hamsi: TPB of 128 give better results (+10kh)

x13/cuda_x13_hamsi512.cu

@@ -688,11 +688,11 @@ void x13_hamsi512_cpu_init(int thr_id, int threads)
 __host__
 void x13_hamsi512_cpu_hash_64(int thr_id, int threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order)
 {
-	const int threadsperblock = 256;
+	const int threadsperblock = 128;
 
 	dim3 grid((threads + threadsperblock-1)/threadsperblock);
 	dim3 block(threadsperblock);
 
 	x13_hamsi512_gpu_hash_64<<<grid, block>>>(threads, startNounce, (uint64_t*)d_hash, d_nonceVector);
-	MyStreamSynchronize(NULL, order, thr_id);
-}
+	//MyStreamSynchronize(NULL, order, thr_id);
+}