Add implementation of helpers for uniform float distribution

Specifically we add
 * `gamma(a, b)`, which returns the magnitude of largest 1-ULP
   step in range [a, b].
 * `count_equidistant_float(a, b, distance)`, which returns the
   number of equi-distant floats in range [a, b].

src/CMakeLists.txt

@@ -109,6 +109,7 @@ set(IMPL_HEADERS
   ${SOURCES_DIR}/internal/catch_polyfills.hpp
   ${SOURCES_DIR}/internal/catch_preprocessor.hpp
   ${SOURCES_DIR}/internal/catch_preprocessor_remove_parens.hpp
+  ${SOURCES_DIR}/internal/catch_random_floating_point_helpers.hpp
   ${SOURCES_DIR}/internal/catch_random_number_generator.hpp
   ${SOURCES_DIR}/internal/catch_random_seed_generation.hpp
   ${SOURCES_DIR}/internal/catch_reporter_registry.hpp

src/catch2/benchmark/detail/catch_stats.cpp

@@ -10,6 +10,7 @@
 #include <catch2/benchmark/detail/catch_stats.hpp>
 
 #include <catch2/internal/catch_compiler_capabilities.hpp>
+#include <catch2/internal/catch_floating_point_helpers.hpp>
 
 #include <algorithm>
 #include <cassert>
@@ -184,20 +185,6 @@ namespace Catch {
                     return std::sqrt( variance );
                 }
 
-#if defined( __GNUC__ ) || defined( __clang__ )
-#    pragma GCC diagnostic push
-#    pragma GCC diagnostic ignored "-Wfloat-equal"
-#endif
-                // Used when we know we want == comparison of two doubles
-                // to centralize warning suppression
-                static bool directCompare( double lhs, double rhs ) {
-                    return lhs == rhs;
-                }
-#if defined( __GNUC__ ) || defined( __clang__ )
-#    pragma GCC diagnostic pop
-#endif
-
-
                 static sample jackknife( double ( *estimator )( double const*,
                                                                 double const* ),
                                          double* first,
@@ -234,7 +221,7 @@ namespace Catch {
                 double g = idx - j;
                 std::nth_element(first, first + j, last);
                 auto xj = first[j];
-                if ( directCompare( g, 0 ) ) {
+                if ( Catch::Detail::directCompare( g, 0 ) ) {
                     return xj;
                 }
 
@@ -338,7 +325,7 @@ namespace Catch {
                                    [point]( double x ) { return x < point; } ) /
                     static_cast<double>( n );
                 // degenerate case with uniform samples
-                if ( directCompare( prob_n, 0. ) ) {
+                if ( Catch::Detail::directCompare( prob_n, 0. ) ) {
                     return { point, point, point, confidence_level };
                 }
 

src/catch2/catch_all.hpp

@@ -91,6 +91,7 @@
 #include <catch2/internal/catch_preprocessor.hpp>
 #include <catch2/internal/catch_preprocessor_internal_stringify.hpp>
 #include <catch2/internal/catch_preprocessor_remove_parens.hpp>
+#include <catch2/internal/catch_random_floating_point_helpers.hpp>
 #include <catch2/internal/catch_random_number_generator.hpp>
 #include <catch2/internal/catch_random_seed_generation.hpp>
 #include <catch2/internal/catch_reporter_registry.hpp>

src/catch2/internal/catch_floating_point_helpers.cpp

@@ -27,6 +27,17 @@ namespace Catch {
             return i;
         }
 
+#if defined( __GNUC__ ) || defined( __clang__ )
+#    pragma GCC diagnostic push
+#    pragma GCC diagnostic ignored "-Wfloat-equal"
+#endif
+        bool directCompare( float lhs, float rhs ) { return lhs == rhs; }
+        bool directCompare( double lhs, double rhs ) { return lhs == rhs; }
+#if defined( __GNUC__ ) || defined( __clang__ )
+#    pragma GCC diagnostic pop
+#endif
+
+
     } // end namespace Detail
 } // end namespace Catch
 

src/catch2/internal/catch_floating_point_helpers.hpp

@@ -22,6 +22,11 @@ namespace Catch {
         uint32_t convertToBits(float f);
         uint64_t convertToBits(double d);
 
+        // Used when we know we want == comparison of two doubles
+        // to centralize warning suppression
+        bool directCompare( float lhs, float rhs );
+        bool directCompare( double lhs, double rhs );
+
     } // end namespace Detail
 
 

src/catch2/internal/catch_random_floating_point_helpers.hpp

@@ -0,0 +1,94 @@
+
+//              Copyright Catch2 Authors
+// Distributed under the Boost Software License, Version 1.0.
+//   (See accompanying file LICENSE.txt or copy at
+//        https://www.boost.org/LICENSE_1_0.txt)
+
+// SPDX-License-Identifier: BSL-1.0
+
+#ifndef CATCH_RANDOM_FLOATING_POINT_HELPERS_HPP_INCLUDED
+#define CATCH_RANDOM_FLOATING_POINT_HELPERS_HPP_INCLUDED
+
+#include <catch2/internal/catch_polyfills.hpp>
+
+#include <cassert>
+#include <cmath>
+#include <cstdint>
+#include <limits>
+#include <type_traits>
+
+namespace Catch {
+
+    namespace Detail {
+        /**
+         * Returns the largest magnitude of 1-ULP distance inside the [a, b] range.
+         *
+         * Assumes `a < b`.
+         */
+        template <typename FloatType>
+        FloatType gamma(FloatType a, FloatType b) {
+            static_assert( std::is_floating_point<FloatType>::value,
+                           "gamma returns the largest ULP magnitude within "
+                           "floating point range [a, b]. This only makes sense "
+                           "for floating point types" );
+            assert( a < b );
+
+            const auto gamma_up = Catch::nextafter( a, std::numeric_limits<FloatType>::infinity() ) - a;
+            const auto gamma_down = b - Catch::nextafter( b, -std::numeric_limits<FloatType>::infinity() );
+
+            return gamma_up < gamma_down ? gamma_down : gamma_up;
+        }
+
+        template <typename FloatingPoint>
+        struct DistanceTypePicker;
+        template <>
+        struct DistanceTypePicker<float> {
+            using type = std::uint32_t;
+        };
+        template <>
+        struct DistanceTypePicker<double> {
+            using type = std::uint64_t;
+        };
+
+        template <typename T>
+        using DistanceType = typename DistanceTypePicker<T>::type;
+
+#if defined( __GNUC__ ) || defined( __clang__ )
+#    pragma GCC diagnostic push
+#    pragma GCC diagnostic ignored "-Wfloat-equal"
+#endif
+        /**
+         * Computes the number of equi-distant floats in [a, b]
+         *
+         * Since not every range can be split into equidistant floats
+         * exactly, we actually compute ceil(b/distance - a/distance),
+         * because in those cases we want to overcount.
+         *
+         * Uses modified Dekker's FastTwoSum algorithm to handle rounding.
+         */
+        template <typename FloatType>
+        DistanceType<FloatType>
+        count_equidistant_floats( FloatType a, FloatType b, FloatType distance ) {
+            assert( a < b );
+            // We get distance as gamma for our uniform float distribution,
+            // so this will round perfectly.
+            const auto ag = a / distance;
+            const auto bg = b / distance;
+
+            const auto s = bg - ag;
+            const auto err = ( std::fabs( a ) <= std::fabs( b ) )
+                                 ? -ag - ( s - bg )
+                                 : bg - ( s + ag );
+            const auto ceil_s = static_cast<DistanceType<FloatType>>( std::ceil( s ) );
+
+            return ( ceil_s != s ) ? ceil_s : ceil_s + ( err > 0 );
+        }
+#if defined( __GNUC__ ) || defined( __clang__ )
+#    pragma GCC diagnostic pop
+#endif
+
+    }
+
+} // end namespace Catch
+
+#endif // CATCH_RANDOM_FLOATING_POINT_HELPERS_HPP_INCLUDED

src/catch2/meson.build

@@ -115,6 +115,7 @@ internal_headers = [
   'internal/catch_preprocessor.hpp',
   'internal/catch_preprocessor_internal_stringify.hpp',
   'internal/catch_preprocessor_remove_parens.hpp',
+  'internal/catch_random_floating_point_helpers.hpp',
   'internal/catch_random_number_generator.hpp',
   'internal/catch_random_seed_generation.hpp',
   'internal/catch_reporter_registry.hpp',

tests/SelfTest/IntrospectiveTests/FloatingPoint.tests.cpp

@@ -9,7 +9,9 @@
 #include <catch2/catch_test_macros.hpp>
 #include <catch2/catch_template_test_macros.hpp>
 #include <catch2/internal/catch_floating_point_helpers.hpp>
+#include <catch2/internal/catch_random_floating_point_helpers.hpp>
 
+#include <limits>
 
 TEST_CASE("convertToBits", "[floating-point][conversion]") {
     using Catch::Detail::convertToBits;
@@ -72,3 +74,60 @@ TEST_CASE("UlpDistance", "[floating-point][ulp][approvals]") {
     CHECK( ulpDistance( 1.f, 2.f ) == 0x80'00'00 );
     CHECK( ulpDistance( -2.f, 2.f ) == 0x80'00'00'00 );
 }
+
+
+
+TEMPLATE_TEST_CASE("gamma", "[approvals][floating-point][ulp][gamma]", float, double) {
+    using Catch::Detail::gamma;
+    using Catch::Detail::directCompare;
+
+    // We need to butcher the equal tests with the directCompare helper,
+    // because the Wfloat-equal triggers in decomposer rather than here,
+    // so we cannot locally disable it. Goddamn GCC.
+    CHECK( directCompare( gamma( TestType( -1. ), TestType( 1. ) ),
+                          gamma( TestType( 0.2332 ), TestType( 1.0 ) ) ) );
+    CHECK( directCompare( gamma( TestType( -2. ), TestType( 0 ) ),
+                          gamma( TestType( 1. ), TestType( 1.5 ) ) ) );
+    CHECK( gamma( TestType( 0. ), TestType( 1.0 ) ) <
+           gamma( TestType( 1.0 ), TestType( 1.5 ) ) );
+    CHECK( gamma( TestType( 0 ), TestType( 1. ) ) <
+           std::numeric_limits<TestType>::epsilon() );
+    CHECK( gamma( TestType( -1. ), TestType( -0. ) ) <
+           std::numeric_limits<TestType>::epsilon() );
+    CHECK( directCompare( gamma( TestType( 1. ), TestType( 2. ) ),
+                          std::numeric_limits<TestType>::epsilon() ) );
+    CHECK( directCompare( gamma( TestType( -2. ), TestType( -1. ) ),
+                          std::numeric_limits<TestType>::epsilon() ) );
+}
+
+TEMPLATE_TEST_CASE("count_equidistant_floats",
+                   "[approvals][floating-point][distance]",
+                   float,
+                   double) {
+    using Catch::Detail::count_equidistant_floats;
+    auto count_steps = []( TestType a, TestType b ) {
+        return count_equidistant_floats( a, b, Catch::Detail::gamma( a, b ) );
+    };
+
+    CHECK( count_steps( TestType( -1. ), TestType( 1. ) ) ==
+           2 * count_steps( TestType( 0. ), TestType( 1. ) ) );
+}
+
+TEST_CASE( "count_equidistant_floats",
+           "[approvals][floating-point][distance]" ) {
+    using Catch::Detail::count_equidistant_floats;
+    auto count_floats_with_scaled_ulp = []( auto a, auto b ) {
+        return count_equidistant_floats( a, b, Catch::Detail::gamma( a, b ) );
+    };
+
+    CHECK( count_floats_with_scaled_ulp( 1., 1.5 ) == 1ull << 51 );
+    CHECK( count_floats_with_scaled_ulp( 1.25, 1.5 ) == 1ull << 50 );
+    CHECK( count_floats_with_scaled_ulp( 1.f, 1.5f ) == 1 << 22 );
+
+    STATIC_REQUIRE( std::is_same<std::uint64_t,
+                                 decltype( count_floats_with_scaled_ulp(
+                                     0., 1. ) )>::value );
+    STATIC_REQUIRE( std::is_same<std::uint32_t,
+                                 decltype( count_floats_with_scaled_ulp(
+                                     0.f, 1.f ) )>::value );
+}