Post B5654sOsM20a0Aycm8 by harold@mastodon.gamedev.place
 (DIR) More posts by harold@mastodon.gamedev.place
 (DIR) Post #B5654sOsM20a0Aycm8 by harold@mastodon.gamedev.place
       0 likes, 0 repeats
       
       Algorithm with unknown (but capped) output size has an a bit of annoying thing when you need to unroll it for ILP: you end up with two streams (merging the outputs likes to result in a high-latency loop-carried dependency). Unless I'm missing a trick there?Here's an idea: start in the middle and go both ways, doing half of it backwards. The output is still two streams, but they join up in the middle so you still get one contiguous span as the result.But I can't be the first to think of it
       
 (DIR) Post #B5654savdDeKbYcFYu by wolf480pl@mstdn.io
       0 likes, 0 repeats
       
       @harold kinda reminds me of Mill's instruction fetch