Monolithic large language models (LLMs) like GPT-4 have paved the way for modern generative AI applications. Training, serving, and maintaining monolithic LLMs at scale, however, remains prohibitively expensive and challenging. The disproportionate increase in compute-to-memory ratio of modern AI accelerators have created a memory wall, necessitating new methods to deploy AI. Recent research has shown that a composition of many smaller expert models, each with several orders of magnitude fewer parameters, can match or exceed the capabilities of monolithic LLMs. Composition of Experts (CoE) is a modular approach that lowers the cost and complexity of training and serving. However, this approach presents two key challenges when using conventional hardware: (1) without fused operations, smaller models have lower operational intensity, which makes high utilization more challenging to achieve; and (2) hosting a large number of models can be either prohibitively expensive or slow when dynamically switching between them. In this paper, we describe how combining CoE, streaming dataflow, and a three-tier memory system scales the AI memory wall. We describe Samba-CoE, a CoE system with 150 experts and a trillion total parameters. We deploy Samba-CoE on the SambaNova SN40L Reconfigurable Dataflow Unit (RDU) -a commercial dataflow accelerator architecture that has been codesigned for enterprise inference and training applications. The chip introduces a new three-tier memory system with on-chip distributed SRAM, on-package HBM, and off-package DDR DRAM. A dedicated inter-RDU network enables scaling up and out over multiple sockets. We demonstrate speedups ranging from 2× to 13× on various benchmarks running on eight RDU sockets compared with an unfused baseline. We show that for CoE inference deployments, the 8-socket RDU Node reduces machine footprint by up to 19 ×, speeds up model switching time by 15× to 31×, and achieves an overall speedup of 3.7× over a DGX H100 and 6.6× over a DGX A100.

Symbolic parallelism is a fresh look at the decade-old problem of turning sequential, imperative, code into associative reduction kernels, based on the realization that map/reduce is at its core a staging problem: how can programs be separated so that part of the computation can be done before loop-carried dependencies become available? Previous work has investigated dynamic approaches that build symbolic summaries while the actual data is processed. In this short paper, we approach the problem from the static side, and show that with simple syntax- or type-driven transformations, we can readily transform large classes of imperative groupby-aggregate programs into map/reduce parallelism with deterministic overhead.

In recent years the computing landscape has seen an increasing shift towards specialized accelerators. Field programmable gate arrays (FPGAs) are particularly promising for the implementation of these accelerators, as they offer significant performance and energy improvements over CPUs for a wide class of applications and are far more flexible than fixed-function ASICs. However, FPGAs are difficult to program. Traditional programming models for reconfigurable logic use low-level hardware description languages like Verilog and VHDL, which have none of the productivity features of modern software languages but produce very efficient designs, and low-level software languages like C and OpenCL coupled with high-level synthesis (HLS) tools that typically produce designs that are far less efficient. Functional languages with parallel patterns are a better fit for hardware generation because they provide high-level abstractions to programmers with little experience in hardware design and avoid many of the problems faced when generating hardware from imperative languages. In this paper, we identify two important optimizations for using parallel patterns to generate efficient hardware: tiling and metapipelining. We present a general representation of tiled parallel patterns, and provide rules for automatically tiling patterns and generating metapipelines. We demonstrate experimentally that these optimizations result in speedups up to 39.4× on a set of benchmarks from the data analytics domain.

High performance in modern computing platforms requires programs to be parallel, distributed, and run on heterogeneous hardware. However programming such architectures is extremely difficult due to the need to implement the application using multiple programming models and combine them together in ad-hoc ways. To optimize distributed applications both for modern hardware and for modern programmers we need a programming model that is sufficiently expressive to support a variety of parallel applications, sufficiently performant to surpass hand-optimized sequential implementations, and sufficiently portable to support a variety of heterogeneous hardware. Unfortunately existing systems tend to fall short of these requirements. In this paper we introduce the Distributed Multiloop Language (DMLL), a new intermediate language based on common parallel patterns that captures the necessary semantic knowledge to efficiently target distributed heterogeneous architectures. We show straightforward analyses that determine what data to distribute based on its usage as well as powerful transformations of nested patterns that restructure computation to enable distribution and optimize for heterogeneous devices. We present experimental results for a range of applications spanning multiple domains and demonstrate highly efficient execution compared to manually-optimized counterparts in multiple distributed programming models.

Field Programmable Gate Arrays (FPGAs) are very versatile devices, but their complicated programming model has stymied their widespread usage. While modern High-Level Synthesis (HLS) tools provide better programming models, the interface they offer is still too low-level. In order to produce good quality hardware designs with these tools, the users are forced to manually perform optimizations that demand detailed knowledge of both the application and the implementation platform. Additionally, many HLS tools only generate isolated hardware modules that the user still needs to integrate into a system design before generating the FPGA bitstream. These problems make HLS tools difficult to use for application developers who have little hardware design knowledge. To address these problems, we propose an automated methodology to generate FPGA bitstreams from high-level programs written in Domain-Specific Languages (DSLs). We leverage the domain-knowledge conveyed by the DSL and its domain-specific semantics to extract application parallelism, perform optimizations and also identify a suitable system-architecture for the implementation, thereby, relieving the user from most of the hardware-level details. We demonstrate the high productivity and high design quality this approach offers by automatically generating hardware systems from applications written in OptiML, a machine-learning DSL. To evaluate our methodology, we use four OptiML applications and show that we can easily generate different solutions which achieve different trade-offs between performance and area. More importantly, the results reveal that our generated hardware achieves much better performance compared to the one obtained from using the HLS tool without platform-specific optimizations.

Recent work has explored using higher level languages to improve programmer productivity on GPUs. These languages often utilize high level computation patterns (e.g., Map and Reduce) that encode parallel semantics to enable automatic compilation to GPU kernels. However, the problem of efficiently mapping patterns to GPU hardware becomes significantly more difficult when the patterns are nested, which is common in non-trivial applications. To address this issue, we present a general analysis framework for automatically and efficiently mapping nested patterns onto GPUs. The analysis maps nested patterns onto a logical multidimensional domain and parameterizes the block size and degree of parallelism in each dimension. We then add GPU-specific hard and soft constraints to prune the space of possible mappings and select the best mapping. We also perform multiple compiler optimizations that are guided by the mapping to avoid dynamic memory allocations and automatically utilize shared memory within GPU kernels. We compare the performance of our automatically selected mappings to hand-optimized implementations on multiple benchmarks and show that the average performance gap on 7 out of 8 benchmarks is 24%. Furthermore, our mapping strategy outperforms simple 1D mappings and existing 2D mappings by up to 28.6x and 9.6x respectively.

Just-in-time (JIT) compilation of running programs provides more optimization opportunities than offline compilation. Modern JIT compilers, such as those in virtual machines like Oracle's HotSpot for Java or Google's V8 for JavaScript, rely on dynamic profiling as their key mechanism to guide optimizations. While these JIT compilers offer good average performance, their behavior is a black box and the achieved performance is highly unpredictable.

Developing high-performance software is a difficult task that requires the use of low-level, architecture-specific programming models (e.g., OpenMP for CMPs, CUDA for GPUs, MPI for clusters). It is typically not possible to write a single application that can run efficiently in different environments, leading to multiple versions and increased complexity. Domain-Specific Languages (DSLs) are a promising avenue to enable programmers to use high-level abstractions and still achieve good performance on a variety of hardware. This is possible because DSLs have higher-level semantics and restrictions than general-purpose languages, so DSL compilers can perform higher-level optimization and translation. However, the cost of developing performance-oriented DSLs is a substantial roadblock to their development and adoption. In this article, we present an overview of the Delite compiler framework and the DSLs that have been developed with it. Delite simplifies the process of DSL development by providing common components, like parallel patterns, optimizations, and code generators, that can be reused in DSL implementations. Delite DSLs are embedded in Scala, a general-purpose programming language, but use metaprogramming to construct an Intermediate Representation (IR) of user programs and compile to multiple languages (including C++, CUDA, and OpenCL). DSL programs are automatically parallelized and different parts of the application can run simultaneously on CPUs and GPUs. We present Delite DSLs for machine learning, data querying, graph analysis, and scientific computing and show that they all achieve performance competitive to or exceeding C++ code.

Domain-specific languages provide a promising path to automatically compile high-level code to parallel, heterogeneous, and distributed hardware. However, in practice high performance DSLs still require considerable software expertise to develop and force users into tool-chains that hinder prototyping and debugging. To address these problems, we present Forge, a new meta DSL for declaratively specifying high performance embedded DSLs. Forge provides DSL authors with high-level abstractions (e.g., data structures, parallel patterns, effects) for specifying their DSL in a way that permits high performance. From this high-level specification, Forge automatically generates both a naïve Scala library implementation of the DSL and a high performance version using the Delite DSL framework. Users of a Forge-generated DSL can prototype their application using the library version, and then switch to the Delite version to run on multicore CPUs, GPUs, and clusters without changing the application code. Forge-generated Delite DSLs perform within 2x of hand-optimized C++ and up to 40x better than Spark, an alternative high-level distributed programming environment. Compared to a manually implemented Delite DSL, Forge provides a factor of 3-6x reduction in lines of code and does not sacrifice any performance. Furthermore, Forge specifications can be generated from existing Scala libraries, are easy to maintain, shield DSL developers from changes in the Delite framework, and enable DSLs to be retargeted to other frameworks transparently.

High level data structures are a cornerstone of modern programming and at the same time stand in the way of compiler optimizations. In order to reason about user- or library-defined data structures compilers need to be extensible. Common mechanisms to extend compilers fall into two categories. Frontend macros, staging or partial evaluation systems can be used to programmatically remove abstraction and specialize programs before they enter the compiler. Alternatively, some compilers allow extending the internal workings by adding new transformation passes at different points in the compile chain or adding new intermediate representation (IR) types. None of these mechanisms alone is sufficient to handle the challenges posed by high level data structures. This paper shows a novel way to combine them to yield benefits that are greater than the sum of the parts.

Computing systems are becoming increasingly parallel and heterogeneous, and therefore new applications must be capable of exploiting parallelism in order to continue achieving high performance. However, targeting these emerging devices often requires using multiple disparate programming models and making decisions that can limit forward scalability. In previous work we proposed the use of domain-specific languages (DSLs) to provide high-level abstractions that enable transformations to high performance parallel code without degrading programmer productivity. In this paper we present a new end-to-end system for building, compiling, and executing DSL applications on parallel heterogeneous hardware, the Delite Compiler Framework and Runtime. The framework lifts embedded DSL applications to an intermediate representation (IR), performs generic, parallel, and domain-specific optimizations, and generates an execution graph that targets multiple heterogeneous hardware devices. Finally we present results comparing the performance of several machine learning applications written in OptiML, a DSL for machine learning that utilizes Delite, to C++ and MATLAB implementations. We find that the implicitly parallel OptiML applications achieve single-threaded performance comparable to C++ and outperform explicitly parallel MATLAB in nearly all cases.

Exploiting heterogeneous parallel hardware currently requires mapping application code to multiple disparate programming models. Unfortunately, general-purpose programming models available today can yield high performance but are too low-level to be accessible to the average programmer. We propose leveraging domain-specific languages (DSLs) to map high-level application code to heterogeneous devices. To demonstrate the potential of this approach we present OptiML, a DSL for machine learning. OptiML programs are implicitly parallel and can achieve high performance on heterogeneous hardware with no modification required to the source code. For such a DSL-based approach to be tractable at large scales, better tools are required for DSL authors to simplify language creation and parallelization. To address this concern, we introduce Delite, a system designed specifically for DSLs that is both a framework for creating an implicitly parallel DSL as well as a dynamic runtime providing automated targeting to heterogeneous parallel hardware. We show that OptiML running on Delite achieves single-threaded, parallel, and GPU performance superior to explicitly parallelized MATLAB code in nearly all cases.