Fix path history info target

[Erlkonigal]

• Align target generation in getPHistInfo with getTarget

Change-Id: I3cf46388e19780eccbaa80656db2f74652b97677

Summary by CodeRabbit

• Bug Fixes

  • Corrected branch target handling for indirect and return branches in prediction so folded path/history uses the overridden targets consistently.

• Tests

  • Added unit tests validating history folding and path-hash behavior with overridden targets, and confirming predictions remain taken—improving reliability of branch-prediction behavior.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 237b2253-5fdf-4e91-84f6-8275bb48c0ef

📥 Commits

Reviewing files that changed from the base of the PR and between bf1efb6 and 382ed1b.

📒 Files selected for processing (3)

• src/cpu/pred/btb/common.hh
• src/cpu/pred/btb/test/btb_mgsc.test.cc
• src/cpu/pred/btb/test/btb_tage.test.cc

🚧 Files skipped from review as they are similar to previous changes (1)

• src/cpu/pred/btb/common.hh

---

📝 Walkthrough

Walkthrough

Refactored taken-target computation in BTB prediction: added getEntryTarget() and updated getTarget()/getPHistInfo() to source indirect-target overrides from indirectTargets (by entry.pc) or returnTarget for indirect returns; added tests exercising folded-history updates with those overrides.

Changes

Cohort / File(s)	Summary
BTB Target Resolution src/cpu/pred/btb/common.hh	Introduced FullBTBPrediction::getEntryTarget(const BTBEntry&); getTarget()/getPHistInfo() now use getEntryTarget() to resolve taken-entry targets, selecting entry.target, indirectTargets[entry.pc], or returnTarget based on isIndirect/isReturn.
BTB MGSC tests src/cpu/pred/btb/test/btb_mgsc.test.cc	Added tests verifying BTBMGSC::specUpdatePHist() uses overridden indirect/return targets from FullBTBPrediction when updating folded history.
BTB TAGE tests src/cpu/pred/btb/test/btb_tage.test.cc	Added TAGE upper-bound path-hash tests ensuring path-hash/folded-history snapshot and predictTAGE behavior when BTB entries override targets via indirectTargets or returnTarget.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Possibly related PRs

• bpu,configs: Add optional BTBTAGE upper-bound mode for kmhv3 #770 — Refactors/specification changes to BTB target selection and specUpdatePHist that overlap this PR's modifications.
• cpu-o3:Implement predwrongSource method #683 — Changes to FullBTBPrediction and specUpdatePHist with related BTB/FetchStream handling.
• Use PHR in ITTAGE #628 — Modifications to specUpdatePHist/target usage affecting indirect/return target semantics.

Suggested labels

align-kmhv3

Suggested reviewers

• jensen-yan
• Yakkhini
• CJ362ff

Poem

🐰 I hop through entries, targets in sight,
Indirect paths and returns made right,
Folded histories shift with care,
Predictions follow me everywhere,
A tiny hop, a cleaner flight.

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title 'Fix path history info target' directly aligns with the PR's stated objective to align target generation in getPHistInfo with getTarget, and accurately reflects the main refactoring of target computation in the BTB prediction logic.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix-phist-target

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[Copilot]

Pull request overview

Aligns path-history (PHR) PC/target selection in FullBTBPrediction::getPHistInfo() with the main fetch-target selection logic in getTarget() to avoid using stale BTB targets for indirect control flow.

Changes:

• Updates getPHistInfo() so unconditional indirect branches can source their target from indirectTargets (ITTAGE) or returnTarget (RAS), matching getTarget().
• Fixes an incorrect comment describing the unconditional-branch PC.

---

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

[github-actions]

🚀 Coremark Smoke Test Results

Branch	IPC	Change
Base (xs-dev)	2.2665	-
This PR	2.2709	📈 +0.0044 (+0.19%)

✅ Difftest smoke test passed!

[github-actions]

🚀 Performance test triggered: gcc12-spec06-0.8c

[jensen-yan]

Thank you very much for your contribution and your PR. This is indeed a serious bug in the PHR update of the model, which may affect the update performance of TAGE, etc. I'm curious how you discovered this bug, whether through testing or code reading. Currently, it seems that XiangShan RTL does not have this bug, and we are evaluating the performance impact of this PR.

[XiangShanRobot]

[Generated by GEM5 Performance Robot]
commit: 382ed1b
workflow: On-Demand SPEC Test (Tier 1.5)

Ideal BTB Performance

Overall Score

	PR	Master	Diff(%)
Score	20.84	20.69	+0.71 🟢

[jensen-yan]

gcc12-spec06-0.8c performance analysis

Compared against the xs-dev Ideal BTB baseline used by the performance robot:

• PR run: 24170827894 (382ed1b5ec397465aa42fbedcf5cca983e68dab4)
• Baseline run: 24170690821 (86825b380fb1d020150e94de658f831c96bbdb98)
• Data processed with python3 run.py ... --slice gcc12

Overall

Metric	Baseline	PR	Delta	Delta %
Int score / GHz	20.6907	20.8375	+0.1468	+0.71%
Total branch wrong MPKI	4.9768	4.8501	-0.1267	-2.55%
Conditional branch MPKI	4.8504	4.7279	-0.1225	-2.53%
Indirect branch MPKI	0.0636	0.0609	-0.0027	-4.29%
Return branch MPKI	0.0122	0.0124	+0.0002	+1.34%

Per-benchmark score / condMPKI

Benchmark	Score (Base)	Score (PR)	Score Delta	Score Delta %	condMPKI (Base)	condMPKI (PR)	condMPKI Delta	condMPKI Delta %
gobmk	15.612	16.143	+0.530	+3.40%	8.829	8.162	-0.666	-7.55%
sjeng	13.169	13.666	+0.497	+3.77%	5.462	4.929	-0.534	-9.77%
gcc	21.177	21.437	+0.260	+1.23%	1.736	1.561	-0.175	-10.08%
h264ref	25.234	25.325	+0.091	+0.36%	0.500	0.476	-0.025	-4.89%
perlbench	16.920	17.004	+0.084	+0.49%	1.184	1.230	+0.046	+3.92%
mcf	34.644	34.675	+0.031	+0.09%	12.709	12.697	-0.012	-0.09%
bzip2	11.265	11.283	+0.017	+0.15%	6.346	6.312	-0.034	-0.54%
omnetpp	21.798	21.807	+0.008	+0.04%	4.086	4.031	-0.056	-1.37%
astar	14.333	14.335	+0.002	+0.02%	16.088	16.083	-0.005	-0.03%
libquantum	46.858	46.858	+0.000	+0.00%	0.043	0.043	+0.000	+0.00%
hmmer	17.065	17.059	-0.006	-0.04%	0.324	0.324	+0.000	+0.00%
xalancbmk	33.961	33.659	-0.301	-0.89%	0.898	0.888	-0.010	-1.11%

Takeaway

• The gain is real and matches the CI robot comment: +0.71% overall Int score on gcc12-spec06-0.8c.
• The improvement is concentrated in gobmk, sjeng, and gcc, and these gains correlate well with lower conditional branch MPKI.
• The aggregate improvement mainly comes from conditional-path prediction quality rather than a large direct improvement in indirect/return prediction. This is consistent with the PHR/path-history fix improving downstream TAGE-style learning instead of only fixing target prediction itself.