[llvm-advisor] Add optimization remarks visualization: relational grid, heatmap, Code Explorer and snapshot diff - #218692
Open
kamini08 wants to merge 41 commits into
Conversation
Signed-off-by: kamini08 <kaminibanait03@gmail.com>
Signed-off-by: kamini08 <kaminibanait03@gmail.com>
Signed-off-by: kamini08 <kaminibanait03@gmail.com>
… renderer Signed-off-by: kamini08 <kaminibanait03@gmail.com>
All six remark types (unknown, passed, missed, analysis, analysis-fp-commute, analysis-aliasing, failure) are now uniformly named across the ingestion pipeline: - add remarkTypeKey() and allRemarkTypeKeys() in RemarksAnalysisUtils as the single source of truth for type -> JSON key mapping; replaces the direct use of LLVM's typeToStr() which returns CamelCase strings inconsistent with the rest of the JSON API - RemarksAnalyzer (llvm.remarks.summary): pre-initialize all six keys at 0 before iterating so the by_type object always has the same shape regardless of which types appear in the file - RemarksDetailAnalyzer (llvm.remarks.detail): use canonical names - PassImpact insight: update hit-rate lookup keys to lowercase - OptimizationDelta insight: fix stale comment - Remarks Explorer (views.js): decode relational type integers using the canonical name array; show all six types as table columns, hiding any column that is all-zero across the top 20 rows - unit-detail.js: remove hardcoded type list from empty-state message Signed-off-by: kamini08 <kaminibanait03@gmail.com>
…ache llvm.remarks.summary and llvm.remarks.detail changed their output format in the previous commit (lowercase type keys, all six pre-initialized). Bump both to version 2 so the run-key hash changes and old cached results with capitalized keys are not returned. Signed-off-by: kamini08 <kaminibanait03@gmail.com>
Signed-off-by: kamini08 <kaminibanait03@gmail.com>
Signed-off-by: kamini08 <kaminibanait03@gmail.com>
Signed-off-by: kamini08 <kaminibanait03@gmail.com>
Signed-off-by: kamini08 <kaminibanait03@gmail.com>
Signed-off-by: kamini08 <kaminibanait03@gmail.com>
…a header Signed-off-by: kamini08 <kaminibanait03@gmail.com>
Signed-off-by: kamini08 <kaminibanait03@gmail.com>
Use sys::fs::real_path in findBuildDir so that symlinked or relative paths to compile_commands.json resolve to their actual directory. Signed-off-by: kamini08 <kaminibanait03@gmail.com>
Add RemarksHotspotAnalyzer that aggregates optimization remarks by function, file, and line to identify hotspots. Register the builtin.remarks_hotspot runner and add the llvm.remarks.hotspot capability to the catalog with dependency on llvm.remarks.summary. Signed-off-by: kamini08 <kaminibanait03@gmail.com>
Add HeatmapView to render hotspot tables from llvm.remarks.hotspot. Add heatmap icon to core.js and sidebar navigation to shell.js with keyboard shortcut g h. Register the /heatmap route in bundle.py and index.html. Signed-off-by: kamini08 <kaminibanait03@gmail.com>
When llvm.remarks.relational is available, render the full triage grid in the unit detail Remarks tab instead of the limited findingList. Add llvm.remarks.relational to the default capabilities queried for units so the relational data is fetched. Signed-off-by: kamini08 <kaminibanait03@gmail.com>
Signed-off-by: kamini08 <kaminibanait03@gmail.com>
Use llvm::demangle() in RemarksRelationalAnalyzer and RemarksHotspotAnalyzer so that mangled C++ function names are presented as human-readable signatures in the UI. Signed-off-by: kamini08 <kaminibanait03@gmail.com>
…ional endpoint The relational endpoint now accepts query parameters: - pass, name, function, file: substring filters on string columns - type: exact match on remark type enum - min_hotness: minimum hotness threshold - offset, limit: pagination (default limit 10000, max 100000) Server scans all units, builds merged columns, applies filters, then writes only the requested slice. Response includes total (filtered count) for pagination. Frontend triage grid is rewritten as server-paginated: fetches one page at a time, filter inputs debounce 300ms then re-fetch from server, page controls step through results. Signed-off-by: kamini08 <kaminibanait03@gmail.com>
…ering Signed-off-by: kamini08 <kaminibanait03@gmail.com>
…-level comparison Signed-off-by: kamini08 <kaminibanait03@gmail.com>
Signed-off-by: kamini08 <kaminibanait03@gmail.com>
Signed-off-by: kamini08 <kaminibanait03@gmail.com>
…nt empty snapshot Signed-off-by: kamini08 <kaminibanait03@gmail.com>
Signed-off-by: kamini08 <kaminibanait03@gmail.com>
…n + fix emitOptRemarks yaml Signed-off-by: kamini08 <kaminibanait03@gmail.com>
…ashboard + red highlight for added remarks Signed-off-by: kamini08 <kaminibanait03@gmail.com>
… at source location Signed-off-by: kamini08 <kaminibanait03@gmail.com>
…s to Code Explorer Signed-off-by: kamini08 <kaminibanait03@gmail.com>
…rer filter wiring Signed-off-by: kamini08 <kaminibanait03@gmail.com>
…s, and project stats sidebar Signed-off-by: kamini08 <kaminibanait03@gmail.com>
…aders Signed-off-by: kamini08 <kaminibanait03@gmail.com>
…le missing source gracefully in Code Explorer Signed-off-by: kamini08 <kaminibanait03@gmail.com>
…ap rows Signed-off-by: kamini08 <kaminibanait03@gmail.com>
Signed-off-by: kamini08 <kaminibanait03@gmail.com>
… and HTTP endpoints
|
Hello @kamini08 👋 Thank you for submitting a Pull Request (PR) to the LLVM Project. Since this is your first PR, here are a few useful links covering our main contribution policies and review practices.
Please reply to this message to confirm that you have read these policies, especially the LLVM AI Tool Use Policy, and that any AI tool usage has been noted in the PR description. Frequently asked questionsHow do I add reviewers? This PR will be automatically labeled, and the relevant teams will be notified. For some parts of the project, reviewers may also be added automatically. You can also add reviewers manually using the Reviewers section on this page. If you cannot use that section, it is probably because you do not have write permissions for the repository. In that case, you can request a review by tagging reviewers in a comment using What if there are no comments? If you have not received any comments on your PR after a week, you can request a review by pinging the PR with a comment such as “Ping”. The common courtesy ping rate is once a week. Please remember that you are asking for volunteer time from other developers. Are any special GitHub settings required to contribute to LLVM? We only require contributors to have a public email address associated with their GitHub commits, see this section of LLVM Developer Policy for details. If you have questions, feel free to leave a comment on this PR, or ask on LLVM Discord or LLVM Discourse. Thank you, |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Add optimization remarks visualization: relational grid, heatmap, Code Explorer and snapshot diff
This PR is part of my GSoC 2026 project on llvm-advisor. It adds end-to-end support for visualizing and comparing Clang optimization remarks: importing
-fsave-optimization-recordfiles, analyzing them with new C++ analyzers, and exposing the results through CLI commands and a web UI with a heatmap, Code Explorer, triage grid, and snapshot diff.What this adds
Import for
-fsave-optimization-recordYAML files.llvm-advisor import foo.opt.yamlparses a Clang optimization-record file, creates a snapshot and unit, runs the remarks analyzers, and stores the results in the CAS-backed store. The importer also reports clear errors for missing, empty, or malformed YAML, and preserves PGOHotnessvalues when present.Four new C++ analyzers registered in the capability catalog:
llvm.remarks.summary- produces overview counts by pass, remark name, and remark type. These numbers back the dashboard cards and the overview page.llvm.remarks.relational- builds a columnar table of every remark. Instead of returning a large array of objects, it returns integer-indexed columns plus deduplicated string tables for function, file, pass, and name, which keeps JSON payloads small and makes browser-side filtering fast even for millions of rows.llvm.remarks.hotspot- aggregates remarks by function, file, and line, tracking total counts and max hotness. This powers the heatmap and the project stats sidebar.llvm.remarks.detail- per-remark detail output used by the detail panel and Code Explorer badges.Heatmap view (
g h). A sortable table of hotspots that shows visual hotness bars, status chips (Critical / Moderate / Low), and a project stats sidebar, so you can see at a glance which functions and lines have the most actionable remarks.Code Explorer (
g e). Displays source files with remark counts overlaid on each line. It supportspassandfunctionfilters and is linked from the heatmap rows and the triage grid, so you can jump straight from a high-level hotspot to the exact source line that produced it.Snapshot comparison. The CLI command
llvm-advisor compare --before <snap> --after <snap>and the Compare UI tab (g c) show matched, changed, added, and removed units between two builds, plus an Optimization Impact table that breaks down per-function deltas in missed and passed remarks.HTTP endpoints. The server exposes REST endpoints for relational remarks, hotspots, per-source-file remarks, and snapshot diffs. The front-end is a single-page app that consumes these endpoints and renders the overview, heatmap, triage grid, Code Explorer, and Compare views.
Testing
There are 9 lit tests under
llvm/test/tools/llvm-advisor/covering import, invalid input, relational/hotspot queries, snapshot compare, and the HTTP endpoints.Documentation