Skip to content

Commit

Permalink
perf boost mode
Browse files Browse the repository at this point in the history
  • Loading branch information
@eggrobin
eggrobin committed Dec 28, 2024
1 parent f11e934 commit b1b7221
Show file tree
Hide file tree
Showing 2 changed files with 166 additions and 15 deletions.
166 changes: 151 additions & 15 deletions nanobenchmarks/main.cpp
View file
Original file line number Diff line number Diff line change
@@ -1,3 +1,7 @@

#include <Windows.h>
#include <Powersetting.h>

#include <algorithm>
#include <iostream>
#include <iomanip>
Expand All @@ -17,10 +21,11 @@

#define BENCHMARK_CALLING_CONVENTION

static std::map<std::string, double(BENCHMARK_CALLING_CONVENTION*)(double)>&
using BenchmarkedFunction = double(BENCHMARK_CALLING_CONVENTION*)(double);

Check warning on line 24 in nanobenchmarks/main.cpp

View workflow job for this annotation

GitHub Actions / check-cpp

readability/casting 

Using C-style cast.  Use reinterpret_cast<BENCHMARK_CALLING_CONVENTION*>(...) instead

static std::map<std::string, BenchmarkedFunction>&
function_registry =
*new std::map<std::string,
double(BENCHMARK_CALLING_CONVENTION*)(double)>();
*new std::map<std::string, BenchmarkedFunction>();

#define BENCHMARK_FUNCTION(f) \
static bool registered_##f = function_registry.emplace(#f, &(f)).second
Expand All @@ -35,7 +40,7 @@ static std::map<std::string, double(BENCHMARK_CALLING_CONVENTION*)(double)>&
static bool registered_##line = \
function_registry.emplace(name, &(__VA_ARGS__)).second

#define BENCHMARKED_FUNCTION(f) \
#define BenchmarkedFunction(f) \
double BENCHMARK_CALLING_CONVENTION f(double x); \
BENCHMARK_FUNCTION(f); \
double BENCHMARK_CALLING_CONVENTION f(double x)
Expand All @@ -50,41 +55,41 @@ BENCHMARK_EXTERN_C_FUNCTION(sqrtsd_xmm0_xmm0);
BENCHMARK_EXTERN_C_FUNCTION(mulsd_xmm0_xmm0);
BENCHMARK_EXTERN_C_FUNCTION(mulsd_xmm0_xmm0_4x);

BENCHMARKED_FUNCTION(twice) {
BenchmarkedFunction(twice) {
return 2 * x;
}

BENCHMARKED_FUNCTION(thrice) {
BenchmarkedFunction(thrice) {
return 3 * x;
}

BENCHMARKED_FUNCTION(inc) {
BenchmarkedFunction(inc) {
return x + 1;
}

BENCHMARKED_FUNCTION(add_4_times) {
BenchmarkedFunction(add_4_times) {
return x * x * x * x * x;
}

BENCHMARKED_FUNCTION(add_16_times) {
BenchmarkedFunction(add_16_times) {

Check warning on line 75 in nanobenchmarks/main.cpp

View workflow job for this annotation

GitHub Actions / check-cpp

whitespace/blank_line 

Redundant blank line at the start of a code block should be deleted.
return x + x + x + x +
x + x + x + x +
x + x + x + x +
x + x + x + x + x;}

BENCHMARKED_FUNCTION(square_root) {
BenchmarkedFunction(square_root) {
__m128d x_0 = _mm_set_sd(x);
return _mm_cvtsd_f64(_mm_sqrt_sd(x_0, x_0));
}

BENCHMARKED_FUNCTION(sqrt_sqrt) {
BenchmarkedFunction(sqrt_sqrt) {
__m128d x_0 = _mm_set_sd(x);
x_0 = _mm_sqrt_sd(x_0, x_0);
return _mm_cvtsd_f64(_mm_sqrt_sd(x_0, x_0));
}

BENCHMARKED_FUNCTION(square_root_division) {
BenchmarkedFunction(square_root_division) {
__m128d x_0 = _mm_set_sd(x);
return _mm_cvtsd_f64(_mm_div_sd(x_0, _mm_sqrt_sd(x_0, x_0)));
}
Expand Down Expand Up @@ -134,7 +139,7 @@ distribution operator+(distribution x, double b) {
return result;
}

__declspec(noinline) distribution benchmark(double (BENCHMARK_CALLING_CONVENTION* f)(double)) {
__declspec(noinline) distribution benchmark(BenchmarkedFunction f) {
constexpr int k = 1'000'000;
static double* durations = new double[k];
int registers[4]{};
Expand Down Expand Up @@ -180,7 +185,124 @@ BENCHMARK_FUNCTION_WITH_NAME(
BENCHMARK_FUNCTION_WITH_NAME("Cbrt",
principia::numerics::_cbrt::Cbrt);

std::pair<DWORD, DWORD> ReadPerfBoostModeACDC(
GUID const* const active_power_scheme) {
#define CASE(value) \
case value: \
return #value
auto perf_boost_mode_to_string = [](DWORD mode) -> std::string {
switch (mode) {
CASE(PROCESSOR_PERF_BOOST_MODE_DISABLED);
CASE(PROCESSOR_PERF_BOOST_MODE_ENABLED);
CASE(PROCESSOR_PERF_BOOST_MODE_AGGRESSIVE);
CASE(PROCESSOR_PERF_BOOST_MODE_EFFICIENT_ENABLED);
CASE(PROCESSOR_PERF_BOOST_MODE_EFFICIENT_AGGRESSIVE);
CASE(PROCESSOR_PERF_BOOST_MODE_AGGRESSIVE_AT_GUARANTEED);
CASE(PROCESSOR_PERF_BOOST_MODE_EFFICIENT_AGGRESSIVE_AT_GUARANTEED);
default:
return std::to_string(mode);
}
};
#undef CASE
DWORD perf_boost_mode_ac;
DWORD perf_boost_mode_dc;
DWORD perf_boost_mode_size = sizeof(perf_boost_mode_ac);
CHECK_EQ(PowerReadACValue(nullptr,
active_power_scheme,
&GUID_PROCESSOR_SETTINGS_SUBGROUP,
&GUID_PROCESSOR_PERF_BOOST_MODE,
nullptr,
reinterpret_cast<LPBYTE>(&perf_boost_mode_ac),
&perf_boost_mode_size),
ERROR_SUCCESS)
<< perf_boost_mode_size;
CHECK_EQ(PowerReadDCValue(nullptr,
active_power_scheme,
&GUID_PROCESSOR_SETTINGS_SUBGROUP,
&GUID_PROCESSOR_PERF_BOOST_MODE,
nullptr,
reinterpret_cast<LPBYTE>(&perf_boost_mode_dc),
&perf_boost_mode_size),
ERROR_SUCCESS)
<< perf_boost_mode_size;
std::println(std::cout,
"PERF_BOOST_MODE AC={} ({})",
perf_boost_mode_ac,
perf_boost_mode_to_string(perf_boost_mode_ac));
std::println(std::cout,
"PERF_BOOST_MODE DC={} ({})",
perf_boost_mode_ac,
perf_boost_mode_to_string(perf_boost_mode_dc));
return {
perf_boost_mode_ac, perf_boost_mode_dc
};
}

int __cdecl main(int argc, const char** argv) {
SYSTEM_POWER_STATUS power_status;
CHECK(GetSystemPowerStatus(&power_status));
std::println(std::cout,
"ACLineStatus={} ({})",
power_status.ACLineStatus,
power_status.ACLineStatus == 0 ? "Offline"
: power_status.ACLineStatus == 1 ? "Online"
: "Unknown");
GUID* active_power_scheme;
CHECK_EQ(PowerGetActiveScheme(nullptr, &active_power_scheme), ERROR_SUCCESS);

// powercfg /query scheme_current sub_processor PERFBOOSTMODE

auto const [perf_boost_mode_ac, perf_boost_mode_dc] =
ReadPerfBoostModeACDC(active_power_scheme);
std::println(std::cout, "Disabling perf boost mode…");
std::println(
std::cout,
R"(If interrupted, restore with
POWERCFG /SETACVALUEINDEX SCHEME_CURRENT SUB_PROCESSOR PERFBOOSTMODE {}
POWERCFG /SETDCVALUEINDEX SCHEME_CURRENT SUB_PROCESSOR PERFBOOSTMODE {})",
perf_boost_mode_ac,
perf_boost_mode_dc);
CHECK_EQ(PowerWriteACValueIndex(nullptr,
active_power_scheme,
&GUID_PROCESSOR_SETTINGS_SUBGROUP,
&GUID_PROCESSOR_PERF_BOOST_MODE,
PROCESSOR_PERF_BOOST_MODE_DISABLED),
ERROR_SUCCESS);
CHECK_EQ(PowerWriteDCValueIndex(nullptr,
active_power_scheme,
&GUID_PROCESSOR_SETTINGS_SUBGROUP,
&GUID_PROCESSOR_PERF_BOOST_MODE,
PROCESSOR_PERF_BOOST_MODE_DISABLED),
ERROR_SUCCESS);
auto const [updated_perf_boost_mode_ac, updated_perf_boost_mode_dc] =
ReadPerfBoostModeACDC(active_power_scheme);
CHECK_EQ(updated_perf_boost_mode_ac, PROCESSOR_PERF_BOOST_MODE_DISABLED);
CHECK_EQ(updated_perf_boost_mode_dc, PROCESSOR_PERF_BOOST_MODE_DISABLED);

{
DWORD perf_boost_mode_ac;
DWORD perf_boost_mode_dc;
DWORD perf_boost_mode_size = sizeof(perf_boost_mode_ac);
CHECK_EQ(PowerReadACValue(nullptr,
active_power_scheme,
&GUID_PROCESSOR_SETTINGS_SUBGROUP,
&GUID_PROCESSOR_PERF_BOOST_MODE,
nullptr,
reinterpret_cast<LPBYTE>(&perf_boost_mode_ac),
&perf_boost_mode_size),
ERROR_SUCCESS)
<< perf_boost_mode_size;
CHECK_EQ(PowerReadDCValue(nullptr,
active_power_scheme,
&GUID_PROCESSOR_SETTINGS_SUBGROUP,
&GUID_PROCESSOR_PERF_BOOST_MODE,
nullptr,
reinterpret_cast<LPBYTE>(&perf_boost_mode_dc),
&perf_boost_mode_size),
ERROR_SUCCESS)
<< perf_boost_mode_size;
}

distribution::quantile_definitions = {1000, 100, 10, 4, 2};
std::cout << principia::base::_cpuid::CPUVendorIdentificationString() << " "
<< principia::base::_cpuid::ProcessorBrandString() << "\nFeatures:"
Expand All @@ -194,7 +316,7 @@ int __cdecl main(int argc, const char** argv) {
std::pair{&mulsd_xmm0_xmm0_4x, 4 * 4},
std::pair{&sqrtps_xmm0_xmm0, 12},
};
std::map<double(BENCHMARK_CALLING_CONVENTION*)(double), distribution>
std::map<BenchmarkedFunction, distribution>
reference_measurements;
std::cout << std::setw(name_width + 2) << "RAW TSC:" << distribution::heading
<< "\n";
Expand Down Expand Up @@ -229,7 +351,7 @@ int __cdecl main(int argc, const char** argv) {
<< "\n";
std::cout << std::setw(name_width + 2) << "Cycles:" << distribution::heading
<< "\n";
auto bm_cycles = [&](double(BENCHMARK_CALLING_CONVENTION * f)(double)) {
auto bm_cycles = [&](BenchmarkedFunction f) {
return a * benchmark(f) + b;
};
for (auto const& [name, f] : function_registry) {
Expand All @@ -242,4 +364,18 @@ int __cdecl main(int argc, const char** argv) {
std::make_format_args(name));
std::cout << a * benchmark(f) + b << "\n";
}
std::println(std::cout, "Restoring perf boost mode…");
CHECK_EQ(PowerWriteACValueIndex(nullptr,
active_power_scheme,
&GUID_PROCESSOR_SETTINGS_SUBGROUP,
&GUID_PROCESSOR_PERF_BOOST_MODE,
perf_boost_mode_ac),
ERROR_SUCCESS);
CHECK_EQ(PowerWriteDCValueIndex(nullptr,
active_power_scheme,
&GUID_PROCESSOR_SETTINGS_SUBGROUP,
&GUID_PROCESSOR_PERF_BOOST_MODE,
perf_boost_mode_dc),
ERROR_SUCCESS);
ReadPerfBoostModeACDC(active_power_scheme);
}

Check warning on line 381 in nanobenchmarks/main.cpp

View workflow job for this annotation

GitHub Actions / check-cpp

whitespace/ending_newline 

Could not find a newline character at the end of the file.
15 changes: 15 additions & 0 deletions nanobenchmarks/nanobenchmarks.vcxproj
View file
Original file line number Diff line number Diff line change
Expand Up @@ -9,6 +9,21 @@
<ImportGroup Label="ExtensionSettings">
<Import Project="$(VCTargetsPath)\BuildCustomizations\masm.props" />
</ImportGroup>
<ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
<Link>
<AdditionalDependencies>gmock.lib;gmock_main.lib;powrprof.lib;%(AdditionalDependencies)</AdditionalDependencies>
</Link>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'">
<Link>
<AdditionalDependencies>gmock.lib;gmock_main.lib;powrprof.lib;%(AdditionalDependencies)</AdditionalDependencies>
</Link>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release_LLVM|x64'">
<Link>
<AdditionalDependencies>gmock.lib;gmock_main.lib;powrprof.lib;%(AdditionalDependencies)</AdditionalDependencies>
</Link>
</ItemDefinitionGroup>
<ItemGroup>
<ClCompile Include="main.cpp" />
</ItemGroup>
Expand Down

0 comments on commit b1b7221

Please sign in to comment.