Add VACASK device models as integration tests#17
Open
robtaylor wants to merge 23 commits into
Open
Conversation
The compiler was producing non-deterministic output due to use of ahash's RandomState, which uses random seeding for its hash function. This caused iteration order of hash-based collections to vary between runs, leading to different compilation results and sporadic CI failures. Root cause: GVN (Global Value Numbering) used ahash::RandomState for hashing expressions. Different hash values across runs caused different equivalence class assignments and leader selections, which cascaded: 1. GVN picks different instruction leaders non-deterministically 2. Values get replaced differently via replace_uses() 3. op_dependent_insts BitSet gets populated differently 4. determine_evaluation() makes different linearization decisions 5. Noise sources/implicit equations are created vs. linearized differently This manifested as entire noise sources or implicit equations appearing or disappearing between runs - not just ordering differences. Changes: - Replace ahash::RandomState with BuildHasherDefault<FxHasher> in GVN - Replace AHashMap/AHashSet with IndexMap/IndexSet (with FxHasher) or HashMap/HashSet (with FxHasher) throughout the codebase - Use FxHasher consistently for deterministic hashing - Remove unused ahash dependencies from several crates Affected crates: mir_opt, mir_autodiff, sim_back, osdi, hir_def, hir_lower, basedb, mir, vfs, typed_indexmap, sourcegen 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
This commit fixes multiple issues related to building and linking with LLVM 18 on macOS ARM64: 1. Use LLVM 18's clang for bitcode generation (build.rs) - Bitcode generated by Apple's clang is incompatible with LLVM 18 - Now uses clang from LLVM_SYS_181_PREFIX when available 2. Configure ld64.lld linker with proper flags (linker/src/lib.rs) - Detect and use ld64.lld from LLVM 18 when available - Automatically add -syslibroot flag using xcrun --show-sdk-path 3. Simplify macOS target configuration (aarch64_apple_darwin.rs) - Removed -lSystem from post_link_args - With -undefined dynamic_lookup, symbols are resolved at runtime - Added -platform_version flags required by ld64.lld 4. Update README with macOS build instructions - Add macOS dependency setup section with Homebrew instructions - Document build-macos.sh and test-macos.sh convenience scripts - Add section explaining integration tests - Clarify how to enable and run integration tests These changes ensure the compiler works correctly when built against LLVM 18, fixing "Unsupported stack probing method" and "library not found for -lSystem" errors. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
Updates: - Configure macOS CI to use llvm@18 instead of latest llvm - Add xtask for setup on Mac. Unfortunatly we can't fix for straight cargo build/test without patching llvm-sys 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
This change makes integration test snapshots platform-agnostic, allowing cross-platform CI testing without spurious failures due to different struct layouts on ARM64 vs x86-64. Removed values: - residual offset numbers (4 values per node) - react_ptr offset values in jacobian entries - instance_size and model_size values These values are calculated by LLVM using platform-specific ABI alignment rules (LLVMABISizeOfType and LLVMOffsetOfElement), causing differences between architectures. All semantic information (parameter names, types, node names, jacobian flags, etc.) is preserved. All integration test snapshots have been regenerated. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
LLVM 18's ld64.lld requires -platform_version flags on both x86_64 and ARM64. Also fixes CI to use correct GitHub runners (macos-14 for ARM64, macos-13 for x86_64). Changes: - Add -platform_version flags to x86_64_apple_darwin target spec - Update CI matrix to use macos-14 for ARM64 builds (macos-13 is x86_64 only) This fixes the linker error: ld64.lld: error: must specify -platform_version ld64.lld: error: missing or unsupported -arch x86_64 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
Root cause: snprintf was declared with ALL parameters as variadic `(...)` instead of having its first 3 parameters fixed: `(ptr, i64, ptr, ...)`. On ARM64, fixed and variadic parameters use different calling conventions: - Fixed parameters: passed in x0-x7 registers - Variadic parameters: special handling, different register/stack allocation When snprintf was declared as `i32 (...)`, LLVM generated code expecting all arguments to use the variadic calling convention. This caused: 1. Arguments passed in wrong registers/memory locations 2. Double values (like 0x3ff0000000000000 = 1.0) interpreted as pointers 3. SIGSEGV when snprintf tried to strlen() these bogus pointer values Fix: Changed intrinsics.rs line 100 to pass `args` instead of `&[]` to `ty_variadic_func()`. Now snprintf is correctly declared as: declare i32 @snprintf(ptr, i64, ptr, ...) This makes the first 3 parameters use the standard calling convention while only format arguments use varargs convention. Also removed duplicate parameter addition code in compilation_unit.rs (lines 452-455) which was likely a failed attempt to work around the incorrect function signature. Fixes: Integration test crashes on ARM64 macOS with LLVM 18 Affects: BSIM3, BSIM4, and other models with large instance structs Testing: All 28 integration tests now pass on ARM64 macOS 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
…eprecation The deprecated IndexSet::remove() method was replaced with swap_remove() in the small-signal network analysis. However, this exposed a latent order- dependency bug where the analysis results differed based on iteration order. Root Cause: The algorithm had a circular dependency where analyze_value() checks if values are in small_signal_vals, but membership in that set depends on the analysis results. With platform-specific hash ordering (ahash::RandomState), this caused reactive/resistive contribution counts to swap on Windows MSYS2. Solution: Implemented an order-independent fixed-point algorithm with four phases: 1. Speculatively add ALL candidate nodes to small_signal_vals 2. Evaluate all candidates against this consistent set state 3. Remove speculative nodes that weren't confirmed 4. Add confirmed flows and remove resolved candidates This ensures all candidates see the same set state during evaluation, making the analysis deterministic regardless of iteration order while still supporting circular dependencies (e.g., noise nodes). 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
…eprecation Use the original algorithm that processes candidates one at a time, temporarily adding each candidate's node to small_signal_vals only during its own analysis. Use swap_remove() instead of the deprecated remove(). Added regression test for issue pascalkuthe#30 to prevent this bug from recurring. Fixes pascalkuthe#30 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
robtaylor
force-pushed
the
vacask-integration
branch
from
December 17, 2025 22:01
48edc66 to
3599959
Compare
- Add VACASK as git submodule at external/vacask - Add integration test discovery for VACASK Verilog-A models: - Basic device models (capacitor, diode, inductor, opamp, resistor, bsim*) - SPICE models (bjt, jfet1/2, mes1, mos1-3/6/9, vdmos, bsim*) - Simplified SPICE models (sn/ variants) - Use prefixed snapshot names to avoid conflicts with OpenVAF models - Generate OSDI descriptor snapshots for all VACASK models This adds ~35 new device model tests from the VACASK project. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
robtaylor
force-pushed
the
vacask-integration
branch
from
December 17, 2025 22:18
3599959 to
7af13ac
Compare
robtaylor
force-pushed
the
branches/macos-patches-v3
branch
2 times, most recently
from
December 18, 2025 17:45
ac4d02b to
1caa8f5
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
2 participants
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 VACASK device models as integration tests for OpenVAF.
Changes
external/vacaskThis adds comprehensive device model coverage from the VACASK project for testing OpenVAF's Verilog-A compilation.
Test plan
🤖 Generated with Claude Code