Abstract
High-Level Synthesis (HLS) is an automated design process that
seeks to improve productivity over traditional design methods by
increasing design abstraction from register transfer level (RTL)
to behavioural level. Various commercial HLS tools have been available
on the market since the 1990s, but only recently they have started to
gain adoption across industry and academia. The slow adoption rate has
mainly stemmed from lower quality of results (QoR) than obtained with conventional hardware description languages (HDLs). However, the latest HLS tool generations have substantially narrowed the QoR gap.
This thesis studies the feasibility of HLS in video codec development. It introduces several HLS implementations for High Efficiency Video Coding (HEVC) , that is the key enabling technology for numerous modern media applications. HEVC doubles the coding efficiency over its predecessor Advanced Video Coding (AVC) standard for the same subjective visual quality, but typically at the cost of considerably higher computational complexity. Therefore, real-time HEVC calls for automated design methodologies that can be used to minimize the HW implementation and verification effort.
This thesis proposes to use HLS throughout the whole encoder design process. From data-intensive coding tools, like intra prediction and discrete transforms, to more control-oriented tools, such as entropy coding. The C source code of the open-source Kvazaar HEVC encoder serves as a design entry point for the HLS flow, and it is also utilized in design verification. The performance results are gathered with and reported for field programmable gate array (FPGA) .
The main contribution of this thesis is an HEVC intra encoder prototype that is built on a Nokia AirFrame Cloud Server equipped with 2.4 GHz dual 14-core Intel Xeon processors and two Intel Arria 10 GX FPGA Development Kits, that can be connected to the server via peripheral component interconnect express (PCIe) generation 3 or 40 Gigabit Ethernet. The proof-of-concept system achieves real-time.
4K coding speed up to 120 fps, which can be further scaled up by adding practically any number of network-connected FPGA cards.
Overcoming the complexity of HEVC and customizing its rich features for a real-time HEVC encoder implementation on hardware is not a trivial task, as hardware development has traditionally turned out to be very time-consuming. This thesis shows that HLS is able to boost the development time, provide previously unseen design scalability, and still result in competitive performance and QoR over state-of-the-art hardware implementations.
This thesis studies the feasibility of HLS in video codec development. It introduces several HLS implementations for High Efficiency Video Coding (HEVC) , that is the key enabling technology for numerous modern media applications. HEVC doubles the coding efficiency over its predecessor Advanced Video Coding (AVC) standard for the same subjective visual quality, but typically at the cost of considerably higher computational complexity. Therefore, real-time HEVC calls for automated design methodologies that can be used to minimize the HW implementation and verification effort.
This thesis proposes to use HLS throughout the whole encoder design process. From data-intensive coding tools, like intra prediction and discrete transforms, to more control-oriented tools, such as entropy coding. The C source code of the open-source Kvazaar HEVC encoder serves as a design entry point for the HLS flow, and it is also utilized in design verification. The performance results are gathered with and reported for field programmable gate array (FPGA) .
The main contribution of this thesis is an HEVC intra encoder prototype that is built on a Nokia AirFrame Cloud Server equipped with 2.4 GHz dual 14-core Intel Xeon processors and two Intel Arria 10 GX FPGA Development Kits, that can be connected to the server via peripheral component interconnect express (PCIe) generation 3 or 40 Gigabit Ethernet. The proof-of-concept system achieves real-time.
4K coding speed up to 120 fps, which can be further scaled up by adding practically any number of network-connected FPGA cards.
Overcoming the complexity of HEVC and customizing its rich features for a real-time HEVC encoder implementation on hardware is not a trivial task, as hardware development has traditionally turned out to be very time-consuming. This thesis shows that HLS is able to boost the development time, provide previously unseen design scalability, and still result in competitive performance and QoR over state-of-the-art hardware implementations.
Original language | English |
---|---|
Place of Publication | Tampere |
Publisher | Tampere University |
ISBN (Electronic) | 978-952-03-2776-7 |
ISBN (Print) | 978-952-03-2775-0 |
Publication status | Published - 2023 |
Publication type | G5 Doctoral dissertation (articles) |
Publication series
Name | Tampere University Dissertations - Tampereen yliopiston väitöskirjat |
---|---|
Volume | 751 |
ISSN (Print) | 2489-9860 |
ISSN (Electronic) | 2490-0028 |