Open
Conversation
Contributor
There was a problem hiding this comment.
Pull request overview
Optimizes multipass JIT lowering for EVM SIGNEXTEND by replacing the previous “sign-bit-position + generic mask” construction with a more direct word/byte decomposition that better matches the interpreter’s structure and reduces intermediate arithmetic/live ranges.
Changes:
- Rewrote
EVMMirBuilder::handleSignextend()lowering to compute(index / 8, index % 8)and operate on the selected 64-bit limb. - Extracted the sign byte, sign-extended it to
i64, rebuilt the selected limb, and filled higher limbs using the sign. - Kept the existing
index >= 31no-op behavior via the existingNoExtensionselect.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| MInstruction *One = createIntConstInstruction(MirI64Type, 1); | ||
| MInstruction *Const3 = createIntConstInstruction(MirI64Type, 3); | ||
| MInstruction *Const8 = createIntConstInstruction(MirI64Type, 8); | ||
| MInstruction *Const63 = createIntConstInstruction(MirI64Type, 63); |
There was a problem hiding this comment.
Const8 is created but never used in this rewritten lowering, which can trigger unused-variable warnings and slightly undermines the stated goal of reducing intermediates. Remove it (or use it if it was intended for computing SignByteOffset).
Suggested change
| MInstruction *Const63 = createIntConstInstruction(MirI64Type, 63); |
⚡ Performance Regression Check Results✅ Performance Check Passed (interpreter)Performance Benchmark Results (threshold: 25%)
Summary: 194 benchmarks, 0 regressions ✅ Performance Check Passed (multipass)Performance Benchmark Results (threshold: 25%)
Summary: 194 benchmarks, 0 regressions |
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 change optimizes EVM
SIGNEXTENDlowering in the multipass JIT.The previous lowering computed the sign bit position with a wider arithmetic
chain, derived the target 64-bit limb through
/ 64and% 64, then built theextended result from a generic bit-mask formulation. That shape was correct, but
it introduced extra intermediate values and a longer live range around the
selected component, sign mask, and extension path.
This patch rewrites the lowering to follow the interpreter-style decomposition
more directly:
sign_word_index = index / 8sign_byte_index = index % 8i64What Changed
handleSignextend()insrc/compiler/evm_frontend/evm_mir_compiler.cppsmaller word/byte-index based lowering
propagation
index >= 31fast return behaviorWhy This Is Better
and reason about
((index * 8 + 7) / 64, % 64)shape((1 << (bit_offset + 1)) - 1)pathreduce unnecessary copies and register pressure
Dependency Note
This commit is intentionally independent from the earlier u64 fast-path /
constant-folding chain.
In particular, it does not depend on:
perf(evm): add u64 constant fast paths and constant folding ...The
SIGNEXTENDoptimization only changeshandleSignextend()insrc/compiler/evm_frontend/evm_mir_compiler.cppand can be applied cleanly ontop of current
upstream/mainby itself.