diff options
author | Jack Lloyd <[email protected]> | 2017-02-04 23:02:08 -0500 |
---|---|---|
committer | Jack Lloyd <[email protected]> | 2017-02-04 23:02:08 -0500 |
commit | 8efc26812b3234acfdee099d6e343722f26bef24 (patch) | |
tree | ece9774291cfffebc256034055bccf618740e11d /src/cli/speed.cpp | |
parent | 90909a2cac9a55d76ab5154ce179eccafd82f1c8 (diff) |
Add toggle to disable SIMD speed test added in 2a7127fea6
Failed to consider need for ISA flags (none required on x86-64 or
aarch64 which are my primary test targets) but handling this correctly
requires splitting up speed.cpp which requires a lot more refactoring
work than I want to take on. Anyway the additional complexity just for
this microbenchmark is not worth it so hide it behind a macro check.
Also add an ECC scalar perf test.
Diffstat (limited to 'src/cli/speed.cpp')
-rw-r--r-- | src/cli/speed.cpp | 141 |
1 files changed, 105 insertions, 36 deletions
diff --git a/src/cli/speed.cpp b/src/cli/speed.cpp index 7404ab12d..772cd6a46 100644 --- a/src/cli/speed.cpp +++ b/src/cli/speed.cpp @@ -20,9 +20,15 @@ #include <botan/mac.h> #include <botan/cipher_mode.h> #include <botan/entropy_src.h> - +#include <botan/cpuid.h> #include <botan/internal/os_utils.h> +//#define INCLUDE_SIMD_PERF + +#if defined(BOTAN_HAS_SIMD_32) && defined(INCLUDE_SIMD_PERF) + #include <botan/internal/simd_32.h> +#endif + #if defined(BOTAN_HAS_AUTO_SEEDING_RNG) #include <botan/auto_rng.h> #endif @@ -103,10 +109,6 @@ #include <botan/chacha.h> #endif -#if defined(BOTAN_HAS_SIMD_32) - #include <botan/internal/simd_32.h> -#endif - namespace Botan_CLI { namespace { @@ -114,29 +116,43 @@ namespace { class Timer { public: - Timer(const std::string& name, uint64_t event_mult = 1) : - m_name(name), m_event_mult(event_mult) {} + Timer(const std::string& name, + uint64_t event_mult = 1, + const std::string& doing = "", + const std::string& provider = "") : + m_name(name + (provider.empty() ? provider : " [" + provider + "]")), + m_doing(doing), + m_event_mult(event_mult) + {} - Timer(const std::string& what, + Timer(const std::string& name, const std::string& provider, const std::string& doing, uint64_t event_mult = 1) : - m_name(what + (provider.empty() ? provider : " [" + provider + "]")), - m_doing(doing), - m_event_mult(event_mult) {} + Timer(name, event_mult, doing, provider) {} + + static uint64_t get_system_timestamp_ns() + { + return Botan::OS::get_system_timestamp_ns(); + } + + static uint64_t get_cpu_cycle_counter() + { + return Botan::OS::get_processor_timestamp(); + } void start() { stop(); - m_timer_start = Botan::OS::get_system_timestamp_ns(); - m_cpu_cycles_start = Botan::OS::get_processor_timestamp(); + m_timer_start = Timer::get_system_timestamp_ns(); + m_cpu_cycles_start = Timer::get_cpu_cycle_counter(); } void stop() { if(m_timer_start) { - const uint64_t now = Botan::OS::get_system_timestamp_ns(); + const uint64_t now = Timer::get_system_timestamp_ns(); if(now > m_timer_start) { @@ -146,7 +162,7 @@ class Timer if(m_cpu_cycles_start != 0) { - uint64_t cycles_taken = Botan::OS::get_processor_timestamp() - m_cpu_cycles_start; + uint64_t cycles_taken = Timer::get_cpu_cycle_counter() - m_cpu_cycles_start; if(cycles_taken > 0) { m_cpu_cycles_used += cycles_taken; @@ -357,8 +373,8 @@ std::vector<std::string> default_benchmark_list() "ECKCDSA", "ECGDSA", "Curve25519", + "NEWHOPE", "McEliece", - "NEWHOPE" }; } @@ -499,6 +515,10 @@ class Speed final : public Command } #endif #if defined(BOTAN_HAS_ECC_GROUP) + else if(algo == "ecc_mult") + { + bench_ecc_mult(msec); + } else if(algo == "os2ecp") { bench_os2ecp(msec); @@ -523,10 +543,17 @@ class Speed final : public Command } #endif } -#if defined(BOTAN_HAS_SIMD_32) +#if defined(BOTAN_HAS_SIMD_32) && defined(INCLUDE_SIMD_PERF) else if(algo == "simd") { - bench_simd32(msec); + if(Botan::CPUID::has_simd_32()) + { + bench_simd32(msec); + } + else + { + output() << "Skipping simd perf test, CPUID indicates SIMD not supported"; + } } #endif else if(algo == "entropy") @@ -605,8 +632,11 @@ class Speed final : public Command const Botan::SymmetricKey key(rng(), cipher.maximum_keylength()); cipher.set_key(key); - const Botan::InitializationVector iv(rng(), 12); - cipher.set_iv(iv.begin(), iv.size()); + if(cipher.valid_iv_length(12)) + { + const Botan::InitializationVector iv(rng(), 12); + cipher.set_iv(iv.begin(), iv.size()); + } while(encrypt_timer.under(runtime)) { @@ -697,7 +727,7 @@ class Speed final : public Command output() << Timer::result_string_bps(timer); } -#if defined(BOTAN_HAS_SIMD_32) +#if defined(BOTAN_HAS_SIMD_32) && defined(INCLUDE_SIMD_PERF) void bench_simd32(const std::chrono::milliseconds msec) { const size_t SIMD_par = 32; @@ -705,15 +735,15 @@ class Speed final : public Command Botan::SIMD_4x32 simd[SIMD_par]; - Timer total_time("", "", "", 0); + Timer total_time(""); - Timer load_le_op("SIMD_4x32", "", "load_le", SIMD_par); - Timer load_be_op("SIMD_4x32", "", "load_be", SIMD_par); - Timer add_op("SIMD_4x32", "", "add", SIMD_par); - Timer sub_op("SIMD_4x32", "", "sub", SIMD_par); - Timer xor_op("SIMD_4x32", "", "xor", SIMD_par); - Timer bswap_op("SIMD_4x32", "", "bswap", SIMD_par); - Timer transpose_op("SIMD_4x32", "", "transpose4", SIMD_par/4); + Timer load_le_op("SIMD_4x32", SIMD_par, "load_le"); + Timer load_be_op("SIMD_4x32", SIMD_par, "load_be"); + Timer add_op("SIMD_4x32", SIMD_par, "add"); + Timer sub_op("SIMD_4x32", SIMD_par, "sub"); + Timer xor_op("SIMD_4x32", SIMD_par, "xor"); + Timer bswap_op("SIMD_4x32", SIMD_par, "bswap"); + Timer transpose_op("SIMD_4x32", SIMD_par/4, "transpose4"); std::chrono::milliseconds msec_part = msec / 5; @@ -724,7 +754,7 @@ class Speed final : public Command { total_time.start(); - load_le_op.run([&simd,rnd] { + load_le_op.run([&] { for(size_t i = 0; i != SIMD_par; ++i) { // Test that unaligned loads work ok @@ -732,42 +762,42 @@ class Speed final : public Command } }); - load_be_op.run([&simd,rnd] { + load_be_op.run([&] { for(size_t i = 0; i != SIMD_par; ++i) { simd[i].load_be(rnd + i); } }); - add_op.run([&simd] { + add_op.run([&] { for(size_t i = 0; i != SIMD_par; ++i) { simd[i] += simd[(i+8) % SIMD_par]; } }); - xor_op.run([&simd] { + xor_op.run([&] { for(size_t i = 0; i != SIMD_par; ++i) { simd[i] ^= simd[(i+8) % SIMD_par]; } }); - transpose_op.run([&simd] { + transpose_op.run([&] { for(size_t i = 0; i != SIMD_par; i += 4) { Botan::SIMD_4x32::transpose(simd[i], simd[i+1], simd[i+2], simd[i+3]); } }); - sub_op.run([&simd] { + sub_op.run([&] { for(size_t i = 0; i != SIMD_par; ++i) { simd[i] -= simd[(i+8) % SIMD_par]; } }); - bswap_op.run([&simd] { + bswap_op.run([&] { for(size_t i = 0; i != SIMD_par; ++i) { simd[i] = simd[i].bswap(); @@ -827,6 +857,40 @@ class Speed final : public Command } #if defined(BOTAN_HAS_ECC_GROUP) + void bench_ecc_mult(const std::chrono::milliseconds runtime) + { + const std::vector<std::string> groups = { + "secp256r1", "brainpool256r1", + "secp384r1", "brainpool384r1", + "secp521r1", "brainpool512r1" + }; + + for(std::string group_name : groups) + { + const Botan::EC_Group group(group_name); + + Timer mult_timer(group_name + " scalar mult"); + Timer blinded_mult_timer(group_name + " blinded scalar mult"); + + const Botan::BigInt scalar(rng(), group.get_curve().get_p().bits()); + const Botan::PointGFp& base_point = group.get_base_point(); + Botan::Blinded_Point_Multiply scalar_mult(base_point, group.get_order(), 4); + + while(blinded_mult_timer.under(runtime)) + { + const Botan::PointGFp r1 = mult_timer.run([&]() { return base_point * scalar; }); + + const Botan::PointGFp r2 = blinded_mult_timer.run( + [&]() { return scalar_mult.blinded_multiply(scalar, rng()); }); + + BOTAN_ASSERT_EQUAL(r1, r2, "Same point computed by both methods"); + } + + output() << Timer::result_string_ops(mult_timer); + output() << Timer::result_string_ops(blinded_mult_timer); + } + } + void bench_os2ecp(const std::chrono::milliseconds runtime) { Timer uncmp_timer("OS2ECP uncompressed"); @@ -1336,6 +1400,11 @@ class Speed final : public Command size_t n = params.first; size_t t = params.second; + if((msec < std::chrono::milliseconds(5000)) && (n >= 3000)) + { + continue; + } + const std::string nm = "McEliece-" + std::to_string(n) + "," + std::to_string(t) + " (WF=" + std::to_string(Botan::mceliece_work_factor(n, t)) + ")"; |