|
|
|
@ -367,9 +367,6 @@ extern "C" int scanhash_decred(int thr_id, struct work* work, uint32_t max_nonce
@@ -367,9 +367,6 @@ extern "C" int scanhash_decred(int thr_id, struct work* work, uint32_t max_nonce
|
|
|
|
|
const dim3 grid((throughput + TPB-1)/(TPB)); |
|
|
|
|
const dim3 block(TPB); |
|
|
|
|
|
|
|
|
|
if (opt_benchmark) { |
|
|
|
|
ptarget[6] = swab32(0xff); |
|
|
|
|
} |
|
|
|
|
if (!init[thr_id]){ |
|
|
|
|
cudaSetDevice(dev_id); |
|
|
|
|
if (opt_cudaschedule == -1 && gpu_threads == 1) { |
|
|
|
@ -431,6 +428,8 @@ extern "C" int scanhash_decred(int thr_id, struct work* work, uint32_t max_nonce
@@ -431,6 +428,8 @@ extern "C" int scanhash_decred(int thr_id, struct work* work, uint32_t max_nonce
|
|
|
|
|
} |
|
|
|
|
*pnonce = work->nonces[0]; |
|
|
|
|
return rc; |
|
|
|
|
} else { |
|
|
|
|
gpulog(LOG_WARNING, thr_id, "result %u for %08x does not validate on CPU!", i, h_resNonce[thr_id][i]); |
|
|
|
|
} |
|
|
|
|
} |
|
|
|
|
} |
|
|
|
|