Conversation
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
📝 WalkthroughWalkthroughDecoupledBPUWithBTB gains a two-fetch mode and max-fetch-bytes config; tick() can produce up to two predictions per cycle. Fetch path now supports an in-cycle 2‑fetch extension, preserves/keeps the next FSQ entry buffered, and changes the fetch-stop semantics. Changes
Sequence Diagram(s)sequenceDiagram
participant Core as DecoupledBPUWithBTB
participant Predictor as BTB/TAGE
participant FTQ as FTQ
participant FSQ as FSQ
participant Fetch as FetchUnit
Note over Core: Up to N = (enable2Fetch ? 2 : 1) iterations per tick
loop per-prediction
Core->>Predictor: requestPrediction()
Predictor-->>Core: provisionalPrediction
Core->>Predictor: generateFinalPrediction()
Predictor-->>Core: finalPrediction (+overrideBubbles)
Core->>FTQ: ftqNext / getTarget()
Core->>FSQ: enqueue(finalPrediction)
alt do 2-fetch
Core->>Fetch: signal do_2fetch (keep next FSQ entry buffered)
Fetch-->>Core: continue fetch without consuming FTQ entry
else single fetch
Core->>Fetch: consume FTQ target / mark used
end
end
Estimated code review effort🎯 4 (Complex) | ⏱️ ~45 minutes Possibly related PRs
Suggested reviewers
Poem
🚥 Pre-merge checks | ✅ 2 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
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. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 1
🤖 Fix all issues with AI agents
In `@src/cpu/pred/btb/decoupled_bpred.cc`:
- Line 140: The local variable tempNumOverrideBubbles is declared but never
used; remove the unused declaration of tempNumOverrideBubbles from the scope
where it's defined (the function containing the line "unsigned
tempNumOverrideBubbles = 0;") to clean up dead code and avoid compiler warnings,
ensuring no other references to tempNumOverrideBubbles remain in the function.
🧹 Nitpick comments (2)
src/cpu/pred/btb/decoupled_bpred.hh (1)
162-162: Consider makingenableTwoTakenconfigurable via params.This feature flag is hardcoded to
truewith no way to disable it through simulation parameters. Other similar options likefetchStreamQueueSize,predictWidth, andresolveBlockThresholdare initialized from theParamsobject in the constructor. For flexibility during experimentation and for consistency with the existing pattern, consider adding this to theDecoupledBPUWithBTBParams.src/cpu/pred/btb/decoupled_bpred.cc (1)
142-184: Add documentation clarifying the multi-prediction loop behavior.The loop logic for producing up to 2 predictions per tick is non-trivial. Consider adding a comment explaining:
- The intended behavior when both predictions succeed (no bubbles)
- What happens when the first prediction generates override bubbles (second iteration essentially becomes a no-op)
- The interaction between
numOverrideBubblesbeing set inside the loop but decremented only once outsideThis will help future maintainers understand the "ideal 2-taken" semantics.
📝 Suggested documentation
int predsRemainsToBeMade = enableTwoTaken ? 2 : 1; - unsigned tempNumOverrideBubbles = 0; + // "Ideal 2-taken" mode: attempt up to 2 predictions per tick. + // - If the first prediction generates override bubbles, the second iteration + // will be blocked by validateFSQEnqueue() until bubbles are consumed. + // - If no bubbles, both predictions can be enqueued in a single tick. + // - Bubble decrement happens once per tick after the loop. while (predsRemainsToBeMade > 0) {
🚀 Coremark Smoke Test Results
✅ Difftest smoke test passed! |
|
🚀 Performance test triggered: spec06-0.8c |
|
[Generated by GEM5 Performance Robot] Ideal BTB PerformanceOverall Score
|
Yakkhini
force-pushed
the
2-taken-ideal
branch
from
0d664b4 to
ebf7386
Compare
🚀 Coremark Smoke Test Results
✅ Difftest smoke test passed! |
|
🚀 Performance test triggered: spec06-0.8c |
|
[Generated by GEM5 Performance Robot] Ideal BTB PerformanceOverall Score
|
Yakkhini
force-pushed
the
2-taken-ideal
branch
from
ebf7386 to
9406572
Compare
|
🚀 Performance test triggered: spec06-0.8c |
🚀 Coremark Smoke Test Results
✅ Difftest smoke test passed! |
|
[Generated by GEM5 Performance Robot] Ideal BTB PerformanceOverall Score
|
|
🚀 Performance test triggered: spec06-0.8c |
Yakkhini
force-pushed
the
2-taken-ideal
branch
from
398987b to
0f5a594
Compare
|
🚀 Performance test triggered: spec06-0.8c |
Yakkhini
force-pushed
the
2-taken-ideal
branch
from
0f5a594 to
0542507
Compare
|
🚀 Performance test triggered: spec06-0.8c |
2026-03-05 Co-Analysis (New vs Ref)1) GoalThis report re-analyzes the latest 2026-03-05 results with reference to answer:
Also included as supplemental findings:
2) Data Sources
Only rows with valid numeric 3) Executive Summary
4) Core Question AnswersQ1. Is frontend bandwidth bound already gone?No. It is much smaller, but not eliminated. Evidence:
Q2. Has bottleneck fully shifted to branch misprediction + backend?Partially, but not fully. TopDown dominant category counts:
Interpretation:
Q3. Is there still improvement ceiling?Yes. Using a rough bound for new data:
This is optimistic and non-orthogonal, but confirms FE bandwidth is not yet fully exhausted. 5) What Changed (New vs Ref)5.1 Throughput and frontend supply
These strongly indicate 2Taken/2Fetch improves frontend supply and feed continuity. 5.2 Trade-offs and side-effects
Interpretation:
5.3 Queue-pressure signals (important)
Normalized examples (new, per 1k committed insts):
This suggests structural queue pressure remains a likely next limiter. 6) Workload-Level PatternsHigh IPC gain and strong FE reduction:
Low/near-flat gains:
Slight regression:
Correlation hint:
7) Practical Conclusions
8) Recommended Next Tuning Priorities
9) Notes
|
Change-Id: I39d54a0621d139cc00a156b02a6d7d888d9b15f0 Co-authored-by: Xu Boran <xuboran@bosc.ac.cn>
Change-Id: Ic203a9694c093034744986309e796b9d66d6f826
Change-Id: I3f0f686000b610c3bf842e62c9b9e91e7188a028
Change-Id: Id8896d4c28af5c11e71e3ef453068cd84231e468
🚀 Coremark Smoke Test Results
✅ Difftest smoke test passed! |
Change-Id: Ib34d2a7b889b5e06fed64bb700f0a9779ec54e69
|
🚀 Performance test triggered: gcc12-spec06-0.8c |
🚀 Coremark Smoke Test Results
✅ Difftest smoke test passed! |
|
[Generated by GEM5 Performance Robot] https://github.com/OpenXiangShan/GEM5/actions/runs/23296346286?pr=736 Ideal BTB PerformanceOverall Score
|
Change-Id: I9523d6623d9e1afc95e15d0baeadbc406b18dfeb
🚀 Coremark Smoke Test Results
✅ Difftest smoke test passed! |
Change-Id: Icdab4d15307e9a1b3be947d26cddc2eafb5c904e
🚀 Coremark Smoke Test Results
✅ Difftest smoke test passed! |
|
🚀 Performance test triggered: gcc12-spec06-0.8c |
Change-Id: Ic3b477418bae5150fec9c7491a0c949e63c01f5f
|
https://github.com/OpenXiangShan/GEM5/actions/runs/23481175392?pr=736 ：ideal fetchBuffer |
|
🚀 Performance test triggered: gcc12-spec06-0.8c |
🚀 Coremark Smoke Test Results
✅ Difftest smoke test passed! |
Change-Id: I6a3ea48aac437181b610155028b02af8fa3c700d
|
Close this since further works switch to a detail predictor implementation. This PR aims to a overview estimates of coarse-grained performance on 2-Taken. There is reports on the latest commit. Access it from |
🚀 Coremark Smoke Test Results
✅ Difftest smoke test passed! |
2 participants
Change-Id: I39d54a0621d139cc00a156b02a6d7d888d9b15f0
Summary by CodeRabbit
New Features
Refactor
Chores