Align LLBP-X provider and enable in kmhv3#919
Open
CaoJiaming776 wants to merge 2 commits into
Open
Conversation
added 2 commits
June 24, 2026 13:15
Change-Id: I08c3756fcc972e6d4170eb7fff4dee654a2320a4
Change-Id: Ided6257af49a9bdbddff49ba29a8a1d2d53e8942
📝 WalkthroughWalkthroughIntroduces ChangesBTBLLBPX BTB Direction-Override Predictor
Sequence Diagram(s)sequenceDiagram
participant DecoupledBPU as DecoupledBPUWithBTB
participant BTBLLBPX
participant ContextStore as SetAssociativeStore[Context]
participant PatternStore as SetAssociativeStore[Pattern]
participant PatternBuffer
DecoupledBPU->>BTBLLBPX: putPCHistory(startAddr, history, stagePreds)
loop each BTB entry in stagePreds
BTBLLBPX->>BTBLLBPX: compute contextKey / patternKey
BTBLLBPX->>ContextStore: find(contextKey, contextTag)
ContextStore-->>BTBLLBPX: ContextEntry* (hit) or nullptr
BTBLLBPX->>PatternStore: find(patternKey, patternTag)
PatternStore-->>BTBLLBPX: PatternEntry* (hit) or nullptr
alt enableTiming
BTBLLBPX->>PatternBuffer: findPatternBuffer(key, tag)
PatternBuffer-->>BTBLLBPX: readyTick check result
end
BTBLLBPX->>BTBLLBPX: apply depth/timing eligibility gates
BTBLLBPX->>DecoupledBPU: override direction in stagePreds (if eligible)
end
BTBLLBPX->>BTBLLBPX: store LLBPXMeta in ThreadState
DecoupledBPU->>BTBLLBPX: update(FetchTarget)
BTBLLBPX->>BTBLLBPX: locate BranchMeta per BTB entry
alt provider was used
BTBLLBPX->>PatternStore: updatePattern (saturating counter)
BTBLLBPX->>PatternBuffer: rememberPattern (dirty/readyTick)
else misprediction or missing context/pattern
BTBLLBPX->>ContextStore: allocate(contextKey, contextTag)
BTBLLBPX->>PatternStore: allocate(patternKey, patternTag)
BTBLLBPX->>PatternBuffer: rememberPattern
end
Estimated code review effort🎯 5 (Critical) | ⏱️ ~120 minutes Possibly related PRs
Suggested labels
Suggested reviewers
Poem
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches📝 Generate docstrings
🧪 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 |
Contributor
There was a problem hiding this comment.
Pull request overview
This PR introduces a new BTB-side direction override component (LLBP-X) into the decoupled BTB branch predictor stack, extends the predictor metadata capacity to accommodate the added component, and enables LLBP-X in the XiangShan kmhv3 configuration (with an optional timing gate).
Changes:
- Add the
BTBLLBPXpredictor component (C++ implementation + SimObject params) and integrate it intoDecoupledBPUWithBTB. - Expand per-stream predictor metadata capacity (and add a component-count guard) to support additional BTB predictor components.
- Enable LLBP-X in
configs/example/kmhv3.pyand add CLI options related to LLBP-X timing.
Reviewed changes
Copilot reviewed 11 out of 11 changed files in this pull request and generated 3 comments.
Show a summary per file
| File | Description |
|---|---|
src/cpu/pred/SConscript |
Registers BTBLLBPX SimObject, builds btb_llbpx.cc, and adds LLBPX debug flag. |
src/cpu/pred/btb/llbpx_cache.hh |
Adds a small set-associative store utility used by LLBP-X. |
src/cpu/pred/btb/decoupled_bpred.hh |
Wires BTBLLBPX into the decoupled BTB predictor composition. |
src/cpu/pred/btb/decoupled_bpred.cc |
Adds LLBP-X to component list, introduces component-count guard, and stores per-component metadata. |
src/cpu/pred/btb/common.hh |
Extends TageInfoForMGSC provider fields; increases FetchTarget::predMetas capacity via MaxBTBPredComponents. |
src/cpu/pred/btb/btb_tage.cc |
Populates new provider-related fields for downstream consumers (e.g., LLBP-X). |
src/cpu/pred/btb/btb_llbpx.hh |
Declares the new BTBLLBPX predictor component. |
src/cpu/pred/btb/btb_llbpx.cc |
Implements LLBP-X lookup/override, timing gate, recovery, and update/allocation logic. |
src/cpu/pred/BranchPredictor.py |
Adds the BTBLLBPX SimObject and plugs it into DecoupledBPUWithBTB params. |
configs/example/kmhv3.py |
Enables LLBP-X by default in kmhv3 and optionally enables its timing gate via CLI. |
configs/common/xiangshan.py |
Adds CLI flags related to LLBP-X enable/timing. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+890
to
+903
| parser.add_argument( | ||
| "--enable-llbpx", | ||
| "--enable-llbp", | ||
| action="store_true", | ||
| default=False, | ||
| help="Enable the experimental BTB-side LLBP-X direction override component", | ||
| ) | ||
| parser.add_argument( | ||
| "--enable-llbpx-timing", | ||
| "--enable-llbp-timing", | ||
| action="store_true", | ||
| default=False, | ||
| help="Enable the LLBP-X Pattern Buffer timing gate", | ||
| ) |
Comment on lines
+107
to
+111
| BTBLLBPX::putPCHistory(Addr startAddr, | ||
| const boost::dynamic_bitset<> &history, | ||
| std::vector<FullBTBPrediction> &stagePreds) | ||
| { | ||
| if (stagePreds.empty()) { |
Comment on lines
+94
to
+96
| panic_if(numComponents > MaxBTBPredComponents, | ||
| "BTB predictor component count %u exceeds predMetas capacity %u", | ||
| numComponents, MaxBTBPredComponents); |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 5
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@configs/example/kmhv3.py`:
- Around line 113-115: The kmhv3 branch predictor setup currently hard-codes
llbpx on, so the new CLI toggle is ignored. Update the predictor initialization
in the kmhv3 config so `cpu.branchPred.llbpx.enabled` is driven by the parsed
args flag (the same one used for `enable_llbpx_timing`), and keep the timing
option tied to that same `args` value in the `cpu.branchPred.llbpx` block.
In `@src/cpu/pred/btb/btb_llbpx.cc`:
- Around line 423-434: Snapshot the old prediction in BTBLLBPX::updatePattern
before calling updateCounter, because pattern->taken() is currently read after
the counter is modified and can misclassify just-corrected weak states as
correct. Store the pre-update taken/not-taken result in a local variable, use
that saved value for the confidence adjustment, and keep the rest of the logic
in updatePattern and replacementScore unchanged.
- Around line 141-145: The LLBPX gating in lookupHelper() is using the wrong
TAGE depth field, so it may reject valid overrides when TAGE falls back from the
main provider to alt/base. Update the base provider depth lookup in btb_llbpx.cc
to read tage_final_provider_table from stagePred.tageInfoForMgscs instead of
tage_provider_table, keeping the existing branchPC lookup and
providerDepthEligible comparison intact.
- Around line 347-350: The recoverPHist() override in BTBLLBPX currently does
not restore threadState[tid].rcr, leaving the recent-control-record state
speculative after a squash when path-history recovery is used. Update
recoverPHist() to mirror the RCR restoration done in recoverHist(), using the
provided history/entry/update data so threadState[tid].rcr is brought back to
the correct committed state before the next contextKey() call.
- Around line 82-86: The BTBLLBPX initialization currently validates tagBits,
keyBits, and patternBufferSize but never checks counterBits, so invalid values
can break updateCounter() with an illegal shift or overflow the short saturation
range. Add a panic_if guard near the existing constructor validation in BTBLLBPX
to require counterBits be within a safe positive bound before any call sites
that rely on updateCounter(), and make sure the check is placed alongside the
other parameter validations in btb_llbpx.cc.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: 6e323617-b9b2-4ceb-95bd-dd48e57b2ae1
📒 Files selected for processing (11)
configs/common/xiangshan.pyconfigs/example/kmhv3.pysrc/cpu/pred/BranchPredictor.pysrc/cpu/pred/SConscriptsrc/cpu/pred/btb/btb_llbpx.ccsrc/cpu/pred/btb/btb_llbpx.hhsrc/cpu/pred/btb/btb_tage.ccsrc/cpu/pred/btb/common.hhsrc/cpu/pred/btb/decoupled_bpred.ccsrc/cpu/pred/btb/decoupled_bpred.hhsrc/cpu/pred/btb/llbpx_cache.hh
Comment on lines
+113
to
+115
| cpu.branchPred.llbpx.enabled = True | ||
| cpu.branchPred.llbpx.enableTiming = bool( | ||
| getattr(args, 'enable_llbpx_timing', False)) |
There was a problem hiding this comment.
🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win
Wire --enable-llbpx into the kmhv3 predictor toggle.
Line 113 hard-codes cpu.branchPred.llbpx.enabled = True, so the new --enable-llbpx/--enable-llbp flag has no effect.
Suggested fix
- cpu.branchPred.llbpx.enabled = True
+ cpu.branchPred.llbpx.enabled = bool(
+ getattr(args, 'enable_llbpx', False))
cpu.branchPred.llbpx.enableTiming = bool(
getattr(args, 'enable_llbpx_timing', False))📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| cpu.branchPred.llbpx.enabled = True | |
| cpu.branchPred.llbpx.enableTiming = bool( | |
| getattr(args, 'enable_llbpx_timing', False)) | |
| cpu.branchPred.llbpx.enabled = bool( | |
| getattr(args, 'enable_llbpx', False)) | |
| cpu.branchPred.llbpx.enableTiming = bool( | |
| getattr(args, 'enable_llbpx_timing', False)) |
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@configs/example/kmhv3.py` around lines 113 - 115, The kmhv3 branch predictor
setup currently hard-codes llbpx on, so the new CLI toggle is ignored. Update
the predictor initialization in the kmhv3 config so
`cpu.branchPred.llbpx.enabled` is driven by the parsed args flag (the same one
used for `enable_llbpx_timing`), and keep the timing option tied to that same
`args` value in the `cpu.branchPred.llbpx` block.
Comment on lines
+82
to
+86
| panic_if(tagBits == 0 || tagBits >= 64, "BTBLLBPX tagBits must be in [1, 63]"); | ||
| panic_if(keyBits == 0 || keyBits >= 64, "BTBLLBPX keyBits must be in [1, 63]"); | ||
| panic_if(enableTiming && patternBufferSize == 0, | ||
| "BTBLLBPX timing mode requires a non-zero patternBufferSize"); | ||
| warn_if(adaptCtxDepth, "BTBLLBPX Phase 1 ignores adaptive context depth"); |
There was a problem hiding this comment.
🩺 Stability & Availability | 🟠 Major
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
set -euo pipefail
echo "Inspect BTBLLBPX parameter definitions and uses:"
fd -a 'BranchPredictor.py|btb_llbpx.hh|btb_llbpx.cc' .
echo
echo "Search for counterBits declarations/defaults/validation:"
rg -n -C3 '\bcounterBits\b|BTBLLBPX' src configsRepository: OpenXiangShan/GEM5
Length of output: 19375
🏁 Script executed:
#!/bin/bash
set -euo pipefail
echo "BTBLLBPX header (parameter/member definitions):"
sed -n '180,230p' src/cpu/pred/btb/btb_llbpx.hh
echo
echo "BTBLLBPX constructor and counter update implementation:"
sed -n '65,90p' src/cpu/pred/btb/btb_llbpx.cc
echo
sed -n '549,565p' src/cpu/pred/btb/btb_llbpx.cc
echo
echo "BTBLLBPX Python params:"
sed -n '1122,1145p' src/cpu/pred/BranchPredictor.pyRepository: OpenXiangShan/GEM5
Length of output: 4879
🏁 Script executed:
#!/bin/bash
set -euo pipefail
python3 - <<'PY'
# Read-only probe of the shift expression's edge cases.
# This mirrors the source expression logically: max = (1 << (counterBits - 1)) - 1
# and min = -(1 << (counterBits - 1))
for bits in [0, 1, 2, 3, 31, 32, 63, 64]:
try:
v = 1 << (bits - 1)
print(bits, v.bit_length())
except Exception as e:
print(bits, type(e).__name__, e)
PYRepository: OpenXiangShan/GEM5
Length of output: 226
🏁 Script executed:
#!/bin/bash
set -euo pipefail
echo "Search for any validation or assumptions on BTBLLBPX counterBits:"
rg -n -C2 'counterBits|Pattern counter width|panic_if|fatal_if|warn_if' src/cpu/pred/btb src/cpu/pred/BranchPredictor.py
echo
echo "Inspect the generated param type declaration, if present:"
fd -a 'BTBLLBPX.hh|BTBLLBPX.py|BTBLLBPX.*' build src params 2>/dev/null || trueRepository: OpenXiangShan/GEM5
Length of output: 5541
🏁 Script executed:
#!/bin/bash
set -euo pipefail
python3 - <<'PY'
# Probe the exact problematic ranges for the source expression.
# We keep this read-only and do not import or execute repository code.
def expr(bits):
return (1 << (bits - 1)) - 1, -(1 << (bits - 1))
for bits in [0, 1, 2, 3, 15, 16, 17, 31, 32, 33]:
try:
print(bits, expr(bits))
except Exception as e:
print(bits, type(e).__name__, e)
PYRepository: OpenXiangShan/GEM5
Length of output: 364
Guard counterBits before updateCounter()
counterBits has no validation. 0 makes (1 << (counterBits - 1)) a negative shift, and larger values overflow the short saturation bounds.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@src/cpu/pred/btb/btb_llbpx.cc` around lines 82 - 86, The BTBLLBPX
initialization currently validates tagBits, keyBits, and patternBufferSize but
never checks counterBits, so invalid values can break updateCounter() with an
illegal shift or overflow the short saturation range. Add a panic_if guard near
the existing constructor validation in BTBLLBPX to require counterBits be within
a safe positive bound before any call sites that rely on updateCounter(), and
make sure the check is placed alongside the other parameter validations in
btb_llbpx.cc.
Source: Linters/SAST tools
Comment on lines
+141
to
+145
| int baseProviderHistIdx = -2; | ||
| auto tageInfoIt = stagePred.tageInfoForMgscs.find(branchPC); | ||
| if (tageInfoIt != stagePred.tageInfoForMgscs.end()) { | ||
| baseProviderHistIdx = tageInfoIt->second.tage_provider_table; | ||
| } |
There was a problem hiding this comment.
🎯 Functional Correctness | 🟠 Major | ⚡ Quick win
Use TAGE’s final provider depth here, not the main provider depth.
lookupHelper() now publishes both tage_provider_table and tage_final_provider_table, but this path still gates LLBPX against tage_provider_table. When TAGE falls back to alt/base, Line 144 makes providerDepthEligible compare against the wrong source and can reject overrides that should be legal.
Suggested fix
- int baseProviderHistIdx = -2;
+ int baseProviderHistIdx = -2;
auto tageInfoIt = stagePred.tageInfoForMgscs.find(branchPC);
if (tageInfoIt != stagePred.tageInfoForMgscs.end()) {
- baseProviderHistIdx = tageInfoIt->second.tage_provider_table;
+ baseProviderHistIdx =
+ tageInfoIt->second.tage_final_provider_table;
}📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| int baseProviderHistIdx = -2; | |
| auto tageInfoIt = stagePred.tageInfoForMgscs.find(branchPC); | |
| if (tageInfoIt != stagePred.tageInfoForMgscs.end()) { | |
| baseProviderHistIdx = tageInfoIt->second.tage_provider_table; | |
| } | |
| int baseProviderHistIdx = -2; | |
| auto tageInfoIt = stagePred.tageInfoForMgscs.find(branchPC); | |
| if (tageInfoIt != stagePred.tageInfoForMgscs.end()) { | |
| baseProviderHistIdx = | |
| tageInfoIt->second.tage_final_provider_table; | |
| } |
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@src/cpu/pred/btb/btb_llbpx.cc` around lines 141 - 145, The LLBPX gating in
lookupHelper() is using the wrong TAGE depth field, so it may reject valid
overrides when TAGE falls back from the main provider to alt/base. Update the
base provider depth lookup in btb_llbpx.cc to read tage_final_provider_table
from stagePred.tageInfoForMgscs instead of tage_provider_table, keeping the
existing branchPC lookup and providerDepthEligible comparison intact.
Comment on lines
+347
to
+350
| BTBLLBPX::recoverPHist(const boost::dynamic_bitset<> &history, | ||
| const FetchTarget &entry, | ||
| const PathHistoryUpdate &update) | ||
| { |
There was a problem hiding this comment.
🎯 Functional Correctness | 🟠 Major | ⚡ Quick win
Recover the RCR in path-history mode too.
This empty override leaves threadState[tid].rcr on speculative state after a squash when recovery flows through recoverPHist() instead of recoverHist(). The next contextKey() then hashes the wrong recent-control record sequence.
Suggested fix
void
BTBLLBPX::recoverPHist(const boost::dynamic_bitset<> &history,
const FetchTarget &entry,
const PathHistoryUpdate &update)
{
+ auto meta = std::static_pointer_cast<LLBPXMeta>(
+ entry.predMetas[getComponentIdx()]);
+ if (!meta || entry.tid >= threadState.size()) {
+ return;
+ }
+
+ restoreRCR(entry.tid, *meta);
+ if (entry.exeTaken) {
+ pushRCR(entry.tid, entry.exeBranchInfo, true);
+ }
+#ifndef UNIT_TEST
+ llbpxStats.rcrRecover++;
+#endif
}📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| BTBLLBPX::recoverPHist(const boost::dynamic_bitset<> &history, | |
| const FetchTarget &entry, | |
| const PathHistoryUpdate &update) | |
| { | |
| BTBLLBPX::recoverPHist(const boost::dynamic_bitset<> &history, | |
| const FetchTarget &entry, | |
| const PathHistoryUpdate &update) | |
| { | |
| auto meta = std::static_pointer_cast<LLBPXMeta>( | |
| entry.predMetas[getComponentIdx()]); | |
| if (!meta || entry.tid >= threadState.size()) { | |
| return; | |
| } | |
| restoreRCR(entry.tid, *meta); | |
| if (entry.exeTaken) { | |
| pushRCR(entry.tid, entry.exeBranchInfo, true); | |
| } | |
| `#ifndef` UNIT_TEST | |
| llbpxStats.rcrRecover++; | |
| `#endif` | |
| } |
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@src/cpu/pred/btb/btb_llbpx.cc` around lines 347 - 350, The recoverPHist()
override in BTBLLBPX currently does not restore threadState[tid].rcr, leaving
the recent-control-record state speculative after a squash when path-history
recovery is used. Update recoverPHist() to mirror the RCR restoration done in
recoverHist(), using the provided history/entry/update data so
threadState[tid].rcr is brought back to the correct committed state before the
next contextKey() call.
Comment on lines
+423
to
+434
| BTBLLBPX::updatePattern(const BranchMeta &meta, bool actualTaken) | ||
| { | ||
| auto *pattern = patterns.find(meta.key, meta.patternTag); | ||
| if (!pattern) { | ||
| return; | ||
| } | ||
| updateCounter(actualTaken, pattern->counter); | ||
| if (pattern->taken() == actualTaken && pattern->confidence < 15) { | ||
| pattern->confidence++; | ||
| } else if (pattern->confidence > 0) { | ||
| pattern->confidence--; | ||
| } |
There was a problem hiding this comment.
🎯 Functional Correctness | 🟠 Major | ⚡ Quick win
Snapshot the old prediction before updating the counter.
Line 430 checks pattern->taken() after updateCounter(). A weak counter that was wrong and just crossed zero will be treated as “correct” and gain confidence, which skews replacementScore() toward recently corrected mistakes.
Suggested fix
void
BTBLLBPX::updatePattern(const BranchMeta &meta, bool actualTaken)
{
auto *pattern = patterns.find(meta.key, meta.patternTag);
if (!pattern) {
return;
}
+ const bool predictedTaken = pattern->taken();
updateCounter(actualTaken, pattern->counter);
- if (pattern->taken() == actualTaken && pattern->confidence < 15) {
+ if (predictedTaken == actualTaken && pattern->confidence < 15) {
pattern->confidence++;
} else if (pattern->confidence > 0) {
pattern->confidence--;
}📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| BTBLLBPX::updatePattern(const BranchMeta &meta, bool actualTaken) | |
| { | |
| auto *pattern = patterns.find(meta.key, meta.patternTag); | |
| if (!pattern) { | |
| return; | |
| } | |
| updateCounter(actualTaken, pattern->counter); | |
| if (pattern->taken() == actualTaken && pattern->confidence < 15) { | |
| pattern->confidence++; | |
| } else if (pattern->confidence > 0) { | |
| pattern->confidence--; | |
| } | |
| BTBLLBPX::updatePattern(const BranchMeta &meta, bool actualTaken) | |
| { | |
| auto *pattern = patterns.find(meta.key, meta.patternTag); | |
| if (!pattern) { | |
| return; | |
| } | |
| const bool predictedTaken = pattern->taken(); | |
| updateCounter(actualTaken, pattern->counter); | |
| if (predictedTaken == actualTaken && pattern->confidence < 15) { | |
| pattern->confidence++; | |
| } else if (pattern->confidence > 0) { | |
| pattern->confidence--; | |
| } |
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@src/cpu/pred/btb/btb_llbpx.cc` around lines 423 - 434, Snapshot the old
prediction in BTBLLBPX::updatePattern before calling updateCounter, because
pattern->taken() is currently read after the counter is modified and can
misclassify just-corrected weak states as correct. Store the pre-update
taken/not-taken result in a local variable, use that saved value for the
confidence adjustment, and keep the rest of the logic in updatePattern and
replacementScore unchanged.
🚀 Coremark Smoke Test Results
✅ Difftest smoke test passed! |
|
[Generated by GEM5 Performance Robot] Align BTB PerformanceOverall Score
[Generated by GEM5 Performance Robot] Align BTB PerformanceOverall Score
|
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
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.
This branch adds an experimental BTB-side LLBP-X direction override component to the XiangShan kmhv3 branch predictor flow, and enables it by default in
kmhv3.py.The implementation is inspired by the upstream LLBP-X project:
This is not a direct patch import. The upstream gem5 integration targets a different predictor structure, while this repository uses the
DecoupledBPUWithBTBcomponent chain. This branch ports the main LLBP-X ideas into the local BTB predictor framework.Summary by CodeRabbit
New Features
Bug Fixes