swiss ok, needs more benchmark

2026-03-06 18:22:30 +08:00
parent 83997f64d0
commit f6e897ed06
70 changed files with 1889 additions and 3371 deletions
@@ -92,7 +92,10 @@ Toolchain=InProcessEmitToolchain  InvocationCount=2048  UnrollFactor=1
 | Hlsf | 76.23 µs | 0.355 µs | 0.315 µs | 0.44 | 0.01 |
 | Libc | 172.98 µs | 3.042 µs | 2.988 µs | 1.00 | 0.02 |

-This run shows HLSF significantly faster than libc under the current configuration. See the HLSF document for details and caveats.
+This benchmark measures steady-state small-block allocation throughput (64B–4KB),  
+which represents the dominant allocation pattern in StgSharp's matrix computation pipeline.  
+HLSF uses a pre-allocated arena with size-class free lists, avoiding kernel-mode allocation calls.  
+Under this workload, HLSF achieves ~2.3× throughput over system malloc. This is not a general-purpose allocator comparison.

 ### Graphics and Rendering