FCCM Sunday (May 11, 2014) Workshop on

Domain Customized Languages (DCLs)

for Custom Computing Machines

Domain Specific Languages (DSLs) are in common use today to increase programmer productivity and code quality. DSLs allows application solutions to be expressed using abstractions that are closer to a problem domain. DSLs take a wide range of forms, from markup languages, modeling languages, specification languages, through tailored programming languages. Familiar DSLs include HTML, Matlab, SQL, SNORT, and YACC grammars.

Custom computing machines can offer higher performance and energy efficiency compared to traditional, general-purpose computing platforms.  The use of field-programmable platforms can avoid the high, fixed manufacturing costs of ASICs. However, most FCCMs (Field-Programmable Custom Computing Machines) are still programmed as if they were ASICs, at the register-transfer level using Verilog or VHDL. This results in long development times and the need for low-level, hardware-centric design expertise.  These two issues have served as a historical barrier for programmers and software engineers as they typically do not possess the requisite hardware design expertise to use these powerful machines. Compounding this problem is the concern that working at the RTL level can be error prone and result in inefficient and non-portable designs.

The objective of this workshop is to explore the possibility that domain-customized languages (DCLs) could provide programmers better access to FCCM’s, allowing them to exploit the same benefits of accessibility, productivity, code quality and design correctness enjoyed by their DSL counterparts. This workshop brings together experts who will share their experience in creating a diverse set of domain-customized languages that target FCCMs. Each expert will first be given 40 minutes to address the following questions:

• What are the application domains their language targets?
• How does their language advance capabilities over traditional HDLs?
• Better accessibility to the FCCM by domain experts?
• Reduced development times?
• Increased quality or design correctness of a solution?
• Better portability, greater optimization, or higher quality mapping
• to FCCM platforms?

In short, we ask the experts to tell a compelling story and illustrate through concrete code examples appropriate for a technical audience why we should use their specific DCL.

Presentations will then be followed by a 60 minute open discussion to explore the broader issues of DCL applicability and acceptability. Specific types of questions might include:

• What are the technical and/or non-technical barriers for DCL approaches to gain wider acceptance?
• How far can DCLs go to address the historical issues of accessibility, portability, and productivity?
• How many different DCLs do we need and are there limits to domain customization? [i.e., in the limit, does every application end up with its own language?]
• What new DCLs are missing?
• What research is needed to enable DCLs to provide the same benefits as DSLs?
• Ecosystem, Interfaces and APIs, IP integration and reuse, Open Source….

3:00pm	Networking – Gordon Brebener, Xilinx
3:40pm	Graphs –Eriko Nurvitadhi, Intel
4:20pm	Break
4:30pm	Model-based FPGA Design Flow – Girish Venkataramani, Mathworks
5:10pm	Dataflow – Rishiyur S. Nikhil, Bluespec
5:50pm	Break
6:00pm	Discussion
7:00pm	FCCM Opening Reception

Organizers: David Andrews and André DeHon