Feature or enhancement
Proposal:
The JIT optimizer initially only allowed a inplace op rewrites. This led to us contorting the JIT uops to fit that.
Now that the JIT optimizer allows more than one op to be written at once, we should clean up and de-duplicate all the composite uops.
E.g.
REPLACE_OP(_POP_TOP_LOAD_CONST_INLINE_BORROW, ..)
should become
ADD_OP(_POP_TOP, ...)
ADD_OP(_LOAD_CONST_INLINE_BORROW, ...)
and _POP_TOP_LOAD_CONST_INLINE_BORROW should then be completely removed from the JIT uops and bytecodes.c. This will allow us to de-duplicate and refactor the code greatly, while making the existing optimizations also more effective as we don't have to do as much repeated work.
Has this already been discussed elsewhere?
This is a minor feature, which does not need previous discussion elsewhere
Links to previous discussion of this feature:
No response
Linked PRs
- gh-148211: decompose
_POP_TOP_LOAD_CONST_INLINE(_BORROW)in JIT #148230 - gh-148211: refactor bool to explicit uops in JIT #148258
- gh-148211: decompose
_INSERT_1_LOAD_CONST_INLINE(_BORROW)in JIT #148283 - gh-148211: decompose
[_POP_CALL_X/_SHUFFLE_2]_LOAD_CONST_INLINE_BORROWin JIT #148313 - gh-148211: decompose
[_POP_TWO/_INSERT_2]_LOAD_CONST_INLINE_BORROWin JIT #148357 - gh-148211: decompose
_POP_TWO/_POP_CALL(_ONE/_TWO)in JIT #148377 - gh-148211: decompose
_SHUFFLE_3_LOAD_CONST_INLINE_BORROWin JIT #148816