SIMD re-convergence at thread frontiers

Abstract

Hardware and compiler techniques for mapping data-parallel programs with divergent control flow to SIMD architectures have recently enabled the emergence of new GPGPU programming models such as CUDA, OpenCL, and DirectX Compute. The impact of branch divergence can be quite different depending upon whether the program's control flow is structured or unstructured. In this paper, we show that unstructured control flow occurs frequently in applications and can lead to significant code expansion when executed using existing approaches for handling branch divergence.This paper proposes a new technique for automatically mapping arbitrary control flow onto SIMD processors that relies on a concept of a Thread Frontier, which is a bounded region of the program containing all threads that have branched away from the current warp. This technique is evaluated on a GPU emulator configured to model i) a commodity GPU (Intel Sandybridge), and ii) custom hardware support not realized in current GPU architectures. It is shown that this new technique performs identically to the best existing method for structured control flow, and re-converges at the earliest possible point when executing unstructured control flow. This leads to i) between 1.5 − 633.2% reductions in dynamic instruction counts for several real applications, ii) simplification of the compilation process, and iii) ability to efficiently add high level unstructured programming constructs (e.g., exceptions) to existing data-parallel languages.

CAPRI: Prediction of compaction-adequacy for handling control-divergence in GPGPU architectures

“…There are two common ways to maintaining the logical PC of each thread. The first, used by the GPU in Intel's Sandy Bridge [7,16], maintains a separate PC for each thread and masks out threads that do not match the current per warp PC. NVIDIA and AMD GPUs use an alternate mechanism in which the active masks are stored on a reconvergence stack, which we explain below.…”

“…This work was extended to dynamic warp subdivision [18], which allows warp subsets to be scheduled independently to enhance latency tolerance. Diamos et al [7] propose Thread Frontier as an alternative to the immediate post dominator reconvergence algorithm. Thread frontiers use the earliest reconvergence point possible in an unstructured control flow [31].…”

Dynamic Inter-Thread Vectorization Architecture: Extracting DLP from TLP

“…They do not support exceptions or interruptions, which prevents their use with a general-purpose system software stack. Various works extend the SIMT model to support more generic code [22], [23] or more flexible execution [24], [25], [26]. However, they all target applications specifically written for GPUs, rather than general-purpose parallel applications.…”

Reconvergence de contrôle implicite pour les architectures SIMT

“…Il n'est donc pas strictement nécessaire de recourir à un mécanisme tel que celui employé par NVIDIA pour exécuter du code arbitraire. Cependant, la technique à base de sauts et annotations permet d'éviter la duplication statique de code en la remplaçant par de la duplication dynamique (Diamos et al, 2011). Le mécanisme utilisé par Tesla peut également être étendu à certains sauts indirects, comme le propose l'architecture Fermi (Nickolls, Dally, 2010).…”

“…Diamos et ses coauteurs formalisent cette approche en présentant un algorithme permettant de calculer l'ordre optimal des blocs de base et proposent une réalisation logicielle (Diamos et al, 2011).…”