A recent development in multicore technology has enabled development of hundreds or thousands core processor. However, on such multicore processor, an efficient hardware cache coherence scheme will become very complex and expensive to develop. This paper proposes a parallelizing compiler directed software coherence scheme for shared memory multicore systems without hardware cache coherence control. The general idea of the proposed method is that an automatic parallelizing compiler analyzes the control dependency and data dependency among coarse grain task in the program. Then based on the obtained information, task parallelization, false sharing detection and data restructuration to prevent false sharing are performed. Next the compiler inserts cache control code to handle stale data problem. The proposed method is built on OSCAR automatic parallelizing compiler and evaluated on Renesas RP2 with 8 SH-4A cores processor. The hardware cache coherence scheme on the RP2 processor is only available for up to 4 cores and the hardware cache coherence can be completely turned off for non-coherence cache mode. Performance evaluation is performed using 10 benchmark program from SPEC2000, SPEC2006, NAS Parallel Benchmark (NPB) and Mediabench II. The proposed method performs as good as or better than hardware cache coherence scheme. For example, 4 cores with the hardware coherence mechanism gave us speed up of 2.52 times against 1 core for SPEC2000 'equake', 2.9 times for SPEC2006 'lbm', 3.34 times for NPB 'cg', and 3.17 times for MediaBench II MPEG2 Encoder. The proposed software cache coherence control gave us 2.63 times for 4 cores and 4.37 for 8 cores for 'equake', 3.28 times for 4 cores and 4.76 times for 8 cores for lbm, 3.71 times for 4 cores and 4.92 times for 8 cores for 'MPEG2 Encoder'.

Embedded multicore processors for hard real-time applications like automobile engine control require the usage of local memory on each processor core to precisely meet the real-time deadline constraints, since cache memory cannot satisfy the deadline requirements due to cache misses. To utilize local memory, programmers or compilers need to explicitly manage data movement and data replacement for local memory considering the limited size. However, such management is extremely difficult and time consuming for programmers. This paper proposes an automatic local memory management method by compilers through (i) multi-dimensional data decomposition techniques to fit working sets onto limited size local memory (ii) suitable block management structures, called Adjustable Blocks, to create application specific fixed size data transfer blocks (iii) multi-dimensional templates to preserve the original multi-dimensional representations of the decomposed multi-dimensional data that are mapped onto one-dimensional Adjustable Blocks (iv) block replacement policies from liveness analysis of the decomposed data, and (v) code size reduction schemes to generate shorter codes. The proposed local memory management method is implemented on the OSCAR multi-grain and multi-platform compiler and evaluated on the Renesas RP2 8 core embedded homogeneous multicore processor equipped with local and shared memory. Evaluations on 5 programs including multimedia and scientific applications show promising results. For instance, speedups on 8 cores compared to single core execution using off-chip shared memory on an AAC encoder program, a MPEG2 encoder program, Tomcatv, and Swim are improved from 7.14 to 20.12, 1.97 to 7.59, 5.73 to 7.38, and 7.40 to 11.30, respectively, when using local memory with the proposed method. These evaluations indicate the usefulness and the validity of the proposed local memory management method on real embedded multicore processors.

One of the applications as a source of big data, there is a sensor network for the environmental monitoring that is designed to detect the deterioration of the infrastructure, erosion control and so on. The specific targets are bridges, buildings, slopes and embankments due to the natural disasters or aging. Basic requirement of this monitoring system is to collect data over a long period of time from a large number of nodes that installed in a wide area. However, in order to apply a wireless sensor network (WSN), using wireless communication and energy harvesting, there are not many cases in the actual monitoring system design. Because of the system must satisfy various conditions measurement location and time specified by the civil engineering communication quality and topology obtained from the network technology the electrical engineering to solve the balance of weather environment and power consumption that depends on the above-mentioned conditions. We propose the whole WSN design methodology especially for the electrical architecture that is affected by the network behavior and the environmental disturbance. It is characterized by determining recursively mutual trade-off of a wireless simulation and a power architecture simulation of the node devices. Furthermore, the system allows the redundancy of the design. In addition, we deployed the actual slope monitoring WSN that is designed by the proposed method to the snow-covered area. A conventional similar monitoring WSN, with 7 Ah Li-battery, it worked only 129 days in a mild climate area. On the other hand, our proposed system, deployed in the heavy snow area has been working more than 6 months (still working) with 3.2 Ah batteries. Finally, it made a contribution to the civil engineering succeeded in the real time observation of the groundwater level displacement at the time of melting snow in the spring season.

Parallelizing compilers employing powerful compiler optimizations are essential tools to fully exploit performance from today's computer systems. These optimizations are supported by both highly sophisticated program analysis techniques and aggressive program restructuring techniques. However, the compilation time for such powerful compilers becomes larger and larger for real commercial application due to these strong program analysis techniques. In this paper, we propose a compilation time reduction technique for parallelizing compilers. The basic idea of the proposed technique is based on an observation that parallelizing compilers apply multiple program analysis passes and restructuring passes to a source program but all program analysis passes do not have to be applied to the whole source program. Thus, there is an opportunity for compilation time reduction by removing redundant program analysis. We describe the removing redundant program analysis techniques considering the inter-procedural propagation of analysis update information in this paper. We implement the proposed technique into OSCAR automatically multigrain parallelizing compiler. We then evaluate the proposed technique by using three proprietary large scale programs. The proposed technique can remove 37.7% of program analysis time on average for basic analysis includes def-use analysis and dependence calculation, and 51.7% for pointer analysis, respectively.

The emergence of multi-core processors in smart devices promises higher performance and low power consumption. The parallelization of applications enables us to improve their performance. However, simultaneously utilizing many cores would drastically drain the device battery life. This paper shows a demonstration system of realtime video processing combined with power reduction controlled by the OSCAR automatic parallelization compiler on ODROID-X2, an open Android development platform based on Samsung Exynos4412 Prime with 4 ARM Cortext- A9 cores. In this paper, we exploited the DVFS framework, core partitioning, and profiling technique and OSCAR parallelization - power control algorithm to reduce the total consumption in a real-time video application. The demonstration results show that it can cut power consumption by 42.8% for MPEG-2 Decoder application and 59.8% for Optical Flow application by using 3 cores in both applications.

Model-based design is a very popular software development method for developing a wide variety of embedded applications such as automotive systems, aircraft systems, and medical systems. Model-based design tools like MATLAB/Simulink typically allow engineers to graphically build models consisting of connected blocks for the purpose of reducing development time. These tools also support automatic C code generation from models with a special tool such as Embedded Coder to map models onto various kinds of embedded CPUs. Since embedded systems require real-time processing, the use of multi-core CPUs poses more opportunities for accelerating program execution to satisfy the real-time constraints. While prior approaches exploit parallelism among blocks by inspecting MATLAB/Simulink models, this may lose an opportunity for fully exploiting parallelism of the whole program because models potentially have parallelism within a block. To unlock this limitation, this paper presents an automatic parallelization technique for auto-generated C code developed by MATLAB/Simulink with Embedded Coder. Specifically, this work (1) exploits multi-level parallelism including inter-block and intra-block parallelism by analyzing the auto-generated C code, and (2) performs static scheduling to reduce dynamic overheads as much as possible. Also, this paper proposes an automatic profiling framework for the auto-generated code for enhancing static scheduling, which leads to improving the performance of MATLAB/Simulink applications. Performance evaluation shows 4.21 times speedup with six processor cores on Intel Xeon X5670 and 3.38 times speedup with four processor cores on ARM Cortex-A15 compared with uniprocessor execution for a road tracking application.

This paper proposes coarse grain task parallelization for a earthquake simulation program using Finite Difference Method to solve the wave equations in 3-D heterogeneous structure or the Ground Motion Simulator (GMS) on various cc-NUMA servers using IBM, Intel and Fujitsu multicore processors. The GMS has been developed by the National Research Institute for Earth Science and Disaster Prevention (NIED) in Japan. Earthquake wave propagation simulations are important numerical applications to save lives through damage predictions of residential areas by earthquakes. Parallel processing with strong scaling has been required to precisely calculate the simulations quickly. The proposed method uses the OSCAR compiler for exploiting coarse grain task parallelism efficiently to get scalable speed-ups with strong scaling. The OSCAR compiler can analyze data dependence and control dependence among coarse grain tasks, such as subroutines, loops and basic blocks. Moreover, locality optimizations considering the boundary calculations of FDM and a new static scheduler that enables more efficient task schedulings on cc-NUMA servers are presented. The performance evaluation shows 110 times speed-up using 128 cores against the sequential execution on a POWER7 based 128 cores cc-NUMA server Hitachi SR16000 VM1, 37.2 times speed-up using 64 cores against the sequential execution on a Xeon E7-8830 based 64 cores cc-NUMA server BS2000, 19.8 times speed-up using 32 cores against the sequential execution on a Xeon X7560 based 32 cores cc-NUMA server HA8000/RS440, 99.3 times speed-up using 128 cores against the sequential execution on a SPARC64 VII based 256 cores cc-NUMA server Fujitsu M9000, 9.42 times speed-up using 12 cores against the sequential execution on a POWER8 based 12 cores cc-NUMA server Power System S812L.

Network intrusion detection system (NIDS) is becoming an important element even in embedded systems as well as in data centers since embedded computers have been increasingly exposed to the Internet. The demand for power budget of these embedded systems is a critical issue in addition to that for performance. In this paper, we propose a technique to minimize power consumption in the NIDS by 2-step power scheduling with the adaptive control interval. In addition, we also propose a CPU-core controlling algorithm so that our scheduling technique can preserve the performance for other applications and NIDS assuming the cases of multiplexing NIDS and them simultaneously on the same device such as a home server or a mobile platform. We implement our 2-step algorithm into Suricata, which is a popular NIDS, as well as a 1-step algorithm with the adaptive interval, and a simple fixed-interval algorithm for evaluations. Experimental results show that our 2-step scheduling with both the adaptive and the fixed 30-millisecond interval achieve 75% power saving comparing with the Ondemand governor and 87% comparing with the Performance governor in Linux, respectively, without affecting their performance capability on four ARM Cortex-A15 cores at the network traffic of 1,000 packets/seconds. In contrast, when the network traffic reaches to 17,000 packets/seconds, our 2-step scheduling and the Ondemand as well as the Performance governor can maintain the packet processing capacity while the fixed 30-milliseconds interval processes only 50% packets with two and three cores, and about 80% packets on four cores.

Architecture simulators play an important role in exploring frontiers in the early stages of the architecture design. However, the execution time of simulators increases with an increase the number of cores. The sampling simulation technique that was originally proposed to simulate single-core processors is a promising approach to reduce simulation time. Two main hurdles for multi/many-core are preparing sampling points and thread skewing at functional simulation time. This paper proposes a very simple and low-error sampling-based acceleration technique for multi/many-core simulators. For a parallelized application, an iteration of a large loop including a parallelizable program part, is defined as a sampling unit. We apply X-means method to a profile result of the collection of iterations derived from a real machine to form clusters of those iterations. Multiple iterations are exploited as sampling points from these clusters. We execute the simulation along the sampling points and calculate the number of total execution cycles. Results from a 16-core simulation show that our proposed simulation technique gives us a maximum of 443x speedup with a 0.52% error and 218x speedup with 1.50% error on an average.

Investigation of the runtime behavior is one of the most important processes for performance tuning on a computer system. Profiling tools have been widely used to detect hot-spots in a program. In addition to them, tracing tools produce valuable information especially from parallelized programs, such as thread scheduling, barrier synchronizations, context switching, thread migration, and jitter by interrupts. Users can optimize a runtime system and hardware configuration in addition to a program itself by utilizing the attained information. However, existing tools provide information per process or per function. Finer information like task-or loop-granularity should be required to understand the program behavior more precisely. This paper has proposed a tracing tool, Annotatable Systrace, to investigate runtime execution behavior of a parallelized program based on an extended Linux ftrace. The Annotatable Systrace can add arbitrary annotations in a trace of a target program. The proposed tool exploits traces from 183.equake, 179.art, and mpeg2enc on Intel Xeon X7560 and ARMv7 as an evaluation. The evaluation shows that the tool enables us to observe load imbalance along with the program execution. It can also generate a trace with the inserted annotations even on a 32-core machine. The overhead of one annotation on Intel Xeon is 1.07 us and the one on ARMv7 is 4.44 us, respectively.

Recently, more safety, comfort and environmental feasibility are required for the automobile. Accordingly, control systems need performance enhancement on microprocessors for real-time software which realize that. However, the improvement of clock frequency has been limited by power consumption and the performance of a single-core processor which controls power has reached the limits. For these factors, multi-core processors will be used for automotive control system. Recently Model-based Design by MATLAB and Simulink has been used for developing automobile systems because of elimination time of development and improvement of reliability. However, auto-generated-code from MATLAB and Simulink has been functioned on only single core processor so far. This paper proposes a parallelization method of engine control C codes for a multi-core processor generated from MATLAB and Simulink using Embedded Coder. The engine control C code which composed of many conditional branches and arithmetic assignment statements and are difficult to parallelize have been parallelized automatically using OSCAR automatic parallel compiler. In this result, it is succeeded to attain performance improvement on RP2 and V850E2R. Maximum 1.9x speedup on two cores and 3.76x speedup on four cores are attained.

In recent years, smart devices are transitioning from single core processors to multicore processors to satisfy the growing demands of higher performance and lower power consumption. However, power consumption of multicore processors is increasing, as usage of smart devices become more intense. This situation is one of the most fundamental and important obstacle that the mobile device industries face, to extend the battery life of smart devices. This paper evaluates the power reduction control by the OSCAR Automatic Parallelizing Compiler on an Android platform with the newly developed precise power measurement environment on the ODROID-X2, a development platform with the Samsung Exynos4412 Prime, which consists of 4 ARM Cortex-A9 cores. The OSCAR Compiler enables automatic exploitation of multigrain parallelism within a sequential program, and automatically generates a parallelized code with the OSCAR Multi-Platform API power reduction directives for the purpose of DVFS (Dynamic Voltage and Frequency Scaling), clock gating, and power gating. The paper also introduces a newly developed micro second order pseudo clock gating method to reduce power consumption using WFI (Wait For Interrupt). By inserting GPIO (General Purpose Input Output) control functions into programs, signals appear on the power waveform indicating the point of where the GPIO control was inserted and provides a precise power measurement of the specified program area. The results of the power evaluation for real-time Mpeg2 Decoder show 86.7% power reduction, namely from 2.79[W] to 0.37[W] and for real-time Optical Flow show 86.5% power reduction, namely from 2.23[W] to 0.36[W] on 3 core execution.

This paper evaluates an automatic power reduction scheme of OSCAR automatic parallelizing compiler having power reduction control capability when multiple media applications parallelized by the OSCAR compiler are executed simultaneously on RP2, a 8-core multicore processor developed by Renesas Electronics, Hitachi, and Waseda University. OSCAR compiler enables the hierarchical multigrain parallel processing and power reduction control using DVFS (Dynamic Voltage and Frequency Scaling), clock gating and power gating for each processor core using the OSCAR multi-platform API. The RP2 has eight SH4A processor cores, each of which has power control mechanisms such as DVFS, clock gating and power gating. First, multiple applications with relatively light computational load are executed simultaneously on the RP2. The average power consumption of power controlled eight AAC encoder programs, each of which was executed on one processor, was reduced by 47%, (to 1.01W), against one AAC encoder execution on one processor (from 1.89W) without power control. Second, when multiple intermediate computational load applications are executed, the power consumptions of an AAC encoder executed on four processors with the power reduction control was reduced by 57% (to 0.84W) against an AAC encoder execution on one processor (from 1.95W). Power consumptions of one MPEG2 decoder on four processors with power reduction control was reduced by 49% (to 1.01W) against one MPEG2 decoder execution on one processor (from 1.99W). Finally, when a combination of a high computational load application program and an intermediate computational load application program are executed simultaneously, the consumed power reduced by 21% by using twice number of cores for each application. This paper confirmed parallel processing and power reduction by OSCAR compiler are efficient for multiple application executions. In execution of multiple light computational load applications, power consumption increases only 12% for one application. Parallel processing being applied to intermediate computational load applications, power consumption of executing one application on one processor core (1.49W) is almost same power consumption of two applications on eight processor cores (1.46W). © 2013 Springer-Verlag.

Currently few mobile applications exploit the power- and performance capabilities of multi-core architectures. As the number of cores increases, the challenges become more pressing. We picked three challenges: application parallelization, performance-predictability/portability and power control for mobile devices. We tackled the challenges with our auto-parallelizing compiler and operating system enhancements.

In the age of dark silicon on-chip power control is a necessity. Upcoming and state of the art embedded- and cloud computer system-on-chips (SoCs) already provide interfaces for fine grained power control. Sometimes both: core- and interconnect-voltage and frequency can be scaled for example. To further reduce power consumption SoCs often have specialized accelerators. Due to the rising specialization of hard- and software general purpose operating systems require changes to exploit the power saving opportunities provided by the hardware. However, they lack detailed hardware- and application-level-information. Application-level power control in turn is still very uncommon and difficult to realize. Now a days vendors of mobile devices are forced to tweak and patch system-level software to enhance the power efficiency of each individual product. This manual process is time consuming and must be re-iterated for each new product. In this paper we explore the opportunities and challenges of automatic application- level power control using compilers. © 2013 IEEE.

Video Games have been a very popular form of digital entertainment in recent years. They have been delivered in state of the art technologies that include multi-core processors that are known to be the leading contributor in enhancing the performance of computer applications. Since parallel programming is a difficult technology to implement, that field in Video Games is still rich with areas for advancements. This paper investigates performance enhancement in Video Games when using parallelizing compilers and the difficulties involved in achieving that. This experiment conducts several stages in attempting to parallelize a well-renowned sequentially written Video Game called ioquake3. First, the Game is profiled for discovering bottlenecks, then examined by hand on how much parallelism could be extracted from those bottlenecks, and what sort of hazards exist in delivering a parallel-friendly version of ioquake3. Then, the Game code is rewritten into a hazard-free version while also modified to comply with the Parallelizable-C rules, which crucially aid parallelizing compilers in extracting parallelism. Next, the program is compiled using a parallelizing compiler called OSCAR (Optimally Scheduled Advanced Multiprocessor) to produce a parallel version of ioquake3. Finally, the performance of the newly produced parallel version of ioquake3 on a Multi-core platform is analyzed.
The following is found: (1) the parallelized game by the compiler from the revised sequential program of the game is found to achieve a 5.1 faster performance at 8-threads than original one on an IBM Power 5+ machine that is equipped with 8-cores, and (2) hazards are caused by thread contentions over globally shared data, and as well as thread private data, and (3) AI driven players are represented very similarly to Human players inside ioquake3 engine, which gives an estimation of the costs for parallelizing Human driven sessions, and (4) 70% of the costs of the experiment is spent in analyzing ioquake3 code, 30% in implementing the changes in the code.

We built a 12.4 mm x 12.4 mm, 45-nm CMOS, chip that integrates eight 648-MHz general purpose cores, two matrix processor (MX-2) cores, four flexible engine (FE) cores and media IP (VPU5) to establish heterogeneous multi-core chip architecture. The general purpose core had its IPC (instructions per cycle) performance enhanced by adding 32-bit instructions to the existing 16-bit fixed-length instruction set and executing up to two 32-bit instructions per cycle. Considering these five-to-seven years of embedded LSI and increasing trend of access-master within LSI, we predict that the memory usage of single core will not exceed 32-bit physical area (i.e. 4 GB), but chip-total memory usage will exceed 4 GB. Based on this prediction, the physical address was expanded from 32-bit to 40-bit. The fabricated chip was tested and a parallel operation of eight general purpose cores and four FE cores and eight data transfer units (DTU) is obtained on AAC (Advanced Audio Coding) encode processing.

Heterogeneous multicores have been attracting much attention to attain high performance keeping power consumption low in wide spread of areas. However, heterogeneous multicores force programmers very difficult programming. The long application program development period lowers product competitiveness. In order to overcome such a situation, this paper proposes a compilation framework which bridges a gap between programmers and heterogeneous multicores. In particular, this paper describes the compilation framework based on OSCAR compiler. It realizes coarse grain task parallel processing, data transfer using a DMA controller, power reduction control from user programs with DVFS and clock gating on various heterogeneous multicores from different vendors. This paper also evaluates processing performance and the power reduction by the proposed framework on a newly developed 15 core heterogeneous multicore chip named RP-X integrating 8 general purpose processor cores and 3 types of accelerator cores which was developed by Renesas Electronics, Hitachi, Tokyo Institute of Technology and Waseda University. The framework attains speedups up to 32x for an optical flow program with eight general purpose processor cores and four DRP(Dynamically Reconfigurable Processor) accelerator cores against sequential execution by a single processor core and 80% of power reduction for the real-time AAC encoding.

Heterogeneous multicore architectures, integrating several kinds of accelerator cores in addition to general purpose processor cores, have been attracting much attention to realize high performance with low power consumption. To attain effective high performance, high application software productivity, and low power consumption on heterogeneous multicores, cooperation between an architecture and a parallelizing compiler is important. This paper proposes a compiler cooperative heterogeneous multicore architecture and parallelizing compilation scheme for it. Performance of the proposed scheme is evaluated on the heterogeneous multicore integrating Hitachi and Renesas' SH4A processor cores and Hitachi's FE-GA accelerator cores, using an MP3 encoder. The heterogeneous multicore gives us 14.34 times speedup with two SH4As and two FE-GAs, and 26.05 times speedup with four SH4As and four FE-GAs against sequential execution with a single SH4A. The cooperation between the heterogeneous multicore architecture and the parallelizing compiler enables to achieve high performance in a short development period. © 2011 Springer-Verlag Berlin Heidelberg.

We develop a heterogeneous multi-core SoC for applications, such as digital TV systems with IP networks (IP-TV) including image recognition and database search. Figure 5.3.1 shows the chip features. This SoC is capable of decoding 1080i audio/video data using a part of SoC (one general-purpose CPU core, video processing unit called VPU5 and sound processing unit called SPU) [1]. Four dynamically reconfigurable processors called FE [2] are integrated and have a total theoretical performance of 41.5GOPS and power consumption of 0.76W. Two 1024-way matrix-processors called MX-2 [3] are integrated and have a total theoretical performance of 36.9GOPS and power consumption of 1.10W. Overall, the performance per watt of our SoC is 37.3GOPS/W at 1.15V, the highest among comparable processors [4-6] excluding special-purpose codecs. The operation granularity of the CPU, FE and MX-2 are 32bit, 16bit, and 4bit respectively, and thus we can assign the appropriate processor for each task in an effective manner. A heterogeneous multi-core approach is one of the most promising approaches to attain high performance with low frequency, or low power, for consumer electronics application and scientific applications, compared to homogeneous multi-core SoCs [4]. For example, for image-recognition application in the IP-TV system, the FEs are assigned to calculate optical flow operation [7] of VGA (640x480) size video data at 15fps, which requires 0.62GOPS. The MX-2s are used for face detection and calculation of the feature quantity of the VGA video data at 15fps, which requires 30.6GOPS. In addition, general-purpose CPU cores are used for database search using the results of the above operations, which requires further enhancement of CPU. The automatic parallelization compilers analyze parallelism of the data flow, generate coarse grain tasks, schedule tasks to minimize execution time considering data transfer overhead for general-purpose CPU and FE. ©2010 IEEE.

OSCAR (Optimally Scheduled Advanced Multiprocessor) API has been designed for real-time embedded low-power multicores to generate parallel programs for various multicores from different vendors by using the OSCAR parallelizing compiler. The OSCAR API has been developed by Waseda University in collaboration with Fujitsu Laboratory, Hitachi, NEC, Panasonic, Renesas Technology, and Toshiba in an METI/NEDO project entitled "Multicore Technology for Realtime Consumer Electronics." By using the OSCAR API as an interface between the OSCAR compiler and backend compilers, the OSCAR compiler enables hierarchical multigrain parallel processing with memory optimization under capacity restriction for cache memory, local memory, distributed shared memory, and on-chip/off-chip shared memory; data transfer using a DMA controller; and power reduction control using DVFS (Dynamic Voltage and Frequency Scaling), clock gating, and power gating for various embedded multicores. In addition, a parallelized program automatically generated by the OSCAR, compiler with OSCAR API can be compiled by the ordinary OpenMP compilers since the OSCAR API is designed on a subset of the OpenMP. This paper describes the OSCAR API and its compatibility with the OSCAR compiler by showing code examples. Performance evaluations of the OSCAR compiler and the OSCAR. API are carried out using an IBM Power5+ workstation, an IBM Power6 high-end SMP server, and a newly developed consumer electronics multicore chip RP2 by Renesas, Hitachi and Waseda. From the results of scalability evaluation, it is found that on an average, the OSCAR compiler with the OSCAR API can exploit 5.8 times speedup over the sequential execution on the Power5+ workstation with eight cores and 2.9 times speedup on RP2 with four cores, respectively. In addition, the OSCAR compiler can accelerate an IBM XL Fortran compiler up to 3.3 times on the Power6 SMP server. Due to low-power optimization on RP2, the OSCAR compiler with the OSCAR API achieves a maximum power reduction of 84% in the real-time execution mode.

We are developing a software-execution framework based on an octo-core chip multiprocessor named RP2 and an automatic multigrain-parallelizing compiler named OSCAR. The main purpose of this framework is to maintain good speed scalability and power efficiency over the number of processor cores under severe hardware restrictions for embedded use. Key to the speed scalability is reduction of a communication overhead with parallelized tasks. A data-categorization scheme enables small-overhead cache-coherency maintenance by using directives and instructions from the compiler. In this scheme, the number of cache-flushing time is minimized and parallelized tasks are quickly synchronized by using flags in local memory. As regards power efficiency, to reduce power consumption, power supply to processor cores waiting for other cores is timely and frequently cut off, even in the middle of an application, by using a timelypower- gating scheme. In this scheme, to achieve quick mode transition between "NORMAL" mode and "RESUME POWEROFF" mode, register values of the processor core are stored in core-local memory, which is active even in "RESUME POWEROFF" mode and can be accessed in one or two clock cycles. Measured speed and power of an application show good speed scalability in execution time and high power efficiency, simultaneously. In the case of a secure AAC-LC encoding program, execution speed when eight processor cores are used can be increased by 4.85 times compared to that of sequential execution. Moreover, power consumption under the same condition can be reduced by 51.0% by parallelizing and timely-power gating. The time for mode transition is less than 20 μsec, which is only 2.5% of the "RESUME POWER-OFF" period. © 2009 IEEE.

A power-aware compiler controllable chip multiprocessor (CMP) is presented and its performance and power consumption are evaluated with the optimally scheduled advanced multiprocessor (OSCAR) parallelizing compiler. The CMP is equipped with power control registers that change clock frequency and power supply voltage to functional units including processor cores, memories, and an interconnection network. The OSCAR compiler carries out coarse-grain task parallelization of programs and reduces power consumption using architectural power control support and the compiler's power saving scheme. The performance evaluation shows that MPEG-2 encoding on the proposed CMP with four CPUs results in 82.6% power reduction in real-time execution mode with a deadline constraint on its sequential execution time. Furthermore, MP3 encoding on a heterogeneous CMP with four CPUs and four accelerators results in 53.9% power reduction at 21.1-fold speed-up in performance against its sequential execution in the fastest execution mode.

This paper describes a heterogeneous multi-core processor (HMCP) architecture that integrates general-purpose processors (CPUs) and accelerators (ACCs) to achieve exceptional performance as well as low-power consumption for the SoCs of embedded systems. The memory architectures of CPUs and ACCs were unified to improve programming and compiling efficiency. Advanced audio codec-low complexity (AAC-LC) stereo audio encoding was parallelized on a heterogeneous multi-core having homogeneous processor cores and dynamically reconfigurable processor (DRP) ACC cores in a preliminary evaluation of the HMCP architecture. The performance evaluation revealed that 54x AAC encoding was achieved on the chip with two CPUs at 600 MHz and two DRPs at 300 MHz, which achieved encoding of an entire CD within 1-2 min.

A 104.8mm2 90nm CMOS 600MHz SoC integrates 8 processor cores and 8 user RAMs in 17 separate power domains and delivers 33.6GFLOPS. An automatic parallelizing compiler assigns tasks to each CPU and controls its power mode including power supply in accordance with its processing load and status. The compiler also uses barrier registers to achieve fast and accurate CPU synchronization. ©2008 IEEE.

Multicore processors, or chip multiprocessors, which allow us to realize low power consumption, high effective performance, good cost performance and short hardware/software development period, are attracting much attention. In order to achieve full potential of multicore processors, cooperation with a parallelizing compiler is very important. The latest compiler extracts multilevel parallelism, such as coarse grain task parallelism, loop parallelism and near fine grain parallelism, to keep parallel execution efficiency high. It also controls voltage and clock frequency of processors carefully to reduce energy consumption during execution of an application program. This paper evaluates performance of compiler controlled power saving scheme which has been implemented in OSCAR multigrain parallelizing compiler. The developed power saving scheme realizes voltage/frequency control and power shutdown of each processor core during coarse grain task parallel processing. In performance evaluation, when static power is assumed as one-tenth of dynamic power, OSCAR compiler with the power saving scheme achieved 61.2 percent energy reduction for SPEC CFP95 applu without performance degradation on 4 processors and 87.4 percent energy reduction for mpeg2encode, 88.1 percent energy reduction for SPEC CFP95 tomcatv and 84.6 percent energy reduction for applu with real-time deadline constraint on 4 processors.

A heterogeneous multi-core processor (HMCP) architecture, which integrates general purpose processors (CPU) and accelerators (ACC) to achieve high-performance as well as low-power consumption with the support of a parallelizing compiler, was developed. The evaluation was performed using an MP3 audio encoder on a simulator that accurately models the HMCP, It showed that 16-frame encoding on the HMCP with four CPUs and four ACCs yielded 24.5-fold speed-up of performance against sequential execution on one CPU. Furthermore, power saving by the compiler reduced energy consumption of the encoding to 0.17 J, namely, by 28.4%.

Multicore processors have become mainstream computer architecture to go beyond the performance and power efficiency limits of single-core processors. To achieve low power consumption and high performance on multicores, parallelizing compilers take on an important role. This paper describes the performance of a compiler-based power reduction scheme cooperating with OSCAR multigrain parallelizing compiler on a newly developed 8-way SH4A low power multicore chip for consumer electronics, which supports DVFS (Dynamic Voltage and Frequency Scaling) and Clock/Power Gating. Using hardware parameters and parallelized program information, OSCAR compiler determines suitable voltage and frequency of each active processor core and appropriate schedule of clock gating and power gating. Performance experiments shows the compiler reduces consumed power by 88.3%, namely from 5.68 W to 0.67 W, for real-time secure AAC Encoding and 73.5%, namely from 5.73 W to 1.52 W, for real-time MPEG2 Decoding on 8 core execution.

Multicore processors have been adopted for consumer electronics like portable electronics, mobile phones, car navigation systems, digital TVs and games to obtain high performance with low power consumption. The OSCAR automatic parallelizing compiler has been developed to utilize these multicores easily. Also, a new Consumer Electronics Multicore Application Program Interface (API) to use the OSCAR compiler with native sequential compilers for various kinds of multicores from different vendors has been developed in NEDO (New Energy and Industrial Technology Development Organization) "Multicore Technology for Realtime Consumer Electronics" project with Japanese 6 IT companies. This paper evaluates the parallel processing performance of multimedia applications using this API by the OSCAR compiler on the FR1000 4 VLIW cores multicore processor developed by Fujitsu Ltd, and the RP1 4 SH-4A cores multicore processor jointly-developed by Renesas Technology Corp., Hitachi Ltd. and Waseda University. As the results, the parallel codes generated by the OSCAR compiler using the API give us 3.27 times speedup on average using 4 cores against 1 core on the FR1000 multicore, and 3.31 times speedup on average using 4 cores against 1 core on the RP1 multicore.

A 4320MIPS four-core SoC that supports both SMP and AMP for embedded applications is designed in 90nm CMOS. Each processor-core can be operated with a different frequency dynamically including clock stop, while keeping data cache coherency, to maintain maximum processing performance and to reduce average operating power. The 97.6mm2 die achieves a floating-point performance of 16.8GFLOPS. © 2007 IEEE.

A heterogeneous multiprocessor on a chip has been designed and implemented. It consists of 2 CPUs and 2 DRPs (Dynamic Reconfigurable Processors). The design of DRP was intended to achieve high-performance in a small area to be integrated on a SoC for embedded systems. Memory architecture of CPUs and DRPs were unified to improve programming and compiling efficiency. 54x AAC-LC stereo encoding has been enabled with 2 DRPs at 300MHz and 2 CPUs at 600MHz.

With the increase of transistors integrated onto a chip, multi core processor architectures have attracted much attention to achieve high effective performance, shorten development period and reduce the power consumption. To this end, the compiler for a multi core processor is expected not only to parallelize program effectively, but also to control the voltage and clock frequency of processors and storages carefully inside an application program. This paper proposes a compilation scheme for reduction of power consumption under the multigrain parallel processing environment that controls Voltage/Frequency and power supply of each processor core on a chip. In the evaluation, the OSCAR compiler with the proposed scheme achieves 60.7 percent energy savings for SPEC CFP95 applu without performance degradation on 4 processors, and 45.4 percent energy savings for SPEC CFP95 tomcatv with real-time deadline constraint on 4 processors, and 46.5 percent energy savings for SPEC CFP95 swim with the deadline constraint on 4 processors. © 2006 Springer-Verlag Berlin Heidelberg.

This paper describes performance of OSCAR multigrain parallelizing compiler on various SMP servers, such as IBM pSeries 690, Sun Fire V880, Sun Ultra 80, NEC TX7/i6010 and SGI Altix 3700. The OSCAR compiler hierarchically exploits the coarse grain task parallelism among loops, subroutines and basic blocks and the near fine grain parallelism among statements inside a basic block in addition to the loop parallelism. Also, it allows us global cache optimization over different loops, or coarse grain tasks, based on data localization technique with interarray padding to reduce memory access overhead. Current performance of OSCAR compiler is evaluated on the above SMP servers. For example, the OSCAR compiler generating OpenMP parallelized programs from ordinary sequential Fortran programs gives us 5.7 times speedup, in the average of seven programs, such as SPEC CFP95 tomcatv, swim, su2cor, hydro2d, mgrid, applu and turb3d, compared with IBM XL Fortran compiler 8.1 on IBM pSeries 690 24 processors SMP server. Also, it gives us 2.6 times speedup compare with Intel Fortran Itanium Compiler 7.1 on SGI Altix 3700 Itanium 2 16 processors server, 1.7 times speedup compared with NEC Fortran Itanium Compiler 3.4 on NEC TX7/i6010 Itanium 2 8 processors server, 2.5 times speedup compared with Sun Forte 7.0 on Sun Ultra 80 UltraSPARC II4 processors desktop work-station, and 2.1 times speedup compare with Sun Forte compiler 7.1 on Sun Fire V880 UltraSPARC III Cu 8 processors server.

This paper describes multigrain parallel processing on a compiler cooperative chip multiprocessor The multigrain parallel processing hierarchically exploits multiple grains of parallelism such as coarse grain task parallelism, loop iteration level parallelism and statement level near-fine grain parallelism. The chip multiprocessor has been designed to attain high effective peformance, cost effectiveness and high software productivity by supporting the optimizations of the multigrain parallelizing compiler, which is developed by Japanese Millennium Project IT21 "Advance Parallelizing Compiler". To achieve full potential of multigrain parallel processing, the chip multiprocessor integrates simple single-issue processors having distributed shared data memory for both optimal use of data locality and scalar data transfer local data memory for processor private data, in addition to centralized shared memory for shared data among processors. This paper focuses on the scalability of the chip multiprocessor having up to eight processors on a chip by exploiting of the multigrain parallelism from SPECfp95 programs. When microSPARC like the simple processor core is used under assumption of 90 nm technology and 2.8 GHz, the evaluation results show the speedups for eight processors and four processors reach 7.1 and 3.9, respectively. Similarly, when 400 MHz is assumed for embedded usage, the speedups reach 7.8 and 4.0, respectively.

Chip Multiprocessor (CMP) architecture has attracting much attention as a next-generation microprocessor architecture and many kinds of CMP are widely being researched. However, CMP architectures several difficulties for effective use of memory, especially cache or local memory near a processor core. The authors have proposed OSCAR CMP architecture, which cooperatively works with multigrain parallelizing compiler which gives us much higher parallelism than instruction level parallelism or loop level parallelism and high productivity of application programs. To support the compiler optimization for effective use of cache or local memory, OSCAR CMP has local data memory (LDM) for processor private data and distributed shared memory (DSM) for synchronization and fine grain data transfers among processors, in addition to centralized shared memory (CSM) to support dynamic task scheduling. This paper proposes a static coarse grain task scheduling scheme for data localization using live variable analysis. Furthermore, remote memory data transfer scheduling scheme using information of live variable analysis is also described. The proposed scheme is implemented on OSCAR FORTRAN multigrain parallelizing compiler and is evaluated on OSCAR CMP using Tomcatv and Swim in SPEC CFP 95 benchmark.

Currently, many people are enjoying multimedia applications with image and audio processing on PCs, PDAs, mobile phones and so on. With the popularization of the multimedia applications, needs for low cost, low power consumption and high performance processors has been increasing. To this end, chip multiprocessor architectures which allow us to attain scalable performance improvement by using multigrain parallelism are attracting much attention. However, in order to extract higher performance on a chip multiprocessor, more sophisticated software techniques are required, such as decomposing a program into adequate grain of tasks, assigning them onto processors considering parallelism, data locality optimization and so on. This paper describes a parallel processing scheme for MPEG2 encoding using data localization which improve execution efficiency assigning coarse grain tasks sharing same data on a same processor consecutively for a chip multiprocessor. The performance evaluation on OSCAR chip multiprocessor architecture shows that proposed scheme gives us 6.97 times speedup using 8 processors and 10.93 times speedup using 16 processors against sequential execution time respectively. Moreover, the proposed scheme gives us 1.61 times speedup using 8 processors and 2.08 times speedup using 16 processors against loop parallel processing which has been widely used for multiprocessor systems using the same number of processors.

Effective use of cache memory is getting more important with increasing gap between the processor speed and memory access speed. Also, use of multigrain parallelism is getting more important to improve effective performance beyond the limitation of loop iteration level parallelism. Considering these factors, this paper proposes a coarse grain task static scheduling scheme considering cache optimization. The proposed scheme schedules coarse grain tasks to threads so that shared data among coarse grain tasks can be passed via cache after task and data decomposition considering cache size at compile time. It is implemented on OSCAR Fortran multigrain parallelizing compiler and evaluated on Sun Ultra80 four-processor SMP workstation using Swim and Tomcatv from the SPEC fp 95. As the results, the proposed scheme gives us 4.56 times speedup for Swim and 2.37 times on 4 processors for Tomcatv respectively against the Sun Forte HPC Ver. 6 update 1 loop parallelizing compiler.

It seems that Instruction Level Parallelism (ILP) approach, which has been used by various superscalar processors and VLIW processors for a long time, reaches its limitation of performance improvement. To obtain scalable performance improvement, cost effectiveness and high productivity even in the era of one billion transistors, the cooperative work between software and hardware is getting increasingly important. For this reason, the authors have developed OSCAR (Optimally SCheduled Advanced multiprocessoR) Chip Multiprocessor (OSCAR CMP) and OSCAR multigrain compiler simultaneously. To preserve the scalability in the future, OSCAR CMP has mechanisms for efficient use of parallelism and data locality, and for hiding data transfer overhead. These mechanisms can be fully controlled by the OSCAR multigrain compiler In this paper, the authors focus on multigrain parallel processing on OSCAR CMP, which enables us to exploit loop iteration level parallelism and coarse grain task parallelism in addition to ILP from the entire of a program. Performance of multigrain parallel processing on OSCAR CMP architecture is evaluated using SPEC fp 2000195 benchmark suite. When microSPARC like single issue core is used, OSCAR CMP gives us from 1.77 to 3.96 times speedup for four processors against single processor In addition, OSCAR CMP is compared with Sun UltraSPARC II like processor to evaluate cost effectiveness. As a result, OSCAR CMP gives us 1.66 times better performance on the average under the condition that OSCAR CMP and UltraSPARC II are built from almost same number of transistors.

With the recent increase of multimedia contents using JPEG and MPEG, low cost, low power consumption and high performance processors for multimedia application have been expected. Particularly, single chip multiprocessor architectures having simple processor cores that will attain scalability and cost performance are attracting much attention to develop such processors. Single chip multiprocessor architectures allow us to exploit coarse grain task level and loop level parallelism in addition to the instruction level parallelism, so parallel processing technology is indispensable to get us scalable performance improvement. This paper describes a multigrain parallel processing scheme for the JPEG encoding for a single chip multiprocessor and its performance. The evaluation shows an OSCAR type single chip multiprocessor having four single-issue simple processor cores gave us 3.59 times speed-up.

Effective use of cache memory is getting more important with increasing gap between the processor speed and memory access speed. Also, use of multigrain parallelism is getting more important to improve effective performance beyond the limitation of loop iteration level parallelism. Considering these factors, this paper proposes a coarse grain task static scheduling scheme considering cache optimization. The proposed scheme schedules coarse grain tasks to threads so that shared data among coarse grain tasks can be passed via cache after task and data decomposition considering cache size at compile time. It is implemented on OSCAR Fortran multigrain parallelizing compiler and evaluated on Sun Ultra80 four-processor SMP workstation, using Swim and Tomcatv from the SPEC fp 95. As the results, the proposed scheme gives us 4.56 times speedup for Swim and 2.37 times on 4 processors for Tomcatv respectively against the Sun Forte HPC 6 loop parallelizing compiler. © 2002 Springer Berlin Heidelberg.

With the recent increase of multimedia contents using JPEG and MPEG, low cost, low power consumption and high performance processors for multimedia application have been expected. Particularly, single chip multiprocessor architecture having simple processor cores that will attain good scalability and cost effectiveness is attracting much attention. To exploit full performance of single chip multiprocessor architecture, multigrain parallel processing, which exploits coarse grain task parallelism, loop parallelism and instruction level parallelism, is attractive. This paper describes a multigrain parallel processing scheme for the JPEG encoding on a single chip multiprocessor and its performance. The evaluation shows an OSCAR type single chip multiprocessor having four single-issue simple processor cores gave us 3.59 times speed-up against sequential execution time.

This paper describes OSCAR multigrain parallelizing compiler which has been developed in Japanese Millennium Project IT21 "Advanced Parallelizing Compiler" project and its performance on SMP machines. The compiler realizes multigrain parallelization for chip-multiprocessors to high-end servers. It hierarchically exploits coarse grain task parallelism among loops, subroutines and basic blocks and near fine grain parallelism among statements inside a basic block in addition to loop parallelism. Also, it globally optimizes cache use over different loops, or coarse grain tasks, based on data localization technique to reduce memory access overhead Current performance of OSCAR compiler for SPEC95fp is evaluated on different SMPs. For example, it gives us 3.7 times speedup for HYDRO2D, 1.8 times for SWIM, 1.7 times for SU2COR, 2.0 times for MGRID, 3.3 times for TURB3D on 8 processor IBM RS6000, against XL Fortran compiler ver:7.1 and 4.2 times speedup for SWIM and 2.2 times speedup for TURB3D on 4 processor Sun Ultra80 workstation against Forte6 update 2.

With the increase of the number of transistors integrated on a chip, efficient use of transistors and scalable improvement of effective performance of a processor are getting important problems. However it has been thought that popular superscalar and VLIW would have difficulty, to obtain scalable improvement of effective performance in future because of the limitation of instruction level parallelism. To cope with this problem, a single chip multiprocessor (SCM) approach,vith multi grain parallelprocessing inside a chip, which hierarchically exploits loop parallelism and coarse grain parallelism among subroutines, loops and basic blocks in addition to instruction level parallelism, is thought one of the most promising approaches. This paper evaluates effectiveness of the single chip multiprocessor architectures with a shared cache, global registers, distributed shared memory and/or local memory for near fine grain parallel processing as the first step of research on SCM architecture to support multi grain parallel processing. The evaluation shows OSCAR (Optimally Scheduled Advanced Multiprocessor architecture having distributed shared memory and local memory in addition to centralized shared memory and attachment of global register gives us significant speed up such as 13.8% to 143.8% for four processors compared with shared cache architecture for applications which have been difficult to extract parallelism effectively.

With the increase of the number of transistors integrated on a chip, efficient use of transistors and scalable improvement of effective performance of a processor are getting im-portant problems. However, it has been thought that popular superscalar and VLIW would have difficulty to obtain scalable improvement of effective performance in future because of the limitation of instruction level parallelism. To cope with this problem, a single chip multiprocessor (SCM) approach with multi grain parallel processing inside a chip, which hierarchically exploits loop parallelism and coarse grain parallelism among subroutines, loops and basic blocks in addition to instruction level parallelism, is thought one of the most promising approaches. This paper evaluates effectiveness of the single chip multiprocessor architectures with a shared cache, global registers, distributed shared memory and/or local memory for near fine grain parallel processing as the first step of research on SCM architecture to support multi grain parallel processing. The evaluation shows OSCAR (Optimally Scheduled Advanced Multiprocessor) architecture having distributed shared memory and local memory in addition to centralized shared memory and attachment of global register gives us significant speed up such as 13.8% to 143.8% for four pro-cessors compared with shared cache architecture for applications which have been difficult to extract parallelism effectively.

OSCAR (Optimally Scheduled Advanced Multiprocessor) was designed to efficiently realize multi-grain parallel processing using static and dynamic scheduling. It is a shared memory multiprocessor system having centralized and distributed shared memories in addition to local memory on each processor with data transfer controller for overlapping of data transfer and task processing. Also, its Fortran multi-grain compiler hierarchically exploits coarse grain parallelism among loops, subroutines and basic blocks, conventional medium grain parallelism among loop-iterations in a Doall loop and near fine grain parallelism among statements. At the coarse grain parallel processing, data localization (automatic data distribution) have been employed to minimize data transfer overhear. In the near fine grain processing of a basic block, explicit synchronization can be removed by use of a clock level accurate code scheduling technique with architectural supports. This paper describes OSCAR's architecture, its compiler and the performance for the multi-grain parallel processing. OSCAR's architecture and compilation technology will be more important in future High Performance Computers and single chip multiprocessors.