Skip to content

docs: expand terminology glossary from 6 to 30+ profiling terms#2859

Open
Ashutosh0x wants to merge 1 commit into
openxla:masterfrom
Ashutosh0x:docs/expand-terminology
Open

docs: expand terminology glossary from 6 to 30+ profiling terms#2859
Ashutosh0x wants to merge 1 commit into
openxla:masterfrom
Ashutosh0x:docs/expand-terminology

Conversation

@Ashutosh0x

Copy link
Copy Markdown

Summary

Expand the erminology.md glossary from 6 basic terms to 30+ comprehensive definitions, organized into logical categories.

Problem

The existing terminology page only defined 6 terms: Profile, Session, Run, Host, Device, Step. However, XProf documentation and UI heavily use terms like FLOPS Utilization, HBM, TensorCore, Duty Cycle, All-Reduce, Roofline Model, HLO, Fusion, etc. — none of which were defined. This creates a barrier for new users trying to understand profiling output.

Changes

Expanded from 6 → 30+ terms across 6 categories:

Category Terms Added
Hardware HBM, VMEM, TensorCore, MXU, SXM/PCIe
Performance Metrics FLOPS Utilization, Duty Cycle, Memory Bandwidth Utilization, Goodput Efficiency
XLA/Compilation HLO, Op, Fusion, Kernel, Custom Call
Communication DCN, ICI, All-Reduce, All-Gather, Collective Permute
Profiling Tools Trace Viewer, Op Profile, Memory Viewer, Roofline Model, Graph Viewer, Kernel Profiling

All definitions are derived from existing XProf documentation pages (overview_page.md, roofline_model.md, trace_viewer.md, etc.) to ensure consistency.

No behavioral changes

Documentation-only change.

The existing terminology page only defined 6 basic terms (Profile,
Session, Run, Host, Device, Step). Expand it with comprehensive
definitions organized into logical categories:

- General Terms (4): Profile, Session, Run, Step
- Hardware Terms (7): Host, Device, HBM, VMEM, TensorCore, MXU, SXM/PCIe
- Performance Metrics (4): FLOPS Utilization, Duty Cycle, Memory BW, Goodput
- XLA/Compilation Terms (5): HLO, Op, Fusion, Kernel, Custom Call
- Communication Terms (5): DCN, ICI, All-Reduce, All-Gather, Collective Permute
- Profiling Tools (6): Trace Viewer, Op Profile, Memory Viewer, Roofline, Graph Viewer, Kernel Profiling

These terms appear throughout XProf documentation and UI but were
previously undefined, creating a barrier for new users.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant