...
This would be cool. What's your plan here to make it faster?
(In reply to comment #1) > This would be cool. What's your plan here to make it faster? Do not collect BytecodeBasicBlock::m_offsets. Sometimes we do not need this information. I think this reduces memory allocation if the basic block is super large. Like, kraken's data initialization.
(In reply to comment #2) > (In reply to comment #1) > > This would be cool. What's your plan here to make it faster? > > Do not collect BytecodeBasicBlock::m_offsets. Sometimes we do not need this > information. > I think this reduces memory allocation if the basic block is super large. > Like, kraken's data initialization. Nice! I measured that it took something like 3-5ms to run the algorithm on the huge program code block from ai-astart. So any improvement for this would be awesome.