Generated TASM can often be simplified before it is executed on the VM. Consider for example
push 14
push 15
add
this can be replaced by
push 29
When we resolve addresses of list items and structs, we generate a lot of TASM code that can be simplified this way.
Optimally, everything that can be calculated at compile-time should be. Such that the VM's trace lengths can be minimized. A good place to start would be to implment TASM-constant-folding for these patterns
[ ] push a push b add
[ ] push a push b mul
[ ] push a push b eq
[ ] push 0 eq push 0 eq push 0 eq -> push 0 eq
[ ] push a push b lt
[ ] push a push b and
[ ] push a push b xor
[ ] push a push b pow
[ ] push a push b div
[ ] push a log_2_floor
[ ] push a invert
[ ] push a pop_count
[ ] All four XFieldElement-specific instructions
Concretely, this optimization step could be implemented as a method on InnerFunctionTasmCode but other interfaces are possible, too.
Generated TASM can often be simplified before it is executed on the VM. Consider for example
this can be replaced by
When we resolve addresses of list items and structs, we generate a lot of TASM code that can be simplified this way.
Optimally, everything that can be calculated at compile-time should be. Such that the VM's trace lengths can be minimized. A good place to start would be to implment TASM-constant-folding for these patterns
push a push b addpush a push b mulpush a push b eqpush 0 eq push 0 eq push 0 eq->push 0 eqpush a push b ltpush a push b andpush a push b xorpush a push b powpush a push b divpush a log_2_floorpush a invertpush a pop_countXFieldElement-specific instructionsConcretely, this optimization step could be implemented as a method on
InnerFunctionTasmCodebut other interfaces are possible, too.