Merge pull request #1285 from cppalliance/conversions

Author: mborland

doc/modules/ROOT/nav.adoc

@@ -3,6 +3,8 @@
 * xref:examples.adoc[]
 ** xref:examples.adoc#examples_construction[Basic Construction]
 ** xref:examples.adoc#examples_basic_math[Basic Arithmetic]
+** xref:examples.adoc#examples_conversions[Conversions]
+*** xref:examples.adoc#examples_binary_floating_conversions[Binary Floating Point]
 ** xref:examples.adoc#examples_promotion[Promotion and Mixed Decimal Arithmetic]
 ** xref:examples.adoc#examples_charconv[`<charconv>`]
 ** xref:examples.adoc#examples_generic_programming[Generic Programming]

doc/modules/ROOT/pages/examples.adoc

@@ -69,6 +69,30 @@ Decimal Result: 100
 ....
 ====
 
+[#examples_conversions]
+== Conversions
+
+[#examples_binary_floating_conversions]
+=== Binary Floating Point Conversions
+
+.This https://github.com/cppalliance/decimal/blob/develop/examples/binary_float_conversions.cpp[example] shows show to construct a binary floating point value from decimal type and vice versa
+====
+[source,c++]
+----
+include::example$binary_float_conversions.cpp[]
+----
+
+.Expected Output:
+....
+Decimal QNAN converts to double QNAN
+Decimal INFINITY converts to double INFINITY
+decimal64_t pi: 3.141592653589793
+     double pi: 3.141592653589793
+  Converted pi: 3.141592653589793
+decimal32_t pi: 3.141593
+....
+====
+
 [#examples_promotion]
 == Promotion and Mixed Decimal Arithmetic
 

examples/binary_float_conversions.cpp

@@ -0,0 +1,79 @@
+// Copyright 2025 Matt Borland
+// Distributed under the Boost Software License, Version 1.0.
+// https://www.boost.org/LICENSE_1_0.txt
+//
+// This file demonstrates how to convert various types to decimal types and back,
+// along with edge case handling
+
+#include <boost/decimal/decimal32_t.hpp>        // For type decimal32_t
+#include <boost/decimal/decimal64_t.hpp>        // For type decimal64_t
+#include <boost/decimal/cmath.hpp>              // For decimal support of cmath functions
+#include <boost/decimal/iostream.hpp>           // For decimal support of <iostream> and <iomanip>
+#include <boost/decimal/numbers.hpp>            // For decimal support of <numbers>
+#include <iostream>
+#include <cmath>
+#include <limits>
+
+int main()
+{
+    using boost::decimal::decimal32_t;      // Type decimal32_t
+    using boost::decimal::decimal64_t;      // Type decimal64_t
+
+    // Non-finite values construct the equivalent non-finite value in binary floating point
+    constexpr decimal64_t decimal_qnan {std::numeric_limits<decimal64_t>::quiet_NaN()};
+    const double double_from_qnan {static_cast<double>(decimal_qnan)};
+
+    // Note here that we must use boost::decimal::isnan for decimal types,
+    // as it is illegal to overload std::isnan
+    if (boost::decimal::isnan(decimal_qnan) && std::isnan(double_from_qnan))
+    {
+        std::cout << "Decimal QNAN converts to double QNAN\n";
+    }
+
+    constexpr decimal64_t decimal_inf {std::numeric_limits<decimal64_t>::infinity()};
+    const double double_from_inf {static_cast<double>(decimal_inf)};
+
+    // Same as the above but with INF instead of NAN
+    if (boost::decimal::isinf(decimal_inf) && std::isinf(double_from_inf))
+    {
+        std::cout << "Decimal INFINITY converts to double INFINITY\n";
+    }
+
+    // For finite values we make a best effort approach to covert to double
+    // We are able to decompose the decimal floating point value into a sign, significand, and exponent.
+    // From there we use the methods outline in Daniel Lemire's "Number Parsing at a Gigabyte a Second",
+    // to construct the binary floating point value.
+    // See: https://arxiv.org/pdf/2101.11408
+
+    // Construct the decimal64_t version of pi using our pre-computed constants from <boost/decimal/numbers.hpp>
+    constexpr decimal64_t decimal_pi {boost::decimal::numbers::pi_v<decimal64_t>};
+    const double double_from_pi {static_cast<double>(decimal_pi)};
+
+    std::cout << std::setprecision(std::numeric_limits<decimal64_t>::digits10)
+              << "decimal64_t pi: " << decimal_pi << '\n'
+              << "     double pi: " << double_from_pi << '\n';
+
+    // To construct a decimal64_t from double we use the methods described in "Ryu: fast float-to-string conversion"
+    // See: https://dl.acm.org/doi/10.1145/3192366.3192369
+    // This paper shows how to decompose a double into it's sign, significand, and exponent
+    // Once we have those components we can use the normal constructors of the decimal types to construct
+    // Since we are using the normal constructors here,
+    // any construction from this conversion is subject to the current rounding mode
+    // Such as with a lossy conversion like shown (double -> decimal32_t)
+
+    const decimal64_t decimal_from_double {static_cast<decimal64_t>(double_from_pi)};
+    const decimal32_t lossy_decimal_from_double {static_cast<decimal32_t>(double_from_pi)};
+
+    std::cout << "  Converted pi: " << decimal_from_double << '\n'
+              << "decimal32_t pi: " << lossy_decimal_from_double << '\n';
+
+
+    // Other than what has already been shown,
+    // there are no other ways in the library to convert between decimal types and binary floating point types
+    // The reason for this is to discourage their use.
+    //
+    // You can use intermediate representations like strings if you want to make these conversions,
+    // and want to be sure about what the resulting value will be
+
+    return 0;
+}

test/Jamfile

@@ -215,6 +215,7 @@ run ../examples/basic_arithmetic.cpp ;
 run ../examples/to_from_file.cpp ;
 run ../examples/addition.cpp ;
 run ../examples/debugger.cpp ;
+run ../examples/binary_float_conversions.cpp ;
 
 # Test compilation of separate headers
 compile compile_tests/bid_conversion.cpp ;