core/vm: use rsh instead of lookup

[holiman]

This PR cleans up the code analysis a bit. There's this ugly lookup there, instead of a bitshift.

For some reason, whenever I try to remove that lookup, like this PR does, there's a performance loss. Results from go v1.17, with go test . -run - -bench JumpdestOp/\(PUSH1$\|STOP\) -count 10

name                        old time/op  new time/op  delta
JumpdestOpAnalysis/PUSH1-6  27.3ns ± 4%  31.0ns ± 0%  +13.41%  (p=0.000 n=10+8)
JumpdestOpAnalysis/STOP-6   22.0ns ± 8%  17.7ns ± 1%  -19.73%  (p=0.000 n=10+8)

So the PUSH1 degrades 13%, and the STOP, which should not be affected at all improves by ~20%.

😕

Do other folks get similar numbers?

[holiman]

Assigning this to @fjl who requested this change

[holiman]

Tested again on top of the benchmark changes from @chfast -- but see the same thing.

name                        old time/op    new time/op    delta
JumpdestOpAnalysis/PUSH1-6    1.01ms ± 3%    1.17ms ± 3%  +16.49%  (p=0.000 n=10+10)
JumpdestOpAnalysis/STOP-6      758µs ± 2%     597µs ± 4%  -21.31%  (p=0.000 n=9+10)

name                        old speed      new speed      delta
JumpdestOpAnalysis/PUSH1-6  1.22GB/s ± 3%  1.05GB/s ± 3%  -14.16%  (p=0.000 n=10+10)
JumpdestOpAnalysis/STOP-6   1.61GB/s ± 3%  2.06GB/s ± 4%  +27.59%  (p=0.000 n=10+10)

For some reason, this change makes the worst-case slower.

[chfast]

I can benchmark this after #23499 is merged.

[chfast]

Please rebase.

[chfast]

Note that codeSegment() does not use the lookup table.

[holiman]

Please rebase.

Done!