Skip to content

[rocprofiler-compute] [Documentation] Add metric descriptions for missing gfx942 metrics#3027

Open
vedithal-amd wants to merge 1 commit intodevelopfrom
users/vedithal/rocprofiler-compute-mi300-metric-descriptions
Open

[rocprofiler-compute] [Documentation] Add metric descriptions for missing gfx942 metrics#3027
vedithal-amd wants to merge 1 commit intodevelopfrom
users/vedithal/rocprofiler-compute-mi300-metric-descriptions

Conversation

@vedithal-amd
Copy link
Contributor

Motivation

This PR adds metric descriptions for 10 gfx942 (MI300) metrics that were previously missing descriptions across the following hardware blocks:

  • Memory Chart (sL1D): 1 metric
  • Compute Pipeline (MFMA): 2 metrics
  • Address Processing Unit (TA): 3 metrics
  • L2 Cache (Fabric Interface): 1 metric
  • Speed-of-Light panels: 3 metrics

These descriptions provide essential context for users profiling and analyzing performance on gfx942 architecture (MI300 series), including details on hardware counters, stall conditions, and architectural behavior specific to MI300's advanced features like F8 MFMA operations.

Technical Details

This PR uses material such as:

  • Counter definitions from counter_defs.yaml
  • Metric formulas used in avg/min/max/value/peak fields
  • Metric descriptions of similar metrics across architectures
  • MI300A/MI300X architecture documentation

Aforementioned information was used to derive accurate metric descriptions following established conventions

Changes:

  • Added 7 metric descriptions to analysis config YAML files for gfx942
  • Added 5 RST-formatted descriptions to per-architecture metric definitions (gfx942_metrics_description.yaml)
  • Updated hash database (src/utils/.config_hashes.json) for configuration integrity verification
  • All metric descriptions follow proper RST formatting with :ref: cross-references
  • Hash consistency check passes successfully

JIRA ID

ROCM-1126

AIPROFCOMP-9

Test Plan

Test Result

Submission Checklist

@vedithal-amd vedithal-amd marked this pull request as ready for review February 2, 2026 22:37
@vedithal-amd vedithal-amd requested review from a team and prbasyal-amd as code owners February 2, 2026 22:37
Copilot AI review requested due to automatic review settings February 2, 2026 22:37
@vedithal-amd vedithal-amd requested a review from a team as a code owner February 2, 2026 22:37
Copy link
Contributor

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

This PR adds documentation for 10 previously undocumented metrics specific to the gfx942 (MI300) architecture, improving the completeness of profiling documentation for this hardware generation.

Changes:

  • Added metric descriptions for gfx942-specific hardware counters across memory, compute, and cache subsystems
  • Updated configuration hash database to reflect the documentation changes
  • Added changelog entry documenting the metric description additions

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated no comments.

Show a summary per file
File Description
gfx942_metrics_description.yaml Added 5 RST-formatted metric descriptions for F8 MFMA ops, cache hit rates, atomic requests, and 128B reads
.config_hashes.json Updated configuration file hashes for modified analysis config files
1700_l2_cache.yaml Added description for L2 128B read requests metric
1500_address_processing_unit_and_data_return_path_ta_td.yaml Added descriptions for 3 sequencer stall metrics in the address processing unit
1100_compute_units_compute_pipeline.yaml Added descriptions for F8 MFMA floating-point operations metrics
0300_memory_chart.yaml Added description for scalar L1D cache latency metric
CHANGELOG.md Documented the addition of missing gfx942 metric descriptions

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

@vedithal-amd vedithal-amd force-pushed the users/vedithal/rocprofiler-compute-fix-metrics-description branch from 07817fc to 87204a2 Compare February 2, 2026 22:49
@vedithal-amd vedithal-amd force-pushed the users/vedithal/rocprofiler-compute-fix-metrics-description branch from 87204a2 to 5358ab3 Compare February 2, 2026 23:36
@vedithal-amd vedithal-amd force-pushed the users/vedithal/rocprofiler-compute-mi300-metric-descriptions branch from 659bd27 to 04590cd Compare February 2, 2026 23:42
Base automatically changed from users/vedithal/rocprofiler-compute-fix-metrics-description to develop February 3, 2026 14:40
Add metric descriptions for gfx942 metrics that were previously missing descriptions across multiple hardware blocks:

- Memory Chart (sL1D): 1 metric
- Compute Pipeline (MFMA): 2 metrics
- Address Processing Unit (TA): 3 metrics
- L2 Cache (Fabric Interface): 1 metric

Additionally, add RST descriptions in per-arch yaml for:
- Speed-of-Light panels: 3 metrics
- Cache access metrics: 2 metrics

Update config hashes for consistency verification.
@vedithal-amd vedithal-amd force-pushed the users/vedithal/rocprofiler-compute-mi300-metric-descriptions branch from 04590cd to f8f8c86 Compare February 3, 2026 14:46
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant