Beispiel #1
0
        public void Run(ProgramParameters parms)
        //int deviceId, int workGroupSize, int workSize, int numThreadsCreateWork, KernelType kernelt, int keysize, IEnumerable<string> patterns)
        {
            int        deviceId             = (int)parms.DeviceId;
            int        workGroupSize        = (int)parms.WorkGroupSize;
            int        workSize             = (int)parms.WorkSize;
            int        numThreadsCreateWork = (int)parms.CpuThreads;
            KernelType kernelt = parms.KernelType;
            int        keysize = (int)parms.KeySize;

            Console.WriteLine("Cooking up some delicions scallions...");
            this.workSize = (uint)workSize;
            profiler      = new Profiler();
            #region init
            profiler.StartRegion("init");

            // Combine patterns into a single regexp and build one of Richard's objects
            var rp = new RegexPattern(parms.Regex);

            // Create bitmasks array for the GPU
            var gpu_bitmasks = rp.GenerateOnionPatternBitmasksForGpu(MIN_CHARS)
                               .Select(t => TorBase32.ToUIntArray(TorBase32.CreateBase32Mask(t)))
                               .SelectMany(t => t).ToArray();
            //Create Hash Table
            uint[]   dataArray;
            ushort[] hashTable;
            uint[][] all_patterns;
            int      max_items_per_key = 0;
            {
                Func <uint[], ushort> fnv =
                    (pattern_arr) =>
                {
                    uint f = Util.FNVHash(pattern_arr[0], pattern_arr[1], pattern_arr[2]);
                    f = ((f >> 10) ^ f) & (uint)1023;
                    return((ushort)f);
                };
                all_patterns = rp.GenerateOnionPatternsForGpu(7)
                               .Select(i => TorBase32.ToUIntArray(TorBase32.FromBase32Str(i.Replace('.', 'a'))))
                               .ToArray();
                var gpu_dict_list = all_patterns
                                    .Select(i => new KeyValuePair <ushort, uint>(fnv(i), Util.FNVHash(i[0], i[1], i[2])))
                                    .GroupBy(i => i.Key)
                                    .OrderBy(i => i.Key)
                                    .ToList();

                dataArray = gpu_dict_list.SelectMany(i => i.Select(j => j.Value)).ToArray();
                hashTable = new ushort[1024];                 //item 1 index, item 2 length
                int currIndex = 0;
                foreach (var item in gpu_dict_list)
                {
                    int len = item.Count();
                    hashTable[item.Key] = (ushort)currIndex;
                    currIndex          += len;
                    if (len > max_items_per_key)
                    {
                        max_items_per_key = len;
                    }
                }

                Console.WriteLine("Putting {0} patterns into {1} buckets.", currIndex, gpu_dict_list.Count);
            }

            // Set the key size
            keySize = keysize;

            // Find kernel name and check key size
            kernel_type = kernelt;
            string kernelFileName = null, kernelName = null;
            switch (kernel_type)
            {
            case KernelType.Normal:
                kernelFileName = "kernel.cl";
                kernelName     = "normal";
                break;

            case KernelType.Optimized4_9:
                if (keySize != 1024)
                {
                    throw new ArgumentException("Kernel only works with keysize 1024.");
                }
                kernelFileName = "kernel.cl";
                kernelName     = "optimized";
                break;

            case KernelType.Optimized4_11:
                if (keySize != 2048 && keySize != 4096)
                {
                    throw new ArgumentException("Kernel only works with keysize 2048 or 4096.");
                }
                kernelFileName = "kernel.cl";
                kernelName     = "optimized";
                break;

            default:
                throw new ArgumentException("Pick a supported kernel.");
            }

            Console.WriteLine("Using kernel {0} from file {1} ({2})", kernelName, kernelFileName, kernel_type);

            //create device context and kernel
            CLDeviceInfo device = GetDevices()[deviceId];
            if ((uint)workGroupSize > device.MaxWorkGroupSize)
            {
                workGroupSize = (int)device.MaxWorkGroupSize;
            }
            Console.WriteLine("Using work group size {0}", workGroupSize);
            CLContext context = new CLContext(device.DeviceId);

            Console.Write("Compiling kernel... ");
            string kernel_text = KernelGenerator.GenerateKernel(parms, gpu_bitmasks.Length / 3, max_items_per_key, gpu_bitmasks.Take(3).ToArray(), all_patterns[0], all_patterns.Length);
            if (parms.SaveGeneratedKernelPath != null)
            {
                System.IO.File.WriteAllText(parms.SaveGeneratedKernelPath, kernel_text);
            }
            IntPtr program = context.CreateAndCompileProgram(kernel_text);

            var hashes_per_win = 0.5 / rp.GenerateAllOnionPatternsForRegex().Select(t => Math.Pow(2, -5 * t.Count(q => q != '.'))).Sum();
            Console.WriteLine("done.");

            CLKernel kernel = context.CreateKernel(program, kernelName);
            //Create buffers
            CLBuffer <uint> bufLastWs;
            CLBuffer <uint> bufMidstates;
            CLBuffer <int>  bufExpIndexes;
            CLBuffer <uint> bufResults;
            {
                int    num_exps   = (get_der_len(EXP_MAX) - get_der_len(EXP_MIN) + 1);
                uint[] LastWs     = new uint[num_exps * 16];
                uint[] Midstates  = new uint[num_exps * 5];
                int[]  ExpIndexes = new int[num_exps];
                uint[] Results    = new uint[128];

                bufLastWs     = context.CreateBuffer(OpenTK.Compute.CL10.MemFlags.MemReadOnly | OpenTK.Compute.CL10.MemFlags.MemCopyHostPtr, LastWs);
                bufMidstates  = context.CreateBuffer(OpenTK.Compute.CL10.MemFlags.MemReadOnly | OpenTK.Compute.CL10.MemFlags.MemCopyHostPtr, Midstates);
                bufExpIndexes = context.CreateBuffer(OpenTK.Compute.CL10.MemFlags.MemReadOnly | OpenTK.Compute.CL10.MemFlags.MemCopyHostPtr, ExpIndexes);
                bufResults    = context.CreateBuffer(OpenTK.Compute.CL10.MemFlags.MemReadWrite | OpenTK.Compute.CL10.MemFlags.MemCopyHostPtr, Results);
            }
            //Create pattern buffers
            CLBuffer <ushort> bufHashTable;
            CLBuffer <uint>   bufDataArray;
            CLBuffer <uint>   bufBitmasks;
            {
                bufHashTable = context.CreateBuffer(OpenTK.Compute.CL10.MemFlags.MemReadOnly | OpenTK.Compute.CL10.MemFlags.MemCopyHostPtr, hashTable);
                bufDataArray = context.CreateBuffer(OpenTK.Compute.CL10.MemFlags.MemReadOnly | OpenTK.Compute.CL10.MemFlags.MemCopyHostPtr, dataArray);
                bufBitmasks  = context.CreateBuffer(OpenTK.Compute.CL10.MemFlags.MemReadOnly | OpenTK.Compute.CL10.MemFlags.MemCopyHostPtr, gpu_bitmasks);
            }
            //Set kernel arguments
            lock (new object()) { }             // Empty lock, resolves (or maybe hides) a race condition in SetKernelArg
            kernel.SetKernelArg(0, bufLastWs);
            kernel.SetKernelArg(1, bufMidstates);
            kernel.SetKernelArg(2, bufResults);
            kernel.SetKernelArg(3, (uint)EXP_MIN);
            kernel.SetKernelArg(4, (byte)get_der_len(EXP_MIN));
            kernel.SetKernelArg(5, bufExpIndexes);
            kernel.SetKernelArg(6, bufBitmasks);
            kernel.SetKernelArg(7, bufHashTable);
            kernel.SetKernelArg(8, bufDataArray);
            profiler.EndRegion("init");

            bufBitmasks.EnqueueWrite(true);
            bufHashTable.EnqueueWrite(true);
            bufDataArray.EnqueueWrite(true);

            //start the thread to generate input data
            for (int i = 0; i < numThreadsCreateWork; i++)
            {
                Thread inputThread = new Thread(CreateInput);
                inputThread.Start();
                inputThreads.Add(inputThread);
            }
            Thread.Sleep(3000);            //wait just a bit so some work is available
            #endregion

            int loop = 0;

            var gpu_runtime_sw = System.Diagnostics.Stopwatch.StartNew();

            profiler.StartRegion("total without init");
            bool success = false;
            while (!success)
            {
                lock (this) { if (this.Abort)
                              {
                                  break;
                              }
                }                                                      //abort flag was set.... bail
                KernelInput input = null;
                lock (_kernelInput)
                {
                    if (_kernelInput.Count > 0)
                    {
                        input = _kernelInput.Pop();
                    }
                }
                if (input == null)                 //If we have run out of work sleep for a bit
                {
                    Console.WriteLine("Lack of work for the GPU!! Taking a nap!!");
                    Thread.Sleep(250);
                    continue;
                }

                profiler.StartRegion("set buffers");
                bufLastWs.Data     = input.LastWs;
                bufMidstates.Data  = input.Midstates;
                bufExpIndexes.Data = input.ExpIndexes;
                bufResults.Data    = input.Results;
                kernel.SetKernelArg(3, input.BaseExp);
                profiler.EndRegion("set buffers");

                profiler.StartRegion("write buffers");
                bufLastWs.EnqueueWrite(true);
                bufMidstates.EnqueueWrite(true);
                bufExpIndexes.EnqueueWrite(true);
                Array.Clear(bufResults.Data, 0, bufResults.Data.Length);
                bufResults.EnqueueWrite(true);
                profiler.EndRegion("write buffers");

                kernel.EnqueueNDRangeKernel(workSize, workGroupSize);

                profiler.StartRegion("read results");
                bufResults.EnqueueRead(false);
                profiler.EndRegion("read results");

                loop++;
                Console.Write("\r");
                long hashes = (long)workSize * (long)loop;

                Console.Write("LoopIteration:{0}  HashCount:{1:0.00}MH  Speed:{2:0.0}MH/s  Runtime:{3}  Predicted:{4}  ",
                              loop, hashes / 1000000.0d, hashes / gpu_runtime_sw.ElapsedMilliseconds / 1000.0d,
                              gpu_runtime_sw.Elapsed.ToString().Split('.')[0],
                              PredictedRuntime(hashes_per_win, hashes * 1000 / gpu_runtime_sw.ElapsedMilliseconds));

                profiler.StartRegion("check results");
                foreach (var result in input.Results)
                {
                    if (result != 0)
                    {
                        try
                        {
                            input.Rsa.ChangePublicExponent((BigNumber)result);

                            string onion_hash = input.Rsa.OnionHash;
                            Console.WriteLine("CPU checking hash: {0}", onion_hash);

                            if (rp.DoesOnionHashMatchPattern(onion_hash))
                            {
                                Console.WriteLine();
                                Console.WriteLine("Ding!! Delicious scallions for you!!");
                                Console.WriteLine();

                                string key = input.Rsa.Rsa.PrivateKeyAsPEM;

                                if (parms.KeyOutputPath != null)
                                {
                                    System.IO.File.AppendAllText(parms.KeyOutputPath, "Generated at: " + System.DateTime.Now.ToString("G") + "\n");
                                    System.IO.File.AppendAllText(parms.KeyOutputPath, "Address/Hash: " + onion_hash + ".onion\n");
                                    System.IO.File.AppendAllText(parms.KeyOutputPath, "RSA key: \n" + key + "\n\n");
                                }

                                Console.WriteLine("Exponent: {0}", result);
                                input.Rsa.ChangePublicExponent((BigNumber)result);
                                Console.WriteLine("Address/Hash: " + onion_hash + ".onion");
                                Console.WriteLine();
                                Console.WriteLine(key);
                                Console.WriteLine();
                                if (!parms.ContinueGeneration)
                                {
                                    success = true;
                                }
                            }
                        }
                        catch (OpenSslException /*ex*/) { }
                    }
                }
                profiler.EndRegion("check results");
            }

            foreach (var thread in inputThreads)
            {
                thread.Abort();
            }
            profiler.EndRegion("total without init");
            Console.WriteLine(profiler.GetSummaryString());
            Console.WriteLine("{0:0.00} million hashes per second", ((long)loop * (long)workSize * (long)1000) / (double)profiler.GetTotalMS("total without init") / (double)1000000);
        }
Beispiel #2
0
        public void Run(ProgramParameters parms)
        //int deviceId, int workGroupSize, int workSize, int numThreadsCreateWork, KernelType kernelt, int keysize, IEnumerable<string> patterns)
        {
            int        deviceId             = (int)parms.DeviceId;
            int        workGroupSize        = (int)parms.WorkGroupSize;
            int        workSize             = (int)parms.WorkSize;
            int        numThreadsCreateWork = (int)parms.CpuThreads;
            KernelType kernelt = parms.KernelType;
            int        keysize = (int)parms.KeySize;

            Console.WriteLine("Cooking up some delicions scallions...");
            this.workSize = (uint)workSize;
            profiler      = new Profiler();
            #region init
            profiler.StartRegion("init");

            // Combine patterns into a single regexp and build one of Richard's objects
            var rp = new RegexPattern(parms.Regex);

            // Create bitmasks array for the GPU
            var gpu_bitmasks = rp.GenerateOnionPatternBitmasksForGpu(MIN_CHARS)
                               .Select(t => TorBase32.ToUIntArray(TorBase32.CreateBase32Mask(t)))
                               .SelectMany(t => t).ToArray();
            //Create Hash Table
            uint[]   dataArray;
            ushort[] hashTable;
            uint[][] all_patterns;
            int      max_items_per_key = 0;
            {
                Func <uint[], ushort> fnv =
                    (pattern_arr) =>
                {
                    uint f = Util.FNVHash(pattern_arr[0], pattern_arr[1], pattern_arr[2]);
                    f = ((f >> 10) ^ f) & (uint)1023;
                    return((ushort)f);
                };
                all_patterns = rp.GenerateOnionPatternsForGpu(7)
                               .Select(i => TorBase32.ToUIntArray(TorBase32.FromBase32Str(i.Replace('.', 'a'))))
                               .ToArray();
                var gpu_dict_list = all_patterns
                                    .Select(i => new KeyValuePair <ushort, uint>(fnv(i), Util.FNVHash(i[0], i[1], i[2])))
                                    .GroupBy(i => i.Key)
                                    .OrderBy(i => i.Key)
                                    .ToList();

                dataArray = gpu_dict_list.SelectMany(i => i.Select(j => j.Value)).ToArray();
                hashTable = new ushort[1024];                 //item 1 index, item 2 length
                int currIndex = 0;
                foreach (var item in gpu_dict_list)
                {
                    int len = item.Count();
                    hashTable[item.Key] = (ushort)currIndex;
                    currIndex          += len;
                    if (len > max_items_per_key)
                    {
                        max_items_per_key = len;
                    }
                }

                Console.WriteLine("Putting {0} patterns into {1} buckets.", currIndex, gpu_dict_list.Count);
            }

            // Set the key size
            keySize = keysize;

            // Find kernel name and check key size
            kernel_type = kernelt;
            string kernelFileName = null, kernelName = null;
            switch (kernel_type)
            {
            case KernelType.Normal:
                kernelFileName = "kernel.cl";
                kernelName     = "normal";
                break;

            case KernelType.Optimized4:
                kernelFileName = "kernel.cl";
                kernelName     = "optimized";
                break;

            default:
                throw new ArgumentException("Pick a supported kernel.");
            }

            Console.WriteLine("Using kernel {0} from file {1} ({2})", kernelName, kernelFileName, kernel_type);

            //create device context and kernel
            CLDeviceInfo device = GetDevices()[deviceId];
            if ((uint)workGroupSize > device.MaxWorkGroupSize)
            {
                workGroupSize = (int)device.MaxWorkGroupSize;
            }
            Console.WriteLine("Using work group size {0}", workGroupSize);
            CLContext context = new CLContext(device.DeviceId);

            Console.Write("Compiling kernel... ");
            string kernel_text = KernelGenerator.GenerateKernel(parms, gpu_bitmasks.Length / 3, max_items_per_key, gpu_bitmasks.Take(3).ToArray(), all_patterns[0], all_patterns.Length, parms.ExponentIndex);
            if (parms.SaveGeneratedKernelPath != null)
            {
                System.IO.File.WriteAllText(parms.SaveGeneratedKernelPath, kernel_text);
            }
            IntPtr program = context.CreateAndCompileProgram(kernel_text);

            var hashes_per_win = 0.5 / rp.GenerateAllOnionPatternsForRegex().Select(t => Math.Pow(2, -5 * t.Count(q => q != '.'))).Sum();
            Console.WriteLine("done.");

            //
            // Test SHA1 algo
            //
            {
                Console.WriteLine("Testing SHA1 hash...");

                CLKernel        shaTestKern = context.CreateKernel(program, "shaTest");
                CLBuffer <uint> bufSuccess  = context.CreateBuffer(OpenTK.Compute.CL10.MemFlags.MemReadWrite | OpenTK.Compute.CL10.MemFlags.MemCopyHostPtr, new uint[5]);
                shaTestKern.SetKernelArg(0, bufSuccess);

                shaTestKern.EnqueueNDRangeKernel(workSize, workGroupSize);

                bufSuccess.EnqueueRead(false);

                // Calculate the SHA1 CPU-side
                System.Security.Cryptography.SHA1 sha = new System.Security.Cryptography.SHA1CryptoServiceProvider();

                String        testdata = "Hello world!";
                byte[]        cpuhash  = sha.ComputeHash(Encoding.ASCII.GetBytes(testdata));
                StringBuilder cpuhex   = new StringBuilder(cpuhash.Length * 2);
                foreach (byte b in cpuhash)
                {
                    cpuhex.AppendFormat("{0:x2}", b);
                }
                Console.WriteLine("CPU SHA-1: {0}", cpuhex.ToString());

                // Convert the SHA1 GPU-side to hex
                String gpuhex = String.Format("{0:x8}{1:x8}{2:x8}{3:x8}{4:x8}", bufSuccess.Data[0], bufSuccess.Data[1], bufSuccess.Data[2], bufSuccess.Data[3], bufSuccess.Data[4]);

                Console.WriteLine("GPU SHA-1: {0}", gpuhex);

                if (gpuhex != cpuhex.ToString())
                {
                    Console.WriteLine();
                    Console.WriteLine("******************************* ERROR ERROR ERROR *******************************");
                    Console.WriteLine("*                                                                               *");
                    Console.WriteLine("* GPU and CPU SHA-1 calculations do NOT match.                                  *");
                    Console.WriteLine("* Hashing will NOT work until this is resolved.                                 *");
                    Console.WriteLine("* The program will continue, but WILL NOT find a valid match.                   *");
                    Console.WriteLine("*                                                                               *");
                    Console.WriteLine("* See https://github.com/lachesis/scallion/issues/11#issuecomment-29046835      *");
                    Console.WriteLine("*                                                                               *");
                    Console.WriteLine("*********************************************************************************");
                    Console.WriteLine();
                }
                else
                {
                    Console.WriteLine("Looks good!");
                }
            }

            CLKernel kernel = context.CreateKernel(program, kernelName);
            //Create buffers
            CLBuffer <uint> bufLastWs;
            CLBuffer <uint> bufMidstates;
            CLBuffer <int>  bufExpIndexes;
            CLBuffer <uint> bufResults;
            {
                int    num_exps   = (get_der_len(EXP_MAX) - get_der_len(EXP_MIN) + 1);
                uint[] LastWs     = new uint[num_exps * 16];
                uint[] Midstates  = new uint[num_exps * 5];
                int[]  ExpIndexes = new int[num_exps];
                uint[] Results    = new uint[128];

                bufLastWs     = context.CreateBuffer(OpenTK.Compute.CL10.MemFlags.MemReadOnly | OpenTK.Compute.CL10.MemFlags.MemCopyHostPtr, LastWs);
                bufMidstates  = context.CreateBuffer(OpenTK.Compute.CL10.MemFlags.MemReadOnly | OpenTK.Compute.CL10.MemFlags.MemCopyHostPtr, Midstates);
                bufExpIndexes = context.CreateBuffer(OpenTK.Compute.CL10.MemFlags.MemReadOnly | OpenTK.Compute.CL10.MemFlags.MemCopyHostPtr, ExpIndexes);
                bufResults    = context.CreateBuffer(OpenTK.Compute.CL10.MemFlags.MemReadWrite | OpenTK.Compute.CL10.MemFlags.MemCopyHostPtr, Results);
            }
            //Create pattern buffers
            CLBuffer <ushort> bufHashTable;
            CLBuffer <uint>   bufDataArray;
            CLBuffer <uint>   bufBitmasks;
            {
                bufHashTable = context.CreateBuffer(OpenTK.Compute.CL10.MemFlags.MemReadOnly | OpenTK.Compute.CL10.MemFlags.MemCopyHostPtr, hashTable);
                bufDataArray = context.CreateBuffer(OpenTK.Compute.CL10.MemFlags.MemReadOnly | OpenTK.Compute.CL10.MemFlags.MemCopyHostPtr, dataArray);
                bufBitmasks  = context.CreateBuffer(OpenTK.Compute.CL10.MemFlags.MemReadOnly | OpenTK.Compute.CL10.MemFlags.MemCopyHostPtr, gpu_bitmasks);
            }
            //Set kernel arguments
            lock (new object()) { }             // Empty lock, resolves (or maybe hides) a race condition in SetKernelArg
            kernel.SetKernelArg(0, bufLastWs);
            kernel.SetKernelArg(1, bufMidstates);
            kernel.SetKernelArg(2, bufResults);
            kernel.SetKernelArg(3, (uint)EXP_MIN);
            kernel.SetKernelArg(4, (byte)get_der_len(EXP_MIN));
            kernel.SetKernelArg(5, bufExpIndexes);
            kernel.SetKernelArg(6, bufBitmasks);
            kernel.SetKernelArg(7, bufHashTable);
            kernel.SetKernelArg(8, bufDataArray);
            profiler.EndRegion("init");

            bufBitmasks.EnqueueWrite(true);
            bufHashTable.EnqueueWrite(true);
            bufDataArray.EnqueueWrite(true);

            //start the thread to generate input data
            for (int i = 0; i < numThreadsCreateWork; i++)
            {
                Thread inputThread = new Thread(CreateInput);
                inputThread.Start();
                inputThreads.Add(inputThread);
            }
            Thread.Sleep(3000);            //wait just a bit so some work is available
            #endregion

            int loop = 0;

            var gpu_runtime_sw = System.Diagnostics.Stopwatch.StartNew();

            profiler.StartRegion("total without init");
            bool success = false;
            while (!success)
            {
                lock (this) { if (this.Abort)
                              {
                                  break;
                              }
                }                                                      //abort flag was set.... bail
                KernelInput input = null;
                lock (_kernelInput)
                {
                    if (_kernelInput.Count > 0)
                    {
                        input = _kernelInput.Pop();
                    }
                }
                if (input == null)                 //If we have run out of work sleep for a bit
                {
                    Console.WriteLine("Lack of work for the GPU!! Taking a nap!!");
                    Thread.Sleep(250);
                    continue;
                }

                profiler.StartRegion("set buffers");
                bufLastWs.Data     = input.LastWs;
                bufMidstates.Data  = input.Midstates;
                bufExpIndexes.Data = input.ExpIndexes;
                bufResults.Data    = input.Results;
                kernel.SetKernelArg(3, input.BaseExp);
                profiler.EndRegion("set buffers");

                profiler.StartRegion("write buffers");
                bufLastWs.EnqueueWrite(true);
                bufMidstates.EnqueueWrite(true);
                bufExpIndexes.EnqueueWrite(true);
                Array.Clear(bufResults.Data, 0, bufResults.Data.Length);
                bufResults.EnqueueWrite(true);
                profiler.EndRegion("write buffers");

                kernel.EnqueueNDRangeKernel(workSize, workGroupSize);

                profiler.StartRegion("read results");
                bufResults.EnqueueRead(false);
                profiler.EndRegion("read results");

                loop++;
                Console.Write("\r");
                long hashes = (long)workSize * (long)loop;

                Console.Write("LoopIteration:{0}  HashCount:{1:0.00}MH  Speed:{2:0.0}MH/s  Runtime:{3}  Predicted:{4}  ",
                              loop, hashes / 1000000.0d, hashes / gpu_runtime_sw.ElapsedMilliseconds / 1000.0d,
                              gpu_runtime_sw.Elapsed.ToString().Split('.')[0],
                              PredictedRuntime(hashes_per_win, hashes * 1000 / gpu_runtime_sw.ElapsedMilliseconds));

                profiler.StartRegion("check results");
                foreach (var result in input.Results)
                {
                    if (result != 0)
                    {
                        try
                        {
                            input.Rsa.Rsa.PublicExponent = (BigNumber)result;

                            string onion_hash = input.Rsa.OnionHash;
                            Console.WriteLine("CPU checking hash: {0}", onion_hash);

                            if (rp.DoesOnionHashMatchPattern(onion_hash))
                            {
                                input.Rsa.ChangePublicExponent(result);
                                OutputKey(input.Rsa);

                                if (!parms.ContinueGeneration)
                                {
                                    success = true;
                                }
                            }
                        }
                        catch (OpenSslException /*ex*/) { }
                    }
                }
                profiler.EndRegion("check results");

                // Mark key as used (if configured)
                if (parms.UsedModuliFile != null)
                {
                    parms.UsedModuliFile.WriteLine(input.Rsa.Rsa.PublicModulus.ToDecimalString());
                }
            }

            foreach (var thread in inputThreads)
            {
                thread.Abort();
            }
            profiler.EndRegion("total without init");
            Console.WriteLine(profiler.GetSummaryString());
            Console.WriteLine("{0:0.00} million hashes per second", ((long)loop * (long)workSize * (long)1000) / (double)profiler.GetTotalMS("total without init") / (double)1000000);
        }