Scrypt opencl kernel written in AMD GCN ISA assembly language
Only tested on my R9 270
By now, the hashrate is slower than the original kernel compiled by AMD APP SDK. On my R9 270, gcn-scrypt hashes at 470KH/s, while the original kernel hashes at 488KH/s. I'm working on optimizing the hashrate.
- Build clyfish/sgminer
- Build clyfish/gcnasm
- Set
gcnasmin PATH, thenmake. - Copy
kernel.binorkernel_lds.bininoutputto sgminer director, rename to correct name, egpswPitcairnglg2tc5121nf10w256l4.bin. - Run sgminer
kernel_lds.bin uses 84 VGPR, so it can run 12 wavefronts per compute unit,
while kernel.bin and the original kernel can only run 8 wavefronts per compute unit.
But, on my R9 270, kernel.bin hashes at 470KH/s, and kernel_lds.bin hashes at 440KH/s.
- BTC: 1N9THnTL2Xp7fDktEZR5QGrr7tjVLkg5oH
- LTC: LZ8G5P553yFBVai4desySw3CNqnU9ynEVm
- DOGE: DL37GuBD8yrbG14Krit9rmXFe9PpXAfB1w