13369064240

Committed 17 Feb 2025 11:20AM UTC coverage: 91.643% (+0.006%) from 91.637%

Build # 13369064240

Build Type

push

github

Committed by

web-flow

Commit Message

Merge pull request #4693 from randombit/jack/speed-compare

Add script to help compare performance changes

Run Details

94979 of 103640 relevant lines covered (91.64%)

11047273.77 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

78.4

/src/cli/speed.cpp

/*
* (C) 2009,2010,2014,2015,2017,2018,2024 Jack Lloyd
* (C) 2015 Simon Warta (Kullo GmbH)
*
* Botan is released under the Simplified BSD License (see license.txt)
*/

#include "cli.h"
#include "perf.h"

#include <algorithm>
#include <chrono>
#include <iomanip>
#include <map>
#include <set>
#include <sstream>

// Always available:
#include <botan/version.h>
#include <botan/internal/cpuid.h>
#include <botan/internal/fmt.h>
#include <botan/internal/stl_util.h>

#if defined(BOTAN_HAS_OS_UTILS)
   #include <botan/internal/os_utils.h>
#endif

#if defined(BOTAN_HAS_ECC_GROUP)
   #include <botan/ec_group.h>
#endif

namespace Botan_CLI {

namespace {

class JSON_Output final {
   public:
      void add(const Timer& timer) { m_results.push_back(timer); }

      std::string print() const {
         std::ostringstream out;

         out << "[\n";

         out << "{"
             << "\"arch\": \"" << BOTAN_TARGET_ARCH << "\", "
             << "\"version\": \"" << Botan::short_version_cstr() << "\", "
             << "\"git\": \"" << BOTAN_VERSION_VC_REVISION << "\", "
             << "\"compiler\": \"" << BOTAN_COMPILER_INVOCATION_STRING << "\""
             << "},\n";

         for(size_t i = 0; i != m_results.size(); ++i) {
            const Timer& t = m_results[i];

            out << "{"
                << "\"algo\": \"" << t.get_name() << "\", "
                << "\"op\": \"" << t.doing() << "\", "
                << "\"events\": " << t.events() << ", ";

            if(t.cycles_consumed() > 0) {
               out << "\"cycles\": " << t.cycles_consumed() << ", ";
            }

            if(t.buf_size() > 0) {
               out << "\"bps\": " << static_cast<uint64_t>(t.events() / (t.value() / 1000000000.0)) << ", ";
               out << "\"buf_size\": " << t.buf_size() << ", ";
            }

            out << "\"nanos\": " << t.value() << "}";

            if(i != m_results.size() - 1) {
               out << ",";
            }

            out << "\n";
         }
         out << "]\n";

         return out.str();
      }

   private:
      std::vector<Timer> m_results;
};

class Summary final {
   public:
      Summary() = default;

      void add(const Timer& t) {
         if(t.buf_size() == 0) {
            m_ops_entries.push_back(t);
         } else {
            m_bps_entries[std::make_pair(t.doing(), t.get_name())].push_back(t);
         }
      }

      std::string print() {
         const size_t name_padding = 35;
         const size_t op_name_padding = 16;
         const size_t op_padding = 16;

         std::ostringstream result_ss;
         result_ss << std::fixed;

         if(!m_bps_entries.empty()) {
            result_ss << "\n";

            // add table header
            result_ss << std::setw(name_padding) << std::left << "algo" << std::setw(op_name_padding) << std::left
                      << "operation";

            for(const Timer& t : m_bps_entries.begin()->second) {
               result_ss << std::setw(op_padding) << std::right << (std::to_string(t.buf_size()) + " bytes");
            }
            result_ss << "\n";

            // add table entries
            for(const auto& entry : m_bps_entries) {
               if(entry.second.empty()) {
                  continue;
               }

               result_ss << std::setw(name_padding) << std::left << (entry.first.second) << std::setw(op_name_padding)
                         << std::left << (entry.first.first);

               for(const Timer& t : entry.second) {
                  if(t.events() == 0) {
                     result_ss << std::setw(op_padding) << std::right << "N/A";
                  } else {
                     result_ss << std::setw(op_padding) << std::right << std::setprecision(2)
                               << (t.bytes_per_second() / 1000.0);
                  }
               }

               result_ss << "\n";
            }

            result_ss << "\n[results are the number of 1000s bytes processed per second]\n";
         }

         if(!m_ops_entries.empty()) {
            result_ss << std::setprecision(6) << "\n";

            // sort entries
            std::sort(m_ops_entries.begin(), m_ops_entries.end());

            // add table header
            result_ss << std::setw(name_padding) << std::left << "algo" << std::setw(op_name_padding) << std::left
                      << "operation" << std::setw(op_padding) << std::right << "sec/op" << std::setw(op_padding)
                      << std::right << "op/sec"
                      << "\n";

            // add table entries
            for(const Timer& entry : m_ops_entries) {
               result_ss << std::setw(name_padding) << std::left << entry.get_name() << std::setw(op_name_padding)
                         << std::left << entry.doing() << std::setw(op_padding) << std::right
                         << entry.seconds_per_event() << std::setw(op_padding) << std::right
                         << entry.events_per_second() << "\n";
            }
         }

         return result_ss.str();
      }

   private:
      std::map<std::pair<std::string, std::string>, std::vector<Timer>> m_bps_entries;
      std::vector<Timer> m_ops_entries;
};

std::vector<size_t> unique_buffer_sizes(const std::string& cmdline_arg) {
   const size_t MAX_BUF_SIZE = 64 * 1024 * 1024;

   std::set<size_t> buf;
   for(const std::string& size_str : Command::split_on(cmdline_arg, ',')) {
      size_t x = 0;
      try {
         size_t converted = 0;
         x = static_cast<size_t>(std::stoul(size_str, &converted, 0));

         if(converted != size_str.size()) {
            throw CLI_Usage_Error("Invalid integer");
         }
      } catch(std::exception&) {
         throw CLI_Usage_Error("Invalid integer value '" + size_str + "' for option buf-size");
      }

      if(x == 0) {
         throw CLI_Usage_Error("Cannot have a zero-sized buffer");
      }

      if(x > MAX_BUF_SIZE) {
         throw CLI_Usage_Error("Specified buffer size is too large");
      }

      buf.insert(x);
   }

   return std::vector<size_t>(buf.begin(), buf.end());
}

std::string format_timer(const Timer& t, size_t time_unit) {
   constexpr size_t MiB = 1024 * 1024;

   std::ostringstream oss;

   oss << t.get_name() << " ";

   const uint64_t events = t.events();

   if(t.buf_size() == 0) {
      // Report operations/time unit

      if(events == 0) {
         oss << "no events ";
      } else {
         oss << static_cast<uint64_t>(t.events_per_second()) << ' ' << t.doing() << "/sec; ";

         if(time_unit == 1000) {
            oss << std::setprecision(2) << std::fixed << (t.milliseconds() / events) << " ms/op ";
         } else if(time_unit == 1000 * 1000) {
            oss << std::setprecision(2) << std::fixed << (t.microseconds() / events) << " us/op ";
         } else if(time_unit == 1000 * 1000 * 1000) {
            oss << std::setprecision(0) << std::fixed << (t.nanoseconds() / events) << " ns/op ";
         }

         if(t.cycles_consumed() != 0 && events > 0) {
            const double cycles_per_op = static_cast<double>(t.cycles_consumed()) / events;
            const int precision = (cycles_per_op < 10000) ? 2 : 0;
            oss << std::fixed << std::setprecision(precision) << cycles_per_op << " cycles/op ";
         }

         oss << "(" << events << " " << (events == 1 ? "op" : "ops") << " in " << t.milliseconds() << " ms)";
      }
   } else {
      // Bulk op - report bytes/time unit

      const double MiB_total = static_cast<double>(events) / MiB;
      const double MiB_per_sec = MiB_total / t.seconds();

      if(!t.doing().empty()) {
         oss << t.doing() << " ";
      }

      if(t.buf_size() > 0) {
         oss << "buffer size " << t.buf_size() << " bytes: ";
      }

      if(events == 0) {
         oss << "N/A ";
      } else {
         oss << std::fixed << std::setprecision(3) << MiB_per_sec << " MiB/sec ";
      }

      if(t.cycles_consumed() != 0 && events > 0) {
         const double cycles_per_byte = static_cast<double>(t.cycles_consumed()) / events;
         oss << std::fixed << std::setprecision(2) << cycles_per_byte << " cycles/byte ";
      }

      oss << "(" << MiB_total << " MiB in " << t.milliseconds() << " ms)";
   }

   return oss.str();
}

}  // namespace

class Speed final : public Command {
   public:
      Speed() :
            Command(
               "speed --msec=500 --format=default --time-unit=ms --ecc-groups= --buf-size=1024 --clear-cpuid= --cpu-clock-speed=0 --cpu-clock-ratio=1.0 *algos") {
      }

      static std::vector<std::string> default_benchmark_list() {
         /*
         This is not intended to be exhaustive: it just hits the high
         points of the most interesting or widely used algorithms.
         */
         // clang-format off
         return {
            /* Block ciphers */
            "AES-128",
            "AES-192",
            "AES-256",
            "ARIA-128",
            "ARIA-192",
            "ARIA-256",
            "Blowfish",
            "CAST-128",
            "Camellia-128",
            "Camellia-192",
            "Camellia-256",
            "DES",
            "TripleDES",
            "GOST-28147-89",
            "IDEA",
            "Noekeon",
            "SHACAL2",
            "SM4",
            "Serpent",
            "Threefish-512",
            "Twofish",

            /* Cipher modes */
            "AES-128/CBC",
            "AES-128/CTR-BE",
            "AES-128/EAX",
            "AES-128/OCB",
            "AES-128/GCM",
            "AES-128/XTS",
            "AES-128/SIV",

            "Serpent/CBC",
            "Serpent/CTR-BE",
            "Serpent/EAX",
            "Serpent/OCB",
            "Serpent/GCM",
            "Serpent/XTS",
            "Serpent/SIV",

            "ChaCha20Poly1305",

            /* Stream ciphers */
            "RC4",
            "Salsa20",
            "ChaCha20",

            /* Hashes */
            "SHA-1",
            "SHA-256",
            "SHA-512",
            "SHA-3(256)",
            "SHA-3(512)",
            "RIPEMD-160",
            "Skein-512",
            "Blake2b",
            "Whirlpool",

            /* XOFs */
            "SHAKE-128",
            "SHAKE-256",

            /* MACs */
            "CMAC(AES-128)",
            "HMAC(SHA-256)",

            /* pubkey */
            "RSA",
            "DH",
            "ECDH",
            "ECDSA",
            "Ed25519",
            "Ed448",
            "X25519",
            "X448",
            "ML-KEM",
            "ML-DSA",
            "SLH-DSA",
            "FrodoKEM",
            "HSS-LMS",
         };
         // clang-format on
      }

      std::string group() const override { return "misc"; }

      std::string description() const override { return "Measures the speed of algorithms"; }

      void go() override {
         std::chrono::milliseconds msec(get_arg_sz("msec"));
         std::vector<std::string> ecc_groups = Command::split_on(get_arg("ecc-groups"), ',');
         const std::string format = get_arg("format");
         const std::string clock_ratio = get_arg("cpu-clock-ratio");

         const size_t clock_speed = get_arg_sz("cpu-clock-speed");

         double clock_cycle_ratio = std::strtod(clock_ratio.c_str(), nullptr);

         m_time_unit = [](std::string_view tu) {
            if(tu == "ms") {
               return 1000;
            } else if(tu == "us") {
               return 1000 * 1000;
            } else if(tu == "ns") {
               return 1000 * 1000 * 1000;
            } else {
               throw CLI_Usage_Error("Unknown time unit (supported: ms, us, ns)");
            }
         }(get_arg("time-unit"));

         /*
         * This argument is intended to be the ratio between the cycle counter
         * and the actual machine cycles. It is extremely unlikely that there is
         * any machine where the cycle counter increments faster than the actual
         * clock.
         */
         if(clock_cycle_ratio < 0.0 || clock_cycle_ratio > 1.0) {
            throw CLI_Usage_Error("Unlikely CPU clock ratio of " + clock_ratio);
         }

         clock_cycle_ratio = 1.0 / clock_cycle_ratio;

#if defined(BOTAN_HAS_OS_UTILS)
         if(clock_speed != 0 && Botan::OS::get_cpu_cycle_counter() != 0) {
            error_output() << "The --cpu-clock-speed option is only intended to be used on "
                              "platforms without access to a cycle counter.\n"
                              "Expect incorrect results\n\n";
         }
#endif

         if(format == "table") {
            m_summary = std::make_unique<Summary>();
         } else if(format == "json") {
            m_json = std::make_unique<JSON_Output>();
         } else if(format != "default") {
            throw CLI_Usage_Error("Unknown --format type '" + format + "'");
         }

#if defined(BOTAN_HAS_ECC_GROUP)
         if(ecc_groups.empty()) {
            ecc_groups = {"secp256r1", "secp384r1", "secp521r1", "brainpool256r1", "brainpool384r1", "brainpool512r1"};
         } else if(ecc_groups.size() == 1 && ecc_groups[0] == "all") {
            auto all = Botan::EC_Group::known_named_groups();
            ecc_groups.assign(all.begin(), all.end());
         }
#endif

         std::vector<std::string> algos = get_arg_list("algos");

         const std::vector<size_t> buf_sizes = unique_buffer_sizes(get_arg("buf-size"));

         for(const std::string& cpuid_to_clear : Command::split_on(get_arg("clear-cpuid"), ',')) {
            auto bits = Botan::CPUID::bit_from_string(cpuid_to_clear);
            if(bits.empty()) {
               error_output() << "Warning don't know CPUID flag '" << cpuid_to_clear << "'\n";
            }

            for(auto bit : bits) {
               Botan::CPUID::clear_cpuid_bit(bit);
            }
         }

         if(verbose() || m_summary) {
            output() << Botan::version_string() << "\n"
                     << "CPUID: " << Botan::CPUID::to_string() << "\n\n";
         }

         const bool using_defaults = (algos.empty());
         if(using_defaults) {
            algos = default_benchmark_list();
         }

         PerfConfig perf_config([&](const Timer& t) { this->record_result(t); },
                                clock_speed,
                                clock_cycle_ratio,
                                msec,
                                ecc_groups,
                                buf_sizes,
                                this->error_output(),
                                this->rng());

         for(const auto& algo : algos) {
            if(auto perf = PerfTest::get(algo)) {
               perf->go(perf_config);
            } else if(verbose() || !using_defaults) {
               error_output() << "Unknown algorithm '" << algo << "'\n";
            }
         }

         if(m_json) {
            output() << m_json->print();
         }
         if(m_summary) {
            output() << m_summary->print() << "\n";
         }

         if(verbose() && clock_speed == 0 && m_cycles_consumed > 0 && m_ns_taken > 0) {
            const double seconds = static_cast<double>(m_ns_taken) / 1000000000;
            const double Hz = static_cast<double>(m_cycles_consumed) / seconds;
            const double MHz = Hz / 1000000;
            output() << "\nEstimated clock speed " << MHz << " MHz\n";
         }
      }

   private:
      size_t m_time_unit = 0;
      uint64_t m_cycles_consumed = 0;
      uint64_t m_ns_taken = 0;
      std::unique_ptr<Summary> m_summary;
      std::unique_ptr<JSON_Output> m_json;

      void record_result(const Timer& t) {
         m_ns_taken += t.value();
         m_cycles_consumed += t.cycles_consumed();
         if(m_json) {
            m_json->add(t);
         } else {
            output() << format_timer(t, m_time_unit) << std::endl;

            if(m_summary) {
               m_summary->add(t);
            }
         }
      }
};

BOTAN_REGISTER_COMMAND("speed", Speed);

}  // namespace Botan_CLI

1	/*
2	* (C) 2009,2010,2014,2015,2017,2018,2024 Jack Lloyd
3	* (C) 2015 Simon Warta (Kullo GmbH)
4	*
5	* Botan is released under the Simplified BSD License (see license.txt)
6	*/
7
8	#include "cli.h"
9	#include "perf.h"
10
11	#include <algorithm>
12	#include <chrono>
13	#include <iomanip>
14	#include <map>
15	#include <set>
16	#include <sstream>
17
18	// Always available:
19	#include <botan/version.h>
20	#include <botan/internal/cpuid.h>
21	#include <botan/internal/fmt.h>
22	#include <botan/internal/stl_util.h>
23
24	#if defined(BOTAN_HAS_OS_UTILS)
25	#include <botan/internal/os_utils.h>
26	#endif
27
28	#if defined(BOTAN_HAS_ECC_GROUP)
29	#include <botan/ec_group.h>
30	#endif
31
32	namespace Botan_CLI {
33
34	namespace {
35
36	class JSON_Output final {	2✔
37	public:
38	void add(const Timer& timer) { m_results.push_back(timer); }	2✔
39
40	std::string print() const {	1✔
41	std::ostringstream out;	1✔
42
43	out << "[\n";	1✔
44
45	out << "{"	1✔
46	<< "\"arch\": \"" << BOTAN_TARGET_ARCH << "\", "
47	<< "\"version\": \"" << Botan::short_version_cstr() << "\", "
48	<< "\"git\": \"" << BOTAN_VERSION_VC_REVISION << "\", "
49	<< "\"compiler\": \"" << BOTAN_COMPILER_INVOCATION_STRING << "\""
50	<< "},\n";	1✔
51
52	for(size_t i = 0; i != m_results.size(); ++i) {	3✔
53	const Timer& t = m_results[i];	2✔
54
55	out << "{"	2✔
56	<< "\"algo\": \"" << t.get_name() << "\", "	2✔
57	<< "\"op\": \"" << t.doing() << "\", "	2✔
58	<< "\"events\": " << t.events() << ", ";	2✔
59
60	if(t.cycles_consumed() > 0) {	4✔
61	out << "\"cycles\": " << t.cycles_consumed() << ", ";	4✔
62	}
63
64	if(t.buf_size() > 0) {	2✔
65	out << "\"bps\": " << static_cast<uint64_t>(t.events() / (t.value() / 1000000000.0)) << ", ";	2✔
66	out << "\"buf_size\": " << t.buf_size() << ", ";	2✔
67	}
68
69	out << "\"nanos\": " << t.value() << "}";	2✔
70
71	if(i != m_results.size() - 1) {	2✔
72	out << ",";	1✔
73	}
74
75	out << "\n";	2✔
76	}
77	out << "]\n";	1✔
78
79	return out.str();	2✔
80	}	1✔
81
82	private:
83	std::vector<Timer> m_results;
84	};
85
86	class Summary final {	1✔
87	public:
88	Summary() = default;	1✔
89
90	void add(const Timer& t) {	2✔
91	if(t.buf_size() == 0) {	4✔
92	m_ops_entries.push_back(t);	×
93	} else {
94	m_bps_entries[std::make_pair(t.doing(), t.get_name())].push_back(t);	4✔
95	}
96	}	2✔
97
98	std::string print() {	1✔
99	const size_t name_padding = 35;	1✔
100	const size_t op_name_padding = 16;	1✔
101	const size_t op_padding = 16;	1✔
102
103	std::ostringstream result_ss;	1✔
104	result_ss << std::fixed;	1✔
105
106	if(!m_bps_entries.empty()) {	1✔
107	result_ss << "\n";	1✔
108
109	// add table header
110	result_ss << std::setw(name_padding) << std::left << "algo" << std::setw(op_name_padding) << std::left	1✔
111	<< "operation";	1✔
112
113	for(const Timer& t : m_bps_entries.begin()->second) {	2✔
114	result_ss << std::setw(op_padding) << std::right << (std::to_string(t.buf_size()) + " bytes");	2✔
115	}
116	result_ss << "\n";	1✔
117
118	// add table entries
119	for(const auto& entry : m_bps_entries) {	3✔
120	if(entry.second.empty()) {	2✔
121	continue;	×
122	}
123
124	result_ss << std::setw(name_padding) << std::left << (entry.first.second) << std::setw(op_name_padding)	2✔
125	<< std::left << (entry.first.first);	2✔
126
127	for(const Timer& t : entry.second) {	4✔
128	if(t.events() == 0) {	2✔
129	result_ss << std::setw(op_padding) << std::right << "N/A";	×
130	} else {
131	result_ss << std::setw(op_padding) << std::right << std::setprecision(2)	2✔
132	<< (t.bytes_per_second() / 1000.0);	2✔
133	}
134	}
135
136	result_ss << "\n";	2✔
137	}
138
139	result_ss << "\n[results are the number of 1000s bytes processed per second]\n";	1✔
140	}
141
142	if(!m_ops_entries.empty()) {	1✔
143	result_ss << std::setprecision(6) << "\n";	×
144
145	// sort entries
146	std::sort(m_ops_entries.begin(), m_ops_entries.end());	×
147
148	// add table header
149	result_ss << std::setw(name_padding) << std::left << "algo" << std::setw(op_name_padding) << std::left	×
150	<< "operation" << std::setw(op_padding) << std::right << "sec/op" << std::setw(op_padding)	×
151	<< std::right << "op/sec"	×
152	<< "\n";	×
153
154	// add table entries
155	for(const Timer& entry : m_ops_entries) {	×
156	result_ss << std::setw(name_padding) << std::left << entry.get_name() << std::setw(op_name_padding)	×
157	<< std::left << entry.doing() << std::setw(op_padding) << std::right	×
158	<< entry.seconds_per_event() << std::setw(op_padding) << std::right	×
159	<< entry.events_per_second() << "\n";	×
160	}
161	}
162
163	return result_ss.str();	2✔
164	}	1✔
165
166	private:
167	std::map<std::pair<std::string, std::string>, std::vector<Timer>> m_bps_entries;
168	std::vector<Timer> m_ops_entries;
169	};
170
171	std::vector<size_t> unique_buffer_sizes(const std::string& cmdline_arg) {	28✔
172	const size_t MAX_BUF_SIZE = 64 * 1024 * 1024;	28✔
173
174	std::set<size_t> buf;	28✔
175	for(const std::string& size_str : Command::split_on(cmdline_arg, ',')) {	54✔
176	size_t x = 0;	29✔
177	try {	29✔
178	size_t converted = 0;	29✔
179	x = static_cast<size_t>(std::stoul(size_str, &converted, 0));	29✔
180
181	if(converted != size_str.size()) {	28✔
182	throw CLI_Usage_Error("Invalid integer");	×
183	}
184	} catch(std::exception&) {	1✔
185	throw CLI_Usage_Error("Invalid integer value '" + size_str + "' for option buf-size");	2✔
186	}	1✔
187
188	if(x == 0) {	28✔
189	throw CLI_Usage_Error("Cannot have a zero-sized buffer");	2✔
190	}
191
192	if(x > MAX_BUF_SIZE) {	27✔
193	throw CLI_Usage_Error("Specified buffer size is too large");	2✔
194	}
195
196	buf.insert(x);	26✔
197	}	28✔
198
199	return std::vector<size_t>(buf.begin(), buf.end());	28✔
200	}	25✔
201
202	std::string format_timer(const Timer& t, size_t time_unit) {	483✔
203	constexpr size_t MiB = 1024 * 1024;	483✔
204
205	std::ostringstream oss;	483✔
206
207	oss << t.get_name() << " ";	483✔
208
209	const uint64_t events = t.events();	483✔
210
211	if(t.buf_size() == 0) {	483✔
212	// Report operations/time unit
213
214	if(events == 0) {	460✔
215	oss << "no events ";	×
216	} else {
217	oss << static_cast<uint64_t>(t.events_per_second()) << ' ' << t.doing() << "/sec; ";	920✔
218
219	if(time_unit == 1000) {	460✔
220	oss << std::setprecision(2) << std::fixed << (t.milliseconds() / events) << " ms/op ";	460✔
221	} else if(time_unit == 1000 * 1000) {	×
222	oss << std::setprecision(2) << std::fixed << (t.microseconds() / events) << " us/op ";	×
223	} else if(time_unit == 1000 * 1000 * 1000) {	×
224	oss << std::setprecision(0) << std::fixed << (t.nanoseconds() / events) << " ns/op ";	×
225	}
226
227	if(t.cycles_consumed() != 0 && events > 0) {	920✔
228	const double cycles_per_op = static_cast<double>(t.cycles_consumed()) / events;	460✔
229	const int precision = (cycles_per_op < 10000) ? 2 : 0;	460✔
230	oss << std::fixed << std::setprecision(precision) << cycles_per_op << " cycles/op ";	460✔
231	}
232
233	oss << "(" << events << " " << (events == 1 ? "op" : "ops") << " in " << t.milliseconds() << " ms)";	711✔
234	}
235	} else {
236	// Bulk op - report bytes/time unit
237
238	const double MiB_total = static_cast<double>(events) / MiB;	23✔
239	const double MiB_per_sec = MiB_total / t.seconds();	23✔
240
241	if(!t.doing().empty()) {	23✔
242	oss << t.doing() << " ";	23✔
243	}
244
245	if(t.buf_size() > 0) {	23✔
246	oss << "buffer size " << t.buf_size() << " bytes: ";	23✔
247	}
248
249	if(events == 0) {	23✔
250	oss << "N/A ";	×
251	} else {
252	oss << std::fixed << std::setprecision(3) << MiB_per_sec << " MiB/sec ";	23✔
253	}
254
255	if(t.cycles_consumed() != 0 && events > 0) {	46✔
256	const double cycles_per_byte = static_cast<double>(t.cycles_consumed()) / events;	23✔
257	oss << std::fixed << std::setprecision(2) << cycles_per_byte << " cycles/byte ";	23✔
258	}
259
260	oss << "(" << MiB_total << " MiB in " << t.milliseconds() << " ms)";	23✔
261	}
262
263	return oss.str();	966✔
264	}	483✔
265
266	} // namespace
267
268	class Speed final : public Command {	×
269	public:
270	Speed() :	29✔
271	Command(
272	"speed --msec=500 --format=default --time-unit=ms --ecc-groups= --buf-size=1024 --clear-cpuid= --cpu-clock-speed=0 --cpu-clock-ratio=1.0 *algos") {	58✔
273	}	29✔
274
275	static std::vector<std::string> default_benchmark_list() {	×
276	/*
277	This is not intended to be exhaustive: it just hits the high
278	points of the most interesting or widely used algorithms.
279	*/
280	// clang-format off
281	return {	×
282	/* Block ciphers */
283	"AES-128",
284	"AES-192",
285	"AES-256",
286	"ARIA-128",
287	"ARIA-192",
288	"ARIA-256",
289	"Blowfish",
290	"CAST-128",
291	"Camellia-128",
292	"Camellia-192",
293	"Camellia-256",
294	"DES",
295	"TripleDES",
296	"GOST-28147-89",
297	"IDEA",
298	"Noekeon",
299	"SHACAL2",
300	"SM4",
301	"Serpent",
302	"Threefish-512",
303	"Twofish",
304
305	/* Cipher modes */
306	"AES-128/CBC",
307	"AES-128/CTR-BE",
308	"AES-128/EAX",
309	"AES-128/OCB",
310	"AES-128/GCM",
311	"AES-128/XTS",
312	"AES-128/SIV",
313
314	"Serpent/CBC",
315	"Serpent/CTR-BE",
316	"Serpent/EAX",
317	"Serpent/OCB",
318	"Serpent/GCM",
319	"Serpent/XTS",
320	"Serpent/SIV",
321
322	"ChaCha20Poly1305",
323
324	/* Stream ciphers */
325	"RC4",
326	"Salsa20",
327	"ChaCha20",
328
329	/* Hashes */
330	"SHA-1",
331	"SHA-256",
332	"SHA-512",
333	"SHA-3(256)",
334	"SHA-3(512)",
335	"RIPEMD-160",
336	"Skein-512",
337	"Blake2b",
338	"Whirlpool",
339
340	/* XOFs */
341	"SHAKE-128",
342	"SHAKE-256",
343
344	/* MACs */
345	"CMAC(AES-128)",
346	"HMAC(SHA-256)",
347
348	/* pubkey */
349	"RSA",
350	"DH",
351	"ECDH",
352	"ECDSA",
353	"Ed25519",
354	"Ed448",
355	"X25519",
356	"X448",
357	"ML-KEM",
358	"ML-DSA",
359	"SLH-DSA",
360	"FrodoKEM",
361	"HSS-LMS",
362	};	×
363	// clang-format on
364	}
365
366	std::string group() const override { return "misc"; }	1✔
367
368	std::string description() const override { return "Measures the speed of algorithms"; }	1✔
369
370	void go() override {	28✔
371	std::chrono::milliseconds msec(get_arg_sz("msec"));	28✔
372	std::vector<std::string> ecc_groups = Command::split_on(get_arg("ecc-groups"), ',');	56✔
373	const std::string format = get_arg("format");	28✔
374	const std::string clock_ratio = get_arg("cpu-clock-ratio");	31✔
375
376	const size_t clock_speed = get_arg_sz("cpu-clock-speed");	28✔
377
378	double clock_cycle_ratio = std::strtod(clock_ratio.c_str(), nullptr);	28✔
379
380	m_time_unit = [](std::string_view tu) {	115✔
381	if(tu == "ms") {	28✔
382	return 1000;	28✔
383	} else if(tu == "us") {	×
384	return 1000 * 1000;	×
385	} else if(tu == "ns") {	×
386	return 1000 * 1000 * 1000;	×
387	} else {
388	throw CLI_Usage_Error("Unknown time unit (supported: ms, us, ns)");	×
389	}
390	}(get_arg("time-unit"));	28✔
391
392	/*
393	* This argument is intended to be the ratio between the cycle counter
394	* and the actual machine cycles. It is extremely unlikely that there is
395	* any machine where the cycle counter increments faster than the actual
396	* clock.
397	*/
398	if(clock_cycle_ratio < 0.0 \|\| clock_cycle_ratio > 1.0) {	28✔
399	throw CLI_Usage_Error("Unlikely CPU clock ratio of " + clock_ratio);	×
400	}
401
402	clock_cycle_ratio = 1.0 / clock_cycle_ratio;	28✔
403
404	#if defined(BOTAN_HAS_OS_UTILS)
405	if(clock_speed != 0 && Botan::OS::get_cpu_cycle_counter() != 0) {	28✔
406	error_output() << "The --cpu-clock-speed option is only intended to be used on "	×
407	"platforms without access to a cycle counter.\n"
408	"Expect incorrect results\n\n";	×
409	}
410	#endif
411
412	if(format == "table") {	28✔
413	m_summary = std::make_unique<Summary>();	1✔
414	} else if(format == "json") {	27✔
415	m_json = std::make_unique<JSON_Output>();	1✔
416	} else if(format != "default") {	26✔
417	throw CLI_Usage_Error("Unknown --format type '" + format + "'");	×
418	}
419
420	#if defined(BOTAN_HAS_ECC_GROUP)
421	if(ecc_groups.empty()) {	28✔
422	ecc_groups = {"secp256r1", "secp384r1", "secp521r1", "brainpool256r1", "brainpool384r1", "brainpool512r1"};	224✔
423	} else if(ecc_groups.size() == 1 && ecc_groups[0] == "all") {	×
424	auto all = Botan::EC_Group::known_named_groups();	×
425	ecc_groups.assign(all.begin(), all.end());	×
426	}	×
427	#endif
428
429	std::vector<std::string> algos = get_arg_list("algos");	31✔
430
431	const std::vector<size_t> buf_sizes = unique_buffer_sizes(get_arg("buf-size"));	59✔
432
433	for(const std::string& cpuid_to_clear : Command::split_on(get_arg("clear-cpuid"), ',')) {	26✔
434	auto bits = Botan::CPUID::bit_from_string(cpuid_to_clear);	1✔
435	if(bits.empty()) {	1✔
436	error_output() << "Warning don't know CPUID flag '" << cpuid_to_clear << "'\n";	1✔
437	}
438
439	for(auto bit : bits) {	1✔
440	Botan::CPUID::clear_cpuid_bit(bit);	×
441	}
442	}	26✔
443
444	if(verbose() \|\| m_summary) {	25✔
445	output() << Botan::version_string() << "\n"	2✔
446	<< "CPUID: " << Botan::CPUID::to_string() << "\n\n";	3✔
447	}
448
449	const bool using_defaults = (algos.empty());	25✔
450	if(using_defaults) {	25✔
451	algos = default_benchmark_list();	×
452	}
453
454	PerfConfig perf_config([&](const Timer& t) { this->record_result(t); },	510✔
455	clock_speed,
456	clock_cycle_ratio,
457	msec,
458	ecc_groups,
459	buf_sizes,
460	this->error_output(),
461	this->rng());	25✔
462
463	for(const auto& algo : algos) {	70✔
464	if(auto perf = PerfTest::get(algo)) {	45✔
465	perf->go(perf_config);	45✔
466	} else if(verbose() \|\| !using_defaults) {	×
467	error_output() << "Unknown algorithm '" << algo << "'\n";	×
468	}	45✔
469	}
470
471	if(m_json) {	25✔
472	output() << m_json->print();	2✔
473	}
474	if(m_summary) {	25✔
475	output() << m_summary->print() << "\n";	3✔
476	}
477
478	if(verbose() && clock_speed == 0 && m_cycles_consumed > 0 && m_ns_taken > 0) {	25✔
479	const double seconds = static_cast<double>(m_ns_taken) / 1000000000;	×
480	const double Hz = static_cast<double>(m_cycles_consumed) / seconds;	×
481	const double MHz = Hz / 1000000;	×
482	output() << "\nEstimated clock speed " << MHz << " MHz\n";	×
483	}
484	}	115✔
485
486	private:
487	size_t m_time_unit = 0;
488	uint64_t m_cycles_consumed = 0;
489	uint64_t m_ns_taken = 0;
490	std::unique_ptr<Summary> m_summary;
491	std::unique_ptr<JSON_Output> m_json;
492
493	void record_result(const Timer& t) {	485✔
494	m_ns_taken += t.value();	485✔
495	m_cycles_consumed += t.cycles_consumed();	485✔
496	if(m_json) {	485✔
497	m_json->add(t);	2✔
498	} else {
499	output() << format_timer(t, m_time_unit) << std::endl;	966✔
500
501	if(m_summary) {	483✔
502	m_summary->add(t);	2✔
503	}
504	}
505	}	485✔
506	};
507
508	BOTAN_REGISTER_COMMAND("speed", Speed);	29✔
509
510	} // namespace Botan_CLI

randombit / botan / 13369064240

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous