ps2/llvm
Watch
2
0
Fork
You've already forked llvm
 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 1

Keep the split-hi16 rename, state what belongs in the fork #13

Merged
techwritescode merged 2 commits from r5900-split-hi16-name into r5900-ee 2026-10-15 16:58:07 +00:00
Conversation 0 Commits 2 Files changed 2 +14 -1
techwritescode commented 2026-10-15 16:56:18 +00:00
Owner
Copy link

Replaces the R5900 loop tuning in #12 with the parts that are PS2-specific.

  • [Mips] Give the split-hi16 pass its own name so opt can start: the one Mips-only commit from #12.
  • UPDATING.md: a "What belongs in the fork" section. Changes must be gated on the R5900/R3000A or __ps2sdk__, or be hidden options that default to off. Generic pass edits, pass-pipeline callbacks, clang driver -O changes and benchmark-tuned thresholds stay out.

Dropped from #12: the generic LoopUnrollAndJam/LoopUnrollPass/TargetTransformInfo.h edits, the R5900 pass pipeline callback, -funroll-loops at -O1, and R5900 partial/runtime unroll. On the console the partial unroll alone made the -O2 bench functions 1.70x GCC's size; without the tuning they are 0.88x.

Bench on the console with this branch (sdk-core tests/bench, all checksums match GCC), LLVM / GCC cycles:

Kernel -O1 -O2 -O3 -Os
packet_build 0.84 1.11 1.05 0.49
physics 1.52 1.16 2.05 1.04
raster 1.06 1.18 1.01 0.87
other five kernels 0.64-1.00 0.80-1.00 0.77-1.00 0.74-1.01

Total function size LLVM / GCC: 0.96, 0.88, 0.78, 0.92. The speed gaps are documented as exceptions in the bench README (PS2-101).

🤖 Generated with Claude Code

Replaces the R5900 loop tuning in #12 with the parts that are PS2-specific. - `[Mips] Give the split-hi16 pass its own name so opt can start`: the one Mips-only commit from #12. - `UPDATING.md`: a "What belongs in the fork" section. Changes must be gated on the R5900/R3000A or `__ps2sdk__`, or be hidden options that default to off. Generic pass edits, pass-pipeline callbacks, clang driver `-O` changes and benchmark-tuned thresholds stay out. Dropped from #12: the generic `LoopUnrollAndJam`/`LoopUnrollPass`/`TargetTransformInfo.h` edits, the R5900 pass pipeline callback, `-funroll-loops` at `-O1`, and R5900 partial/runtime unroll. On the console the partial unroll alone made the `-O2` bench functions 1.70x GCC's size; without the tuning they are 0.88x. Bench on the console with this branch (sdk-core `tests/bench`, all checksums match GCC), LLVM / GCC cycles: | Kernel | -O1 | -O2 | -O3 | -Os | |---|---|---|---|---| | packet_build | 0.84 | 1.11 | 1.05 | 0.49 | | physics | 1.52 | 1.16 | 2.05 | 1.04 | | raster | 1.06 | 1.18 | 1.01 | 0.87 | | other five kernels | 0.64-1.00 | 0.80-1.00 | 0.77-1.00 | 0.74-1.01 | Total function size LLVM / GCC: 0.96, 0.88, 0.78, 0.92. The speed gaps are documented as exceptions in the bench README (PS2-101). 🤖 Generated with [Claude Code](https://claude.com/claude-code) 
The pass argument matched the -mips-split-hi16 option, so opt aborted with a duplicate option registration.

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
UPDATING.md: state what belongs in the fork
Some checks failed
action.yml / UPDATING.md: state what belongs in the fork (push) Failing after 0s
Details
release-binaries-all.yml / UPDATING.md: state what belongs in the fork (push) Failing after 0s
Details
release-binaries.yml / UPDATING.md: state what belongs in the fork (push) Failing after 0s
Details
release-documentation.yml / UPDATING.md: state what belongs in the fork (push) Failing after 0s
Details
release-sources.yml / UPDATING.md: state what belongs in the fork (push) Failing after 0s
Details
Check for private emails used in PRs / validate_email (pull_request) Has been skipped
Details
action.yml / UPDATING.md: state what belongs in the fork (pull_request) Failing after 0s
Details
Check LLVM ABI annotations / Check LLVM_ABI annotations with ids (pull_request) Has been skipped
Details
CI Checks / Build and Test Linux (pull_request) Has been skipped
Details
CI Checks / Build and Test Windows (pull_request) Has been skipped
Details
CI Checks / Compute macOS Projects (pull_request) Has been skipped
Details
CI Checks / Build and Test macOS arm64 (pull_request) Has been skipped
Details
release-binaries-all.yml / UPDATING.md: state what belongs in the fork (pull_request) Failing after 0s
Details
release-binaries.yml / UPDATING.md: state what belongs in the fork (pull_request) Failing after 0s
Details
release-documentation.yml / UPDATING.md: state what belongs in the fork (pull_request) Failing after 0s
Details
release-sources.yml / UPDATING.md: state what belongs in the fork (pull_request) Failing after 0s
Details
Add buildbot information to first PRs from new contributors / buildbot_comment (pull_request_target) Has been skipped
Details
Labelling new pull requests / greeter (pull_request_target) Has been cancelled
Details
Labelling new pull requests / automate-prs-labels (pull_request_target) Has been cancelled
Details
f6c4ab0141 
Only PS2-CPU-specific changes: target code gated on the R5900 or
R3000A, runtime code gated on __ps2sdk__ or _MIPS_ARCH_R5900, and
hidden options that default to off. Generic pass edits, pipeline
changes, driver -O changes and benchmark-tuned thresholds stay out.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
techwritescode deleted branch r5900-split-hi16-name 2026-10-15 16:58:09 +00:00
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
ps2/llvm!13
No description provided.