perf(compiler): add EVM peepholes and branch shaping - #435
Merged
Conversation
Contributor
There was a problem hiding this comment.
Pull request overview
This PR adds a set of EVM frontend and x86 backend peephole optimizations aimed at reducing redundant MIR materialization (notably around ISZERO/NOT) and producing tighter x86 compare/branch sequences and short-diamond branch layouts.
Changes:
- Add EVM MIR builder constant-identity folds (e.g.,
+0,*0/*1,&0,&~0,|0,^0) and introduce deferred operand kinds for bitwise-NOT and zero-test to enable collapsingISZEROchains. - Materialize deferred values only when needed by teaching
extractU256Operand()/ argument conversion paths to realize deferred operands and by extendinghandleCompareEQZ()with a negation mode. - Improve x86 output by (a) folding
cmp/test -> setcc -> [movzx] -> test -> jneinto a directjcc, and (b) shaping “short diamond” branches to prefer fallthrough to the true block.
Reviewed changes
Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.
| File | Description |
|---|---|
| src/compiler/target/x86/x86lowering.cpp | Adds short-diamond detection and inverts branch emission to make the true block fall through when safe. |
| src/compiler/target/x86/x86_cg_peephole.cpp | Extends compare-chain peephole to handle optional movzx and shortens compare+branch sequences. |
| src/compiler/evm_frontend/evm_mir_compiler.h | Introduces deferred operand kinds and constant helper predicates; adds identity folds in arithmetic/bitwise handlers; updates handleCompareEQZ() signature. |
| src/compiler/evm_frontend/evm_mir_compiler.cpp | Implements MUL identity folds, deferred NOT/ISZERO materialization, and realizes deferred operands in extraction and arg-lowering paths. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
abmcar
marked this pull request as draft
March 27, 2026 09:43
⚡ 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 |
Contributor
There was a problem hiding this comment.
Pull request overview
Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
abmcar
force-pushed
the
perf/evm-peephole-rules
branch
from
April 8, 2026 11:36
6f3f6fb to
28b6771
Compare
Contributor
There was a problem hiding this comment.
Pull request overview
Copilot reviewed 7 out of 7 changed files in this pull request and generated 1 comment.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
…st_cast Add begin()/end() const overloads to MBasicBlock so read-only queries can iterate statements without casting away const. Remove the const_cast in isShortDiamondTrueFallthrough() that was only needed because const iterators were missing. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The compare-chain peephole (cmp->setcc->[movzx]->test->jne => cmp->jcc) was erasing SETCC/MOVZX instructions without checking if their virtual registers had other uses. The lowering cache (_expr_reg_map) can share these registers across consumers, so erasing them could leave dangling references. Add hasOneNonDBGUse guards to bail out when the registers are still live. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…nd enum Replace the separate bool field with ZERO_TEST_EQ/ZERO_TEST_NE enum variants, eliminating a redundant state that could silently be inert when DeferredValueKind was BITWISE_NOT. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
abmcar
force-pushed
the
perf/evm-peephole-rules
branch
from
April 10, 2026 13:58
d781522 to
e39b95f
Compare
This was referenced Apr 13, 2026
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.
Summary
ADD,SUB,MUL,AND,OR, andXORNOTand zero-test materialization soISZEROchains can collapse without building redundant 256-bit MIR valuesmovzx/testsequencesImplementation
src/compiler/evm_frontend/evm_mir_compiler.hOperandsrc/compiler/evm_frontend/evm_mir_compiler.cppMULzero/one fast pathshandleCompareEQZ()to emit eithereqornefor deferred zero testsU256Instis actually neededsrc/compiler/target/x86/x86_cg_peephole.cppcmp/test -> setcc -> [movzx] -> test -> jneinto a shorter compare/branch formsrc/compiler/target/x86/x86lowering.cppValidation
tools/format.sh check./build/evmStateTests1798/1798Benchmark
Compared with the same-session
fix2baseline onevmone-benchexternal totals:11.4%3.8%Representative wins:
external/total/main/jump_around/empty: about48%fasterexternal/total/main/memory_grow_mload/by32: improvedexternal/total/main/memory_grow_mstore/by16: improvedKnown remaining hot spots:
JUMPDEST_n0loop_with_many_jumpdests