specialized virtual configurable arrays dominique lavenier - frederic raimbault irisa rennes, france...
DESCRIPTION
Overview Introduction –exemplified from F. Raimbault talk Virtual Configurable Arrays –implementation - Advantages / Disadvantages Specialized Virtual Configurable Arrays –2 examples : Genome / hyperspectral images ConclusionTRANSCRIPT
Specialized Virtual Configurable Arrays
Dominique Lavenier - Frederic Raimbault
IRISARennes, France
UBSVannes, France
SVCASpecialized Virtual Configurable Arrays
• Warning– Just ideas - no work (yet) performed
• The talk mainly aims to– get feedback - positive or negative !– open discussion / collaborations ?
Overview• Introduction
– exemplified from F. Raimbault talk• Virtual Configurable Arrays
– implementation - Advantages / Disadvantages
• Specialized Virtual Configurable Arrays– 2 examples : Genome / hyperspectral images
• Conclusion
FPGA support for Java
PCHardware
JAVAmachine
network
hardware support:any reconfigurable boards
Boards are different• architecture• FPGA family• power computation
We want to define a hardware support• independent of the FPGA boards• allowing fast implementation
Architecture
PE PE PECTRL
Application dependant• Nb of PEs
• PE functionality
Java Hardware Support
• We want an hardware support with the following features:– platform independent
• all FPGA boards can be targeted– fast implementation
• depending on the application and the available resources, an architecture must be synthesize in a very short time
Challenge
• The hardware support must provide:– A platform independent hardware
– A fast design implementation
Virtual Configurable Array
Specialization
Virtual Configurable ArrayIntroduction
Fixe implementation(Applicationindependent)
Applicationdependentimplementation
VirtualConfigurableArray
Virtual Configurable Array Implementation (1)
• Virtual CLBs– one virtual CLB is made of several physical CLBs
Virtual Configurable Array Implementation (2)
• Routing– physical CLBs are used as switches
Virtual Configurable Arrays
• Advantages– applications are portables– common design tools - open architecture
• Disadvantages– less resources / lower speed (how much ?)– no concept evolution
• still the same problems for programming, routing, ...
Virtual Configurable Arrays
• Platform independent
• Fast implementation
YES
NO
Specialized Virtual Configurable ArraysIntroduction
SpecializationofVirtual Configurable Arrays
CLB functionality
CLB interconnection
2 Examples
• Genome Computation
• Hyperspectral image processing
Genome Computation
• Data – DNA or Protein sequences– large databases
• Computation– data retrieval, classification, ...
– mostly based on sequence comparison
– time consuming but highly parallel
Genome Computation
• Needs:– high computation power– rapid test of new algorithms
• Features:– integer arithmetic
• 8,12,16 bits - no multiplication– efficient parallelization on linear arrays
Specialized Virtual Configurable Array for Genome Computation
CLB:• N-bit operators
Regular Routing• N-bit wires
VCLB
VC
LB i
CM CMCM
n-bit operator(32 operations)
5 bits7 bits7 bits
from i+3from i+2from i+1to i+3to i+2to i+1
to i-1to i-2to i-3
from i-1from i-2from i-3
routing switch
configurationmemories
Hyperspectral images processing
• Data– 3D cube– one image = qq 100 Mbytes
• Computation– compression, segmentation, …– very time consuming, but high level of
parallelsim
A few hundredspectrum
HyperSpectral Image Processing
• Needs:– high computation power– rapid test of new algorithms
• Features:– integer arithmetic– efficient parallelization on 2D arrays
Specialized Virtual Configurable Array for Hyperspectral Image Processing
MemoryVCLBRouting
Specialized Virtual Configurable ArrayConclusion
SVCA
FPGA
Architecture
Specialized Virtual Configurable ArraysConclusion
• One SVCA class of algorithms• Advantages
– platform independent - fast programming• Disadvantages
– small array - slow