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

Phase 7-Step4: Replace runtime checks with config macros (+1.1% improvement)

Browse Source 
**What Changed**:
Replace 3 runtime checks with compile-time config macros in hot path:
- `g_fastcache_enable` → `TINY_FRONT_FASTCACHE_ENABLED` (line 421)
- `tiny_heap_v2_enabled()` → `TINY_FRONT_HEAP_V2_ENABLED` (line 809)
- `ultra_slim_mode_enabled()` → `TINY_FRONT_ULTRA_SLIM_ENABLED` (line 757)

**Why This Works**:
PGO mode (-DHAKMEM_TINY_FRONT_PGO=1 in bench builds):
- Config macros become compile-time constants (0 or 1)
- Compiler eliminates dead branches: if (0) { ... } → removed
- Smaller code size, better instruction cache locality
- Fewer branch mispredictions in hot path

Normal mode (default, backward compatible):
- Config macros expand to runtime function calls
- Preserves ENV variable control (e.g., HAKMEM_TINY_FRONT_V2=1)

**Performance**:
bench_random_mixed (ws=256):
- Before (Step 3): 80.6 M ops/s
- After (Step 4):  81.0 / 81.0 / 82.4 M ops/s
- Average: ~81.5 M ops/s (+1.1%, +0.9 M ops/s)

**Dead Code Elimination Benefit**:
- FastCache check eliminated (PGO mode: TINY_FRONT_FASTCACHE_ENABLED = 0)
- Heap V2 check eliminated (PGO mode: TINY_FRONT_HEAP_V2_ENABLED = 0)
- Ultra SLIM check eliminated (PGO mode: TINY_FRONT_ULTRA_SLIM_ENABLED = 0)

**Files Modified**:
- core/tiny_alloc_fast.inc.h (+6 lines comments, 3 lines changed)

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

Co-Authored-By: Claude <noreply@anthropic.com>
This commit is contained in:
Moe Charm (CI)
2025-11-29 17:04:24 +09:00
parent 09942d5a08
commit 21f7b35503
1 changed files with 6 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
core/tiny_alloc_fast.inc.h
View File

@ -417,7 +417,8 @@ static inline void* tiny_alloc_fast_pop(int class_idx) {
#endif
// Phase 1: Try array stack (FastCache) first for hottest tiny classes (C0–C3)
if (__builtin_expect(g_fastcache_enable && class_idx <= 3, 1)) {
// Phase 7-Step4: Use config macro for dead code elimination in PGO mode
if (__builtin_expect(TINY_FRONT_FASTCACHE_ENABLED && class_idx <= 3, 1)) {
void* fc = fastcache_pop(class_idx);
if (__builtin_expect(fc != NULL, 1)) {
// Frontend FastCache hit (already tracked by g_front_fc_hit)
@ -752,7 +753,8 @@ static inline void* tiny_alloc_fast(size_t size) {
}
#endif
if (__builtin_expect(ultra_slim_mode_enabled(), 0)) {
// Phase 7-Step4: Use config macro for dead code elimination in PGO mode
if (__builtin_expect(TINY_FRONT_ULTRA_SLIM_ENABLED, 0)) {
return ultra_slim_alloc_with_refill(size);
}
// ========== End Phase 19-2: Ultra SLIM ==========
@ -804,7 +806,8 @@ static inline void* tiny_alloc_fast(size_t size) {
void* ptr = NULL;
// Front-V2: TLS magazine front (A/B, default OFF)
if (__builtin_expect(tiny_heap_v2_enabled() && front_prune_heapv2_enabled() && class_idx <= 3, 0)) {
// Phase 7-Step4: Use config macro for dead code elimination in PGO mode
if (__builtin_expect(TINY_FRONT_HEAP_V2_ENABLED && front_prune_heapv2_enabled() && class_idx <= 3, 0)) {
void* hv2 = tiny_heap_v2_alloc_by_class(class_idx);
if (hv2) {
front_metrics_heapv2_hit(class_idx);