gh-143421: Use new buffer to save optimized uops

[cocolato]

• Issue: JIT optimizer cleanups #143421

[python-cla-bot]

All commit authors signed the Contributor License Agreement.

[cocolato]

@Fidget-Spinner There may be many details that need fixing, and I hope you can give me some suggestions.

[cocolato]

I haven't yet removed the cases of REPLACE_OP((inst+1),...), perhaps there's a better way to achieve this.

cpython/Python/optimizer_analysis.c

Line 244 in 03e6457

REPLACE_OP((this_instr+1), _EXIT_TRACE, 0, 0);

[Fidget-Spinner]

Thanks, this is really valuable work. We definitely need this.

[Fidget-Spinner]

All these REPLACE_OP should be left untouched. As you already have a mechanism to copy over the current op if there's no ADD_OP called right :) ?

[cocolato]

Do you mean we need the _NOP to appear in both the old buffer and the new buffer?

[Fidget-Spinner]

No, you can tell the add_op function to skip if we see _NOP, and also tell the automatic add op check the same thing.

[cocolato]

No, you can tell the add_op function to skip if we see _NOP, and also tell the automatic add op check the same thing.

If i skip add _NOP to out_buffer in add_op, the original _GURAD_XX will be added automatically. If we use REPLACE_OP here, it will modify the original trace. So perhaps can use ADD_OP here?

[Fidget-Spinner]

All these REPLACE_OP should be left untouched as well. They should stay the same.

[cocolato]

Thanks for taking the time to review this! I will rewrite them later.