Hey JeanLuc
tried on linux got no compilation error even with CCAP=8.6 ... windows is driving me nuts
on the illegal memory access issue when searching complete addresses..
does it help you to know that every gridsize x1 is working but every x2 x128 whatever is throwing that error?
i can take 83968x1 and it is working (slow but working) 1x1 working 1x2 error
cuda-memcheck.. same func call comp_keys_comp
GPU: GPU #0 GeForce RTX 3090 (82x0 cores) Grid(656x128)
========= Misaligned Shared or Local Address
========= at 0x00008980 in comp_keys_comp(unsigned short*, unsigned int*, unsigned long*, unsigned int, unsigned int*)
========= by thread (86,0,0) in block (300,0,0)
i'm still not good enough to degug that correctly
Try remove all compute_XX and sm_XX in line #63 VanitySearch.vcxproj and try to add only compute_86,sm_86. In new line after #156 in file /gpu/GPUEngine.cu add "{0x86, 82},"
Now youcan try compile in Windows. In Visual Studio you must select ReleaseX64
By the way: You can try set in command line when you try start searching parameter -g 1920,512 and check is it better than before? if yes - you can try to search the best value in place when now is 1920. If it not work - change 512 to 256.
Sorry for bad value of this post but i add them from phone😉
I hope that tips was help solve your issue.