Skip to content

Latest commit

History

54 Commits

Folders and files

NameName
Last commit message
Last commit date

Repository files navigation

AVS3-RDOQ

Introduction

Pending...

Project Structure

The tree map of this hardware architecture is shown as below:

  • result #test vectors
  • RTL #all RTL code
    • sims_1 #simulation files
    • sources_1 #source code
    • constrs_1 #timing constraint file

Tips

The simulation file in this project uses relative paths to read the test vectors from the result file. When establishing the simulation project, pay attention to whether the path to the read file is placed correctly.

Version Record

V0.0 ,单尺寸流水
时间备注LUTFFRAM(bits)DSPFmax(MHz)
2023.4.1单尺寸RDOQ流水结构89854349121039233.1
pre_quant3146031855032177
ocd101924836430038233.1
lnpd3299343500189.25
时间备注LUTFFRAM(bits)DSPFmax(MHz)
2023.4.3单尺寸RDOQ流水结构413446493090400
pre_quant40216553400188.2
ocd25483640527040047.58
lnpd2767318700189.25
时间备注LUTFFRAM(bits)DSPFmax(MHz)
2023.4.6单尺寸RDOQ流水结构108330233640368
pre_quant927816207032
ocd94603190850288110
lnpd3143258400
时间备注LUTFFRAM(bits)DSPFmax(MHz)
2023.4.7单尺寸RDOQ流水结构93994301110320109.72
pre_quant948111692032350.2
ocd82127144290288110
lnpd2371395000
v0.1 ,单尺寸流水

1.ocd优化, 优化计算式,提前计算err值

2.pre_quant优化,减少乘法器位宽

3.Reconfirm the bit width of each variable.

4.Reduce the array dCurrCost_tem by half.

5.Reduce the array rdoq_est_lastby half, [2 6 12 2]->[1 6 12 2].

时间备注LUTFFRAM(bits)DSPFmax(MHz)
2023.4.10单尺寸RDOQ流水结构76846246590352109.72(s)
pre_quant64469225032350.2(v)
ocd6739712967032064.11(v)
lnpd2989242700189.25(v)

1.Replace the multiplication operation with addition and shift operations in "temp_coef_abs_d2" for values between 2 and 9.

2.Reduce the bit width of "i64Delta" to reduce the number of DSPs used in the calculation of "err" by half.

时间备注LUTFFRAM(bits)DSPFmax(MHz)
2023.4.12单尺寸RDOQ流水结构79711242590224126.85(s)
pre_quant63749879032350.2(v)
ocd69714147570192
lnpd2985242700189.25(v)
v1.0 ,多并行流水
时间备注LUTFFRAM(bits)DSPFmax(MHz)
2023.4.16多并行RDOQ流水结构78330219650224113.57(s)
pre_quant65237349032335.6(s)
ocd76350158570192113.6(s)
lnpd2854253800394.1(s)
v1.1,多并行流水

Delete unnecessary delay operations in the PQ module because there are many unused signals passing through this module. (The deleted signal was provided by the testbench file to the next stage OCD.)

时间备注LUTFFRAM(bits)DSPFmax(MHz)
2023.4.21多并行RDOQ流水结构72017234400224116.27(s)
pre_quant61153289032335.6(s)
ocd63559166440192113.6(s)
lnpd2338347600394.1(s)

About

This work presents the RTL design from our paper published in the journal TCSVT: 'Parallelized RDOQ Algorithm and Fully Pipelined Hardware Architecture for AVS3 Video Coding.'

Topics

Resources

Stars

2 stars

Watchers

1 watching

Forks

Releases

Packages

Used by

Contributors

Languages