core/vm: use pointers to operations vs. copy by value

[karalabe]

This is a funky heuristic optimization based on the ideas from #21180. I'm not entirely sure why this makes our tight loop faster, but the snail tracer has a 5% speed gain vs master.

I was benchmarking via

$ evm --statdump --codefile snailtracer.hex run  1>/dev/null

Where the snail tracer is https://gist.github.com/karalabe/dc2fd6add66341c60e7cc96b91ba0f76

[holiman]

LGTM

[holiman]

I did a little experiment. master versus this pr, versus #21193

[user@work go-ethereum]$ for i in {1..5}; do build/bin/evm-master --statdump --codefile snailtracer.hex run  2>&1 1>/dev/null  | grep time ; done
execution time:  392.978976ms
execution time:  388.336157ms
execution time:  386.650593ms
execution time:  582.198802ms
execution time:  436.742533ms
[user@work go-ethereum]$ for i in {1..5}; do build/bin/evm-karalabe --statdump --codefile snailtracer.hex run  2>&1 1>/dev/null  | grep time ; done
execution time:  336.529811ms
execution time:  442.506244ms
execution time:  406.039679ms
execution time:  358.390552ms
execution time:  493.00324ms
[user@work go-ethereum]$ for i in {1..5}; do build/bin/evm-holiman --statdump --codefile snailtracer.hex run  2>&1 1>/dev/null  | grep time ; done
execution time:  345.169556ms
execution time:  333.998188ms
execution time:  433.470609ms
execution time:  439.249263ms
execution time:  340.805428ms

[holiman]

So the stats are :

• master: 437.4ms
• this PR: 407.3ms
• other PR: 378.5ms

[holiman]

With both PRs, I get this:

[user@work go-ethereum]$ for i in {1..5}; do build/bin/evm-both --statdump --codefile snailtracer.hex run  2>&1 1>/dev/null  | grep time ; done
execution time:  331.148506ms
execution time:  422.554855ms
execution time:  333.043146ms
execution time:  386.965078ms
execution time:  364.806642ms

average 367ms