A branch of sgminer optimized with GCN cross lane instructions on AMD (ethash, phi2, lyra2Z[z], allium, x25x, lyra2REv2/v3, argon2d, yescrypt, neoscrypt, 0x10)
Sharing ulong[192] in 4 threads make speed faster. Sharing it in 2 threads make speed faster. So "sharing" it in 1 thread should be the best? I wonder how vgpr spill can slow down the speed but worth trying. However, according to @mikerodey 's info -- "Compiler will generate a lot of replays into Global Memory, resulting in huge pipeline stalls. This will be very slow, especially compared to current closed-sourced lyra2z and phi2 implementations." So the result seems uncertain.
Sharing ulong[192] in 4 threads make speed faster. Sharing it in 2 threads make speed faster. So "sharing" it in 1 thread should be the best? I wonder how vgpr spill can slow down the speed but worth trying. However, according to @mikerodey 's info -- "Compiler will generate a lot of replays into Global Memory, resulting in huge pipeline stalls. This will be very slow, especially compared to current closed-sourced lyra2z and phi2 implementations." So the result seems uncertain.