chore: import upstream snapshot with attribution

test/AssemblyTests.cmake

@@ -0,0 +1,67 @@
+set(CLANG_SUPPORTED_VERSION "5.0.0")
+set(GCC_SUPPORTED_VERSION "5.5.0")
+
+if (CMAKE_CXX_COMPILER_ID MATCHES "Clang")
+  if (NOT CMAKE_CXX_COMPILER_VERSION VERSION_EQUAL ${CLANG_SUPPORTED_VERSION})
+    message (WARNING
+      "Unsupported Clang version " ${CMAKE_CXX_COMPILER_VERSION}
+      ". Expected is " ${CLANG_SUPPORTED_VERSION}
+      ". Assembly tests may be broken.")
+  endif()
+elseif(CMAKE_CXX_COMPILER_ID MATCHES "GNU")
+  if (NOT CMAKE_CXX_COMPILER_VERSION VERSION_EQUAL ${GCC_SUPPORTED_VERSION})
+    message (WARNING
+      "Unsupported GCC version " ${CMAKE_CXX_COMPILER_VERSION}
+      ". Expected is " ${GCC_SUPPORTED_VERSION}
+      ". Assembly tests may be broken.")
+  endif()
+else()
+  message (WARNING "Unsupported compiler. Assembly tests may be broken.")
+endif()
+
+include(split_list)
+
+set(ASM_TEST_FLAGS "")
+check_cxx_compiler_flag(-O3 BENCHMARK_HAS_O3_FLAG)
+if (BENCHMARK_HAS_O3_FLAG)
+  list(APPEND ASM_TEST_FLAGS -O3)
+endif()
+
+check_cxx_compiler_flag(-g0 BENCHMARK_HAS_G0_FLAG)
+if (BENCHMARK_HAS_G0_FLAG)
+  list(APPEND ASM_TEST_FLAGS -g0)
+endif()
+
+check_cxx_compiler_flag(-fno-stack-protector BENCHMARK_HAS_FNO_STACK_PROTECTOR_FLAG)
+if (BENCHMARK_HAS_FNO_STACK_PROTECTOR_FLAG)
+  list(APPEND ASM_TEST_FLAGS -fno-stack-protector)
+endif()
+
+split_list(ASM_TEST_FLAGS)
+string(TOUPPER "${CMAKE_CXX_COMPILER_ID}" ASM_TEST_COMPILER)
+
+macro(add_filecheck_test name)
+  cmake_parse_arguments(ARG "" "" "CHECK_PREFIXES" ${ARGV})
+  add_library(${name} OBJECT ${name}.cc)
+  target_link_libraries(${name} PRIVATE benchmark::benchmark)
+  set_target_properties(${name} PROPERTIES COMPILE_FLAGS "-S ${ASM_TEST_FLAGS}")
+  set(ASM_OUTPUT_FILE "${CMAKE_CURRENT_BINARY_DIR}/${name}.s")
+  add_custom_target(copy_${name} ALL
+      COMMAND ${PROJECT_SOURCE_DIR}/tools/strip_asm.py
+        $<TARGET_OBJECTS:${name}>
+        ${ASM_OUTPUT_FILE}
+      BYPRODUCTS ${ASM_OUTPUT_FILE})
+  add_dependencies(copy_${name} ${name})
+  if (NOT ARG_CHECK_PREFIXES)
+    set(ARG_CHECK_PREFIXES "CHECK")
+  endif()
+  foreach(prefix ${ARG_CHECK_PREFIXES})
+    add_test(NAME run_${name}_${prefix}
+        COMMAND
+          ${LLVM_FILECHECK_EXE} ${name}.cc
+          --input-file=${ASM_OUTPUT_FILE}
+          --check-prefixes=CHECK,CHECK-${ASM_TEST_COMPILER}
+        WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR})
+  endforeach()
+endmacro()
+

test/BUILD

@@ -0,0 +1,140 @@
+load("@rules_cc//cc:defs.bzl", "cc_library", "cc_test")
+
+platform(
+    name = "windows",
+    constraint_values = [
+        "@platforms//os:windows",
+    ],
+)
+
+TEST_COPTS = [
+    "-pedantic",
+    "-pedantic-errors",
+    "-std=c++17",
+    "-Wall",
+    "-Wconversion",
+    "-Wextra",
+    "-Wshadow",
+    #    "-Wshorten-64-to-32",
+    "-Wfloat-equal",
+    "-fstrict-aliasing",
+    ## assert() are used a lot in tests upstream, which may be optimised out leading to
+    ## unused-variable warning.
+    "-Wno-unused-variable",
+    "-Werror=old-style-cast",
+]
+
+TEST_MSVC_OPTS = [
+    "/std:c++17",
+]
+
+# Some of the issues with DoNotOptimize only occur when optimization is enabled
+PER_SRC_COPTS = {
+    "donotoptimize_test.cc": ["-O3"],
+}
+
+TEST_ARGS = ["--benchmark_min_time=0.01s"]
+
+PER_SRC_TEST_ARGS = {
+    "user_counters_tabular_test.cc": [
+        "--benchmark_counters_tabular=true",
+        "--benchmark_min_time=0.2s",
+    ],
+    "repetitions_test.cc": [" --benchmark_repetitions=3"],
+    "spec_arg_test.cc": ["--benchmark_filter=BM_NotChosen"],
+    "spec_arg_verbosity_test.cc": ["--v=42"],
+    "complexity_test.cc": ["--benchmark_min_time=1000000x"],
+    "user_counters_test.cc": ["--benchmark_min_time=0.2s"],
+    "user_counters_threads_test.cc": ["--benchmark_min_time=0.2s"],
+}
+
+cc_library(
+    name = "output_test_helper",
+    testonly = 1,
+    srcs = ["output_test_helper.cc"],
+    hdrs = ["output_test.h"],
+    copts = select({
+        "//:windows": TEST_MSVC_OPTS,
+        "//conditions:default": TEST_COPTS,
+    }),
+    deps = [
+        "//:benchmark",
+        "//:benchmark_internal_headers",
+    ],
+)
+
+# Tests that use gtest.  These rely on `gtest_main`.
+[
+    cc_test(
+        name = test_src[:-len(".cc")],
+        size = "small",
+        srcs = [test_src],
+        copts = select({
+            "//:windows": TEST_MSVC_OPTS,
+            "//conditions:default": TEST_COPTS,
+        }) + PER_SRC_COPTS.get(test_src, []),
+        deps = [
+            "//:benchmark",
+            "//:benchmark_internal_headers",
+            "@com_google_googletest//:gtest",
+            "@com_google_googletest//:gtest_main",
+        ],
+    )
+    for test_src in glob(["*_gtest.cc"])
+]
+
+# Tests that do not use gtest.  These have their own `main` defined.
+[
+    cc_test(
+        name = test_src[:-len(".cc")],
+        size = "small",
+        srcs = [test_src],
+        args = TEST_ARGS + PER_SRC_TEST_ARGS.get(test_src, []),
+        copts = select({
+            "//:windows": TEST_MSVC_OPTS,
+            "//conditions:default": TEST_COPTS,
+        }) + PER_SRC_COPTS.get(test_src, []),
+        deps = [
+            ":output_test_helper",
+            "//:benchmark",
+            "//:benchmark_internal_headers",
+        ],
+        # FIXME: Add support for assembly tests to bazel.
+        # See Issue #556
+        # https://github.com/google/benchmark/issues/556
+    )
+    for test_src in glob(
+        ["*_test.cc"],
+        exclude = [
+            "*_assembly_test.cc",
+            "cxx11_test.cc",
+            "link_main_test.cc",
+        ],
+    )
+]
+
+cc_test(
+    name = "cxx11_test",
+    size = "small",
+    srcs = ["cxx11_test.cc"],
+    copts = TEST_COPTS + ["-std=c++11"],
+    target_compatible_with = select({
+        "//:windows": ["@platforms//:incompatible"],
+        "//conditions:default": [],
+    }),
+    deps = [
+        ":output_test_helper",
+        "//:benchmark_main",
+    ],
+)
+
+cc_test(
+    name = "link_main_test",
+    size = "small",
+    srcs = ["link_main_test.cc"],
+    copts = select({
+        "//:windows": TEST_MSVC_OPTS,
+        "//conditions:default": TEST_COPTS,
+    }),
+    deps = ["//:benchmark_main"],
+)

test/CMakeLists.txt

@@ -0,0 +1,341 @@
+#Enable the tests
+
+set(THREADS_PREFER_PTHREAD_FLAG ON)
+
+find_package(Threads REQUIRED)
+include(CheckCXXCompilerFlag)
+
+add_cxx_compiler_flag(-Wno-unused-variable)
+
+# NOTE: Some tests use `<cassert>` to perform the test. Therefore we must
+# strip -DNDEBUG from the default CMake flags in DEBUG mode.
+string(TOUPPER "${CMAKE_BUILD_TYPE}" uppercase_CMAKE_BUILD_TYPE)
+if( NOT uppercase_CMAKE_BUILD_TYPE STREQUAL "DEBUG" )
+  add_definitions( -UNDEBUG )
+  add_definitions(-DTEST_BENCHMARK_LIBRARY_HAS_NO_ASSERTIONS)
+  # Also remove /D NDEBUG to avoid MSVC warnings about conflicting defines.
+  foreach (flags_var_to_scrub
+      CMAKE_CXX_FLAGS_RELEASE
+      CMAKE_CXX_FLAGS_RELWITHDEBINFO
+      CMAKE_CXX_FLAGS_MINSIZEREL
+      CMAKE_C_FLAGS_RELEASE
+      CMAKE_C_FLAGS_RELWITHDEBINFO
+      CMAKE_C_FLAGS_MINSIZEREL)
+    string (REGEX REPLACE "(^| )[/-]D *NDEBUG($| )" " "
+      "${flags_var_to_scrub}" "${${flags_var_to_scrub}}")
+  endforeach()
+endif()
+
+if (NOT BUILD_SHARED_LIBS)
+  add_definitions(-DBENCHMARK_STATIC_DEFINE)
+endif()
+
+check_cxx_compiler_flag(-O3 BENCHMARK_HAS_O3_FLAG)
+set(BENCHMARK_O3_FLAG "")
+if (BENCHMARK_HAS_O3_FLAG)
+  set(BENCHMARK_O3_FLAG "-O3")
+endif()
+
+# NOTE: These flags must be added after find_package(Threads REQUIRED) otherwise
+# they will break the configuration check.
+if (DEFINED BENCHMARK_CXX_LINKER_FLAGS)
+  list(APPEND CMAKE_EXE_LINKER_FLAGS ${BENCHMARK_CXX_LINKER_FLAGS})
+endif()
+
+add_library(output_test_helper STATIC output_test_helper.cc output_test.h)
+target_link_libraries(output_test_helper PRIVATE benchmark::benchmark)
+
+macro(compile_benchmark_test name)
+  add_executable(${name} "${name}.cc")
+  target_link_libraries(${name} benchmark::benchmark ${CMAKE_THREAD_LIBS_INIT})
+  if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "NVHPC")
+  target_compile_options( ${name} PRIVATE --diag_suppress partial_override )
+  endif()
+endmacro(compile_benchmark_test)
+
+macro(compile_benchmark_test_with_main name)
+  add_executable(${name} "${name}.cc")
+  target_link_libraries(${name} benchmark::benchmark_main)
+endmacro(compile_benchmark_test_with_main)
+
+macro(compile_output_test name)
+  add_executable(${name} "${name}.cc" output_test.h)
+  target_link_libraries(${name} output_test_helper benchmark::benchmark_main
+          ${BENCHMARK_CXX_LIBRARIES} ${CMAKE_THREAD_LIBS_INIT})
+endmacro(compile_output_test)
+
+macro(benchmark_add_test)
+  add_test(${ARGV})
+  if(WIN32 AND BUILD_SHARED_LIBS)
+    cmake_parse_arguments(TEST "" "NAME" "" ${ARGN})
+    set_tests_properties(${TEST_NAME} PROPERTIES ENVIRONMENT_MODIFICATION "PATH=path_list_prepend:$<TARGET_FILE_DIR:benchmark::benchmark>")
+  endif()
+endmacro(benchmark_add_test)
+
+# Demonstration executable
+
+compile_benchmark_test_with_main(cxx11_test)
+if(DEFINED MSVC)
+  # MSVC does not really support C++11.
+  set_property(TARGET cxx11_test PROPERTY CXX_STANDARD 14)
+else()
+  set_property(TARGET cxx11_test PROPERTY CXX_STANDARD 11)
+endif()
+set_property(TARGET cxx11_test PROPERTY CXX_STANDARD_REQUIRED ON)
+set_property(TARGET cxx11_test PROPERTY CXX_EXTENSIONS OFF)
+benchmark_add_test(NAME cxx11_test COMMAND cxx11_test --benchmark_min_time=0.01s)
+
+compile_benchmark_test(benchmark_test)
+benchmark_add_test(NAME benchmark COMMAND benchmark_test --benchmark_min_time=0.01s)
+
+compile_benchmark_test(spec_arg_test)
+benchmark_add_test(NAME spec_arg COMMAND spec_arg_test --benchmark_filter=BM_NotChosen)
+
+compile_benchmark_test(spec_arg_verbosity_test)
+benchmark_add_test(NAME spec_arg_verbosity COMMAND spec_arg_verbosity_test --v=42)
+
+compile_benchmark_test(benchmark_setup_teardown_test)
+benchmark_add_test(NAME benchmark_setup_teardown COMMAND benchmark_setup_teardown_test)
+
+compile_benchmark_test(filter_test)
+macro(add_filter_test name filter expect)
+  benchmark_add_test(NAME ${name} COMMAND filter_test --benchmark_min_time=0.01s --benchmark_filter=${filter} ${expect})
+  benchmark_add_test(NAME ${name}_list_only COMMAND filter_test --benchmark_list_tests --benchmark_filter=${filter} ${expect})
+endmacro(add_filter_test)
+
+compile_benchmark_test(benchmark_min_time_flag_time_test)
+benchmark_add_test(NAME min_time_flag_time COMMAND benchmark_min_time_flag_time_test)
+
+compile_benchmark_test(benchmark_min_time_flag_iters_test)
+benchmark_add_test(NAME min_time_flag_iters COMMAND benchmark_min_time_flag_iters_test)
+
+add_filter_test(filter_simple "Foo" 3)
+add_filter_test(filter_simple_negative "-Foo" 2)
+add_filter_test(filter_suffix "BM_.*" 4)
+add_filter_test(filter_suffix_negative "-BM_.*" 1)
+add_filter_test(filter_regex_all ".*" 5)
+add_filter_test(filter_regex_all_negative "-.*" 0)
+add_filter_test(filter_regex_blank "" 5)
+add_filter_test(filter_regex_blank_negative "-" 0)
+add_filter_test(filter_regex_none "monkey" 0)
+add_filter_test(filter_regex_none_negative "-monkey" 5)
+add_filter_test(filter_regex_wildcard ".*Foo.*" 3)
+add_filter_test(filter_regex_wildcard_negative "-.*Foo.*" 2)
+add_filter_test(filter_regex_begin "^BM_.*" 4)
+add_filter_test(filter_regex_begin_negative "-^BM_.*" 1)
+add_filter_test(filter_regex_begin2 "^N" 1)
+add_filter_test(filter_regex_begin2_negative "-^N" 4)
+add_filter_test(filter_regex_end ".*Ba$" 1)
+add_filter_test(filter_regex_end_negative "-.*Ba$" 4)
+
+compile_benchmark_test(options_test)
+benchmark_add_test(NAME options_benchmarks COMMAND options_test --benchmark_min_time=0.01s)
+
+compile_benchmark_test(basic_test)
+benchmark_add_test(NAME basic_benchmark COMMAND basic_test --benchmark_min_time=0.01s)
+
+compile_output_test(repetitions_test)
+benchmark_add_test(NAME repetitions_benchmark COMMAND repetitions_test --benchmark_min_time=0.01s --benchmark_repetitions=3)
+
+compile_benchmark_test(diagnostics_test)
+benchmark_add_test(NAME diagnostics_test COMMAND diagnostics_test --benchmark_min_time=0.01s)
+
+compile_benchmark_test(skip_with_error_test)
+benchmark_add_test(NAME skip_with_error_test COMMAND skip_with_error_test --benchmark_min_time=0.01s)
+
+compile_benchmark_test(donotoptimize_test)
+# Enable errors for deprecated deprecations (DoNotOptimize(Tp const& value)).
+check_cxx_compiler_flag(-Werror=deprecated-declarations BENCHMARK_HAS_DEPRECATED_DECLARATIONS_FLAG)
+if (BENCHMARK_HAS_DEPRECATED_DECLARATIONS_FLAG)
+  target_compile_options (donotoptimize_test PRIVATE "-Werror=deprecated-declarations")
+endif()
+# Some of the issues with DoNotOptimize only occur when optimization is enabled
+check_cxx_compiler_flag(-O3 BENCHMARK_HAS_O3_FLAG)
+if (BENCHMARK_HAS_O3_FLAG)
+  set_target_properties(donotoptimize_test PROPERTIES COMPILE_FLAGS "-O3")
+endif()
+benchmark_add_test(NAME donotoptimize_test COMMAND donotoptimize_test --benchmark_min_time=0.01s)
+
+compile_benchmark_test(fixture_test)
+benchmark_add_test(NAME fixture_test COMMAND fixture_test --benchmark_min_time=0.01s)
+
+compile_benchmark_test(register_benchmark_test)
+benchmark_add_test(NAME register_benchmark_test COMMAND register_benchmark_test --benchmark_min_time=0.01s)
+
+compile_benchmark_test(map_test)
+benchmark_add_test(NAME map_test COMMAND map_test --benchmark_min_time=0.01s)
+
+compile_benchmark_test(multiple_ranges_test)
+benchmark_add_test(NAME multiple_ranges_test COMMAND multiple_ranges_test --benchmark_min_time=0.01s)
+
+compile_benchmark_test(args_product_test)
+benchmark_add_test(NAME args_product_test COMMAND args_product_test --benchmark_min_time=0.01s)
+
+compile_benchmark_test_with_main(link_main_test)
+benchmark_add_test(NAME link_main_test COMMAND link_main_test --benchmark_min_time=0.01s)
+
+compile_output_test(reporter_output_test)
+benchmark_add_test(NAME reporter_output_test COMMAND reporter_output_test --benchmark_min_time=0.01s)
+
+compile_output_test(templated_fixture_test)
+benchmark_add_test(NAME templated_fixture_test COMMAND templated_fixture_test --benchmark_min_time=0.01s)
+
+compile_output_test(templated_fixture_method_test)
+benchmark_add_test(NAME templated_fixture_method_test COMMAND templated_fixture_method_test --benchmark_min_time=0.01s)
+
+compile_output_test(user_counters_test)
+benchmark_add_test(NAME user_counters_test COMMAND user_counters_test --benchmark_min_time=0.2s)
+
+compile_output_test(user_counters_threads_test)
+benchmark_add_test(NAME user_counters_threads_test COMMAND user_counters_threads_test --benchmark_min_time=0.2s)
+
+compile_output_test(perf_counters_test)
+benchmark_add_test(NAME perf_counters_test COMMAND perf_counters_test --benchmark_min_time=0.01s --benchmark_perf_counters=CYCLES,INSTRUCTIONS)
+
+compile_output_test(internal_threading_test)
+benchmark_add_test(NAME internal_threading_test COMMAND internal_threading_test --benchmark_min_time=0.01s)
+
+compile_output_test(manual_threading_test)
+benchmark_add_test(NAME manual_threading_test COMMAND manual_threading_test --benchmark_min_time=0.01s)
+
+compile_output_test(report_aggregates_only_test)
+benchmark_add_test(NAME report_aggregates_only_test COMMAND report_aggregates_only_test --benchmark_min_time=0.01s)
+
+compile_output_test(display_aggregates_only_test)
+benchmark_add_test(NAME display_aggregates_only_test COMMAND display_aggregates_only_test --benchmark_min_time=0.01s)
+
+compile_output_test(user_counters_tabular_test)
+benchmark_add_test(NAME user_counters_tabular_test COMMAND user_counters_tabular_test --benchmark_counters_tabular=true --benchmark_min_time=0.2s)
+
+compile_output_test(user_counters_thousands_test)
+benchmark_add_test(NAME user_counters_thousands_test COMMAND user_counters_thousands_test --benchmark_min_time=0.01s)
+
+compile_output_test(memory_manager_test)
+benchmark_add_test(NAME memory_manager_test COMMAND memory_manager_test --benchmark_min_time=0.01s)
+
+compile_output_test(profiler_manager_test)
+benchmark_add_test(NAME profiler_manager_test COMMAND profiler_manager_test --benchmark_min_time=0.01s)
+
+compile_benchmark_test(profiler_manager_iterations_test)
+benchmark_add_test(NAME profiler_manager_iterations COMMAND profiler_manager_iterations_test)
+
+compile_output_test(complexity_test)
+benchmark_add_test(NAME complexity_benchmark COMMAND complexity_test --benchmark_min_time=1000000x)
+
+compile_output_test(locale_impermeability_test)
+benchmark_add_test(NAME locale_impermeability_test COMMAND locale_impermeability_test)
+
+compile_output_test(scoped_pause_test)
+benchmark_add_test(NAME scoped_pause_test COMMAND scoped_pause_test)
+
+###############################################################################
+# GoogleTest Unit Tests
+###############################################################################
+
+if (BENCHMARK_ENABLE_GTEST_TESTS)
+  macro(compile_gtest name)
+    add_executable(${name} "${name}.cc")
+    target_link_libraries(${name} benchmark::benchmark
+        gmock_main ${CMAKE_THREAD_LIBS_INIT})
+  endmacro(compile_gtest)
+
+  macro(add_gtest name)
+    compile_gtest(${name})
+    benchmark_add_test(NAME ${name} COMMAND ${name})
+    if(WIN32 AND BUILD_SHARED_LIBS)
+      set_tests_properties(${name} PROPERTIES
+        ENVIRONMENT_MODIFICATION "PATH=path_list_prepend:$<TARGET_FILE_DIR:benchmark::benchmark>;PATH=path_list_prepend:$<TARGET_FILE_DIR:gmock_main>"
+      )
+    endif()
+  endmacro()
+
+  add_gtest(benchmark_gtest)
+  add_gtest(benchmark_name_gtest)
+  add_gtest(benchmark_random_interleaving_gtest)
+  add_gtest(commandlineflags_gtest)
+  add_gtest(statistics_gtest)
+  add_gtest(string_util_gtest)
+  add_gtest(perf_counters_gtest)
+  add_gtest(time_unit_gtest)
+  add_gtest(min_time_parse_gtest)
+  add_gtest(profiler_manager_gtest)
+  add_gtest(benchmark_setup_teardown_cb_types_gtest)
+  add_gtest(memory_results_gtest)
+endif(BENCHMARK_ENABLE_GTEST_TESTS)
+
+###############################################################################
+# Assembly Unit Tests
+###############################################################################
+
+if (BENCHMARK_ENABLE_ASSEMBLY_TESTS)
+  if (NOT LLVM_FILECHECK_EXE)
+    message(FATAL_ERROR "LLVM FileCheck is required when including this file")
+  endif()
+  include(AssemblyTests.cmake)
+  add_filecheck_test(donotoptimize_assembly_test)
+  add_filecheck_test(state_assembly_test)
+  add_filecheck_test(clobber_memory_assembly_test)
+endif()
+
+
+
+###############################################################################
+# Code Coverage Configuration
+###############################################################################
+
+# Add the coverage command(s)
+if(CMAKE_BUILD_TYPE)
+  string(TOLOWER ${CMAKE_BUILD_TYPE} CMAKE_BUILD_TYPE_LOWER)
+endif()
+if (${CMAKE_BUILD_TYPE_LOWER} MATCHES "coverage")
+  find_program(GCOV gcov)
+  find_program(LCOV lcov)
+  find_program(GENHTML genhtml)
+  find_program(CTEST ctest)
+  if (GCOV AND LCOV AND GENHTML AND CTEST AND HAVE_CXX_FLAG_COVERAGE)
+    add_custom_command(
+      OUTPUT ${CMAKE_BINARY_DIR}/lcov/index.html
+      COMMAND ${LCOV} -q -z -d .
+      COMMAND ${LCOV} -q --no-external -c -b "${CMAKE_SOURCE_DIR}" -d . -o before.lcov -i
+      COMMAND ${CTEST} --force-new-ctest-process
+      COMMAND ${LCOV} -q --no-external -c -b "${CMAKE_SOURCE_DIR}" -d . -o after.lcov
+      COMMAND ${LCOV} -q -a before.lcov -a after.lcov --output-file final.lcov
+      COMMAND ${LCOV} -q -r final.lcov "'${CMAKE_SOURCE_DIR}/test/*'" -o final.lcov
+      COMMAND ${GENHTML} final.lcov -o lcov --demangle-cpp --sort -p "${CMAKE_BINARY_DIR}" -t benchmark
+      DEPENDS filter_test benchmark_test options_test basic_test fixture_test complexity_test
+      WORKING_DIRECTORY ${CMAKE_BINARY_DIR}
+      COMMENT "Running LCOV"
+    )
+    add_custom_target(coverage
+      DEPENDS ${CMAKE_BINARY_DIR}/lcov/index.html
+      COMMENT "LCOV report at lcov/index.html"
+    )
+    message(STATUS "Coverage command added")
+  else()
+    if (HAVE_CXX_FLAG_COVERAGE)
+      set(CXX_FLAG_COVERAGE_MESSAGE supported)
+    else()
+      set(CXX_FLAG_COVERAGE_MESSAGE unavailable)
+    endif()
+    message(WARNING
+      "Coverage not available:\n"
+      "  gcov: ${GCOV}\n"
+      "  lcov: ${LCOV}\n"
+      "  genhtml: ${GENHTML}\n"
+      "  ctest: ${CTEST}\n"
+      "  --coverage flag: ${CXX_FLAG_COVERAGE_MESSAGE}")
+  endif()
+endif()
+
+if (BENCHMARK_ENABLE_RUST_BINDINGS)
+  find_program(CARGO_EXECUTABLE cargo REQUIRED)
+  add_custom_target(build_rust_bindings ALL
+    COMMAND ${CARGO_EXECUTABLE} build --tests
+    WORKING_DIRECTORY ${PROJECT_SOURCE_DIR}/bindings/rust
+    COMMENT "Building Rust bindings and tests"
+  )
+  add_test(NAME rust_bindings_tests
+    COMMAND ${CARGO_EXECUTABLE} test
+    WORKING_DIRECTORY ${PROJECT_SOURCE_DIR}/bindings/rust
+  )
+endif()

test/args_product_test.cc

@@ -0,0 +1,80 @@
+#include <cassert>
+#include <iostream>
+#include <set>
+#include <vector>
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+#include "benchmark/utils.h"
+
+class ArgsProductFixture : public ::benchmark::Fixture {
+ public:
+  ArgsProductFixture()
+      : expectedValues({{0, 100, 2000, 30000},
+                        {1, 15, 3, 8},
+                        {1, 15, 3, 9},
+                        {1, 15, 7, 8},
+                        {1, 15, 7, 9},
+                        {1, 15, 10, 8},
+                        {1, 15, 10, 9},
+                        {2, 15, 3, 8},
+                        {2, 15, 3, 9},
+                        {2, 15, 7, 8},
+                        {2, 15, 7, 9},
+                        {2, 15, 10, 8},
+                        {2, 15, 10, 9},
+                        {4, 5, 6, 11}}) {}
+
+  void SetUp(const ::benchmark::State& state) override {
+    std::vector<int64_t> ranges = {state.range(0), state.range(1),
+                                   state.range(2), state.range(3)};
+
+    assert(expectedValues.find(ranges) != expectedValues.end());
+
+    actualValues.insert(ranges);
+  }
+
+  // NOTE: This is not TearDown as we want to check after _all_ runs are
+  // complete.
+  ~ArgsProductFixture() override {
+    if (actualValues != expectedValues) {
+      std::cout << "EXPECTED\n";
+      for (const auto& v : expectedValues) {
+        std::cout << "{";
+        for (int64_t iv : v) {
+          std::cout << iv << ", ";
+        }
+        std::cout << "}\n";
+      }
+      std::cout << "ACTUAL\n";
+      for (const auto& v : actualValues) {
+        std::cout << "{";
+        for (int64_t iv : v) {
+          std::cout << iv << ", ";
+        }
+        std::cout << "}\n";
+      }
+    }
+  }
+
+  std::set<std::vector<int64_t>> expectedValues;
+  std::set<std::vector<int64_t>> actualValues;
+};
+
+BENCHMARK_DEFINE_F(ArgsProductFixture, Empty)(benchmark::State& state) {
+  for (auto _ : state) {
+    int64_t product =
+        state.range(0) * state.range(1) * state.range(2) * state.range(3);
+    for (int64_t x = 0; x < product; x++) {
+      benchmark::DoNotOptimize(x);
+    }
+  }
+}
+
+BENCHMARK_REGISTER_F(ArgsProductFixture, Empty)
+    ->Args({0, 100, 2000, 30000})
+    ->ArgsProduct({{1, 2}, {15}, {3, 7, 10}, {8, 9}})
+    ->Args({4, 5, 6, 11});
+
+BENCHMARK_MAIN();

test/basic_test.cc

@@ -0,0 +1,184 @@
+
+#include "benchmark/macros.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+#include "benchmark/types.h"
+#include "benchmark/utils.h"
+
+#define BASIC_BENCHMARK_TEST(x) BENCHMARK(x)->Arg(8)->Arg(512)->Arg(8192)
+
+namespace {
+void BM_empty(benchmark::State& state) {
+  for (auto _ : state) {
+    auto iterations = static_cast<double>(state.iterations()) *
+                      static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(iterations);
+  }
+}
+BENCHMARK(BM_empty);
+BENCHMARK(BM_empty)->ThreadPerCpu();
+
+void BM_spin_empty(benchmark::State& state) {
+  for (auto _ : state) {
+    for (auto x = 0; x < state.range(0); ++x) {
+      benchmark::DoNotOptimize(x);
+    }
+  }
+}
+BASIC_BENCHMARK_TEST(BM_spin_empty);
+BASIC_BENCHMARK_TEST(BM_spin_empty)->ThreadPerCpu();
+
+void BM_spin_pause_before(benchmark::State& state) {
+  for (auto i = 0; i < state.range(0); ++i) {
+    benchmark::DoNotOptimize(i);
+  }
+  for (auto _ : state) {
+    for (auto i = 0; i < state.range(0); ++i) {
+      benchmark::DoNotOptimize(i);
+    }
+  }
+}
+BASIC_BENCHMARK_TEST(BM_spin_pause_before);
+BASIC_BENCHMARK_TEST(BM_spin_pause_before)->ThreadPerCpu();
+
+void BM_spin_pause_during(benchmark::State& state) {
+  for (auto _ : state) {
+    state.PauseTiming();
+    for (auto i = 0; i < state.range(0); ++i) {
+      benchmark::DoNotOptimize(i);
+    }
+    state.ResumeTiming();
+    for (auto i = 0; i < state.range(0); ++i) {
+      benchmark::DoNotOptimize(i);
+    }
+  }
+}
+BASIC_BENCHMARK_TEST(BM_spin_pause_during);
+BASIC_BENCHMARK_TEST(BM_spin_pause_during)->ThreadPerCpu();
+
+void BM_pause_during(benchmark::State& state) {
+  for (auto _ : state) {
+    state.PauseTiming();
+    state.ResumeTiming();
+  }
+}
+BENCHMARK(BM_pause_during);
+BENCHMARK(BM_pause_during)->ThreadPerCpu();
+BENCHMARK(BM_pause_during)->UseRealTime();
+BENCHMARK(BM_pause_during)->UseRealTime()->ThreadPerCpu();
+
+void BM_spin_pause_after(benchmark::State& state) {
+  for (auto _ : state) {
+    for (auto i = 0; i < state.range(0); ++i) {
+      benchmark::DoNotOptimize(i);
+    }
+  }
+  for (auto i = 0; i < state.range(0); ++i) {
+    benchmark::DoNotOptimize(i);
+  }
+}
+BASIC_BENCHMARK_TEST(BM_spin_pause_after);
+BASIC_BENCHMARK_TEST(BM_spin_pause_after)->ThreadPerCpu();
+
+void BM_spin_pause_before_and_after(benchmark::State& state) {
+  for (auto i = 0; i < state.range(0); ++i) {
+    benchmark::DoNotOptimize(i);
+  }
+  for (auto _ : state) {
+    for (auto i = 0; i < state.range(0); ++i) {
+      benchmark::DoNotOptimize(i);
+    }
+  }
+  for (auto i = 0; i < state.range(0); ++i) {
+    benchmark::DoNotOptimize(i);
+  }
+}
+BASIC_BENCHMARK_TEST(BM_spin_pause_before_and_after);
+BASIC_BENCHMARK_TEST(BM_spin_pause_before_and_after)->ThreadPerCpu();
+
+void BM_empty_stop_start(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+}
+BENCHMARK(BM_empty_stop_start);
+BENCHMARK(BM_empty_stop_start)->ThreadPerCpu();
+
+void BM_KeepRunning(benchmark::State& state) {
+  benchmark::IterationCount iter_count = 0;
+  assert(iter_count == state.iterations());
+  while (state.KeepRunning()) {
+    ++iter_count;
+  }
+  assert(iter_count == state.iterations());
+}
+BENCHMARK(BM_KeepRunning);
+
+void BM_KeepRunningBatch(benchmark::State& state) {
+  // Choose a batch size >1000 to skip the typical runs with iteration
+  // targets of 10, 100 and 1000.  If these are not actually skipped the
+  // bug would be detectable as consecutive runs with the same iteration
+  // count.  Below we assert that this does not happen.
+  const benchmark::IterationCount batch_size = 1009;
+
+  static benchmark::IterationCount prior_iter_count = 0;
+  benchmark::IterationCount iter_count = 0;
+  while (state.KeepRunningBatch(batch_size)) {
+    iter_count += batch_size;
+  }
+  assert(state.iterations() == iter_count);
+
+  // Verify that the iteration count always increases across runs (see
+  // comment above).
+  assert(iter_count == batch_size            // max_iterations == 1
+         || iter_count > prior_iter_count);  // max_iterations > batch_size
+  prior_iter_count = iter_count;
+}
+// Register with a fixed repetition count to establish the invariant that
+// the iteration count should always change across runs.  This overrides
+// the --benchmark_repetitions command line flag, which would otherwise
+// cause this test to fail if set > 1.
+BENCHMARK(BM_KeepRunningBatch)->Repetitions(1);
+
+void BM_RangedFor(benchmark::State& state) {
+  benchmark::IterationCount iter_count = 0;
+  for (auto _ : state) {
+    ++iter_count;
+  }
+  assert(iter_count == state.max_iterations);
+}
+BENCHMARK(BM_RangedFor);
+
+template <typename T>
+void BM_OneTemplateFunc(benchmark::State& state) {
+  auto arg = state.range(0);
+  BENCHMARK_UNUSED T sum = 0;
+  for (auto _ : state) {
+    sum += static_cast<T>(arg);
+  }
+}
+BENCHMARK(BM_OneTemplateFunc<int>)->Arg(1);
+BENCHMARK(BM_OneTemplateFunc<double>)->Arg(1);
+
+template <typename A, typename B>
+void BM_TwoTemplateFunc(benchmark::State& state) {
+  auto arg = state.range(0);
+  BENCHMARK_UNUSED A sum = 0;
+  BENCHMARK_UNUSED B prod = 1;
+  for (auto _ : state) {
+    sum += static_cast<A>(arg);
+    prod *= static_cast<B>(arg);
+  }
+}
+BENCHMARK(BM_TwoTemplateFunc<int, double>)->Arg(1);
+BENCHMARK(BM_TwoTemplateFunc<double, int>)->Arg(1);
+
+// Ensure that StateIterator provides all the necessary typedefs required to
+// instantiate std::iterator_traits.
+static_assert(
+    std::is_same<typename std::iterator_traits<
+                     benchmark::State::StateIterator>::value_type,
+                 typename benchmark::State::StateIterator::value_type>::value,
+    "");
+}  // end namespace
+
+BENCHMARK_MAIN();

test/benchmark_gtest.cc

@@ -0,0 +1,169 @@
+#include <map>
+#include <string>
+#include <vector>
+
+#include "../src/benchmark_register.h"
+#include "benchmark/benchmark_api.h"
+#include "gmock/gmock.h"
+#include "gtest/gtest.h"
+
+namespace benchmark {
+namespace internal {
+
+namespace {
+
+TEST(AddRangeTest, Simple) {
+  std::vector<int> dst;
+  AddRange(&dst, 1, 2, 2);
+  EXPECT_THAT(dst, testing::ElementsAre(1, 2));
+}
+
+TEST(AddRangeTest, Simple64) {
+  std::vector<int64_t> dst;
+  AddRange(&dst, static_cast<int64_t>(1), static_cast<int64_t>(2), 2);
+  EXPECT_THAT(dst, testing::ElementsAre(1, 2));
+}
+
+TEST(AddRangeTest, Advanced) {
+  std::vector<int> dst;
+  AddRange(&dst, 5, 15, 2);
+  EXPECT_THAT(dst, testing::ElementsAre(5, 8, 15));
+}
+
+TEST(AddRangeTest, Advanced64) {
+  std::vector<int64_t> dst;
+  AddRange(&dst, static_cast<int64_t>(5), static_cast<int64_t>(15), 2);
+  EXPECT_THAT(dst, testing::ElementsAre(5, 8, 15));
+}
+
+TEST(AddRangeTest, FullRange8) {
+  std::vector<int8_t> dst;
+  AddRange(&dst, int8_t{1}, std::numeric_limits<int8_t>::max(), 8);
+  EXPECT_THAT(
+      dst, testing::ElementsAre(int8_t{1}, int8_t{8}, int8_t{64}, int8_t{127}));
+}
+
+TEST(AddRangeTest, FullRange64) {
+  std::vector<int64_t> dst;
+  AddRange(&dst, int64_t{1}, std::numeric_limits<int64_t>::max(), 1024);
+  EXPECT_THAT(
+      dst, testing::ElementsAre(1LL, 1024LL, 1048576LL, 1073741824LL,
+                                1099511627776LL, 1125899906842624LL,
+                                1152921504606846976LL, 9223372036854775807LL));
+}
+
+TEST(AddRangeTest, NegativeRanges) {
+  std::vector<int> dst;
+  AddRange(&dst, -8, 0, 2);
+  EXPECT_THAT(dst, testing::ElementsAre(-8, -4, -2, -1, 0));
+}
+
+TEST(AddRangeTest, StrictlyNegative) {
+  std::vector<int> dst;
+  AddRange(&dst, -8, -1, 2);
+  EXPECT_THAT(dst, testing::ElementsAre(-8, -4, -2, -1));
+}
+
+TEST(AddRangeTest, SymmetricNegativeRanges) {
+  std::vector<int> dst;
+  AddRange(&dst, -8, 8, 2);
+  EXPECT_THAT(dst, testing::ElementsAre(-8, -4, -2, -1, 0, 1, 2, 4, 8));
+}
+
+TEST(AddRangeTest, SymmetricNegativeRangesOddMult) {
+  std::vector<int> dst;
+  AddRange(&dst, -30, 32, 5);
+  EXPECT_THAT(dst, testing::ElementsAre(-30, -25, -5, -1, 0, 1, 5, 25, 32));
+}
+
+TEST(AddRangeTest, NegativeRangesAsymmetric) {
+  std::vector<int> dst;
+  AddRange(&dst, -3, 5, 2);
+  EXPECT_THAT(dst, testing::ElementsAre(-3, -2, -1, 0, 1, 2, 4, 5));
+}
+
+TEST(AddRangeTest, NegativeRangesLargeStep) {
+  // Always include -1, 0, 1 when crossing zero.
+  std::vector<int> dst;
+  AddRange(&dst, -8, 8, 10);
+  EXPECT_THAT(dst, testing::ElementsAre(-8, -1, 0, 1, 8));
+}
+
+TEST(AddRangeTest, ZeroOnlyRange) {
+  std::vector<int> dst;
+  AddRange(&dst, 0, 0, 2);
+  EXPECT_THAT(dst, testing::ElementsAre(0));
+}
+
+TEST(AddRangeTest, ZeroStartingRange) {
+  std::vector<int> dst;
+  AddRange(&dst, 0, 2, 2);
+  EXPECT_THAT(dst, testing::ElementsAre(0, 1, 2));
+}
+
+TEST(AddRangeTest, NegativeRange64) {
+  std::vector<int64_t> dst;
+  AddRange<int64_t>(&dst, -4, 4, 2);
+  EXPECT_THAT(dst, testing::ElementsAre(-4, -2, -1, 0, 1, 2, 4));
+}
+
+TEST(AddRangeTest, NegativeRangePreservesExistingOrder) {
+  // If elements already exist in the range, ensure we don't change
+  // their ordering by adding negative values.
+  std::vector<int64_t> dst = {1, 2, 3};
+  AddRange<int64_t>(&dst, -2, 2, 2);
+  EXPECT_THAT(dst, testing::ElementsAre(1, 2, 3, -2, -1, 0, 1, 2));
+}
+
+TEST(AddRangeTest, FullNegativeRange64) {
+  std::vector<int64_t> dst;
+  const auto min = std::numeric_limits<int64_t>::min();
+  const auto max = std::numeric_limits<int64_t>::max();
+  AddRange(&dst, min, max, 1024);
+  EXPECT_THAT(
+      dst, testing::ElementsAreArray(std::vector<int64_t>{
+               min, -1152921504606846976LL, -1125899906842624LL,
+               -1099511627776LL, -1073741824LL, -1048576LL, -1024LL, -1LL, 0LL,
+               1LL, 1024LL, 1048576LL, 1073741824LL, 1099511627776LL,
+               1125899906842624LL, 1152921504606846976LL, max}));
+}
+
+TEST(AddRangeTest, Simple8) {
+  std::vector<int8_t> dst;
+  AddRange<int8_t>(&dst, int8_t{1}, int8_t{8}, int8_t{2});
+  EXPECT_THAT(dst,
+              testing::ElementsAre(int8_t{1}, int8_t{2}, int8_t{4}, int8_t{8}));
+}
+
+TEST(AddCustomContext, Simple) {
+  std::map<std::string, std::string>*& global_context = GetGlobalContext();
+  EXPECT_THAT(global_context, nullptr);
+
+  AddCustomContext("foo", "bar");
+  AddCustomContext("baz", "qux");
+
+  EXPECT_THAT(*global_context,
+              testing::UnorderedElementsAre(testing::Pair("foo", "bar"),
+                                            testing::Pair("baz", "qux")));
+
+  delete global_context;
+  global_context = nullptr;
+}
+
+TEST(AddCustomContext, DuplicateKey) {
+  std::map<std::string, std::string>*& global_context = GetGlobalContext();
+  EXPECT_THAT(global_context, nullptr);
+
+  AddCustomContext("foo", "bar");
+  AddCustomContext("foo", "qux");
+
+  EXPECT_THAT(*global_context,
+              testing::UnorderedElementsAre(testing::Pair("foo", "bar")));
+
+  delete global_context;
+  global_context = nullptr;
+}
+
+}  // namespace
+}  // namespace internal
+}  // namespace benchmark

test/benchmark_min_time_flag_iters_test.cc

@@ -0,0 +1,71 @@
+#include <cassert>
+#include <cstdlib>
+#include <cstring>
+#include <string>
+#include <vector>
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/registration.h"
+#include "benchmark/reporter.h"
+#include "benchmark/state.h"
+
+// Tests that we can specify the number of iterations with
+// --benchmark_min_time=<NUM>x.
+namespace {
+
+class TestReporter : public benchmark::ConsoleReporter {
+ public:
+  bool ReportContext(const Context& context) override {
+    return ConsoleReporter::ReportContext(context);
+  };
+
+  void ReportRuns(const std::vector<Run>& report) override {
+    assert(report.size() == 1);
+    iter_nums_.push_back(report[0].iterations);
+    ConsoleReporter::ReportRuns(report);
+  };
+
+  TestReporter() {}
+
+  ~TestReporter() override {}
+
+  const std::vector<benchmark::IterationCount>& GetIters() const {
+    return iter_nums_;
+  }
+
+ private:
+  std::vector<benchmark::IterationCount> iter_nums_;
+};
+
+void BM_MyBench(benchmark::State& state) {
+  for (auto s : state) {
+  }
+}
+}  // end namespace
+
+BENCHMARK(BM_MyBench);
+
+int main(int argc, char** argv) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+
+  // Make a fake argv and append the new --benchmark_min_time=<foo> to it.
+  int fake_argc = argc + 1;
+  std::vector<const char*> fake_argv(static_cast<size_t>(fake_argc));
+  for (size_t i = 0; i < static_cast<size_t>(argc); ++i) {
+    fake_argv[i] = argv[i];
+  }
+  fake_argv[static_cast<size_t>(argc)] = "--benchmark_min_time=4x";
+
+  benchmark::Initialize(&fake_argc, const_cast<char**>(fake_argv.data()));
+
+  TestReporter test_reporter;
+  const size_t returned_count =
+      benchmark::RunSpecifiedBenchmarks(&test_reporter, "BM_MyBench");
+  assert(returned_count == 1);
+
+  // Check the executed iters.
+  const std::vector<benchmark::IterationCount> iters = test_reporter.GetIters();
+  assert(!iters.empty() && iters[0] == 4);
+
+  return 0;
+}

test/benchmark_min_time_flag_time_test.cc

@@ -0,0 +1,96 @@
+#include <cassert>
+#include <climits>
+#include <cmath>
+#include <cstdlib>
+#include <cstring>
+#include <iostream>
+#include <string>
+#include <vector>
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/registration.h"
+#include "benchmark/reporter.h"
+#include "benchmark/state.h"
+
+// Tests that we can specify the min time with
+// --benchmark_min_time=<NUM> (no suffix needed) OR
+// --benchmark_min_time=<NUM>s
+namespace {
+
+// This is from benchmark.h
+typedef int64_t IterationCount;
+
+class TestReporter : public benchmark::ConsoleReporter {
+ public:
+  bool ReportContext(const Context& context) override {
+    return ConsoleReporter::ReportContext(context);
+  };
+
+  void ReportRuns(const std::vector<Run>& report) override {
+    assert(report.size() == 1);
+    ConsoleReporter::ReportRuns(report);
+  };
+
+  void ReportRunsConfig(double min_time, bool /* has_explicit_iters */,
+                        IterationCount /* iters */) override {
+    min_times_.push_back(min_time);
+  }
+
+  TestReporter() {}
+
+  ~TestReporter() override {}
+
+  const std::vector<double>& GetMinTimes() const { return min_times_; }
+
+ private:
+  std::vector<double> min_times_;
+};
+
+bool AlmostEqual(double a, double b) {
+  return std::fabs(a - b) < std::numeric_limits<double>::epsilon();
+}
+
+void DoTestHelper(int* argc, const char** argv, double expected) {
+  benchmark::Initialize(argc, const_cast<char**>(argv));
+
+  TestReporter test_reporter;
+  const size_t returned_count =
+      benchmark::RunSpecifiedBenchmarks(&test_reporter, "BM_MyBench");
+  assert(returned_count == 1);
+
+  // Check the min_time
+  const std::vector<double>& min_times = test_reporter.GetMinTimes();
+  assert(!min_times.empty() && AlmostEqual(min_times[0], expected));
+}
+
+void BM_MyBench(benchmark::State& state) {
+  for (auto s : state) {
+  }
+}
+BENCHMARK(BM_MyBench);
+
+}  // end namespace
+
+int main(int argc, char** argv) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+
+  // Make a fake argv and append the new --benchmark_min_time=<foo> to it.
+  int fake_argc = argc + 1;
+  std::vector<const char*> fake_argv(static_cast<size_t>(fake_argc));
+
+  for (size_t i = 0; i < static_cast<size_t>(argc); ++i) {
+    fake_argv[i] = argv[i];
+  }
+
+  const char* no_suffix = "--benchmark_min_time=4";
+  const char* with_suffix = "--benchmark_min_time=4.0s";
+  double expected = 4.0;
+
+  fake_argv[static_cast<size_t>(argc)] = no_suffix;
+  DoTestHelper(&fake_argc, fake_argv.data(), expected);
+
+  fake_argv[static_cast<size_t>(argc)] = with_suffix;
+  DoTestHelper(&fake_argc, fake_argv.data(), expected);
+
+  return 0;
+}

test/benchmark_name_gtest.cc

@@ -0,0 +1,82 @@
+#include "benchmark/reporter.h"
+#include "gtest/gtest.h"
+
+namespace {
+
+using namespace benchmark;
+using namespace benchmark::internal;
+
+TEST(BenchmarkNameTest, Empty) {
+  const auto name = BenchmarkName();
+  EXPECT_EQ(name.str(), std::string());
+}
+
+TEST(BenchmarkNameTest, FunctionName) {
+  auto name = BenchmarkName();
+  name.function_name = "function_name";
+  EXPECT_EQ(name.str(), "function_name");
+}
+
+TEST(BenchmarkNameTest, FunctionNameAndArgs) {
+  auto name = BenchmarkName();
+  name.function_name = "function_name";
+  name.args = "some_args:3/4/5";
+  EXPECT_EQ(name.str(), "function_name/some_args:3/4/5");
+}
+
+TEST(BenchmarkNameTest, MinTime) {
+  auto name = BenchmarkName();
+  name.function_name = "function_name";
+  name.args = "some_args:3/4";
+  name.min_time = "min_time:3.4s";
+  EXPECT_EQ(name.str(), "function_name/some_args:3/4/min_time:3.4s");
+}
+
+TEST(BenchmarkNameTest, MinWarmUpTime) {
+  auto name = BenchmarkName();
+  name.function_name = "function_name";
+  name.args = "some_args:3/4";
+  name.min_warmup_time = "min_warmup_time:3.5s";
+  EXPECT_EQ(name.str(), "function_name/some_args:3/4/min_warmup_time:3.5s");
+}
+
+TEST(BenchmarkNameTest, Iterations) {
+  auto name = BenchmarkName();
+  name.function_name = "function_name";
+  name.min_time = "min_time:3.4s";
+  name.iterations = "iterations:42";
+  EXPECT_EQ(name.str(), "function_name/min_time:3.4s/iterations:42");
+}
+
+TEST(BenchmarkNameTest, Repetitions) {
+  auto name = BenchmarkName();
+  name.function_name = "function_name";
+  name.min_time = "min_time:3.4s";
+  name.repetitions = "repetitions:24";
+  EXPECT_EQ(name.str(), "function_name/min_time:3.4s/repetitions:24");
+}
+
+TEST(BenchmarkNameTest, TimeType) {
+  auto name = BenchmarkName();
+  name.function_name = "function_name";
+  name.min_time = "min_time:3.4s";
+  name.time_type = "hammer_time";
+  EXPECT_EQ(name.str(), "function_name/min_time:3.4s/hammer_time");
+}
+
+TEST(BenchmarkNameTest, Threads) {
+  auto name = BenchmarkName();
+  name.function_name = "function_name";
+  name.min_time = "min_time:3.4s";
+  name.threads = "threads:256";
+  EXPECT_EQ(name.str(), "function_name/min_time:3.4s/threads:256");
+}
+
+TEST(BenchmarkNameTest, TestEmptyFunctionName) {
+  auto name = BenchmarkName();
+  name.args = "first:3/second:4";
+  name.threads = "threads:22";
+  EXPECT_EQ(name.str(), "first:3/second:4/threads:22");
+}
+
+}  // end namespace

test/benchmark_random_interleaving_gtest.cc

@@ -0,0 +1,130 @@
+#include <queue>
+#include <string>
+#include <vector>
+
+#include "../src/commandlineflags.h"
+#include "../src/string_util.h"
+#include "benchmark/benchmark_api.h"
+#include "benchmark/registration.h"
+#include "benchmark/reporter.h"
+#include "benchmark/state.h"
+#include "gmock/gmock.h"
+#include "gtest/gtest.h"
+
+namespace benchmark {
+
+BM_DECLARE_bool(benchmark_enable_random_interleaving);
+BM_DECLARE_string(benchmark_filter);
+BM_DECLARE_int32(benchmark_repetitions);
+
+namespace internal {
+namespace {
+
+class EventQueue : public std::queue<std::string> {
+ public:
+  void Put(const std::string& event) { push(event); }
+
+  void Clear() {
+    while (!empty()) {
+      pop();
+    }
+  }
+
+  std::string Get() {
+    std::string event = front();
+    pop();
+    return event;
+  }
+};
+
+// NOLINTNEXTLINE(cppcoreguidelines-avoid-non-const-global-variables)
+EventQueue* const queue = new EventQueue();
+
+class NullReporter : public BenchmarkReporter {
+ public:
+  bool ReportContext(const Context& /*context*/) override { return true; }
+  void ReportRuns(const std::vector<Run>& /* report */) override {}
+};
+
+class BenchmarkTest : public testing::Test {
+ public:
+  static void SetupHook(int /* num_threads */) { queue->push("Setup"); }
+
+  static void TeardownHook(int /* num_threads */) { queue->push("Teardown"); }
+
+  static void Execute(const std::string& pattern) {
+    queue->Clear();
+
+    std::unique_ptr<BenchmarkReporter> reporter(new NullReporter());
+    FLAGS_benchmark_filter = pattern;
+    RunSpecifiedBenchmarks(reporter.get());
+
+    queue->Put("DONE");  // End marker
+  }
+};
+
+void BM_Match1(benchmark::State& state) {
+  const int64_t arg = state.range(0);
+
+  for (auto _ : state) {
+  }
+  queue->Put(StrFormat("BM_Match1/%d", static_cast<int>(arg)));
+}
+BENCHMARK(BM_Match1)
+    ->Iterations(100)
+    ->Arg(1)
+    ->Arg(2)
+    ->Arg(3)
+    ->Range(10, 80)
+    ->Args({90})
+    ->Args({100});
+
+TEST_F(BenchmarkTest, Match1) {
+  Execute("BM_Match1");
+  ASSERT_EQ("BM_Match1/1", queue->Get());
+  ASSERT_EQ("BM_Match1/2", queue->Get());
+  ASSERT_EQ("BM_Match1/3", queue->Get());
+  ASSERT_EQ("BM_Match1/10", queue->Get());
+  ASSERT_EQ("BM_Match1/64", queue->Get());
+  ASSERT_EQ("BM_Match1/80", queue->Get());
+  ASSERT_EQ("BM_Match1/90", queue->Get());
+  ASSERT_EQ("BM_Match1/100", queue->Get());
+  ASSERT_EQ("DONE", queue->Get());
+}
+
+TEST_F(BenchmarkTest, Match1WithRepetition) {
+  FLAGS_benchmark_repetitions = 2;
+
+  Execute("BM_Match1/(64|80)");
+  ASSERT_EQ("BM_Match1/64", queue->Get());
+  ASSERT_EQ("BM_Match1/64", queue->Get());
+  ASSERT_EQ("BM_Match1/80", queue->Get());
+  ASSERT_EQ("BM_Match1/80", queue->Get());
+  ASSERT_EQ("DONE", queue->Get());
+}
+
+TEST_F(BenchmarkTest, Match1WithRandomInterleaving) {
+  FLAGS_benchmark_enable_random_interleaving = true;
+  FLAGS_benchmark_repetitions = 100;
+
+  std::map<std::string, int> element_count;
+  std::map<std::string, int> interleaving_count;
+  Execute("BM_Match1/(64|80)");
+  for (int i = 0; i < 100; ++i) {
+    std::vector<std::string> interleaving;
+    interleaving.push_back(queue->Get());
+    interleaving.push_back(queue->Get());
+    element_count[interleaving[0]]++;
+    element_count[interleaving[1]]++;
+    interleaving_count[StrFormat("%s,%s", interleaving[0].c_str(),
+                                 interleaving[1].c_str())]++;
+  }
+  EXPECT_EQ(element_count["BM_Match1/64"], 100) << "Unexpected repetitions.";
+  EXPECT_EQ(element_count["BM_Match1/80"], 100) << "Unexpected repetitions.";
+  EXPECT_GE(interleaving_count.size(), 2) << "Interleaving was not randomized.";
+  ASSERT_EQ("DONE", queue->Get());
+}
+
+}  // namespace
+}  // namespace internal
+}  // namespace benchmark

test/benchmark_setup_teardown_cb_types_gtest.cc

@@ -0,0 +1,129 @@
+#include "benchmark/benchmark_api.h"
+#include "benchmark/reporter.h"
+#include "benchmark/state.h"
+#include "benchmark/types.h"
+#include "gtest/gtest.h"
+
+using benchmark::Benchmark;
+using benchmark::BenchmarkReporter;
+using benchmark::callback_function;
+using benchmark::ClearRegisteredBenchmarks;
+using benchmark::RegisterBenchmark;
+using benchmark::RunSpecifiedBenchmarks;
+using benchmark::State;
+
+static int functor_called = 0;
+struct Functor {
+  void operator()(const benchmark::State& /*unused*/) { functor_called++; }
+};
+
+class NullReporter : public BenchmarkReporter {
+ public:
+  bool ReportContext(const Context& /*context*/) override { return true; }
+  void ReportRuns(const std::vector<Run>& /* report */) override {}
+};
+
+class BenchmarkTest : public testing::Test {
+ public:
+  Benchmark* bm;
+  NullReporter null_reporter;
+
+  int setup_calls;
+  int teardown_calls;
+
+  void SetUp() override {
+    setup_calls = 0;
+    teardown_calls = 0;
+    functor_called = 0;
+
+    bm = RegisterBenchmark("BM", [](State& st) {
+      for (auto _ : st) {
+      }
+    });
+    bm->Iterations(1);
+  }
+
+  void TearDown() override { ClearRegisteredBenchmarks(); }
+};
+
+// Test that Setup/Teardown can correctly take a lambda expressions
+TEST_F(BenchmarkTest, LambdaTestCopy) {
+  auto setup_lambda = [this](const State&) { setup_calls++; };
+  auto teardown_lambda = [this](const State&) { teardown_calls++; };
+  bm->Setup(setup_lambda);
+  bm->Teardown(teardown_lambda);
+  RunSpecifiedBenchmarks(&null_reporter);
+  EXPECT_EQ(setup_calls, 1);
+  EXPECT_EQ(teardown_calls, 1);
+}
+
+// Test that Setup/Teardown can correctly take a lambda expressions
+TEST_F(BenchmarkTest, LambdaTestMove) {
+  auto setup_lambda = [this](const State&) { setup_calls++; };
+  auto teardown_lambda = [this](const State&) { teardown_calls++; };
+  bm->Setup(std::move(setup_lambda));
+  bm->Teardown(std::move(teardown_lambda));
+  RunSpecifiedBenchmarks(&null_reporter);
+  EXPECT_EQ(setup_calls, 1);
+  EXPECT_EQ(teardown_calls, 1);
+}
+
+// Test that Setup/Teardown can correctly take std::function
+TEST_F(BenchmarkTest, CallbackFunctionCopy) {
+  callback_function setup_lambda = [this](const State&) { setup_calls++; };
+  callback_function teardown_lambda = [this](const State&) {
+    teardown_calls++;
+  };
+  bm->Setup(setup_lambda);
+  bm->Teardown(teardown_lambda);
+  RunSpecifiedBenchmarks(&null_reporter);
+  EXPECT_EQ(setup_calls, 1);
+  EXPECT_EQ(teardown_calls, 1);
+}
+
+// Test that Setup/Teardown can correctly take std::function
+TEST_F(BenchmarkTest, CallbackFunctionMove) {
+  callback_function setup_lambda = [this](const State&) { setup_calls++; };
+  callback_function teardown_lambda = [this](const State&) {
+    teardown_calls++;
+  };
+  bm->Setup(std::move(setup_lambda));
+  bm->Teardown(std::move(teardown_lambda));
+  RunSpecifiedBenchmarks(&null_reporter);
+  EXPECT_EQ(setup_calls, 1);
+  EXPECT_EQ(teardown_calls, 1);
+}
+
+// Test that Setup/Teardown can correctly take functors
+TEST_F(BenchmarkTest, FunctorCopy) {
+  Functor func;
+  bm->Setup(func);
+  bm->Teardown(func);
+  RunSpecifiedBenchmarks(&null_reporter);
+  EXPECT_EQ(functor_called, 2);
+}
+
+// Test that Setup/Teardown can correctly take functors
+TEST_F(BenchmarkTest, FunctorMove) {
+  Functor func1;
+  Functor func2;
+  bm->Setup(std::move(func1));
+  bm->Teardown(std::move(func2));
+  RunSpecifiedBenchmarks(&null_reporter);
+  EXPECT_EQ(functor_called, 2);
+}
+
+// Test that Setup/Teardown can not take nullptr
+TEST_F(BenchmarkTest, NullptrTest) {
+#if GTEST_HAS_DEATH_TEST
+  // Tests only runnable in debug mode (when BM_CHECK is enabled).
+#ifndef NDEBUG
+#ifndef TEST_BENCHMARK_LIBRARY_HAS_NO_ASSERTIONS
+  EXPECT_DEATH(bm->Setup(nullptr), "setup != nullptr");
+  EXPECT_DEATH(bm->Teardown(nullptr), "teardown != nullptr");
+#else
+  GTEST_SKIP() << "Test skipped because BM_CHECK is disabled";
+#endif
+#endif
+#endif
+}

test/benchmark_setup_teardown_test.cc

@@ -0,0 +1,163 @@
+#include <atomic>
+#include <cassert>
+#include <cstdlib>
+#include <cstring>
+#include <string>
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+
+// Test that Setup() and Teardown() are called exactly once
+// for each benchmark run (single-threaded).
+namespace {
+namespace singlethreaded {
+static int setup_call = 0;
+static int teardown_call = 0;
+}  // namespace singlethreaded
+}  // namespace
+static void DoSetup1(const benchmark::State& state) {
+  ++singlethreaded::setup_call;
+
+  // Setup/Teardown should never be called with any thread_idx != 0.
+  assert(state.thread_index() == 0);
+}
+
+static void DoTeardown1(const benchmark::State& state) {
+  ++singlethreaded::teardown_call;
+  assert(state.thread_index() == 0);
+}
+
+static void BM_with_setup(benchmark::State& state) {
+  for (auto s : state) {
+  }
+}
+BENCHMARK(BM_with_setup)
+    ->Arg(1)
+    ->Arg(3)
+    ->Arg(5)
+    ->Arg(7)
+    ->Iterations(100)
+    ->Setup(DoSetup1)
+    ->Teardown(DoTeardown1);
+
+// Test that Setup() and Teardown() are called once for each group of threads.
+namespace {
+namespace concurrent {
+static std::atomic<int> setup_call(0);
+static std::atomic<int> teardown_call(0);
+static std::atomic<int> func_call(0);
+}  // namespace concurrent
+
+void DoSetup2(const benchmark::State& state) {
+  concurrent::setup_call.fetch_add(1, std::memory_order_acquire);
+  assert(state.thread_index() == 0);
+}
+
+void DoTeardown2(const benchmark::State& state) {
+  concurrent::teardown_call.fetch_add(1, std::memory_order_acquire);
+  assert(state.thread_index() == 0);
+}
+
+void BM_concurrent(benchmark::State& state) {
+  for (auto s : state) {
+  }
+  concurrent::func_call.fetch_add(1, std::memory_order_acquire);
+}
+
+BENCHMARK(BM_concurrent)
+    ->Setup(DoSetup2)
+    ->Teardown(DoTeardown2)
+    ->Iterations(100)
+    ->Threads(5)
+    ->Threads(10)
+    ->Threads(15);
+
+// Testing interaction with Fixture::Setup/Teardown
+namespace fixture_interaction {
+int setup = 0;
+int fixture_setup = 0;
+}  // namespace fixture_interaction
+
+#define FIXTURE_BECHMARK_NAME MyFixture
+
+class FIXTURE_BECHMARK_NAME : public ::benchmark::Fixture {
+ public:
+  void SetUp(const ::benchmark::State& /*unused*/) override {
+    fixture_interaction::fixture_setup++;
+  }
+
+  ~FIXTURE_BECHMARK_NAME() override {}
+};
+
+BENCHMARK_F(FIXTURE_BECHMARK_NAME, BM_WithFixture)(benchmark::State& st) {
+  for (auto _ : st) {
+  }
+}
+
+void DoSetupWithFixture(const benchmark::State& /*unused*/) {
+  fixture_interaction::setup++;
+}
+
+BENCHMARK_REGISTER_F(FIXTURE_BECHMARK_NAME, BM_WithFixture)
+    ->Arg(1)
+    ->Arg(3)
+    ->Arg(5)
+    ->Arg(7)
+    ->Setup(DoSetupWithFixture)
+    ->Repetitions(1)
+    ->Iterations(100);
+
+// Testing repetitions.
+namespace repetitions {
+int setup = 0;
+}
+
+void DoSetupWithRepetitions(const benchmark::State& /*unused*/) {
+  repetitions::setup++;
+}
+void BM_WithRep(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+}
+
+BENCHMARK(BM_WithRep)
+    ->Arg(1)
+    ->Arg(3)
+    ->Arg(5)
+    ->Arg(7)
+    ->Setup(DoSetupWithRepetitions)
+    ->Iterations(100)
+    ->Repetitions(4);
+}  // namespace
+
+int main(int argc, char** argv) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+
+  benchmark::Initialize(&argc, argv);
+
+  size_t ret = benchmark::RunSpecifiedBenchmarks(".");
+  assert(ret > 0);
+
+  // Setup/Teardown is called once for each arg group (1,3,5,7).
+  assert(singlethreaded::setup_call == 4);
+  assert(singlethreaded::teardown_call == 4);
+
+  // 3 group of threads calling this function (3,5,10).
+  assert(concurrent::setup_call.load(std::memory_order_relaxed) == 3);
+  assert(concurrent::teardown_call.load(std::memory_order_relaxed) == 3);
+  assert((5 + 10 + 15) ==
+         concurrent::func_call.load(std::memory_order_relaxed));
+
+  // Setup is called 4 times, once for each arg group (1,3,5,7)
+  assert(fixture_interaction::setup == 4);
+  // Fixture::Setup is called every time the bm routine is run.
+  // The exact number is indeterministic, so we just assert that
+  // it's more than setup.
+  assert(fixture_interaction::fixture_setup > fixture_interaction::setup);
+
+  // Setup is call once for each repetition * num_arg =  4 * 4 = 16.
+  assert(repetitions::setup == 16);
+
+  return 0;
+}

test/benchmark_test.cc

@@ -0,0 +1,308 @@
+#include <assert.h>
+#include <math.h>
+#include <stdint.h>
+
+#include <chrono>
+#include <complex>
+#include <cstdlib>
+#include <iostream>
+#include <list>
+#include <mutex>
+#include <optional>
+#include <set>
+#include <sstream>
+#include <string>
+#include <thread>
+#include <type_traits>
+#include <utility>
+#include <vector>
+
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+#include "benchmark/utils.h"
+
+#if defined(__GNUC__)
+#define BENCHMARK_NOINLINE __attribute__((noinline))
+#else
+#define BENCHMARK_NOINLINE
+#endif
+
+namespace {
+
+int BENCHMARK_NOINLINE Factorial(int n) {
+  return (n == 1) ? 1 : n * Factorial(n - 1);
+}
+
+double CalculatePi(int depth) {
+  double pi = 0.0;
+  for (int i = 0; i < depth; ++i) {
+    double numerator = static_cast<double>(((i % 2) * 2) - 1);
+    double denominator = static_cast<double>((2 * i) - 1);
+    pi += numerator / denominator;
+  }
+  return (pi - 1.0) * 4;
+}
+
+std::set<int64_t> ConstructRandomSet(int64_t size) {
+  std::set<int64_t> s;
+  for (int i = 0; i < size; ++i) {
+    s.insert(s.end(), i);
+  }
+  return s;
+}
+
+// NOLINTBEGIN(cppcoreguidelines-avoid-non-const-global-variables)
+std::mutex test_vector_mu;
+std::optional<std::vector<int>> test_vector;
+// NOLINTEND(cppcoreguidelines-avoid-non-const-global-variables)
+
+void BM_Factorial(benchmark::State& state) {
+  int fac_42 = 0;
+  for (auto _ : state) {
+    fac_42 = Factorial(8);
+  }
+  // Prevent compiler optimizations
+  std::stringstream ss;
+  ss << fac_42;
+  state.SetLabel(ss.str());
+}
+BENCHMARK(BM_Factorial);
+BENCHMARK(BM_Factorial)->UseRealTime();
+
+void BM_CalculatePiRange(benchmark::State& state) {
+  double pi = 0.0;
+  for (auto _ : state) {
+    pi = CalculatePi(static_cast<int>(state.range(0)));
+  }
+  std::stringstream ss;
+  ss << pi;
+  state.SetLabel(ss.str());
+}
+BENCHMARK_RANGE(BM_CalculatePiRange, 1, 1024 * 1024);
+
+void BM_CalculatePi(benchmark::State& state) {
+  static const int depth = 1024;
+  for (auto _ : state) {
+    double pi = CalculatePi(static_cast<int>(depth));
+    benchmark::DoNotOptimize(pi);
+  }
+}
+BENCHMARK(BM_CalculatePi)->Threads(8);
+BENCHMARK(BM_CalculatePi)->ThreadRange(1, 32);
+BENCHMARK(BM_CalculatePi)->ThreadPerCpu();
+
+void BM_SetInsert(benchmark::State& state) {
+  std::set<int64_t> data;
+  for (auto _ : state) {
+    state.PauseTiming();
+    data = ConstructRandomSet(state.range(0));
+    state.ResumeTiming();
+    for (int j = 0; j < state.range(1); ++j) {
+      data.insert(rand());
+    }
+  }
+  state.SetItemsProcessed(state.iterations() * state.range(1));
+  state.SetBytesProcessed(state.iterations() * state.range(1) *
+                          static_cast<int64_t>(sizeof(int)));
+}
+
+// Test many inserts at once to reduce the total iterations needed. Otherwise,
+// the slower, non-timed part of each iteration will make the benchmark take
+// forever.
+BENCHMARK(BM_SetInsert)->Ranges({{1 << 10, 8 << 10}, {128, 512}});
+
+template <typename Container,
+          typename ValueType = typename Container::value_type>
+void BM_Sequential(benchmark::State& state) {
+  ValueType v = 42;
+  for (auto _ : state) {
+    Container c;
+    for (int64_t i = state.range(0); --i;) {
+      c.push_back(v);
+    }
+  }
+  const int64_t items_processed = state.iterations() * state.range(0);
+  state.SetItemsProcessed(items_processed);
+  state.SetBytesProcessed(items_processed * static_cast<int64_t>(sizeof(v)));
+}
+BENCHMARK_TEMPLATE2(BM_Sequential, std::vector<int>, int)
+    ->Range(1 << 0, 1 << 10);
+BENCHMARK_TEMPLATE(BM_Sequential, std::list<int>)->Range(1 << 0, 1 << 10);
+// Test the variadic version of BENCHMARK_TEMPLATE in C++11 and beyond.
+BENCHMARK_TEMPLATE(BM_Sequential, std::vector<int>, int)->Arg(512);
+
+void BM_StringCompare(benchmark::State& state) {
+  size_t len = static_cast<size_t>(state.range(0));
+  std::string s1(len, '-');
+  std::string s2(len, '-');
+  for (auto _ : state) {
+    auto comp = s1.compare(s2);
+    benchmark::DoNotOptimize(comp);
+  }
+}
+BENCHMARK(BM_StringCompare)->Range(1, 1 << 20);
+
+void BM_SetupTeardown(benchmark::State& state) {
+  if (state.thread_index() == 0) {
+    // No need to lock test_vector_mu here as this is running single-threaded.
+    test_vector = std::vector<int>();
+  }
+  int i = 0;
+  for (auto _ : state) {
+    std::lock_guard<std::mutex> l(test_vector_mu);
+    if (i % 2 == 0) {
+      test_vector->push_back(i);
+    } else {
+      test_vector->pop_back();
+    }
+    ++i;
+  }
+  if (state.thread_index() == 0) {
+    test_vector.reset();
+  }
+}
+BENCHMARK(BM_SetupTeardown)->ThreadPerCpu();
+
+void BM_LongTest(benchmark::State& state) {
+  double tracker = 0.0;
+  for (auto _ : state) {
+    for (int i = 0; i < state.range(0); ++i) {
+      benchmark::DoNotOptimize(tracker += i);
+    }
+  }
+}
+BENCHMARK(BM_LongTest)->Range(1 << 16, 1 << 28);
+
+void BM_ParallelMemset(benchmark::State& state) {
+  int64_t size = state.range(0) / static_cast<int64_t>(sizeof(int));
+  int thread_size = static_cast<int>(size) / state.threads();
+  int from = thread_size * state.thread_index();
+  int to = from + thread_size;
+
+  if (state.thread_index() == 0) {
+    test_vector = std::vector<int>(static_cast<size_t>(size));
+  }
+
+  for (auto _ : state) {
+    for (int i = from; i < to; i++) {
+      // No need to lock test_vector_mu as ranges
+      // do not overlap between threads.
+      benchmark::DoNotOptimize(test_vector->at(static_cast<size_t>(i)) = 1);
+    }
+  }
+
+  if (state.thread_index() == 0) {
+    test_vector.reset();
+  }
+}
+BENCHMARK(BM_ParallelMemset)->Arg(10 << 20)->ThreadRange(1, 4);
+
+void BM_ManualTiming(benchmark::State& state) {
+  int64_t slept_for = 0;
+  int64_t microseconds = state.range(0);
+  std::chrono::duration<double, std::micro> sleep_duration{
+      static_cast<double>(microseconds)};
+
+  for (auto _ : state) {
+    auto start = std::chrono::high_resolution_clock::now();
+    // Simulate some useful workload with a sleep
+    std::this_thread::sleep_for(
+        std::chrono::duration_cast<std::chrono::nanoseconds>(sleep_duration));
+    auto end = std::chrono::high_resolution_clock::now();
+
+    auto elapsed =
+        std::chrono::duration_cast<std::chrono::duration<double>>(end - start);
+
+    state.SetIterationTime(elapsed.count());
+    slept_for += microseconds;
+  }
+  state.SetItemsProcessed(slept_for);
+}
+BENCHMARK(BM_ManualTiming)->Range(1, 1 << 14)->UseRealTime();
+BENCHMARK(BM_ManualTiming)->Range(1, 1 << 14)->UseManualTime();
+
+template <class... Args>
+void BM_with_args(benchmark::State& state, Args&&...) {
+  for (auto _ : state) {
+  }
+}
+BENCHMARK_CAPTURE(BM_with_args, int_test, 42, 43, 44);
+BENCHMARK_CAPTURE(BM_with_args, string_and_pair_test, std::string("abc"),
+                  std::pair<int, double>(42, 3.8));
+
+void BM_non_template_args(benchmark::State& state, int, double) {
+  while (state.KeepRunning()) {
+  }
+}
+BENCHMARK_CAPTURE(BM_non_template_args, basic_test, 0, 0);
+
+template <class T, class U, class... ExtraArgs>
+void BM_template2_capture(benchmark::State& state, ExtraArgs&&... extra_args) {
+  static_assert(std::is_same<T, void>::value, "");
+  static_assert(std::is_same<U, char*>::value, "");
+  static_assert(std::is_same<ExtraArgs..., unsigned int>::value, "");
+  unsigned int dummy[sizeof...(ExtraArgs)] = {extra_args...};
+  assert(dummy[0] == 42);
+  for (auto _ : state) {
+  }
+}
+BENCHMARK_TEMPLATE2_CAPTURE(BM_template2_capture, void, char*, foo, 42U);
+BENCHMARK_CAPTURE((BM_template2_capture<void, char*>), foo, 42U);
+
+template <class T, class... ExtraArgs>
+void BM_template1_capture(benchmark::State& state, ExtraArgs&&... extra_args) {
+  static_assert(std::is_same<T, void>::value, "");
+  static_assert(std::is_same<ExtraArgs..., unsigned long>::value, "");
+  unsigned long dummy[sizeof...(ExtraArgs)] = {extra_args...};
+  assert(dummy[0] == 24);
+  for (auto _ : state) {
+  }
+}
+BENCHMARK_TEMPLATE1_CAPTURE(BM_template1_capture, void, foo, 24UL);
+BENCHMARK_CAPTURE(BM_template1_capture<void>, foo, 24UL);
+
+void BM_DenseThreadRanges(benchmark::State& st) {
+  switch (st.range(0)) {
+    case 1:
+      assert(st.threads() == 1 || st.threads() == 2 || st.threads() == 3);
+      break;
+    case 2:
+      assert(st.threads() == 1 || st.threads() == 3 || st.threads() == 4);
+      break;
+    case 3:
+      assert(st.threads() == 5 || st.threads() == 8 || st.threads() == 11 ||
+             st.threads() == 14);
+      break;
+    default:
+      assert(false && "Invalid test case number");
+  }
+  while (st.KeepRunning()) {
+  }
+}
+BENCHMARK(BM_DenseThreadRanges)->Arg(1)->DenseThreadRange(1, 3);
+BENCHMARK(BM_DenseThreadRanges)->Arg(2)->DenseThreadRange(1, 4, 2);
+BENCHMARK(BM_DenseThreadRanges)->Arg(3)->DenseThreadRange(5, 14, 3);
+
+void BM_BenchmarkName(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+
+  // Check that the benchmark name is passed correctly to `state`.
+  assert("BM_BenchmarkName" == state.name());
+}
+BENCHMARK(BM_BenchmarkName);
+
+// regression test for #1446
+template <typename type>
+void BM_templated_test(benchmark::State& state) {
+  for (auto _ : state) {
+    type created_string;
+    benchmark::DoNotOptimize(created_string);
+  }
+}
+
+const auto BM_templated_test_double = BM_templated_test<std::complex<double>>;
+BENCHMARK(BM_templated_test_double);
+}  // end namespace
+
+BENCHMARK_MAIN();

test/clobber_memory_assembly_test.cc

@@ -0,0 +1,65 @@
+#include "benchmark/macros.h"
+#include "benchmark/utils.h"
+
+#ifdef __clang__
+#pragma clang diagnostic ignored "-Wreturn-type"
+#endif
+BENCHMARK_DISABLE_DEPRECATED_WARNING
+
+extern "C" {
+
+extern int ExternInt;
+extern int ExternInt2;
+extern int ExternInt3;
+}
+
+// CHECK-LABEL: test_basic:
+extern "C" void test_basic() {
+  int x;
+  benchmark::DoNotOptimize(&x);
+  x = 101;
+  benchmark::ClobberMemory();
+  // CHECK: leaq [[DEST:[^,]+]], %rax
+  // CHECK: movl $101, [[DEST]]
+  // CHECK: ret
+}
+
+// CHECK-LABEL: test_redundant_store:
+extern "C" void test_redundant_store() {
+  ExternInt = 3;
+  benchmark::ClobberMemory();
+  ExternInt = 51;
+  // CHECK-DAG: ExternInt
+  // CHECK-DAG: movl $3
+  // CHECK: movl $51
+}
+
+// CHECK-LABEL: test_redundant_read:
+extern "C" void test_redundant_read() {
+  int x;
+  benchmark::DoNotOptimize(&x);
+  x = ExternInt;
+  benchmark::ClobberMemory();
+  x = ExternInt2;
+  // CHECK: leaq [[DEST:[^,]+]], %rax
+  // CHECK: ExternInt(%rip)
+  // CHECK: movl %eax, [[DEST]]
+  // CHECK-NOT: ExternInt2
+  // CHECK: ret
+}
+
+// CHECK-LABEL: test_redundant_read2:
+extern "C" void test_redundant_read2() {
+  int x;
+  benchmark::DoNotOptimize(&x);
+  x = ExternInt;
+  benchmark::ClobberMemory();
+  x = ExternInt2;
+  benchmark::ClobberMemory();
+  // CHECK: leaq [[DEST:[^,]+]], %rax
+  // CHECK: ExternInt(%rip)
+  // CHECK: movl %eax, [[DEST]]
+  // CHECK: ExternInt2(%rip)
+  // CHECK: movl %eax, [[DEST]]
+  // CHECK: ret
+}

test/commandlineflags_gtest.cc

@@ -0,0 +1,228 @@
+#include <cstdlib>
+
+#include "../src/commandlineflags.h"
+#include "../src/internal_macros.h"
+#include "gmock/gmock.h"
+#include "gtest/gtest.h"
+
+namespace benchmark {
+namespace {
+
+#if defined(BENCHMARK_OS_WINDOWS)
+int setenv(const char* name, const char* value, int overwrite) {
+  if (!overwrite) {
+    // NOTE: getenv_s is far superior but not available under mingw.
+    char* env_value = getenv(name);
+    if (env_value == nullptr) {
+      return -1;
+    }
+  }
+  return _putenv_s(name, value);
+}
+
+int unsetenv(const char* name) { return _putenv_s(name, ""); }
+
+#endif  // BENCHMARK_OS_WINDOWS
+
+TEST(BoolFromEnv, Default) {
+  ASSERT_EQ(unsetenv("NOT_IN_ENV"), 0);
+  EXPECT_EQ(BoolFromEnv("not_in_env", true), true);
+}
+
+TEST(BoolFromEnv, False) {
+  ASSERT_EQ(setenv("IN_ENV", "0", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", true), false);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "N", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", true), false);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "n", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", true), false);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "NO", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", true), false);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "No", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", true), false);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "no", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", true), false);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "F", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", true), false);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "f", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", true), false);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "FALSE", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", true), false);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "False", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", true), false);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "false", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", true), false);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "OFF", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", true), false);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "Off", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", true), false);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "off", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", true), false);
+  unsetenv("IN_ENV");
+}
+
+TEST(BoolFromEnv, True) {
+  ASSERT_EQ(setenv("IN_ENV", "1", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", false), true);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "Y", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", false), true);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "y", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", false), true);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "YES", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", false), true);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "Yes", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", false), true);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "yes", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", false), true);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "T", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", false), true);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "t", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", false), true);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "TRUE", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", false), true);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "True", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", false), true);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "true", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", false), true);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "ON", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", false), true);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "On", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", false), true);
+  unsetenv("IN_ENV");
+
+  ASSERT_EQ(setenv("IN_ENV", "on", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", false), true);
+  unsetenv("IN_ENV");
+
+#ifndef BENCHMARK_OS_WINDOWS
+  ASSERT_EQ(setenv("IN_ENV", "", 1), 0);
+  EXPECT_EQ(BoolFromEnv("in_env", false), true);
+  unsetenv("IN_ENV");
+#endif
+}
+
+TEST(Int32FromEnv, NotInEnv) {
+  ASSERT_EQ(unsetenv("NOT_IN_ENV"), 0);
+  EXPECT_EQ(Int32FromEnv("not_in_env", 42), 42);
+}
+
+TEST(Int32FromEnv, InvalidInteger) {
+  ASSERT_EQ(setenv("IN_ENV", "foo", 1), 0);
+  EXPECT_EQ(Int32FromEnv("in_env", 42), 42);
+  unsetenv("IN_ENV");
+}
+
+TEST(Int32FromEnv, ValidInteger) {
+  ASSERT_EQ(setenv("IN_ENV", "42", 1), 0);
+  EXPECT_EQ(Int32FromEnv("in_env", 64), 42);
+  unsetenv("IN_ENV");
+}
+
+TEST(DoubleFromEnv, NotInEnv) {
+  ASSERT_EQ(unsetenv("NOT_IN_ENV"), 0);
+  EXPECT_EQ(DoubleFromEnv("not_in_env", 0.51), 0.51);
+}
+
+TEST(DoubleFromEnv, InvalidReal) {
+  ASSERT_EQ(setenv("IN_ENV", "foo", 1), 0);
+  EXPECT_EQ(DoubleFromEnv("in_env", 0.51), 0.51);
+  unsetenv("IN_ENV");
+}
+
+TEST(DoubleFromEnv, ValidReal) {
+  ASSERT_EQ(setenv("IN_ENV", "0.51", 1), 0);
+  EXPECT_EQ(DoubleFromEnv("in_env", 0.71), 0.51);
+  unsetenv("IN_ENV");
+}
+
+TEST(StringFromEnv, Default) {
+  ASSERT_EQ(unsetenv("NOT_IN_ENV"), 0);
+  EXPECT_STREQ(StringFromEnv("not_in_env", "foo"), "foo");
+}
+
+TEST(StringFromEnv, Valid) {
+  ASSERT_EQ(setenv("IN_ENV", "foo", 1), 0);
+  EXPECT_STREQ(StringFromEnv("in_env", "bar"), "foo");
+  unsetenv("IN_ENV");
+}
+
+TEST(KvPairsFromEnv, Default) {
+  ASSERT_EQ(unsetenv("NOT_IN_ENV"), 0);
+  EXPECT_THAT(KvPairsFromEnv("not_in_env", {{"foo", "bar"}}),
+              testing::ElementsAre(testing::Pair("foo", "bar")));
+}
+
+TEST(KvPairsFromEnv, MalformedReturnsDefault) {
+  ASSERT_EQ(setenv("IN_ENV", "foo", 1), 0);
+  EXPECT_THAT(KvPairsFromEnv("in_env", {{"foo", "bar"}}),
+              testing::ElementsAre(testing::Pair("foo", "bar")));
+  unsetenv("IN_ENV");
+}
+
+TEST(KvPairsFromEnv, Single) {
+  ASSERT_EQ(setenv("IN_ENV", "foo=bar", 1), 0);
+  EXPECT_THAT(KvPairsFromEnv("in_env", {}),
+              testing::ElementsAre(testing::Pair("foo", "bar")));
+  unsetenv("IN_ENV");
+}
+
+TEST(KvPairsFromEnv, Multiple) {
+  ASSERT_EQ(setenv("IN_ENV", "foo=bar,baz=qux", 1), 0);
+  EXPECT_THAT(KvPairsFromEnv("in_env", {}),
+              testing::UnorderedElementsAre(testing::Pair("foo", "bar"),
+                                            testing::Pair("baz", "qux")));
+  unsetenv("IN_ENV");
+}
+
+}  // namespace
+}  // namespace benchmark

test/complexity_test.cc

@@ -0,0 +1,278 @@
+#undef NDEBUG
+#include <cassert>
+#include <cmath>
+#include <cstdlib>
+#include <vector>
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+#include "benchmark/statistics.h"
+#include "benchmark/types.h"
+#include "benchmark/utils.h"
+#include "output_test.h"
+
+namespace {
+
+#define ADD_COMPLEXITY_CASES(...) \
+  const int CONCAT(dummy, __LINE__) = AddComplexityTest(__VA_ARGS__)
+
+int AddComplexityTest(const std::string& test_name,
+                      const std::string& big_o_test_name,
+                      const std::string& rms_test_name,
+                      const std::string& big_o, int family_index) {
+  SetSubstitutions({{"%name", test_name},
+                    {"%bigo_name", big_o_test_name},
+                    {"%rms_name", rms_test_name},
+                    {"%bigo_str", "[ ]* %float " + big_o},
+                    {"%bigo", big_o},
+                    {"%rms", "[ ]*[0-9]+ %"}});
+  AddCases(
+      TC_ConsoleOut,
+      {{"^%bigo_name %bigo_str %bigo_str[ ]*$"},
+       {"^%bigo_name", MR_Not},  // Assert we we didn't only matched a name.
+       {"^%rms_name %rms %rms[ ]*$", MR_Next}});
+  AddCases(
+      TC_JSONOut,
+      {{"\"name\": \"%bigo_name\",$"},
+       {"\"family_index\": " + std::to_string(family_index) + ",$", MR_Next},
+       {"\"per_family_instance_index\": 0,$", MR_Next},
+       {"\"run_name\": \"%name\",$", MR_Next},
+       {"\"run_type\": \"aggregate\",$", MR_Next},
+       {"\"repetitions\": %int,$", MR_Next},
+       {"\"threads\": 1,$", MR_Next},
+       {"\"aggregate_name\": \"BigO\",$", MR_Next},
+       {"\"aggregate_unit\": \"time\",$", MR_Next},
+       {"\"cpu_coefficient\": %float,$", MR_Next},
+       {"\"real_coefficient\": %float,$", MR_Next},
+       {"\"big_o\": \"%bigo\",$", MR_Next},
+       {"\"time_unit\": \"ns\"$", MR_Next},
+       {"}", MR_Next},
+       {"\"name\": \"%rms_name\",$"},
+       {"\"family_index\": " + std::to_string(family_index) + ",$", MR_Next},
+       {"\"per_family_instance_index\": 0,$", MR_Next},
+       {"\"run_name\": \"%name\",$", MR_Next},
+       {"\"run_type\": \"aggregate\",$", MR_Next},
+       {"\"repetitions\": %int,$", MR_Next},
+       {"\"threads\": 1,$", MR_Next},
+       {"\"aggregate_name\": \"RMS\",$", MR_Next},
+       {"\"aggregate_unit\": \"percentage\",$", MR_Next},
+       {"\"rms\": %float$", MR_Next},
+       {"}", MR_Next}});
+  AddCases(TC_CSVOut, {{"^\"%bigo_name\",,%float,%float,%bigo,,,,,$"},
+                       {"^\"%bigo_name\"", MR_Not},
+                       {"^\"%rms_name\",,%float,%float,,,,,,$", MR_Next}});
+  return 0;
+}
+
+// ========================================================================= //
+// --------------------------- Testing BigO O(1) --------------------------- //
+// ========================================================================= //
+
+void BM_Complexity_O1(benchmark::State& state) {
+  for (auto _ : state) {
+    // This test requires a non-zero CPU time to avoid divide-by-zero
+    benchmark::DoNotOptimize(state.iterations());
+    double tmp = static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(tmp);
+    for (benchmark::IterationCount i = 0; i < state.iterations(); ++i) {
+      benchmark::DoNotOptimize(state.iterations());
+      tmp *= static_cast<double>(state.iterations());
+      benchmark::DoNotOptimize(tmp);
+    }
+
+    // always 1ns per iteration
+    state.SetIterationTime(42 * 1e-9);
+  }
+  state.SetComplexityN(state.range(0));
+}
+BENCHMARK(BM_Complexity_O1)
+    ->Range(1, 1 << 18)
+    ->UseManualTime()
+    ->Complexity(benchmark::o1);
+BENCHMARK(BM_Complexity_O1)->Range(1, 1 << 18)->UseManualTime()->Complexity();
+BENCHMARK(BM_Complexity_O1)
+    ->Range(1, 1 << 18)
+    ->UseManualTime()
+    ->Complexity([](benchmark::IterationCount) { return 1.0; });
+
+constexpr char one_test_name[] = "BM_Complexity_O1/manual_time";
+constexpr char big_o_1_test_name[] = "BM_Complexity_O1/manual_time_BigO";
+constexpr char rms_o_1_test_name[] = "BM_Complexity_O1/manual_time_RMS";
+constexpr char enum_auto_big_o_1[] = "\\([0-9]+\\)";
+constexpr char lambda_big_o_1[] = "f\\(N\\)";
+
+// Add enum tests
+ADD_COMPLEXITY_CASES(one_test_name, big_o_1_test_name, rms_o_1_test_name,
+                     enum_auto_big_o_1, /*family_index=*/0);
+
+// Add auto tests
+ADD_COMPLEXITY_CASES(one_test_name, big_o_1_test_name, rms_o_1_test_name,
+                     enum_auto_big_o_1, /*family_index=*/1);
+
+// Add lambda tests
+ADD_COMPLEXITY_CASES(one_test_name, big_o_1_test_name, rms_o_1_test_name,
+                     lambda_big_o_1, /*family_index=*/2);
+
+// ========================================================================= //
+// --------------------------- Testing BigO O(N) --------------------------- //
+// ========================================================================= //
+
+void BM_Complexity_O_N(benchmark::State& state) {
+  for (auto _ : state) {
+    // This test requires a non-zero CPU time to avoid divide-by-zero
+    benchmark::DoNotOptimize(state.iterations());
+    double tmp = static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(tmp);
+    for (benchmark::IterationCount i = 0; i < state.iterations(); ++i) {
+      benchmark::DoNotOptimize(state.iterations());
+      tmp *= static_cast<double>(state.iterations());
+      benchmark::DoNotOptimize(tmp);
+    }
+
+    // 1ns per iteration per entry
+    state.SetIterationTime(static_cast<double>(state.range(0)) * 42 * 1e-9);
+  }
+  state.SetComplexityN(state.range(0));
+}
+BENCHMARK(BM_Complexity_O_N)
+    ->RangeMultiplier(2)
+    ->Range(1 << 10, 1 << 20)
+    ->UseManualTime()
+    ->Complexity(benchmark::oN);
+BENCHMARK(BM_Complexity_O_N)
+    ->RangeMultiplier(2)
+    ->Range(1 << 10, 1 << 20)
+    ->UseManualTime()
+    ->Complexity();
+BENCHMARK(BM_Complexity_O_N)
+    ->RangeMultiplier(2)
+    ->Range(1 << 10, 1 << 20)
+    ->UseManualTime()
+    ->Complexity([](benchmark::IterationCount n) -> double {
+      return static_cast<double>(n);
+    });
+
+constexpr char n_test_name[] = "BM_Complexity_O_N/manual_time";
+constexpr char big_o_n_test_name[] = "BM_Complexity_O_N/manual_time_BigO";
+constexpr char rms_o_n_test_name[] = "BM_Complexity_O_N/manual_time_RMS";
+constexpr char enum_auto_big_o_n[] = "N";
+constexpr char lambda_big_o_n[] = "f\\(N\\)";
+
+// Add enum tests
+ADD_COMPLEXITY_CASES(n_test_name, big_o_n_test_name, rms_o_n_test_name,
+                     enum_auto_big_o_n, /*family_index=*/3);
+
+// Add auto tests
+ADD_COMPLEXITY_CASES(n_test_name, big_o_n_test_name, rms_o_n_test_name,
+                     enum_auto_big_o_n, /*family_index=*/4);
+
+// Add lambda tests
+ADD_COMPLEXITY_CASES(n_test_name, big_o_n_test_name, rms_o_n_test_name,
+                     lambda_big_o_n, /*family_index=*/5);
+
+// ========================================================================= //
+// ------------------------- Testing BigO O(NlgN) ------------------------- //
+// ========================================================================= //
+
+const double kLog2E = 1.44269504088896340736;
+void BM_Complexity_O_N_log_N(benchmark::State& state) {
+  for (auto _ : state) {
+    // This test requires a non-zero CPU time to avoid divide-by-zero
+    benchmark::DoNotOptimize(state.iterations());
+    double tmp = static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(tmp);
+    for (benchmark::IterationCount i = 0; i < state.iterations(); ++i) {
+      benchmark::DoNotOptimize(state.iterations());
+      tmp *= static_cast<double>(state.iterations());
+      benchmark::DoNotOptimize(tmp);
+    }
+
+    state.SetIterationTime(static_cast<double>(state.range(0)) * kLog2E *
+                           std::log(state.range(0)) * 42 * 1e-9);
+  }
+  state.SetComplexityN(state.range(0));
+}
+BENCHMARK(BM_Complexity_O_N_log_N)
+    ->RangeMultiplier(2)
+    ->Range(1 << 10, 1U << 24)
+    ->UseManualTime()
+    ->Complexity(benchmark::oNLogN);
+BENCHMARK(BM_Complexity_O_N_log_N)
+    ->RangeMultiplier(2)
+    ->Range(1 << 10, 1U << 24)
+    ->UseManualTime()
+    ->Complexity();
+BENCHMARK(BM_Complexity_O_N_log_N)
+    ->RangeMultiplier(2)
+    ->Range(1 << 10, 1U << 24)
+    ->UseManualTime()
+    ->Complexity([](benchmark::IterationCount n) {
+      return kLog2E * static_cast<double>(n) * std::log(static_cast<double>(n));
+    });
+
+constexpr char n_lg_n_test_name[] = "BM_Complexity_O_N_log_N/manual_time";
+constexpr char big_o_n_lg_n_test_name[] =
+    "BM_Complexity_O_N_log_N/manual_time_BigO";
+constexpr char rms_o_n_lg_n_test_name[] =
+    "BM_Complexity_O_N_log_N/manual_time_RMS";
+constexpr char enum_auto_big_o_n_lg_n[] = "NlgN";
+constexpr char lambda_big_o_n_lg_n[] = "f\\(N\\)";
+
+// Add enum tests
+ADD_COMPLEXITY_CASES(n_lg_n_test_name, big_o_n_lg_n_test_name,
+                     rms_o_n_lg_n_test_name, enum_auto_big_o_n_lg_n,
+                     /*family_index=*/6);
+
+// NOTE: auto big-o is wron.g
+ADD_COMPLEXITY_CASES(n_lg_n_test_name, big_o_n_lg_n_test_name,
+                     rms_o_n_lg_n_test_name, enum_auto_big_o_n_lg_n,
+                     /*family_index=*/7);
+
+//// Add lambda tests
+ADD_COMPLEXITY_CASES(n_lg_n_test_name, big_o_n_lg_n_test_name,
+                     rms_o_n_lg_n_test_name, lambda_big_o_n_lg_n,
+                     /*family_index=*/8);
+
+// ========================================================================= //
+// -------- Testing formatting of Complexity with captured args ------------ //
+// ========================================================================= //
+
+void BM_ComplexityCaptureArgs(benchmark::State& state, int n) {
+  for (auto _ : state) {
+    // This test requires a non-zero CPU time to avoid divide-by-zero
+    benchmark::DoNotOptimize(state.iterations());
+    double tmp = static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(tmp);
+    for (benchmark::IterationCount i = 0; i < state.iterations(); ++i) {
+      benchmark::DoNotOptimize(state.iterations());
+      tmp *= static_cast<double>(state.iterations());
+      benchmark::DoNotOptimize(tmp);
+    }
+
+    state.SetIterationTime(static_cast<double>(state.range(0)) * 42 * 1e-9);
+  }
+  state.SetComplexityN(n);
+}
+
+BENCHMARK_CAPTURE(BM_ComplexityCaptureArgs, capture_test, 100)
+    ->UseManualTime()
+    ->Complexity(benchmark::oN)
+    ->Ranges({{1, 2}, {3, 4}});
+
+const std::string complexity_capture_name =
+    "BM_ComplexityCaptureArgs/capture_test/manual_time";
+
+ADD_COMPLEXITY_CASES(complexity_capture_name, complexity_capture_name + "_BigO",
+                     complexity_capture_name + "_RMS", "N",
+                     /*family_index=*/9);
+}  // end namespace
+
+// ========================================================================= //
+// --------------------------- TEST CASES END ------------------------------ //
+// ========================================================================= //
+
+int main(int argc, char* argv[]) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+  RunOutputTests(argc, argv);
+}

test/cxx11_test.cc

@@ -0,0 +1,12 @@
+#include "benchmark/benchmark_api.h"
+
+#if defined(_MSC_VER)
+#if _MSVC_LANG != 201402L
+// MSVC, even in C++11 mode, dooes not claim to be in C++11 mode.
+#error "Trying to compile C++11 test with wrong C++ standard"
+#endif  //  _MSVC_LANG
+#else   // Non-MSVC
+#if __cplusplus != 201103L
+#error "Trying to compile C++11 test with wrong C++ standard"
+#endif  // Non-MSVC
+#endif

test/diagnostics_test.cc

@@ -0,0 +1,107 @@
+// Testing:
+//   State::PauseTiming()
+//   State::ResumeTiming()
+// Test that CHECK's within these function diagnose when they are called
+// outside of the KeepRunning() loop.
+//
+// NOTE: Users should NOT include or use src/check.h. This is only done in
+// order to test library internals.
+
+#include <cstdlib>
+#include <stdexcept>
+
+#include "../src/check.h"
+#include "benchmark/benchmark_api.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+#include "benchmark/utils.h"
+
+#if defined(__GNUC__) && !defined(__EXCEPTIONS)
+#define TEST_HAS_NO_EXCEPTIONS
+#endif
+
+namespace {
+void TestHandler() {
+#ifndef TEST_HAS_NO_EXCEPTIONS
+  throw std::logic_error("");
+#else
+  std::abort();
+#endif
+}
+
+void try_invalid_pause_resume(benchmark::State& state) {
+#if !defined(TEST_BENCHMARK_LIBRARY_HAS_NO_ASSERTIONS) && \
+    !defined(TEST_HAS_NO_EXCEPTIONS)
+  try {
+    state.PauseTiming();
+    std::abort();
+  } catch (std::logic_error const&) {
+  }
+  try {
+    state.ResumeTiming();
+    std::abort();
+  } catch (std::logic_error const&) {
+  }
+#else
+  (void)state;  // avoid unused warning
+#endif
+}
+
+void BM_diagnostic_test(benchmark::State& state) {
+  static bool called_once = false;
+
+  if (!called_once) {
+    try_invalid_pause_resume(state);
+  }
+
+  for (auto _ : state) {
+    auto iterations = static_cast<double>(state.iterations()) *
+                      static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(iterations);
+  }
+
+  if (!called_once) {
+    try_invalid_pause_resume(state);
+  }
+
+  called_once = true;
+}
+BENCHMARK(BM_diagnostic_test);
+
+void BM_diagnostic_test_keep_running(benchmark::State& state) {
+  static bool called_once = false;
+
+  if (!called_once) {
+    try_invalid_pause_resume(state);
+  }
+
+  while (state.KeepRunning()) {
+    auto iterations = static_cast<double>(state.iterations()) *
+                      static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(iterations);
+  }
+
+  if (!called_once) {
+    try_invalid_pause_resume(state);
+  }
+
+  called_once = true;
+}
+BENCHMARK(BM_diagnostic_test_keep_running);
+}  // end namespace
+
+int main(int argc, char* argv[]) {
+#ifdef NDEBUG
+  // This test is exercising functionality for debug builds, which are not
+  // available in release builds. Skip the test if we are in that environment
+  // to avoid a test failure.
+  std::cout << "Diagnostic test disabled in release build\n";
+  (void)argc;
+  (void)argv;
+#else
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+  benchmark::internal::GetAbortHandler() = &TestHandler;
+  benchmark::Initialize(&argc, argv);
+  benchmark::RunSpecifiedBenchmarks();
+#endif
+}

test/display_aggregates_only_test.cc

@@ -0,0 +1,51 @@
+
+#undef NDEBUG
+#include <cstdio>
+#include <string>
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+#include "output_test.h"
+
+// Ok this test is super ugly. We want to check what happens with the file
+// reporter in the presence of DisplayAggregatesOnly().
+// We do not care about console output, the normal tests check that already.
+
+namespace {
+void BM_SummaryRepeat(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+}
+BENCHMARK(BM_SummaryRepeat)->Repetitions(3)->DisplayAggregatesOnly();
+}  // end namespace
+
+int main(int argc, char* argv[]) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+
+  const std::string output = GetFileReporterOutput(argc, argv);
+
+  if (SubstrCnt(output, "\"name\": \"BM_SummaryRepeat/repeats:3") != 7 ||
+      SubstrCnt(output, "\"name\": \"BM_SummaryRepeat/repeats:3\"") != 3 ||
+      SubstrCnt(output, "\"name\": \"BM_SummaryRepeat/repeats:3_mean\"") != 1 ||
+      SubstrCnt(output, "\"name\": \"BM_SummaryRepeat/repeats:3_median\"") !=
+          1 ||
+      SubstrCnt(output, "\"name\": \"BM_SummaryRepeat/repeats:3_stddev\"") !=
+          1 ||
+      SubstrCnt(output, "\"name\": \"BM_SummaryRepeat/repeats:3_cv\"") != 1) {
+    std::cout << "Precondition mismatch. Expected to only find 8 "
+                 "occurrences of \"BM_SummaryRepeat/repeats:3\" substring:\n"
+                 "\"name\": \"BM_SummaryRepeat/repeats:3\", "
+                 "\"name\": \"BM_SummaryRepeat/repeats:3\", "
+                 "\"name\": \"BM_SummaryRepeat/repeats:3\", "
+                 "\"name\": \"BM_SummaryRepeat/repeats:3_mean\", "
+                 "\"name\": \"BM_SummaryRepeat/repeats:3_median\", "
+                 "\"name\": \"BM_SummaryRepeat/repeats:3_stddev\", "
+                 "\"name\": \"BM_SummaryRepeat/repeats:3_cv\"\nThe entire "
+                 "output:\n";
+    std::cout << output;
+    return 1;
+  }
+
+  return 0;
+}

test/donotoptimize_assembly_test.cc

@@ -0,0 +1,203 @@
+#include "benchmark/macros.h"
+#include "benchmark/utils.h"
+
+#ifdef __clang__
+#pragma clang diagnostic ignored "-Wreturn-type"
+#pragma clang diagnostic ignored "-Wmissing-prototypes"
+#endif
+BENCHMARK_DISABLE_DEPRECATED_WARNING
+
+extern "C" {
+
+extern int ExternInt;
+extern int ExternInt2;
+extern int ExternInt3;
+extern int BigArray[2049];
+
+const int ConstBigArray[2049]{};
+
+inline int Add42(int x) { return x + 42; }
+
+struct NotTriviallyCopyable {
+  NotTriviallyCopyable();
+  explicit NotTriviallyCopyable(int x) : value(x) {}
+  NotTriviallyCopyable(NotTriviallyCopyable const&);
+  int value;
+};
+
+struct Large {
+  int value;
+  int data[2];
+};
+
+struct ExtraLarge {
+  int arr[2049];
+};
+}
+
+extern ExtraLarge ExtraLargeObj;
+const ExtraLarge ConstExtraLargeObj{};
+
+// CHECK-LABEL: test_with_rvalue:
+extern "C" void test_with_rvalue() {
+  benchmark::DoNotOptimize(Add42(0));
+  // CHECK: movl $42, %eax
+  // CHECK: ret
+}
+
+// CHECK-LABEL: test_with_large_rvalue:
+extern "C" void test_with_large_rvalue() {
+  benchmark::DoNotOptimize(Large{ExternInt, {ExternInt, ExternInt}});
+  // CHECK: ExternInt(%rip)
+  // CHECK: movl %eax, -{{[0-9]+}}(%[[REG:[a-z]+]]
+  // CHECK: movl %eax, -{{[0-9]+}}(%[[REG]])
+  // CHECK: movl %eax, -{{[0-9]+}}(%[[REG]])
+  // CHECK: ret
+}
+
+// CHECK-LABEL: test_with_non_trivial_rvalue:
+extern "C" void test_with_non_trivial_rvalue() {
+  benchmark::DoNotOptimize(NotTriviallyCopyable(ExternInt));
+  // CHECK: mov{{l|q}} ExternInt(%rip)
+  // CHECK: ret
+}
+
+// CHECK-LABEL: test_with_lvalue:
+extern "C" void test_with_lvalue() {
+  int x = 101;
+  benchmark::DoNotOptimize(x);
+  // CHECK-GNU: movl $101, %eax
+  // CHECK-CLANG: movl $101, -{{[0-9]+}}(%[[REG:[a-z]+]])
+  // CHECK: ret
+}
+
+// CHECK-LABEL: test_with_large_lvalue:
+extern "C" void test_with_large_lvalue() {
+  Large L{ExternInt, {ExternInt, ExternInt}};
+  benchmark::DoNotOptimize(L);
+  // CHECK: ExternInt(%rip)
+  // CHECK: movl %eax, -{{[0-9]+}}(%[[REG:[a-z]+]])
+  // CHECK: movl %eax, -{{[0-9]+}}(%[[REG]])
+  // CHECK: movl %eax, -{{[0-9]+}}(%[[REG]])
+  // CHECK: ret
+}
+
+// CHECK-LABEL: test_with_extra_large_lvalue_with_op:
+extern "C" void test_with_extra_large_lvalue_with_op() {
+  ExtraLargeObj.arr[16] = 42;
+  benchmark::DoNotOptimize(ExtraLargeObj);
+  // CHECK: movl $42, ExtraLargeObj+64(%rip)
+  // CHECK: ret
+}
+
+// CHECK-LABEL: test_with_big_array_with_op
+extern "C" void test_with_big_array_with_op() {
+  BigArray[16] = 42;
+  benchmark::DoNotOptimize(BigArray);
+  // CHECK: movl $42, BigArray+64(%rip)
+  // CHECK: ret
+}
+
+// CHECK-LABEL: test_with_non_trivial_lvalue:
+extern "C" void test_with_non_trivial_lvalue() {
+  NotTriviallyCopyable NTC(ExternInt);
+  benchmark::DoNotOptimize(NTC);
+  // CHECK: ExternInt(%rip)
+  // CHECK: movl %eax, -{{[0-9]+}}(%[[REG:[a-z]+]])
+  // CHECK: ret
+}
+
+// CHECK-LABEL: test_with_const_lvalue:
+extern "C" void test_with_const_lvalue() {
+  const int x = 123;
+  benchmark::DoNotOptimize(x);
+  // CHECK: movl $123, %eax
+  // CHECK: ret
+}
+
+// CHECK-LABEL: test_with_large_const_lvalue:
+extern "C" void test_with_large_const_lvalue() {
+  const Large L{ExternInt, {ExternInt, ExternInt}};
+  benchmark::DoNotOptimize(L);
+  // CHECK: ExternInt(%rip)
+  // CHECK: movl %eax, -{{[0-9]+}}(%[[REG:[a-z]+]])
+  // CHECK: movl %eax, -{{[0-9]+}}(%[[REG]])
+  // CHECK: movl %eax, -{{[0-9]+}}(%[[REG]])
+  // CHECK: ret
+}
+
+// CHECK-LABEL: test_with_const_extra_large_obj:
+extern "C" void test_with_const_extra_large_obj() {
+  benchmark::DoNotOptimize(ConstExtraLargeObj);
+  // CHECK: ret
+}
+
+// CHECK-LABEL: test_with_const_big_array
+extern "C" void test_with_const_big_array() {
+  benchmark::DoNotOptimize(ConstBigArray);
+  // CHECK: ret
+}
+
+// CHECK-LABEL: test_with_non_trivial_const_lvalue:
+extern "C" void test_with_non_trivial_const_lvalue() {
+  const NotTriviallyCopyable Obj(ExternInt);
+  benchmark::DoNotOptimize(Obj);
+  // CHECK: mov{{q|l}} ExternInt(%rip)
+  // CHECK: ret
+}
+
+// CHECK-LABEL: test_div_by_two:
+extern "C" int test_div_by_two(int input) {
+  int divisor = 2;
+  benchmark::DoNotOptimize(divisor);
+  return input / divisor;
+  // CHECK: movl $2, [[DEST:.*]]
+  // CHECK: idivl [[DEST]]
+  // CHECK: ret
+}
+
+// CHECK-LABEL: test_inc_integer:
+extern "C" int test_inc_integer() {
+  int x = 0;
+  for (int i = 0; i < 5; ++i) benchmark::DoNotOptimize(++x);
+  // CHECK: movl $1, [[DEST:.*]]
+  // CHECK: {{(addl \$1,|incl)}} [[DEST]]
+  // CHECK: {{(addl \$1,|incl)}} [[DEST]]
+  // CHECK: {{(addl \$1,|incl)}} [[DEST]]
+  // CHECK: {{(addl \$1,|incl)}} [[DEST]]
+  // CHECK-CLANG: movl [[DEST]], %eax
+  // CHECK: ret
+  return x;
+}
+
+// CHECK-LABEL: test_pointer_rvalue
+extern "C" void test_pointer_rvalue() {
+  // CHECK: movl $42, [[DEST:.*]]
+  // CHECK: leaq [[DEST]], %rax
+  // CHECK-CLANG: movq %rax, -{{[0-9]+}}(%[[REG:[a-z]+]])
+  // CHECK: ret
+  int x = 42;
+  benchmark::DoNotOptimize(&x);
+}
+
+// CHECK-LABEL: test_pointer_const_lvalue:
+extern "C" void test_pointer_const_lvalue() {
+  // CHECK: movl $42, [[DEST:.*]]
+  // CHECK: leaq [[DEST]], %rax
+  // CHECK-CLANG: movq %rax, -{{[0-9]+}}(%[[REG:[a-z]+]])
+  // CHECK: ret
+  int x = 42;
+  int* const xp = &x;
+  benchmark::DoNotOptimize(xp);
+}
+
+// CHECK-LABEL: test_pointer_lvalue:
+extern "C" void test_pointer_lvalue() {
+  // CHECK: movl $42, [[DEST:.*]]
+  // CHECK: leaq [[DEST]], %rax
+  // CHECK-CLANG: movq %rax, -{{[0-9]+}}(%[[REG:[a-z+]+]])
+  // CHECK: ret
+  int x = 42;
+  int* xp = &x;
+  benchmark::DoNotOptimize(xp);
+}

test/donotoptimize_test.cc

@@ -0,0 +1,70 @@
+#include <cstdint>
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/utils.h"
+
+namespace {
+#if defined(__GNUC__)
+std::int64_t double_up(std::int64_t x) __attribute__((const));
+#endif
+std::int64_t double_up(const std::int64_t x) { return x * 2; }
+}  // namespace
+
+// Using DoNotOptimize on types like BitRef seem to cause a lot of problems
+// with the inline assembly on both GCC and Clang.
+struct BitRef {
+  int index;
+  unsigned char& byte;
+
+ public:
+  static BitRef Make() {
+    static unsigned char arr[2] = {};
+    BitRef b(1, arr[0]);
+    return b;
+  }
+
+ private:
+  BitRef(int i, unsigned char& b) : index(i), byte(b) {}
+};
+
+int main(int argc, char* argv[]) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+
+  // this test verifies compilation of DoNotOptimize() for some types
+
+  char buffer1[1] = "";
+  benchmark::DoNotOptimize(buffer1);
+
+  char buffer2[2] = "";
+  benchmark::DoNotOptimize(buffer2);
+
+  char buffer3[3] = "";
+  benchmark::DoNotOptimize(buffer3);
+
+  char buffer8[8] = "";
+  benchmark::DoNotOptimize(buffer8);
+
+  char buffer20[20] = "";
+  benchmark::DoNotOptimize(buffer20);
+
+  char buffer1024[1024] = "";
+  benchmark::DoNotOptimize(buffer1024);
+  char* bptr = &buffer1024[0];
+  benchmark::DoNotOptimize(bptr);
+
+  int x = 123;
+  benchmark::DoNotOptimize(x);
+  int* xp = &x;
+  benchmark::DoNotOptimize(xp);
+  benchmark::DoNotOptimize(x += 42);
+
+  std::int64_t y = double_up(x);
+  benchmark::DoNotOptimize(y);
+
+  // These tests are to e
+  BitRef lval = BitRef::Make();
+  benchmark::DoNotOptimize(lval);
+
+  // Check that accept rvalue.
+  benchmark::DoNotOptimize(BitRef::Make());
+}

test/filter_test.cc

@@ -0,0 +1,122 @@
+#include <algorithm>
+#include <cassert>
+#include <cmath>
+#include <cstdint>
+#include <cstdlib>
+#include <iostream>
+#include <limits>
+#include <sstream>
+#include <string>
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/registration.h"
+#include "benchmark/reporter.h"
+#include "benchmark/state.h"
+
+namespace {
+
+class TestReporter : public benchmark::ConsoleReporter {
+ public:
+  bool ReportContext(const Context& context) override {
+    return ConsoleReporter::ReportContext(context);
+  };
+
+  void ReportRuns(const std::vector<Run>& report) override {
+    ++count_;
+    max_family_index_ = std::max(max_family_index_, report[0].family_index);
+    ConsoleReporter::ReportRuns(report);
+  };
+
+  TestReporter() : count_(0), max_family_index_(0) {}
+
+  ~TestReporter() override {}
+
+  int GetCount() const { return count_; }
+
+  int64_t GetMaxFamilyIndex() const { return max_family_index_; }
+
+ private:
+  mutable int count_;
+  mutable int64_t max_family_index_;
+};
+
+void NoPrefix(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+}
+BENCHMARK(NoPrefix);
+
+void BM_Foo(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+}
+BENCHMARK(BM_Foo);
+
+void BM_Bar(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+}
+BENCHMARK(BM_Bar);
+
+void BM_FooBar(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+}
+BENCHMARK(BM_FooBar);
+
+void BM_FooBa(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+}
+BENCHMARK(BM_FooBa);
+}  // end namespace
+
+int main(int argc, char** argv) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+
+  bool list_only = false;
+  for (int i = 0; i < argc; ++i) {
+    list_only |= std::string(argv[i]).find("--benchmark_list_tests") !=
+                 std::string::npos;
+  }
+
+  benchmark::Initialize(&argc, argv);
+
+  TestReporter test_reporter;
+  const int64_t returned_count =
+      static_cast<int64_t>(benchmark::RunSpecifiedBenchmarks(&test_reporter));
+
+  if (argc == 2) {
+    // Make sure we ran all of the tests
+    std::stringstream ss(argv[1]);
+    int64_t expected_return = 0;
+    ss >> expected_return;
+
+    if (returned_count != expected_return) {
+      std::cerr << "ERROR: Expected " << expected_return
+                << " tests to match the filter but returned_count = "
+                << returned_count << '\n';
+      return -1;
+    }
+
+    const int64_t expected_reports = list_only ? 0 : expected_return;
+    const int64_t reports_count = test_reporter.GetCount();
+    if (reports_count != expected_reports) {
+      std::cerr << "ERROR: Expected " << expected_reports
+                << " tests to be run but reported_count = " << reports_count
+                << '\n';
+      return -1;
+    }
+
+    const int64_t max_family_index = test_reporter.GetMaxFamilyIndex();
+    const int64_t num_families = reports_count == 0 ? 0 : 1 + max_family_index;
+    if (num_families != expected_reports) {
+      std::cerr << "ERROR: Expected " << expected_reports
+                << " test families to be run but num_families = "
+                << num_families << '\n';
+      return -1;
+    }
+  }
+
+  return 0;
+}

test/fixture_test.cc

@@ -0,0 +1,53 @@
+
+#include <cassert>
+#include <memory>
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+
+#define FIXTURE_BECHMARK_NAME MyFixture
+
+class FIXTURE_BECHMARK_NAME : public ::benchmark::Fixture {
+ public:
+  void SetUp(const ::benchmark::State& state) override {
+    if (state.thread_index() == 0) {
+      assert(data.get() == nullptr);
+      data.reset(new int(42));
+    }
+  }
+
+  void TearDown(const ::benchmark::State& state) override {
+    if (state.thread_index() == 0) {
+      assert(data.get() != nullptr);
+      data.reset();
+    }
+  }
+
+  ~FIXTURE_BECHMARK_NAME() override { assert(data == nullptr); }
+
+  std::unique_ptr<int> data;
+};
+
+BENCHMARK_F(FIXTURE_BECHMARK_NAME, Foo)(benchmark::State& st) {
+  assert(data.get() != nullptr);
+  assert(*data == 42);
+  for (auto _ : st) {
+  }
+}
+
+BENCHMARK_DEFINE_F(FIXTURE_BECHMARK_NAME, Bar)(benchmark::State& st) {
+  if (st.thread_index() == 0) {
+    assert(data.get() != nullptr);
+    assert(*data == 42);
+  }
+  for (auto _ : st) {
+    assert(data.get() != nullptr);
+    assert(*data == 42);
+  }
+  st.SetItemsProcessed(st.range(0));
+}
+BENCHMARK_REGISTER_F(FIXTURE_BECHMARK_NAME, Bar)->Arg(42);
+BENCHMARK_REGISTER_F(FIXTURE_BECHMARK_NAME, Bar)->Arg(42)->ThreadPerCpu();
+
+BENCHMARK_MAIN();

test/internal_threading_test.cc

@@ -0,0 +1,195 @@
+
+#undef NDEBUG
+
+#include <chrono>
+#include <thread>
+
+#include "../src/timers.h"
+#include "benchmark/benchmark_api.h"
+#include "benchmark/counter.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+#include "output_test.h"
+
+namespace {
+const std::chrono::duration<double, std::milli> time_frame(50);
+const double time_frame_in_sec(
+    std::chrono::duration_cast<std::chrono::duration<double, std::ratio<1, 1>>>(
+        time_frame)
+        .count());
+
+void MyBusySpinwait() {
+  const auto start = benchmark::ChronoClockNow();
+
+  while (true) {
+    const auto now = benchmark::ChronoClockNow();
+    const auto elapsed = now - start;
+
+    if (std::chrono::duration<double, std::chrono::seconds::period>(elapsed) >=
+        time_frame) {
+      return;
+    }
+  }
+}
+
+// ========================================================================= //
+// --------------------------- TEST CASES BEGIN ---------------------------- //
+// ========================================================================= //
+
+// ========================================================================= //
+// BM_MainThread
+
+void BM_MainThread(benchmark::State& state) {
+  for (auto _ : state) {
+    MyBusySpinwait();
+    state.SetIterationTime(time_frame_in_sec);
+  }
+  state.counters["invtime"] =
+      benchmark::Counter{1, benchmark::Counter::kIsRate};
+}
+
+BENCHMARK(BM_MainThread)->Iterations(1)->Threads(1);
+BENCHMARK(BM_MainThread)->Iterations(1)->Threads(1)->UseRealTime();
+BENCHMARK(BM_MainThread)->Iterations(1)->Threads(1)->UseManualTime();
+BENCHMARK(BM_MainThread)->Iterations(1)->Threads(1)->MeasureProcessCPUTime();
+BENCHMARK(BM_MainThread)
+    ->Iterations(1)
+    ->Threads(1)
+    ->MeasureProcessCPUTime()
+    ->UseRealTime();
+BENCHMARK(BM_MainThread)
+    ->Iterations(1)
+    ->Threads(1)
+    ->MeasureProcessCPUTime()
+    ->UseManualTime();
+
+BENCHMARK(BM_MainThread)->Iterations(1)->Threads(2);
+BENCHMARK(BM_MainThread)->Iterations(1)->Threads(2)->UseRealTime();
+BENCHMARK(BM_MainThread)->Iterations(1)->Threads(2)->UseManualTime();
+BENCHMARK(BM_MainThread)->Iterations(1)->Threads(2)->MeasureProcessCPUTime();
+BENCHMARK(BM_MainThread)
+    ->Iterations(1)
+    ->Threads(2)
+    ->MeasureProcessCPUTime()
+    ->UseRealTime();
+BENCHMARK(BM_MainThread)
+    ->Iterations(1)
+    ->Threads(2)
+    ->MeasureProcessCPUTime()
+    ->UseManualTime();
+
+// ========================================================================= //
+// BM_WorkerThread
+
+void BM_WorkerThread(benchmark::State& state) {
+  for (auto _ : state) {
+    std::thread Worker(&MyBusySpinwait);
+    Worker.join();
+    state.SetIterationTime(time_frame_in_sec);
+  }
+  state.counters["invtime"] =
+      benchmark::Counter{1, benchmark::Counter::kIsRate};
+}
+
+BENCHMARK(BM_WorkerThread)->Iterations(1)->Threads(1);
+BENCHMARK(BM_WorkerThread)->Iterations(1)->Threads(1)->UseRealTime();
+BENCHMARK(BM_WorkerThread)->Iterations(1)->Threads(1)->UseManualTime();
+BENCHMARK(BM_WorkerThread)->Iterations(1)->Threads(1)->MeasureProcessCPUTime();
+BENCHMARK(BM_WorkerThread)
+    ->Iterations(1)
+    ->Threads(1)
+    ->MeasureProcessCPUTime()
+    ->UseRealTime();
+BENCHMARK(BM_WorkerThread)
+    ->Iterations(1)
+    ->Threads(1)
+    ->MeasureProcessCPUTime()
+    ->UseManualTime();
+
+BENCHMARK(BM_WorkerThread)->Iterations(1)->Threads(2);
+BENCHMARK(BM_WorkerThread)->Iterations(1)->Threads(2)->UseRealTime();
+BENCHMARK(BM_WorkerThread)->Iterations(1)->Threads(2)->UseManualTime();
+BENCHMARK(BM_WorkerThread)->Iterations(1)->Threads(2)->MeasureProcessCPUTime();
+BENCHMARK(BM_WorkerThread)
+    ->Iterations(1)
+    ->Threads(2)
+    ->MeasureProcessCPUTime()
+    ->UseRealTime();
+BENCHMARK(BM_WorkerThread)
+    ->Iterations(1)
+    ->Threads(2)
+    ->MeasureProcessCPUTime()
+    ->UseManualTime();
+
+// ========================================================================= //
+// BM_MainThreadAndWorkerThread
+
+void BM_MainThreadAndWorkerThread(benchmark::State& state) {
+  for (auto _ : state) {
+    std::thread Worker(&MyBusySpinwait);
+    MyBusySpinwait();
+    Worker.join();
+    state.SetIterationTime(time_frame_in_sec);
+  }
+  state.counters["invtime"] =
+      benchmark::Counter{1, benchmark::Counter::kIsRate};
+}
+
+BENCHMARK(BM_MainThreadAndWorkerThread)->Iterations(1)->Threads(1);
+BENCHMARK(BM_MainThreadAndWorkerThread)
+    ->Iterations(1)
+    ->Threads(1)
+    ->UseRealTime();
+BENCHMARK(BM_MainThreadAndWorkerThread)
+    ->Iterations(1)
+    ->Threads(1)
+    ->UseManualTime();
+BENCHMARK(BM_MainThreadAndWorkerThread)
+    ->Iterations(1)
+    ->Threads(1)
+    ->MeasureProcessCPUTime();
+BENCHMARK(BM_MainThreadAndWorkerThread)
+    ->Iterations(1)
+    ->Threads(1)
+    ->MeasureProcessCPUTime()
+    ->UseRealTime();
+BENCHMARK(BM_MainThreadAndWorkerThread)
+    ->Iterations(1)
+    ->Threads(1)
+    ->MeasureProcessCPUTime()
+    ->UseManualTime();
+
+BENCHMARK(BM_MainThreadAndWorkerThread)->Iterations(1)->Threads(2);
+BENCHMARK(BM_MainThreadAndWorkerThread)
+    ->Iterations(1)
+    ->Threads(2)
+    ->UseRealTime();
+BENCHMARK(BM_MainThreadAndWorkerThread)
+    ->Iterations(1)
+    ->Threads(2)
+    ->UseManualTime();
+BENCHMARK(BM_MainThreadAndWorkerThread)
+    ->Iterations(1)
+    ->Threads(2)
+    ->MeasureProcessCPUTime();
+BENCHMARK(BM_MainThreadAndWorkerThread)
+    ->Iterations(1)
+    ->Threads(2)
+    ->MeasureProcessCPUTime()
+    ->UseRealTime();
+BENCHMARK(BM_MainThreadAndWorkerThread)
+    ->Iterations(1)
+    ->Threads(2)
+    ->MeasureProcessCPUTime()
+    ->UseManualTime();
+}  // end namespace
+
+// ========================================================================= //
+// ---------------------------- TEST CASES END ----------------------------- //
+// ========================================================================= //
+
+int main(int argc, char* argv[]) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+
+  RunOutputTests(argc, argv);
+}

test/link_main_test.cc

@@ -0,0 +1,14 @@
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+#include "benchmark/utils.h"
+
+namespace {
+void BM_empty(benchmark::State& state) {
+  for (auto _ : state) {
+    auto iterations = static_cast<double>(state.iterations()) *
+                      static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(iterations);
+  }
+}
+BENCHMARK(BM_empty);
+}  // end namespace

test/locale_impermeability_test.cc

@@ -0,0 +1,50 @@
+#undef NDEBUG
+#include <cassert>
+#include <cmath>
+#include <cstdlib>
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+#include "output_test.h"
+
+namespace {
+void BM_ostream(benchmark::State& state) {
+#if !defined(__GLIBCXX__) || defined(__GLIBC__) || defined(__DragonFly__)
+  // libstdc++ only supports locale manipulations on GNU and
+  // DragonflyBSD platforms at the time of writing, don't run the test
+  // on other platforms when using libstdc++.
+  std::locale::global(std::locale("en_US.UTF-8"));
+#endif
+  while (state.KeepRunning()) {
+    state.SetIterationTime(1e-6);
+  }
+}
+BENCHMARK(BM_ostream)->UseManualTime()->Iterations(1000000);
+
+ADD_CASES(TC_ConsoleOut, {{"^BM_ostream/iterations:1000000/manual_time"
+                           " %console_report$"}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_ostream/iterations:1000000/manual_time\",$"},
+           {"\"family_index\": 0,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": "
+            "\"BM_ostream/iterations:1000000/manual_time\",$",
+            MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 1,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"iterations\": 1000000,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\"$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_ostream/iterations:1000000/"
+                       "manual_time\",1000000,%float,%float,ns,,,,,$"}});
+}  // end namespace
+
+int main(int argc, char* argv[]) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+  RunOutputTests(argc, argv);
+}

test/manual_threading_test.cc

@@ -0,0 +1,178 @@
+
+#include <memory>
+#undef NDEBUG
+
+#include <chrono>
+#include <thread>
+
+#include "../src/timers.h"
+#include "benchmark/benchmark_api.h"
+#include "benchmark/counter.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+
+namespace {
+
+const std::chrono::duration<double, std::milli> time_frame(50);
+const double time_frame_in_sec(
+    std::chrono::duration_cast<std::chrono::duration<double, std::ratio<1, 1>>>(
+        time_frame)
+        .count());
+
+void MyBusySpinwait() {
+  const auto start = benchmark::ChronoClockNow();
+
+  while (true) {
+    const auto now = benchmark::ChronoClockNow();
+    const auto elapsed = now - start;
+
+    if (std::chrono::duration<double, std::chrono::seconds::period>(elapsed) >=
+        time_frame) {
+      return;
+    }
+  }
+}
+
+int numRunThreadsCalled_ = 0;
+
+class ManualThreadRunner : public benchmark::ThreadRunnerBase {
+ public:
+  explicit ManualThreadRunner(int num_threads)
+      : pool(static_cast<size_t>(num_threads - 1)) {}
+
+  void RunThreads(const std::function<void(int)>& fn) final {
+    for (std::size_t ti = 0; ti < pool.size(); ++ti) {
+      pool[ti] = std::thread(fn, static_cast<int>(ti + 1));
+    }
+
+    fn(0);
+
+    for (std::thread& thread : pool) {
+      thread.join();
+    }
+
+    ++numRunThreadsCalled_;
+  }
+
+ private:
+  std::vector<std::thread> pool;
+};
+
+// ========================================================================= //
+// --------------------------- TEST CASES BEGIN ---------------------------- //
+// ========================================================================= //
+
+// ========================================================================= //
+// BM_ManualThreading
+// Creation of threads is done before the start of the measurement,
+// joining after the finish of the measurement.
+void BM_ManualThreading(benchmark::State& state) {
+  for (auto _ : state) {
+    MyBusySpinwait();
+    state.SetIterationTime(time_frame_in_sec);
+  }
+  state.counters["invtime"] =
+      benchmark::Counter{1, benchmark::Counter::kIsRate};
+}
+
+}  // end namespace
+
+BENCHMARK(BM_ManualThreading)
+    ->Iterations(1)
+    ->ThreadRunner([](int num_threads) {
+      return std::make_unique<ManualThreadRunner>(num_threads);
+    })
+    ->Threads(1);
+BENCHMARK(BM_ManualThreading)
+    ->Iterations(1)
+    ->ThreadRunner([](int num_threads) {
+      return std::make_unique<ManualThreadRunner>(num_threads);
+    })
+    ->Threads(1)
+    ->UseRealTime();
+BENCHMARK(BM_ManualThreading)
+    ->Iterations(1)
+    ->ThreadRunner([](int num_threads) {
+      return std::make_unique<ManualThreadRunner>(num_threads);
+    })
+    ->Threads(1)
+    ->UseManualTime();
+BENCHMARK(BM_ManualThreading)
+    ->Iterations(1)
+    ->ThreadRunner([](int num_threads) {
+      return std::make_unique<ManualThreadRunner>(num_threads);
+    })
+    ->Threads(1)
+    ->MeasureProcessCPUTime();
+BENCHMARK(BM_ManualThreading)
+    ->Iterations(1)
+    ->ThreadRunner([](int num_threads) {
+      return std::make_unique<ManualThreadRunner>(num_threads);
+    })
+    ->Threads(1)
+    ->MeasureProcessCPUTime()
+    ->UseRealTime();
+BENCHMARK(BM_ManualThreading)
+    ->Iterations(1)
+    ->ThreadRunner([](int num_threads) {
+      return std::make_unique<ManualThreadRunner>(num_threads);
+    })
+    ->Threads(1)
+    ->MeasureProcessCPUTime()
+    ->UseManualTime();
+
+BENCHMARK(BM_ManualThreading)
+    ->Iterations(1)
+    ->ThreadRunner([](int num_threads) {
+      return std::make_unique<ManualThreadRunner>(num_threads);
+    })
+    ->Threads(2);
+BENCHMARK(BM_ManualThreading)
+    ->Iterations(1)
+    ->ThreadRunner([](int num_threads) {
+      return std::make_unique<ManualThreadRunner>(num_threads);
+    })
+    ->Threads(2)
+    ->UseRealTime();
+BENCHMARK(BM_ManualThreading)
+    ->Iterations(1)
+    ->ThreadRunner([](int num_threads) {
+      return std::make_unique<ManualThreadRunner>(num_threads);
+    })
+    ->Threads(2)
+    ->UseManualTime();
+BENCHMARK(BM_ManualThreading)
+    ->Iterations(1)
+    ->ThreadRunner([](int num_threads) {
+      return std::make_unique<ManualThreadRunner>(num_threads);
+    })
+    ->Threads(2)
+    ->MeasureProcessCPUTime();
+BENCHMARK(BM_ManualThreading)
+    ->Iterations(1)
+    ->ThreadRunner([](int num_threads) {
+      return std::make_unique<ManualThreadRunner>(num_threads);
+    })
+    ->Threads(2)
+    ->MeasureProcessCPUTime()
+    ->UseRealTime();
+BENCHMARK(BM_ManualThreading)
+    ->Iterations(1)
+    ->ThreadRunner([](int num_threads) {
+      return std::make_unique<ManualThreadRunner>(num_threads);
+    })
+    ->Threads(2)
+    ->MeasureProcessCPUTime()
+    ->UseManualTime();
+
+// ========================================================================= //
+// ---------------------------- TEST CASES END ----------------------------- //
+// ========================================================================= //
+
+int main(int argc, char* argv[]) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+  benchmark::Initialize(&argc, argv);
+  benchmark::RunSpecifiedBenchmarks();
+  benchmark::Shutdown();
+  assert(numRunThreadsCalled_ > 0);
+}

test/map_test.cc

@@ -0,0 +1,61 @@
+#include <cstdlib>
+#include <map>
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+#include "benchmark/utils.h"
+
+namespace {
+
+std::map<int, int> ConstructRandomMap(int size) {
+  std::map<int, int> m;
+  for (int i = 0; i < size; ++i) {
+    m.insert(std::make_pair(std::rand() % size, std::rand() % size));
+  }
+  return m;
+}
+
+// Basic version.
+void BM_MapLookup(benchmark::State& state) {
+  const int size = static_cast<int>(state.range(0));
+  std::map<int, int> m;
+  for (auto _ : state) {
+    state.PauseTiming();
+    m = ConstructRandomMap(size);
+    state.ResumeTiming();
+    for (int i = 0; i < size; ++i) {
+      auto it = m.find(std::rand() % size);
+      benchmark::DoNotOptimize(it);
+    }
+  }
+  state.SetItemsProcessed(state.iterations() * size);
+}
+BENCHMARK(BM_MapLookup)->Range(1 << 3, 1 << 12);
+}  // namespace
+
+// Using fixtures.
+class MapFixture : public ::benchmark::Fixture {
+ public:
+  void SetUp(const ::benchmark::State& st) override {
+    m = ConstructRandomMap(static_cast<int>(st.range(0)));
+  }
+
+  void TearDown(const ::benchmark::State& /*unused*/) override { m.clear(); }
+
+  std::map<int, int> m;
+};
+
+BENCHMARK_DEFINE_F(MapFixture, Lookup)(benchmark::State& state) {
+  const int size = static_cast<int>(state.range(0));
+  for (auto _ : state) {
+    for (int i = 0; i < size; ++i) {
+      auto it = m.find(std::rand() % size);
+      benchmark::DoNotOptimize(it);
+    }
+  }
+  state.SetItemsProcessed(state.iterations() * size);
+}
+BENCHMARK_REGISTER_F(MapFixture, Lookup)->Range(1 << 3, 1 << 12);
+
+BENCHMARK_MAIN();

test/memory_manager_test.cc

@@ -0,0 +1,54 @@
+#include <memory>
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/managers.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+#include "benchmark/utils.h"
+#include "output_test.h"
+
+namespace {
+class TestMemoryManager : public benchmark::MemoryManager {
+  void Start() override {}
+  void Stop(Result& result) override {
+    result.num_allocs = 42;
+    result.max_bytes_used = 42000;
+  }
+};
+
+void BM_empty(benchmark::State& state) {
+  for (auto _ : state) {
+    auto iterations = static_cast<double>(state.iterations()) *
+                      static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(iterations);
+  }
+}
+BENCHMARK(BM_empty);
+}  // end namespace
+
+ADD_CASES(TC_ConsoleOut, {{"^BM_empty %console_report$"}});
+ADD_CASES(TC_JSONOut, {{"\"name\": \"BM_empty\",$"},
+                       {"\"family_index\": 0,$", MR_Next},
+                       {"\"per_family_instance_index\": 0,$", MR_Next},
+                       {"\"run_name\": \"BM_empty\",$", MR_Next},
+                       {"\"run_type\": \"iteration\",$", MR_Next},
+                       {"\"repetitions\": 1,$", MR_Next},
+                       {"\"repetition_index\": 0,$", MR_Next},
+                       {"\"threads\": 1,$", MR_Next},
+                       {"\"iterations\": %int,$", MR_Next},
+                       {"\"real_time\": %float,$", MR_Next},
+                       {"\"cpu_time\": %float,$", MR_Next},
+                       {"\"time_unit\": \"ns\",$", MR_Next},
+                       {"\"allocs_per_iter\": %float,$", MR_Next},
+                       {"\"max_bytes_used\": 42000$", MR_Next},
+                       {"}", MR_Next}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_empty\",%csv_report$"}});
+
+int main(int argc, char* argv[]) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+  std::unique_ptr<benchmark::MemoryManager> mm(new TestMemoryManager());
+
+  benchmark::RegisterMemoryManager(mm.get());
+  RunOutputTests(argc, argv);
+  benchmark::RegisterMemoryManager(nullptr);
+}

test/memory_results_gtest.cc

@@ -0,0 +1,104 @@
+#include <vector>
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/managers.h"
+#include "benchmark/reporter.h"
+#include "benchmark/state.h"
+#include "gtest/gtest.h"
+
+namespace {
+
+using benchmark::Benchmark;
+using benchmark::ClearRegisteredBenchmarks;
+using benchmark::ConsoleReporter;
+using benchmark::MemoryManager;
+using benchmark::RegisterBenchmark;
+using benchmark::RunSpecifiedBenchmarks;
+using benchmark::State;
+
+constexpr int N_REPETITIONS = 100;
+constexpr int N_ITERATIONS = 1;
+
+int num_allocs = 0;
+int max_bytes_used = 0;
+int total_allocated_bytes = 0;
+int net_heap_growth = 0;
+
+void reset() {
+  num_allocs = 0;
+  max_bytes_used = 0;
+  total_allocated_bytes = 0;
+  net_heap_growth = 0;
+}
+class TestMemoryManager : public MemoryManager {
+  void Start() override {}
+  void Stop(Result& result) override {
+    result.num_allocs = num_allocs;
+    result.net_heap_growth = net_heap_growth;
+    result.max_bytes_used = max_bytes_used;
+    result.total_allocated_bytes = total_allocated_bytes;
+
+    num_allocs += 1;
+    max_bytes_used += 2;
+    net_heap_growth += 4;
+    total_allocated_bytes += 10;
+  }
+};
+
+class TestReporter : public ConsoleReporter {
+ public:
+  TestReporter() = default;
+  virtual ~TestReporter() = default;
+
+  bool ReportContext(const Context& /*unused*/) override { return true; }
+
+  void PrintHeader(const Run&) override {}
+  void PrintRunData(const Run& run) override {
+    if (run.repetition_index == -1) return;
+    if (!run.memory_result.memory_iterations) return;
+
+    store.push_back(run.memory_result);
+  }
+
+  std::vector<MemoryManager::Result> store;
+};
+
+class MemoryResultsTest : public testing::Test {
+ public:
+  Benchmark* bm;
+  TestReporter reporter;
+
+  void SetUp() override {
+    bm = RegisterBenchmark("BM", [](State& st) {
+      for (auto _ : st) {
+      }
+    });
+    bm->Repetitions(N_REPETITIONS);
+    bm->Iterations(N_ITERATIONS);
+    reset();
+  }
+  void TearDown() override { ClearRegisteredBenchmarks(); }
+};
+
+TEST_F(MemoryResultsTest, NoMMTest) {
+  RunSpecifiedBenchmarks(&reporter);
+  EXPECT_EQ(reporter.store.size(), 0);
+}
+
+TEST_F(MemoryResultsTest, ResultsTest) {
+  auto mm = std::make_unique<TestMemoryManager>();
+  RegisterMemoryManager(mm.get());
+
+  RunSpecifiedBenchmarks(&reporter);
+  EXPECT_EQ(reporter.store.size(), N_REPETITIONS);
+
+  for (size_t i = 0; i < reporter.store.size(); i++) {
+    EXPECT_EQ(reporter.store[i].num_allocs, static_cast<int64_t>(i));
+    EXPECT_EQ(reporter.store[i].max_bytes_used, static_cast<int64_t>(i) * 2);
+    EXPECT_EQ(reporter.store[i].net_heap_growth, static_cast<int64_t>(i) * 4);
+    EXPECT_EQ(reporter.store[i].total_allocated_bytes,
+              static_cast<int64_t>(i) * 10);
+  }
+}
+
+}  // namespace

test/min_time_parse_gtest.cc

@@ -0,0 +1,30 @@
+#include "../src/benchmark_runner.h"
+#include "gtest/gtest.h"
+
+namespace {
+
+TEST(ParseMinTimeTest, InvalidInput) {
+#if GTEST_HAS_DEATH_TEST
+  // Tests only runnable in debug mode (when BM_CHECK is enabled).
+#ifndef NDEBUG
+#ifndef TEST_BENCHMARK_LIBRARY_HAS_NO_ASSERTIONS
+  ASSERT_DEATH_IF_SUPPORTED(
+      { benchmark::internal::ParseBenchMinTime("abc"); },
+      "Malformed seconds value passed to --benchmark_min_time: `abc`");
+
+  ASSERT_DEATH_IF_SUPPORTED(
+      { benchmark::internal::ParseBenchMinTime("123ms"); },
+      "Malformed seconds value passed to --benchmark_min_time: `123ms`");
+
+  ASSERT_DEATH_IF_SUPPORTED(
+      { benchmark::internal::ParseBenchMinTime("1z"); },
+      "Malformed seconds value passed to --benchmark_min_time: `1z`");
+
+  ASSERT_DEATH_IF_SUPPORTED(
+      { benchmark::internal::ParseBenchMinTime("1hs"); },
+      "Malformed seconds value passed to --benchmark_min_time: `1hs`");
+#endif
+#endif
+#endif
+}
+}  // namespace

test/multiple_ranges_test.cc

@@ -0,0 +1,101 @@
+#include <cassert>
+#include <iostream>
+#include <set>
+#include <vector>
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+#include "benchmark/utils.h"
+
+namespace {
+class MultipleRangesFixture : public ::benchmark::Fixture {
+ public:
+  MultipleRangesFixture()
+      : expectedValues({{1, 3, 5},
+                        {1, 3, 8},
+                        {1, 3, 15},
+                        {2, 3, 5},
+                        {2, 3, 8},
+                        {2, 3, 15},
+                        {1, 4, 5},
+                        {1, 4, 8},
+                        {1, 4, 15},
+                        {2, 4, 5},
+                        {2, 4, 8},
+                        {2, 4, 15},
+                        {1, 7, 5},
+                        {1, 7, 8},
+                        {1, 7, 15},
+                        {2, 7, 5},
+                        {2, 7, 8},
+                        {2, 7, 15},
+                        {7, 6, 3}}) {}
+
+  void SetUp(const ::benchmark::State& state) override {
+    std::vector<int64_t> ranges = {state.range(0), state.range(1),
+                                   state.range(2)};
+
+    assert(expectedValues.find(ranges) != expectedValues.end());
+
+    actualValues.insert(ranges);
+  }
+
+  // NOTE: This is not TearDown as we want to check after _all_ runs are
+  // complete.
+  ~MultipleRangesFixture() override {
+    if (actualValues != expectedValues) {
+      std::cout << "EXPECTED\n";
+      for (const auto& v : expectedValues) {
+        std::cout << "{";
+        for (int64_t iv : v) {
+          std::cout << iv << ", ";
+        }
+        std::cout << "}\n";
+      }
+      std::cout << "ACTUAL\n";
+      for (const auto& v : actualValues) {
+        std::cout << "{";
+        for (int64_t iv : v) {
+          std::cout << iv << ", ";
+        }
+        std::cout << "}\n";
+      }
+    }
+  }
+
+  std::set<std::vector<int64_t>> expectedValues;
+  std::set<std::vector<int64_t>> actualValues;
+};
+
+BENCHMARK_DEFINE_F(MultipleRangesFixture, Empty)(benchmark::State& state) {
+  for (auto _ : state) {
+    int64_t product = state.range(0) * state.range(1) * state.range(2);
+    for (int64_t x = 0; x < product; x++) {
+      benchmark::DoNotOptimize(x);
+    }
+  }
+}
+
+BENCHMARK_REGISTER_F(MultipleRangesFixture, Empty)
+    ->RangeMultiplier(2)
+    ->Ranges({{1, 2}, {3, 7}, {5, 15}})
+    ->Args({7, 6, 3});
+
+void BM_CheckDefaultArgument(benchmark::State& state) {
+  // Test that the 'range()' without an argument is the same as 'range(0)'.
+  assert(state.range() == state.range(0));
+  assert(state.range() != state.range(1));
+  for (auto _ : state) {
+  }
+}
+BENCHMARK(BM_CheckDefaultArgument)->Ranges({{1, 5}, {6, 10}});
+
+void BM_MultipleRanges(benchmark::State& st) {
+  for (auto _ : st) {
+  }
+}
+BENCHMARK(BM_MultipleRanges)->Ranges({{5, 5}, {6, 6}});
+}  // end namespace
+
+BENCHMARK_MAIN();

test/options_test.cc

@@ -0,0 +1,82 @@
+#include <chrono>
+#include <thread>
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+#include "benchmark/types.h"
+
+#if defined(NDEBUG)
+#undef NDEBUG
+#endif
+#include <cassert>
+
+namespace {
+void BM_basic(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+}
+
+void BM_basic_slow(benchmark::State& state) {
+  std::chrono::milliseconds sleep_duration(state.range(0));
+  for (auto _ : state) {
+    std::this_thread::sleep_for(
+        std::chrono::duration_cast<std::chrono::nanoseconds>(sleep_duration));
+  }
+}
+
+BENCHMARK(BM_basic);
+BENCHMARK(BM_basic)->Arg(42);
+BENCHMARK(BM_basic_slow)->Arg(10)->Unit(benchmark::kNanosecond);
+BENCHMARK(BM_basic_slow)->Arg(100)->Unit(benchmark::kMicrosecond);
+BENCHMARK(BM_basic_slow)->Arg(1000)->Unit(benchmark::kMillisecond);
+BENCHMARK(BM_basic_slow)->Arg(1000)->Unit(benchmark::kSecond);
+BENCHMARK(BM_basic)->Range(1, 8);
+BENCHMARK(BM_basic)->RangeMultiplier(2)->Range(1, 8);
+BENCHMARK(BM_basic)->DenseRange(10, 15);
+BENCHMARK(BM_basic)->Args({42, 42});
+BENCHMARK(BM_basic)->Ranges({{64, 512}, {64, 512}});
+BENCHMARK(BM_basic)->MinTime(0.7);
+BENCHMARK(BM_basic)->MinWarmUpTime(0.8);
+BENCHMARK(BM_basic)->MinTime(0.1)->MinWarmUpTime(0.2);
+BENCHMARK(BM_basic)->UseRealTime();
+BENCHMARK(BM_basic)->ThreadRange(2, 4);
+BENCHMARK(BM_basic)->ThreadPerCpu();
+BENCHMARK(BM_basic)->Repetitions(3);
+BENCHMARK(BM_basic)
+    ->RangeMultiplier(std::numeric_limits<int>::max())
+    ->Range(std::numeric_limits<int64_t>::min(),
+            std::numeric_limits<int64_t>::max());
+
+// Negative ranges
+BENCHMARK(BM_basic)->Range(-64, -1);
+BENCHMARK(BM_basic)->RangeMultiplier(4)->Range(-8, 8);
+BENCHMARK(BM_basic)->DenseRange(-2, 2, 1);
+BENCHMARK(BM_basic)->Ranges({{-64, 1}, {-8, -1}});
+
+void CustomArgs(benchmark::Benchmark* b) {
+  for (int i = 0; i < 10; ++i) {
+    b->Arg(i);
+  }
+}
+
+BENCHMARK(BM_basic)->Apply(CustomArgs);
+
+void BM_explicit_iteration_count(benchmark::State& state) {
+  // Test that benchmarks specified with an explicit iteration count are
+  // only run once.
+  static bool invoked_before = false;
+  assert(!invoked_before);
+  invoked_before = true;
+
+  // Test that the requested iteration count is respected.
+  assert(state.max_iterations == 42);
+  for (auto _ : state) {
+  }
+  assert(state.iterations() == state.max_iterations);
+  assert(state.iterations() == 42);
+}
+BENCHMARK(BM_explicit_iteration_count)->Iterations(42);
+}  // end namespace
+
+BENCHMARK_MAIN();

test/output_test.h

@@ -0,0 +1,213 @@
+#ifndef TEST_OUTPUT_TEST_H
+#define TEST_OUTPUT_TEST_H
+
+#undef NDEBUG
+#include <functional>
+#include <initializer_list>
+#include <memory>
+#include <sstream>
+#include <string>
+#include <utility>
+#include <vector>
+
+#include "../src/re.h"
+#include "benchmark/benchmark_api.h"
+
+#define CONCAT2(x, y) x##y
+#define CONCAT(x, y) CONCAT2(x, y)
+
+#define ADD_CASES(...) \
+  const int CONCAT(dummy, __LINE__) = ::AddCases(__VA_ARGS__)
+
+#define SET_SUBSTITUTIONS(...) \
+  const int CONCAT(dummy, __LINE__) = ::SetSubstitutions(__VA_ARGS__)
+
+enum MatchRules : uint8_t {
+  MR_Default,  // Skip non-matching lines until a match is found.
+  MR_Next,     // Match must occur on the next line.
+  MR_Not  // No line between the current position and the next match matches
+          // the regex
+};
+
+struct TestCase {
+  TestCase(std::string re, int rule = MR_Default);
+
+  std::string regex_str;
+  int match_rule;
+  std::string substituted_regex;
+  std::shared_ptr<benchmark::Regex> regex;
+};
+
+enum TestCaseID : uint8_t {
+  TC_ConsoleOut,
+  TC_ConsoleErr,
+  TC_JSONOut,
+  TC_JSONErr,
+  TC_CSVOut,
+  TC_CSVErr,
+
+  TC_NumID  // PRIVATE
+};
+
+// Add a list of test cases to be run against the output specified by
+// 'ID'
+int AddCases(TestCaseID ID, std::initializer_list<TestCase> il);
+
+// Add or set a list of substitutions to be performed on constructed regex's
+// See 'output_test_helper.cc' for a list of default substitutions.
+int SetSubstitutions(
+    std::initializer_list<std::pair<std::string, std::string>> il);
+
+// Run all output tests.
+void RunOutputTests(int argc, char* argv[]);
+
+// Count the number of 'pat' substrings in the 'haystack' string.
+int SubstrCnt(const std::string& haystack, const std::string& pat);
+
+// Run registered benchmarks with file reporter enabled, and return the content
+// outputted by the file reporter.
+std::string GetFileReporterOutput(int argc, char* argv[]);
+
+// ========================================================================= //
+// ------------------------- Results checking ------------------------------ //
+// ========================================================================= //
+
+// Call this macro to register a benchmark for checking its results. This
+// should be all that's needed. It subscribes a function to check the (CSV)
+// results of a benchmark. This is done only after verifying that the output
+// strings are really as expected.
+// bm_name_pattern: a name or a regex pattern which will be matched against
+//                  all the benchmark names. Matching benchmarks
+//                  will be the subject of a call to checker_function
+// checker_function: should be of type ResultsCheckFn (see below)
+#define CHECK_BENCHMARK_RESULTS(bm_name_pattern, checker_function) \
+  const size_t CONCAT(dummy, __LINE__) =                           \
+      AddChecker(bm_name_pattern, checker_function)
+
+struct Results;
+typedef std::function<void(Results const&)> ResultsCheckFn;
+
+size_t AddChecker(const std::string& bm_name_pattern, const ResultsCheckFn& fn);
+
+// Class holding the results of a benchmark.
+// It is passed in calls to checker functions.
+struct Results {
+  // the benchmark name
+  std::string name;
+  // the benchmark fields
+  std::map<std::string, std::string> values;
+
+  Results(const std::string& n) : name(n) {}
+
+  int NumThreads() const;
+
+  double NumIterations() const;
+
+  typedef enum : uint8_t { kCpuTime, kRealTime } BenchmarkTime;
+
+  // get cpu_time or real_time in seconds
+  double GetTime(BenchmarkTime which) const;
+
+  // get the real_time duration of the benchmark in seconds.
+  // it is better to use fuzzy float checks for this, as the float
+  // ASCII formatting is lossy.
+  double DurationRealTime() const {
+    return NumIterations() * GetTime(kRealTime);
+  }
+  // get the cpu_time duration of the benchmark in seconds
+  double DurationCPUTime() const { return NumIterations() * GetTime(kCpuTime); }
+
+  // get the string for a result by name, or nullptr if the name
+  // is not found
+  const std::string* Get(const std::string& entry_name) const {
+    auto it = values.find(entry_name);
+    if (it == values.end()) return nullptr;
+    return &it->second;
+  }
+
+  // get a result by name, parsed as a specific type.
+  // NOTE: for counters, use GetCounterAs instead.
+  template <class T>
+  T GetAs(const std::string& entry_name) const;
+
+  // counters are written as doubles, so they have to be read first
+  // as a double, and only then converted to the asked type.
+  template <class T>
+  T GetCounterAs(const std::string& entry_name) const {
+    double dval = GetAs<double>(entry_name);
+    T tval = static_cast<T>(dval);
+    return tval;
+  }
+};
+
+template <class T>
+T Results::GetAs(const std::string& entry_name) const {
+  auto* sv = Get(entry_name);
+  BM_CHECK(sv != nullptr && !sv->empty());
+  std::stringstream ss;
+  ss << *sv;
+  T out;
+  ss >> out;
+  BM_CHECK(!ss.fail());
+  return out;
+}
+
+//----------------------------------
+// Macros to help in result checking. Do not use them with arguments causing
+// side-effects.
+
+// clang-format off
+
+#define CHECK_RESULT_VALUE_IMPL(entry, getfn, var_type, var_name, relationship, value) \
+    CONCAT(BM_CHECK_, relationship)                                        \
+    (entry.getfn< var_type >(var_name), (value)) << "\n"                \
+    << __FILE__ << ":" << __LINE__ << ": " << (entry).name << ":\n"     \
+    << __FILE__ << ":" << __LINE__ << ": "                              \
+    << "expected (" << #var_type << ")" << (var_name)                   \
+    << "=" << (entry).getfn< var_type >(var_name)                       \
+    << " to be " #relationship " to " << (value) << "\n"
+
+// check with tolerance. eps_factor is the tolerance window, which is
+// interpreted relative to value (eg, 0.1 means 10% of value).
+#define CHECK_FLOAT_RESULT_VALUE_IMPL(entry, getfn, var_type, var_name, relationship, value, eps_factor) \
+    CONCAT(BM_CHECK_FLOAT_, relationship)                                  \
+    (entry.getfn< var_type >(var_name), (value), (eps_factor) * (value)) << "\n" \
+    << __FILE__ << ":" << __LINE__ << ": " << (entry).name << ":\n"     \
+    << __FILE__ << ":" << __LINE__ << ": "                              \
+    << "expected (" << #var_type << ")" << (var_name)                   \
+    << "=" << (entry).getfn< var_type >(var_name)                       \
+    << " to be " #relationship " to " << (value) << "\n"                \
+    << __FILE__ << ":" << __LINE__ << ": "                              \
+    << "with tolerance of " << (eps_factor) * (value)                   \
+    << " (" << (eps_factor)*100. << "%), "                              \
+    << "but delta was " << ((entry).getfn< var_type >(var_name) - (value)) \
+    << " (" << (((entry).getfn< var_type >(var_name) - (value))         \
+               /                                                        \
+               ((value) > 1.e-5 || value < -1.e-5 ? value : 1.e-5)*100.) \
+    << "%)"
+
+#define CHECK_RESULT_VALUE(entry, var_type, var_name, relationship, value) \
+    CHECK_RESULT_VALUE_IMPL(entry, GetAs, var_type, var_name, relationship, value)
+
+#define CHECK_COUNTER_VALUE(entry, var_type, var_name, relationship, value) \
+    CHECK_RESULT_VALUE_IMPL(entry, GetCounterAs, var_type, var_name, relationship, value)
+
+#define CHECK_FLOAT_RESULT_VALUE(entry, var_name, relationship, value, eps_factor) \
+    CHECK_FLOAT_RESULT_VALUE_IMPL(entry, GetAs, double, var_name, relationship, value, eps_factor)
+
+#define CHECK_FLOAT_COUNTER_VALUE(entry, var_name, relationship, value, eps_factor) \
+    CHECK_FLOAT_RESULT_VALUE_IMPL(entry, GetCounterAs, double, var_name, relationship, value, eps_factor)
+
+// clang-format on
+
+// ========================================================================= //
+// --------------------------- Misc Utilities ------------------------------ //
+// ========================================================================= //
+
+namespace {
+
+const char* const dec_re = "[0-9]*[.]?[0-9]+([eE][-+][0-9]+)?";
+
+}  //  end namespace
+
+#endif  // TEST_OUTPUT_TEST_H

test/output_test_helper.cc

@@ -0,0 +1,561 @@
+#include <cstdio>
+#include <cstring>
+#include <fstream>
+#include <iostream>
+#include <map>
+#include <memory>
+#include <random>
+#include <sstream>
+#include <streambuf>
+
+#include "../src/benchmark_api_internal.h"
+#include "../src/check.h"  // NOTE: check.h is for internal use only!
+#include "../src/log.h"    // NOTE: log.h is for internal use only
+#include "../src/re.h"     // NOTE: re.h is for internal use only
+#include "output_test.h"
+
+// ========================================================================= //
+// ------------------------------ Internals -------------------------------- //
+// ========================================================================= //
+namespace internal {
+namespace {
+
+using TestCaseList = std::vector<TestCase>;
+
+// Use a vector because the order elements are added matters during iteration.
+// std::map/unordered_map don't guarantee that.
+// For example:
+//  SetSubstitutions({{"%HelloWorld", "Hello"}, {"%Hello", "Hi"}});
+//     Substitute("%HelloWorld") // Always expands to Hello.
+using SubMap = std::vector<std::pair<std::string, std::string>>;
+
+TestCaseList& GetTestCaseList(TestCaseID ID) {
+  // Uses function-local statics to ensure initialization occurs
+  // before first use.
+  static TestCaseList lists[TC_NumID];
+  return lists[ID];
+}
+
+SubMap& GetSubstitutions() {
+  // Don't use 'dec_re' from header because it may not yet be initialized.
+  // clang-format off
+  static std::string safe_dec_re = "[0-9]*[.]?[0-9]+([eE][-+][0-9]+)?";
+  static std::string time_re = "([0-9]+[.])?[0-9]+";
+  static std::string percentage_re = "[0-9]+[.][0-9]{2}";
+  static SubMap map = {
+      {"%float", "[0-9]*[.]?[0-9]+([eE][-+][0-9]+)?"},
+      // human-readable float
+      {"%hrfloat", "[0-9]*[.]?[0-9]+([eE][-+][0-9]+)?[kKMGTPEZYmunpfazy]?i?"},
+      {"%percentage", percentage_re},
+      {"%int", "[ ]*[0-9]+"},
+      {" %s ", "[ ]+"},
+      {"%time", "[ ]*" + time_re + "[ ]+ns"},
+      {"%console_report", "[ ]*" + time_re + "[ ]+ns [ ]*" + time_re + "[ ]+ns [ ]*[0-9]+"},
+      {"%console_percentage_report", "[ ]*" + percentage_re + "[ ]+% [ ]*" + percentage_re + "[ ]+% [ ]*[0-9]+"},
+      {"%console_us_report", "[ ]*" + time_re + "[ ]+us [ ]*" + time_re + "[ ]+us [ ]*[0-9]+"},
+      {"%console_ms_report", "[ ]*" + time_re + "[ ]+ms [ ]*" + time_re + "[ ]+ms [ ]*[0-9]+"},
+      {"%console_s_report", "[ ]*" + time_re + "[ ]+s [ ]*" + time_re + "[ ]+s [ ]*[0-9]+"},
+      {"%console_time_only_report", "[ ]*" + time_re + "[ ]+ns [ ]*" + time_re + "[ ]+ns"},
+      {"%console_us_report", "[ ]*" + time_re + "[ ]+us [ ]*" + time_re + "[ ]+us [ ]*[0-9]+"},
+      {"%console_us_time_only_report", "[ ]*" + time_re + "[ ]+us [ ]*" + time_re + "[ ]+us"},
+      {"%csv_header",
+       "name,iterations,real_time,cpu_time,time_unit,bytes_per_second,"
+       "items_per_second,label,error_occurred,error_message"},
+      {"%csv_report", "[0-9]+," + safe_dec_re + "," + safe_dec_re + ",ns,,,,,"},
+      {"%csv_us_report", "[0-9]+," + safe_dec_re + "," + safe_dec_re + ",us,,,,,"},
+      {"%csv_ms_report", "[0-9]+," + safe_dec_re + "," + safe_dec_re + ",ms,,,,,"},
+      {"%csv_s_report", "[0-9]+," + safe_dec_re + "," + safe_dec_re + ",s,,,,,"},
+      {"%csv_cv_report", "[0-9]+," + safe_dec_re + "," + safe_dec_re + ",,,,,,"},
+      {"%csv_bytes_report",
+       "[0-9]+," + safe_dec_re + "," + safe_dec_re + ",ns," + safe_dec_re + ",,,,"},
+      {"%csv_items_report",
+       "[0-9]+," + safe_dec_re + "," + safe_dec_re + ",ns,," + safe_dec_re + ",,,"},
+      {"%csv_bytes_items_report",
+       "[0-9]+," + safe_dec_re + "," + safe_dec_re + ",ns," + safe_dec_re +
+       "," + safe_dec_re + ",,,"},
+      {"%csv_label_report_begin", "[0-9]+," + safe_dec_re + "," + safe_dec_re + ",ns,,,"},
+      {"%csv_label_report_end", ",,"}};
+  // clang-format on
+  return map;
+}
+
+std::string PerformSubstitutions(std::string source) {
+  SubMap const& subs = GetSubstitutions();
+  using SizeT = std::string::size_type;
+  for (auto const& KV : subs) {
+    SizeT pos = 0;
+    SizeT next_start = 0;
+    while ((pos = source.find(KV.first, next_start)) != std::string::npos) {
+      next_start = pos + KV.second.size();
+      source.replace(pos, KV.first.size(), KV.second);
+    }
+  }
+  return source;
+}
+
+void CheckCase(std::stringstream& remaining_output, TestCase const& TC,
+               TestCaseList const& not_checks) {
+  std::string first_line;
+  bool on_first = true;
+  std::string line;
+  while (!remaining_output.eof()) {
+    BM_CHECK(remaining_output.good());
+    std::getline(remaining_output, line);
+    if (on_first) {
+      first_line = line;
+      on_first = false;
+    }
+    for (const auto& NC : not_checks) {
+      BM_CHECK(!NC.regex->Match(line))
+          << "Unexpected match for line \"" << line << "\" for MR_Not regex \""
+          << NC.regex_str << "\""
+          << "\n    actual regex string \"" << TC.substituted_regex << "\""
+          << "\n    started matching near: " << first_line;
+    }
+    if (TC.regex->Match(line)) {
+      return;
+    }
+    BM_CHECK(TC.match_rule != MR_Next)
+        << "Expected line \"" << line << "\" to match regex \"" << TC.regex_str
+        << "\""
+        << "\n    actual regex string \"" << TC.substituted_regex << "\""
+        << "\n    started matching near: " << first_line;
+  }
+  BM_CHECK(remaining_output.eof() == false)
+      << "End of output reached before match for regex \"" << TC.regex_str
+      << "\" was found"
+      << "\n    actual regex string \"" << TC.substituted_regex << "\""
+      << "\n    started matching near: " << first_line;
+}
+
+void CheckCases(TestCaseList const& checks, std::stringstream& output) {
+  std::vector<TestCase> not_checks;
+  for (size_t i = 0; i < checks.size(); ++i) {
+    const auto& TC = checks[i];
+    if (TC.match_rule == MR_Not) {
+      not_checks.push_back(TC);
+      continue;
+    }
+    CheckCase(output, TC, not_checks);
+    not_checks.clear();
+  }
+}
+
+class TestReporter : public benchmark::BenchmarkReporter {
+ public:
+  TestReporter(std::vector<benchmark::BenchmarkReporter*> reps)
+      : reporters_(std::move(reps)) {}
+
+  bool ReportContext(const Context& context) override {
+    bool last_ret = false;
+    bool first = true;
+    for (auto* rep : reporters_) {
+      bool new_ret = rep->ReportContext(context);
+      BM_CHECK(first || new_ret == last_ret)
+          << "Reports return different values for ReportContext";
+      first = false;
+      last_ret = new_ret;
+    }
+    (void)first;
+    return last_ret;
+  }
+
+  void ReportRuns(const std::vector<Run>& report) override {
+    for (auto* rep : reporters_) {
+      rep->ReportRuns(report);
+    }
+  }
+  void Finalize() override {
+    for (auto* rep : reporters_) {
+      rep->Finalize();
+    }
+  }
+
+ private:
+  std::vector<benchmark::BenchmarkReporter*> reporters_;
+};
+}  // namespace
+
+}  // end namespace internal
+
+// ========================================================================= //
+// -------------------------- Results checking ----------------------------- //
+// ========================================================================= //
+
+namespace internal {
+
+// Utility class to manage subscribers for checking benchmark results.
+// It works by parsing the CSV output to read the results.
+class ResultsChecker {
+ public:
+  struct PatternAndFn : public TestCase {  // reusing TestCase for its regexes
+    PatternAndFn(const std::string& rx, ResultsCheckFn fn_)
+        : TestCase(rx), fn(std::move(fn_)) {}
+    ResultsCheckFn fn;
+  };
+
+  std::vector<PatternAndFn> check_patterns;
+  std::vector<Results> results;
+  std::vector<std::string> field_names;
+
+  void Add(const std::string& entry_pattern, const ResultsCheckFn& fn);
+
+  void CheckResults(std::stringstream& output);
+
+ private:
+  void SetHeader_(const std::string& csv_header);
+  void SetValues_(const std::string& entry_csv_line);
+
+  std::vector<std::string> SplitCsv_(const std::string& line) const;
+};
+
+namespace {
+// store the static ResultsChecker in a function to prevent initialization
+// order problems
+ResultsChecker& GetResultsChecker() {
+  static ResultsChecker rc;
+  return rc;
+}
+}  // end namespace
+
+// add a results checker for a benchmark
+void ResultsChecker::Add(const std::string& entry_pattern,
+                         const ResultsCheckFn& fn) {
+  check_patterns.emplace_back(entry_pattern, fn);
+}
+
+// check the results of all subscribed benchmarks
+void ResultsChecker::CheckResults(std::stringstream& output) {
+  // first reset the stream to the start
+  {
+    auto start = std::stringstream::pos_type(0);
+    // clear before calling tellg()
+    output.clear();
+    // seek to zero only when needed
+    if (output.tellg() > start) {
+      output.seekg(start);
+    }
+    // and just in case
+    output.clear();
+  }
+  // now go over every line and publish it to the ResultsChecker
+  std::string line;
+  bool on_first = true;
+  while (!output.eof()) {
+    BM_CHECK(output.good());
+    std::getline(output, line);
+    if (on_first) {
+      SetHeader_(line);  // this is important
+      on_first = false;
+      continue;
+    }
+    SetValues_(line);
+  }
+  // finally we can call the subscribed check functions
+  for (const auto& p : check_patterns) {
+    BM_VLOG(2) << "--------------------------------\n";
+    BM_VLOG(2) << "checking for benchmarks matching " << p.regex_str << "...\n";
+    for (const auto& r : results) {
+      if (!p.regex->Match(r.name)) {
+        BM_VLOG(2) << p.regex_str << " is not matched by " << r.name << "\n";
+        continue;
+      }
+      BM_VLOG(2) << p.regex_str << " is matched by " << r.name << "\n";
+      BM_VLOG(1) << "Checking results of " << r.name << ": ... \n";
+      p.fn(r);
+      BM_VLOG(1) << "Checking results of " << r.name << ": OK.\n";
+    }
+  }
+}
+
+// prepare for the names in this header
+void ResultsChecker::SetHeader_(const std::string& csv_header) {
+  field_names = SplitCsv_(csv_header);
+}
+
+// set the values for a benchmark
+void ResultsChecker::SetValues_(const std::string& entry_csv_line) {
+  if (entry_csv_line.empty()) {
+    return;
+  }  // some lines are empty
+  BM_CHECK(!field_names.empty());
+  auto vals = SplitCsv_(entry_csv_line);
+  BM_CHECK_EQ(vals.size(), field_names.size());
+  results.emplace_back(vals[0]);  // vals[0] is the benchmark name
+  auto& entry = results.back();
+  for (size_t i = 1, e = vals.size(); i < e; ++i) {
+    entry.values[field_names[i]] = vals[i];
+  }
+}
+
+// a quick'n'dirty csv splitter (eliminating quotes)
+std::vector<std::string> ResultsChecker::SplitCsv_(
+    const std::string& line) const {
+  std::vector<std::string> out;
+  if (line.empty()) {
+    return out;
+  }
+  if (!field_names.empty()) {
+    out.reserve(field_names.size());
+  }
+  size_t prev = 0;
+  size_t pos = line.find_first_of(',');
+  size_t curr = pos;
+  while (pos != std::string::npos) {
+    BM_CHECK(curr > 0);
+    if (line[prev] == '"') {
+      ++prev;
+    }
+    if (line[curr - 1] == '"') {
+      --curr;
+    }
+    out.push_back(line.substr(prev, curr - prev));
+    prev = pos + 1;
+    pos = line.find_first_of(',', pos + 1);
+    curr = pos;
+  }
+  curr = line.size();
+  if (line[prev] == '"') {
+    ++prev;
+  }
+  if (line[curr - 1] == '"') {
+    --curr;
+  }
+  out.push_back(line.substr(prev, curr - prev));
+  return out;
+}
+
+}  // end namespace internal
+
+size_t AddChecker(const std::string& bm_name, const ResultsCheckFn& fn) {
+  auto& rc = internal::GetResultsChecker();
+  rc.Add(bm_name, fn);
+  return rc.results.size();
+}
+
+int Results::NumThreads() const {
+  auto pos = name.find("/threads:");
+  if (pos == std::string::npos) {
+    return 1;
+  }
+  auto end = name.find('/', pos + 9);
+  std::stringstream ss;
+  ss << name.substr(pos + 9, end);
+  int num = 1;
+  ss >> num;
+  BM_CHECK(!ss.fail());
+  return num;
+}
+
+double Results::NumIterations() const { return GetAs<double>("iterations"); }
+
+double Results::GetTime(BenchmarkTime which) const {
+  BM_CHECK(which == kCpuTime || which == kRealTime);
+  const char* which_str = which == kCpuTime ? "cpu_time" : "real_time";
+  double val = GetAs<double>(which_str);
+  const auto* unit = Get("time_unit");
+  BM_CHECK(unit);
+  if (*unit == "ns") {
+    return val * 1.e-9;
+  }
+  if (*unit == "us") {
+    return val * 1.e-6;
+  }
+  if (*unit == "ms") {
+    return val * 1.e-3;
+  }
+  if (*unit == "s") {
+    return val;
+  }
+  BM_CHECK(1 == 0) << "unknown time unit: " << *unit;
+  return 0;
+}
+
+// ========================================================================= //
+// -------------------------- Public API Definitions------------------------ //
+// ========================================================================= //
+
+TestCase::TestCase(std::string re, int rule)
+    : regex_str(std::move(re)),
+      match_rule(rule),
+      substituted_regex(internal::PerformSubstitutions(regex_str)),
+      regex(std::make_shared<benchmark::Regex>()) {
+  std::string err_str;
+  regex->Init(substituted_regex, &err_str);
+  BM_CHECK(err_str.empty())
+      << "Could not construct regex \"" << substituted_regex << "\""
+      << "\n    originally \"" << regex_str << "\""
+      << "\n    got error: " << err_str;
+}
+
+int AddCases(TestCaseID ID, std::initializer_list<TestCase> il) {
+  auto& L = internal::GetTestCaseList(ID);
+  L.insert(L.end(), il);
+  return 0;
+}
+
+int SetSubstitutions(
+    std::initializer_list<std::pair<std::string, std::string>> il) {
+  auto& subs = internal::GetSubstitutions();
+  for (auto KV : il) {
+    bool exists = false;
+    KV.second = internal::PerformSubstitutions(KV.second);
+    for (auto& EKV : subs) {
+      if (EKV.first == KV.first) {
+        EKV.second = std::move(KV.second);
+        exists = true;
+        break;
+      }
+    }
+    if (!exists) {
+      subs.push_back(std::move(KV));
+    }
+  }
+  return 0;
+}
+
+// Disable deprecated warnings temporarily because we need to reference
+// CSVReporter but don't want to trigger -Werror=-Wdeprecated-declarations
+BENCHMARK_DISABLE_DEPRECATED_WARNING
+
+void RunOutputTests(int argc, char* argv[]) {
+  using internal::GetTestCaseList;
+  benchmark::Initialize(&argc, argv);
+  auto options = benchmark::internal::GetOutputOptions(/*force_no_color*/ true);
+  benchmark::ConsoleReporter CR(options);
+  benchmark::JSONReporter JR;
+  benchmark::CSVReporter CSVR;
+  struct ReporterTest {
+    std::string name;
+    std::vector<TestCase>& output_cases;
+    std::vector<TestCase>& error_cases;
+    benchmark::BenchmarkReporter& reporter;
+    std::stringstream out_stream;
+    std::stringstream err_stream;
+
+    ReporterTest(const std::string& n, std::vector<TestCase>& out_tc,
+                 std::vector<TestCase>& err_tc,
+                 benchmark::BenchmarkReporter& br)
+        : name(n), output_cases(out_tc), error_cases(err_tc), reporter(br) {
+      reporter.SetOutputStream(&out_stream);
+      reporter.SetErrorStream(&err_stream);
+    }
+  } TestCases[] = {
+      {std::string("ConsoleReporter"), GetTestCaseList(TC_ConsoleOut),
+       GetTestCaseList(TC_ConsoleErr), CR},
+      {std::string("JSONReporter"), GetTestCaseList(TC_JSONOut),
+       GetTestCaseList(TC_JSONErr), JR},
+      {std::string("CSVReporter"), GetTestCaseList(TC_CSVOut),
+       GetTestCaseList(TC_CSVErr), CSVR},
+  };
+
+  // Create the test reporter and run the benchmarks.
+  std::cout << "Running benchmarks...\n";
+  internal::TestReporter test_rep({&CR, &JR, &CSVR});
+  benchmark::RunSpecifiedBenchmarks(&test_rep);
+
+  for (auto& rep_test : TestCases) {
+    std::string msg =
+        std::string("\nTesting ") + rep_test.name + std::string(" Output\n");
+    std::string banner(msg.size() - 1, '-');
+    std::cout << banner << msg << banner << "\n";
+
+    std::cerr << rep_test.err_stream.str();
+    std::cout << rep_test.out_stream.str();
+
+    internal::CheckCases(rep_test.error_cases, rep_test.err_stream);
+    internal::CheckCases(rep_test.output_cases, rep_test.out_stream);
+
+    std::cout << "\n";
+  }
+
+  // now that we know the output is as expected, we can dispatch
+  // the checks to subscribees.
+  auto& csv = TestCases[2];
+  // would use == but gcc spits a warning
+  BM_CHECK(csv.name == std::string("CSVReporter"));
+  internal::GetResultsChecker().CheckResults(csv.out_stream);
+}
+
+BENCHMARK_RESTORE_DEPRECATED_WARNING
+
+int SubstrCnt(const std::string& haystack, const std::string& pat) {
+  if (pat.length() == 0) {
+    return 0;
+  }
+  int count = 0;
+  for (size_t offset = haystack.find(pat); offset != std::string::npos;
+       offset = haystack.find(pat, offset + pat.length())) {
+    ++count;
+  }
+  return count;
+}
+
+namespace {
+char ToHex(int ch) {
+  return ch < 10 ? static_cast<char>('0' + ch)
+                 : static_cast<char>('a' + (ch - 10));
+}
+
+char RandomHexChar() {
+  static std::mt19937 rd{std::random_device{}()};
+  static std::uniform_int_distribution<int> mrand{0, 15};
+  return ToHex(mrand(rd));
+}
+
+std::string GetRandomFileName() {
+  std::string model = "test.%%%%%%";
+  for (auto& ch : model) {
+    if (ch == '%') {
+      ch = RandomHexChar();
+    }
+  }
+  return model;
+}
+
+bool FileExists(std::string const& name) {
+  std::ifstream in(name.c_str());
+  return in.good();
+}
+
+std::string GetTempFileName() {
+  // This function attempts to avoid race conditions where two tests
+  // create the same file at the same time. However, it still introduces races
+  // similar to tmpnam.
+  int retries = 3;
+  while (--retries != 0) {
+    std::string name = GetRandomFileName();
+    if (!FileExists(name)) {
+      return name;
+    }
+  }
+  std::cerr << "Failed to create unique temporary file name\n";
+  std::flush(std::cerr);
+  std::exit(1);
+}
+}  // end namespace
+
+std::string GetFileReporterOutput(int argc, char* argv[]) {
+  std::vector<char*> new_argv(argv, argv + argc);
+  assert(static_cast<decltype(new_argv)::size_type>(argc) == new_argv.size());
+
+  std::string tmp_file_name = GetTempFileName();
+  std::cout << "Will be using this as the tmp file: " << tmp_file_name << '\n';
+
+  std::string tmp = "--benchmark_out=";
+  tmp += tmp_file_name;
+  new_argv.emplace_back(const_cast<char*>(tmp.c_str()));
+
+  argc = static_cast<int>(new_argv.size());
+
+  benchmark::Initialize(&argc, new_argv.data());
+  benchmark::RunSpecifiedBenchmarks();
+
+  // Read the output back from the file, and delete the file.
+  std::ifstream tmp_stream(tmp_file_name);
+  std::string output = std::string((std::istreambuf_iterator<char>(tmp_stream)),
+                                   std::istreambuf_iterator<char>());
+  std::remove(tmp_file_name.c_str());
+
+  return output;
+}

test/overload_test.cc

@@ -0,0 +1,37 @@
+#include "benchmark/macros.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+
+namespace {
+// Simulate an overloaded function name.
+// This version does nothing and is just here to create ambiguity for
+// MyOverloadedBenchmark.
+BENCHMARK_UNUSED void MyOverloadedBenchmark() {}
+
+// This is the actual benchmark function we want to register.
+// It has the signature void(benchmark::State&) required by the library.
+void MyOverloadedBenchmark(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+}
+
+// This macro invocation should compile correctly if benchmark.h
+// contains the fix (using static_cast), but would fail to compile
+// if the benchmark name were ambiguous (e.g., when using + or no cast
+// with an overloaded function).
+BENCHMARK(MyOverloadedBenchmark);
+
+// Also test BENCHMARK_TEMPLATE with an overloaded name.
+template <int N>
+void MyTemplatedOverloadedBenchmark() {}
+
+template <int N>
+void MyTemplatedOverloadedBenchmark(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+}
+
+BENCHMARK_TEMPLATE(MyTemplatedOverloadedBenchmark, 1);
+}  // end namespace
+
+BENCHMARK_MAIN();

test/perf_counters_gtest.cc

@@ -0,0 +1,401 @@
+#include <mutex>
+#include <random>
+#include <set>
+#include <string>
+#include <thread>
+#include <vector>
+
+#include "../src/perf_counters.h"
+#include "gmock/gmock.h"
+#include "gtest/gtest.h"
+
+#ifndef GTEST_SKIP
+struct MsgHandler {
+  void operator=(std::ostream&) {}
+};
+#define GTEST_SKIP() return MsgHandler() = std::cout
+#endif
+
+using benchmark::internal::PerfCounters;
+using benchmark::internal::PerfCountersMeasurement;
+using benchmark::internal::PerfCounterValues;
+using ::testing::AllOf;
+using ::testing::Gt;
+using ::testing::Lt;
+
+namespace {
+const char kGenericPerfEvent1[] = "CYCLES";
+const char kGenericPerfEvent2[] = "INSTRUCTIONS";
+
+std::set<std::string> UniqueCounterNames(const PerfCounters& counters) {
+  return {counters.names().begin(), counters.names().end()};
+}
+
+bool HasRequiredPerfCounters(const std::vector<std::string>& names) {
+  if (!PerfCounters::kSupported) {
+    return false;
+  }
+  auto counters = PerfCounters::Create(names);
+  auto actual_names = UniqueCounterNames(counters);
+  for (const auto& name : names) {
+    if (actual_names.find(name) == actual_names.end()) {
+      return false;
+    }
+  }
+  return true;
+}
+
+TEST(PerfCountersTest, Init) {
+  EXPECT_EQ(PerfCounters::Initialize(), PerfCounters::kSupported);
+}
+
+TEST(PerfCountersTest, OneCounter) {
+  if (!HasRequiredPerfCounters({kGenericPerfEvent1})) {
+    GTEST_SKIP() << "Requested performance counters are not available.";
+  }
+  auto counter = PerfCounters::Create({kGenericPerfEvent1});
+  EXPECT_EQ(UniqueCounterNames(counter).size(), 1);
+}
+
+TEST(PerfCountersTest, NegativeTest) {
+  if (!PerfCounters::kSupported) {
+    EXPECT_FALSE(PerfCounters::Initialize());
+    return;
+  }
+  if (!HasRequiredPerfCounters({kGenericPerfEvent2, kGenericPerfEvent1})) {
+    GTEST_SKIP() << "Requested performance counters are not available.";
+  }
+  // Safety checks
+  // Create() will always create a valid object, even if passed no or
+  // wrong arguments as the new behavior is to warn and drop unsupported
+  // counters
+  EXPECT_EQ(PerfCounters::Create({}).num_counters(), 0);
+  EXPECT_EQ(PerfCounters::Create({""}).num_counters(), 0);
+  EXPECT_EQ(PerfCounters::Create({"not a counter name"}).num_counters(), 0);
+  {
+    // Try sneaking in a bad egg to see if it is filtered out. The
+    // number of counters has to be two, not zero
+    auto counter =
+        PerfCounters::Create({kGenericPerfEvent2, "", kGenericPerfEvent1});
+    auto names = UniqueCounterNames(counter);
+    EXPECT_EQ(names.size(), 2);
+    EXPECT_EQ(names,
+              std::set<std::string>({kGenericPerfEvent2, kGenericPerfEvent1}));
+  }
+  {
+    // Try sneaking in an outrageous counter, like a fat finger mistake
+    auto counter = PerfCounters::Create(
+        {kGenericPerfEvent2, "not a counter name", kGenericPerfEvent1});
+    auto names = UniqueCounterNames(counter);
+    EXPECT_EQ(names.size(), 2);
+    EXPECT_EQ(names,
+              std::set<std::string>({kGenericPerfEvent2, kGenericPerfEvent1}));
+  }
+  {
+    // Finally try a golden input - it should like both of them
+    EXPECT_EQ(UniqueCounterNames(PerfCounters::Create(
+                                     {kGenericPerfEvent1, kGenericPerfEvent2}))
+                  .size(),
+              2);
+  }
+  {
+    // Add a bad apple in the end of the chain to check the edges
+    auto counter = PerfCounters::Create(
+        {kGenericPerfEvent1, kGenericPerfEvent2, "bad event name"});
+    auto names = UniqueCounterNames(counter);
+    EXPECT_EQ(names.size(), 2);
+    EXPECT_EQ(names,
+              std::set<std::string>({kGenericPerfEvent1, kGenericPerfEvent2}));
+  }
+}
+
+static std::map<std::string, uint64_t> SnapshotAndCombine(
+    PerfCounters& counters) {
+  PerfCounterValues values(counters.num_counters());
+  std::map<std::string, uint64_t> value_map;
+
+  if (counters.Snapshot(&values)) {
+    for (size_t i = 0; i != counters.num_counters(); ++i) {
+      value_map[counters.names()[i]] += values[i];
+    }
+  }
+  return value_map;
+}
+
+TEST(PerfCountersTest, Read1Counter) {
+  if (!HasRequiredPerfCounters({kGenericPerfEvent1})) {
+    GTEST_SKIP() << "Requested performance counters are not available.";
+  }
+  auto counters = PerfCounters::Create({kGenericPerfEvent1});
+  auto values1 = SnapshotAndCombine(counters);
+  EXPECT_EQ(values1.size(), 1);
+  EXPECT_GT(values1.begin()->second, 0);
+  auto values2 = SnapshotAndCombine(counters);
+  EXPECT_EQ(values2.size(), 1);
+  EXPECT_GT(values2.begin()->second, 0);
+  EXPECT_GT(values2.begin()->second, values1.begin()->second);
+}
+
+TEST(PerfCountersTest, Read1CounterEachCPU) {
+  if (!HasRequiredPerfCounters({kGenericPerfEvent1})) {
+    GTEST_SKIP() << "Requested performance counters are not available.";
+  }
+#ifdef __linux__
+  cpu_set_t saved_set;
+  if (sched_getaffinity(0, sizeof(saved_set), &saved_set) != 0) {
+    // This can happen e.g. if there are more than CPU_SETSIZE CPUs.
+    GTEST_SKIP() << "Could not save CPU affinity mask.";
+  }
+
+  for (size_t cpu = 0; cpu != CPU_SETSIZE; ++cpu) {
+    cpu_set_t set;
+    CPU_ZERO(&set);
+    CPU_SET(cpu, &set);
+    if (sched_setaffinity(0, sizeof(set), &set) != 0) {
+      break;
+    }
+
+    auto counters = PerfCounters::Create({kGenericPerfEvent1});
+    auto values1 = SnapshotAndCombine(counters);
+    EXPECT_EQ(values1.size(), 1);
+    EXPECT_GT(values1.begin()->second, 0);
+    auto values2 = SnapshotAndCombine(counters);
+    EXPECT_EQ(values2.size(), 1);
+    EXPECT_GT(values2.begin()->second, 0);
+    EXPECT_GT(values2.begin()->second, values1.begin()->second);
+  }
+
+  EXPECT_EQ(sched_setaffinity(0, sizeof(saved_set), &saved_set), 0);
+#else
+  GTEST_SKIP() << "Test skipped on non-Linux.";
+#endif
+}
+
+TEST(PerfCountersTest, Read2Counters) {
+  if (!HasRequiredPerfCounters({kGenericPerfEvent1, kGenericPerfEvent2})) {
+    GTEST_SKIP() << "Requested performance counters are not available.";
+  }
+  auto counters =
+      PerfCounters::Create({kGenericPerfEvent1, kGenericPerfEvent2});
+  auto values1 = SnapshotAndCombine(counters);
+  EXPECT_EQ(values1.size(), 2);
+  for (auto& kv : values1) {
+    EXPECT_GT(kv.second, 0);
+  }
+  auto values2 = SnapshotAndCombine(counters);
+  EXPECT_EQ(values1.size(), 2);
+  for (auto& kv : values2) {
+    EXPECT_GT(kv.second, 0);
+    EXPECT_GT(kv.second, values1[kv.first]);
+  }
+}
+
+TEST(PerfCountersTest, ReopenExistingCounters) {
+  // This test works in recent and old Intel hardware, Pixel 3, and Pixel 6.
+  // However we cannot make assumptions beyond 2 HW counters due to Pixel 6.
+  if (!HasRequiredPerfCounters({kGenericPerfEvent1})) {
+    GTEST_SKIP() << "Requested performance counters are not available.";
+  }
+  std::vector<std::string> kMetrics({kGenericPerfEvent1});
+  std::vector<PerfCounters> counters(2);
+  for (auto& counter : counters) {
+    counter = PerfCounters::Create(kMetrics);
+  }
+  PerfCounterValues values(counters[0].num_counters());
+  EXPECT_TRUE(counters[0].Snapshot(&values));
+  EXPECT_TRUE(counters[1].Snapshot(&values));
+}
+
+TEST(PerfCountersTest, CreateExistingMeasurements) {
+  // The test works (i.e. causes read to fail) for the assumptions
+  // about hardware capabilities (i.e. small number (2) hardware
+  // counters) at this date,
+  // the same as previous test ReopenExistingCounters.
+  if (!PerfCounters::kSupported) {
+    GTEST_SKIP() << "Test skipped because libpfm is not supported.";
+  }
+
+  // This means we will try 10 counters but we can only guarantee
+  // for sure at this time that only 3 will work. Perhaps in the future
+  // we could use libpfm to query for the hardware limits on this
+  // particular platform.
+  const int kMaxCounters = 10;
+  const int kMinValidCounters = 2;
+
+  // Let's use a ubiquitous counter that is guaranteed to work
+  // on all platforms
+  const std::vector<std::string> kMetrics{"cycles"};
+  if (!HasRequiredPerfCounters(kMetrics)) {
+    GTEST_SKIP() << "Requested performance counters are not available.";
+  }
+
+  // Cannot create a vector of actual objects because the
+  // copy constructor of PerfCounters is deleted - and so is
+  // implicitly deleted on PerfCountersMeasurement too
+  std::vector<std::unique_ptr<PerfCountersMeasurement>>
+      perf_counter_measurements;
+
+  perf_counter_measurements.reserve(kMaxCounters);
+  for (int j = 0; j < kMaxCounters; ++j) {
+    perf_counter_measurements.emplace_back(
+        new PerfCountersMeasurement(kMetrics));
+  }
+
+  std::vector<std::pair<std::string, double>> measurements;
+
+  // Start all counters together to see if they hold
+  size_t max_counters = kMaxCounters;
+  for (size_t i = 0; i < kMaxCounters; ++i) {
+    auto& counter(*perf_counter_measurements[i]);
+    std::set<std::string> names{counter.names().begin(), counter.names().end()};
+    EXPECT_EQ(names.size(), 1);
+    if (!counter.Start()) {
+      max_counters = i;
+      break;
+    };
+  }
+
+  ASSERT_GE(max_counters, kMinValidCounters);
+
+  // Start all together
+  for (size_t i = 0; i < max_counters; ++i) {
+    auto& counter(*perf_counter_measurements[i]);
+    EXPECT_TRUE(counter.Stop(measurements) || (i >= kMinValidCounters));
+  }
+
+  // Start/stop individually
+  for (size_t i = 0; i < max_counters; ++i) {
+    auto& counter(*perf_counter_measurements[i]);
+    measurements.clear();
+    counter.Start();
+    EXPECT_TRUE(counter.Stop(measurements) || (i >= kMinValidCounters));
+  }
+}
+
+// We try to do some meaningful work here but the compiler
+// insists in optimizing away our loop so we had to add a
+// no-optimize macro. In case it fails, we added some entropy
+// to this pool as well.
+
+BENCHMARK_DONT_OPTIMIZE size_t do_work() {
+  static std::mt19937 rd{std::random_device{}()};
+  static std::uniform_int_distribution<size_t> mrand(0, 10);
+  const size_t kNumLoops = 1000000;
+  size_t sum = 0;
+  for (size_t j = 0; j < kNumLoops; ++j) {
+    sum += mrand(rd);
+  }
+  benchmark::DoNotOptimize(sum);
+  return sum;
+}
+
+void measure(size_t threadcount, std::map<std::string, uint64_t>* before,
+             std::map<std::string, uint64_t>* after) {
+  BM_CHECK_NE(before, nullptr);
+  BM_CHECK_NE(after, nullptr);
+  std::vector<std::thread> threads(threadcount);
+  // Because we do not care whether the threads execute concurrently, but we do
+  // care that they do all of their work between the SnapshotAndCombine calls,
+  // we serialize them with a mutex. See
+  // https://github.com/google/benchmark/issues/2173.
+  std::mutex mutex;
+  auto work = [&mutex]() {
+    mutex.lock();
+    BM_CHECK(do_work() > 1000);
+    mutex.unlock();
+  };
+
+  // We need to first set up the counters, then start the threads, so the
+  // threads would inherit the counters. But later, we need to first destroy
+  // the thread pool (so all the work finishes), then measure the counters. So
+  // the scopes overlap, and we need to explicitly control the scope of the
+  // threadpool.
+  auto counters =
+      PerfCounters::Create({kGenericPerfEvent1, kGenericPerfEvent2});
+  mutex.lock();
+  for (auto& t : threads) {
+    t = std::thread(work);
+  }
+  *before = SnapshotAndCombine(counters);
+  mutex.unlock();
+  for (auto& t : threads) {
+    t.join();
+  }
+  *after = SnapshotAndCombine(counters);
+}
+
+TEST(PerfCountersTest, MultiThreaded) {
+  if (!HasRequiredPerfCounters({kGenericPerfEvent1, kGenericPerfEvent2})) {
+    GTEST_SKIP() << "Requested performance counters are not available.";
+  }
+  std::map<std::string, uint64_t> before, after;
+
+  // Notice that this test will work even if we taskset it to a single CPU
+  // In this case the threads will run sequentially
+  // Start two threads and measure the number of combined cycles and
+  // instructions
+  measure(2, &before, &after);
+  std::vector<double> Elapsed2Threads{
+      static_cast<double>(after[kGenericPerfEvent1] -
+                          before[kGenericPerfEvent1]),
+      static_cast<double>(after[kGenericPerfEvent2] -
+                          before[kGenericPerfEvent2])};
+
+  // Start four threads and measure the number of combined cycles and
+  // instructions
+  measure(4, &before, &after);
+  std::vector<double> Elapsed4Threads{
+      static_cast<double>(after[kGenericPerfEvent1] -
+                          before[kGenericPerfEvent1]),
+      static_cast<double>(after[kGenericPerfEvent2] -
+                          before[kGenericPerfEvent2])};
+
+  // The following expectations fail (at least on a beefy workstation with lots
+  // of cpus) - it seems that in some circumstances the runtime of 4 threads
+  // can even be better than with 2.
+  // So instead of expecting 4 threads to be slower, let's just make sure they
+  // do not differ too much in general (one is not more than 10x than the
+  // other).
+  EXPECT_THAT(Elapsed4Threads[0] / Elapsed2Threads[0], AllOf(Gt(0.1), Lt(10)));
+  EXPECT_THAT(Elapsed4Threads[1] / Elapsed2Threads[1], AllOf(Gt(0.1), Lt(10)));
+}
+
+TEST(PerfCountersTest, HardwareLimits) {
+  // The test works (i.e. causes read to fail) for the assumptions
+  // about hardware capabilities (i.e. small number (3-4) hardware
+  // counters) at this date,
+  // the same as previous test ReopenExistingCounters.
+  if (!PerfCounters::kSupported) {
+    GTEST_SKIP() << "Test skipped because libpfm is not supported.";
+  }
+  EXPECT_TRUE(PerfCounters::Initialize());
+
+  // Taken from `perf list`, but focusses only on those HW events that actually
+  // were reported when running `sudo perf stat -a sleep 10`, intersected over
+  // several platforms. All HW events listed in the first command not reported
+  // in the second seem to not work. This is sad as we don't really get to test
+  // the grouping here (groups can contain up to 6 members)...
+  std::vector<std::string> counter_names{
+      "cycles",         // leader
+      "instructions",   //
+      "branch-misses",  //
+  };
+
+  // In the off-chance that some of these values are not supported,
+  // we filter them out so the test will complete without failure
+  // albeit it might not actually test the grouping on that platform
+  std::vector<std::string> valid_names;
+  for (const std::string& name : counter_names) {
+    if (PerfCounters::IsCounterSupported(name)) {
+      valid_names.push_back(name);
+    }
+  }
+  PerfCountersMeasurement counter(valid_names);
+
+  std::vector<std::pair<std::string, double>> measurements;
+
+  counter.Start();
+  EXPECT_TRUE(counter.Stop(measurements));
+}
+
+}  // namespace

test/perf_counters_test.cc

@@ -0,0 +1,123 @@
+#include <cstdarg>
+#include <set>
+#include <string>
+#include <vector>
+#undef NDEBUG
+
+#include "../src/commandlineflags.h"
+#include "../src/perf_counters.h"
+#include "benchmark/benchmark_api.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+#include "benchmark/utils.h"
+#include "output_test.h"
+
+namespace benchmark {
+
+BM_DECLARE_string(benchmark_perf_counters);
+
+}  // namespace benchmark
+namespace {
+const char kGenericPerfEvent1[] = "CYCLES";
+const char kGenericPerfEvent2[] = "INSTRUCTIONS";
+
+std::set<std::string> UniqueCounterNames(
+    const benchmark::internal::PerfCounters& counters) {
+  return {counters.names().begin(), counters.names().end()};
+}
+
+bool HasRequiredPerfCounters(const std::vector<std::string>& names) {
+  if (!benchmark::internal::PerfCounters::kSupported) {
+    return false;
+  }
+  auto counters = benchmark::internal::PerfCounters::Create(names);
+  auto actual_names = UniqueCounterNames(counters);
+  for (const auto& name : names) {
+    if (actual_names.find(name) == actual_names.end()) {
+      return false;
+    }
+  }
+  return true;
+}
+
+void BM_Simple(benchmark::State& state) {
+  for (auto _ : state) {
+    auto iterations = double(state.iterations()) * double(state.iterations());
+    benchmark::DoNotOptimize(iterations);
+  }
+}
+BENCHMARK(BM_Simple);
+ADD_CASES(TC_JSONOut, {{"\"name\": \"BM_Simple\",$"}});
+
+const int kIters = 1000000;
+
+void BM_WithoutPauseResume(benchmark::State& state) {
+  int n = 0;
+
+  for (auto _ : state) {
+    for (auto i = 0; i < kIters; ++i) {
+      n = 1 - n;
+      benchmark::DoNotOptimize(n);
+    }
+  }
+}
+
+BENCHMARK(BM_WithoutPauseResume);
+ADD_CASES(TC_JSONOut, {{"\"name\": \"BM_WithoutPauseResume\",$"}});
+
+void BM_WithPauseResume(benchmark::State& state) {
+  int m = 0, n = 0;
+
+  for (auto _ : state) {
+    for (auto i = 0; i < kIters; ++i) {
+      n = 1 - n;
+      benchmark::DoNotOptimize(n);
+    }
+
+    state.PauseTiming();
+    for (auto j = 0; j < kIters; ++j) {
+      m = 1 - m;
+      benchmark::DoNotOptimize(m);
+    }
+    state.ResumeTiming();
+  }
+}
+
+BENCHMARK(BM_WithPauseResume);
+
+ADD_CASES(TC_JSONOut, {{"\"name\": \"BM_WithPauseResume\",$"}});
+
+static void CheckSimple(Results const& e) {
+  CHECK_COUNTER_VALUE(e, double, kGenericPerfEvent1, GT, 0);
+}
+
+double withoutPauseResumeInstrCount = 0.0;
+double withPauseResumeInstrCount = 0.0;
+
+void SaveInstrCountWithoutResume(Results const& e) {
+  withoutPauseResumeInstrCount = e.GetAs<double>(kGenericPerfEvent2);
+}
+
+void SaveInstrCountWithResume(Results const& e) {
+  withPauseResumeInstrCount = e.GetAs<double>(kGenericPerfEvent2);
+}
+
+CHECK_BENCHMARK_RESULTS("BM_Simple", &CheckSimple);
+CHECK_BENCHMARK_RESULTS("BM_WithoutPauseResume", &SaveInstrCountWithoutResume);
+CHECK_BENCHMARK_RESULTS("BM_WithPauseResume", &SaveInstrCountWithResume);
+}  // end namespace
+
+int main(int argc, char* argv[]) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+  if (!HasRequiredPerfCounters({kGenericPerfEvent1, kGenericPerfEvent2})) {
+    return 0;
+  }
+  benchmark::FLAGS_benchmark_perf_counters =
+      std::string(kGenericPerfEvent1) + "," + kGenericPerfEvent2;
+  benchmark::internal::PerfCounters::Initialize();
+  RunOutputTests(argc, argv);
+
+  BM_CHECK_GT(withPauseResumeInstrCount, kIters);
+  BM_CHECK_GT(withoutPauseResumeInstrCount, kIters);
+  BM_CHECK_LT(withPauseResumeInstrCount, 1.5 * withoutPauseResumeInstrCount);
+}

test/profiler_manager_gtest.cc

@@ -0,0 +1,45 @@
+#include <memory>
+
+#include "benchmark/managers.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+#include "benchmark/utils.h"
+#include "gtest/gtest.h"
+
+namespace {
+
+class TestProfilerManager : public benchmark::ProfilerManager {
+ public:
+  void AfterSetupStart() override { ++start_called; }
+  void BeforeTeardownStop() override { ++stop_called; }
+
+  int start_called = 0;
+  int stop_called = 0;
+};
+
+void BM_empty(benchmark::State& state) {
+  for (auto _ : state) {
+    auto iterations = state.iterations();
+    benchmark::DoNotOptimize(iterations);
+  }
+}
+BENCHMARK(BM_empty);
+
+TEST(ProfilerManager, ReregisterManager) {
+#if GTEST_HAS_DEATH_TEST
+  // Tests only runnable in debug mode (when BM_CHECK is enabled).
+#ifndef NDEBUG
+#ifndef TEST_BENCHMARK_LIBRARY_HAS_NO_ASSERTIONS
+  ASSERT_DEATH_IF_SUPPORTED(
+      {
+        std::unique_ptr<TestProfilerManager> pm(new TestProfilerManager());
+        benchmark::RegisterProfilerManager(pm.get());
+        benchmark::RegisterProfilerManager(pm.get());
+      },
+      "RegisterProfilerManager");
+#endif
+#endif
+#endif
+}
+
+}  // namespace

test/profiler_manager_iterations_test.cc

@@ -0,0 +1,66 @@
+#include <cassert>
+#include <cstdlib>
+#include <memory>
+#include <vector>
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/managers.h"
+#include "benchmark/registration.h"
+#include "benchmark/reporter.h"
+#include "benchmark/state.h"
+
+// Tests that we can specify the number of profiler iterations with
+// --benchmark_min_time=<NUM>x.
+namespace {
+
+int iteration_count = 0;
+int end_profiler_iteration_count = 0;
+
+class TestProfilerManager : public benchmark::ProfilerManager {
+  void AfterSetupStart() override { iteration_count = 0; }
+  void BeforeTeardownStop() override {
+    end_profiler_iteration_count = iteration_count;
+  }
+};
+
+class NullReporter : public benchmark::BenchmarkReporter {
+ public:
+  bool ReportContext(const Context& /*context*/) override { return true; }
+  void ReportRuns(const std::vector<Run>& /* report */) override {}
+};
+
+void BM_MyBench(benchmark::State& state) {
+  for (auto s : state) {
+    ++iteration_count;
+  }
+}
+BENCHMARK(BM_MyBench);
+}  // end namespace
+
+int main(int argc, char** argv) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+  // Make a fake argv and append the new --benchmark_profiler_iterations=<foo>
+  // to it.
+  int fake_argc = argc + 1;
+  std::vector<const char*> fake_argv(static_cast<size_t>(fake_argc));
+  for (size_t i = 0; i < static_cast<size_t>(argc); ++i) {
+    fake_argv[i] = argv[i];
+  }
+  fake_argv[static_cast<size_t>(argc)] = "--benchmark_min_time=4x";
+
+  std::unique_ptr<benchmark::ProfilerManager> pm(new TestProfilerManager());
+  benchmark::RegisterProfilerManager(pm.get());
+
+  benchmark::Initialize(&fake_argc, const_cast<char**>(fake_argv.data()));
+
+  NullReporter null_reporter;
+  const size_t returned_count =
+      benchmark::RunSpecifiedBenchmarks(&null_reporter, "BM_MyBench");
+  assert(returned_count == 1);
+
+  // Check the executed iters.
+  assert(end_profiler_iteration_count == 4);
+
+  benchmark::RegisterProfilerManager(nullptr);
+  return 0;
+}

test/profiler_manager_test.cc

@@ -0,0 +1,58 @@
+// FIXME: WIP
+
+#include <cassert>
+#include <memory>
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/managers.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+#include "benchmark/utils.h"
+#include "output_test.h"
+
+namespace {
+class TestProfilerManager : public benchmark::ProfilerManager {
+ public:
+  void AfterSetupStart() override { ++start_called; }
+  void BeforeTeardownStop() override { ++stop_called; }
+
+  int start_called = 0;
+  int stop_called = 0;
+};
+
+void BM_empty(benchmark::State& state) {
+  for (auto _ : state) {
+    auto iterations = state.iterations();
+    benchmark::DoNotOptimize(iterations);
+  }
+}
+BENCHMARK(BM_empty);
+
+ADD_CASES(TC_ConsoleOut, {{"^BM_empty %console_report$"}});
+ADD_CASES(TC_JSONOut, {{"\"name\": \"BM_empty\",$"},
+                       {"\"family_index\": 0,$", MR_Next},
+                       {"\"per_family_instance_index\": 0,$", MR_Next},
+                       {"\"run_name\": \"BM_empty\",$", MR_Next},
+                       {"\"run_type\": \"iteration\",$", MR_Next},
+                       {"\"repetitions\": 1,$", MR_Next},
+                       {"\"repetition_index\": 0,$", MR_Next},
+                       {"\"threads\": 1,$", MR_Next},
+                       {"\"iterations\": %int,$", MR_Next},
+                       {"\"real_time\": %float,$", MR_Next},
+                       {"\"cpu_time\": %float,$", MR_Next},
+                       {"\"time_unit\": \"ns\"$", MR_Next},
+                       {"}", MR_Next}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_empty\",%csv_report$"}});
+}  // end namespace
+
+int main(int argc, char* argv[]) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+  std::unique_ptr<TestProfilerManager> pm(new TestProfilerManager());
+
+  benchmark::RegisterProfilerManager(pm.get());
+  RunOutputTests(argc, argv);
+  benchmark::RegisterProfilerManager(nullptr);
+
+  assert(pm->start_called == 1);
+  assert(pm->stop_called == 1);
+}

test/register_benchmark_test.cc

@@ -0,0 +1,208 @@
+
+#undef NDEBUG
+#include <cassert>
+#include <vector>
+
+#include "../src/check.h"  // NOTE: check.h is for internal use only!
+#include "benchmark/benchmark_api.h"
+#include "benchmark/registration.h"
+#include "benchmark/reporter.h"
+#include "benchmark/state.h"
+
+namespace {
+
+class TestReporter : public benchmark::ConsoleReporter {
+ public:
+  void ReportRuns(const std::vector<Run>& report) override {
+    all_runs_.insert(all_runs_.end(), begin(report), end(report));
+    ConsoleReporter::ReportRuns(report);
+  }
+
+  std::vector<Run> all_runs_;
+};
+
+struct TestCase {
+  const std::string name;
+  const std::string label;
+  // Note: not explicit as we rely on it being converted through ADD_CASES.
+  TestCase(const std::string& xname) : TestCase(xname, "") {}
+  TestCase(const std::string& xname, const std::string& xlabel)
+      : name(xname), label(xlabel) {}
+
+  typedef benchmark::BenchmarkReporter::Run Run;
+
+  void CheckRun(Run const& run) const {
+    // clang-format off
+    BM_CHECK(name == run.benchmark_name()) << "expected " << name << " got "
+                                      << run.benchmark_name();
+    if (!label.empty()) {
+      BM_CHECK(run.report_label == label) << "expected " << label << " got "
+                                       << run.report_label;
+    } else {
+      BM_CHECK(run.report_label.empty());
+    }
+    // clang-format on
+  }
+};
+
+std::vector<TestCase> ExpectedResults;
+
+int AddCases(std::initializer_list<TestCase> const& v) {
+  for (const auto& N : v) {
+    ExpectedResults.push_back(N);
+  }
+  return 0;
+}
+
+#define CONCAT(x, y) CONCAT2(x, y)
+#define CONCAT2(x, y) x##y
+#define ADD_CASES(...) \
+  const int CONCAT(dummy, __LINE__) = AddCases({__VA_ARGS__})
+
+using ReturnVal = benchmark::Benchmark const* const;
+
+//----------------------------------------------------------------------------//
+// Test RegisterBenchmark with no additional arguments
+//----------------------------------------------------------------------------//
+void BM_function(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+}
+BENCHMARK(BM_function);
+ReturnVal dummy = benchmark::RegisterBenchmark(
+    "BM_function_manual_registration", BM_function);
+ADD_CASES({"BM_function"}, {"BM_function_manual_registration"});
+
+//----------------------------------------------------------------------------//
+// Test RegisterBenchmark with additional arguments
+// Note: GCC <= 4.8 do not support this form of RegisterBenchmark because they
+//       reject the variadic pack expansion of lambda captures.
+//----------------------------------------------------------------------------//
+
+void BM_extra_args(benchmark::State& st, const char* label) {
+  for (auto _ : st) {
+  }
+  st.SetLabel(label);
+}
+int RegisterFromFunction() {
+  std::pair<const char*, const char*> cases[] = {
+      {"test1", "One"}, {"test2", "Two"}, {"test3", "Three"}};
+  for (auto const& c : cases) {
+    benchmark::RegisterBenchmark(c.first, &BM_extra_args, c.second);
+  }
+  return 0;
+}
+const int dummy2 = RegisterFromFunction();
+ADD_CASES({"test1", "One"}, {"test2", "Two"}, {"test3", "Three"});
+
+//----------------------------------------------------------------------------//
+// Test RegisterBenchmark with DISABLED_ benchmark
+//----------------------------------------------------------------------------//
+void DISABLED_BM_function(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+}
+BENCHMARK(DISABLED_BM_function);
+ReturnVal dummy3 = benchmark::RegisterBenchmark("DISABLED_BM_function_manual",
+                                                DISABLED_BM_function);
+// No need to add cases because we don't expect them to run.
+
+//----------------------------------------------------------------------------//
+// Test BENCHMARK_NAMED: verifies name format "func/test_case_name" and that
+// chaining (e.g. ->Threads()) works, without introducing a lambda.
+//----------------------------------------------------------------------------//
+void BM_named(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+}
+BENCHMARK_NAMED(BM_named, variant_a);
+BENCHMARK_NAMED(BM_named, variant_b);
+BENCHMARK_NAMED(BM_named, variant_c)->Threads(2);
+ADD_CASES({"BM_named/variant_a"}, {"BM_named/variant_b"},
+          {"BM_named/variant_c/threads:2"});
+
+//----------------------------------------------------------------------------//
+// Test RegisterBenchmark with different callable types
+//----------------------------------------------------------------------------//
+
+struct CustomFixture {
+  void operator()(benchmark::State& st) {
+    for (auto _ : st) {
+    }
+  }
+};
+
+void TestRegistrationAtRuntime() {
+  {
+    CustomFixture fx;
+    benchmark::RegisterBenchmark("custom_fixture", fx);
+    AddCases({std::string("custom_fixture")});
+  }
+  {
+    const char* x = "42";
+    auto capturing_lam = [=](benchmark::State& st) {
+      for (auto _ : st) {
+      }
+      st.SetLabel(x);
+    };
+    benchmark::RegisterBenchmark("lambda_benchmark", capturing_lam);
+    AddCases({{"lambda_benchmark", x}});
+  }
+}
+
+// Test that all benchmarks, registered at either during static init or runtime,
+// are run and the results are passed to the reported.
+void RunTestOne() {
+  TestRegistrationAtRuntime();
+
+  TestReporter test_reporter;
+  benchmark::RunSpecifiedBenchmarks(&test_reporter);
+
+  typedef benchmark::BenchmarkReporter::Run Run;
+  auto EB = ExpectedResults.begin();
+
+  for (Run const& run : test_reporter.all_runs_) {
+    assert(EB != ExpectedResults.end());
+    EB->CheckRun(run);
+    ++EB;
+  }
+  assert(EB == ExpectedResults.end());
+}
+
+// Test that ClearRegisteredBenchmarks() clears all previously registered
+// benchmarks.
+// Also test that new benchmarks can be registered and ran afterwards.
+void RunTestTwo() {
+  assert(!ExpectedResults.empty() &&
+         "must have at least one registered benchmark");
+  ExpectedResults.clear();
+  benchmark::ClearRegisteredBenchmarks();
+
+  TestReporter test_reporter;
+  size_t num_ran = benchmark::RunSpecifiedBenchmarks(&test_reporter);
+  assert(num_ran == 0);
+  assert(test_reporter.all_runs_.begin() == test_reporter.all_runs_.end());
+
+  TestRegistrationAtRuntime();
+  num_ran = benchmark::RunSpecifiedBenchmarks(&test_reporter);
+  assert(num_ran == ExpectedResults.size());
+
+  typedef benchmark::BenchmarkReporter::Run Run;
+  auto EB = ExpectedResults.begin();
+
+  for (Run const& run : test_reporter.all_runs_) {
+    assert(EB != ExpectedResults.end());
+    EB->CheckRun(run);
+    ++EB;
+  }
+  assert(EB == ExpectedResults.end());
+}
+}  // end namespace
+
+int main(int argc, char* argv[]) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+  benchmark::Initialize(&argc, argv);
+
+  RunTestOne();
+  RunTestTwo();
+}

test/repetitions_test.cc

@@ -0,0 +1,258 @@
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+#include "output_test.h"
+
+namespace {
+// ========================================================================= //
+// ------------------------ Testing Basic Output --------------------------- //
+// ========================================================================= //
+
+void BM_ExplicitRepetitions(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+}
+BENCHMARK(BM_ExplicitRepetitions)->Repetitions(2);
+
+ADD_CASES(TC_ConsoleOut,
+          {{"^BM_ExplicitRepetitions/repeats:2 %console_report$"}});
+ADD_CASES(TC_ConsoleOut,
+          {{"^BM_ExplicitRepetitions/repeats:2 %console_report$"}});
+ADD_CASES(TC_ConsoleOut,
+          {{"^BM_ExplicitRepetitions/repeats:2_mean %console_report$"}});
+ADD_CASES(TC_ConsoleOut,
+          {{"^BM_ExplicitRepetitions/repeats:2_median %console_report$"}});
+ADD_CASES(TC_ConsoleOut,
+          {{"^BM_ExplicitRepetitions/repeats:2_stddev %console_report$"}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_ExplicitRepetitions/repeats:2\",$"},
+           {"\"family_index\": 0,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_ExplicitRepetitions/repeats:2\",$", MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 2,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\"$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_ExplicitRepetitions/repeats:2\",$"},
+           {"\"family_index\": 0,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_ExplicitRepetitions/repeats:2\",$", MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 2,$", MR_Next},
+           {"\"repetition_index\": 1,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\"$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_ExplicitRepetitions/repeats:2_mean\",$"},
+           {"\"family_index\": 0,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_ExplicitRepetitions/repeats:2\",$", MR_Next},
+           {"\"run_type\": \"aggregate\",$", MR_Next},
+           {"\"repetitions\": 2,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"aggregate_name\": \"mean\",$", MR_Next},
+           {"\"aggregate_unit\": \"time\",$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\"$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_ExplicitRepetitions/repeats:2_median\",$"},
+           {"\"family_index\": 0,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_ExplicitRepetitions/repeats:2\",$", MR_Next},
+           {"\"run_type\": \"aggregate\",$", MR_Next},
+           {"\"repetitions\": 2,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"aggregate_name\": \"median\",$", MR_Next},
+           {"\"aggregate_unit\": \"time\",$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\"$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_ExplicitRepetitions/repeats:2_stddev\",$"},
+           {"\"family_index\": 0,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_ExplicitRepetitions/repeats:2\",$", MR_Next},
+           {"\"run_type\": \"aggregate\",$", MR_Next},
+           {"\"repetitions\": 2,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"aggregate_name\": \"stddev\",$", MR_Next},
+           {"\"aggregate_unit\": \"time\",$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\"$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_ExplicitRepetitions/repeats:2\",%csv_report$"}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_ExplicitRepetitions/repeats:2\",%csv_report$"}});
+ADD_CASES(TC_CSVOut,
+          {{"^\"BM_ExplicitRepetitions/repeats:2_mean\",%csv_report$"}});
+ADD_CASES(TC_CSVOut,
+          {{"^\"BM_ExplicitRepetitions/repeats:2_median\",%csv_report$"}});
+ADD_CASES(TC_CSVOut,
+          {{"^\"BM_ExplicitRepetitions/repeats:2_stddev\",%csv_report$"}});
+
+// ========================================================================= //
+// ------------------------ Testing Basic Output --------------------------- //
+// ========================================================================= //
+
+void BM_ImplicitRepetitions(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+}
+BENCHMARK(BM_ImplicitRepetitions);
+
+ADD_CASES(TC_ConsoleOut, {{"^BM_ImplicitRepetitions %console_report$"}});
+ADD_CASES(TC_ConsoleOut, {{"^BM_ImplicitRepetitions %console_report$"}});
+ADD_CASES(TC_ConsoleOut, {{"^BM_ImplicitRepetitions %console_report$"}});
+ADD_CASES(TC_ConsoleOut, {{"^BM_ImplicitRepetitions_mean %console_report$"}});
+ADD_CASES(TC_ConsoleOut, {{"^BM_ImplicitRepetitions_median %console_report$"}});
+ADD_CASES(TC_ConsoleOut, {{"^BM_ImplicitRepetitions_stddev %console_report$"}});
+ADD_CASES(TC_JSONOut, {{"\"name\": \"BM_ImplicitRepetitions\",$"},
+                       {"\"family_index\": 1,$", MR_Next},
+                       {"\"per_family_instance_index\": 0,$", MR_Next},
+                       {"\"run_name\": \"BM_ImplicitRepetitions\",$", MR_Next},
+                       {"\"run_type\": \"iteration\",$", MR_Next},
+                       {"\"repetitions\": 3,$", MR_Next},
+                       {"\"repetition_index\": 0,$", MR_Next},
+                       {"\"threads\": 1,$", MR_Next},
+                       {"\"iterations\": %int,$", MR_Next},
+                       {"\"real_time\": %float,$", MR_Next},
+                       {"\"cpu_time\": %float,$", MR_Next},
+                       {"\"time_unit\": \"ns\"$", MR_Next},
+                       {"}", MR_Next}});
+ADD_CASES(TC_JSONOut, {{"\"name\": \"BM_ImplicitRepetitions\",$"},
+                       {"\"family_index\": 1,$", MR_Next},
+                       {"\"per_family_instance_index\": 0,$", MR_Next},
+                       {"\"run_name\": \"BM_ImplicitRepetitions\",$", MR_Next},
+                       {"\"run_type\": \"iteration\",$", MR_Next},
+                       {"\"repetitions\": 3,$", MR_Next},
+                       {"\"repetition_index\": 1,$", MR_Next},
+                       {"\"threads\": 1,$", MR_Next},
+                       {"\"iterations\": %int,$", MR_Next},
+                       {"\"real_time\": %float,$", MR_Next},
+                       {"\"cpu_time\": %float,$", MR_Next},
+                       {"\"time_unit\": \"ns\"$", MR_Next},
+                       {"}", MR_Next}});
+ADD_CASES(TC_JSONOut, {{"\"name\": \"BM_ImplicitRepetitions\",$"},
+                       {"\"family_index\": 1,$", MR_Next},
+                       {"\"per_family_instance_index\": 0,$", MR_Next},
+                       {"\"run_name\": \"BM_ImplicitRepetitions\",$", MR_Next},
+                       {"\"run_type\": \"iteration\",$", MR_Next},
+                       {"\"repetitions\": 3,$", MR_Next},
+                       {"\"repetition_index\": 2,$", MR_Next},
+                       {"\"threads\": 1,$", MR_Next},
+                       {"\"iterations\": %int,$", MR_Next},
+                       {"\"real_time\": %float,$", MR_Next},
+                       {"\"cpu_time\": %float,$", MR_Next},
+                       {"\"time_unit\": \"ns\"$", MR_Next},
+                       {"}", MR_Next}});
+ADD_CASES(TC_JSONOut, {{"\"name\": \"BM_ImplicitRepetitions_mean\",$"},
+                       {"\"family_index\": 1,$", MR_Next},
+                       {"\"per_family_instance_index\": 0,$", MR_Next},
+                       {"\"run_name\": \"BM_ImplicitRepetitions\",$", MR_Next},
+                       {"\"run_type\": \"aggregate\",$", MR_Next},
+                       {"\"repetitions\": 3,$", MR_Next},
+                       {"\"threads\": 1,$", MR_Next},
+                       {"\"aggregate_name\": \"mean\",$", MR_Next},
+                       {"\"aggregate_unit\": \"time\",$", MR_Next},
+                       {"\"iterations\": %int,$", MR_Next},
+                       {"\"real_time\": %float,$", MR_Next},
+                       {"\"cpu_time\": %float,$", MR_Next},
+                       {"\"time_unit\": \"ns\"$", MR_Next},
+                       {"}", MR_Next}});
+ADD_CASES(TC_JSONOut, {{"\"name\": \"BM_ImplicitRepetitions_median\",$"},
+                       {"\"family_index\": 1,$", MR_Next},
+                       {"\"per_family_instance_index\": 0,$", MR_Next},
+                       {"\"run_name\": \"BM_ImplicitRepetitions\",$", MR_Next},
+                       {"\"run_type\": \"aggregate\",$", MR_Next},
+                       {"\"repetitions\": 3,$", MR_Next},
+                       {"\"threads\": 1,$", MR_Next},
+                       {"\"aggregate_name\": \"median\",$", MR_Next},
+                       {"\"aggregate_unit\": \"time\",$", MR_Next},
+                       {"\"iterations\": %int,$", MR_Next},
+                       {"\"real_time\": %float,$", MR_Next},
+                       {"\"cpu_time\": %float,$", MR_Next},
+                       {"\"time_unit\": \"ns\"$", MR_Next},
+                       {"}", MR_Next}});
+ADD_CASES(TC_JSONOut, {{"\"name\": \"BM_ImplicitRepetitions_stddev\",$"},
+                       {"\"family_index\": 1,$", MR_Next},
+                       {"\"per_family_instance_index\": 0,$", MR_Next},
+                       {"\"run_name\": \"BM_ImplicitRepetitions\",$", MR_Next},
+                       {"\"run_type\": \"aggregate\",$", MR_Next},
+                       {"\"repetitions\": 3,$", MR_Next},
+                       {"\"threads\": 1,$", MR_Next},
+                       {"\"aggregate_name\": \"stddev\",$", MR_Next},
+                       {"\"aggregate_unit\": \"time\",$", MR_Next},
+                       {"\"iterations\": %int,$", MR_Next},
+                       {"\"real_time\": %float,$", MR_Next},
+                       {"\"cpu_time\": %float,$", MR_Next},
+                       {"\"time_unit\": \"ns\"$", MR_Next},
+                       {"}", MR_Next}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_ImplicitRepetitions\",%csv_report$"}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_ImplicitRepetitions\",%csv_report$"}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_ImplicitRepetitions_mean\",%csv_report$"}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_ImplicitRepetitions_median\",%csv_report$"}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_ImplicitRepetitions_stddev\",%csv_report$"}});
+
+// ========================================================================= //
+// --------------------- Testing Skipped Repetitions ------------------------ //
+// ========================================================================= //
+
+void BM_SkippedFirstRepetition(benchmark::State& state) {
+  static int repetition_index = 0;
+  if (repetition_index++ == 0) {
+    state.SkipWithError("skipped first repetition");
+    return;
+  }
+  for (auto _ : state) {
+  }
+}
+BENCHMARK(BM_SkippedFirstRepetition)->Repetitions(3);
+
+ADD_CASES(
+    TC_ConsoleOut,
+    {{"^BM_SkippedFirstRepetition/repeats:3[ ]+ERROR OCCURRED: "
+      "'skipped first repetition'$"},
+     {"^BM_SkippedFirstRepetition/repeats:3 %console_report$", MR_Next},
+     {"^BM_SkippedFirstRepetition/repeats:3 %console_report$", MR_Next},
+     {"^BM_SkippedFirstRepetition/repeats:3_mean %console_report$", MR_Next},
+     {"^BM_SkippedFirstRepetition/repeats:3_median %console_report$", MR_Next},
+     {"^BM_SkippedFirstRepetition/repeats:3_stddev %console_report$",
+      MR_Next}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_SkippedFirstRepetition/repeats:3_mean\",$"},
+           {"\"run_type\": \"aggregate\",$"},
+           {"\"repetitions\": 3,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"aggregate_name\": \"mean\",$", MR_Next},
+           {"\"aggregate_unit\": \"time\",$", MR_Next},
+           {"\"iterations\": 2,$", MR_Next}});
+ADD_CASES(TC_CSVOut,
+          {{"^\"BM_SkippedFirstRepetition/repeats:3_mean\",2,%float,%float,"
+            "ns,,,,,$"}});
+}  // end namespace
+
+// ========================================================================= //
+// --------------------------- TEST CASES END ------------------------------ //
+// ========================================================================= //
+
+int main(int argc, char* argv[]) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+  RunOutputTests(argc, argv);
+}

test/report_aggregates_only_test.cc

@@ -0,0 +1,46 @@
+
+#undef NDEBUG
+#include <cstdio>
+#include <string>
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+#include "output_test.h"
+
+namespace {
+// Ok this test is super ugly. We want to check what happens with the file
+// reporter in the presence of ReportAggregatesOnly().
+// We do not care about console output, the normal tests check that already.
+
+void BM_SummaryRepeat(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+}
+BENCHMARK(BM_SummaryRepeat)->Repetitions(3)->ReportAggregatesOnly();
+}  // end namespace
+
+int main(int argc, char* argv[]) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+  const std::string output = GetFileReporterOutput(argc, argv);
+
+  if (SubstrCnt(output, "\"name\": \"BM_SummaryRepeat/repeats:3") != 4 ||
+      SubstrCnt(output, "\"name\": \"BM_SummaryRepeat/repeats:3_mean\"") != 1 ||
+      SubstrCnt(output, "\"name\": \"BM_SummaryRepeat/repeats:3_median\"") !=
+          1 ||
+      SubstrCnt(output, "\"name\": \"BM_SummaryRepeat/repeats:3_stddev\"") !=
+          1 ||
+      SubstrCnt(output, "\"name\": \"BM_SummaryRepeat/repeats:3_cv\"") != 1) {
+    std::cout << "Precondition mismatch. Expected to only find four "
+                 "occurrences of \"BM_SummaryRepeat/repeats:3\" substring:\n"
+                 "\"name\": \"BM_SummaryRepeat/repeats:3_mean\", "
+                 "\"name\": \"BM_SummaryRepeat/repeats:3_median\", "
+                 "\"name\": \"BM_SummaryRepeat/repeats:3_stddev\", "
+                 "\"name\": \"BM_SummaryRepeat/repeats:3_cv\"\nThe entire "
+                 "output:\n";
+    std::cout << output;
+    return 1;
+  }
+
+  return 0;
+}

test/scoped_pause_test.cc

@@ -0,0 +1,30 @@
+
+#include <chrono>
+#include <thread>
+
+#include "benchmark/benchmark.h"
+#include "output_test.h"
+
+// BM_ScopedPause sleeps for 10ms in a ScopedPauseTiming block.
+// The reported time should be much less than 10ms.
+void BM_ScopedPause(benchmark::State& state) {
+  for (auto _ : state) {
+    benchmark::ScopedPauseTiming pause(state);
+    std::this_thread::sleep_for(std::chrono::milliseconds(10));
+    state.SetIterationTime(0.0);
+  }
+}
+BENCHMARK(BM_ScopedPause)->UseManualTime()->Iterations(1);
+
+void CheckResults(Results const& results) {
+  // Check that the real time is much less than the 10ms sleep time.
+  // Allow for up to 1ms of timing noise/overhead.
+  CHECK_FLOAT_RESULT_VALUE(results, "real_time", LT, 1e6, 0.0);
+}
+CHECK_BENCHMARK_RESULTS("BM_ScopedPause", &CheckResults);
+
+int main(int argc, char* argv[]) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+  RunOutputTests(argc, argv);
+  return 0;
+}

test/skip_with_error_test.cc

@@ -0,0 +1,215 @@
+
+#undef NDEBUG
+#include <cassert>
+#include <vector>
+
+#include "../src/check.h"  // NOTE: check.h is for internal use only!
+#include "benchmark/benchmark_api.h"
+#include "benchmark/registration.h"
+#include "benchmark/reporter.h"
+#include "benchmark/state.h"
+#include "benchmark/utils.h"
+
+namespace {
+
+class TestReporter : public benchmark::ConsoleReporter {
+ public:
+  bool ReportContext(const Context& context) override {
+    return ConsoleReporter::ReportContext(context);
+  };
+
+  void ReportRuns(const std::vector<Run>& report) override {
+    all_runs_.insert(all_runs_.end(), begin(report), end(report));
+    ConsoleReporter::ReportRuns(report);
+  }
+
+  TestReporter() {}
+  ~TestReporter() override {}
+
+  mutable std::vector<Run> all_runs_;
+};
+
+struct TestCase {
+  std::string name;
+  bool error_occurred;
+  std::string error_message;
+
+  typedef benchmark::BenchmarkReporter::Run Run;
+
+  void CheckRun(Run const& run) const {
+    BM_CHECK(name == run.benchmark_name())
+        << "expected " << name << " got " << run.benchmark_name();
+    BM_CHECK_EQ(error_occurred,
+                benchmark::internal::SkippedWithError == run.skipped);
+    BM_CHECK(error_message == run.skip_message);
+    if (error_occurred) {
+      // BM_CHECK(run.iterations == 0);
+    } else {
+      BM_CHECK(run.iterations != 0);
+    }
+  }
+};
+
+// NOLINTNEXTLINE(cppcoreguidelines-avoid-non-const-global-variables)
+std::vector<TestCase> ExpectedResults;
+
+int AddCases(const std::string& base_name,
+             std::initializer_list<TestCase> const& v) {
+  for (auto TC : v) {
+    TC.name = base_name + TC.name;
+    ExpectedResults.push_back(std::move(TC));
+  }
+  return 0;
+}
+
+#define CONCAT(x, y) CONCAT2(x, y)
+#define CONCAT2(x, y) x##y
+#define ADD_CASES(...) const int CONCAT(dummy, __LINE__) = AddCases(__VA_ARGS__)
+
+void BM_error_no_running(benchmark::State& state) {
+  state.SkipWithError("error message");
+}
+BENCHMARK(BM_error_no_running);
+ADD_CASES("BM_error_no_running", {{"", true, "error message"}});
+
+void BM_error_before_running(benchmark::State& state) {
+  state.SkipWithError("error message");
+  while (state.KeepRunning()) {
+    assert(false);
+  }
+}
+BENCHMARK(BM_error_before_running);
+ADD_CASES("BM_error_before_running", {{"", true, "error message"}});
+
+void BM_error_before_running_batch(benchmark::State& state) {
+  state.SkipWithError("error message");
+  while (state.KeepRunningBatch(17)) {
+    assert(false);
+  }
+}
+BENCHMARK(BM_error_before_running_batch);
+ADD_CASES("BM_error_before_running_batch", {{"", true, "error message"}});
+
+void BM_error_before_running_range_for(benchmark::State& state) {
+  state.SkipWithError("error message");
+  for (auto _ : state) {
+    assert(false);
+  }
+}
+BENCHMARK(BM_error_before_running_range_for);
+ADD_CASES("BM_error_before_running_range_for", {{"", true, "error message"}});
+
+void BM_error_during_running(benchmark::State& state) {
+  int first_iter = 1;
+  while (state.KeepRunning()) {
+    if (state.range(0) == 1 && state.thread_index() <= (state.threads() / 2)) {
+      assert(first_iter);
+      first_iter = 0;
+      state.SkipWithError("error message");
+    } else {
+      state.PauseTiming();
+      state.ResumeTiming();
+    }
+  }
+}
+BENCHMARK(BM_error_during_running)->Arg(1)->Arg(2)->ThreadRange(1, 8);
+ADD_CASES("BM_error_during_running", {{"/1/threads:1", true, "error message"},
+                                      {"/1/threads:2", true, "error message"},
+                                      {"/1/threads:4", true, "error message"},
+                                      {"/1/threads:8", true, "error message"},
+                                      {"/2/threads:1", false, ""},
+                                      {"/2/threads:2", false, ""},
+                                      {"/2/threads:4", false, ""},
+                                      {"/2/threads:8", false, ""}});
+
+void BM_error_during_running_ranged_for(benchmark::State& state) {
+  assert(state.max_iterations > 3 && "test requires at least a few iterations");
+  bool first_iter = true;
+  // NOTE: Users should not write the for loop explicitly.
+  for (auto It = state.begin(), End = state.end(); It != End; ++It) {
+    if (state.range(0) == 1) {
+      assert(first_iter);
+      first_iter = false;
+      (void)first_iter;
+      state.SkipWithError("error message");
+      // Test the unfortunate but documented behavior that the ranged-for loop
+      // doesn't automatically terminate when SkipWithError is set.
+      assert(++It != End);
+      break;  // Required behavior
+    }
+  }
+}
+BENCHMARK(BM_error_during_running_ranged_for)->Arg(1)->Arg(2)->Iterations(5);
+ADD_CASES("BM_error_during_running_ranged_for",
+          {{"/1/iterations:5", true, "error message"},
+           {"/2/iterations:5", false, ""}});
+
+void BM_error_after_running(benchmark::State& state) {
+  for (auto _ : state) {
+    auto iterations = static_cast<double>(state.iterations()) *
+                      static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(iterations);
+  }
+  if (state.thread_index() <= (state.threads() / 2)) {
+    state.SkipWithError("error message");
+  }
+}
+BENCHMARK(BM_error_after_running)->ThreadRange(1, 8);
+ADD_CASES("BM_error_after_running", {{"/threads:1", true, "error message"},
+                                     {"/threads:2", true, "error message"},
+                                     {"/threads:4", true, "error message"},
+                                     {"/threads:8", true, "error message"}});
+
+void BM_error_while_paused(benchmark::State& state) {
+  bool first_iter = true;
+  while (state.KeepRunning()) {
+    if (state.range(0) == 1 && state.thread_index() <= (state.threads() / 2)) {
+      assert(first_iter);
+      first_iter = false;
+      state.PauseTiming();
+      state.SkipWithError("error message");
+    } else {
+      state.PauseTiming();
+      state.ResumeTiming();
+    }
+  }
+}
+BENCHMARK(BM_error_while_paused)->Arg(1)->Arg(2)->ThreadRange(1, 8);
+ADD_CASES("BM_error_while_paused", {{"/1/threads:1", true, "error message"},
+                                    {"/1/threads:2", true, "error message"},
+                                    {"/1/threads:4", true, "error message"},
+                                    {"/1/threads:8", true, "error message"},
+                                    {"/2/threads:1", false, ""},
+                                    {"/2/threads:2", false, ""},
+                                    {"/2/threads:4", false, ""},
+                                    {"/2/threads:8", false, ""}});
+
+void BM_malformed(benchmark::State& /*unused*/) {
+  // NOTE: empty body wanted. No thing else.
+}
+BENCHMARK(BM_malformed);
+ADD_CASES("BM_malformed",
+          {{"", true,
+            "The benchmark didn't run, nor was it explicitly skipped. Please "
+            "call 'SkipWithXXX` in your benchmark as appropriate."}});
+}  // end namespace
+
+int main(int argc, char* argv[]) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+  benchmark::Initialize(&argc, argv);
+
+  TestReporter test_reporter;
+  benchmark::RunSpecifiedBenchmarks(&test_reporter);
+
+  typedef benchmark::BenchmarkReporter::Run Run;
+  auto EB = ExpectedResults.begin();
+
+  for (Run const& run : test_reporter.all_runs_) {
+    assert(EB != ExpectedResults.end());
+    EB->CheckRun(run);
+    ++EB;
+  }
+  assert(EB == ExpectedResults.end());
+
+  return 0;
+}

test/spec_arg_test.cc

@@ -0,0 +1,110 @@
+#include <algorithm>
+#include <cassert>
+#include <cstdint>
+#include <cstdlib>
+#include <cstring>
+#include <iostream>
+#include <limits>
+#include <string>
+#include <vector>
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/registration.h"
+#include "benchmark/reporter.h"
+#include "benchmark/state.h"
+
+// Tests that we can override benchmark-spec value from FLAGS_benchmark_filter
+// with argument to RunSpecifiedBenchmarks(...).
+
+namespace {
+
+class TestReporter : public benchmark::ConsoleReporter {
+ public:
+  bool ReportContext(const Context& context) override {
+    return ConsoleReporter::ReportContext(context);
+  };
+
+  void ReportRuns(const std::vector<Run>& report) override {
+    assert(report.size() == 1);
+    matched_functions.push_back(report[0].run_name.function_name);
+    ConsoleReporter::ReportRuns(report);
+  };
+
+  TestReporter() {}
+
+  ~TestReporter() override {}
+
+  const std::vector<std::string>& GetMatchedFunctions() const {
+    return matched_functions;
+  }
+
+ private:
+  std::vector<std::string> matched_functions;
+};
+
+void BM_NotChosen(benchmark::State& state) {
+  assert(false && "SHOULD NOT BE CALLED");
+  for (auto _ : state) {
+  }
+}
+BENCHMARK(BM_NotChosen);
+
+void BM_Chosen(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+}
+BENCHMARK(BM_Chosen);
+
+}  // end namespace
+
+int main(int argc, char** argv) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+
+  const std::string flag = "BM_NotChosen";
+
+  // Verify that argv specify --benchmark_filter=BM_NotChosen.
+  bool found = false;
+  for (int i = 0; i < argc; ++i) {
+    if (strcmp("--benchmark_filter=BM_NotChosen", argv[i]) == 0) {
+      found = true;
+      break;
+    }
+  }
+  assert(found);
+
+  benchmark::Initialize(&argc, argv);
+
+  // Check that the current flag value is reported accurately via the
+  // GetBenchmarkFilter() function.
+  if (flag != benchmark::GetBenchmarkFilter()) {
+    std::cerr
+        << "Seeing different value for flags. GetBenchmarkFilter() returns ["
+        << benchmark::GetBenchmarkFilter() << "] expected flag=[" << flag
+        << "]\n";
+    return 1;
+  }
+  TestReporter test_reporter;
+  const char* const spec = "BM_Chosen";
+  const size_t returned_count =
+      benchmark::RunSpecifiedBenchmarks(&test_reporter, spec);
+  assert(returned_count == 1);
+  const std::vector<std::string> matched_functions =
+      test_reporter.GetMatchedFunctions();
+  assert(matched_functions.size() == 1);
+  if (strcmp(spec, matched_functions.front().c_str()) != 0) {
+    std::cerr << "Expected benchmark [" << spec << "] to run, but got ["
+              << matched_functions.front() << "]\n";
+    return 2;
+  }
+
+  // Test that SetBenchmarkFilter works.
+  const std::string golden_value = "golden_value";
+  benchmark::SetBenchmarkFilter(golden_value);
+  std::string current_value = benchmark::GetBenchmarkFilter();
+  if (golden_value != current_value) {
+    std::cerr << "Expected [" << golden_value
+              << "] for --benchmark_filter but got [" << current_value << "]\n";
+    return 3;
+  }
+  return 0;
+}

test/spec_arg_verbosity_test.cc

@@ -0,0 +1,49 @@
+#include <string.h>
+
+#include <iostream>
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+
+namespace {
+// Tests that the user specified verbosity level can be get.
+void BM_Verbosity(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+}
+BENCHMARK(BM_Verbosity);
+}  // end namespace
+
+int main(int argc, char** argv) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+
+  const int32_t flagv = 42;
+
+  // Verify that argv specify --v=42.
+  bool found = false;
+  for (int i = 0; i < argc; ++i) {
+    if (strcmp("--v=42", argv[i]) == 0) {
+      found = true;
+      break;
+    }
+  }
+  if (!found) {
+    std::cerr << "This test requires '--v=42' to be passed as a command-line "
+              << "argument.\n";
+    return 1;
+  }
+
+  benchmark::Initialize(&argc, argv);
+
+  // Check that the current flag value is reported accurately via the
+  // GetBenchmarkVerbosity() function.
+  if (flagv != benchmark::GetBenchmarkVerbosity()) {
+    std::cerr
+        << "Seeing different value for flags. GetBenchmarkVerbosity() returns ["
+        << benchmark::GetBenchmarkVerbosity() << "] expected flag=[" << flagv
+        << "]\n";
+    return 1;
+  }
+  return 0;
+}

test/state_assembly_test.cc

@@ -0,0 +1,70 @@
+#include "benchmark/state.h"
+#include "benchmark/utils.h"
+
+#ifdef __clang__
+#pragma clang diagnostic ignored "-Wreturn-type"
+#pragma clang diagnostic ignored "-Wmissing-prototypes"
+#endif
+
+// clang-format off
+extern "C" {
+  extern int ExternInt;
+  benchmark::State& GetState();
+  void Fn();
+}
+// clang-format on
+
+using benchmark::State;
+
+// CHECK-LABEL: test_for_auto_loop:
+extern "C" int test_for_auto_loop() {
+  State& S = GetState();
+  int x = 42;
+  // CHECK: 	[[CALL:call(q)*]]	_ZN9benchmark5State16StartKeepRunningEv
+  // CHECK-NEXT: testq %rbx, %rbx
+  // CHECK-NEXT: je [[LOOP_END:.*]]
+
+  for (auto _ : S) {
+    // CHECK: .L[[LOOP_HEAD:[a-zA-Z0-9_]+]]:
+    // CHECK-GNU-NEXT: subq $1, %rbx
+    // CHECK-CLANG-NEXT: {{(addq \$1, %rax|incq %rax|addq \$-1, %rbx)}}
+    // CHECK-NEXT: jne .L[[LOOP_HEAD]]
+    benchmark::DoNotOptimize(x);
+  }
+  // CHECK: [[LOOP_END]]:
+  // CHECK: [[CALL]]	_ZN9benchmark5State17FinishKeepRunningEv
+
+  // CHECK: movl $101, %eax
+  // CHECK: ret
+  return 101;
+}
+
+// CHECK-LABEL: test_while_loop:
+extern "C" int test_while_loop() {
+  State& S = GetState();
+  int x = 42;
+
+  // CHECK: j{{(e|mp)}} .L[[LOOP_HEADER:[a-zA-Z0-9_]+]]
+  // CHECK-NEXT: .L[[LOOP_BODY:[a-zA-Z0-9_]+]]:
+  while (S.KeepRunning()) {
+    // CHECK-GNU-NEXT: subq $1, %[[IREG:[a-z]+]]
+    // CHECK-CLANG-NEXT: {{(addq \$-1,|decq)}} %[[IREG:[a-z]+]]
+    // CHECK: movq %[[IREG]], [[DEST:.*]]
+    benchmark::DoNotOptimize(x);
+  }
+  // CHECK-DAG: movq [[DEST]], %[[IREG]]
+  // CHECK-DAG: testq %[[IREG]], %[[IREG]]
+  // CHECK-DAG: jne .L[[LOOP_BODY]]
+  // CHECK-DAG: .L[[LOOP_HEADER]]:
+
+  // CHECK: cmpb $0
+  // CHECK-NEXT: jne .L[[LOOP_END:[a-zA-Z0-9_]+]]
+  // CHECK: [[CALL:call(q)*]] _ZN9benchmark5State16StartKeepRunningEv
+
+  // CHECK: .L[[LOOP_END]]:
+  // CHECK: [[CALL]] _ZN9benchmark5State17FinishKeepRunningEv
+
+  // CHECK: movl $101, %eax
+  // CHECK: ret
+  return 101;
+}

test/statistics_gtest.cc

@@ -0,0 +1,35 @@
+//===---------------------------------------------------------------------===//
+// statistics_test - Unit tests for src/statistics.cc
+//===---------------------------------------------------------------------===//
+
+#include "../src/statistics.h"
+#include "gtest/gtest.h"
+
+namespace {
+TEST(StatisticsTest, Mean) {
+  EXPECT_DOUBLE_EQ(benchmark::StatisticsMean({42, 42, 42, 42}), 42.0);
+  EXPECT_DOUBLE_EQ(benchmark::StatisticsMean({1, 2, 3, 4}), 2.5);
+  EXPECT_DOUBLE_EQ(benchmark::StatisticsMean({1, 2, 5, 10, 10, 14}), 7.0);
+}
+
+TEST(StatisticsTest, Median) {
+  EXPECT_DOUBLE_EQ(benchmark::StatisticsMedian({42, 42, 42, 42}), 42.0);
+  EXPECT_DOUBLE_EQ(benchmark::StatisticsMedian({1, 2, 3, 4}), 2.5);
+  EXPECT_DOUBLE_EQ(benchmark::StatisticsMedian({1, 2, 5, 10, 10}), 5.0);
+}
+
+TEST(StatisticsTest, StdDev) {
+  EXPECT_DOUBLE_EQ(benchmark::StatisticsStdDev({101, 101, 101, 101}), 0.0);
+  EXPECT_DOUBLE_EQ(benchmark::StatisticsStdDev({1, 2, 3}), 1.0);
+  EXPECT_DOUBLE_EQ(benchmark::StatisticsStdDev({2.5, 2.4, 3.3, 4.2, 5.1}),
+                   1.151086443322134);
+}
+
+TEST(StatisticsTest, CV) {
+  EXPECT_DOUBLE_EQ(benchmark::StatisticsCV({101, 101, 101, 101}), 0.0);
+  EXPECT_DOUBLE_EQ(benchmark::StatisticsCV({1, 2, 3}), 1. / 2.);
+  ASSERT_NEAR(benchmark::StatisticsCV({2.5, 2.4, 3.3, 4.2, 5.1}),
+              0.32888184094918121, 1e-15);
+}
+
+}  // end namespace

test/string_util_gtest.cc

@@ -0,0 +1,203 @@
+//===---------------------------------------------------------------------===//
+// string_util_test - Unit tests for src/string_util.cc
+//===---------------------------------------------------------------------===//
+
+#include <tuple>
+
+#include "../src/internal_macros.h"
+#include "../src/string_util.h"
+#include "gmock/gmock.h"
+#include "gtest/gtest.h"
+
+namespace {
+TEST(StringUtilTest, stoul) {
+  {
+    size_t pos = 0;
+    EXPECT_EQ(0UL, benchmark::stoul("0", &pos));
+    EXPECT_EQ(1UL, pos);
+  }
+  {
+    size_t pos = 0;
+    EXPECT_EQ(7UL, benchmark::stoul("7", &pos));
+    EXPECT_EQ(1UL, pos);
+  }
+  {
+    size_t pos = 0;
+    EXPECT_EQ(135UL, benchmark::stoul("135", &pos));
+    EXPECT_EQ(3UL, pos);
+  }
+#if ULONG_MAX == 0xFFFFFFFFul
+  {
+    size_t pos = 0;
+    EXPECT_EQ(0xFFFFFFFFul, benchmark::stoul("4294967295", &pos));
+    EXPECT_EQ(10ul, pos);
+  }
+#elif ULONG_MAX == 0xFFFFFFFFFFFFFFFFul
+  {
+    size_t pos = 0;
+    EXPECT_EQ(0xFFFFFFFFFFFFFFFFUL,
+              benchmark::stoul("18446744073709551615", &pos));
+    EXPECT_EQ(20UL, pos);
+  }
+#endif
+  {
+    size_t pos = 0;
+    EXPECT_EQ(10UL, benchmark::stoul("1010", &pos, 2));
+    EXPECT_EQ(4UL, pos);
+  }
+  {
+    size_t pos = 0;
+    EXPECT_EQ(520UL, benchmark::stoul("1010", &pos, 8));
+    EXPECT_EQ(4UL, pos);
+  }
+  {
+    size_t pos = 0;
+    EXPECT_EQ(1010UL, benchmark::stoul("1010", &pos, 10));
+    EXPECT_EQ(4UL, pos);
+  }
+  {
+    size_t pos = 0;
+    EXPECT_EQ(4112UL, benchmark::stoul("1010", &pos, 16));
+    EXPECT_EQ(4UL, pos);
+  }
+  {
+    size_t pos = 0;
+    EXPECT_EQ(0xBEEFUL, benchmark::stoul("BEEF", &pos, 16));
+    EXPECT_EQ(4UL, pos);
+  }
+#ifndef BENCHMARK_HAS_NO_EXCEPTIONS
+  {
+    ASSERT_THROW(std::ignore = benchmark::stoul("this is a test"),
+                 std::invalid_argument);
+  }
+#endif
+}
+
+TEST(StringUtilTest, stoi) {
+  {
+    size_t pos = 0;
+    EXPECT_EQ(0, benchmark::stoi("0", &pos));
+    EXPECT_EQ(1UL, pos);
+  }  // namespace
+  {
+    size_t pos = 0;
+    EXPECT_EQ(-17, benchmark::stoi("-17", &pos));
+    EXPECT_EQ(3UL, pos);
+  }
+  {
+    size_t pos = 0;
+    EXPECT_EQ(1357, benchmark::stoi("1357", &pos));
+    EXPECT_EQ(4UL, pos);
+  }
+  {
+    size_t pos = 0;
+    EXPECT_EQ(10, benchmark::stoi("1010", &pos, 2));
+    EXPECT_EQ(4UL, pos);
+  }
+  {
+    size_t pos = 0;
+    EXPECT_EQ(520, benchmark::stoi("1010", &pos, 8));
+    EXPECT_EQ(4UL, pos);
+  }
+  {
+    size_t pos = 0;
+    EXPECT_EQ(1010, benchmark::stoi("1010", &pos, 10));
+    EXPECT_EQ(4UL, pos);
+  }
+  {
+    size_t pos = 0;
+    EXPECT_EQ(4112, benchmark::stoi("1010", &pos, 16));
+    EXPECT_EQ(4UL, pos);
+  }
+  {
+    size_t pos = 0;
+    EXPECT_EQ(0xBEEF, benchmark::stoi("BEEF", &pos, 16));
+    EXPECT_EQ(4UL, pos);
+  }
+#ifndef BENCHMARK_HAS_NO_EXCEPTIONS
+  {
+    ASSERT_THROW(std::ignore = benchmark::stoi("this is a test"),
+                 std::invalid_argument);
+  }
+#endif
+}
+
+TEST(StringUtilTest, stod) {
+  {
+    size_t pos = 0;
+    EXPECT_EQ(0.0, benchmark::stod("0", &pos));
+    EXPECT_EQ(1UL, pos);
+  }
+  {
+    size_t pos = 0;
+    EXPECT_EQ(-84.0, benchmark::stod("-84", &pos));
+    EXPECT_EQ(3UL, pos);
+  }
+  {
+    size_t pos = 0;
+    EXPECT_EQ(1234.0, benchmark::stod("1234", &pos));
+    EXPECT_EQ(4UL, pos);
+  }
+  {
+    size_t pos = 0;
+    EXPECT_EQ(1.5, benchmark::stod("1.5", &pos));
+    EXPECT_EQ(3UL, pos);
+  }
+  {
+    size_t pos = 0;
+    /* Note: exactly representable as double */
+    EXPECT_EQ(-1.25e+9, benchmark::stod("-1.25e+9", &pos));
+    EXPECT_EQ(8UL, pos);
+  }
+#ifndef BENCHMARK_HAS_NO_EXCEPTIONS
+  {
+    ASSERT_THROW(std::ignore = benchmark::stod("this is a test"),
+                 std::invalid_argument);
+  }
+#endif
+}
+
+TEST(StringUtilTest, StrSplit) {
+  EXPECT_EQ(benchmark::StrSplit("", ','), std::vector<std::string>{});
+  EXPECT_EQ(benchmark::StrSplit("hello", ','),
+            std::vector<std::string>({"hello"}));
+  EXPECT_EQ(benchmark::StrSplit("hello,there,is,more", ','),
+            std::vector<std::string>({"hello", "there", "is", "more"}));
+}
+
+using HumanReadableFixture = ::testing::TestWithParam<
+    std::tuple<double, benchmark::Counter::OneK, std::string>>;
+
+INSTANTIATE_TEST_SUITE_P(
+    HumanReadableTests, HumanReadableFixture,
+    ::testing::Values(
+        std::make_tuple(0.0, benchmark::Counter::kIs1024, "0"),
+        std::make_tuple(999.0, benchmark::Counter::kIs1024, "999"),
+        std::make_tuple(1000.0, benchmark::Counter::kIs1024, "1000"),
+        std::make_tuple(1024.0, benchmark::Counter::kIs1024, "1Ki"),
+        std::make_tuple(1000 * 1000.0, benchmark::Counter::kIs1024,
+                        "976\\.56.Ki"),
+        std::make_tuple(1024 * 1024.0, benchmark::Counter::kIs1024, "1Mi"),
+        std::make_tuple(1000 * 1000 * 1000.0, benchmark::Counter::kIs1024,
+                        "953\\.674Mi"),
+        std::make_tuple(1024 * 1024 * 1024.0, benchmark::Counter::kIs1024,
+                        "1Gi"),
+        std::make_tuple(0.0, benchmark::Counter::kIs1000, "0"),
+        std::make_tuple(999.0, benchmark::Counter::kIs1000, "999"),
+        std::make_tuple(1000.0, benchmark::Counter::kIs1000, "1k"),
+        std::make_tuple(1024.0, benchmark::Counter::kIs1000, "1.024k"),
+        std::make_tuple(1000 * 1000.0, benchmark::Counter::kIs1000, "1M"),
+        std::make_tuple(1024 * 1024.0, benchmark::Counter::kIs1000,
+                        "1\\.04858M"),
+        std::make_tuple(1000 * 1000 * 1000.0, benchmark::Counter::kIs1000,
+                        "1G"),
+        std::make_tuple(1024 * 1024 * 1024.0, benchmark::Counter::kIs1000,
+                        "1\\.07374G")));
+
+TEST_P(HumanReadableFixture, HumanReadableNumber) {
+  std::string str = benchmark::HumanReadableNumber(std::get<0>(GetParam()),
+                                                   std::get<1>(GetParam()));
+  ASSERT_THAT(str, ::testing::MatchesRegex(std::get<2>(GetParam())));
+}
+
+}  // end namespace

test/templated_fixture_method_test.cc

@@ -0,0 +1,28 @@
+
+#include <cassert>
+#include <memory>
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+
+template <typename T>
+class MyFixture : public ::benchmark::Fixture {
+ public:
+  MyFixture() : data(0) {}
+
+  T data;
+
+  using type = T;
+};
+
+BENCHMARK_TEMPLATE_METHOD_F(MyFixture, Foo)(benchmark::State& st) {
+  for (auto _ : st) {
+    this->data += typename Base::type(1);
+  }
+}
+
+BENCHMARK_TEMPLATE_INSTANTIATE_F(MyFixture, Foo, int);
+BENCHMARK_TEMPLATE_INSTANTIATE_F(MyFixture, Foo, double);
+
+BENCHMARK_MAIN();

test/templated_fixture_test.cc

@@ -0,0 +1,30 @@
+
+#include <cassert>
+#include <memory>
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+
+template <typename T>
+class MyFixture : public ::benchmark::Fixture {
+ public:
+  MyFixture() : data(0) {}
+
+  T data;
+};
+
+BENCHMARK_TEMPLATE_F(MyFixture, Foo, int)(benchmark::State& st) {
+  for (auto _ : st) {
+    data += 1;
+  }
+}
+
+BENCHMARK_TEMPLATE_DEFINE_F(MyFixture, Bar, double)(benchmark::State& st) {
+  for (auto _ : st) {
+    data += 1.0;
+  }
+}
+BENCHMARK_REGISTER_F(MyFixture, Bar);
+
+BENCHMARK_MAIN();

test/time_unit_gtest.cc

@@ -0,0 +1,38 @@
+#include "benchmark/benchmark_api.h"
+#include "benchmark/state.h"
+#include "gtest/gtest.h"
+
+namespace benchmark {
+namespace internal {
+
+namespace {
+
+class DummyBenchmark : public benchmark::Benchmark {
+ public:
+  DummyBenchmark() : Benchmark("dummy") {}
+  void Run(State& /*state*/) override {}
+};
+
+TEST(DefaultTimeUnitTest, TimeUnitIsNotSet) {
+  DummyBenchmark benchmark;
+  EXPECT_EQ(benchmark.GetTimeUnit(), kNanosecond);
+}
+
+TEST(DefaultTimeUnitTest, DefaultIsSet) {
+  DummyBenchmark benchmark;
+  EXPECT_EQ(benchmark.GetTimeUnit(), kNanosecond);
+  SetDefaultTimeUnit(kMillisecond);
+  EXPECT_EQ(benchmark.GetTimeUnit(), kMillisecond);
+}
+
+TEST(DefaultTimeUnitTest, DefaultAndExplicitUnitIsSet) {
+  DummyBenchmark benchmark;
+  benchmark.Unit(kMillisecond);
+  SetDefaultTimeUnit(kMicrosecond);
+
+  EXPECT_EQ(benchmark.GetTimeUnit(), kMillisecond);
+}
+
+}  // namespace
+}  // namespace internal
+}  // namespace benchmark

test/user_counters_tabular_test.cc

@@ -0,0 +1,572 @@
+
+#undef NDEBUG
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/counter.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+#include "benchmark/utils.h"
+#include "output_test.h"
+
+namespace {
+// @todo: <jpmag> this checks the full output at once; the rule for
+// CounterSet1 was failing because it was not matching "^[-]+$".
+// @todo: <jpmag> check that the counters are vertically aligned.
+ADD_CASES(TC_ConsoleOut,
+          {
+              // keeping these lines long improves readability, so:
+              // clang-format off
+    {"^[-]+$", MR_Next},
+    {"^Benchmark %s Time %s CPU %s Iterations %s Bar %s Bat %s Baz %s Foo %s Frob %s Lob$", MR_Next},
+    {"^[-]+$", MR_Next},
+      {"^BM_Counters_Tabular/repeats:2/threads:1 %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$", MR_Next},
+      {"^BM_Counters_Tabular/repeats:2/threads:1 %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$", MR_Next},
+      {"^BM_Counters_Tabular/repeats:2/threads:1_mean %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$", MR_Next},
+      {"^BM_Counters_Tabular/repeats:2/threads:1_median %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$", MR_Next},
+            {"^BM_Counters_Tabular/repeats:2/threads:1_stddev %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$", MR_Next},
+            {"^BM_Counters_Tabular/repeats:2/threads:1_cv %console_percentage_report [ ]*%percentage[ ]*% [ ]*%percentage[ ]*% [ ]*%percentage[ ]*% [ ]*%percentage[ ]*% [ ]*%percentage[ ]*% [ ]*%percentage[ ]*%$", MR_Next},
+      {"^BM_Counters_Tabular/repeats:2/threads:2 %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$", MR_Next},
+      {"^BM_Counters_Tabular/repeats:2/threads:2 %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$", MR_Next},
+      {"^BM_Counters_Tabular/repeats:2/threads:2_mean %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$", MR_Next},
+      {"^BM_Counters_Tabular/repeats:2/threads:2_median %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$", MR_Next},
+            {"^BM_Counters_Tabular/repeats:2/threads:2_stddev %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$", MR_Next},
+            {"^BM_Counters_Tabular/repeats:2/threads:2_cv %console_percentage_report [ ]*%percentage[ ]*% [ ]*%percentage[ ]*% [ ]*%percentage[ ]*% [ ]*%percentage[ ]*% [ ]*%percentage[ ]*% [ ]*%percentage[ ]*%$", MR_Next},
+    {"^BM_CounterRates_Tabular/threads:%int %console_report [ ]*%hrfloat/s [ ]*%hrfloat/s [ ]*%hrfloat/s [ ]*%hrfloat/s [ ]*%hrfloat/s [ ]*%hrfloat/s$", MR_Next},
+    {"^BM_CounterRates_Tabular/threads:%int %console_report [ ]*%hrfloat/s [ ]*%hrfloat/s [ ]*%hrfloat/s [ ]*%hrfloat/s [ ]*%hrfloat/s [ ]*%hrfloat/s$", MR_Next},
+    {"^BM_CounterRates_Tabular/threads:%int %console_report [ ]*%hrfloat/s [ ]*%hrfloat/s [ ]*%hrfloat/s [ ]*%hrfloat/s [ ]*%hrfloat/s [ ]*%hrfloat/s$", MR_Next},
+    {"^BM_CounterRates_Tabular/threads:%int %console_report [ ]*%hrfloat/s [ ]*%hrfloat/s [ ]*%hrfloat/s [ ]*%hrfloat/s [ ]*%hrfloat/s [ ]*%hrfloat/s$", MR_Next},
+    {"^BM_CounterRates_Tabular/threads:%int %console_report [ ]*%hrfloat/s [ ]*%hrfloat/s [ ]*%hrfloat/s [ ]*%hrfloat/s [ ]*%hrfloat/s [ ]*%hrfloat/s$", MR_Next},
+    {"^[-]+$", MR_Next},
+    {"^Benchmark %s Time %s CPU %s Iterations %s Bar %s Baz %s Foo$", MR_Next},
+    {"^[-]+$", MR_Next},
+    {"^BM_CounterSet0_Tabular/threads:%int %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$", MR_Next},
+    {"^BM_CounterSet0_Tabular/threads:%int %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$", MR_Next},
+    {"^BM_CounterSet0_Tabular/threads:%int %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$", MR_Next},
+    {"^BM_CounterSet0_Tabular/threads:%int %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$", MR_Next},
+    {"^BM_CounterSet0_Tabular/threads:%int %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$", MR_Next},
+    {"^BM_CounterSet1_Tabular/threads:%int %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$", MR_Next},
+    {"^BM_CounterSet1_Tabular/threads:%int %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$", MR_Next},
+    {"^BM_CounterSet1_Tabular/threads:%int %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$", MR_Next},
+    {"^BM_CounterSet1_Tabular/threads:%int %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$", MR_Next},
+    {"^BM_CounterSet1_Tabular/threads:%int %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$", MR_Next},
+    {"^[-]+$", MR_Next},
+    {"^Benchmark %s Time %s CPU %s Iterations %s Bat %s Baz %s Foo$", MR_Next},
+    {"^[-]+$", MR_Next},
+    {"^BM_CounterSet2_Tabular/threads:%int %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$", MR_Next},
+    {"^BM_CounterSet2_Tabular/threads:%int %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$", MR_Next},
+    {"^BM_CounterSet2_Tabular/threads:%int %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$", MR_Next},
+    {"^BM_CounterSet2_Tabular/threads:%int %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$", MR_Next},
+    {"^BM_CounterSet2_Tabular/threads:%int %console_report [ ]*%hrfloat [ ]*%hrfloat [ ]*%hrfloat$"},
+              // clang-format on
+          });
+ADD_CASES(TC_CSVOut, {{"%csv_header,"
+                       "\"Bar\",\"Bat\",\"Baz\",\"Foo\",\"Frob\",\"Lob\""}});
+
+// ========================================================================= //
+// ------------------------- Tabular Counters Output ----------------------- //
+// ========================================================================= //
+
+void BM_Counters_Tabular(benchmark::State& state) {
+  for (auto _ : state) {
+    // This test requires a non-zero CPU time to avoid divide-by-zero
+    auto iterations = static_cast<double>(state.iterations()) *
+                      static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(iterations);
+  }
+  namespace bm = benchmark;
+  state.counters.insert({
+      {"Foo", {1, bm::Counter::kAvgThreads}},
+      {"Bar", {2, bm::Counter::kAvgThreads}},
+      {"Baz", {4, bm::Counter::kAvgThreads}},
+      {"Bat", {8, bm::Counter::kAvgThreads}},
+      {"Frob", {16, bm::Counter::kAvgThreads}},
+      {"Lob", {32, bm::Counter::kAvgThreads}},
+  });
+}
+BENCHMARK(BM_Counters_Tabular)->ThreadRange(1, 2)->Repetitions(2);
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_Tabular/repeats:2/threads:1\",$"},
+           {"\"family_index\": 0,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_Tabular/repeats:2/threads:1\",$",
+            MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 2,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"Bar\": %float,$", MR_Next},
+           {"\"Bat\": %float,$", MR_Next},
+           {"\"Baz\": %float,$", MR_Next},
+           {"\"Foo\": %float,$", MR_Next},
+           {"\"Frob\": %float,$", MR_Next},
+           {"\"Lob\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_Tabular/repeats:2/threads:1\",$"},
+           {"\"family_index\": 0,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_Tabular/repeats:2/threads:1\",$",
+            MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 2,$", MR_Next},
+           {"\"repetition_index\": 1,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"Bar\": %float,$", MR_Next},
+           {"\"Bat\": %float,$", MR_Next},
+           {"\"Baz\": %float,$", MR_Next},
+           {"\"Foo\": %float,$", MR_Next},
+           {"\"Frob\": %float,$", MR_Next},
+           {"\"Lob\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_Tabular/repeats:2/threads:1_mean\",$"},
+           {"\"family_index\": 0,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_Tabular/repeats:2/threads:1\",$",
+            MR_Next},
+           {"\"run_type\": \"aggregate\",$", MR_Next},
+           {"\"repetitions\": 2,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"aggregate_name\": \"mean\",$", MR_Next},
+           {"\"aggregate_unit\": \"time\",$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"Bar\": %float,$", MR_Next},
+           {"\"Bat\": %float,$", MR_Next},
+           {"\"Baz\": %float,$", MR_Next},
+           {"\"Foo\": %float,$", MR_Next},
+           {"\"Frob\": %float,$", MR_Next},
+           {"\"Lob\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_Tabular/repeats:2/threads:1_median\",$"},
+           {"\"family_index\": 0,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_Tabular/repeats:2/threads:1\",$",
+            MR_Next},
+           {"\"run_type\": \"aggregate\",$", MR_Next},
+           {"\"repetitions\": 2,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"aggregate_name\": \"median\",$", MR_Next},
+           {"\"aggregate_unit\": \"time\",$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"Bar\": %float,$", MR_Next},
+           {"\"Bat\": %float,$", MR_Next},
+           {"\"Baz\": %float,$", MR_Next},
+           {"\"Foo\": %float,$", MR_Next},
+           {"\"Frob\": %float,$", MR_Next},
+           {"\"Lob\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_Tabular/repeats:2/threads:1_stddev\",$"},
+           {"\"family_index\": 0,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_Tabular/repeats:2/threads:1\",$",
+            MR_Next},
+           {"\"run_type\": \"aggregate\",$", MR_Next},
+           {"\"repetitions\": 2,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"aggregate_name\": \"stddev\",$", MR_Next},
+           {"\"aggregate_unit\": \"time\",$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"Bar\": %float,$", MR_Next},
+           {"\"Bat\": %float,$", MR_Next},
+           {"\"Baz\": %float,$", MR_Next},
+           {"\"Foo\": %float,$", MR_Next},
+           {"\"Frob\": %float,$", MR_Next},
+           {"\"Lob\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_Tabular/repeats:2/threads:1_cv\",$"},
+           {"\"family_index\": 0,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_Tabular/repeats:2/threads:1\",$",
+            MR_Next},
+           {"\"run_type\": \"aggregate\",$", MR_Next},
+           {"\"repetitions\": 2,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"aggregate_name\": \"cv\",$", MR_Next},
+           {"\"aggregate_unit\": \"percentage\",$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"Bar\": %float,$", MR_Next},
+           {"\"Bat\": %float,$", MR_Next},
+           {"\"Baz\": %float,$", MR_Next},
+           {"\"Foo\": %float,$", MR_Next},
+           {"\"Frob\": %float,$", MR_Next},
+           {"\"Lob\": %float$", MR_Next},
+           {"}", MR_Next}});
+
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_Tabular/repeats:2/threads:2\",$"},
+           {"\"family_index\": 0,$", MR_Next},
+           {"\"per_family_instance_index\": 1,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_Tabular/repeats:2/threads:2\",$",
+            MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 2,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": 2,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"Bar\": %float,$", MR_Next},
+           {"\"Bat\": %float,$", MR_Next},
+           {"\"Baz\": %float,$", MR_Next},
+           {"\"Foo\": %float,$", MR_Next},
+           {"\"Frob\": %float,$", MR_Next},
+           {"\"Lob\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_Tabular/repeats:2/threads:2\",$"},
+           {"\"family_index\": 0,$", MR_Next},
+           {"\"per_family_instance_index\": 1,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_Tabular/repeats:2/threads:2\",$",
+            MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 2,$", MR_Next},
+           {"\"repetition_index\": 1,$", MR_Next},
+           {"\"threads\": 2,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"Bar\": %float,$", MR_Next},
+           {"\"Bat\": %float,$", MR_Next},
+           {"\"Baz\": %float,$", MR_Next},
+           {"\"Foo\": %float,$", MR_Next},
+           {"\"Frob\": %float,$", MR_Next},
+           {"\"Lob\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_Tabular/repeats:2/threads:2_median\",$"},
+           {"\"family_index\": 0,$", MR_Next},
+           {"\"per_family_instance_index\": 1,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_Tabular/repeats:2/threads:2\",$",
+            MR_Next},
+           {"\"run_type\": \"aggregate\",$", MR_Next},
+           {"\"repetitions\": 2,$", MR_Next},
+           {"\"threads\": 2,$", MR_Next},
+           {"\"aggregate_name\": \"median\",$", MR_Next},
+           {"\"aggregate_unit\": \"time\",$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"Bar\": %float,$", MR_Next},
+           {"\"Bat\": %float,$", MR_Next},
+           {"\"Baz\": %float,$", MR_Next},
+           {"\"Foo\": %float,$", MR_Next},
+           {"\"Frob\": %float,$", MR_Next},
+           {"\"Lob\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_Tabular/repeats:2/threads:2_stddev\",$"},
+           {"\"family_index\": 0,$", MR_Next},
+           {"\"per_family_instance_index\": 1,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_Tabular/repeats:2/threads:2\",$",
+            MR_Next},
+           {"\"run_type\": \"aggregate\",$", MR_Next},
+           {"\"repetitions\": 2,$", MR_Next},
+           {"\"threads\": 2,$", MR_Next},
+           {"\"aggregate_name\": \"stddev\",$", MR_Next},
+           {"\"aggregate_unit\": \"time\",$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"Bar\": %float,$", MR_Next},
+           {"\"Bat\": %float,$", MR_Next},
+           {"\"Baz\": %float,$", MR_Next},
+           {"\"Foo\": %float,$", MR_Next},
+           {"\"Frob\": %float,$", MR_Next},
+           {"\"Lob\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_Tabular/repeats:2/threads:2_cv\",$"},
+           {"\"family_index\": 0,$", MR_Next},
+           {"\"per_family_instance_index\": 1,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_Tabular/repeats:2/threads:2\",$",
+            MR_Next},
+           {"\"run_type\": \"aggregate\",$", MR_Next},
+           {"\"repetitions\": 2,$", MR_Next},
+           {"\"threads\": 2,$", MR_Next},
+           {"\"aggregate_name\": \"cv\",$", MR_Next},
+           {"\"aggregate_unit\": \"percentage\",$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"Bar\": %float,$", MR_Next},
+           {"\"Bat\": %float,$", MR_Next},
+           {"\"Baz\": %float,$", MR_Next},
+           {"\"Foo\": %float,$", MR_Next},
+           {"\"Frob\": %float,$", MR_Next},
+           {"\"Lob\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_CSVOut,
+          {{"^\"BM_Counters_Tabular/repeats:2/threads:1\",%csv_report,"
+            "%float,%float,%float,%float,%float,%float$"}});
+ADD_CASES(TC_CSVOut,
+          {{"^\"BM_Counters_Tabular/repeats:2/threads:1\",%csv_report,"
+            "%float,%float,%float,%float,%float,%float$"}});
+ADD_CASES(TC_CSVOut,
+          {{"^\"BM_Counters_Tabular/repeats:2/threads:1_mean\",%csv_report,"
+            "%float,%float,%float,%float,%float,%float$"}});
+ADD_CASES(TC_CSVOut,
+          {{"^\"BM_Counters_Tabular/repeats:2/threads:1_median\",%csv_report,"
+            "%float,%float,%float,%float,%float,%float$"}});
+ADD_CASES(TC_CSVOut,
+          {{"^\"BM_Counters_Tabular/repeats:2/threads:1_stddev\",%csv_report,"
+            "%float,%float,%float,%float,%float,%float$"}});
+ADD_CASES(TC_CSVOut,
+          {{"^\"BM_Counters_Tabular/repeats:2/threads:1_cv\",%csv_cv_report,"
+            "%float,%float,%float,%float,%float,%float$"}});
+ADD_CASES(TC_CSVOut,
+          {{"^\"BM_Counters_Tabular/repeats:2/threads:2\",%csv_report,"
+            "%float,%float,%float,%float,%float,%float$"}});
+ADD_CASES(TC_CSVOut,
+          {{"^\"BM_Counters_Tabular/repeats:2/threads:2\",%csv_report,"
+            "%float,%float,%float,%float,%float,%float$"}});
+ADD_CASES(TC_CSVOut,
+          {{"^\"BM_Counters_Tabular/repeats:2/threads:2_mean\",%csv_report,"
+            "%float,%float,%float,%float,%float,%float$"}});
+ADD_CASES(TC_CSVOut,
+          {{"^\"BM_Counters_Tabular/repeats:2/threads:2_median\",%csv_report,"
+            "%float,%float,%float,%float,%float,%float$"}});
+ADD_CASES(TC_CSVOut,
+          {{"^\"BM_Counters_Tabular/repeats:2/threads:2_stddev\",%csv_report,"
+            "%float,%float,%float,%float,%float,%float$"}});
+ADD_CASES(TC_CSVOut,
+          {{"^\"BM_Counters_Tabular/repeats:2/threads:2_cv\",%csv_cv_report,"
+            "%float,%float,%float,%float,%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckTabular(Results const& e) {
+  CHECK_COUNTER_VALUE(e, int, "Foo", EQ, 1);
+  CHECK_COUNTER_VALUE(e, int, "Bar", EQ, 2);
+  CHECK_COUNTER_VALUE(e, int, "Baz", EQ, 4);
+  CHECK_COUNTER_VALUE(e, int, "Bat", EQ, 8);
+  CHECK_COUNTER_VALUE(e, int, "Frob", EQ, 16);
+  CHECK_COUNTER_VALUE(e, int, "Lob", EQ, 32);
+}
+CHECK_BENCHMARK_RESULTS("BM_Counters_Tabular/repeats:2/threads:1$",
+                        &CheckTabular);
+CHECK_BENCHMARK_RESULTS("BM_Counters_Tabular/repeats:2/threads:2$",
+                        &CheckTabular);
+
+// ========================================================================= //
+// -------------------- Tabular+Rate Counters Output ----------------------- //
+// ========================================================================= //
+
+void BM_CounterRates_Tabular(benchmark::State& state) {
+  for (auto _ : state) {
+    // This test requires a non-zero CPU time to avoid divide-by-zero
+    auto iterations = static_cast<double>(state.iterations()) *
+                      static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(iterations);
+  }
+  namespace bm = benchmark;
+  state.counters.insert({
+      {"Foo", {1, bm::Counter::kAvgThreadsRate}},
+      {"Bar", {2, bm::Counter::kAvgThreadsRate}},
+      {"Baz", {4, bm::Counter::kAvgThreadsRate}},
+      {"Bat", {8, bm::Counter::kAvgThreadsRate}},
+      {"Frob", {16, bm::Counter::kAvgThreadsRate}},
+      {"Lob", {32, bm::Counter::kAvgThreadsRate}},
+  });
+}
+BENCHMARK(BM_CounterRates_Tabular)->ThreadRange(1, 16);
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_CounterRates_Tabular/threads:%int\",$"},
+           {"\"family_index\": 1,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_CounterRates_Tabular/threads:%int\",$",
+            MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 1,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"Bar\": %float,$", MR_Next},
+           {"\"Bat\": %float,$", MR_Next},
+           {"\"Baz\": %float,$", MR_Next},
+           {"\"Foo\": %float,$", MR_Next},
+           {"\"Frob\": %float,$", MR_Next},
+           {"\"Lob\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_CounterRates_Tabular/threads:%int\",%csv_report,"
+                       "%float,%float,%float,%float,%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckTabularRate(Results const& e) {
+  double t = e.DurationCPUTime() / e.NumThreads();
+  CHECK_FLOAT_COUNTER_VALUE(e, "Foo", EQ, 1. / t, 0.001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "Bar", EQ, 2. / t, 0.001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "Baz", EQ, 4. / t, 0.001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "Bat", EQ, 8. / t, 0.001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "Frob", EQ, 16. / t, 0.001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "Lob", EQ, 32. / t, 0.001);
+}
+CHECK_BENCHMARK_RESULTS("BM_CounterRates_Tabular/threads:%int",
+                        &CheckTabularRate);
+
+// ========================================================================= //
+// ------------------------- Tabular Counters Output ----------------------- //
+// ========================================================================= //
+
+// set only some of the counters
+void BM_CounterSet0_Tabular(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+  namespace bm = benchmark;
+  state.counters.insert({
+      {"Foo", {10, bm::Counter::kAvgThreads}},
+      {"Bar", {20, bm::Counter::kAvgThreads}},
+      {"Baz", {40, bm::Counter::kAvgThreads}},
+  });
+}
+BENCHMARK(BM_CounterSet0_Tabular)->ThreadRange(1, 16);
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_CounterSet0_Tabular/threads:%int\",$"},
+           {"\"family_index\": 2,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_CounterSet0_Tabular/threads:%int\",$", MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 1,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"Bar\": %float,$", MR_Next},
+           {"\"Baz\": %float,$", MR_Next},
+           {"\"Foo\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_CounterSet0_Tabular/threads:%int\",%csv_report,"
+                       "%float,,%float,%float,,"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckSet0(Results const& e) {
+  CHECK_COUNTER_VALUE(e, int, "Foo", EQ, 10);
+  CHECK_COUNTER_VALUE(e, int, "Bar", EQ, 20);
+  CHECK_COUNTER_VALUE(e, int, "Baz", EQ, 40);
+}
+CHECK_BENCHMARK_RESULTS("BM_CounterSet0_Tabular", &CheckSet0);
+
+// again.
+void BM_CounterSet1_Tabular(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+  namespace bm = benchmark;
+  state.counters.insert({
+      {"Foo", {15, bm::Counter::kAvgThreads}},
+      {"Bar", {25, bm::Counter::kAvgThreads}},
+      {"Baz", {45, bm::Counter::kAvgThreads}},
+  });
+}
+BENCHMARK(BM_CounterSet1_Tabular)->ThreadRange(1, 16);
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_CounterSet1_Tabular/threads:%int\",$"},
+           {"\"family_index\": 3,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_CounterSet1_Tabular/threads:%int\",$", MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 1,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"Bar\": %float,$", MR_Next},
+           {"\"Baz\": %float,$", MR_Next},
+           {"\"Foo\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_CounterSet1_Tabular/threads:%int\",%csv_report,"
+                       "%float,,%float,%float,,"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckSet1(Results const& e) {
+  CHECK_COUNTER_VALUE(e, int, "Foo", EQ, 15);
+  CHECK_COUNTER_VALUE(e, int, "Bar", EQ, 25);
+  CHECK_COUNTER_VALUE(e, int, "Baz", EQ, 45);
+}
+CHECK_BENCHMARK_RESULTS("BM_CounterSet1_Tabular/threads:%int", &CheckSet1);
+
+// ========================================================================= //
+// ------------------------- Tabular Counters Output ----------------------- //
+// ========================================================================= //
+
+// set only some of the counters, different set now.
+void BM_CounterSet2_Tabular(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+  namespace bm = benchmark;
+  state.counters.insert({
+      {"Foo", {10, bm::Counter::kAvgThreads}},
+      {"Bat", {30, bm::Counter::kAvgThreads}},
+      {"Baz", {40, bm::Counter::kAvgThreads}},
+  });
+}
+BENCHMARK(BM_CounterSet2_Tabular)->ThreadRange(1, 16);
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_CounterSet2_Tabular/threads:%int\",$"},
+           {"\"family_index\": 4,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_CounterSet2_Tabular/threads:%int\",$", MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 1,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"Bat\": %float,$", MR_Next},
+           {"\"Baz\": %float,$", MR_Next},
+           {"\"Foo\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_CounterSet2_Tabular/threads:%int\",%csv_report,"
+                       ",%float,%float,%float,,"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckSet2(Results const& e) {
+  CHECK_COUNTER_VALUE(e, int, "Foo", EQ, 10);
+  CHECK_COUNTER_VALUE(e, int, "Bat", EQ, 30);
+  CHECK_COUNTER_VALUE(e, int, "Baz", EQ, 40);
+}
+CHECK_BENCHMARK_RESULTS("BM_CounterSet2_Tabular", &CheckSet2);
+}  // end namespace
+
+// ========================================================================= //
+// --------------------------- TEST CASES END ------------------------------ //
+// ========================================================================= //
+
+int main(int argc, char* argv[]) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+  RunOutputTests(argc, argv);
+}

test/user_counters_test.cc

@@ -0,0 +1,597 @@
+
+#undef NDEBUG
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/counter.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+#include "benchmark/utils.h"
+#include "output_test.h"
+
+// ========================================================================= //
+// ---------------------- Testing Prologue Output -------------------------- //
+// ========================================================================= //
+
+// clang-format off
+
+ADD_CASES(TC_ConsoleOut,
+          {{"^[-]+$", MR_Next},
+           {"^Benchmark %s Time %s CPU %s Iterations UserCounters...$", MR_Next},
+           {"^[-]+$", MR_Next}});
+ADD_CASES(TC_CSVOut, {{"%csv_header,\"bar\",\"foo\""}});
+
+// clang-format on
+
+// ========================================================================= //
+// ------------------------- Simple Counters Output ------------------------ //
+// ========================================================================= //
+namespace {
+void BM_Counters_Simple(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+  state.counters["foo"] = 1;
+  state.counters["bar"] = 2 * static_cast<double>(state.iterations());
+}
+BENCHMARK(BM_Counters_Simple);
+ADD_CASES(TC_ConsoleOut,
+          {{"^BM_Counters_Simple %console_report bar=%hrfloat foo=%hrfloat$"}});
+ADD_CASES(TC_JSONOut, {{"\"name\": \"BM_Counters_Simple\",$"},
+                       {"\"family_index\": 0,$", MR_Next},
+                       {"\"per_family_instance_index\": 0,$", MR_Next},
+                       {"\"run_name\": \"BM_Counters_Simple\",$", MR_Next},
+                       {"\"run_type\": \"iteration\",$", MR_Next},
+                       {"\"repetitions\": 1,$", MR_Next},
+                       {"\"repetition_index\": 0,$", MR_Next},
+                       {"\"threads\": 1,$", MR_Next},
+                       {"\"iterations\": %int,$", MR_Next},
+                       {"\"real_time\": %float,$", MR_Next},
+                       {"\"cpu_time\": %float,$", MR_Next},
+                       {"\"time_unit\": \"ns\",$", MR_Next},
+                       {"\"bar\": %float,$", MR_Next},
+                       {"\"foo\": %float$", MR_Next},
+                       {"}", MR_Next}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_Counters_Simple\",%csv_report,%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckSimple(Results const& e) {
+  double its = e.NumIterations();
+  CHECK_COUNTER_VALUE(e, int, "foo", EQ, 1);
+  // check that the value of bar is within 0.1% of the expected value
+  CHECK_FLOAT_COUNTER_VALUE(e, "bar", EQ, 2. * its, 0.001);
+}
+CHECK_BENCHMARK_RESULTS("BM_Counters_Simple", &CheckSimple);
+}  // end namespace
+
+// ========================================================================= //
+// --------------------- Counters+Items+Bytes/s Output --------------------- //
+// ========================================================================= //
+
+namespace {
+int num_calls1 = 0;
+void BM_Counters_WithBytesAndItemsPSec(benchmark::State& state) {
+  for (auto _ : state) {
+    // This test requires a non-zero CPU time to avoid divide-by-zero
+    auto iterations = static_cast<double>(state.iterations()) *
+                      static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(iterations);
+  }
+  state.counters["foo"] = 1;
+  state.counters["bar"] = ++num_calls1;
+  state.SetBytesProcessed(364);
+  state.SetItemsProcessed(150);
+}
+BENCHMARK(BM_Counters_WithBytesAndItemsPSec);
+ADD_CASES(TC_ConsoleOut, {{"^BM_Counters_WithBytesAndItemsPSec %console_report "
+                           "bar=%hrfloat bytes_per_second=%hrfloat/s "
+                           "foo=%hrfloat items_per_second=%hrfloat/s$"}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_WithBytesAndItemsPSec\",$"},
+           {"\"family_index\": 1,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_WithBytesAndItemsPSec\",$", MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 1,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"bar\": %float,$", MR_Next},
+           {"\"bytes_per_second\": %float,$", MR_Next},
+           {"\"foo\": %float,$", MR_Next},
+           {"\"items_per_second\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_Counters_WithBytesAndItemsPSec\","
+                       "%csv_bytes_items_report,%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckBytesAndItemsPSec(Results const& e) {
+  double t = e.DurationCPUTime();  // this (and not real time) is the time used
+  CHECK_COUNTER_VALUE(e, int, "foo", EQ, 1);
+  CHECK_COUNTER_VALUE(e, int, "bar", EQ, num_calls1);
+  // check that the values are within 0.1% of the expected values
+  CHECK_FLOAT_RESULT_VALUE(e, "bytes_per_second", EQ, 364. / t, 0.001);
+  CHECK_FLOAT_RESULT_VALUE(e, "items_per_second", EQ, 150. / t, 0.001);
+}
+CHECK_BENCHMARK_RESULTS("BM_Counters_WithBytesAndItemsPSec",
+                        &CheckBytesAndItemsPSec);
+}  // end namespace
+
+// ========================================================================= //
+// ------------------------- Rate Counters Output -------------------------- //
+// ========================================================================= //
+namespace {
+void BM_Counters_Rate(benchmark::State& state) {
+  for (auto _ : state) {
+    // This test requires a non-zero CPU time to avoid divide-by-zero
+    auto iterations = static_cast<double>(state.iterations()) *
+                      static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(iterations);
+  }
+  namespace bm = benchmark;
+  state.counters["foo"] = bm::Counter{1, bm::Counter::kIsRate};
+  state.counters["bar"] = bm::Counter{2, bm::Counter::kIsRate};
+}
+BENCHMARK(BM_Counters_Rate);
+ADD_CASES(
+    TC_ConsoleOut,
+    {{"^BM_Counters_Rate %console_report bar=%hrfloat/s foo=%hrfloat/s$"}});
+ADD_CASES(TC_JSONOut, {{"\"name\": \"BM_Counters_Rate\",$"},
+                       {"\"family_index\": 2,$", MR_Next},
+                       {"\"per_family_instance_index\": 0,$", MR_Next},
+                       {"\"run_name\": \"BM_Counters_Rate\",$", MR_Next},
+                       {"\"run_type\": \"iteration\",$", MR_Next},
+                       {"\"repetitions\": 1,$", MR_Next},
+                       {"\"repetition_index\": 0,$", MR_Next},
+                       {"\"threads\": 1,$", MR_Next},
+                       {"\"iterations\": %int,$", MR_Next},
+                       {"\"real_time\": %float,$", MR_Next},
+                       {"\"cpu_time\": %float,$", MR_Next},
+                       {"\"time_unit\": \"ns\",$", MR_Next},
+                       {"\"bar\": %float,$", MR_Next},
+                       {"\"foo\": %float$", MR_Next},
+                       {"}", MR_Next}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_Counters_Rate\",%csv_report,%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckRate(Results const& e) {
+  double t = e.DurationCPUTime();  // this (and not real time) is the time used
+  // check that the values are within 0.1% of the expected values
+  CHECK_FLOAT_COUNTER_VALUE(e, "foo", EQ, 1. / t, 0.001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "bar", EQ, 2. / t, 0.001);
+}
+CHECK_BENCHMARK_RESULTS("BM_Counters_Rate", &CheckRate);
+}  // end namespace
+
+// ========================================================================= //
+// ----------------------- Inverted Counters Output ------------------------ //
+// ========================================================================= //
+
+namespace {
+void BM_Invert(benchmark::State& state) {
+  for (auto _ : state) {
+    // This test requires a non-zero CPU time to avoid divide-by-zero
+    auto iterations = static_cast<double>(state.iterations()) *
+                      static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(iterations);
+  }
+  namespace bm = benchmark;
+  state.counters["foo"] = bm::Counter{0.0001, bm::Counter::kInvert};
+  state.counters["bar"] = bm::Counter{10000, bm::Counter::kInvert};
+}
+BENCHMARK(BM_Invert);
+ADD_CASES(TC_ConsoleOut,
+          {{"^BM_Invert %console_report bar=%hrfloatu foo=%hrfloatk$"}});
+ADD_CASES(TC_JSONOut, {{"\"name\": \"BM_Invert\",$"},
+                       {"\"family_index\": 3,$", MR_Next},
+                       {"\"per_family_instance_index\": 0,$", MR_Next},
+                       {"\"run_name\": \"BM_Invert\",$", MR_Next},
+                       {"\"run_type\": \"iteration\",$", MR_Next},
+                       {"\"repetitions\": 1,$", MR_Next},
+                       {"\"repetition_index\": 0,$", MR_Next},
+                       {"\"threads\": 1,$", MR_Next},
+                       {"\"iterations\": %int,$", MR_Next},
+                       {"\"real_time\": %float,$", MR_Next},
+                       {"\"cpu_time\": %float,$", MR_Next},
+                       {"\"time_unit\": \"ns\",$", MR_Next},
+                       {"\"bar\": %float,$", MR_Next},
+                       {"\"foo\": %float$", MR_Next},
+                       {"}", MR_Next}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_Invert\",%csv_report,%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckInvert(Results const& e) {
+  CHECK_FLOAT_COUNTER_VALUE(e, "foo", EQ, 10000, 0.0001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "bar", EQ, 0.0001, 0.0001);
+}
+CHECK_BENCHMARK_RESULTS("BM_Invert", &CheckInvert);
+}  // end namespace
+
+// ========================================================================= //
+// --------------------- InvertedRate Counters Output ---------------------- //
+// ========================================================================= //
+
+namespace {
+void BM_Counters_InvertedRate(benchmark::State& state) {
+  for (auto _ : state) {
+    // This test requires a non-zero CPU time to avoid divide-by-zero
+    auto iterations = static_cast<double>(state.iterations()) *
+                      static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(iterations);
+  }
+  namespace bm = benchmark;
+  state.counters["foo"] =
+      bm::Counter{1, bm::Counter::kIsRate | bm::Counter::kInvert};
+  state.counters["bar"] =
+      bm::Counter{8192, bm::Counter::kIsRate | bm::Counter::kInvert};
+}
+BENCHMARK(BM_Counters_InvertedRate);
+ADD_CASES(TC_ConsoleOut, {{"^BM_Counters_InvertedRate %console_report "
+                           "bar=%hrfloats foo=%hrfloats$"}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_InvertedRate\",$"},
+           {"\"family_index\": 4,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_InvertedRate\",$", MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 1,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"bar\": %float,$", MR_Next},
+           {"\"foo\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_CSVOut,
+          {{"^\"BM_Counters_InvertedRate\",%csv_report,%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckInvertedRate(Results const& e) {
+  double t = e.DurationCPUTime();  // this (and not real time) is the time used
+  // check that the values are within 0.1% of the expected values
+  CHECK_FLOAT_COUNTER_VALUE(e, "foo", EQ, t, 0.001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "bar", EQ, t / 8192.0, 0.001);
+}
+CHECK_BENCHMARK_RESULTS("BM_Counters_InvertedRate", &CheckInvertedRate);
+}  // end namespace
+
+// ========================================================================= //
+// ------------------------- Thread Counters Output ------------------------ //
+// ========================================================================= //
+
+namespace {
+void BM_Counters_Threads(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+  state.counters["foo"] = 1;
+  state.counters["bar"] = 2;
+}
+BENCHMARK(BM_Counters_Threads)->ThreadRange(1, 8);
+ADD_CASES(TC_ConsoleOut, {{"^BM_Counters_Threads/threads:%int %console_report "
+                           "bar=%hrfloat foo=%hrfloat$"}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_Threads/threads:%int\",$"},
+           {"\"family_index\": 5,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_Threads/threads:%int\",$", MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 1,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"bar\": %float,$", MR_Next},
+           {"\"foo\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(
+    TC_CSVOut,
+    {{"^\"BM_Counters_Threads/threads:%int\",%csv_report,%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckThreads(Results const& e) {
+  CHECK_COUNTER_VALUE(e, int, "foo", EQ, e.NumThreads());
+  CHECK_COUNTER_VALUE(e, int, "bar", EQ, 2 * e.NumThreads());
+}
+CHECK_BENCHMARK_RESULTS("BM_Counters_Threads/threads:%int", &CheckThreads);
+}  // end namespace
+
+// ========================================================================= //
+// ---------------------- ThreadAvg Counters Output ------------------------ //
+// ========================================================================= //
+
+namespace {
+void BM_Counters_AvgThreads(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+  namespace bm = benchmark;
+  state.counters["foo"] = bm::Counter{1, bm::Counter::kAvgThreads};
+  state.counters["bar"] = bm::Counter{2, bm::Counter::kAvgThreads};
+}
+BENCHMARK(BM_Counters_AvgThreads)->ThreadRange(1, 8);
+ADD_CASES(TC_ConsoleOut, {{"^BM_Counters_AvgThreads/threads:%int "
+                           "%console_report bar=%hrfloat foo=%hrfloat$"}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_AvgThreads/threads:%int\",$"},
+           {"\"family_index\": 6,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_AvgThreads/threads:%int\",$", MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 1,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"bar\": %float,$", MR_Next},
+           {"\"foo\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(
+    TC_CSVOut,
+    {{"^\"BM_Counters_AvgThreads/threads:%int\",%csv_report,%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckAvgThreads(Results const& e) {
+  CHECK_COUNTER_VALUE(e, int, "foo", EQ, 1);
+  CHECK_COUNTER_VALUE(e, int, "bar", EQ, 2);
+}
+CHECK_BENCHMARK_RESULTS("BM_Counters_AvgThreads/threads:%int",
+                        &CheckAvgThreads);
+}  // end namespace
+
+// ========================================================================= //
+// ---------------------- ThreadAvg Counters Output ------------------------ //
+// ========================================================================= //
+
+namespace {
+void BM_Counters_AvgThreadsRate(benchmark::State& state) {
+  for (auto _ : state) {
+    // This test requires a non-zero CPU time to avoid divide-by-zero
+    auto iterations = static_cast<double>(state.iterations()) *
+                      static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(iterations);
+  }
+  namespace bm = benchmark;
+  state.counters["foo"] = bm::Counter{1, bm::Counter::kAvgThreadsRate};
+  state.counters["bar"] = bm::Counter{2, bm::Counter::kAvgThreadsRate};
+}
+BENCHMARK(BM_Counters_AvgThreadsRate)->ThreadRange(1, 8);
+ADD_CASES(TC_ConsoleOut, {{"^BM_Counters_AvgThreadsRate/threads:%int "
+                           "%console_report bar=%hrfloat/s foo=%hrfloat/s$"}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_AvgThreadsRate/threads:%int\",$"},
+           {"\"family_index\": 7,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_AvgThreadsRate/threads:%int\",$",
+            MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 1,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"bar\": %float,$", MR_Next},
+           {"\"foo\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_Counters_AvgThreadsRate/"
+                       "threads:%int\",%csv_report,%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckAvgThreadsRate(Results const& e) {
+  // this (and not real time) is the time used
+  double t = e.DurationCPUTime() / e.NumThreads();
+  CHECK_FLOAT_COUNTER_VALUE(e, "foo", EQ, 1. / t, 0.001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "bar", EQ, 2. / t, 0.001);
+}
+CHECK_BENCHMARK_RESULTS("BM_Counters_AvgThreadsRate/threads:%int",
+                        &CheckAvgThreadsRate);
+}  // end namespace
+
+// ========================================================================= //
+// ------------------- IterationInvariant Counters Output ------------------ //
+// ========================================================================= //
+
+namespace {
+void BM_Counters_IterationInvariant(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+  namespace bm = benchmark;
+  state.counters["foo"] = bm::Counter{1, bm::Counter::kIsIterationInvariant};
+  state.counters["bar"] = bm::Counter{2, bm::Counter::kIsIterationInvariant};
+}
+BENCHMARK(BM_Counters_IterationInvariant);
+ADD_CASES(TC_ConsoleOut, {{"^BM_Counters_IterationInvariant %console_report "
+                           "bar=%hrfloat foo=%hrfloat$"}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_IterationInvariant\",$"},
+           {"\"family_index\": 8,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_IterationInvariant\",$", MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 1,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"bar\": %float,$", MR_Next},
+           {"\"foo\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_CSVOut,
+          {{"^\"BM_Counters_IterationInvariant\",%csv_report,%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckIterationInvariant(Results const& e) {
+  double its = e.NumIterations();
+  // check that the values are within 0.1% of the expected value
+  CHECK_FLOAT_COUNTER_VALUE(e, "foo", EQ, its, 0.001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "bar", EQ, 2. * its, 0.001);
+}
+CHECK_BENCHMARK_RESULTS("BM_Counters_IterationInvariant",
+                        &CheckIterationInvariant);
+}  // end namespace
+
+// ========================================================================= //
+// ----------------- IterationInvariantRate Counters Output ---------------- //
+// ========================================================================= //
+
+namespace {
+void BM_Counters_kIsIterationInvariantRate(benchmark::State& state) {
+  for (auto _ : state) {
+    // This test requires a non-zero CPU time to avoid divide-by-zero
+    auto iterations = static_cast<double>(state.iterations()) *
+                      static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(iterations);
+  }
+  namespace bm = benchmark;
+  state.counters["foo"] =
+      bm::Counter{1, bm::Counter::kIsIterationInvariantRate};
+  state.counters["bar"] =
+      bm::Counter{2, bm::Counter::kIsRate | bm::Counter::kIsIterationInvariant};
+}
+BENCHMARK(BM_Counters_kIsIterationInvariantRate);
+ADD_CASES(TC_ConsoleOut, {{"^BM_Counters_kIsIterationInvariantRate "
+                           "%console_report bar=%hrfloat/s foo=%hrfloat/s$"}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_kIsIterationInvariantRate\",$"},
+           {"\"family_index\": 9,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_kIsIterationInvariantRate\",$",
+            MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 1,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"bar\": %float,$", MR_Next},
+           {"\"foo\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_Counters_kIsIterationInvariantRate\",%csv_report,"
+                       "%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckIsIterationInvariantRate(Results const& e) {
+  double its = e.NumIterations();
+  double t = e.DurationCPUTime();  // this (and not real time) is the time used
+  // check that the values are within 0.1% of the expected values
+  CHECK_FLOAT_COUNTER_VALUE(e, "foo", EQ, its * 1. / t, 0.001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "bar", EQ, its * 2. / t, 0.001);
+}
+CHECK_BENCHMARK_RESULTS("BM_Counters_kIsIterationInvariantRate",
+                        &CheckIsIterationInvariantRate);
+}  // end namespace
+
+// ========================================================================= //
+// --------------------- AvgIterations Counters Output --------------------- //
+// ========================================================================= //
+
+namespace {
+void BM_Counters_AvgIterations(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+  namespace bm = benchmark;
+  state.counters["foo"] = bm::Counter{1, bm::Counter::kAvgIterations};
+  state.counters["bar"] = bm::Counter{2, bm::Counter::kAvgIterations};
+}
+BENCHMARK(BM_Counters_AvgIterations);
+ADD_CASES(TC_ConsoleOut, {{"^BM_Counters_AvgIterations %console_report "
+                           "bar=%hrfloat foo=%hrfloat$"}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_AvgIterations\",$"},
+           {"\"family_index\": 10,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_AvgIterations\",$", MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 1,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"bar\": %float,$", MR_Next},
+           {"\"foo\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_CSVOut,
+          {{"^\"BM_Counters_AvgIterations\",%csv_report,%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckAvgIterations(Results const& e) {
+  double its = e.NumIterations();
+  // check that the values are within 0.1% of the expected value
+  CHECK_FLOAT_COUNTER_VALUE(e, "foo", EQ, 1. / its, 0.001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "bar", EQ, 2. / its, 0.001);
+}
+CHECK_BENCHMARK_RESULTS("BM_Counters_AvgIterations", &CheckAvgIterations);
+}  // end namespace
+
+// ========================================================================= //
+// ------------------- AvgIterationsRate Counters Output ------------------- //
+// ========================================================================= //
+
+namespace {
+void BM_Counters_kAvgIterationsRate(benchmark::State& state) {
+  for (auto _ : state) {
+    // This test requires a non-zero CPU time to avoid divide-by-zero
+    auto iterations = static_cast<double>(state.iterations()) *
+                      static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(iterations);
+  }
+  namespace bm = benchmark;
+  state.counters["foo"] = bm::Counter{1, bm::Counter::kAvgIterationsRate};
+  state.counters["bar"] =
+      bm::Counter{2, bm::Counter::kIsRate | bm::Counter::kAvgIterations};
+}
+BENCHMARK(BM_Counters_kAvgIterationsRate);
+ADD_CASES(TC_ConsoleOut, {{"^BM_Counters_kAvgIterationsRate "
+                           "%console_report bar=%hrfloat/s foo=%hrfloat/s$"}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_kAvgIterationsRate\",$"},
+           {"\"family_index\": 11,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_kAvgIterationsRate\",$", MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 1,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"bar\": %float,$", MR_Next},
+           {"\"foo\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_Counters_kAvgIterationsRate\",%csv_report,"
+                       "%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckAvgIterationsRate(Results const& e) {
+  double its = e.NumIterations();
+  double t = e.DurationCPUTime();  // this (and not real time) is the time used
+  // check that the values are within 0.1% of the expected values
+  CHECK_FLOAT_COUNTER_VALUE(e, "foo", EQ, 1. / its / t, 0.001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "bar", EQ, 2. / its / t, 0.001);
+}
+CHECK_BENCHMARK_RESULTS("BM_Counters_kAvgIterationsRate",
+                        &CheckAvgIterationsRate);
+}  // end namespace
+
+// ========================================================================= //
+// --------------------------- TEST CASES END ------------------------------ //
+// ========================================================================= //
+
+int main(int argc, char* argv[]) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+  RunOutputTests(argc, argv);
+}

test/user_counters_thousands_test.cc

@@ -0,0 +1,195 @@
+
+#undef NDEBUG
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/counter.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+#include "output_test.h"
+
+namespace {
+// ========================================================================= //
+// ------------------------ Thousands Customisation ------------------------ //
+// ========================================================================= //
+
+void BM_Counters_Thousands(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+  namespace bm = benchmark;
+  state.counters.insert({
+      {"t0_1000000DefaultBase",
+       bm::Counter(1000 * 1000, bm::Counter::kDefaults)},
+      {"t1_1000000Base1000", bm::Counter(1000 * 1000, bm::Counter::kDefaults,
+                                         bm::Counter::OneK::kIs1000)},
+      {"t2_1000000Base1024", bm::Counter(1000 * 1000, bm::Counter::kDefaults,
+                                         bm::Counter::OneK::kIs1024)},
+      {"t3_1048576Base1000", bm::Counter(1024 * 1024, bm::Counter::kDefaults,
+                                         bm::Counter::OneK::kIs1000)},
+      {"t4_1048576Base1024", bm::Counter(1024 * 1024, bm::Counter::kDefaults,
+                                         bm::Counter::OneK::kIs1024)},
+  });
+}
+BENCHMARK(BM_Counters_Thousands)->Repetitions(2);
+ADD_CASES(
+    TC_ConsoleOut,
+    {
+        {"^BM_Counters_Thousands/repeats:2 %console_report "
+         "t0_1000000DefaultBase=1M "
+         "t1_1000000Base1000=1M t2_1000000Base1024=976.56[23]Ki "
+         "t3_1048576Base1000=1.04858M t4_1048576Base1024=1Mi$"},
+        {"^BM_Counters_Thousands/repeats:2 %console_report "
+         "t0_1000000DefaultBase=1M "
+         "t1_1000000Base1000=1M t2_1000000Base1024=976.56[23]Ki "
+         "t3_1048576Base1000=1.04858M t4_1048576Base1024=1Mi$"},
+        {"^BM_Counters_Thousands/repeats:2_mean %console_report "
+         "t0_1000000DefaultBase=1M t1_1000000Base1000=1M "
+         "t2_1000000Base1024=976.56[23]Ki t3_1048576Base1000=1.04858M "
+         "t4_1048576Base1024=1Mi$"},
+        {"^BM_Counters_Thousands/repeats:2_median %console_report "
+         "t0_1000000DefaultBase=1M t1_1000000Base1000=1M "
+         "t2_1000000Base1024=976.56[23]Ki t3_1048576Base1000=1.04858M "
+         "t4_1048576Base1024=1Mi$"},
+        {"^BM_Counters_Thousands/repeats:2_stddev %console_time_only_report [ "
+         "]*2 t0_1000000DefaultBase=0 t1_1000000Base1000=0 "
+         "t2_1000000Base1024=0 t3_1048576Base1000=0 t4_1048576Base1024=0$"},
+    });
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_Thousands/repeats:2\",$"},
+           {"\"family_index\": 0,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_Thousands/repeats:2\",$", MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 2,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"t0_1000000DefaultBase\": 1\\.(0)*e\\+(0)*6,$", MR_Next},
+           {"\"t1_1000000Base1000\": 1\\.(0)*e\\+(0)*6,$", MR_Next},
+           {"\"t2_1000000Base1024\": 1\\.(0)*e\\+(0)*6,$", MR_Next},
+           {"\"t3_1048576Base1000\": 1\\.048576(0)*e\\+(0)*6,$", MR_Next},
+           {"\"t4_1048576Base1024\": 1\\.048576(0)*e\\+(0)*6$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_Thousands/repeats:2\",$"},
+           {"\"family_index\": 0,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_Thousands/repeats:2\",$", MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 2,$", MR_Next},
+           {"\"repetition_index\": 1,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"t0_1000000DefaultBase\": 1\\.(0)*e\\+(0)*6,$", MR_Next},
+           {"\"t1_1000000Base1000\": 1\\.(0)*e\\+(0)*6,$", MR_Next},
+           {"\"t2_1000000Base1024\": 1\\.(0)*e\\+(0)*6,$", MR_Next},
+           {"\"t3_1048576Base1000\": 1\\.048576(0)*e\\+(0)*6,$", MR_Next},
+           {"\"t4_1048576Base1024\": 1\\.048576(0)*e\\+(0)*6$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_Thousands/repeats:2_mean\",$"},
+           {"\"family_index\": 0,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_Thousands/repeats:2\",$", MR_Next},
+           {"\"run_type\": \"aggregate\",$", MR_Next},
+           {"\"repetitions\": 2,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"aggregate_name\": \"mean\",$", MR_Next},
+           {"\"aggregate_unit\": \"time\",$", MR_Next},
+           {"\"iterations\": 2,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"t0_1000000DefaultBase\": 1\\.(0)*e\\+(0)*6,$", MR_Next},
+           {"\"t1_1000000Base1000\": 1\\.(0)*e\\+(0)*6,$", MR_Next},
+           {"\"t2_1000000Base1024\": 1\\.(0)*e\\+(0)*6,$", MR_Next},
+           {"\"t3_1048576Base1000\": 1\\.048576(0)*e\\+(0)*6,$", MR_Next},
+           {"\"t4_1048576Base1024\": 1\\.048576(0)*e\\+(0)*6$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_Thousands/repeats:2_median\",$"},
+           {"\"family_index\": 0,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_Thousands/repeats:2\",$", MR_Next},
+           {"\"run_type\": \"aggregate\",$", MR_Next},
+           {"\"repetitions\": 2,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"aggregate_name\": \"median\",$", MR_Next},
+           {"\"aggregate_unit\": \"time\",$", MR_Next},
+           {"\"iterations\": 2,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"t0_1000000DefaultBase\": 1\\.(0)*e\\+(0)*6,$", MR_Next},
+           {"\"t1_1000000Base1000\": 1\\.(0)*e\\+(0)*6,$", MR_Next},
+           {"\"t2_1000000Base1024\": 1\\.(0)*e\\+(0)*6,$", MR_Next},
+           {"\"t3_1048576Base1000\": 1\\.048576(0)*e\\+(0)*6,$", MR_Next},
+           {"\"t4_1048576Base1024\": 1\\.048576(0)*e\\+(0)*6$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_Thousands/repeats:2_stddev\",$"},
+           {"\"family_index\": 0,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_Thousands/repeats:2\",$", MR_Next},
+           {"\"run_type\": \"aggregate\",$", MR_Next},
+           {"\"repetitions\": 2,$", MR_Next},
+           {"\"threads\": 1,$", MR_Next},
+           {"\"aggregate_name\": \"stddev\",$", MR_Next},
+           {"\"aggregate_unit\": \"time\",$", MR_Next},
+           {"\"iterations\": 2,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"t0_1000000DefaultBase\": 0\\.(0)*e\\+(0)*,$", MR_Next},
+           {"\"t1_1000000Base1000\": 0\\.(0)*e\\+(0)*,$", MR_Next},
+           {"\"t2_1000000Base1024\": 0\\.(0)*e\\+(0)*,$", MR_Next},
+           {"\"t3_1048576Base1000\": 0\\.(0)*e\\+(0)*,$", MR_Next},
+           {"\"t4_1048576Base1024\": 0\\.(0)*e\\+(0)*$", MR_Next},
+           {"}", MR_Next}});
+
+ADD_CASES(
+    TC_CSVOut,
+    {{"^\"BM_Counters_Thousands/"
+      "repeats:2\",%csv_report,1e\\+(0)*6,1e\\+(0)*6,1e\\+(0)*6,1\\.04858e\\+("
+      "0)*6,1\\.04858e\\+(0)*6$"},
+     {"^\"BM_Counters_Thousands/"
+      "repeats:2\",%csv_report,1e\\+(0)*6,1e\\+(0)*6,1e\\+(0)*6,1\\.04858e\\+("
+      "0)*6,1\\.04858e\\+(0)*6$"},
+     {"^\"BM_Counters_Thousands/"
+      "repeats:2_mean\",%csv_report,1e\\+(0)*6,1e\\+(0)*6,1e\\+(0)*6,1\\."
+      "04858e\\+(0)*6,1\\.04858e\\+(0)*6$"},
+     {"^\"BM_Counters_Thousands/"
+      "repeats:2_median\",%csv_report,1e\\+(0)*6,1e\\+(0)*6,1e\\+(0)*6,1\\."
+      "04858e\\+(0)*6,1\\.04858e\\+(0)*6$"},
+     {"^\"BM_Counters_Thousands/repeats:2_stddev\",%csv_report,0,0,0,0,0$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckThousands(Results const& e) {
+  if (e.name != "BM_Counters_Thousands/repeats:2") {
+    return;  // Do not check the aggregates!
+  }
+
+  // check that the values are within 0.01% of the expected values
+  CHECK_FLOAT_COUNTER_VALUE(e, "t0_1000000DefaultBase", EQ, 1000 * 1000,
+                            0.0001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "t1_1000000Base1000", EQ, 1000 * 1000, 0.0001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "t2_1000000Base1024", EQ, 1000 * 1000, 0.0001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "t3_1048576Base1000", EQ, 1024 * 1024, 0.0001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "t4_1048576Base1024", EQ, 1024 * 1024, 0.0001);
+}
+CHECK_BENCHMARK_RESULTS("BM_Counters_Thousands", &CheckThousands);
+}  // end namespace
+
+// ========================================================================= //
+// --------------------------- TEST CASES END ------------------------------ //
+// ========================================================================= //
+
+int main(int argc, char* argv[]) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+  RunOutputTests(argc, argv);
+}

test/user_counters_threads_test.cc

@@ -0,0 +1,625 @@
+
+#undef NDEBUG
+
+#include "benchmark/benchmark_api.h"
+#include "benchmark/registration.h"
+#include "benchmark/state.h"
+#include "benchmark/utils.h"
+#include "output_test.h"
+
+// ========================================================================= //
+// ---------------------- Testing Prologue Output -------------------------- //
+// ========================================================================= //
+
+// clang-format off
+
+ADD_CASES(TC_ConsoleOut,
+          {{"^[-]+$", MR_Next},
+           {"^Benchmark %s Time %s CPU %s Iterations UserCounters...$", MR_Next},
+           {"^[-]+$", MR_Next}});
+ADD_CASES(TC_CSVOut, {{"%csv_header,\"bar\",\"foo\""}});
+
+// clang-format on
+
+// ========================================================================= //
+// ------------------------- Simple Counters Output ------------------------ //
+// ========================================================================= //
+
+namespace {
+void BM_Counters_Simple(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+  state.counters["foo"] = 1;
+  state.counters["bar"] = 2 * static_cast<double>(state.iterations());
+}
+BENCHMARK(BM_Counters_Simple)->ThreadRange(1, 8);
+ADD_CASES(TC_ConsoleOut, {{"^BM_Counters_Simple/threads:%int %console_report "
+                           "bar=%hrfloat foo=%hrfloat$"}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_Simple/threads:%int\",$"},
+           {"\"family_index\": 0,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_Simple/threads:%int\",$", MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 1,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": %int,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"bar\": %float,$", MR_Next},
+           {"\"foo\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(
+    TC_CSVOut,
+    {{"^\"BM_Counters_Simple/threads:%int\",%csv_report,%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckSimple(Results const& e) {
+  double its = e.NumIterations();
+  CHECK_COUNTER_VALUE(e, int, "foo", EQ, 1 * e.NumThreads());
+  // check that the value of bar is within 0.1% of the expected value
+  CHECK_FLOAT_COUNTER_VALUE(e, "bar", EQ, 2. * its, 0.001);
+}
+CHECK_BENCHMARK_RESULTS("BM_Counters_Simple/threads:%int", &CheckSimple);
+}  // end namespace
+
+// ========================================================================= //
+// --------------------- Counters+Items+Bytes/s Output --------------------- //
+// ========================================================================= //
+
+namespace {
+void BM_Counters_WithBytesAndItemsPSec(benchmark::State& state) {
+  for (auto _ : state) {
+    // This test requires a non-zero CPU time to avoid divide-by-zero
+    auto iterations = static_cast<double>(state.iterations()) *
+                      static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(iterations);
+  }
+  state.counters["foo"] = 1;
+  state.SetBytesProcessed(364);
+  state.SetItemsProcessed(150);
+}
+BENCHMARK(BM_Counters_WithBytesAndItemsPSec)->ThreadRange(1, 8);
+ADD_CASES(TC_ConsoleOut,
+          {{"^BM_Counters_WithBytesAndItemsPSec/threads:%int %console_report "
+            "bytes_per_second=%hrfloat/s "
+            "foo=%hrfloat items_per_second=%hrfloat/s$"}});
+ADD_CASES(
+    TC_JSONOut,
+    {{"\"name\": \"BM_Counters_WithBytesAndItemsPSec/threads:%int\",$"},
+     {"\"family_index\": 1,$", MR_Next},
+     {"\"per_family_instance_index\": 0,$", MR_Next},
+     {"\"run_name\": \"BM_Counters_WithBytesAndItemsPSec/threads:%int\",$",
+      MR_Next},
+     {"\"run_type\": \"iteration\",$", MR_Next},
+     {"\"repetitions\": 1,$", MR_Next},
+     {"\"repetition_index\": 0,$", MR_Next},
+     {"\"threads\": %int,$", MR_Next},
+     {"\"iterations\": %int,$", MR_Next},
+     {"\"real_time\": %float,$", MR_Next},
+     {"\"cpu_time\": %float,$", MR_Next},
+     {"\"time_unit\": \"ns\",$", MR_Next},
+     {"\"bytes_per_second\": %float,$", MR_Next},
+     {"\"foo\": %float,$", MR_Next},
+     {"\"items_per_second\": %float$", MR_Next},
+     {"}", MR_Next}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_Counters_WithBytesAndItemsPSec/threads:%int\","
+                       "%csv_bytes_items_report,,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckBytesAndItemsPSec(Results const& e) {
+  // this (and not real time) is the time used
+  double t = e.DurationCPUTime() / e.NumThreads();
+  CHECK_COUNTER_VALUE(e, int, "foo", EQ, 1 * e.NumThreads());
+  // check that the values are within 0.1% of the expected values
+  CHECK_FLOAT_RESULT_VALUE(e, "bytes_per_second", EQ,
+                           (364. * e.NumThreads()) / t, 0.001);
+  CHECK_FLOAT_RESULT_VALUE(e, "items_per_second", EQ,
+                           (150. * e.NumThreads()) / t, 0.001);
+}
+CHECK_BENCHMARK_RESULTS("BM_Counters_WithBytesAndItemsPSec/threads:%int",
+                        &CheckBytesAndItemsPSec);
+}  // end namespace
+
+// ========================================================================= //
+// ------------------------- Rate Counters Output -------------------------- //
+// ========================================================================= //
+namespace {
+void BM_Counters_Rate(benchmark::State& state) {
+  for (auto _ : state) {
+    // This test requires a non-zero CPU time to avoid divide-by-zero
+    auto iterations = static_cast<double>(state.iterations()) *
+                      static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(iterations);
+  }
+  namespace bm = benchmark;
+  state.counters["foo"] = bm::Counter{1, bm::Counter::kIsRate};
+  state.counters["bar"] = bm::Counter{2, bm::Counter::kIsRate};
+}
+BENCHMARK(BM_Counters_Rate)->ThreadRange(1, 8);
+ADD_CASES(TC_ConsoleOut, {{"^BM_Counters_Rate/threads:%int %console_report "
+                           "bar=%hrfloat/s foo=%hrfloat/s$"}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_Rate/threads:%int\",$"},
+           {"\"family_index\": 2,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_Rate/threads:%int\",$", MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 1,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": %int,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"bar\": %float,$", MR_Next},
+           {"\"foo\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_CSVOut,
+          {{"^\"BM_Counters_Rate/threads:%int\",%csv_report,%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckRate(Results const& e) {
+  // this (and not real time) is the time used
+  double t = e.DurationCPUTime() / e.NumThreads();
+  // check that the values are within 0.1% of the expected values
+  CHECK_FLOAT_COUNTER_VALUE(e, "foo", EQ, (1. * e.NumThreads()) / t, 0.001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "bar", EQ, (2. * e.NumThreads()) / t, 0.001);
+}
+CHECK_BENCHMARK_RESULTS("BM_Counters_Rate/threads:%int", &CheckRate);
+}  // end namespace
+
+// ========================================================================= //
+// ----------------------- Inverted Counters Output ------------------------ //
+// ========================================================================= //
+
+namespace {
+void BM_Invert(benchmark::State& state) {
+  for (auto _ : state) {
+    // This test requires a non-zero CPU time to avoid divide-by-zero
+    auto iterations = static_cast<double>(state.iterations()) *
+                      static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(iterations);
+  }
+  namespace bm = benchmark;
+  state.counters["foo"] = bm::Counter{0.0001, bm::Counter::kInvert};
+  state.counters["bar"] = bm::Counter{10000, bm::Counter::kInvert};
+}
+BENCHMARK(BM_Invert)->ThreadRange(1, 8);
+ADD_CASES(
+    TC_ConsoleOut,
+    {{"^BM_Invert/threads:%int %console_report bar=%hrfloatu foo=%hrfloatk$"}});
+ADD_CASES(TC_JSONOut, {{"\"name\": \"BM_Invert/threads:%int\",$"},
+                       {"\"family_index\": 3,$", MR_Next},
+                       {"\"per_family_instance_index\": 0,$", MR_Next},
+                       {"\"run_name\": \"BM_Invert/threads:%int\",$", MR_Next},
+                       {"\"run_type\": \"iteration\",$", MR_Next},
+                       {"\"repetitions\": 1,$", MR_Next},
+                       {"\"repetition_index\": 0,$", MR_Next},
+                       {"\"threads\": %int,$", MR_Next},
+                       {"\"iterations\": %int,$", MR_Next},
+                       {"\"real_time\": %float,$", MR_Next},
+                       {"\"cpu_time\": %float,$", MR_Next},
+                       {"\"time_unit\": \"ns\",$", MR_Next},
+                       {"\"bar\": %float,$", MR_Next},
+                       {"\"foo\": %float$", MR_Next},
+                       {"}", MR_Next}});
+ADD_CASES(TC_CSVOut,
+          {{"^\"BM_Invert/threads:%int\",%csv_report,%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckInvert(Results const& e) {
+  CHECK_FLOAT_COUNTER_VALUE(e, "foo", EQ, 1. / (0.0001 * e.NumThreads()),
+                            0.0001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "bar", EQ, 1. / (10000 * e.NumThreads()),
+                            0.0001);
+}
+CHECK_BENCHMARK_RESULTS("BM_Invert/threads:%int", &CheckInvert);
+}  // end namespace
+
+// ========================================================================= //
+// --------------------- InvertedRate Counters Output ---------------------- //
+// ========================================================================= //
+
+namespace {
+void BM_Counters_InvertedRate(benchmark::State& state) {
+  for (auto _ : state) {
+    // This test requires a non-zero CPU time to avoid divide-by-zero
+    auto iterations = static_cast<double>(state.iterations()) *
+                      static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(iterations);
+  }
+  namespace bm = benchmark;
+  state.counters["foo"] =
+      bm::Counter{1, bm::Counter::kIsRate | bm::Counter::kInvert};
+  state.counters["bar"] =
+      bm::Counter{8192, bm::Counter::kIsRate | bm::Counter::kInvert};
+}
+BENCHMARK(BM_Counters_InvertedRate)->ThreadRange(1, 8);
+ADD_CASES(TC_ConsoleOut,
+          {{"^BM_Counters_InvertedRate/threads:%int %console_report "
+            "bar=%hrfloats foo=%hrfloats$"}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_InvertedRate/threads:%int\",$"},
+           {"\"family_index\": 4,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_InvertedRate/threads:%int\",$",
+            MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 1,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": %int,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"bar\": %float,$", MR_Next},
+           {"\"foo\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_Counters_InvertedRate/"
+                       "threads:%int\",%csv_report,%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckInvertedRate(Results const& e) {
+  // this (and not real time) is the time used
+  double t = e.DurationCPUTime() / e.NumThreads();
+  // check that the values are within 0.1% of the expected values
+  CHECK_FLOAT_COUNTER_VALUE(e, "foo", EQ, t / (e.NumThreads()), 0.001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "bar", EQ, t / (8192.0 * e.NumThreads()), 0.001);
+}
+CHECK_BENCHMARK_RESULTS("BM_Counters_InvertedRate/threads:%int",
+                        &CheckInvertedRate);
+}  // end namespace
+
+// ========================================================================= //
+// ------------------------- Thread Counters Output ------------------------ //
+// ========================================================================= //
+
+namespace {
+void BM_Counters_Threads(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+  state.counters["foo"] = 1;
+  state.counters["bar"] = 2;
+}
+BENCHMARK(BM_Counters_Threads)->ThreadRange(1, 8);
+ADD_CASES(TC_ConsoleOut, {{"^BM_Counters_Threads/threads:%int %console_report "
+                           "bar=%hrfloat foo=%hrfloat$"}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_Threads/threads:%int\",$"},
+           {"\"family_index\": 5,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_Threads/threads:%int\",$", MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 1,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": %int,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"bar\": %float,$", MR_Next},
+           {"\"foo\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(
+    TC_CSVOut,
+    {{"^\"BM_Counters_Threads/threads:%int\",%csv_report,%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckThreads(Results const& e) {
+  CHECK_COUNTER_VALUE(e, int, "foo", EQ, e.NumThreads());
+  CHECK_COUNTER_VALUE(e, int, "bar", EQ, 2 * e.NumThreads());
+}
+CHECK_BENCHMARK_RESULTS("BM_Counters_Threads/threads:%int", &CheckThreads);
+}  // end namespace
+
+// ========================================================================= //
+// ---------------------- ThreadAvg Counters Output ------------------------ //
+// ========================================================================= //
+
+namespace {
+void BM_Counters_AvgThreads(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+  namespace bm = benchmark;
+  state.counters["foo"] = bm::Counter{1, bm::Counter::kAvgThreads};
+  state.counters["bar"] = bm::Counter{2, bm::Counter::kAvgThreads};
+}
+BENCHMARK(BM_Counters_AvgThreads)->ThreadRange(1, 8);
+ADD_CASES(TC_ConsoleOut, {{"^BM_Counters_AvgThreads/threads:%int "
+                           "%console_report bar=%hrfloat foo=%hrfloat$"}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_AvgThreads/threads:%int\",$"},
+           {"\"family_index\": 6,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_AvgThreads/threads:%int\",$", MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 1,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": %int,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"bar\": %float,$", MR_Next},
+           {"\"foo\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(
+    TC_CSVOut,
+    {{"^\"BM_Counters_AvgThreads/threads:%int\",%csv_report,%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckAvgThreads(Results const& e) {
+  CHECK_COUNTER_VALUE(e, int, "foo", EQ, 1);
+  CHECK_COUNTER_VALUE(e, int, "bar", EQ, 2);
+}
+CHECK_BENCHMARK_RESULTS("BM_Counters_AvgThreads/threads:%int",
+                        &CheckAvgThreads);
+}  // end namespace
+
+// ========================================================================= //
+// ---------------------- ThreadAvg Counters Output ------------------------ //
+// ========================================================================= //
+
+namespace {
+void BM_Counters_AvgThreadsRate(benchmark::State& state) {
+  for (auto _ : state) {
+    // This test requires a non-zero CPU time to avoid divide-by-zero
+    auto iterations = static_cast<double>(state.iterations()) *
+                      static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(iterations);
+  }
+  namespace bm = benchmark;
+  state.counters["foo"] = bm::Counter{1, bm::Counter::kAvgThreadsRate};
+  state.counters["bar"] = bm::Counter{2, bm::Counter::kAvgThreadsRate};
+}
+BENCHMARK(BM_Counters_AvgThreadsRate)->ThreadRange(1, 8);
+ADD_CASES(TC_ConsoleOut, {{"^BM_Counters_AvgThreadsRate/threads:%int "
+                           "%console_report bar=%hrfloat/s foo=%hrfloat/s$"}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_AvgThreadsRate/threads:%int\",$"},
+           {"\"family_index\": 7,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_AvgThreadsRate/threads:%int\",$",
+            MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 1,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": %int,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"bar\": %float,$", MR_Next},
+           {"\"foo\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_Counters_AvgThreadsRate/"
+                       "threads:%int\",%csv_report,%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckAvgThreadsRate(Results const& e) {
+  // this (and not real time) is the time used
+  double t = e.DurationCPUTime() / e.NumThreads();
+  CHECK_FLOAT_COUNTER_VALUE(e, "foo", EQ, 1. / t, 0.001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "bar", EQ, 2. / t, 0.001);
+}
+CHECK_BENCHMARK_RESULTS("BM_Counters_AvgThreadsRate/threads:%int",
+                        &CheckAvgThreadsRate);
+}  // end namespace
+
+// ========================================================================= //
+// ------------------- IterationInvariant Counters Output ------------------ //
+// ========================================================================= //
+
+namespace {
+void BM_Counters_IterationInvariant(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+  namespace bm = benchmark;
+  state.counters["foo"] = bm::Counter{1, bm::Counter::kIsIterationInvariant};
+  state.counters["bar"] = bm::Counter{2, bm::Counter::kIsIterationInvariant};
+}
+BENCHMARK(BM_Counters_IterationInvariant)->ThreadRange(1, 8);
+ADD_CASES(TC_ConsoleOut,
+          {{"^BM_Counters_IterationInvariant/threads:%int %console_report "
+            "bar=%hrfloat foo=%hrfloat$"}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_IterationInvariant/threads:%int\",$"},
+           {"\"family_index\": 8,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_IterationInvariant/threads:%int\",$",
+            MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 1,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": %int,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"bar\": %float,$", MR_Next},
+           {"\"foo\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_Counters_IterationInvariant/"
+                       "threads:%int\",%csv_report,%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckIterationInvariant(Results const& e) {
+  double its = e.NumIterations();
+  // check that the values are within 0.1% of the expected value
+  CHECK_FLOAT_COUNTER_VALUE(e, "foo", EQ, its * e.NumThreads(), 0.001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "bar", EQ, 2. * its * e.NumThreads(), 0.001);
+}
+CHECK_BENCHMARK_RESULTS("BM_Counters_IterationInvariant/threads:%int",
+                        &CheckIterationInvariant);
+}  // end namespace
+
+// ========================================================================= //
+// ----------------- IterationInvariantRate Counters Output ---------------- //
+// ========================================================================= //
+
+namespace {
+void BM_Counters_kIsIterationInvariantRate(benchmark::State& state) {
+  for (auto _ : state) {
+    // This test requires a non-zero CPU time to avoid divide-by-zero
+    auto iterations = static_cast<double>(state.iterations()) *
+                      static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(iterations);
+  }
+  namespace bm = benchmark;
+  state.counters["foo"] =
+      bm::Counter{1, bm::Counter::kIsIterationInvariantRate};
+  state.counters["bar"] =
+      bm::Counter{2, bm::Counter::kIsRate | bm::Counter::kIsIterationInvariant};
+}
+BENCHMARK(BM_Counters_kIsIterationInvariantRate)->ThreadRange(1, 8);
+ADD_CASES(TC_ConsoleOut,
+          {{"^BM_Counters_kIsIterationInvariantRate/threads:%int "
+            "%console_report bar=%hrfloat/s foo=%hrfloat/s$"}});
+ADD_CASES(
+    TC_JSONOut,
+    {{"\"name\": \"BM_Counters_kIsIterationInvariantRate/threads:%int\",$"},
+     {"\"family_index\": 9,$", MR_Next},
+     {"\"per_family_instance_index\": 0,$", MR_Next},
+     {"\"run_name\": \"BM_Counters_kIsIterationInvariantRate/threads:%int\",$",
+      MR_Next},
+     {"\"run_type\": \"iteration\",$", MR_Next},
+     {"\"repetitions\": 1,$", MR_Next},
+     {"\"repetition_index\": 0,$", MR_Next},
+     {"\"threads\": %int,$", MR_Next},
+     {"\"iterations\": %int,$", MR_Next},
+     {"\"real_time\": %float,$", MR_Next},
+     {"\"cpu_time\": %float,$", MR_Next},
+     {"\"time_unit\": \"ns\",$", MR_Next},
+     {"\"bar\": %float,$", MR_Next},
+     {"\"foo\": %float$", MR_Next},
+     {"}", MR_Next}});
+ADD_CASES(
+    TC_CSVOut,
+    {{"^\"BM_Counters_kIsIterationInvariantRate/threads:%int\",%csv_report,"
+      "%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckIsIterationInvariantRate(Results const& e) {
+  double its = e.NumIterations();
+  // this (and not real time) is the time used
+  double t = e.DurationCPUTime() / e.NumThreads();
+  // check that the values are within 0.1% of the expected values
+  CHECK_FLOAT_COUNTER_VALUE(e, "foo", EQ, its * 1. * e.NumThreads() / t, 0.001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "bar", EQ, its * 2. * e.NumThreads() / t, 0.001);
+}
+CHECK_BENCHMARK_RESULTS("BM_Counters_kIsIterationInvariantRate/threads:%int",
+                        &CheckIsIterationInvariantRate);
+}  // end namespace
+
+// ========================================================================= //
+// --------------------- AvgIterations Counters Output --------------------- //
+// ========================================================================= //
+
+namespace {
+void BM_Counters_AvgIterations(benchmark::State& state) {
+  for (auto _ : state) {
+  }
+  namespace bm = benchmark;
+  state.counters["foo"] = bm::Counter{1, bm::Counter::kAvgIterations};
+  state.counters["bar"] = bm::Counter{2, bm::Counter::kAvgIterations};
+}
+BENCHMARK(BM_Counters_AvgIterations)->ThreadRange(1, 8);
+ADD_CASES(TC_ConsoleOut,
+          {{"^BM_Counters_AvgIterations/threads:%int %console_report "
+            "bar=%hrfloat foo=%hrfloat$"}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_AvgIterations/threads:%int\",$"},
+           {"\"family_index\": 10,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_AvgIterations/threads:%int\",$",
+            MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 1,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": %int,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"bar\": %float,$", MR_Next},
+           {"\"foo\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_CSVOut, {{"^\"BM_Counters_AvgIterations/"
+                       "threads:%int\",%csv_report,%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckAvgIterations(Results const& e) {
+  double its = e.NumIterations();
+  // check that the values are within 0.1% of the expected value
+  CHECK_FLOAT_COUNTER_VALUE(e, "foo", EQ, 1. * e.NumThreads() / its, 0.001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "bar", EQ, 2. * e.NumThreads() / its, 0.001);
+}
+CHECK_BENCHMARK_RESULTS("BM_Counters_AvgIterations/threads:%int",
+                        &CheckAvgIterations);
+}  // end namespace
+
+// ========================================================================= //
+// ------------------- AvgIterationsRate Counters Output ------------------- //
+// ========================================================================= //
+
+namespace {
+void BM_Counters_kAvgIterationsRate(benchmark::State& state) {
+  for (auto _ : state) {
+    // This test requires a non-zero CPU time to avoid divide-by-zero
+    auto iterations = static_cast<double>(state.iterations()) *
+                      static_cast<double>(state.iterations());
+    benchmark::DoNotOptimize(iterations);
+  }
+  namespace bm = benchmark;
+  state.counters["foo"] = bm::Counter{1, bm::Counter::kAvgIterationsRate};
+  state.counters["bar"] =
+      bm::Counter{2, bm::Counter::kIsRate | bm::Counter::kAvgIterations};
+}
+BENCHMARK(BM_Counters_kAvgIterationsRate)->ThreadRange(1, 8);
+ADD_CASES(TC_ConsoleOut, {{"^BM_Counters_kAvgIterationsRate/threads:%int "
+                           "%console_report bar=%hrfloat/s foo=%hrfloat/s$"}});
+ADD_CASES(TC_JSONOut,
+          {{"\"name\": \"BM_Counters_kAvgIterationsRate/threads:%int\",$"},
+           {"\"family_index\": 11,$", MR_Next},
+           {"\"per_family_instance_index\": 0,$", MR_Next},
+           {"\"run_name\": \"BM_Counters_kAvgIterationsRate/threads:%int\",$",
+            MR_Next},
+           {"\"run_type\": \"iteration\",$", MR_Next},
+           {"\"repetitions\": 1,$", MR_Next},
+           {"\"repetition_index\": 0,$", MR_Next},
+           {"\"threads\": %int,$", MR_Next},
+           {"\"iterations\": %int,$", MR_Next},
+           {"\"real_time\": %float,$", MR_Next},
+           {"\"cpu_time\": %float,$", MR_Next},
+           {"\"time_unit\": \"ns\",$", MR_Next},
+           {"\"bar\": %float,$", MR_Next},
+           {"\"foo\": %float$", MR_Next},
+           {"}", MR_Next}});
+ADD_CASES(TC_CSVOut,
+          {{"^\"BM_Counters_kAvgIterationsRate/threads:%int\",%csv_report,"
+            "%float,%float$"}});
+// VS2013 does not allow this function to be passed as a lambda argument
+// to CHECK_BENCHMARK_RESULTS()
+void CheckAvgIterationsRate(Results const& e) {
+  double its = e.NumIterations();
+  // this (and not real time) is the time used
+  double t = e.DurationCPUTime() / e.NumThreads();
+  // check that the values are within 0.1% of the expected values
+  CHECK_FLOAT_COUNTER_VALUE(e, "foo", EQ, 1. * e.NumThreads() / its / t, 0.001);
+  CHECK_FLOAT_COUNTER_VALUE(e, "bar", EQ, 2. * e.NumThreads() / its / t, 0.001);
+}
+CHECK_BENCHMARK_RESULTS("BM_Counters_kAvgIterationsRate/threads:%int",
+                        &CheckAvgIterationsRate);
+}  // end namespace
+
+// ========================================================================= //
+// --------------------------- TEST CASES END ------------------------------ //
+// ========================================================================= //
+
+int main(int argc, char* argv[]) {
+  benchmark::MaybeReenterWithoutASLR(argc, argv);
+  RunOutputTests(argc, argv);
+}