tomoaki/hakorune
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

feat(llvm): Phase 131-4 P1 完了 - PHI ordering 修正（multi-pass architecture）

Browse Source 
Phase 131-4 P1: PHI After Terminator Bug 修正

問題:
- LLVM IR で PHI が terminator の後に出現（LLVM invariant 違反）
- Case B (loop_min_while.hako) が TAG-EMIT で失敗

修正:
- Multi-pass block lowering architecture 実装:
  - Pass A: non-terminator instructions のみ emit
  - Pass B: PHI finalization（block head に確実に配置）
  - Pass C: deferred terminators を最後に emit

変更ファイル:
- src/llvm_py/builders/block_lower.py (~40行):
  - lower_blocks() で terminator を defer
  - lower_terminators() 新設（Pass C）
  - _deferred_terminators dict で管理
- src/llvm_py/builders/function_lower.py (3行):
  - Pass 順序更新: A→B→C
- src/llvm_py/instructions/ret.py (5行):
  - _disable_phi_synthesis flag で Pass C 中の PHI 生成を抑制

テスト結果:
- Case B EMIT: ❌→✅ (修正成功)
- Case B LINK: ❌ (新 TAG-LINK: undefined nyash_console_log)
- Case A/B2: ✅ (退行なし)

箱化モジュール化:
- ✅ Multi-pass で責務分離
- ✅ Flag mechanism で構造的制御
- ✅ ハードコード禁止原則遵守

Next: Phase 131-5 (TAG-LINK 修正)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
This commit is contained in:
nyash-codex
2025-12-14 06:12:31 +09:00
parent 5709026812
commit 73613dcef0
4 changed files with 138 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

79
docs/development/current/main/phase131-3-llvm-lowering-inventory.md
View File

@ -8,7 +8,7 @@
| Case | File | Emit | Link | Run | Notes |
|------|------|------|------|-----|-------|
| A | `apps/tests/phase87_llvm_exe_min.hako` | ✅ | ✅ | ✅ | **PASS** - Simple return 42, no BoxCall, exit code verified |
| B | `apps/tests/loop_min_while.hako` | ❌ | - | - | **TAG-EMIT** - Loop generates invalid LLVM IR (observed: PHI placement/order issue; also mentions empty block) |
| B | `apps/tests/loop_min_while.hako` | ✅ | ❌ | - | **TAG-LINK** - EMIT fixed (Phase 131-4), LINK fails (undefined nyash_console_log) |
| B2 | `/tmp/case_b_simple.hako` | ✅ | ✅ | ✅ | **PASS** - Simple print(42) without loop works |
| C | `apps/tests/llvm_stage3_loop_only.hako` | ❌ | - | - | **TAG-EMIT** - Complex loop (break/continue) fails JoinIR pattern matching |
@ -72,9 +72,51 @@ This strongly suggests an **emission ordering / insertion-position** problem in
- PHI insertion rules + debug: `src/llvm_py/phi_wiring/wiring.py` (`NYASH_PHI_ORDERING_DEBUG=1`)
- “Empty block” safety pass (separate concern): `src/llvm_py/builders/function_lower.py:_enforce_terminators`
**✅ FIXED (Phase 131-4)**: Multi-pass block lowering architecture
**Solution implemented**:
- **Pass A**: Lower non-terminator instructions (body ops only)
- **Pass B**: Finalize PHIs (wire incoming edges) - happens in `function_lower.py`
- **Pass C**: Lower deferred terminators (after PHIs are placed)
**Key changes**:
1. `src/llvm_py/builders/block_lower.py`:
- Split `lower_blocks()` to defer terminators
- Added `lower_terminators()` function for Pass C
- Deferred terminators stored in `builder._deferred_terminators`
2. `src/llvm_py/builders/function_lower.py`:
- Updated pass ordering: Pass A → Pass B → Pass C
- Added call to `_lower_terminators()` after `_finalize_phis()`
3. `src/llvm_py/instructions/ret.py`:
- Added `_disable_phi_synthesis` flag check
- Prevents PHI creation during Pass C (terminators should only use existing values)
**Result**:
- Case B EMIT now succeeds ✅
- Generated LLVM IR is valid (PHIs before terminators)
- No regression in Cases A and B2
---
### 2. TAG-EMIT: JoinIR Pattern Mismatch (Case C)
### 2. TAG-LINK: Missing runtime symbols (Case B)
**File**: `apps/tests/loop_min_while.hako`
**Link Error**:
```
/usr/bin/ld: /home/tomoaki/git/hakorune-selfhost/target/aot_objects/loop_min_while.o: in function `condition_fn':
<string>:(.text+0x99): undefined reference to `nyash_console_log'
```
**Root Cause**: ExternCall lowering emits calls to runtime functions (e.g., `nyash_console_log`) but these symbols are not provided by NyKernel (`libnyash_kernel.a`).
**Next Steps**: Map ExternCall names to actual NyKernel symbols or add missing runtime functions.
---
### 3. TAG-EMIT: JoinIR Pattern Mismatch (Case C)
**File**: `apps/tests/llvm_stage3_loop_only.hako`
@ -127,24 +169,33 @@ Hint: This loop pattern is not supported. All loops must use JoinIR lowering.
## Next Steps
### Priority 1: Fix TAG-EMIT (PHI After Terminator Bug) ⚠️ CRITICAL
### ✅ Priority 1: COMPLETED - Fix TAG-EMIT (PHI After Terminator Bug)
**Target**: Case B (`loop_min_while.hako`)
**Goal**: Ensure PHIs are always emitted/inserted before any terminator in the same basic block.
**Status**: ✅ FIXED in Phase 131-4 (see Root Cause #1 above)
**Candidate approach** (docs-only; implementation to be decided):
- Split lowering into multi-pass so that PHI placeholders exist before terminators are emitted, or delay terminator emission until after PHI finalization:
- (A) Predeclare PHIs at block creation time (placeholders), then emit body ops, then wire incomings, then emit terminators.
- (B) Keep current finalize order, but guarantee `ensure_phi()` always inserts at head even when a terminator exists (verify llvmlite positioning behavior).
**Primary files to look at for the fix**:
- `src/llvm_py/builders/function_lower.py` (pass ordering)
- `src/llvm_py/builders/block_lower.py` (terminator emission split point)
- `src/llvm_py/phi_wiring/wiring.py` (PHI insertion positioning)
**Result**: Case B EMIT now succeeds. LINK still fails (TAG-LINK), but that's a separate issue (Priority 2).
---
### Priority 2: Fix TAG-EMIT (JoinIR Pattern Coverage)
### Priority 2: Fix TAG-LINK (Missing Runtime Symbols)
**Target**: Case B (`loop_min_while.hako`)
**Approach**:
1. Identify all ExternCall lowering paths in Python harness
2. Map to actual NyKernel symbols (e.g., `nyash_console_log` → `ny_console_log` or similar)
3. Update ExternCall lowering to use correct symbol names
4. OR: Add wrapper functions in NyKernel to provide missing symbols
**Files**:
- `src/llvm_py/instructions/externcall.py` - ExternCall lowering
- `crates/nyash_kernel/src/lib.rs` - NyKernel runtime symbols
**Expected**: Case B should LINK ✅ RUN ✅ after fix
---
### Priority 3: Fix TAG-EMIT (JoinIR Pattern Coverage)
**Target**: Case C (`llvm_stage3_loop_only.hako`)
**Approach**:

76
src/llvm_py/builders/block_lower.py
View File

@ -1,10 +1,20 @@
from typing import Dict, Any, List
from typing import Dict, Any, List, Tuple
from llvmlite import ir
from trace import debug as trace_debug
from trace import phi_json as trace_phi_json
def lower_blocks(builder, func: ir.Function, block_by_id: Dict[int, Dict[str, Any]], order: List[int], loop_plan: Dict[str, Any] | None):
"""Lower blocks in multi-pass to ensure PHIs are always before terminators.
Phase 131-4: Multi-pass block lowering architecture
- Pass A: Lower non-terminator instructions only (terminators deferred)
- Pass B: PHI finalization happens in function_lower.py
- Pass C: Lower terminators (happens after PHI finalization)
This ensures LLVM IR invariant: PHI nodes must be at block head before any
other instructions, and terminators must be last.
"""
skipped: set[int] = set()
if loop_plan is not None:
try:
@ -218,19 +228,12 @@ def lower_blocks(builder, func: ir.Function, block_by_id: Dict[int, Dict[str, An
created_ids.append(dst)
except Exception:
pass
# Lower terminators
for inst in term_ops:
try:
trace_debug(f"[llvm-py] term op: {inst.get('op')} dst={inst.get('dst')} cond={inst.get('cond')}")
except Exception:
pass
try:
if bb.terminator is not None:
break
except Exception:
pass
ib.position_at_end(bb)
builder.lower_instruction(ib, inst, func)
# Phase 131-4 Pass A: DEFER terminators until after PHI finalization
# Store terminators for Pass C (will be lowered in lower_terminators)
if not hasattr(builder, '_deferred_terminators'):
builder._deferred_terminators = {}
if term_ops:
builder._deferred_terminators[bid] = (bb, term_ops)
try:
for vid in created_ids:
val = vmap_cur.get(vid)
@ -256,3 +259,48 @@ def lower_blocks(builder, func: ir.Function, block_by_id: Dict[int, Dict[str, An
delattr(builder, '_current_vmap')
except Exception:
pass
def lower_terminators(builder, func: ir.Function):
"""Phase 131-4 Pass C: Lower deferred terminators after PHI finalization.
This ensures PHI nodes are always at block heads before terminators are added,
maintaining LLVM IR's invariant: PHIs first, then other instructions, then terminators.
"""
if not hasattr(builder, '_deferred_terminators'):
return
deferred = builder._deferred_terminators
trace_debug(f"[llvm-py/pass-c] Lowering {len(deferred)} blocks with deferred terminators")
for bid, (bb, term_ops) in deferred.items():
ib = ir.IRBuilder(bb)
try:
builder.resolver.builder = ib
builder.resolver.module = builder.module
# Phase 131-4: Disable PHI synthesis during terminator lowering
# Terminators should only use values that already exist (from Pass A/B)
builder.resolver._disable_phi_synthesis = True
except Exception:
pass
for inst in term_ops:
try:
trace_debug(f"[llvm-py/pass-c] term op: {inst.get('op')} dst={inst.get('dst')} in bb{bid}")
except Exception:
pass
try:
if bb.terminator is not None:
# Terminator already exists (e.g., from loop lowering), skip
trace_debug(f"[llvm-py/pass-c] bb{bid} already has terminator, skipping")
break
except Exception:
pass
ib.position_at_end(bb)
builder.lower_instruction(ib, inst, func)
# Clean up deferred state
try:
delattr(builder, '_deferred_terminators')
except Exception:
pass

7
src/llvm_py/builders/function_lower.py
View File

@ -275,6 +275,7 @@ def lower_function(builder, func_data: Dict[str, Any]):
except Exception:
loop_plan = None
# Phase 131-4 Pass A: Lower non-terminator instructions (terminators deferred)
from builders.block_lower import lower_blocks as _lower_blocks
_lower_blocks(builder, func, block_by_id, order, loop_plan)
@ -299,9 +300,13 @@ def lower_function(builder, func_data: Dict[str, Any]):
except Exception:
pass
# Finalize PHIs for this function
# Phase 131-4 Pass B: Finalize PHIs (wires incoming edges)
_finalize_phis(builder)
# Phase 131-4 Pass C: Lower deferred terminators (after PHIs are placed)
from builders.block_lower import lower_terminators as _lower_terminators
_lower_terminators(builder, func)
# Safety pass: ensure every basic block ends with a terminator.
# This avoids llvmlite IR parse errors like "expected instruction opcode" on empty blocks.
try:

6
src/llvm_py/instructions/ret.py
View File

@ -111,7 +111,11 @@ def lower_return(
ret_val = ir.Constant(return_type, None)
# If still zero-like (typed zero) and we have predecessor snapshots, synthesize a minimal PHI at block head.
# Phase 131-4: Skip PHI synthesis if disabled (e.g., during Pass C terminator lowering)
try:
disable_phi = False
if resolver is not None and hasattr(resolver, '_disable_phi_synthesis'):
disable_phi = getattr(resolver, '_disable_phi_synthesis', False)
zero_like = False
if isinstance(ret_val, ir.Constant):
if isinstance(return_type, ir.IntType):
@ -121,7 +125,7 @@ def lower_return(
elif isinstance(return_type, ir.PointerType):
zero_like = (str(ret_val) == str(ir.Constant(return_type, None)))
# Synthesize a PHI for return at the BLOCK HEAD (grouped), not inline.
if zero_like and preds is not None and block_end_values is not None and bb_map is not None and isinstance(value_id, int):
if not disable_phi and zero_like and preds is not None and block_end_values is not None and bb_map is not None and isinstance(value_id, int):
# Derive current block id from name like 'bb3'
cur_bid = None
try: