26995937053

Committed 04 Jun 2026 09:38PM UTC coverage: 89.394% (-2.3%) from 91.672%

Build # 26995937053

Build Type

push

github

Committed by

web-flow

Commit Message

Merge pull request #5642 from randombit/jack/prefetch-in-ks

Improve prefetching for table based implementations

Coverage Stats

110588 of 123708 relevant lines covered (89.39%)

11056434.37 hits per line

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

71.0

/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 <iomanip>
#include <map>
#include <set>
#include <sstream>

// Always available:
#include <botan/version.h>
#include <botan/internal/parsing.h>
#include <botan/internal/target_info.h>

#if defined(BOTAN_HAS_CPUID)
   #include <botan/internal/cpuid.h>
#endif

#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 << "{"
             << R"("arch": ")" << BOTAN_TARGET_ARCH << "\", "
             << R"("version": ")" << Botan::short_version_cstr() << "\", ";

         if(auto vc_revision = Botan::version_vc_revision()) {
            out << R"("git": ")" << *vc_revision << "\", ";
         }

         out << R"("compiler": ")" << BOTAN_COMPILER_INVOCATION_STRING << "\""
             << "},\n";

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

            out << "{"
                << R"("algo": ")" << t.get_name() << "\", "
                << R"("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.nanoseconds() / 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, ',')) {
      if(const auto sz = Botan::parse_sz(size_str)) {
         if(*sz == 0) {
            throw CLI_Usage_Error("Cannot have a zero-sized buffer");
         }
         if(*sz > MAX_BUF_SIZE) {
            throw CLI_Usage_Error("Specified buffer size is too large");
         }

         buf.insert(*sz);
      } else {
         throw CLI_Usage_Error("Invalid integer value '" + size_str + "' for option buf-size");
      }
   }

   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();
}

std::vector<std::string> interpret_ecc_groups(const std::string& arg) {
   if(arg.empty()) {
      return {"secp256r1", "secp384r1", "secp521r1", "brainpool256r1", "brainpool384r1", "brainpool512r1"};
   }
   if(arg == "nist") {
      return {"secp224r1", "secp256r1", "secp384r1", "secp521r1"};
   }

#if defined(BOTAN_HAS_ECC_GROUP)
   if(arg == "all") {
      const auto& all = Botan::EC_Group::known_named_groups();
      return std::vector<std::string>(all.begin(), all.end());
   }

   if(arg == "generic") {
      std::vector<std::string> groups;
      for(const auto& group_name : Botan::EC_Group::known_named_groups()) {
         const Botan::EC_Group group(group_name);
         if(group.engine() == Botan::EC_Group_Engine::Generic) {
            groups.push_back(group_name);
         }
      }
      return groups;
   }

   if(arg == "pcurves") {
      std::vector<std::string> groups;
      for(const auto& group_name : Botan::EC_Group::known_named_groups()) {
         const Botan::EC_Group group(group_name);
         if(group.engine() == Botan::EC_Group_Engine::Optimized) {
            groups.push_back(group_name);
         }
      }
      return groups;
   }
#endif

   return Command::split_on(arg, ',');
}

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",
            "Ascon-AEAD128",

            "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)",
            "Ascon-Hash256",
            "RIPEMD-160",
            "Skein-512",
            "Blake2b",
            "Whirlpool",

            /* XOFs */
            "SHAKE-128",
            "SHAKE-256",
            "Ascon-XOF128",

            /* 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 {
         const uint64_t milliseconds = get_arg_sz("msec");
         const std::string ecc_groups_arg = 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 + "'");
         }

         const auto ecc_groups = interpret_ecc_groups(ecc_groups_arg);

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

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

#if defined(BOTAN_HAS_CPUID)
         for(const std::string& cpuid_to_clear : Command::split_on(get_arg("clear-cpuid"), ',')) {
            if(auto bit = Botan::CPUID::bit_from_string(cpuid_to_clear)) {
               Botan::CPUID::clear_cpuid_bit(*bit);
            } else {
               error_output() << "Warning don't know CPUID flag '" << cpuid_to_clear << "'\n";
            }
         }
#endif

         if(verbose() || m_summary) {
#if defined(BOTAN_HAS_CPUID)
            output() << Botan::version_string() << "\n"
                     << "CPUID: " << Botan::CPUID::to_string() << "\n\n";
#else
            output() << Botan::version_string() << "\n\n";
#endif
         }

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

         const PerfConfig perf_config([&](const Timer& t) { this->record_result(t); },
                                      clock_speed,
                                      clock_cycle_ratio,
                                      milliseconds,
                                      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) << "\n" << std::flush;

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

BOTAN_REGISTER_COMMAND("speed", Speed);

}  // namespace

}  // 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 <iomanip>
13	#include <map>
14	#include <set>
15	#include <sstream>
16
17	// Always available:
18	#include <botan/version.h>
19	#include <botan/internal/parsing.h>
20	#include <botan/internal/target_info.h>
21
22	#if defined(BOTAN_HAS_CPUID)
23	#include <botan/internal/cpuid.h>
24	#endif
25
26	#if defined(BOTAN_HAS_OS_UTILS)
27	#include <botan/internal/os_utils.h>
28	#endif
29
30	#if defined(BOTAN_HAS_ECC_GROUP)
31	#include <botan/ec_group.h>
32	#endif
33
34	namespace Botan_CLI {
35
36	namespace {
37
38	class JSON_Output final {	2✔
39	public:
40	void add(const Timer& timer) { m_results.push_back(timer); }	2✔
41
42	std::string print() const {	1✔
43	std::ostringstream out;	1✔
44
45	out << "[\n";	1✔
46
47	out << "{"	1✔
48	<< R"("arch": ")" << BOTAN_TARGET_ARCH << "\", "
49	<< R"("version": ")" << Botan::short_version_cstr() << "\", ";	1✔
50
51	if(auto vc_revision = Botan::version_vc_revision()) {	1✔
52	out << R"("git": ")" << *vc_revision << "\", ";	×
53	}	×
54
55	out << R"("compiler": ")" << BOTAN_COMPILER_INVOCATION_STRING << "\""	1✔
56	<< "},\n";	1✔
57
58	for(size_t i = 0; i != m_results.size(); ++i) {	3✔
59	const Timer& t = m_results[i];	2✔
60
61	out << "{"	2✔
62	<< R"("algo": ")" << t.get_name() << "\", "	2✔
63	<< R"("op": ")" << t.doing() << "\", "	2✔
64	<< "\"events\": " << t.events() << ", ";	2✔
65
66	if(t.cycles_consumed() > 0) {	4✔
67	out << "\"cycles\": " << t.cycles_consumed() << ", ";	4✔
68	}
69
70	if(t.buf_size() > 0) {	2✔
71	out << "\"bps\": " << static_cast<uint64_t>(t.events() / (t.nanoseconds() / 1000000000.0)) << ", ";	2✔
72	out << "\"buf_size\": " << t.buf_size() << ", ";	2✔
73	}
74
75	out << "\"nanos\": " << t.value() << "}";	2✔
76
77	if(i != m_results.size() - 1) {	2✔
78	out << ",";	1✔
79	}
80
81	out << "\n";	2✔
82	}
83	out << "]\n";	1✔
84
85	return out.str();	2✔
86	}	1✔
87
88	private:
89	std::vector<Timer> m_results;
90	};
91
92	class Summary final {	1✔
93	public:
94	Summary() = default;	1✔
95
96	void add(const Timer& t) {	2✔
97	if(t.buf_size() == 0) {	2✔
98	m_ops_entries.push_back(t);	×
99	} else {
100	m_bps_entries[std::make_pair(t.doing(), t.get_name())].push_back(t);	4✔
101	}
102	}	2✔
103
104	std::string print() {	1✔
105	const size_t name_padding = 35;	1✔
106	const size_t op_name_padding = 16;	1✔
107	const size_t op_padding = 16;	1✔
108
109	std::ostringstream result_ss;	1✔
110	result_ss << std::fixed;	1✔
111
112	if(!m_bps_entries.empty()) {	1✔
113	result_ss << "\n";	1✔
114
115	// add table header
116	result_ss << std::setw(name_padding) << std::left << "algo" << std::setw(op_name_padding) << std::left	1✔
117	<< "operation";	1✔
118
119	for(const Timer& t : m_bps_entries.begin()->second) {	2✔
120	result_ss << std::setw(op_padding) << std::right << (std::to_string(t.buf_size()) + " bytes");	2✔
121	}
122	result_ss << "\n";	1✔
123
124	// add table entries
125	for(const auto& entry : m_bps_entries) {	3✔
126	if(entry.second.empty()) {	2✔
127	continue;	×
128	}
129
130	result_ss << std::setw(name_padding) << std::left << (entry.first.second) << std::setw(op_name_padding)	2✔
131	<< std::left << (entry.first.first);	2✔
132
133	for(const Timer& t : entry.second) {	4✔
134	if(t.events() == 0) {	2✔
135	result_ss << std::setw(op_padding) << std::right << "N/A";	×
136	} else {
137	result_ss << std::setw(op_padding) << std::right << std::setprecision(2)	2✔
138	<< (t.bytes_per_second() / 1000.0);	2✔
139	}
140	}
141
142	result_ss << "\n";	2✔
143	}
144
145	result_ss << "\n[results are the number of 1000s bytes processed per second]\n";	1✔
146	}
147
148	if(!m_ops_entries.empty()) {	1✔
149	result_ss << std::setprecision(6) << "\n";	×
150
151	// sort entries
152	std::sort(m_ops_entries.begin(), m_ops_entries.end());	×
153
154	// add table header
155	result_ss << std::setw(name_padding) << std::left << "algo" << std::setw(op_name_padding) << std::left	×
156	<< "operation" << std::setw(op_padding) << std::right << "sec/op" << std::setw(op_padding)	×
157	<< std::right << "op/sec"	×
158	<< "\n";	×
159
160	// add table entries
161	for(const Timer& entry : m_ops_entries) {	×
162	result_ss << std::setw(name_padding) << std::left << entry.get_name() << std::setw(op_name_padding)	×
163	<< std::left << entry.doing() << std::setw(op_padding) << std::right	×
164	<< entry.seconds_per_event() << std::setw(op_padding) << std::right	×
165	<< entry.events_per_second() << "\n";	×
166	}
167	}
168
169	return result_ss.str();	2✔
170	}	1✔
171
172	private:
173	std::map<std::pair<std::string, std::string>, std::vector<Timer>> m_bps_entries;
174	std::vector<Timer> m_ops_entries;
175	};
176
177	std::vector<size_t> unique_buffer_sizes(const std::string& cmdline_arg) {	29✔
178	const size_t MAX_BUF_SIZE = 64 * 1024 * 1024;	29✔
179
180	std::set<size_t> buf;	29✔
181	for(const std::string& size_str : Command::split_on(cmdline_arg, ',')) {	56✔
182	if(const auto sz = Botan::parse_sz(size_str)) {	30✔
183	if(*sz == 0) {	29✔
184	throw CLI_Usage_Error("Cannot have a zero-sized buffer");	2✔
185	}
186	if(*sz > MAX_BUF_SIZE) {	28✔
187	throw CLI_Usage_Error("Specified buffer size is too large");	2✔
188	}
189
190	buf.insert(*sz);	27✔
191	} else {
192	throw CLI_Usage_Error("Invalid integer value '" + size_str + "' for option buf-size");	2✔
193	}
194	}	29✔
195
196	return std::vector<size_t>(buf.begin(), buf.end());	29✔
197	}	26✔
198
199	std::string format_timer(const Timer& t, size_t time_unit) {	415✔
200	constexpr size_t MiB = 1024 * 1024;	415✔
201
202	std::ostringstream oss;	415✔
203
204	oss << t.get_name() << " ";	415✔
205
206	const uint64_t events = t.events();	415✔
207
208	if(t.buf_size() == 0) {	415✔
209	// Report operations/time unit
210
211	if(events == 0) {	392✔
212	oss << "no events ";	×
213	} else {
214	oss << static_cast<uint64_t>(t.events_per_second()) << ' ' << t.doing() << "/sec; ";	784✔
215
216	if(time_unit == 1000) {	392✔
217	oss << std::setprecision(2) << std::fixed << (t.milliseconds() / events) << " ms/op ";	392✔
218	} else if(time_unit == 1000 * 1000) {	×
219	oss << std::setprecision(2) << std::fixed << (t.microseconds() / events) << " us/op ";	×
220	} else if(time_unit == 1000 * 1000 * 1000) {	×
221	oss << std::setprecision(0) << std::fixed << (t.nanoseconds() / events) << " ns/op ";	×
222	}
223
224	if(t.cycles_consumed() != 0 && events > 0) {	784✔
225	const double cycles_per_op = static_cast<double>(t.cycles_consumed()) / events;	392✔
226	const int precision = (cycles_per_op < 10000) ? 2 : 0;	392✔
227	oss << std::fixed << std::setprecision(precision) << cycles_per_op << " cycles/op ";	392✔
228	}
229
230	oss << "(" << events << " " << (events == 1 ? "op" : "ops") << " in " << t.milliseconds() << " ms)";	666✔
231	}
232	} else {
233	// Bulk op - report bytes/time unit
234
235	const double MiB_total = static_cast<double>(events) / MiB;	23✔
236	const double MiB_per_sec = MiB_total / t.seconds();	23✔
237
238	if(!t.doing().empty()) {	23✔
239	oss << t.doing() << " ";	23✔
240	}
241
242	if(t.buf_size() > 0) {	23✔
243	oss << "buffer size " << t.buf_size() << " bytes: ";	23✔
244	}
245
246	if(events == 0) {	23✔
247	oss << "N/A ";	×
248	} else {
249	oss << std::fixed << std::setprecision(3) << MiB_per_sec << " MiB/sec ";	23✔
250	}
251
252	if(t.cycles_consumed() != 0 && events > 0) {	46✔
253	const double cycles_per_byte = static_cast<double>(t.cycles_consumed()) / events;	23✔
254	oss << std::fixed << std::setprecision(2) << cycles_per_byte << " cycles/byte ";	23✔
255	}
256
257	oss << "(" << MiB_total << " MiB in " << t.milliseconds() << " ms)";	23✔
258	}
259
260	return oss.str();	830✔
261	}	415✔
262
263	std::vector<std::string> interpret_ecc_groups(const std::string& arg) {	29✔
264	if(arg.empty()) {	29✔
265	return {"secp256r1", "secp384r1", "secp521r1", "brainpool256r1", "brainpool384r1", "brainpool512r1"};	29✔
266	}
267	if(arg == "nist") {	×
268	return {"secp224r1", "secp256r1", "secp384r1", "secp521r1"};	×
269	}
270
271	#if defined(BOTAN_HAS_ECC_GROUP)
272	if(arg == "all") {	×
273	const auto& all = Botan::EC_Group::known_named_groups();	×
274	return std::vector<std::string>(all.begin(), all.end());	×
275	}
276
277	if(arg == "generic") {	×
278	std::vector<std::string> groups;	×
279	for(const auto& group_name : Botan::EC_Group::known_named_groups()) {	×
280	const Botan::EC_Group group(group_name);	×
281	if(group.engine() == Botan::EC_Group_Engine::Generic) {	×
282	groups.push_back(group_name);	×
283	}
284	}	×
285	return groups;	×
286	}	×
287
288	if(arg == "pcurves") {	×
289	std::vector<std::string> groups;	×
290	for(const auto& group_name : Botan::EC_Group::known_named_groups()) {	×
291	const Botan::EC_Group group(group_name);	×
292	if(group.engine() == Botan::EC_Group_Engine::Optimized) {	×
293	groups.push_back(group_name);	×
294	}
295	}	×
296	return groups;	×
297	}	×
298	#endif
299
300	return Command::split_on(arg, ',');	×
301	}
302
303	class Speed final : public Command {	×
304	public:
305	Speed() :	30✔
306	Command(
307	"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") {	60✔
308	}	30✔
309
310	static std::vector<std::string> default_benchmark_list() {	×
311	/*
312	This is not intended to be exhaustive: it just hits the high
313	points of the most interesting or widely used algorithms.
314	*/
315	// clang-format off
316	return {	×
317	/* Block ciphers */
318	"AES-128",
319	"AES-192",
320	"AES-256",
321	"ARIA-128",
322	"ARIA-192",
323	"ARIA-256",
324	"Blowfish",
325	"CAST-128",
326	"Camellia-128",
327	"Camellia-192",
328	"Camellia-256",
329	"DES",
330	"TripleDES",
331	"GOST-28147-89",
332	"IDEA",
333	"Noekeon",
334	"SHACAL2",
335	"SM4",
336	"Serpent",
337	"Threefish-512",
338	"Twofish",
339
340	/* Cipher modes */
341	"AES-128/CBC",
342	"AES-128/CTR-BE",
343	"AES-128/EAX",
344	"AES-128/OCB",
345	"AES-128/GCM",
346	"AES-128/XTS",
347	"AES-128/SIV",
348	"Ascon-AEAD128",
349
350	"Serpent/CBC",
351	"Serpent/CTR-BE",
352	"Serpent/EAX",
353	"Serpent/OCB",
354	"Serpent/GCM",
355	"Serpent/XTS",
356	"Serpent/SIV",
357
358	"ChaCha20Poly1305",
359
360	/* Stream ciphers */
361	"RC4",
362	"Salsa20",
363	"ChaCha20",
364
365	/* Hashes */
366	"SHA-1",
367	"SHA-256",
368	"SHA-512",
369	"SHA-3(256)",
370	"SHA-3(512)",
371	"Ascon-Hash256",
372	"RIPEMD-160",
373	"Skein-512",
374	"Blake2b",
375	"Whirlpool",
376
377	/* XOFs */
378	"SHAKE-128",
379	"SHAKE-256",
380	"Ascon-XOF128",
381
382	/* MACs */
383	"CMAC(AES-128)",
384	"HMAC(SHA-256)",
385
386	/* pubkey */
387	"RSA",
388	"DH",
389	"ECDH",
390	"ECDSA",
391	"Ed25519",
392	"Ed448",
393	"X25519",
394	"X448",
395	"ML-KEM",
396	"ML-DSA",
397	"SLH-DSA",
398	"FrodoKEM",
399	"HSS-LMS",
400	};	×
401	// clang-format on
402	}
403
404	std::string group() const override { return "misc"; }	1✔
405
406	std::string description() const override { return "Measures the speed of algorithms"; }	1✔
407
408	void go() override {	29✔
409	const uint64_t milliseconds = get_arg_sz("msec");	29✔
410	const std::string ecc_groups_arg = get_arg("ecc-groups");	29✔
411	const std::string format = get_arg("format");	32✔
412	const std::string clock_ratio = get_arg("cpu-clock-ratio");	32✔
413
414	const size_t clock_speed = get_arg_sz("cpu-clock-speed");	29✔
415
416	double clock_cycle_ratio = std::strtod(clock_ratio.c_str(), nullptr);	29✔
417
418	m_time_unit = [](std::string_view tu) {	119✔
419	if(tu == "ms") {	29✔
420	return 1000;	29✔
421	} else if(tu == "us") {	×
422	return 1000 * 1000;	×
423	} else if(tu == "ns") {	×
424	return 1000 * 1000 * 1000;	×
425	} else {
426	throw CLI_Usage_Error("Unknown time unit (supported: ms, us, ns)");	×
427	}
428	}(get_arg("time-unit"));	29✔
429
430	/*
431	* This argument is intended to be the ratio between the cycle counter
432	* and the actual machine cycles. It is extremely unlikely that there is
433	* any machine where the cycle counter increments faster than the actual
434	* clock.
435	*/
436	if(clock_cycle_ratio < 0.0 \|\| clock_cycle_ratio > 1.0) {	29✔
437	throw CLI_Usage_Error("Unlikely CPU clock ratio of " + clock_ratio);	×
438	}
439
440	clock_cycle_ratio = 1.0 / clock_cycle_ratio;	29✔
441
442	#if defined(BOTAN_HAS_OS_UTILS)
443	if(clock_speed != 0 && Botan::OS::get_cpu_cycle_counter() != 0) {	29✔
444	error_output() << "The --cpu-clock-speed option is only intended to be used on "	×
445	"platforms without access to a cycle counter.\n"
446	"Expect incorrect results\n\n";	×
447	}
448	#endif
449
450	if(format == "table") {	29✔
451	m_summary = std::make_unique<Summary>();	1✔
452	} else if(format == "json") {	28✔
453	m_json = std::make_unique<JSON_Output>();	4✔
454	} else if(format != "default") {	27✔
455	throw CLI_Usage_Error("Unknown --format type '" + format + "'");	×
456	}
457
458	const auto ecc_groups = interpret_ecc_groups(ecc_groups_arg);	29✔
459
460	std::vector<std::string> algos = get_arg_list("algos");	29✔
461
462	const std::vector<size_t> buf_sizes = unique_buffer_sizes(get_arg("buf-size"));	61✔
463
464	#if defined(BOTAN_HAS_CPUID)
465	for(const std::string& cpuid_to_clear : Command::split_on(get_arg("clear-cpuid"), ',')) {	27✔
466	if(auto bit = Botan::CPUID::bit_from_string(cpuid_to_clear)) {	1✔
467	Botan::CPUID::clear_cpuid_bit(*bit);	×
468	} else {
469	error_output() << "Warning don't know CPUID flag '" << cpuid_to_clear << "'\n";	1✔
470	}
471	}	26✔
472	#endif
473
474	if(verbose() \|\| m_summary) {	26✔
475	#if defined(BOTAN_HAS_CPUID)
476	output() << Botan::version_string() << "\n"	2✔
477	<< "CPUID: " << Botan::CPUID::to_string() << "\n\n";	3✔
478	#else
479	output() << Botan::version_string() << "\n\n";
480	#endif
481	}
482
483	const bool using_defaults = (algos.empty());	26✔
484	if(using_defaults) {	26✔
485	algos = default_benchmark_list();	×
486	}
487
488	const PerfConfig perf_config([&](const Timer& t) { this->record_result(t); },	443✔
489	clock_speed,
490	clock_cycle_ratio,
491	milliseconds,
492	ecc_groups,
493	buf_sizes,
494	this->error_output(),
495	this->rng());	26✔
496
497	for(const auto& algo : algos) {	68✔
498	if(auto perf = PerfTest::get(algo)) {	42✔
499	perf->go(perf_config);	42✔
500	} else if(verbose() \|\| !using_defaults) {	×
501	error_output() << "Unknown algorithm '" << algo << "'\n";	×
502	}	42✔
503	}
504
505	if(m_json) {	26✔
506	output() << m_json->print();	2✔
507	}
508	if(m_summary) {	26✔
509	output() << m_summary->print() << "\n";	3✔
510	}
511
512	if(verbose() && clock_speed == 0 && m_cycles_consumed > 0 && m_ns_taken > 0) {	26✔
513	const double seconds = static_cast<double>(m_ns_taken) / 1000000000;	×
514	const double Hz = static_cast<double>(m_cycles_consumed) / seconds;	×
515	const double MHz = Hz / 1000000;	×
516	output() << "\nEstimated clock speed " << MHz << " MHz\n";	×
517	}
518	}	64✔
519
520	private:
521	size_t m_time_unit = 0;
522	uint64_t m_cycles_consumed = 0;
523	uint64_t m_ns_taken = 0;
524	std::unique_ptr<Summary> m_summary;
525	std::unique_ptr<JSON_Output> m_json;
526
527	void record_result(const Timer& t) {	417✔
528	m_ns_taken += t.value();	417✔
529	m_cycles_consumed += t.cycles_consumed();	417✔
530	if(m_json) {	417✔
531	m_json->add(t);	2✔
532	} else {
533	output() << format_timer(t, m_time_unit) << "\n" << std::flush;	830✔
534
535	if(m_summary) {	415✔
536	m_summary->add(t);	2✔
537	}
538	}
539	}	417✔
540	};
541
542	BOTAN_REGISTER_COMMAND("speed", Speed);	30✔
543
544	} // namespace
545
546	} // namespace Botan_CLI

randombit / botan / 26995937053

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