feat: complete safe RV32 constant-load merge optimization

.github/workflows/ci.yml

@@ -11,18 +11,29 @@ env:
 
 permissions:
   contents: read
-  pages: write
-  id-token: write
 
 jobs:
   # ═════════════════════════════════════════════════════════════════════════
   # 课题功能测试：所有14个模块的单元测试 + 集成测试
   # ═════════════════════════════════════════════════════════════════════════
   test:
-    runs-on: self-hosted
+    runs-on: ${{ github.event_name == 'pull_request' && 'ubuntu-latest' || 'self-hosted' }}
     timeout-minutes: 20
     steps:
+      - name: Checkout pull request on isolated runner
+        if: github.event_name == 'pull_request'
+        uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4
+        with:
+          persist-credentials: false
+
+      - name: Set up Python for pull request
+        if: github.event_name == 'pull_request'
+        uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5
+        with:
+          python-version: "3.12"
+
       - name: Sync from local mirror (skip unstable GitHub checkout)
+        if: github.event_name != 'pull_request'
         run: |
           echo "=== CI Sync ==="
           retry() {
@@ -46,12 +57,6 @@ jobs:
           cd "$WORKSPACE"
           echo "Fetching origin main..."
           retry "git fetch origin main" git fetch origin main || exit 1
-          if [ "$GITHUB_EVENT_NAME" = "pull_request" ]; then
-            PR_NUMBER="${GITHUB_REF#refs/pull/}"
-            PR_NUMBER="${PR_NUMBER%/merge}"
-            echo "Fetching PR #${PR_NUMBER} merge ref..."
-            retry "git fetch PR #${PR_NUMBER}" git fetch origin "refs/pull/${PR_NUMBER}/merge" || exit 1
-          fi
           echo "Checking out $GITHUB_SHA..."
           if git -c advice.detachedHead=false checkout -f "$GITHUB_SHA"; then
             echo "::notice::Checkout successful: $(git log -1 --format='%h %ai %s')"
@@ -64,6 +69,7 @@ jobs:
         run: |
           python3.12 -m pip install --upgrade pip
           python3.12 -m pip install -e ".[all]"
+          python3.12 -m pip install "pytest>=7,<10"
 
       - name: Run all topic tests
         run: |
@@ -72,6 +78,14 @@ jobs:
             --junit-xml=benchmark_reports/test_results.xml \
             --ignore=tests/test_simulator.py
 
+      - name: Run PR #35 constant-merge regressions
+        run: |
+          python3.12 -m pytest \
+            tests/test_const_merge.py \
+            tests/test_backend.py::TestConstMergeIntegration \
+            tests/test_simulator.py::TestStubProfiledMachine \
+            -v --tb=short
+
       - name: Generate test visualization page
         if: github.ref == 'refs/heads/main'
         run: |
@@ -81,7 +95,7 @@ jobs:
 
       - name: Upload test reports
         if: github.ref == 'refs/heads/main'
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4
         with:
           name: test-reports
           path: |
@@ -93,10 +107,23 @@ jobs:
   # 模型性能测试：ONNX模型管线 + DSL用例 + CNN RISC-V编译
   # ═════════════════════════════════════════════════════════════════════════
   benchmark:
-    runs-on: self-hosted
+    runs-on: ${{ github.event_name == 'pull_request' && 'ubuntu-latest' || 'self-hosted' }}
     timeout-minutes: 30
     steps:
+      - name: Checkout pull request on isolated runner
+        if: github.event_name == 'pull_request'
+        uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4
+        with:
+          persist-credentials: false
+
+      - name: Set up Python for pull request
+        if: github.event_name == 'pull_request'
+        uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5
+        with:
+          python-version: "3.12"
+
       - name: Sync from local mirror (skip unstable GitHub checkout)
+        if: github.event_name != 'pull_request'
         run: |
           echo "=== CI Sync ==="
           retry() {
@@ -120,12 +147,6 @@ jobs:
           cd "$WORKSPACE"
           echo "Fetching origin main..."
           retry "git fetch origin main" git fetch origin main || exit 1
-          if [ "$GITHUB_EVENT_NAME" = "pull_request" ]; then
-            PR_NUMBER="${GITHUB_REF#refs/pull/}"
-            PR_NUMBER="${PR_NUMBER%/merge}"
-            echo "Fetching PR #${PR_NUMBER} merge ref..."
-            retry "git fetch PR #${PR_NUMBER}" git fetch origin "refs/pull/${PR_NUMBER}/merge" || exit 1
-          fi
           echo "Checking out $GITHUB_SHA..."
           if git -c advice.detachedHead=false checkout -f "$GITHUB_SHA"; then
             echo "::notice::Checkout successful: $(git log -1 --format='%h %ai %s')"
@@ -138,7 +159,7 @@ jobs:
         run: |
           python3.12 -m pip install --upgrade pip
           python3.12 -m pip install -e ".[all]"
-          python3.12 -m pip install markdown
+          python3.12 -m pip install markdown "pytest>=7,<10"
 
       # ── 3.1 ONNX 模型管线基准测试 ──────────────────────────────────────
       - name: ONNX model pipeline benchmarks
@@ -208,7 +229,7 @@ jobs:
 
       # ── 上报 ──────────────────────────────────────────────────────────
       - name: Upload benchmark reports
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4
         with:
           name: benchmark-reports
           path: benchmark_reports/
@@ -253,7 +274,7 @@ jobs:
       # ── 上传 Pages 产物 ──────────────────────────────────────────────
       - name: Upload Pages artifact
         if: github.ref == 'refs/heads/main'
-        uses: actions/upload-pages-artifact@v3
+        uses: actions/upload-pages-artifact@56afc609e74202658d3ffba0e8f6dda462b719fa # v3
         with:
           path: benchmark_reports/
 
@@ -284,4 +305,4 @@ jobs:
     steps:
       - name: Deploy to GitHub Pages
         id: deployment
-        uses: actions/deploy-pages@v4
+        uses: actions/deploy-pages@d6db90164ac5ed86f2b6aed7e0febac5b3c0c03e # v4

benchmarks/bench_const_merge.py

@@ -12,56 +12,91 @@
 from __future__ import annotations
 
 import argparse
+import os
 import statistics
+import sys
 import time
-from typing import Optional
 
-from scratchv.backend.const_merge import merge_constants
+BENCH_DIR = os.path.dirname(__file__)
+PROJ_DIR = os.path.dirname(BENCH_DIR)
+sys.path.insert(0, PROJ_DIR)
 
+from scratchv.backend._asm_parser import parse_asm
+from scratchv.backend.const_merge import merge_constants_detailed
 
-def _gen_synthetic_asm(num_instrs: int, seed: int = 42,
-                        lui_ratio: float = 0.3) -> str:
-    """Generate synthetic assembly with lui+addi patterns.
+
+def _gen_synthetic_asm(
+    num_instructions: int,
+    seed: int = 42,
+    pair_density: float = 0.3,
+    redundant_lui_density: float = 0.1,
+) -> str:
+    """Generate controlled synthetic assembly covering both optimization rules.
 
     Parameters
     ----------
-    num_instrs:
+    num_instructions:
         Target number of instructions.
     seed:
         Random seed for reproducibility.
-    lui_ratio:
+    pair_density:
         Fraction of instructions that form lui+addi pairs.
+    redundant_lui_density:
+        Fraction of generated groups that contain a redundant LUI pattern.
     """
     import random
-    random.seed(seed)
+
+    if num_instructions < 0:
+        raise ValueError("num_instructions must be non-negative")
+    if not 0.0 <= pair_density <= 1.0:
+        raise ValueError("pair_density must be between 0 and 1")
+    if not 0.0 <= redundant_lui_density <= 1.0:
+        raise ValueError("redundant_lui_density must be between 0 and 1")
+    if pair_density + redundant_lui_density > 1.0:
+        raise ValueError(
+            "pair_density + redundant_lui_density must not exceed 1",
+        )
+
+    rng = random.Random(seed)
 
     lines = [".text", "synthetic_func:"]
     i = 0
-    while i < num_instrs:
-        use_lui = random.random() < lui_ratio
-
-        if use_lui and i + 1 < num_instrs:
+    while i < num_instructions:
+        choice = rng.random()
+
+        if choice < redundant_lui_density and i + 2 < num_instructions:
+            regs = ["t0", "t1", "t2", "s0", "s1", "a0", "a1"]
+            reg = rng.choice(regs)
+            imm_hi = rng.choice([0x10000, 0x20000, 0x12345])
+            lines.append(f"  lui {reg}, {hex(imm_hi)}")
+            lines.append("  add a4, a5, a6")
+            lines.append(f"  lui {reg}, {hex(imm_hi)}")
+            i += 3
+        elif (
+            choice < redundant_lui_density + pair_density
+            and i + 1 < num_instructions
+        ):
             regs = ["t0", "t1", "t2", "s0", "s1", "a0", "a1", "a2", "a3"]
-            r = random.choice(regs)
-            imm_hi = random.choice([0x10000, 0x20000, 0x12345, 0xABCDE, 0xFFFFF])
-            imm_lo = random.choice([0x000, 0x100, 0x678, 0xFFF, 0x800])
+            r = rng.choice(regs)
+            imm_hi = rng.choice([0x10000, 0x20000, 0x12345, 0xABCDE, 0xFFFFF])
+            imm_lo = rng.choice([0x000, 0x100, 0x678, 0xFFF, 0x800])
             lines.append(f"  lui {r}, {hex(imm_hi)}")
             lines.append(f"  addi {r}, {r}, {hex(imm_lo)}")
             i += 2
         else:
-            op = random.choice(["add", "sub", "lw", "sw", "mv", "mul", "xor",
-                                "li", "addi", "beq", "j", "ret"])
+            op = rng.choice(["add", "sub", "lw", "sw", "mv", "mul", "xor",
+                             "li", "addi", "beq", "j", "ret"])
             regs = ["t0", "t1", "t2", "t3", "t4", "s0", "s1",
                     "a0", "a1", "a2", "a3", "sp", "ra"]
-            r1 = random.choice(regs)
-            r2 = random.choice(regs)
-            r3 = random.choice(regs)
+            r1 = rng.choice(regs)
+            r2 = rng.choice(regs)
+            r3 = rng.choice(regs)
             if op == "li":
-                lines.append(f"  {op} {r1}, {random.randint(0, 4096)}")
+                lines.append(f"  {op} {r1}, {rng.randint(0, 4096)}")
             elif op == "addi":
-                lines.append(f"  {op} {r1}, {r2}, {random.randint(-2048, 2047)}")
+                lines.append(f"  {op} {r1}, {r2}, {rng.randint(-2048, 2047)}")
             elif op in ("lw", "sw"):
-                lines.append(f"  {op} {r1}, {random.randint(0, 16)}(sp)")
+                lines.append(f"  {op} {r1}, {rng.randint(0, 16)}(sp)")
             elif op in ("beq", "bne", "blt", "bge"):
                 lines.append(f"  {op} {r1}, {r2}, label_{i}")
             elif op == "j":
@@ -78,24 +113,39 @@ def _gen_synthetic_asm(num_instrs: int, seed: int = 42,
 
 def bench_merge(asm_text: str, repeats: int = 50) -> dict:
     """Benchmark the constant merge optimizer."""
+    if repeats < 1:
+        raise ValueError("repeats must be at least 1")
     times = []
     results = []
 
     for _ in range(repeats):
         t0 = time.perf_counter()
-        result, changes = merge_constants(asm_text)
+        result, stats = merge_constants_detailed(asm_text)
         t1 = time.perf_counter()
         times.append(t1 - t0)
-        results.append((result, changes))
-
-    changes_list = [r[1] for r in results]
-    input_lines = asm_text.count("\n")
-    output_lines = results[0][0].count("\n") if results else 0
+        results.append((result, stats))
+
+    changes_list = [r[1].total_changes for r in results]
+    first_stats = results[0][1]
+    parsed_input = parse_asm(asm_text)
+    parsed_output = parse_asm(results[0][0])
+    input_instructions = sum(
+        line.opcode is not None and not line.is_directive
+        for line in parsed_input
+    )
+    output_instructions = sum(
+        line.opcode is not None and not line.is_directive
+        for line in parsed_output
+    )
 
     return {
-        "input_lines": input_lines,
-        "output_lines": output_lines,
-        "line_reduction": input_lines - output_lines,
+        "benchmark_type": "synthetic",
+        "input_instructions": input_instructions,
+        "output_instructions": output_instructions,
+        "instruction_reduction": input_instructions - output_instructions,
+        "candidate_pairs": first_stats.candidate_pairs,
+        "merged_pairs": first_stats.merged_pairs,
+        "redundant_lui_removed": first_stats.redundant_lui_removed,
         "changes_mean": statistics.mean(changes_list),
         "changes_stdev": statistics.stdev(changes_list) if len(changes_list) > 1 else 0,
         "repeats": repeats,
@@ -111,35 +161,50 @@ def main():
     parser = argparse.ArgumentParser(description="Constant Merge Benchmark")
     parser.add_argument("--repeats", type=int, default=50,
                         help="Number of repeat measurements")
+    parser.add_argument("--seed", type=int, default=42)
+    parser.add_argument("--pair-density", type=float, default=0.3)
+    parser.add_argument("--redundant-lui-density", type=float, default=0.1)
     args = parser.parse_args()
 
     sizes = [100, 500, 1000, 2000, 5000]
     print("=" * 80)
     print("RISC-V Constant Load Merge Optimizer Benchmark")
+    print("benchmark_type=synthetic")
     print("=" * 80)
 
     print(f"\n{'Size':>8} {'Mean(ms)':>10} {'Stdev(ms)':>10} "
-          f"{'Changes':>8} {'InpLines':>10} {'OutLines':>10} {'Reduc':>8}")
+          f"{'Pairs':>8} {'RedLUI':>8} {'InpInst':>10} {'OutInst':>10}")
     print("-" * 80)
 
     for size in sizes:
-        asm = _gen_synthetic_asm(size, lui_ratio=0.3)
+        asm = _gen_synthetic_asm(
+            size,
+            seed=args.seed,
+            pair_density=args.pair_density,
+            redundant_lui_density=args.redundant_lui_density,
+        )
         stats = bench_merge(asm, repeats=args.repeats)
         print(f"{size:>8} {stats['mean_s'] * 1000:>10.3f} "
               f"{stats['stdev_s'] * 1000:>10.3f} "
-              f"{stats['changes_mean']:>8.1f} "
-              f"{stats['input_lines']:>10} {stats['output_lines']:>10} "
-              f"{stats['line_reduction']:>8}")
+              f"{stats['merged_pairs']:>8} "
+              f"{stats['redundant_lui_removed']:>8} "
+              f"{stats['input_instructions']:>10} "
+              f"{stats['output_instructions']:>10}")
 
     # Test different lui densities
     print(f"\nLUI Density Impact (2000 instructions):")
     print("-" * 60)
     for ratio in [0.0, 0.1, 0.3, 0.5]:
-        asm = _gen_synthetic_asm(2000, lui_ratio=ratio)
+        asm = _gen_synthetic_asm(
+            2000,
+            seed=args.seed,
+            pair_density=ratio,
+            redundant_lui_density=args.redundant_lui_density,
+        )
         stats = bench_merge(asm, repeats=args.repeats)
         print(f"  ratio={ratio:.1f}  {stats['mean_s'] * 1000:.3f} ms  "
               f"changes: {stats['changes_mean']:.1f}  "
-              f"reduction: {stats['line_reduction']}")
+              f"reduction: {stats['instruction_reduction']}")
 
 
 if __name__ == "__main__":