<img style="width: 810px;" src="clipboard-202604251344-sv2ro.png"><br> The Student T test shows that the performance is almost identical, but the memory consumption is confidently lowered by about 3%.dsh0416 (Delton Ding)
The regression got better controlled with the new patch.   ![]...dsh0416 (Delton Ding)
From the bench results, we could see some very positive signals in some benches, but it also introduces some regressions, I would try if I could further narrowing down the regressions.dsh0416 (Delton Ding)
Here are the ruby benches results.   | Area | interp | yjit...dsh0416 (Delton Ding)
I see that the benchmarks might be affected, since st_numhash/symbol hash path looks like returning constant 128 on bits 25..31, which increases the hash collision, under investigating.dsh0416 (Delton Ding)
I found some crash cases, that I need to fix it first. Also, I am running the full ruby-bench against the master branch on linux and windows to see if I've missed anything.dsh0416 (Delton Ding)
This change adds a Swiss-table-inspired probing layer to Ruby's core `st_table`, and shrinks `st_table_entry` from 24 B to 16 B by moving the stored hash into a parallel array. It is built and enabled by default; `--disable-swiss-st` rev...dsh0416 (Delton Ding)