Search

The commercial release of byte-addressable persistent memories, such as Intel/Micron 3D XPoint memory, is imminent. Ongoing research has sought mechanisms to allow programmers to implement recoverable data structures in these new main memories. Ensuring recoverability requires programmer control of the order of persistent stores; recent work proposes persistency models as an extension to memory consistency to specify such ordering. Prior work has considered persistency models at the abstraction of the instruction set architecture. Instead, we argue for extending the language-level memory model to provide guarantees on the order of persistent writes. We explore a taxonomy of guarantees a language-level persistency model might provide, considering both atomicity and ordering constraints on groups of persistent stores. Then, we propose and evaluate Acquire-Release Persistency (ARP), a language-level persistency model for C++11. We describe how to compile code written for ARP to a state-of-the-art ISA-level persistency model. We then consider enhancements to the ISA-level persistency model that can distinguish memory consistency constraints required for proper synchronization but unnecessary for correct recovery. With these optimizations, we show that ARP increases performance by up to 33.2% (19.8% avg.) over coding directly to the baseline ISA-level persistency model for a suite of persistent-write-intensive workloads.

As technology scales, GPUs are forecasted to incorporate an ever-increasing amount of computing resources to support thread-level parallelism. But even with the best effort, exposing massive thread-level parallelism from a single GPU kernel, particularly from general purpose applications, is going to be a difficult challenge. In some cases, even if there is sufficient thread-level parallelism in a kernel, there may not be enough available memory bandwidth to support such massive concurrent thread execution. Hence, GPU resources may be underutilized as more general purpose applications are ported to execute on GPUs. In this paper, we explore multiprogramming GPUs as a way to resolve the resource underutilization issue. There is a growing hardware support for multiprogramming on GPUs. Hyper-Q has been introduced in the Kepler architecture which enables multiple kernels to be invoked via tens of hardware queue streams. Spatial multitasking has been proposed to partition GPU resources across multiple kernels. But the partitioning is done at the coarse granularity of streaming multiprocessors (SMs) where each kernel is assigned to a subset of SMs. In this paper, we advocate for partitioning a single SM across multiple kernels, which we term as intra-SM slicing. We explore various intra-SM slicing strategies that slice resources within each SM to concurrently run multiple kernels on the SM. Our results show that there is not one intra-SM slicing strategy that derives the best performance for all application pairs. We propose Warped-Slicer , a dynamic intra-SM slicing strategy that uses an analytical method for calculating the SM resource partitioning across different kernels that maximizes performance. The model relies on a set of short online profile runs to determine how each kernel's performance varies as more thread blocks from each kernel are assigned to an SM. The model takes into account the interference effect of shared resource usage across multiple kernels. The model is also computationally efficient and can determine the resource partitioning quickly to enable dynamic decision making as new kernels enter the system. We demonstrate that the proposed Warped-Slicer approach improves performance by 23% over the baseline multiprogramming approach with minimal hardware overhead.

Recent developments in GPU execution models and architectures have introduced dynamic parallelism to facilitate the execution of irregular applications where control flow and memory behavior can be unstructured, time-varying, and hierarchical. The changes brought about by this extension to the traditional bulk synchronous parallel (BSP) model also creates new challenges in exploiting the current GPU memory hierarchy. One of the major challenges is that the reference locality that exists between the parent and child thread blocks (TBs) created during dynamic nested kernel and thread block launches cannot be fully leveraged using the current TB scheduling strategies. These strategies were designed for the current implementations of the BSP model but fall short when dynamic parallelism is introduced since they are oblivious to the hierarchical reference locality. We propose LaPerm , a new locality-aware TB scheduler that exploits such parent-child locality, both spatial and temporal. LaPerm adopts three different scheduling decisions to i) prioritize the execution of the child TBs, ii) bind them to the stream multiprocessors (SMXs) occupied by their parents TBs, and iii) maintain workload balance across compute units. Experiments with a set of irregular CUDA applications executed on a cycle-level simulator employing dynamic parallelism demonstrate that LaPerm is able to achieve an average of 27% performance improvement over the baseline round-robin TB scheduler commonly used in modern GPUs.

Data races make parallel programs hard to understand. Precise race detection that stops an execution on first occurrence of a race addresses this problem, but it comes with significant overhead. In this work, we exploit the insight that precisely detecting only write-after-write (WAW) and read-after-write (RAW) races suffices to provide cleaner semantics for racy programs. We demonstrate that stopping an execution only when these races occur ensures that synchronization-free-regions appear to be executed in isolation and that their writes appear atomic. Additionally, the undetected racy executions can be given certain deterministic guarantees with efficient mechanisms. We present C lean , a system that precisely detects WAW and RAW races and deterministically orders synchronization. We demonstrate that the combination of these two relatively inexpensive mechanisms provides cleaner semantics for racy programs. We evaluate both software-only and hardware-supported C lean . The software-only C lean runs all Pthread benchmarks from the SPLASH-2 and PARSEC suites with an average 7.8x slowdown. The overhead of precise WAW and RAW detection (5.8x) constitutes the majority of this slowdown. Simple hardware extensions reduce the slowdown of C lean 's race detection to on average 10.4% and never more than 46.7%.

GPUs are being increasingly adopted as compute accelerators in many domains, spanning environments from mobile systems to cloud computing. These systems are usually running multiple applications, from one or several users. However GPUs do not provide the support for resource sharing traditionally expected in these scenarios. Thus, such systems are unable to provide key multiprogrammed workload requirements, such as responsiveness, fairness or quality of service. In this paper, we propose a set of hardware extensions that allow GPUs to efficiently support multiprogrammed GPU workloads. We argue for preemptive multitasking and design two preemption mechanisms that can be used to implement GPU scheduling policies. We extend the architecture to allow concurrent execution of GPU kernels from different user processes and implement a scheduling policy that dynamically distributes the GPU cores among concurrently running kernels, according to their priorities. We extend the NVIDIA GK110 (Kepler) like GPU architecture with our proposals and evaluate them on a set of multiprogrammed workloads with up to eight concurrent processes. Our proposals improve execution time of high-priority processes by 15.6x, the average application turnaround time between 1.5x to 2x, and system fairness up to 3.4x. We would like to thank the anonymous reviewers, Alexan- der Veidenbaum, Carlos Villavieja, Lluis Vilanova, Lluc Al- varez, and Marc Jorda on their comments and help improving our work and this paper. This work is supported by Euro- pean Commission through TERAFLUX (FP7-249013), Mont- Blanc (FP7-288777), and RoMoL (GA-321253) projects, NVIDIA through the CUDA Center of Excellence program, Spanish Government through Programa Severo Ochoa (SEV-2011-0067) and Spanish Ministry of Science and Technology through TIN2007-60625 and TIN2012-34557 projects.

Real-world programs often thrash in the instruction cache, especially when SMT methods are used. The Mill™ split-stream encoding doubles the effective capacity of the instruction cache at no increase in perinstruction power usage or cache access latency, while also sharply increasing the potential maximal decode rate for instruction sets that use variable-length encoding.

A significant portion of the energy dissipated in modern integrated circuits is consumed by the overhead associated with timing guardbands that ensure reliable execution. Timing speculation, where the pipeline operates at an unsafe voltage with any rare errors detected and resolved by the architecture, has been demonstrated to significantly improve the energy-efficiency of scalar processor designs. Unfortunately, applying the same timing-speculative approach to wide-SIMD architectures, such as those used in highly-efficient GPUs, may not provide similar gains. In this work, we make two important contributions. The first is a set of models describing a parametrized general error probability function that is based on measurements of a fabricated chip and the expected efficiency benefits of timing speculation in a SIMD context. The second contribution is a decoupled SIMD pipeline that more effectively utilizes timing speculation and recovery, when compared with a standard SIMD design that uses only conventional timing speculation. The proposed lane decoupling enables each SIMD lane to tolerate timing errors independent of other adjacent lanes, resulting in higher throughput and improved scalability. We validate our models and evaluate our design using a cycle-based GPU simulator, describe the conditions where efficiency improvements can be obtained, and explore the benefits of decoupling across a wide range of parameters. Our results show that timing speculation can achieve up to 10.3% improvement in efficiency.

Since the introduction of fully programmable vertex shader hardware, GPU computing has made tremendous advances. Exception support and speculative execution are the next steps to expand the scope and improve the usability of GPUs. However, traditional mechanisms to support exceptions and speculative execution are highly intrusive to GPU hardware design. This paper builds on two related insights to provide a unified lightweight mechanism for supporting exceptions and speculation on GPUs. First, we observe that GPU programs can be broken into code regions that contain little or no live register state at their entry point. We then also recognize that it is simple to generate these regions in such a way that they are idempotent, allowing their entry points to function as program recovery points and enabling support for exception handling, fast context switches, and speculation, all with very low overhead. We call the architecture of GPUs executing these idempotent regions the iGPU architecture. The hardware extensions required are minimal and the construction of idempotent code regions is fully transparent under the typical dynamic compilation framework of GPUs. We demonstrate how iGPU exception support enables virtual memory paging with very low overhead (1% to 4%), and how speculation support enables circuit-speculation techniques that can provide over 25% reduction in energy.

Modern memory controllers employ sophisticated address mapping, command scheduling, and power management optimizations to alleviate the adverse effects of DRAM timing and resource constraints on system performance. A promising way of improving the versatility and efficiency of these controllers is to make them programmable---a proven technique that has seen wide use in other control tasks ranging from DMA scheduling to NAND Flash and directory control. Unfortunately, the stringent latency and throughput requirements of modern DDRx devices have rendered such programmability largely impractical, confining DDRx controllers to fixed-function hardware. This paper presents the instruction set architecture (ISA) and hardware implementation of PARDIS, a programmable memory controller that can meet the performance requirements of a high-speed DDRx interface. The proposed controller is evaluated by mapping previously proposed DRAM scheduling, address mapping, refresh scheduling, and power management algorithms onto PARDIS. Simulation results show that the average performance of PARDIS comes within 8% of fixed-function hardware for each of these techniques; moreover, by enabling application-specific optimizations, PARDIS improves system performance by 6--17% and reduces DRAM energy by 9--22% over four existing memory controllers.

Efficient execution of well-parallelized applications is central to performance in the multicore era. Program analysis tools support the hardware and software sides of this effort by exposing relevant features of multithreaded applications. This paper describes parallel block vectors, which uncover previously unseen characteristics of parallel programs. Parallel block vectors provide block execution profiles per concurrency phase (e.g., the block execution profile of all serial regions of a program). This information provides a direct and fine-grained mapping between an application's runtime parallel phases and the static code that makes up those phases. This paper also demonstrates how to collect parallel block vectors with minimal application perturbation using Harmony. Harmony is an instrumentation pass for the LLVM compiler that introduces just 16-21% overhead on average across eight Parsec benchmarks. We apply parallel block vectors to uncover several novel insights about parallel applications with direct consequences for architectural design. First, that the serial and parallel phases of execution used in Amdahl's Law are often composed of many of the same basic blocks. Second, that program features, such as instruction mix, vary based on the degree of parallelism, with serial phases in particular displaying different instruction mixes from the program as a whole. Third, that dynamic execution frequencies do not necessarily correlate with a block's parallelism.

International audience; Single-Instruction Multiple-Thread (SIMT) micro-architectures implemented in Graphics Processing Units (GPUs) run fine-grained threads in lockstep by grouping them into units, referred to as warps, to amortize the cost of instruction fetch, decode and control logic over multiple execution units. As individual threads take divergent execution paths, their processing takes place sequentially, defeating part of the efficiency advantage of SIMD execution. We present two complementary techniques that mitigate the impact of thread divergence on SIMT micro-architectures. Both techniques relax the SIMD execution model by allowing two distinct instructions to be scheduled to disjoint subsets of the the same row of execution units, instead of one single instruction. They increase flexibility by providing more thread grouping opportunities than SIMD, while preserving the affinity between threads to avoid introducing extra memory divergence. We consider (1) co-issuing instructions from different divergent paths of the same warp and (2) co-issuing instructions from different warps. To support (1), we introduce a novel thread reconvergence technique that ensures threads are run back in lockstep at control-flow reconvergence points without hindering their ability to run branches in parallel. We propose a lane shuffling technique to allow solution (2) to benefit from inter-warp correlations in divergence patterns. The combination of all these techniques improves performance by 23% on a set of regular GPGPU applications and by 40% on irregular applications, while maintaining the same instruction-fetch and processing-unit resource requirements as the contemporary Fermi GPU architecture.

This paper introduces a simple method called multimode custom instructions, which aims at reducing the power consumption of the register file of tightly coupled dynamically reconfigurable application specific instruction set processors (DR-ASIPs). To this end, it proposes to divide custom instructions into two sets depending on criteria related to their size, distribution and reuse rate. Performance is measured on a RISC DR-ASIP with a subset of MiBench using an original automatic custom instruction generator from assembly based on the dancing link algorithm. The results show that this method makes it possible to drop the accesses to the register file made by CIs by up to 47 % with no performance hit, and to reduce the use of the context memory by up to 40 % compared to a reference architecture.

This paper presents the domain-specific programmable design of custom computing machines for high-performance stencil computation. Stencil computation is one of the typical kernels in scientific computations, however its low operational-intensity makes the sustained performance limited by memory bandwidth on recent microprocessors and GPUs. So far we have proposed a scalable streaming-array (SSA) of processing elements, which provides almost linear scalability by increasing FPGAs with a constant externalmemory bandwidth. In order to facilitate custom computing and efficiently utilize hardware resources for various and complex stencil-computations, we design programmable SSA with limited but necessary functionality. We show the design concept, the programmable structure and the SIMD instruction set for SSA. Prototype implementation with nine FPGAs demonstrates that our programmable design with a lot of floating-point units exploits hardware resources well, efficiently achieving 260 GFlop/s, which is 87.4% of the peak, at 1295 MFlop/sW.

Dynamic data race detectors are an important mechanism for creating robust parallel programs. Software race detectors instrument the program under test, observe each memory access, and watch for inter-thread data sharing that could lead to concurrency errors. While this method of bug hunting can find races that are normally difficult to observe, it also suffers from high runtime overheads. It is not uncommon for commercial race detectors to experience 300x slowdowns, limiting their usage. This paper presents a hardware-assisted demand-driven race detector. We are able to observe cache events that are indicative of data sharing between threads by taking advantage of hardware available on modern commercial microprocessors. We use these to build a race detector that is only enabled when it is likely that inter-thread data sharing is occurring. When little sharing takes place, this demand-driven analysis is much faster than contemporary continuous-analysis tools without a large loss of detection accuracy. We modified the race detector in Intel(R) Inspector XE to utilize our hardware-based sharing indicator and were able to achieve performance increases of 3x and 10x in two parallel benchmark suites and 51x for one particular program.

The desire to create novel computing systems, paired with recent advances in neuroscientific understanding of the brain, has led researchers to develop neuromorphic architectures that emulate the brain. To date, such models are developed, trained, and deployed on the same substrate. However, excessive co-dependence between the substrate and the algorithm prevents portability, or at the very least requires reconstructing and retraining the model whenever the substrate changes. This paper proposes a well-defined abstraction layer -- the Neuromorphic instruction set architecture, or NISA -- that separates a neural application's algorithmic specification from the underlying execution substrate, and describes the Aivo framework, which demonstrates the concrete advantages of such an abstraction layer. Aivo consists of a NISA implementation for a rate-encoded neuromorphic system based on the cortical column abstraction, a state-of-the-art integrated development and runtime environment (IDE), and various profile-based optimization tools. Aivo's IDE generates code for emulating cortical networks on the host CPU, multiple GPGPUs, or as boolean functions. Its runtime system can deploy and adaptively optimize cortical networks in a manner similar to conventional just-in-time compilers in managed runtime systems (e.g. Java, C#). We demonstrate the abilities of the NISA abstraction by constructing a cortical network model of the mammalian visual cortex, deploying on multiple execution substrates, and utilizing the various optimization tools we have created. For this hierarchical configuration, Aivo's profiling based network optimization tools reduce the memory footprint by 50% and improve the execution time by a factor of 3x on the host CPU. Deploying the same network on a single GPGPU results in a 30x speedup. We further demonstrate that a speedup of 480x can be achieved by deploying a massively scaled cortical network across three GPGPUs. Finally, converting a trained hierarchical network to C/C++ boolean constructs on the host CPU results in 44x speedup.

By adjusting the design of the ISA and enabling circuit timing-sensitive optimizations in a compiler, we can more effectively exploit timing speculation. While there has been growing interest in systems that leverage circuit-level timing speculation to improve the performance and power-efficiency of processors, most of the innovation has been at the microarchitectural level. We make the observation that some code sequences place greater demand on circuit timing deadlines than others. Furthermore, by selectively replacing these codes with instruction sequences which are semantically equivalent but reduce activity on timing critical circuit paths, we can trigger fewer timing errors and hence reduce recovery costs.

Previous research has shown that Explicit Data Graph Execution (EDGE) instruction set architectures (ISA) allow for power efficient performance scaling. In this paper we describe the preliminary design of a new dynamic multicore processor called E2 that utilizes an EDGE ISA to allow for the dynamic composition of physical cores into logical processors. We provide details of E2's support for dynamic reconfigurability and show how the EDGE ISA facilities outof- order vector execution.

Performance improvement solely through transistor scaling is becoming more and more difficult, thus it is increasingly common to see domain specific accelerators used in conjunction with general purpose processors to achieve future performance goals. There is a serious drawback to accelerators, though: binary compatibility. An application compiled to utilize an accelerator cannot run on a processor without that accelerator, and applications that do not utilize an accelerator will never use it. To overcome this problem, we propose decoupling the instruction set architecture from the underlying accelerators. Computation to be accelerated is expressed using a processor’s baseline instruction set, and light-weight dynamic translation maps the representation to whatever accelerators are available in the system. In this paper, we describe the changes to a compilation framework and processor system needed to support this abstraction for an important set of accelerator designs that support innermost loops. In this analysis, we investigate the dynamic overheads associated with abstraction as well as the static/dynamic tradeoffs to improve the dynamic mapping of loop-nests. As part of the exploration, we also provide a quantitative analysis of the hardware characteristics of effective loop accelerators. We conclude that using a hybrid static-dynamic compilation approach to map computation on to loop-level accelerators is an practical way to increase computation efficiency, without the overheads associated with instruction set modification.

Modern processors rely on memory dependence prediction to execute load instructions as early as possible, speculating that they are not dependent on an earlier, unissued store. To date, the most sophisticated dependence predictors, such as Store Sets, have been tightly coupled to the fetch and execution streams, requiring global knowledge of the in-flight stream of stores to synchronize loads with specific stores. This paper proposes a new dependence predictor design, called a Counting Dependence Predictor (CDP). The key feature of CDPs is that the prediction mechanism predicts some set of events for which a particular dynamic load should wait, which may include some number of matching stores. By waiting for local events only, this dependence predictor can work effectively in a distributed microarchitecture where centralized fetch and execution streams are infeasible or undesirable. We describe and evaluate a distributed Counting Dependence Predictor and protocol that achieves 92% of the performance of perfect memory disambiguation. It outperforms a load-wait table, similar to the Alpha 21264, by 11%. Idealized, centralized implementations of Store Sets and the Exclusive Collision Predictor, both of which would be difficult to implement in a distributed microarchitecture, achieve 97% and 94% of oracular performance, respectively.

This paper introduces the Camino Compiler Infrastructure. Camino implements several types of profiling, including basic block counts, edge profiling, interprocedural path profiling, and a special technique that allows using a SimPoint-like methodology to do efficient and precise fine-grained power behavior characterization. It also supports a growing set of code placement optimizations such as branch alignment and pattern history table partitioning. In its current implementation, Camino works as a post-processor for the Gnu Compiler Collection (GCC). The goal of Camino is to serve as a testbed for various low-level performance optimizations as well as power and energy optimizations. It currently supports the x86 instruction set.

The theoretical study of quantum computation has yielded efficient algorithms for some traditionally hard problems. Correspondingly, experimental work on the underlying physical implementation technology has progressed steadily. However, almost no work has yet been done which explores the architecture design space of large scale quantum computing systems. In this paper, we present a set of tools that enable the quantitative evaluation of architectures for quantum computers. The infrastructure we created comprises a complete compilation and simulation system for computers containing thousands of quantum bits. We begin by compiling complete algorithms into a quantum instruction set. This ISA enables the simple manipulation of quantum state. Another tool we developed automatically transforms quantum software into an equivalent, fault-tolerant version required to operate on real quantum devices. Next, our infrastructure transforms the ISA into a set of low-level micro architecture specific control operations. In the future, these operations can be used to directly control a quantum computer. For now, our simulation framework quickly uses them to determine the reliability of the application for the target micro architecture. Finally, we propose a simple, regular architecture for ion-trap based quantum computers. Using our software infrastructure, we evaluate the design trade offs of this micro architecture.

This paper presents a hardware-based dynamic optimizer that continuously optimizes an applicationýs instruction stream. In continuous optimization, dataflow optimizations are performed using simple, table-based hardware placed in the rename stage of the processor pipeline. The continuous optimizer reduces dataflow height by performing constant propagation, reassociation, redundant load elimination, store forwarding, and silent store removal. To enhance the impact of the optimizations, the optimizer integrates values generated by the execution units back into the optimization process. Continuous optimization allows instructions with input values known at optimization time to be executed in the optimizer, leaving less work for the out-of-order portion of the pipeline. Continuous optimization can detect branch mispredictions earlier and thus reduce the misprediction penalty. In this paper, we present a detailed description of a hardware optimizer and evaluate it in the context of a contemporary microarchitecture running current workloads. Our analysis of SPECint, SPECfp, and mediabench workloads reveals that a hardware optimizer can directly execute 33% of instructions, resolve 29% of mispredicted branches, and generate addresses for 76% of memory operations. These positive effects combine to provide speed ups in the range 0.99 to 1.27.

New processors, both embedded and general purpose, often have conflicting design requirements involving space, power, and performance. Architectural features and compiler optimizations often target one or more design goals at the expense of the others. This paper presents a novel architectural and compiler approach to simultaneously reduce power requirements, decrease code size, and improve performance by integrating an instruction register file (IRF) into the architecture. Frequently occurring instructions are placed in the IRF. Multiple entries in the IRF can be referenced by a single packed instruction in ROM or L1 instruction cache. Unlike conventional code compression, our approach allows the frequent instructions to be referenced in arbitrary combinations. The experimental results show significant improvements in space and power, as well as some improvement in execution time when using only 32 entries. These advantages make packing instructions into registers an effective approach for improving overall efficiency.

Runahead execution is a technique that improves processor performance by pre-executing the running application instead of stalling the processor when a long-latency cache miss occurs. Previous research has shown that this technique significantly improves processor performance. However, the efficiency of runahead execution, which directly affects the dynamic energy consumed by a runahead processor, has not been explored. A runahead processor executes significantly more instructions than a traditionalout-of-order processor, sometimes without providing any performance benefit, which makes it inefficient. In this paper, we describe the causes of inefficiency in runahead execution and propose techniques to make a runahead processor more efficient, thereby reducing its energy consumption and possibly increasing its performance. Our analyses and results provide two major insights: (1) the efficiency of runahead execution can be greatly improved with simple techniques that reduce the number of short, overlapping, and useless runahead periods, which we identify as the three major causes of inefficiency, (2) simple optimizations targeting the increase of useful prefetches generated in runahead mode can increase both the performance and efficiency of a runahead processor. The techniques we propose reduce the increase in the number of instructions executed due to runahead execution from 26.5% to 6.2%, on average, without significantly affecting the performance improvement provided by runahead execution.

Instruction set customization is an effective way to improve processor performance. Critical portions of applicationdata-flow graphs are collapsed for accelerated execution on specialized hardware. Collapsing dataflow subgraphs will compress the latency along critical paths and reduces the number of intermediate results stored in the register file. While custom instructions can be effective, the time and cost of designing a new processor for each application is immense. To overcome this roadblock, this paper proposes a flexible architectural framework to transparently integrate custom instructions into a general-purpose processor. Hardware accelerators are added to the processor to execute the collapsed subgraphs. A simple microarchitectural interface is provided to support a plug-and-play model for integrating a wide range of accelerators into a pre-designed and verified processor core. The accelerators are exploited using an approach of static identification and dynamic realization. The compiler is responsible for identifying profitable subgraphs, while the hardware handles discovery, mapping, and execution of compatible subgraphs. This paper presents the design of a plug-and-play transparent accelerator system and evaluates the cost/performance implications of the design.

Embedded processors for video image recognition require to address both the cost (die size and power) versus real-time performance issue, and also to achieve high flexibility due to the immense diversity of recognition targets, situations, and applications. This paper describes IMAP, a highly parallel SIMD linear processor and memory array architecture that addresses these trading-off requirements. By using parallel and systolic algorithmic techniques, despite of its simple architecture IMAP achieves to exploit not only the straightforward per image row data level parallelism (DLP), but also the inherent DLP of other memory access patterns frequently found in various image recognition tasks, under the use of an explicit parallel C language (1DC). We describe and evaluate IMAP-CE, a latest IMAP processor, which integrates 128 of 100MHz 8 bit4-way VLIW PEs, 128 of 2KByte RAMs, and one 16 bit RISC control processor, into a single chip. The PE instruction set is enhanced for supporting 1DC codes. IMAP-CE is evaluated mainly by comparing its performance running 1DC codes with that of a 2.4GHz Intel P4 running optimized C codes. Based on the use of parallelizing techniques, benchmark results show a speedup of up to 20 for image filter kernels, and of 4 for a full image recognition application.

Significant time is spent by companies trying to reproduce and fix the bugs that occur for released code. To assist developers, we propose the BugNet architecture to continuously record information on production runs. The information collected before the crash of a program can be used by the developers working in their execution environment to deterministically replay the last several million instructions executed before the crash. BugNet is based on the insight that recording the register file contents at any point in time, and then recording the load values that occur after that point can enable deterministic replaying of a programýs execution. BugNet focuses on being able to replay the applicationýs execution and the libraries it uses, but not the operating system. But our approach provides the ability to replay an applicationýs execution across context switches and interrupts. Hence, BugNet obviates the need for tracking program I/O, interrupts and DMA transfers, which would have otherwise required more complex hardware support. In addition, BugNet does not require a final core dump of the system state for replaying, which significantly reduces the amount of data that must be sent back to the developer.

HISC (High-level Instruction Set Computer) is proposed to provide hardware support for OOP. It applies the concept of object-orientation in the architecture and provides special instructions for object manipulation. An operating system is under developing for HISC. It divided into several modules by using the object-oriented paradigm. It includes some special modules for object-manipulation such as object management and garbage collection which traditional operating system do not support.In order to do the functional verification of each module, a test bench is needed. Its aims are to ease the effort to written the testing program and to simulate the working environment. It can make it easier to evaluate if an algorithm is suitable for HISC or not. In this paper, we discuss the design of the test bench. The architecture of the system will be described, together with the implementation of the modules and the pluggable modules architecture.

In Object-Oriented Programming (OOP), programmers divide the software applications into small objects, each of which is responsible for performing part of the work. The communications between different objects are through method invocation. Method manipulation is taking a big role in an OOP system. The performance of method invocation would greatly affect the performance of an OOP system. To speed up method communication, an object-oriented computer architecture is proposed at hardware level. High Level Instruction Set Computer (HISC) is one of the OOP architectures proposed. It applies hardware readable data structures to provide mapping of OOP features directly from architectural level. One of the architectural designs that use the concept of HISC is jHISC v3, which is currently under development. It targets to support Java at architecture level.In this paper, we will discuss the method manipulation procedures in jHISC v3 system. By defining the hardware readable data structures for method context, we will see how jHISC v3 processor can use such information to provide a secure method manipulation mechanism for OOP.

Current-generation microprocessors are designed to process instructions with one and two source operands at equal cost. Handling two source operands requires multiple ports for each instruction in structures--such as the register file and wakeup logic--which are often in the processor's critical timing paths. We argue that these structures are overdesigned since only a small fraction of instructions require two source operands to be processed simultaneously. In this paper, we propose the half-price architecture that judiciously removes this overdesign by restricting the processor's capability to handle two source operands in certain timing-critical cases. Two techniques are proposed and evaluated: one for the wakeup logic is sequential wakeup, which decouples half of the tag matching logic from the wakeup bus to reduce the load capacitance of the bus. The other technique for the register file is sequential register access, which halves the register read ports by sequentially accessing two values using a single port when needed. We show that a pipeline that optimizes scheduling and register access for a single operand achieves nearly the same performance as an ideal base machine that fully handles two operands, with 2.2% (worst case 4.8%) IPC degradation.

Dynamic Instruction Stream Editing (DISE) is a cooperative software-hardware scheme for efficiently adding customization functionality---e.g, safety/security checking, profiling, dynamic code decompression, and dynamic optimization---to an application. In DISE, application customization functions (ACFs) are formulated as rules for macro-expanding certain instructions into parameterized instruction sequences. The processor executes the rules on the fetched instructions, feeding the execution engine an instruction stream that contains ACF code. Dynamic instruction macro-expansion is widely used in many of today's processors to convert a complex ISA to an easier-to-execute, finer-grained internal form. DISE co-opts this technology and adds a programming interface to it.DISE unifies the implementation of a large class of ACFs that would otherwise require either special-purpose hardware widgets or static binary rewriting. We show DISE implementations of two ACFs---memory fault isolation and dynamic code decompression---and their composition. Simulation shows that DISE ACFs have better performance than their software counterparts, and more flexibility (which sometimes translates into performance) than hardware implementations.

Trends in microprocessors of increasing die size and clock speed and decreasing feature sizes have fueled rapidly increasing performance. However, the limited improvements in DRAM latency and bandwidth and diminishing returns of increasing superscalar ILP and cache sizes have led to the proposal of new microprocessor architectures that implement processor-in- memory, stream processing, and tiled processing. Each architecture is typically evaluated separately and compared to a baseline architecture. In this paper, we evaluate the performance of processors that implement these architectures on a common set of signal processing kernels.The implementation results are compared with the measured performance of a conventional system based on the PowerPC with Altivec. The results show that these new processors show significant improvements over conventional systems and that each architecture has its own strengths and weaknesses.

Tarantula is an aggressive floating point machine targeted at technical, scientific and bioinformatics workloads, originally planned as a follow-on candidate to the EV8 processor [6, 5]. Tarantula adds to the EV8 core a vector unit capable of 32 double-precision flops per cycle. The vector unit fetches data directly from a 16 MByte second level cache with a peak bandwidth of sixty four 64-bit values per cycle. The whole chip is backed by a memory controller capable of delivering over 64 GBytes/s of raw band- width. Tarantula extends the Alpha ISA with new vector instructions that operate on new architectural state. Salient features of the architecture and implementation are: (1) it fully integrates into a virtual-memory cache-coherent system without changes to its coherency protocol, (2) provides high bandwidth for non-unit stride memory accesses, (3) supports gather/scatter instructions efficiently, (4) fully integrates with the EV8 core with a narrow, streamlined interface, rather than acting as a co-processor, (5) can achieve a peak of 104 operations per cycle, and (6) achieves excellent "real-computation" per transistor and per watt ratios. Our detailed simulations show that Tarantula achieves an average speedup of 5X over EV8, out of a peak speedup in terms of flops of 8X. Furthermore, performance on gather/scatter intensive benchmarks such as Radix Sort is also remarkable: a speedup of almost 3X over EV8 and 15 sustained operations per cycle. Several benchmarks exceed 20 operations per cycle.

Increasing the number of instruction queue (IQ) entries in a dynamically scheduled processor exposes more instruction-level parallelism, leading to higher performance. However, increasing a conventional IQ's physical size leads to larger latencies and slower clock speeds. We introduce a new IQ design that divides a large queue into small segments, which can be clocked at high frequencies. We use dynamic dependence-based scheduling to promote instructions from segment to segment until they reach a small issue buffer. Our segmented IQ is designed specifically to accommodate variable-latency instructions such as loads. Despite its roughly similar circuit complexity; simulation results indicate that our segmented instruction queue with 512 entries and 128 chains improves performance by up to 69% over a 32-entry conventional instruction queue for SpecINT 2000 benchmarks, and up to 398% for SpecFP 2000 benchmarks. The segmented IQ achieves from 55% to 98% of the performance of a monolithic 512-entry queue while providing the potential for much higher clock speeds.

An instruction set architecture (ISA) suitable for future microprocessor design constraints is proposed. The ISA has hierarchical register files with a small number of accumulators at the top. The instruction stream is divided into chains of dependent instructions (strands) where intra-strand dependences are passed through the accumulator. The general-purpose register file is used for communication between strands and for holding global values that have many consumers.A microarchitecture to support the proposed ISA is proposed and evaluated. The microarchitecture consists of multiple, distributed processing elements. Each PE contains an instruction issue FIFO, a local register (accumulator) and local copy of register file. The overall simplicity, hierarchical value communication, and distributed implementation will provide a very high clock speed and a relatively short pipeline while maintaining a form of superscalar out-of-order execution.Detailed timing simulations using translated program traces show the proposed microarchitecture is tolerant of global wire latencies. Ignoring the significant clock frequency advantages, a microarchitecture that supports a 4-wide fetch/decode pipeline, 8 serial PEs, and a two-cycle inter-PE communication latency performs as well as a conventional 4-way out-of-order superscalar processor.

The number of pipeline stages separating dynamic instruction scheduling from instruction execution has increased considerably in recent out-of-order microprocessor implementations, forcing the scheduler to allocate functional units and other execution resources several cycles before they are actually used. Unfortunately, several proposed microarchitectural optimizations become less desirable or even impossible in such an environment, since they require instantaneous or near-instantaneous changes in execution behavior and resource usage in response to dynamic events that occur during instruction execution. Since they are detected several cycles after scheduling decisions have already been made, such dynamic responses are infeasible. To overcome this limitation, we propose to implement optimizations by performing what we call speculative decode. Speculative decode alters the mapping between user-visible instructions and the implemented core instructions based on observed runtime characteristics and generates speculative instruction sequences. In these sequences, optimizations are pre-scheduled in a manner compatible with realistic pipelines with multicycle scheduling latency. We present case studies on memory reference combining and silent store squashing, and demonstrate that speculative decode performs comparably or even better than impractical in-core implementations that require zero-cycle scheduling latency.

We present the design of a PowerPC-based simulation infrastructure for architectural research. Our infrastructure uses an execution-driven out-of-order processor timing simulator from the SimpleScalar tool set. While porting SimpleScalar to the PowerPC architecture, we would like to remain compatible with other versions of SimpleScalar. We accomplish this by performing dynamic binary translation of the PowerPC instruction set architecture to the SimpleScalar instruction set architecture, and by mapping the PowerPC architectural state onto the SimpleScalar register set. Using this infrastructure, we execute unmodified PowerPC binaries on an out-of-order processor timing simulator which implements the SimpleScalar architecture. We describe and investigate trade-offs in the translation of some complex PowerPC instructions and advocate adoption of speculative decode to optimize instruction translations for the common case. We find that simple decode predictors can reach better than 90% accuracy for guiding speculative decode.

Vector instruction sets are receiving renewed interest because of their applicability to multimedia. Current multimedia instruction sets use short vectors with SIMD implementations, but long vector, pipelined implementations have a number of advantages and arc a logical next step in multimedia ISA development. Support for conditional operations (as occur in loops containing IF statements) is an important aspect of a vector ISA. Seven ISA alternatives for implementing conditional operations are systematically explored. Performance considerations are discussed through evaluation of a typical IF loop over a range of vector lengths and true conditional values. An approach using masked operations is shown to be one of the better methods, especially if its implementation is able to skip over blocks of false mask bits. Additional analyses of complex IF loops and parallel pipeline implementations support the masked operation approach. The paper concludes with a practical implementation of masked operations that skips over power-of-2-length blocks of false values. This implementation is simpler than skipping arbitrary-length blocks and provides similar performance.

A program’s execution profile is an increasingly important source of information for optimizations. Along with its use for high-level optimizations, profile information can be used to take advantage of advanced ISA features such as branch hint bits. Profile information is collected either by instrumenting an executable with profile collection code or by running the program in a simulator. Instrumentation has the advantage that it is relatively fast, with the disadvantage that the instrumentation code can interfere with the measurements (e.g. branch prediction). Traditional simulation is slower but a more accurate process. However, since most simulators simulate the low level machine code of a program, relating the results to the high-level program structure can be problematic. MirvSim differs from other simulators – such as SimpleScalar – that are used in computer architecture in that instead of simulating a machine code level instruction set, it simulates the Mirv language, a high-level intermediate program representation (IR). This allows the simulator to be closely coupled with the rest of the compiler infrastructure and to communicate information with other compiler passes easily.

The architecture described in the following articel is a direct successor of µ-EP-1 (cf. [1]) and was developed by the author and Robert Linden (Universität Bonn, FB Informatik). NICE is a 32-bit processor, utilizing a fixed instruction format, a register set of sixteen general purpose registers and sat extremely powerful but simple and easy to use instruction set which is supported by a variety of addressing modes. The smallest direct addressable data item in main memory is the 32-bit machine word thus utilizing a simple addressing scheme for instruction operands.The aim of this design is reflected by the acronym "NICE" which means "NICE is charmingly elegant" (Robert Linden).At the moment NICE contains neither any hardware support for memory management nor floating point instructions, but using the instruction selection scheme described below features like these can be added without changing the overall machine design.

Improvements in main memory speeds have not kept pace with increasing processor clock frequency and improved exploitation of instruction-level parallelism. Consequently, the gap between processor and main memory performance is expected to grow, increasing the number of execution cycles spent waiting for memory accesses to complete. One solution to this growing problem is to reduce the number of cache misses by increasing the effectiveness of the cache hierarchy. In this paper we present a technique for dynamic analysis of program data access behavior, which is then used to proactively guide the placement of data within the cache hierarchy in a location-sensitive manner. We introduce the concept of a macroblock, which allows us to feasibly characterize the memory locations accessed by a program, and a Memory Address Table, which performs the dynamic reference analysis. Our technique is fully compatible with existing Instruction Set Architectures. Results from detailed simulations of several integer programs show significant speedups.

This work presents an academic RISC processor architecture, named DMN-6 that executes every instruction in the datapath. It concentrates all the movement, branch and alu instructions in the arithmetic-logic unit. The idea is to normalize the control signal generation for an integer functional unit. This is obtained by implementing a number of queue registers of different deepness around the datapath unit. These queues will control an assigned logic corresponding to a stage in the pipeline. The architecture reduces even more the complexity of a pipelined program execution. The main features are : Load/Store architecture, 4-stage pipeline, integer arithmetic, sixteen byte registers, internal separate data and instruction main memories, thirteen 16-bit instruction words.

Previous research has shown that the SPEC benchmarks achieve low miss ratios in relatively small instruction caches. This paper presents evidence that current software-development practices produce applications that exhibit substantially higher instruction-cache miss ratios than do the SPEC benchmarks. To represent these trends, we have assembled a collection of applications, called the Instruction Benchmark Suite (IBS), that provides a better test of instruction-cache performance. We discuss the rationale behind the design of IBS and characterize its behavior relative to the SPEC benchmark suite. Our analysis is based on trace-driven and trap-driven simulations and takes into full account both the application and operating-system components of the workloads.This paper then reexamines a collection of previously-proposed hardware mechanisms for improving instruction-fetch performance in the context of the IBS workloads. We study the impact of cache organization, transfer bandwidth, prefetching, and pipelined memory systems on machines that rely on the use of relatively small primary instruction caches to facilitate increased clock rates. We find that, although of little use for SPEC, the right combination of these techniques substantially benefits IBS. Even so, under IBS, a stubborn lower bound on the instruction-fetch CPI remains as an obstacle to improving overall processor performance.

One can effectively utilize predicated execution to improve branch handling in instruction-level parallel processors. Although the potential benefits of predicated execution are high, the tradeoffs involved in the design of an instruction set to support predicated execution can be difficult. On one end of the design spectrum, architectural support for full predicated execution requires increasing the number of source operands for all instructions. Full predicate support provides for the most flexibility and the largest potential performance improvements. On the other end, partial predicated execution support, such as conditional moves, requires very little change to existing architectures. This paper presents a preliminary study to qualitatively and quantitatively address the benefit of full and partial predicated execution support. With our current compiler technology, we show that the compiler can use both partial and full predication to achieve speedup in large control-intensive programs. Some details of the code generation techniques are shown to provide insight into the benefit of going from partial to full predication. Preliminary experimental results are very encouraging: partial predication provides an average of 33% performance improvement for an 8-issue processor with no predicate support while full predication provides an additional 30% improvement.

Since 1964, we had the notion that the instruction set architecture (ISA) is a useful and fairly opaque abstraction layer between hardware and software. Software rode hardware's performance wave while remaining gloriously oblivious to hardware's growing complexity. Unfortunately, the jig is up. We still have ISAs, but the abstraction no longer offers seamless portability---parallel software needs to be tuned for different core counts, and heterogeneous processing elements (CPUs, GPUs, accelerators) further complicate programmability. We are better at building large-scale heterogeneous processors than we are at programming them. Maintaining software across multiple current platforms is difficult and porting to future platforms is also difficult. There have been many technical responses: virtual ISAs (e.g., NVIDIA's PTX), higher-level programming interfaces (e.g., CUDA or OpenCL), and late-stage compilation and platform-specific tailoring (e.g., Android ART), etc. A team of opinionated experts, drawn from the three ASPLOS communities will examine the problem of programmer productivity in the post-ISA world, first from the perspective of their area of expertise and then noting the contributions from the other two communities. What research will save us and how? This wide-ranging debate will frame important research areas for future work while being grounded in frank discussion about what has succeeded in the past. Attendees can expect actionable insight into important research issues as well an entertaining discussion.

The demand of increasing computation power for symbolic processing has given a strong impulse to the development of ASICs dedicated to the execution of prolog. Unlike past microcoded implementation based on the Warren machine model, novel trends in high performance Prolog processors suggest the implementation of RISC-based processors committed to prolog only through the adoption of a few basic dedicated features, like the Berkeley Abstract Machine (BAM) processor. Following the idea of using a smart compiler for a simple instruction set, the SYMBOL project represents an experiment in applying global compaction techniques and VLIW design philosophy to the static exploitation of instruction-level parallelism in Prolog. This paper presents code analysis results and shows how we can approach the theoretical speed-up limit (about 3) imposed by Amdahl's law on shared memory models, by means of global code optimizations and a suitable architectural support. In addition, we show implementation details and some preliminary data of a VLSI prototype architecture.

Although they are powerful intermediate representations for compilers, pure dataflow graphs are incomplete, and perhaps even undesirable, machine languages. They are incomplete because it is hard to encode critical sections and imperative operations which are essential for the efficient execution of operating system functions, such as resource management. They may be undesirable because they imply a uniform dynamic scheduling policy for all instructions, preventing a compiler from expressing a static schedule which could result in greater run time efficiency, both by reducing redundant operand synchronization, and by using high speed registers to communicate state between instructions. In this paper, we develop a new machine-level programming model which builds upon two previous improvements to the dataflow execution model: sequential scheduling of instructions, and multiported registers for expression temporaries. Surprisingly, these improvements have required almost no architectural changes to explicit token store (ETS) dataflow hardware, only a shift in mindset when reasoning about how that hardware works. Rather than viewing computational progress as the consumption of tokens and the firing of enabled instructions, we instead reason about the evolution of multiple, interacting sequential threads, where forking and joining are extremely efficient. Because this new paradigm has proven so valuable in coding resource management operations and in improving code efficiency, it is now the cornerstone of the Monsoon instruction set architecture and macro assembly language. In retrospect, this suggests that there is a continuum of multithreaded architectures, with pure ETS dataflow and single threaded von Neumann at the extrema. We use this new perspective to better understand the relative strengths and weaknesses of the Monsoon implement ation.

A single processing computer system using only half-adder circuits is described. In addition, it is shown that only a single hard-wired instruction is needed in the control unit to obtain a complete [1] instruction set for this general purpose computer. Such a system has several advantages. First it is intrinsically a RISC machine-in fact the "ultimate RISC" machine. Second, because only a single type of logic element is employed the entire computer system can be easily realized on a single, highly integrated chip. Finally, due to the homogeneous nature of the computer's logic elements, the computer has possible implementations as an optical or chemical machine. This in turn suggests possible paradigms for neural computing and artificial intelligence. This approach was inspired by a paper by van der Poel [5] in which a single instruction is used to furnish a complete set [1] of general purpose instructions.

This paper describes the architecture for issuing multiple instructions per clock in the NonStop Cyclone Processor. Pairs of instructions are fetched and decoded by a dual two-stage prefetch pipeline and passed to a dual six-stage pipeline for execution. Dynamic branch prediction is used to reduce branch penalties. A unique microcode routine for each pair is stored in the large duplexed control store. The microcode controls parallel data paths optimized for executing the most frequent instruction pairs. Other features of the architecture include cache support for unaligned double-precision accesses, a virtually-addressed main memory, and a novel precise exception mechanism.