The Experts below are selected from a list of 6432 Experts worldwide ranked by ideXlab platform
Gianni De Fabritiis - One of the best experts on this subject based on the ideXlab platform.
-
Swan: A Tool for porting CUDA programs to OpenCL
Computer Physics Communications, 2011Co-Authors: Matthew J. Harvey, Gianni De FabritiisAbstract:Abstract The use of modern, high-performance graphical processing units (GPUs) for acceleration of scientific computation has been widely reported. The majority of this work has used the CUDA programming model supported exclusively by GPUs manufactured by NVIDIA. An industry standardisation effort has recently produced the OpenCL specification for GPU programming. This offers the benefits of hardware-independence and reduced dependence on Proprietary Tool-chains. Here we describe a source-to-source translation Tool, “Swan” for facilitating the conversion of an existing CUDA code to use the OpenCL model, as a means to aid programmers experienced with CUDA in evaluating OpenCL and alternative hardware. While the performance of equivalent OpenCL and CUDA code on fixed hardware should be comparable, we find that a real-world CUDA application ported to OpenCL exhibits an overall 50% increase in runtime, a reduction in performance attributable to the immaturity of contemporary compilers. The ported application is shown to have platform independence, running on both NVIDIA and AMD GPUs without modification. We conclude that OpenCL is a viable platform for developing portable GPU applications but that the more mature CUDA Tools continue to provide best performance. Program summary Program title: Swan Catalogue identifier: AEIH_v1_0 Program summary URL: http://cpc.cs.qub.ac.uk/summaries/AEIH_v1_0.html Program obtainable from: CPC Program Library, Queen's University, Belfast, N. Ireland Licensing provisions: GNU Public License version 2 No. of lines in distributed program, including test data, etc.: 17 736 No. of bytes in distributed program, including test data, etc.: 131 177 Distribution format: tar.gz Programming language: C Computer: PC Operating system: Linux RAM: 256 Mbytes Classification: 6.5 External routines: NVIDIA CUDA, OpenCL Nature of problem: Graphical Processing Units (GPUs) from NVIDIA are preferentially programed with the Proprietary CUDA programming Toolkit. An alternative programming model promoted as an industry standard, OpenCL, provides similar capabilities to CUDA and is also supported on non-NVIDIA hardware (including multicore ×86 CPUs, AMD GPUs and IBM Cell processors). The adaptation of a program from CUDA to OpenCL is relatively straightforward but laborious. The Swan Tool facilitates this conversion. Solution method: Swan performs a translation of CUDA kernel source code into an OpenCL equivalent. It also generates the C source code for entry point functions, simplifying kernel invocation from the host program. A concise host-side API abstracts the CUDA and OpenCL APIs. A program adapted to use Swan has no dependency on the CUDA compiler for the host-side program. The converted program may be built for either CUDA or OpenCL, with the selection made at compile time. Restrictions: No support for CUDA C++ features Running time: Nominal
Matthew J. Harvey - One of the best experts on this subject based on the ideXlab platform.
-
Swan: A Tool for porting CUDA programs to OpenCL
Computer Physics Communications, 2011Co-Authors: Matthew J. Harvey, Gianni De FabritiisAbstract:Abstract The use of modern, high-performance graphical processing units (GPUs) for acceleration of scientific computation has been widely reported. The majority of this work has used the CUDA programming model supported exclusively by GPUs manufactured by NVIDIA. An industry standardisation effort has recently produced the OpenCL specification for GPU programming. This offers the benefits of hardware-independence and reduced dependence on Proprietary Tool-chains. Here we describe a source-to-source translation Tool, “Swan” for facilitating the conversion of an existing CUDA code to use the OpenCL model, as a means to aid programmers experienced with CUDA in evaluating OpenCL and alternative hardware. While the performance of equivalent OpenCL and CUDA code on fixed hardware should be comparable, we find that a real-world CUDA application ported to OpenCL exhibits an overall 50% increase in runtime, a reduction in performance attributable to the immaturity of contemporary compilers. The ported application is shown to have platform independence, running on both NVIDIA and AMD GPUs without modification. We conclude that OpenCL is a viable platform for developing portable GPU applications but that the more mature CUDA Tools continue to provide best performance. Program summary Program title: Swan Catalogue identifier: AEIH_v1_0 Program summary URL: http://cpc.cs.qub.ac.uk/summaries/AEIH_v1_0.html Program obtainable from: CPC Program Library, Queen's University, Belfast, N. Ireland Licensing provisions: GNU Public License version 2 No. of lines in distributed program, including test data, etc.: 17 736 No. of bytes in distributed program, including test data, etc.: 131 177 Distribution format: tar.gz Programming language: C Computer: PC Operating system: Linux RAM: 256 Mbytes Classification: 6.5 External routines: NVIDIA CUDA, OpenCL Nature of problem: Graphical Processing Units (GPUs) from NVIDIA are preferentially programed with the Proprietary CUDA programming Toolkit. An alternative programming model promoted as an industry standard, OpenCL, provides similar capabilities to CUDA and is also supported on non-NVIDIA hardware (including multicore ×86 CPUs, AMD GPUs and IBM Cell processors). The adaptation of a program from CUDA to OpenCL is relatively straightforward but laborious. The Swan Tool facilitates this conversion. Solution method: Swan performs a translation of CUDA kernel source code into an OpenCL equivalent. It also generates the C source code for entry point functions, simplifying kernel invocation from the host program. A concise host-side API abstracts the CUDA and OpenCL APIs. A program adapted to use Swan has no dependency on the CUDA compiler for the host-side program. The converted program may be built for either CUDA or OpenCL, with the selection made at compile time. Restrictions: No support for CUDA C++ features Running time: Nominal
Harvey, Mendeley M Data) - One of the best experts on this subject based on the ideXlab platform.
-
Swan: A Tool for porting CUDA programs to OpenCL
2019Co-Authors: Harvey, Mendeley M Data)Abstract:This program has been imported from the CPC Program Library held at Queen's University Belfast (1969-2018) Abstract The use of modern, high-performance graphical processing units (GPUs) for acceleration of scientific computation has been widely reported. The majority of this work has used the CUDA programming model supported exclusively by GPUs manufactured by NVIDIA. An industry standardisation effort has recently produced the OpenCL specification for GPU programming. This offers the benefits of hardware-independence and reduced dependence on Proprietary Tool-chains. Here we describe a source-to-source tr... Title of program: Swan Catalogue Id: AEIH_v1_0 Nature of problem Graphical Processing Units (GPUs) from NVIDIA are preferentially programed with the Proprietary CUDA programming Toolkit. An alternative programming model promoted as an industry standard, OpenCL, provides similar capabilities to CUDA and is also supported on non-NIVIDA hardware (including multicore x86 CPUs, AMD GPUs and IBM Cell processors). The adaptation of a program from CUDA to OpenCL is relatively straightforward but laborious. The Swan Tool facilitates this conversion. Versions of this program held in the CPC repository in Mendeley Data AEIH_v1_0; Swan; 10.1016/j.cpc.2010.12.05
Mähne Torsten - One of the best experts on this subject based on the ideXlab platform.
-
Efficient Modelling and Simulation Methodology for the Design of Heterogeneous Mixed-Signal Systems on Chip
Lausanne EPFL, 2011Co-Authors: Mähne TorstenAbstract:Systems on Chip (SoCs) and Systems in Package (SiPs) are key parts of a continuously broadening range of products, from chip cards and mobile phones to cars. Besides an increasing amount of digital hardware and software for data processing and storage, they integrate more and more analogue/RF circuits, sensors, and actuators to interact with their (analogue) environment. This trend towards more complex and heterogeneous systems with more intertwined functionalities is made possible by the continuous advances in the manufacturing technologies and pushed by market demand for new products and product variants. Therefore, the reuse and retargeting of existing component designs becomes more and more important. However, all these factors make the design process increasingly complex and multidisciplinary. Nowadays, the design of the individual components is usually well understood and optimised through the usage of a diversity of CAD/EDA Tools, design languages, and data formats. These are based on applying specific modelling/abstraction concepts, description formalisms (also called Models of Computation (MoCs)) and analysis/simulation methods. The designer has to bridge the gaps between Tools and methodologies using manual conversion of models and Proprietary Tool couplings/integrations, which is error-prone and time-consuming. A common design methodology and platform to manage, exchange, and collaboratively develop models of different formats and of different levels of abstraction is missing. The verification of the overall system is a big problem, as it requires the availability of compatible models for each component at the right level of abstraction to achieve satisfying results with respect to the system functionality and test coverage, but at the same time acceptable simulation performance in terms of accuracy and speed. Thus, the big challenge is the parallel integration of these very different part design processes. Therefore, the designers need a common design and simulation platform to create and refine an executable specification of the overall system (a virtual prototype) on a high level of abstraction, which supports different MoCs. This makes possible the exploration of different architecture options, estimation of the performance, validation of re-used parts, verification of the interfaces between heterogeneous components and interoperability with other systems as well as the assessment of the impacts of the future working environment and the manufacturing technologies used to realise the system. For embedded Analogue and Mixed-Signal (AMS) systems, the C++-based SystemC with its AMS extensions, to which recent standardisation the author contributed, is currently establishing itself as such a platform. This thesis describes the author's contribution to solve the modelling and simulation challenges mentioned above in three thematic phases. In the first phase, the prototype of a web-based platform to collect models from different domains and levels of abstraction together with their associated structural and semantical meta information has been developed and is called ModelLib. This work included the implementation of a hierarchical access control mechanism, which is able to protect the Intellectual Property (IP) constituted by the model at different levels of detail. The use cases developed for this Tool show how it can support the AMS SoC design process by fostering the reuse and collaborative development of models for tasks like architecture exploration, system validation, and creation of more and more elaborated models of the system. The experiences from the ModelLib development delivered insight into which aspects need to be especially addressed throughout the development of models to make them reusable: mainly flexibility, documentation, and validation. This was the starting point for the development of an efficient modelling methodology for the top-down design and bottom-up verification of RF Systems based on the systematic usage of behavioural models in the second phase. One outcome is the developed library of well documented, parameterisable, and pin-accurate VHDL-AMS models of typical analogue/digital/RF components of a transceiver. The models offer the designer two sets of parameters: one based on the performance specifications and one based on the device parameters back-annotated from the transistor-level implementation. The abstraction level used for the description of the respective analogue/digital/RF component behaviour has been chosen to achieve a good trade-off between accuracy, fidelity, and simulation performance. The pin-accurate model interfaces facilitate the integration of transistor-level models for the validation of the behavioural models or the verification of a component implementation in the system context. These properties make the models suitable for different design tasks such as architecture exploration or overall system validation. This is demonstrated on a model of a binary Frequency-Shift Keying (FSK) transmitter parameterised to meet very different target specifications. This project showed also the limits in terms of abstraction and simulation performance of the "classical" AMS Hardware Description Languages (HDLs). Therefore, the third and last phase was dedicated to further raise the abstraction level for the description of complex and heterogeneous AMS SoCs and thus enable their efficient simulation using different synchronised MoCs. This work uses the C++-based simulation framework SystemC with its AMS extensions. New modelling capabilities going beyond the standardised SystemC AMS extensions have been introduced to describe energy conserving multi-domain systems in a formal and consistent way at a high level of abstraction. To this end, all constants, variables, and parameters of the system model, which represent a physical quantity, can now declare their dimension and associated system of units as an intrinsic part of their data type. Assignments to them need to contain besides the value also the correct measurement unit. This allows a much more precise but still compact definition of the models' interfaces and equations. Thus, the C++ compiler can check the correct assembly of the components and the coherency of the equations by means of dimensional analysis. The implementation is based on the Boost.Units library, which employs template metaprogramming techniques. A dedicated filter for the measurement units data types has been implemented to simplify the compiler messages and thus facilitate the localisation of unit errors. To ensure the reusability of models despite precisely defined interfaces, their interfaces and behaviours need to be parametrisable in a well-defined manner. The enabling implementation techniques for this have been demonstrated with the developed library of generic block diagram component models for the Timed Data Flow (TDF) MoC of the SystemC AMS extensions. These techniques are also the key to integrate a new MoC based on the bond graph formalism into the SystemC AMS extensions. Bond graphs facilitate the unified description of the energy conserving parts of heterogeneous systems with the help of a small set of modelling primitives parametrisable to the physical domain. The resulting models have a simulation performance comparable to an equivalent signal flow model
Radermacher Reinhard - One of the best experts on this subject based on the ideXlab platform.
-
Comparison of Two Object-Oriented Modeling Environments for the Dynamic Simulations of a Residential Heat Pump
'Purdue University (bepress)', 2018Co-Authors: Bhanot Viren, Dhumane Rohit, Cioncolini Andrea, Petagna Paolo, Ling Jiazhen, Aute, Vikrant Chandramohan, Radermacher ReinhardAbstract:Object-oriented physical-modelling platforms greatly facilitate the task of the modelling engineer by abstracting away a lot of the complexity associated with sorting the governing equations and also the nuances of the numerical methods used for solving the differential-algebraic equations (DAEs). For this reason, they have been steadily gaining in popularity in the field of thermofluid simulations. In this study, we compare two platforms of this type: Dymola and EcosimPro. Dymola is a physical modelling environment originally developed at Lund University and now being developed by Dassault Systèmes, and is a commercially available implementation of the open-source physical modelling language Modelica. EcosimPro is a Proprietary Tool developed by Empresarios Agrupados A.I.E originally for the European Space Agency and now sold to the general public. Both platforms utilise object-oriented modelling paradigms such as multiple inheritance, encapsulation (of behaviour within classes), abstraction (hiding model complexity from the user) and acausal equation handling (equations may be written in any order). We use these platforms to conduct a realistic exercise of modelling and simulating a relatively complex residential heat pump system in both heating and cooling modes and comparing the results against measured data. Component libraries have been prepared in both the platforms for modelling system components. Two-phase flow has been accounted for using slip-ratio based void fraction correlations. In general, the component models have been kept as similar as possible between the two platforms. The heat pump under investigation is a residential, 3-ton unit with a scroll compressor. The cooling mode uses a thermostatic expansion valve (TXV) as the expansion device while the heating mode uses a short-tube orifice. A reversing valve controls the flow direction. The heat pump has been tested under both heating and cooling modes as per ASHRAE’s Standard 116-2010 cyclic test conditions. The measured values have been compared against simulations results from both platforms. The refrigerant pressures and temperatures and the heat exchanger air outlet temperatures are compared. The indoor unit air-side capacity and the compressor power consumption integrated over the on-period are also compared. Additionally, the Seasonal Energy Efficiency Ratio (SEER), the Cooling Load Factor (CLF) and the Cyclic Degradation Coefficient (Cd) are compared which help quantify the performance of the heat pump. Finally, qualitative comparisons of the transients associated with the refrigerant charge migration after shutdown have been made, as this migration is responsible for cycling losses associated with dynamic heat pump operation. The two platforms prove to be similarly capable at simulating an advanced cycle. Both platforms can predict the pressure and temperature transients during the on-off cycling of the heat pump, as well as the performance parameters such as accumulated capacities and the SEER rating. Finally, both models predict the simulated charge to be within 80% of the actual charge, which enables a more realistic depiction of system transients
