stranged result to CUDA

I got a Tesla S870.

use it for cryto-analysis. Currently, implemented DES Algorithm.

But, have some problems.

First of all, result of des implemented CUDA(device) always is the same. Initial an ttempt have obvious result. however, from the second time, return the same value(result) if ever key change. I always perform initialization of a storage for result, subkey, and so on. what’s problem?

Second,

fllowing is content to my .cubin file.

code {

      name = _Z8des_testPh

      lmem = 256

      smem = 24

      reg = 68

      bar = 0

      bincode {.....}

}

maximum register 8192 per multiprocessor, 8192/68 = 120.47…

Therefore number of threads is 120.

Multiprocessor is set of thread blocks. Is right?

So, because of 8192 registers per multiprocessor, if number of threads is 120, number of thread blocks is 1. Is right?

I’m not very good at English. But I hope you understand.

wait for your reply…