Register Allocation and Optimal Spill code Scheduling in Software Pipelined Loops using 0-1 Integer Linear Programming Formulation

In achieving higher instruction level parallelism, software pipelining increases the register pressure in the loop. The usefulness of the generated schedule may be restricted to cases where the register pressure is less than the available number of registers. Spill instructions need to be introduced otherwise. But scheduling these spill instructions in the compact schedule is a difficult task. Several heuristics have been proposed to schedule spill code. These heuristics may generate more spill code than necessary, and scheduling them may necessitate increasing the initiation interval. We model the problem of register allocation with spill code generation and scheduling in software pipelined loops as a 0-1 integer linear program. The formulation minimizes the increase in initiation interval (II) by optimally placing spill code and simultaneously minimizes the amount of spill code produced. To the best of our knowledge, this is the first integrated formulation for register allocation, optimal spill code generation and scheduling for software pipelined loops. The proposed formulation performs better than the existing heuristics by preventing an increase in II in 11.11% of the loops and generating 18.48% less spill code on average among the loops extracted from Perfect Club and SPEC benchmarks with a moderate increase in compilation time.

Partial Flow Sensitivity

Compiler optimizations need precise and scalable analyses to discover program properties. We propose a partially flow-sensitive framework that tries to draw on the scalability of flow-insensitive algorithms while providing more precision at some specific program points. Provided with a set of critical nodes — basic blocks at which more precise information is desired — our partially flow-sensitive algorithm computes a reduced control-flow graph by collapsing some sets of non-critical nodes. The algorithm is more scalable than a fully flow-sensitive one as, assuming that the number of critical nodes is small, the reduced flow-graph is much smaller than the original flow-graph. At the same time, a much more precise information is obtained at certain program points than would had been obtained from a flow-insensitive algorithm.

Variable Granularity Access Tracking Scheme for Improving the Performance of Software Transactional Memory

Software transactional memory (STM) has been proposed as a promising programming paradigm for shared memory multi-threaded programs as an alternative to conventional lock based synchronization primitives. Typical STM implementations employ a conflict detection scheme, which works with uniform access granularity, tracking shared data accesses either at word/cache line or at object level. It is well known that a single fixed access tracking granularity cannot meet the conflicting goals of reducing false conflicts without impacting concurrency adversely. A fine grained granularity while improving concurrency can have an adverse impact on performance due to lock aliasing, lock validation overheads, and additional cache pressure. On the other hand, a coarse grained granularity can impact performance due to reduced concurrency. Thus, in general, a fixed or uniform granularity access tracking (UGAT) scheme is application-unaware and rarely matches the access patterns of individual application or parts of an application, leading to sub-optimal performance for different parts of the application(s). In order to mitigate the disadvantages associated with UGAT scheme, we propose a Variable Granularity Access Tracking (VGAT) scheme in this paper. We propose a compiler based approach wherein the compiler uses inter-procedural whole program static analysis to select the access tracking granularity for different shared data structures of the application based on the application's data access pattern. We describe our prototype VGAT scheme, using TL2 as our STM implementation. Our experimental results reveal that VGAT-STM scheme can improve the application performance of STAMP benchmarks from 1.87% to up to 21.2%.

Enhancing Speedup in Network Processing Applications by Exploiting Instruction Reuse with Flow Aggregation

Instruction reuse is a microarchitectural technique that improves the execution time of a program by removing redundant computations at run-time. Although this is the job of an optimizing compiler, they do not succeed many a time due to limited knowledge of run-time data. In this paper we examine instruction reuse of integer ALU and load instructions in network processing applications. Specifically, this paper attempts to answer the following questions: (1) How much of instruction reuse is inherent in network processing applications?, (2) Can reuse be improved by reducing interference in the reuse buffer?, (3) What characteristics of network applications can be exploited to improve reuse?, and (4) What is the effect of reuse on resource contention and memory accesses? We propose an aggregation scheme that combines the high-level concept of network traffic i.e. "flows" with a low level microarchitectural feature of programs i.e. repetition of instructions and data along with an architecture that exploits temporal locality in incoming packet data to improve reuse. We find that for the benchmarks considered, 1% to 50% of instructions are reused while the speedup achieved varies between 1% and 24%. As a side effect, instruction reuse reduces memory traffic and can therefore be considered as a scheme for low power.

Exploiting the Structure of the Constraint Graph for Efficient Points-to Analysis

Points-to analysis is a key compiler analysis. Several memory related optimizations use points-to information to improve their effectiveness. Points-to analysis is performed by building a constraint graph of pointer variables and dynamically updating it to propagate more and more points-to information across its subset edges. So far, the structure of the constraint graph has been only trivially exploited for efficient propagation of information, e.g., in identifying cyclic components or to propagate information in topological order. We perform a careful study of its structure and propose a new inclusion-based flow-insensitive context-sensitive points-to analysis algorithm based on the notion of dominant pointers. We also propose a new kind of pointer-equivalence based on dominant pointers which provides significantly more opportunities for reducing the number of pointers tracked during the analysis. Based on this hitherto unexplored form of pointer-equivalence, we develop a new context-sensitive flow-insensitive points-to analysis algorithm which uses incremental dominator update to efficiently compute points-to information. Using a large suite of programs consisting of SPEC 2000 benchmarks and five large open source programs we show that our points-to analysis is 88% faster than BDD-based Lazy Cycle Detection and 2x faster than Deep Propagation. We argue that our approach of detecting dominator-based pointer-equivalence is a key to improve points-to analysis efficiency.

Reuse, recycle to de-bloat software

Most Java programmers would agree that Java is a language that promotes a philosophy of “create and go forth”. By design, temporary objects are meant to be created on the heap, possibly used and then abandoned to be collected by the garbage collector. Excessive generation of temporary objects is termed “object churn” and is a form of software bloat that often leads to performance and memory problems. To mitigate this problem, many compiler optimizations aim at identifying objects that may be allocated on the stack. However, most such optimizations miss large opportunities for memory reuse when dealing with objects inside loops or when dealing with container objects. In this paper, we describe a novel algorithm that detects bloat caused by the creation of temporary container and String objects within a loop. Our analysis determines which objects created within a loop can be reused. Then we describe a source-to-source transformation that efficiently reuses such objects. Empirical evaluation indicates that our solution can reduce upto 40% of temporary object allocations in large programs, resulting in a performance improvement that can be as high as a 20% reduction in the run time, specifically when a program has a high churn rate or when the program is memory intensive and needs to run the GC often.

Automatic compilation of MATLAB programs for synergistic execution on heterogeneous processors

MATLAB is an array language, initially popular for rapid prototyping, but is now being increasingly used to develop production code for numerical and scientific applications. Typical MATLAB programs have abundant data parallelism. These programs also have control flow dominated scalar regions that have an impact on the program's execution time. Today's computer systems have tremendous computing power in the form of traditional CPU cores and throughput oriented accelerators such as graphics processing units(GPUs). Thus, an approach that maps the control flow dominated regions to the CPU and the data parallel regions to the GPU can significantly improve program performance. In this paper, we present the design and implementation of MEGHA, a compiler that automatically compiles MATLAB programs to enable synergistic execution on heterogeneous processors. Our solution is fully automated and does not require programmer input for identifying data parallel regions. We propose a set of compiler optimizations tailored for MATLAB. Our compiler identifies data parallel regions of the program and composes them into kernels. The problem of combining statements into kernels is formulated as a constrained graph clustering problem. Heuristics are presented to map identified kernels to either the CPU or GPU so that kernel execution on the CPU and the GPU happens synergistically and the amount of data transfer needed is minimized. In order to ensure required data movement for dependencies across basic blocks, we propose a data flow analysis and edge splitting strategy. Thus our compiler automatically handles composition of kernels, mapping of kernels to CPU and GPU, scheduling and insertion of required data transfer. The proposed compiler was implemented and experimental evaluation using a set of MATLAB benchmarks shows that our approach achieves a geometric mean speedup of 19.8X for data parallel benchmarks over native execution of MATLAB.

Tiling stencil computations to maximize parallelism

Most stencil computations allow tile-wise concurrent start, i.e., there always exists a face of the iteration space and a set of tiling hyperplanes such that all tiles along that face can be started concurrently. This provides load balance and maximizes parallelism. However, existing automatic tiling frameworks often choose hyperplanes that lead to pipelined start-up and load imbalance. We address this issue with a new tiling technique that ensures concurrent start-up as well as perfect load-balance whenever possible. We first provide necessary and sufficient conditions on tiling hyperplanes to enable concurrent start for programs with affine data accesses. We then provide an approach to find such hyperplanes. Experimental evaluation on a 12-core Intel Westmere shows that our code is able to outperform a tuned domain-specific stencil code generator by 4% to 27%, and previous compiler techniques by a factor of 2x to 10.14x.

CUDA-for-clusters: a system for efficient execution of CUDA kernels on multi-core clusters

Rapid advancements in multi-core processor architectures coupled with low-cost, low-latency, high-bandwidth interconnects have made clusters of multi-core machines a common computing resource. Unfortunately, writing good parallel programs that efficiently utilize all the resources in such a cluster is still a major challenge. Various programming languages have been proposed as a solution to this problem, but are yet to be adopted widely to run performance-critical code mainly due to the relatively immature software framework and the effort involved in re-writing existing code in the new language. In this paper, we motivate and describe our initial study in exploring CUDA as a programming language for a cluster of multi-cores. We develop CUDA-For-Clusters (CFC), a framework that transparently orchestrates execution of CUDA kernels on a cluster of multi-core machines. The well-structured nature of a CUDA kernel, the growing popularity, support and stability of the CUDA software stack collectively make CUDA a good candidate to be considered as a programming language for a cluster. CFC uses a mixture of source-to-source compiler transformations, a work distribution runtime and a light-weight software distributed shared memory to manage parallel executions. Initial results on running several standard CUDA benchmark programs achieve impressive speedups of up to 7.5X on a cluster with 8 nodes, thereby opening up an interesting direction of research for further investigation.

Automatic Data Allocation and Buffer Management for Multi-GPU Machines

Multi-GPU machines are being increasingly used in high-performance computing. Each GPU in such a machine has its own memory and does not share the address space either with the host CPU or other GPUs. Hence, applications utilizing multiple GPUs have to manually allocate and manage data on each GPU. Existing works that propose to automate data allocations for GPUs have limitations and inefficiencies in terms of allocation sizes, exploiting reuse, transfer costs, and scalability. We propose a scalable and fully automatic data allocation and buffer management scheme for affine loop nests on multi-GPU machines. We call it the Bounding-Box-based Memory Manager (BBMM). BBMM can perform at runtime, during standard set operations like union, intersection, and difference, finding subset and superset relations on hyperrectangular regions of array data (bounding boxes). It uses these operations along with some compiler assistance to identify, allocate, and manage data required by applications in terms of disjoint bounding boxes. This allows it to (1) allocate exactly or nearly as much data as is required by computations running on each GPU, (2) efficiently track buffer allocations and hence maximize data reuse across tiles and minimize data transfer overhead, and (3) and as a result, maximize utilization of the combined memory on multi-GPU machines. BBMM can work with any choice of parallelizing transformations, computation placement, and scheduling schemes, whether static or dynamic. Experiments run on a four-GPU machine with various scientific programs showed that BBMM reduces data allocations on each GPU by up to 75% compared to current allocation schemes, yields performance of at least 88% of manually written code, and allows excellent weak scaling.

Generating Efficient Data Movement Code for Heterogeneous Architectures with Distributed-Memory

Programming for parallel architectures that do not have a shared address space is extremely difficult due to the need for explicit communication between memories of different compute devices. A heterogeneous system with CPUs and multiple GPUs, or a distributed-memory cluster are examples of such systems. Past works that try to automate data movement for distributed-memory architectures can lead to excessive redundant communication. In this paper, we propose an automatic data movement scheme that minimizes the volume of communication between compute devices in heterogeneous and distributed-memory systems. We show that by partitioning data dependences in a particular non-trivial way, one can generate data movement code that results in the minimum volume for a vast majority of cases. The techniques are applicable to any sequence of affine loop nests and works on top of any choice of loop transformations, parallelization, and computation placement. The data movement code generated minimizes the volume of communication for a particular configuration of these. We use a combination of powerful static analyses relying on the polyhedral compiler framework and lightweight runtime routines they generate, to build a source-to-source transformation tool that automatically generates communication code. We demonstrate that the tool is scalable and leads to substantial gains in efficiency. On a heterogeneous system, the communication volume is reduced by a factor of 11X to 83X over state-of-the-art, translating into a mean execution time speedup of 1.53X. On a distributed-memory cluster, our scheme reduces the communication volume by a factor of 1.4X to 63.5X over state-of-the-art, resulting in a mean speedup of 1.55X. In addition, our scheme yields a mean speedup of 2.19X over hand-optimized UPC codes.

Parallel Flow-Sensitive Pointer Analysis by Graph-Rewriting

Precise pointer analysis is a problem of interest to both the compiler and the program verification community. Flow-sensitivity is an important dimension of pointer analysis that affects the precision of the final result computed. Scaling flow-sensitive pointer analysis to millions of lines of code is a major challenge. Recently, staged flow-sensitive pointer analysis has been proposed, which exploits a sparse representation of program code created by staged analysis. In this paper we formulate the staged flow-sensitive pointer analysis as a graph-rewriting problem. Graph-rewriting has already been used for flow-insensitive analysis. However, formulating flow-sensitive pointer analysis as a graph-rewriting problem adds additional challenges due to the nature of flow-sensitivity. We implement our parallel algorithm using Intel Threading Building Blocks and demonstrate considerable scaling (upto 2.6x) for 8 threads on a set of 10 benchmarks. Compared to the sequential implementation of staged flow-sensitive analysis, a single threaded execution of our implementation performs better in 8 of the benchmarks.

PolyGLoT: A Polyhedral Loop Transformation Framework for a Graphical Dataflow Language

Polyhedral techniques for program transformation are now used in several proprietary and open source compilers. However, most of the research on polyhedral compilation has focused on imperative languages such as C, where the computation is specified in terms of statements with zero or more nested loops and other control structures around them. Graphical dataflow languages, where there is no notion of statements or a schedule specifying their relative execution order, have so far not been studied using a powerful transformation or optimization approach. The execution semantics and referential transparency of dataflow languages impose a different set of challenges. In this paper, we attempt to bridge this gap by presenting techniques that can be used to extract polyhedral representation from dataflow programs and to synthesize them from their equivalent polyhedral representation. We then describe PolyGLoT, a framework for automatic transformation of dataflow programs which we built using our techniques and other popular research tools such as Clan and Pluto. For the purpose of experimental evaluation, we used our tools to compile LabVIEW, one of the most widely used dataflow programming languages. Results show that dataflow programs transformed using our framework are able to outperform those compiled otherwise by up to a factor of seventeen, with a mean speed-up of 2.30x while running on an 8-core Intel system.

Taming Control Divergence in GPUs through Control Flow Linearization

Branch divergence is a very commonly occurring performance problem in GPGPU in which the execution of diverging branches is serialized to execute only one control flow path at a time. Existing hardware mechanism to reconverge threads using a stack causes duplicate execution of code for unstructured control flow graphs. Also the stack mechanism cannot effectively utilize the available parallelism among diverging branches. Further, the amount of nested divergence allowed is also limited by depth of the branch divergence stack. In this paper we propose a simple and elegant transformation to handle all of the above mentioned problems. The transformation converts an unstructured CFG to a structured CFG without duplicating user code. It incurs only a linear increase in the number of basic blocks and also the number of instructions. Our solution linearizes the CFG using a predicate variable. This mechanism reconverges the divergent threads as early as possible. It also reduces the depth of the reconvergence stack. The available parallelism in nested branches can be effectively extracted by scheduling the basic blocks to reduce the effect of stalls due to memory accesses. It can also increase execution efficiency of nested loops with different trip counts for different threads. We implemented the proposed transformation at PTX level using the Ocelot compiler infrastructure. We evaluated the technique using various benchmarks to show that it can be effective in handling the performance problem due to divergence in unstructured CFGs.

PolyMage: Automatic Optimization for Image Processing Pipelines

This paper presents the design and implementation of PolyMage, a domain-specific language and compiler for image processing pipelines. An image processing pipeline can be viewed as a graph of interconnected stages which process images successively. Each stage typically performs one of point-wise, stencil, reduction or data-dependent operations on image pixels. Individual stages in a pipeline typically exhibit abundant data parallelism that can be exploited with relative ease. However, the stages also require high memory bandwidth preventing effective utilization of parallelism available on modern architectures. For applications that demand high performance, the traditional options are to use optimized libraries like OpenCV or to optimize manually. While using libraries precludes optimization across library routines, manual optimization accounting for both parallelism and locality is very tedious. The focus of our system, PolyMage, is on automatically generating high-performance implementations of image processing pipelines expressed in a high-level declarative language. Our optimization approach primarily relies on the transformation and code generation capabilities of the polyhedral compiler framework. To the best of our knowledge, this is the first model-driven compiler for image processing pipelines that performs complex fusion, tiling, and storage optimization automatically. Experimental results on a modern multicore system show that the performance achieved by our automatic approach is up to 1.81x better than that achieved through manual tuning in Halide, a state-of-the-art language and compiler for image processing pipelines. For a camera raw image processing pipeline, our performance is comparable to that of a hand-tuned implementation.