View Single Post
Old 2021-08-25, 17:34   #27
bsquared
 
bsquared's Avatar
 
"Ben"
Feb 2007

3×1,193 Posts
Default

Quote:
Originally Posted by bsquared View Post
Thanks to the autoreconf -si hint, I've progressed a little further and have run through a configure process for a sm_70 card. But now I'm getting a bunch of these errors:

Code:
ptxas /tmp/tmpxft_00008eaa_00000000-5_cudakernel.ptx, line 2378; error   : Instruction 'vote' without '.sync' is not supported on .target sm_70 and higher from PTX ISA version 6.4
which probably has nothing to do with your cbgn enhancements. Anyone know if gpu-ecm will build and run on a sm_70 card or hints on how to proceed?
Fixed it.

On line 10 of cudakernel_default.cu replace this line:

Code:
while(__any(cy[threadIdx.x])!=0)
with this one

Code:
while(__any_sync(__activemask(),cy[threadIdx.x])!=0)
Now I have a working gpu-ecm! Later I'll try to get the cbgn part working.

Code:
% echo "(2^997-1)" | ./ecm -gpu -sigma 3:1000 20000
GMP-ECM 7.0.5-dev [configured with GMP 6.2.0, --enable-asm-redc, --enable-gpu, --enable-assert] [ECM]
Input number is (2^997-1) (301 digits)
Using B1=20000, B2=3804582, sigma=3:1000-3:6119 (5120 curves)
GPU: Block: 32x32x1 Grid: 160x1x1 (5120 parallel curves)
Computing 5120 Step 1 took 485ms of CPU time / 4990ms of GPU time
Computing 5120 Step 2 on CPU took 247281ms

Last fiddled with by bsquared on 2021-08-25 at 17:52
bsquared is offline   Reply With Quote