Add JIT-time array map lookup inline optimization

[yunwei37]

Summary

• JIT-time inline optimization for bpf_map_lookup_elem on array maps. When a BPF program calls helper-1 (bpf_map_lookup_elem) with a constant map handle that has been registered as an array map, the JIT rewrites the call into direct address arithmetic — eliminating the helper call entirely.
• New register_array_map() API lets the host register an array map's metadata (fd/handle, max_entries, value_size, and data base pointer) before JIT compilation. During the LLVM IR rewrite pass, calls to _bpf_helper_ext_0001 where arg0 matches a registered handle are replaced with an inline bounds-check and pointer computation: key < max_entries ? base + key * value_size : nullptr.
• Result: 1.30x speedup — execution time drops from 7.51 ns to 5.77 ns per iteration on an array-map-lookup micro-benchmark. Helper calls go from 10M to 0 (verified via bpf_get_helper_call_count()).
• Limitations: JIT path only (not AOT), helper-1 only, array maps only, key_size == 4, constant map handle only. Does not cover per-CPU array maps or hash maps.

Changed files (7 files, +445 lines)

File	Change
include/llvmbpf.hpp	New register_array_map() public API on llvmbpf_vm
src/vm.cpp	Stores array map metadata, passes to JIT context
src/llvm_jit_context.cpp	LLVM IR rewrite: scan for helper-1 calls, replace with inline arithmetic
src/llvm_jit_context.hpp	Internal struct for array map info
example/array_map_inline_bench.cpp	End-to-end benchmark (10M iterations, before/after comparison)
example/CMakeLists.txt	Build target for the benchmark
test/unit-test/vm_test.cpp	Unit test verifying inline correctness and zero helper calls

Test plan

• Unit test (vm_test.cpp): verify that exec() returns the correct looked-up value and bpf_get_helper_call_count() reports 0 helper-1 calls after inlining
• Benchmark (array_map_inline_bench.cpp): run ./build/example/array_map_inline_bench and confirm ~1.3x speedup vs baseline (no inline)
• Build: cmake --build build -j succeeds with no warnings on the new files
• Existing tests: ctest --test-dir build passes (no regressions)

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 79.56989% with 19 lines in your changes missing coverage. Please review.
✅ Project coverage is 74.47%. Comparing base (f037a7e) to head (7f19854).
⚠️ Report is 2 commits behind head on main.

Files with missing lines	Patch %	Lines
src/vm.cpp	50.00%	10 Missing ⚠️
src/llvm_jit_context.cpp	87.67%	9 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #46      +/-   ##
==========================================
+ Coverage   74.12%   74.47%   +0.35%     
==========================================
  Files           5        5              
  Lines        1341     1434      +93     
  Branches      153      176      +23     
==========================================
+ Hits          994     1068      +74     
- Misses        347      366      +19     

Flag	Coverage Δ
bpf_conformance	55.61% <4.30%> (-3.71%)	⬇️
spirv_opencl	36.48% <4.30%> (-2.27%)	⬇️
unittests	50.97% <79.56%> (+2.06%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Officeyutong]

Why calling optimizeModule two times?