id;
+ else
+ return 0;
+ }
+ obj_surface = SURFACE(surface_id);
+ if(obj_surface && obj_surface->private_data)
+ {
+ avc_priv_surface = obj_surface->private_data;
+ return avc_priv_surface->qp_value;
+ }else
+ {
+ return 0;
+ }
+}
+
+static void
+gen9_avc_load_mb_brc_const_data(VADriverContextP ctx,
+ struct encode_state *encode_state,
+ struct intel_encoder_context *encoder_context)
+{
+ struct encoder_vme_mfc_context * vme_context = (struct encoder_vme_mfc_context *)encoder_context->vme_context;
+ struct gen9_avc_encoder_context * avc_ctx = (struct gen9_avc_encoder_context * )vme_context->private_enc_ctx;
+ struct generic_enc_codec_state * generic_state = (struct generic_enc_codec_state * )vme_context->generic_enc_state;
+ struct avc_enc_state * avc_state = (struct avc_enc_state * )vme_context->private_enc_state;
+ VAEncPictureParameterBufferH264 *pic_param = avc_state->pic_param;
+
+ struct i965_gpe_resource *gpe_resource = NULL;
+ unsigned int * data =NULL;
+ unsigned int * data_tmp = NULL;
+ unsigned int size = 16 * 52;
+ unsigned int table_idx = 0;
+ unsigned int block_based_skip_enable = avc_state->block_based_skip_enable;
+ unsigned int transform_8x8_mode_flag = pic_param->pic_fields.bits.transform_8x8_mode_flag;
+ int i = 0;
+
+ gpe_resource = &(avc_ctx->res_mbbrc_const_data_buffer);
+ assert(gpe_resource);
+ data = i965_map_gpe_resource(gpe_resource);
+ assert(data);
+
+ table_idx = slice_type_kernel[generic_state->frame_type];
+
+ memcpy(data,gen9_avc_mb_brc_const_data[table_idx][0],size*sizeof(unsigned int));
+
+ data_tmp = data;
+
+ switch(generic_state->frame_type)
+ {
+ case SLICE_TYPE_I:
+ for(i = 0; i < 52 ; i++)
+ {
+ if(avc_state->old_mode_cost_enable)
+ *data = (unsigned int)gen9_avc_old_intra_mode_cost[i];
+ data += 16;
+ }
+ break;
+ case SLICE_TYPE_P:
+ case SLICE_TYPE_B:
+ for(i = 0; i < 52 ; i++)
+ {
+ if(generic_state->frame_type == SLICE_TYPE_P)
+ {
+ if(avc_state->skip_bias_adjustment_enable)
+ *(data + 3) = (unsigned int)gen9_avc_mv_cost_p_skip_adjustment[i];
+ }
+ if(avc_state->non_ftq_skip_threshold_lut_input_enable)
+ {
+ *(data + 9) = (unsigned int)i965_avc_calc_skip_value(block_based_skip_enable,transform_8x8_mode_flag,avc_state->non_ftq_skip_threshold_lut[i]);
+ }else if(generic_state->frame_type == SLICE_TYPE_P)
+ {
+ *(data + 9) = (unsigned int)gen9_avc_skip_value_p[block_based_skip_enable][transform_8x8_mode_flag][i];
+ }else
+ {
+ *(data + 9) = (unsigned int)gen9_avc_skip_value_b[block_based_skip_enable][transform_8x8_mode_flag][i];
+ }
+
+ if(avc_state->adaptive_intra_scaling_enable)
+ {
+ *(data + 10) = (unsigned int)gen9_avc_adaptive_intra_scaling_factor[i];
+ }else
+ {
+ *(data + 10) = (unsigned int)gen9_avc_intra_scaling_factor[i];
+
+ }
+ data += 16;
+
+ }
+ break;
+ default:
+ assert(0);
+ }
+
+ data = data_tmp;
+ for(i = 0; i < 52 ; i++)
+ {
+ if(avc_state->ftq_skip_threshold_lut_input_enable)
+ {
+ *(data + 6) = (avc_state->ftq_skip_threshold_lut[i] |
+ (avc_state->ftq_skip_threshold_lut[i] <<16) |
+ (avc_state->ftq_skip_threshold_lut[i] <<24) );
+ *(data + 7) = (avc_state->ftq_skip_threshold_lut[i] |
+ (avc_state->ftq_skip_threshold_lut[i] <<8) |
+ (avc_state->ftq_skip_threshold_lut[i] <<16) |
+ (avc_state->ftq_skip_threshold_lut[i] <<24) );
+ }
+
+ if(avc_state->kernel_trellis_enable)
+ {
+ *(data + 11) = (unsigned int)avc_state->lamda_value_lut[i][0];
+ *(data + 12) = (unsigned int)avc_state->lamda_value_lut[i][1];
+
+ }
+ data += 16;
+
+ }
+ i965_unmap_gpe_resource(gpe_resource);
+}
--
2.11.0
More information about the Libva
mailing list