expr: add stack-machine compiled evaluation for MirScalarExpr #34947
Conversation
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
antiguru
force-pushed
the
claude/heuristic-pare
branch
2 times, most recently
from
735b806 to
afd4bb6
Compare
Introduce `CompiledMirScalarExpr`, a flat instruction sequence compiled
from `MirScalarExpr` trees. This replaces recursive AST descent during
evaluation with a linear instruction stream using a value stack and
program counter, improving evaluation performance.
Key changes:
1. Split `LazyUnaryFunc::eval` into `eval` + `eval_input`, where
`eval_input` takes a pre-evaluated `Result<Datum, EvalError>` instead
of requiring expression evaluation. Updated all 27 direct
`LazyUnaryFunc` implementations.
2. Add `embedded_exprs()` / `embedded_exprs_mut()` methods to
`LazyUnaryFunc` for the 9 compound-cast functions that contain
embedded `MirScalarExpr` sub-expressions.
3. Extract `VariadicFunc::eval_eager()` from `VariadicFunc::eval()` for
calling eager variadic functions with pre-evaluated datum slices.
4. Implement `CompiledMirScalarExpr` in `static_eval.rs` with:
- Flat instruction sequence with label-based compilation
- Inline evaluation for unary (via `eval_input`) and binary
(via static column references) functions
- Sub-program compilation for variadic operands and compound-cast
inner expressions (MapListElements, MapArrayElements, MapRecord,
MapListToJsonb, MapArrayToJsonb, ParseAndCast)
- Correct If/else handling with two-offset SkipIfNotTrue for proper
error propagation
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace the CallVariadic instruction (which compiled each operand into a sub-program) with 7 dedicated inline instructions: AndStep, OrStep, SkipIfNotNull, GreatestStep, LeastStep, RaiseIfNullError, and CallEagerVariadic. This eliminates sub-program indirection for all variadic functions and enables short-circuit branching for And, Or, Coalesce, Greatest, Least, and ErrorIfNull. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Pre-compute max_stack_depth during compilation and expose eval_with_stack so callers can amortize the stack allocation across rows. Also optimize accumulator patterns (And/Or/Greatest/Least) to use in-place mutation instead of pop+push, remove unused return_ty from MapRecord, and add benchmarks comparing tree vs compiled eval. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Mirror the existing LazyUnaryFunc::eval_input pattern for binary functions. Add eval_input to LazyBinaryFunc trait (taking pre-evaluated Result<Datum, EvalError> inputs), update the EagerBinaryFunc blanket impl, and add a hand-written BinaryFunc::eval_input dispatch method. This lets the stack machine in static_eval call binary functions directly with pre-evaluated datums instead of going through fake Column(0)/Column(1) expression references and a temporary datum slice. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Adds a benchmark for a 1024-deep left-leaning add chain ((((1 + 1) + 1) + 1) ... + 1) to measure compiled vs tree eval on deeply nested expressions where recursion overhead dominates. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
antiguru
force-pushed
the
claude/heuristic-pare
branch
from
afd4bb6 to
d2fe34e
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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
LazyUnaryFunc::evalintoeval+eval_inputso unary functions can be called with pre-evaluated datums, avoiding re-evaluation in a stack machine contextembedded_exprs()/embedded_exprs_mut()toLazyUnaryFuncfor the 9 compound-cast functions that contain embeddedMirScalarExprsub-expressionsVariadicFunc::eval_eager()for calling eager variadic functions with pre-evaluated datum slicesCompiledMirScalarExpr— a flat instruction sequence compiled fromMirScalarExprtrees that replaces recursive AST descent with a linear instruction stream using a value stack and program counterDetails
The
CompiledMirScalarExprhandles allMirScalarExprvariants:eval_inputTest plan
cargo check -p mz-expr -p mz-transform -p mz-computebuilds cleanly🤖 Generated with Claude Code