Add SSAFunction type, refactor pipeline to accept SSA input#80
Open
dpsanders wants to merge 4 commits into
Open
Add SSAFunction type, refactor pipeline to accept SSA input#80dpsanders wants to merge 4 commits into
dpsanders wants to merge 4 commits into
Conversation
This was referenced Apr 14, 2026
dpsanders
added a commit
that referenced
this pull request
Apr 19, 2026
Five independent fixes plus a registry-wide regression test: - Drop `sign`, `max`, `min` from `unary_functions`. `sign_rev` is not defined anywhere (the registration has always been broken), and `max`/`min` are binary, not unary — ChainRules' adjoint for them synthesises boolean branch selectors (`!(y<x)`) which aren't reversible ops. - Override ChainRules' `abs` adjoint with `x/abs(x)`; the default uses `sign(x)`, whose `_rev_sign` we just removed. - Override ChainRules' `exp2`/`exp10`/`log2`/`log10` adjoints to use plain `Float64` logs instead of the `IrrationalConstants.Logtwo`/ `Logten` that `IntervalArithmetic` refuses to accept. - Extend `IntervalArithmetic.intersect_interval` with `(Real, Interval)` and `(Real, BareInterval)` methods so the constraint intersect works when the output variable comes through as a plain `Real` (e.g. from `d/dx(x^exact(2))`). - Promote integer seeds (`_ā := 1`, unassigned `:= 0`) in the binarized derivative SSA to intervals. Without this, identity-propagating gradients such as `d/dx(x-y) = 1` leave `Int`-valued entries that drive `IntervalContractors`' `mul_rev(a,b,c) = mul_rev(promote(a,b,c)...)` fallback into infinite recursion on `(Int, Int, Int)`. Also fixes PR 81's stale destructuring of `gradient_code` (it now returns an `SSAFunction` rather than a tuple, post-#80). Adds a regression test that loops over every entry in the binary and unary registries and builds+runs a `derivative_contractor` — catches each of these regressions. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Introduce SSAFunction struct to represent decomposed expressions in SSA form (a vector of Assignments + output variable). This enables passing pre-computed SSA (e.g. from gradient_code) into forward_backward_contractor, which is needed for derivative constraints. - cse_equations now returns SSAFunction instead of (code, final) tuple - forward_backward_code/expr/contractor accept SSAFunction - gradient_code accepts SSAFunction - Expression-accepting methods become thin wrappers that call cse_equations - Fix stale istree import (renamed to iscall in newer SymbolicUtils) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
These functions now wrap their output in SSAFunction instead of returning raw (code, final_var, ...) tuples. Also adds a text/plain show method for SSAFunction that displays assignments as `_a := y^2`. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add `variables::NamedTuple` field to SSAFunction for extra named outputs (constraint, gradient). forward_backward_code and gradient_code now return a single SSAFunction instead of a tuple. - Display reverse operations as rev(+)(...) instead of raw generated names. - Bump version to 0.4.2. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Breaking changes to the SSA pipeline (cse_equations, forward_backward_code, gradient_code now return SSAFunction instead of tuples) warrant a minor bump under Julia's 0.x semver convention. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
dpsanders
added a commit
that referenced
this pull request
Apr 19, 2026
Five independent fixes plus a registry-wide regression test: - Drop `sign`, `max`, `min` from `unary_functions`. `sign_rev` is not defined anywhere (the registration has always been broken), and `max`/`min` are binary, not unary — ChainRules' adjoint for them synthesises boolean branch selectors (`!(y<x)`) which aren't reversible ops. - Override ChainRules' `abs` adjoint with `x/abs(x)`; the default uses `sign(x)`, whose `_rev_sign` we just removed. - Override ChainRules' `exp2`/`exp10`/`log2`/`log10` adjoints to use plain `Float64` logs instead of the `IrrationalConstants.Logtwo`/ `Logten` that `IntervalArithmetic` refuses to accept. - Extend `IntervalArithmetic.intersect_interval` with `(Real, Interval)` and `(Real, BareInterval)` methods so the constraint intersect works when the output variable comes through as a plain `Real` (e.g. from `d/dx(x^exact(2))`). - Promote integer seeds (`_ā := 1`, unassigned `:= 0`) in the binarized derivative SSA to intervals. Without this, identity-propagating gradients such as `d/dx(x-y) = 1` leave `Int`-valued entries that drive `IntervalContractors`' `mul_rev(a,b,c) = mul_rev(promote(a,b,c)...)` fallback into infinite recursion on `(Int, Int, Int)`. Also fixes PR 81's stale destructuring of `gradient_code` (it now returns an `SSAFunction` rather than a tuple, post-#80). Adds a regression test that loops over every entry in the binary and unary registries and builds+runs a `derivative_contractor` — catches each of these regressions. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
dpsanders
force-pushed
the
add-ssa-function
branch
from
f1ea4a5 to
4ce0682
Compare
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
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
SSAFunctionstruct to represent decomposed expressions in SSA form (a vector ofAssignments + output variable)cse_equationsnow returnsSSAFunctioninstead of a(code, final)tupleforward_backward_code,forward_backward_expr, andforward_backward_contractoreach get anSSAFunction-accepting method; expression-accepting methods become thin wrappersgradient_codesimilarly acceptsSSAFunctionistreeimport →iscallThis enables passing pre-computed SSA (e.g. from
gradient_code) intoforward_backward_contractor, which is needed for derivative constraints (next PR).Test plan
🤖 Generated with Claude Code