Timeloop v3.0.3 (Ruby)

What's Changed

Several incremental but critical updates:

• Fix bugs in "precise" multicast hop-count estimation.
• Overhaul distributed-multicast modeling approach and make it work with precise hop-count estimation (in prior releases it would only work with probabilistic hop-count estimation).
• Deprecate probabilistic hop-count estimation.

Full Changelog: v3.0.1...v3.0.3

Timeloop v3.0.2 (Ruby)

Add the ability to minimum factors (>=) to mapspace constraints.

Full Changelog: v3.0.1...v3.0.2

Timeloop v3.0.1 (Ruby)

Hotfix release. Fixes bug in outermost-level bypass processing.

Full Changelog: v3.0...v3.0.1

Timeloop v3.0 (Ruby)

Ruby is the third (v3.0) release of Timeloop. The headline feature of this release is support for Imperfect Factorization (described in Horeni et. al., ISPASS 2022), with @MarkHoreni serving as the primary contributor.

Other major changes include support for flattened mappings and spatially-skewed mappings, and countless bugfixes.

v3.0's inputs are backward compatible with those of v2.0.

Full Changelog: v2.0...v3.0

Timeloop v2.0 (Sparseloop)

Sparseloop is the second release (v2.0) of Timeloop. In this release, we added support for modeling sparsity-related hardware optimizations: gating, skipping, and compression. The impact of these optimizations is modeled using statistical characterization of workload densities (fixed structured and hypergeometric).

v2.0’s inputs are backward compatible with those of v1.0.

Full Changelog: v1.0...v2.0

Timeloop v1.0

First stable release. Includes mapper (with linear, random and hybrid search heuristics) and model (with Accelergy integration and support for imperfectly-factorized mappings). Supports modeling of dense (uncompressed) workloads.