Stack-less SIMT reconvergence at low cost

Sylvain Collange 1, *
* Corresponding author
1 ARENAIRE - Computer arithmetic
Inria Grenoble - Rhône-Alpes, LIP - Laboratoire de l'Informatique du Parallélisme
Abstract : Parallel architectures following the SIMT model such as GPUs benefit from application regularity by issuing concurrent threads running in lockstep on SIMD units. As threads take different paths across the control-flow graph, lockstep execution is partially lost, and must be regained whenever possible in order to maximize the occupancy of SIMD units. In this paper, we propose a technique to handle SIMT control divergence that operates in constant space and handles indirect jumps and recursion. We describe a possible implementation which leverage the existing memory divergence management unit, ensuring a low hardware cost. In terms of performance, this solution is at least as efficient as existing techniques.
Document type :
Reports
2011


https://hal.archives-ouvertes.fr/hal-00622654
Contributor : Sylvain Collange <>
Submitted on : Monday, September 12, 2011 - 4:14:16 PM
Last modification on : Wednesday, July 29, 2015 - 1:20:13 AM
Document(s) archivé(s) le : Tuesday, November 13, 2012 - 10:25:17 AM

File

collange_sympa2011_en.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-00622654, version 1

Collections

Citation

Sylvain Collange. Stack-less SIMT reconvergence at low cost. 2011. <hal-00622654>

Export

Share

Metrics

Record views

342

Document downloads

443