We present a scalable, end-to-end workflow for protein design. By augmenting protein sequences with natural language descriptions of their biochemical properties, we train generative models that can be preferentially aligned with protein fitness landscapes. Through complex experimental-and simulation-based observations, we integrate these measures as preferred parameters for generating new protein variants and demonstrate our workflow on five diverse supercomputers. We achieve >1 ExaFLOPS sustained performance in mixed precision on each supercomputer and a maximum sustained performance of 4.11 Ex-aFLOPS and peak performance of 5.57 ExaFLOPS. We establish the scientific performance of our model on two tasks: (1) across a predetermined benchmark dataset of deep mutational scanning experiments to optimize the fitness-determining mutations in the yeast protein HIS7, and (2) in optimizing the design of the enzyme malate dehydrogenase to achieve lower activation barriers (and therefore increased catalytic rates) using simulation data. Our implementation thus sets high watermarks for multimodal protein design workflows.

We propose a scheme for reduced-precision representation of floating point data on a continuum between IEEE-754 floating point types. Our scheme enables the use of lower precision formats for a reduction in storage space requirements and data transfer volume. We describe how our scheme can be accelerated using existing hardware vector units on two general-purpose processor (GPP) microarchitectures (Intel Ivy Bridge and Haswell), as well as on a numerical accelerator (Intel Xeon Phi). Our evaluation demonstrates that supporting reduced precision by exploiting native vector instructions can yield a low overhead custom-precision floating point solution that does not require specialized hardware support. In our experiments we find cases where our scheme is actually faster than native floating point types where the underlying vector instruction set supports efficient byte-level permutations.

We propose a new language-neutral primitive for the LLVM compiler, which provides efficient context switching and message passing between lightweight threads of control. The primitive, called Swapstack, can be used by any language implementation based on LLVM to build higher-level language structures such as continuations, coroutines, and lightweight threads. As part of adding the primitives to LLVM, we have also added compiler support for passing parameters across context switches. Our modified LLVM compiler produces highly efficient code through a combination of exposing the context switching code to existing compiler optimizations, and adding novel compiler optimizations to further reduce the cost of context switches. To demonstrate the generality and efficiency of our primitives, we add one-shot continuations to C++, and provide a simple fiber library that allows millions of fibers to run on multiple cores, with a work-stealing scheduler and fast inter-fiber sychronization. We argue that compiler-supported lightweight context switching can be significantly faster than using a library to switch between contexts, and provide experimental evidence to support the position.