Compilers and More: Exascale Programming Requirements

By Michael Wolfe

April 14, 2011

Programming at Exascale, Part 3

In an earlier column, I discussed six levels of parallelism that we’ll have in exascale systems: node, socket, core, vector, instruction, and pipeline levels, and said that to reach exascale performance, we need to take advantage of all these levels, since the final performance is the product of them all. In my most recent column, I argued that to be successful at that, we need to effectively expose, express, and exploit parallelism: expose it in the application and algorithms, express it in the language and program, and exploit it in the generated code and at runtime. Exposing parallelism is mostly a creative task, and thus must be done by humans. Expressing parallelism is where we mostly get sidetracked: what language, what kind of parallelism, how will it work with legacy software? Since parallel programming is all about performance, we need to focus on those aspects that would hinder performance, specifically locality and synchronization. Finally, successfully exploiting parallelism means mapping the parallelism exposed in the application and expressed in the program to the parallelism in the hardware. I discussed five dimensions of flexibility: scalability, dynamic parallelism, composability, load balancing, and productivity. In this column, the last of a three-part series, I’ll give my views on what programming at the exascale level is likely to require, and how we can get there from where we are today. My belief is that it will take some work, but it’s not a wholesale rewrite of 50 years of high performance expertise.

Exascale Programming: What It Won’t Be

What are the characteristics of a programming strategy for the coming exascale computers? It’s easier to say what it isn’t.

It’s not a library. Encapsulation is a well-known, often used, and important technique to building large systems. By design, encapsulation hides information about the implementation of the encapsulated object (data structure, algorithm, service) from the user of that object. Encapsulation will continue to be important for many reasons. But information hiding obscures not just the algorithm and data structures, but performance aspects, such as what kinds of parallelism are used within the encapsulated object and how that interacts with parallelism of the user of that object, or low level information such as how the data is laid out and how that affects locality in an algorithm. In particular, opaque low-level libraries (e.g., MPI for data distribution and message passing) hide too much information from the system, preventing any system-level tuning. That’s not to say a useful system won’t be built using MPI as the transport layer, but MPI or POSIX threads or other low-level libraries should not be directly used in the application.

It’s not a C++ class hierarchy or template library. Here, I’m again going out on a limb; there have been and continue to be many sets of useful C++ class libraries intended to raise the level of application programming. Take the C++ standard template library for vector; the intent of such a template is to allow a user to define a data structure and get the benefit of reusing any routines in the STL or from elsewhere built on the vector template. But you don’t really understand the performance of the vector datatype; that information hiding means you don’t know if accesses to vector V; are efficient or not. Compare that to an array access in a loop, with the corresponding vector access V[i]; the array access can often be optimized down to two instructions: load, and increment the pointer to the next address. Moreover, two-dimensional objects using the vector type (vector>) become even more opaque.

Or take Thrust, an STL-like implementation providing a high-level interface to GPU programming, built on CUDA. You can define two vectors in Thrust as

   using namespace thrust;    device_vector x(1000);    device_vector y(1000); 

Multiplying two such vectors and then accumulating the result can be done as:

   transform( x.begin(), x.end(), y.begin(), z.begin(), multiplies() );    r = reduce( z.begin(), z.end(), 0, plus() );

This is certainly easier (more productive?) to write than the equivalent CUDA C (or CUDA Fortran) code, but it’s still far easier to write the Fortran:

   r = sum( x(:) * y(:) ) 

Moreover, when the constructs are part of the language, the compiler can compose and optimize them together. As mentioned in my last column, in the Fortran case, the compiler can generate code for the multiply then accumulate the result without requiring an intermediate vector result. With the C++ library, the code for the transform method doesn’t know that its result will immediately be accumulated, so the method or (as in this case) the user has to provide a result vector. The only tool the compiler has to optimize class library calls is inlining, and it’s simply not enough to recover the performance lost by the abstraction. There have been some efforts to use run-time code generation, building the expression tree from the method calls, then generating the optimized (and composed) code from the whole expression tree; this was the technology behind Rapidmind, which is now being used in Intel’s Array Building Blocks (ArBB). Such mechanisms are promising, but what we really want is a way to define new data types and describe operations to the compiler in a way that the compiler can reason about them, compose them, reorder them, and so on; currently, the definition is basically in terms of C code, which is not expressive enough. There’s a research project just waiting to happen.

It’s not a domain-specific language. I really like the idea of DSLs, of embedding domain knowledge in the language and using that knowledge when generating and optimizing the code. However, languages, real languages, are big project; DSLs are (by definition) specialized, and hence don’t have a large enough user community to support production, maintenance and continuing development of the language and all the tools needed to support a language. We can’t expect language implementors (like PGI) to take on the development and continuing support of a plethora of languages, any more than we should expect user communities to each design, implement, and then continue to update, enhance, tune and optimize the language implementation with each new processor release from Intel. A possible alternative approach would be to implement a language to support DSLs, supported by a language vendor, including interfacing to debuggers, performance tools, editors, and so on. The various user communities would then be somewhat insulated from the details of a performance-oriented solution, and the vendor would avoid falling into the many-languages trap. There’s another potential research project.

It’s not OpenCL. OpenCL may be a necessary step towards heterogeneous programming, but it’s not the final answer. It’s very low level, “close to the metal”, as even the language designers admit. As with MPI, we may be able to build on OpenCL, but it’s not sufficient.

It’s not a whole new language. New languages have a high barrier to entry; most programmers avoid adopting a new language for fear that it will die, unless the language meets some need better than anything else, or until it has survived along enough to ameliorate the fear. But I think a new language is not called for here. We may benefit from some new features in existing languages, and maybe new ways to make programs in those languages, but most new languages really don’t add semantically much beyond managed memory.

It’s not easy. I’ve argued before that parallel programming is not easy, won’t be, and can’t be made easy. The idea of making parallel programming easy is silly.

It’s not just parallelism. Parallelism is an important aspect, perhaps the dominant aspect, but the key isn’t parallelism, it’s performance. A bad parallel algorithm doesn’t run fast just because it’s parallel. A bad implementation of a good parallel algorithm will also be slow. It’s quite easy to write slow parallel programs; this was the key failure (my opinion) of High Performance Fortran. So our programming mechanism will focus on performance, where parallelism is one aspect (locality and synchronization being two more).

Exascale Programming: What It Is

So what do we want and need when programming at exascale from whatever programming environment we get? Here is my bucket list:

  • It supports all levels of parallelism, from node parallelism down to vector and pipeline parallelism, effectively. Support is a big word here; it has to allow for a programming model that an application developer can use to think about what kinds of parallelism will map well at different levels, that a programmer can use to write a program that can be mapped well at different levels, and that the implementation (compiler and runtime) can use to exploit the parallelism. We have this today, clumsily, with different mechanisms for different levels; a bit more integration would take us a long way.
  • It can map an expression of program parallelism (a parallel loop, say) to different levels of hardware parallelism (across nodes, or to a vector unit) depending on the target. This will make it scalable up and down, from exascale to laptop. There was a great deal of work on the SISAL language to efficiently scalarize an implicitly parallel language, which turned out to be largely the dual of the parallelizing compiler problem. Such work will be part of this parallelism remapping. Remapping node-level parallelism may require changing the data distribution per node; today, this is done at the application level. We should be able to specify what parameters of the program depend on which aspects of the target machine, so the system can do the remapping.
  • It supports the programmer with lots of feedback. Vectorizing compilers have been very successful for over 35 years in delivering good vector performance from sequential loops because the compilers tell the programmer when they are successful, and more importantly, when and why they fail. This is essentially performance feedback. We are in the business of developing high performance applications, and we should be notified when we are using constructs that will restrict our performance. Static feedback and useful dynamic feedback will both be critical.
  • It supports dynamic parallelism, creating parallel tasks and threads when needed. There are many successful and useful implementations of dynamic parallelism, some limited (OpenMP) and some more aggressive (Cilk). Dynamic parallelism is somewhat at odds with locality and synchronization optimization. Using a work-stealing scheduler, an idle worker will steal a work item from the queue of another worker. However, that work item may have been placed on that worker’s queue because that’s where its data is, or because that work item depends on some other work item also assigned to that worker. However, without constructs for dynamic parallelism, we end up micromanaging thread-level parallelism in the constructs we do have.
  • It efficiently composes abstract operations, as I discussed in my previous column; whether these are native to the language, or abstract operations defined by a user or in a library, the implementation must be able to combine them naturally. Perhaps, when we define abstract operations, we need a mechanism to describe how they can compose with others. Many now-standard compiler optimizations fall into composition, such as loop vectorization and loop fusion. We need more investigation about what composing abstract operations means, beyond simply inlining.
  • It is self-balancing and self-tuning. This involves runtime introspection and behavior modification, and means the parameters or data and work distribution must be exposed to the system in order to be modified. Examples include changing the tile sizes for tiled nested loops when optimized for cache locality, or changing the data distributions when the work load is not uniform across the domain. Such behavior modification has been demonstrated in many systems, though not many integrated with the programming language and its implementation.
  • It must be resilient. The big systems are, many believe, going to be in partial failure mode much of the time. This presents challenges for the system manager and programmer. Expecting the entire system to be working, taking checkpoints and restoring from a failure point will not be efficient if failures are the norm. Some of the necessary features must be supported by the hardware (getting data off a node with a failed processor; early failure detection). Other features could be supported by some of the runtime features we develop for other reasons (redistributing data to working nodes; reserving some nodes to serve as online replacements). Such a system can survive and continue beyond many failures.

Most of these points (except for the last) have been researched and implemented in some form already, and could be reproduced with current technology (and enough motivation) in Fortran, C++, or whatever language you want. We have to extend the programming model to expose performance aspects and perhaps resilience aspects, so the user can guide how the system (compiler plus runtime) implements the program. We often get focused on either abstracting away so much that we lose sight of performance (as happened with High Performance Fortran), or we get so tied up with performance that we focus too much on details of each target machine (as happens today with OpenCL and CUDA). We need to let the programmer do the creative parts, and let the system do the mechanical work.

Final Note: This series of columns is an expanded form of the material from the PGI Exhibitor Forum presentation at SC10 in New Orleans. If you were there, you can tell me whether it’s more informative (or entertaining) in written or verbal form.

About the Author

Michael Wolfe has developed compilers for over 30 years in both academia and industry, and is now a senior compiler engineer at The Portland Group, Inc. (www.pgroup.com), a wholly-owned subsidiary of STMicroelectronics, Inc. The opinions stated here are those of the author, and do not represent opinions of The Portland Group, Inc. or STMicroelectronics, Inc.

Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industy updates delivered to you every week!

2022 Road Trip: NASA Ames Takes Off

November 25, 2022

I left Dallas very early Friday morning after the conclusion of SC22. I had a race with the devil to get from Dallas to Mountain View, Calif., by Sunday. According to Google Maps, this 1,957 mile jaunt would be the longe Read more…

2022 Road Trip: Sandia Brain Trust Sounds Off

November 24, 2022

As the 2022 Great American Supercomputing Road Trip carries on, it’s Sandia’s turn. It was a bright sunny day when I rolled into Albuquerque after a high-speed run from Los Alamos National Laboratory. My interview su Read more…

2022 HPC Road Trip: Los Alamos

November 23, 2022

With SC22 in the rearview mirror, it’s time to get back to the 2022 Great American Supercomputing Road Trip. To refresh everyone’s memory, I jumped in the car on November 3rd and headed towards SC22 in Dallas, stoppi Read more…

Chipmakers Looking at New Architecture to Drive Computing Ahead

November 23, 2022

The ability to scale current computing designs is reaching a breaking point, and chipmakers such as Intel, Qualcomm and AMD are putting their brains together on an alternate architecture to push computing forward. The chipmakers are coalescing around the new concept of sparse computing, which involves bringing computing to data... Read more…

QuEra’s Quest: Build a Flexible Neutral Atom-based Quantum Computer

November 23, 2022

Last month, QuEra Computing began providing access to its 256-qubit, neutral atom-based quantum system, Aquila, from Amazon Braket. Founded in 2018, and built on technology developed at Harvard and MIT, QuEra, is one of Read more…

AWS Solution Channel

Shutterstock 1648511269

Avoid overspending with AWS Batch using a serverless cost guardian monitoring architecture

Pay-as-you-go resources are a compelling but daunting concept for budget conscious research customers. Uncertainty of cloud costs is a barrier-to-entry for most, and having near real-time cost visibility is critical. Read more…

 

shutterstock_1431394361

AI and the need for purpose-built cloud infrastructure

Modern AI solutions augment human understanding, preferences, intent, and even spoken language. AI improves our knowledge and understanding by delivering faster, more informed insights that fuel transformation beyond anything previously imagined. Read more…

SC22’s ‘HPC Accelerates’ Plenary Stresses Need for Collaboration

November 21, 2022

Every year, SC has a theme. For SC22 – held last week in Dallas – it was “HPC Accelerates”: a theme that conference chair Candace Culhane said reflected “how supercomputing is continuously changing the world by Read more…

Chipmakers Looking at New Architecture to Drive Computing Ahead

November 23, 2022

The ability to scale current computing designs is reaching a breaking point, and chipmakers such as Intel, Qualcomm and AMD are putting their brains together on an alternate architecture to push computing forward. The chipmakers are coalescing around the new concept of sparse computing, which involves bringing computing to data... Read more…

QuEra’s Quest: Build a Flexible Neutral Atom-based Quantum Computer

November 23, 2022

Last month, QuEra Computing began providing access to its 256-qubit, neutral atom-based quantum system, Aquila, from Amazon Braket. Founded in 2018, and built o Read more…

SC22’s ‘HPC Accelerates’ Plenary Stresses Need for Collaboration

November 21, 2022

Every year, SC has a theme. For SC22 – held last week in Dallas – it was “HPC Accelerates”: a theme that conference chair Candace Culhane said reflected Read more…

Quantum – Are We There (or Close) Yet? No, Says the Panel

November 19, 2022

For all of its politeness, a fascinating panel on the last day of SC22 – Quantum Computing: A Future for HPC Acceleration? – mostly served to illustrate the Read more…

RISC-V Is Far from Being an Alternative to x86 and Arm in HPC

November 18, 2022

One of the original RISC-V designers this week boldly predicted that the open architecture will surpass rival chip architectures in performance. "The prediction is two or three years we'll be surpassing your architectures and available performance with... Read more…

Gordon Bell Special Prize Goes to LLM-Based Covid Variant Prediction

November 17, 2022

For three years running, ACM has awarded not only its long-standing Gordon Bell Prize (read more about this year’s winner here!) but also its Gordon Bell Spec Read more…

2022 Gordon Bell Prize Goes to Plasma Accelerator Research

November 17, 2022

At the awards ceremony at SC22 in Dallas today, ACM awarded the 2022 ACM Gordon Bell Prize to a team of researchers who used four major supercomputers – inclu Read more…

Gordon Bell Nominee Used LLMs, HPC, Cerebras CS-2 to Predict Covid Variants

November 17, 2022

Large language models (LLMs) have taken the tech world by storm over the past couple of years, dominating headlines with their ability to generate convincing hu Read more…

Nvidia Shuts Out RISC-V Software Support for GPUs 

September 23, 2022

Nvidia is not interested in bringing software support to its GPUs for the RISC-V architecture despite being an early adopter of the open-source technology in its GPU controllers. Nvidia has no plans to add RISC-V support for CUDA, which is the proprietary GPU software platform, a company representative... Read more…

RISC-V Is Far from Being an Alternative to x86 and Arm in HPC

November 18, 2022

One of the original RISC-V designers this week boldly predicted that the open architecture will surpass rival chip architectures in performance. "The prediction is two or three years we'll be surpassing your architectures and available performance with... Read more…

AWS Takes the Short and Long View of Quantum Computing

August 30, 2022

It is perhaps not surprising that the big cloud providers – a poor term really – have jumped into quantum computing. Amazon, Microsoft Azure, Google, and th Read more…

Chinese Startup Biren Details BR100 GPU

August 22, 2022

Amid the high-performance GPU turf tussle between AMD and Nvidia (and soon, Intel), a new, China-based player is emerging: Biren Technology, founded in 2019 and headquartered in Shanghai. At Hot Chips 34, Biren co-founder and president Lingjie Xu and Biren CTO Mike Hong took the (virtual) stage to detail the company’s inaugural product: the Biren BR100 general-purpose GPU (GPGPU). “It is my honor to present... Read more…

Tesla Bulks Up Its GPU-Powered AI Super – Is Dojo Next?

August 16, 2022

Tesla has revealed that its biggest in-house AI supercomputer – which we wrote about last year – now has a total of 7,360 A100 GPUs, a nearly 28 percent uplift from its previous total of 5,760 GPUs. That’s enough GPU oomph for a top seven spot on the Top500, although the tech company best known for its electric vehicles has not publicly benchmarked the system. If it had, it would... Read more…

AMD Thrives in Servers amid Intel Restructuring, Layoffs

November 12, 2022

Chipmakers regularly indulge in a game of brinkmanship, with an example being Intel and AMD trying to upstage one another with server chip launches this week. But each of those companies are in different positions, with AMD playing its traditional role of a scrappy underdog trying to unseat the behemoth Intel... Read more…

JPMorgan Chase Bets Big on Quantum Computing

October 12, 2022

Most talk about quantum computing today, at least in HPC circles, focuses on advancing technology and the hurdles that remain. There are plenty of the latter. F Read more…

UCIe Consortium Incorporates, Nvidia and Alibaba Round Out Board

August 2, 2022

The Universal Chiplet Interconnect Express (UCIe) consortium is moving ahead with its effort to standardize a universal interconnect at the package level. The c Read more…

Leading Solution Providers

Contributors

Using Exascale Supercomputers to Make Clean Fusion Energy Possible

September 2, 2022

Fusion, the nuclear reaction that powers the Sun and the stars, has incredible potential as a source of safe, carbon-free and essentially limitless energy. But Read more…

Nvidia, Qualcomm Shine in MLPerf Inference; Intel’s Sapphire Rapids Makes an Appearance.

September 8, 2022

The steady maturation of MLCommons/MLPerf as an AI benchmarking tool was apparent in today’s release of MLPerf v2.1 Inference results. Twenty-one organization Read more…

Not Just Cash for Chips – The New Chips and Science Act Boosts NSF, DOE, NIST

August 3, 2022

After two-plus years of contentious debate, several different names, and final passage by the House (243-187) and Senate (64-33) last week, the Chips and Science Act will soon become law. Besides the $54.2 billion provided to boost US-based chip manufacturing, the act reshapes US science policy in meaningful ways. NSF’s proposed budget... Read more…

SC22 Unveils ACM Gordon Bell Prize Finalists

August 12, 2022

Courtesy of the schedule for the SC22 conference, we now have our first glimpse at the finalists for this year’s coveted Gordon Bell Prize. The Gordon Bell Pr Read more…

Intel Is Opening up Its Chip Factories to Academia

October 6, 2022

Intel is opening up its fabs for academic institutions so researchers can get their hands on physical versions of its chips, with the end goal of boosting semic Read more…

AMD Previews 400 Gig Adaptive SmartNIC SOC at Hot Chips

August 24, 2022

Fresh from finalizing its acquisitions of FPGA provider Xilinx (Feb. 2022) and DPU provider Pensando (May 2022) ), AMD previewed what it calls a 400 Gig Adaptive smartNIC SOC yesterday at Hot Chips. It is another contender in the increasingly crowded and blurry smartNIC/DPU space where distinguishing between the two isn’t always easy. The motivation for these device types... Read more…

Google Program to Free Chips Boosts University Semiconductor Design

August 11, 2022

A Google-led program to design and manufacture chips for free is becoming popular among researchers and computer enthusiasts. The search giant's open silicon program is providing the tools for anyone to design chips, which then get manufactured. Google foots the entire bill, from a chip's conception to delivery of the final product in a user's hand. Google's... Read more…

AMD’s Genoa CPUs Offer Up to 96 5nm Cores Across 12 Chiplets

November 10, 2022

AMD’s fourth-generation Epyc processor line has arrived, starting with the “general-purpose” architecture, called “Genoa,” the successor to third-gen Eypc Milan, which debuted in March of last year. At a launch event held today in San Francisco, AMD announced the general availability of the latest Epyc CPUs with up to 96 TSMC 5nm Zen 4 cores... Read more…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire