Add and test CUDA support for library free functions

doc/modules/ROOT/pages/bit.adoc

@@ -26,7 +26,7 @@ For `u128`, the functions delegate to the `boost::int128` implementations.
 [source,c++]
 ----
 template <unsigned_library_type UnsignedInt>
-[[nodiscard]] constexpr auto has_single_bit(UnsignedInt x) noexcept -> bool;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto has_single_bit(UnsignedInt x) noexcept -> bool;
 ----
 
 Returns `true` if `x` is a power of two.
@@ -37,7 +37,7 @@ See https://en.cppreference.com/w/cpp/numeric/has_single_bit.html[`std::has_sing
 [source,c++]
 ----
 template <unsigned_library_type UnsignedInt>
-[[nodiscard]] constexpr auto bit_ceil(UnsignedInt x) noexcept -> UnsignedInt;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto bit_ceil(UnsignedInt x) noexcept -> UnsignedInt;
 ----
 
 Returns the smallest power of two not less than `x`.
@@ -49,7 +49,7 @@ See https://en.cppreference.com/w/cpp/numeric/bit_ceil.html[`std::bit_ceil`].
 [source,c++]
 ----
 template <unsigned_library_type UnsignedInt>
-[[nodiscard]] constexpr auto bit_floor(UnsignedInt x) noexcept -> UnsignedInt;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto bit_floor(UnsignedInt x) noexcept -> UnsignedInt;
 ----
 
 Returns the largest power of two not greater than `x`.
@@ -61,7 +61,7 @@ See https://en.cppreference.com/w/cpp/numeric/bit_floor.html[`std::bit_floor`].
 [source,c++]
 ----
 template <unsigned_library_type UnsignedInt>
-[[nodiscard]] constexpr auto bit_width(UnsignedInt x) noexcept -> int;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto bit_width(UnsignedInt x) noexcept -> int;
 ----
 
 Returns the number of bits needed to represent `x` (i.e., 1 + floor(log2(x)) for x > 0, or 0 for x == 0).
@@ -74,7 +74,7 @@ See https://en.cppreference.com/w/cpp/numeric/bit_width.html[`std::bit_width`].
 [source,c++]
 ----
 template <non_bounded_unsigned_library_type UnsignedInt>
-[[nodiscard]] constexpr auto rotl(UnsignedInt x, int s) noexcept -> UnsignedInt;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto rotl(UnsignedInt x, int s) noexcept -> UnsignedInt;
 ----
 
 Computes the result of bitwise left-rotating `x` by `s` positions.
@@ -87,7 +87,7 @@ NOTE: `rotl` is not available for `bounded_uint` types. Bit rotation can produce
 [source,c++]
 ----
 template <non_bounded_unsigned_library_type UnsignedInt>
-[[nodiscard]] constexpr auto rotr(UnsignedInt x, int s) noexcept -> UnsignedInt;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto rotr(UnsignedInt x, int s) noexcept -> UnsignedInt;
 ----
 
 Computes the result of bitwise right-rotating `x` by `s` positions.
@@ -102,7 +102,7 @@ NOTE: `rotr` is not available for `bounded_uint` types. Bit rotation can produce
 [source,c++]
 ----
 template <unsigned_library_type UnsignedInt>
-[[nodiscard]] constexpr auto countl_zero(UnsignedInt x) noexcept -> int;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto countl_zero(UnsignedInt x) noexcept -> int;
 ----
 
 Returns the number of consecutive 0-bits starting from the most significant bit.
@@ -113,7 +113,7 @@ See https://en.cppreference.com/w/cpp/numeric/countl_zero.html[`std::countl_zero
 [source,c++]
 ----
 template <unsigned_library_type UnsignedInt>
-[[nodiscard]] constexpr auto countl_one(UnsignedInt x) noexcept -> int;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto countl_one(UnsignedInt x) noexcept -> int;
 ----
 
 Returns the number of consecutive 1-bits starting from the most significant bit.
@@ -124,7 +124,7 @@ See https://en.cppreference.com/w/cpp/numeric/countl_one.html[`std::countl_one`]
 [source,c++]
 ----
 template <unsigned_library_type UnsignedInt>
-[[nodiscard]] constexpr auto countr_zero(UnsignedInt x) noexcept -> int;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto countr_zero(UnsignedInt x) noexcept -> int;
 ----
 
 Returns the number of consecutive 0-bits starting from the least significant bit.
@@ -135,7 +135,7 @@ See https://en.cppreference.com/w/cpp/numeric/countr_zero.html[`std::countr_zero
 [source,c++]
 ----
 template <unsigned_library_type UnsignedInt>
-[[nodiscard]] constexpr auto countr_one(UnsignedInt x) noexcept -> int;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto countr_one(UnsignedInt x) noexcept -> int;
 ----
 
 Returns the number of consecutive 1-bits starting from the least significant bit.
@@ -146,7 +146,7 @@ See https://en.cppreference.com/w/cpp/numeric/countr_one.html[`std::countr_one`]
 [source,c++]
 ----
 template <unsigned_library_type UnsignedInt>
-[[nodiscard]] constexpr auto popcount(UnsignedInt x) noexcept -> int;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto popcount(UnsignedInt x) noexcept -> int;
 ----
 
 Returns the number of 1-bits in `x`.
@@ -159,7 +159,7 @@ See https://en.cppreference.com/w/cpp/numeric/popcount.html[`std::popcount`].
 [source,c++]
 ----
 template <non_bounded_integral_library_type Int>
-[[nodiscard]] constexpr auto byteswap(Int x) noexcept -> Int;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto byteswap(Int x) noexcept -> Int;
 ----
 
 Reverses the bytes of `x`.
@@ -172,7 +172,7 @@ NOTE: `byteswap` is not available for `bounded_uint` types. Byte reversal can pr
 [source,c++]
 ----
 template <non_bounded_integral_library_type Int>
-[[nodiscard]] constexpr auto bitswap(Int x) noexcept -> Int;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto bitswap(Int x) noexcept -> Int;
 ----
 
 Reverses all bits of `x`.

doc/modules/ROOT/pages/byte_conversions.adoc

@@ -170,7 +170,7 @@ The value is first converted to big-endian byte order using `to_be`, then reinte
 [source,c++]
 ----
 template <non_bounded_integral_library_type T>
-[[nodiscard]] constexpr auto to_be_bytes(const T value) noexcept -> std::array<std::byte, sizeof(T)>;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto to_be_bytes(const T value) noexcept -> std::array<std::byte, sizeof(T)>;
 ----
 
 === Parameters
@@ -203,7 +203,7 @@ The bytes are reinterpreted as the underlying type and then converted from big-e
 [source,c++]
 ----
 template <non_bounded_integral_library_type T, std::size_t N>
-[[nodiscard]] constexpr auto from_be_bytes(const std::span<const std::byte, N> bytes) -> T;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto from_be_bytes(const std::span<const std::byte, N> bytes) -> T;
 ----
 
 === Parameters
@@ -251,7 +251,7 @@ The value is first converted to little-endian byte order using `to_le`, then rei
 [source,c++]
 ----
 template <non_bounded_integral_library_type T>
-[[nodiscard]] constexpr auto to_le_bytes(const T value) noexcept -> std::array<std::byte, sizeof(T)>;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto to_le_bytes(const T value) noexcept -> std::array<std::byte, sizeof(T)>;
 ----
 
 === Parameters
@@ -284,7 +284,7 @@ The bytes are reinterpreted as the underlying type and then converted from littl
 [source,c++]
 ----
 template <non_bounded_integral_library_type T, std::size_t N>
-[[nodiscard]] constexpr auto from_le_bytes(const std::span<const std::byte, N> bytes) -> T;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto from_le_bytes(const std::span<const std::byte, N> bytes) -> T;
 ----
 
 === Parameters
@@ -334,7 +334,7 @@ The result is equivalent to `std::bit_cast<std::array<std::byte, sizeof(T)>>(val
 [source,c++]
 ----
 template <non_bounded_integral_library_type T>
-[[nodiscard]] constexpr auto to_ne_bytes(const T value) noexcept -> std::array<std::byte, sizeof(T)>;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto to_ne_bytes(const T value) noexcept -> std::array<std::byte, sizeof(T)>;
 ----
 
 === Parameters
@@ -368,7 +368,7 @@ Delegates to `from_le_bytes` on little-endian platforms and `from_be_bytes` on b
 [source,c++]
 ----
 template <non_bounded_integral_library_type T, std::size_t N>
-[[nodiscard]] constexpr auto from_ne_bytes(const std::span<const std::byte, N> bytes) -> T;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto from_ne_bytes(const std::span<const std::byte, N> bytes) -> T;
 ----
 
 === Parameters

doc/modules/ROOT/pages/charconv.adoc

@@ -33,12 +33,14 @@ namespace boost::charconv {
 
 // Convert safe integer to character string
 template <safe_numbers::detail::library_type T>
+BOOST_SAFE_NUMBERS_HOST_DEVICE
 constexpr auto to_chars(char* first, char* last,
                         T value,
                         int base = 10) -> charconv::to_chars_result;
 
 // Convert character string to safe integer
 template <safe_numbers::detail::library_type T>
+BOOST_SAFE_NUMBERS_HOST_DEVICE
 constexpr auto from_chars(const char* first, const char* last,
                           T& value,
                           int base = 10) -> charconv::from_chars_result;
@@ -57,8 +59,10 @@ struct to_chars_result
     char* ptr;
     std::errc ec;
 
+    BOOST_SAFE_NUMBERS_HOST_DEVICE
     friend constexpr bool operator==(const to_chars_result& lhs,
                                      const to_chars_result& rhs) noexcept = default;
+    BOOST_SAFE_NUMBERS_HOST_DEVICE
     constexpr explicit operator bool() const noexcept { return ec == std::errc{}; }
 };
 
@@ -83,8 +87,11 @@ struct from_chars_result
     const char* ptr;
     std::errc ec;
 
+    BOOST_SAFE_NUMBERS_HOST_DEVICE
     friend constexpr bool operator==(const from_chars_result& lhs,
                                      const from_chars_result& rhs) noexcept = default;
+
+    BOOST_SAFE_NUMBERS_HOST_DEVICE
     constexpr explicit operator bool() const noexcept { return ec == std::errc{}; }
 };
 
@@ -103,6 +110,7 @@ struct from_chars_result
 [source,c++]
 ----
 template <safe_numbers::detail::library_type T>
+BOOST_SAFE_NUMBERS_HOST_DEVICE
 constexpr auto to_chars(char* first, char* last,
                         T value,
                         int base = 10) -> charconv::to_chars_result;
@@ -133,6 +141,7 @@ Returns `boost::charconv::to_chars_result` with:
 [source,c++]
 ----
 template <safe_numbers::detail::library_type T>
+BOOST_SAFE_NUMBERS_HOST_DEVICE
 constexpr auto from_chars(const char* first, const char* last,
                           T& value,
                           int base = 10) -> charconv::from_chars_result;

doc/modules/ROOT/pages/cuda.adoc

@@ -92,5 +92,5 @@ Device error on thread 256 at /home/runner/work/safe_numbers/boost-root/libs/saf
 
 The `device_error_context` will also attempt to `printf` the error into the terminal.
 This works when compiling with verbose mode `-V`.
-`printf` error messages will look the same as the message displayed by
+`printf` error messages will look the same as the message displayed by the thrown exception
 

doc/modules/ROOT/pages/integer_utilities.adoc

@@ -23,7 +23,7 @@ These operate on the non-bounded unsigned types (`u8`, `u16`, `u32`, `u64`, `u12
 [source,c++]
 ----
 template <non_bounded_unsigned_library_type T>
-[[nodiscard]] constexpr auto isqrt(const T val) -> T;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto isqrt(const T val) -> T;
 ----
 
 Returns the integer square root of `val`, i.e., the largest integer `r` such that `r * r \<= val`.
@@ -74,7 +74,7 @@ struct remove_trailing_zeros_return
 [source,c++]
 ----
 template <non_bounded_unsigned_library_type T>
-[[nodiscard]] constexpr auto remove_trailing_zeros(const T n);
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto remove_trailing_zeros(const T n);
 ----
 
 Removes all trailing decimal zeros from `n`.
@@ -141,7 +141,7 @@ Tests whether an unsigned integer value is an exact power of 10 (i.e., one of 1,
 [source,c++]
 ----
 template <non_bounded_unsigned_library_type T>
-[[nodiscard]] constexpr auto is_power_10(const T n) -> bool;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto is_power_10(const T n) -> bool;
 ----
 
 === Parameters
@@ -171,7 +171,7 @@ Returns the integer base-2 logarithm (floor of log~2~) of a value.
 [source,c++]
 ----
 template <non_bounded_unsigned_library_type T>
-[[nodiscard]] constexpr auto ilog2(const T n) -> int;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto ilog2(const T n) -> int;
 ----
 
 Computes `floor(log~2~(n))` using `bit_width(n) - 1`.
@@ -213,7 +213,7 @@ Uses an O(1) algorithm based on the most significant bit position to approximate
 [source,c++]
 ----
 template <non_bounded_unsigned_library_type T>
-[[nodiscard]] constexpr auto ilog10(const T n) -> int;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto ilog10(const T n) -> int;
 ----
 
 Computes `floor(log~10~(n))` using `num_digits(n) - 1`, where `num_digits` approximates the digit count via `log~10~(x) ~= log~2~(x) / log~2~(10)` and refines with at most two comparisons against a power-of-10 lookup table.
@@ -249,7 +249,7 @@ Returns the integer logarithm in an arbitrary base (floor of log~base~) of a val
 [source,c++]
 ----
 template <non_bounded_unsigned_library_type T>
-[[nodiscard]] constexpr auto ilog(const T n, const T base) -> int;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto ilog(const T n, const T base) -> int;
 ----
 
 Computes `floor(log~base~(n))` by repeated division.
@@ -294,7 +294,7 @@ Integer exponentiation using the exponentiation-by-squaring algorithm.
 [source,c++]
 ----
 template <non_bounded_unsigned_library_type T>
-[[nodiscard]] constexpr auto ipow(const T a, const T b) noexcept -> T;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto ipow(const T a, const T b) noexcept -> T;
 ----
 
 Computes `a` raised to the power `b` using exponentiation by squaring.
@@ -339,7 +339,7 @@ Tests whether an unsigned integer value is an exact power of 2 (i.e., has exactl
 [source,c++]
 ----
 template <non_bounded_unsigned_library_type T>
-[[nodiscard]] constexpr auto is_power_2(const T n) noexcept -> bool;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto is_power_2(const T n) noexcept -> bool;
 ----
 
 === Parameters
@@ -370,7 +370,7 @@ For unsigned types, naive subtraction `a - b` when `b > a` would underflow; `abs
 [source,c++]
 ----
 template <integral_library_type T>
-[[nodiscard]] constexpr auto abs_diff(const T a, const T b) noexcept -> T;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto abs_diff(const T a, const T b) noexcept -> T;
 ----
 
 Returns `|a - b|`, computed as `a - b` if `a >= b`, or `b - a` otherwise.
@@ -411,7 +411,7 @@ For unsigned types, this is equivalent to `(a + b - 1) / b` but computed without
 [source,c++]
 ----
 template <integral_library_type T>
-[[nodiscard]] constexpr auto div_ceil(const T a, const T b) noexcept -> T;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto div_ceil(const T a, const T b) noexcept -> T;
 ----
 
 Returns the ceiling of `a / b`.
@@ -454,7 +454,7 @@ This is useful for alignment calculations (e.g., aligning a size to a page bound
 [source,c++]
 ----
 template <integral_library_type T>
-[[nodiscard]] constexpr auto next_multiple_of(const T a, const T b) noexcept -> T;
+BOOST_SAFE_NUMBERS_HOST_DEVICE [[nodiscard]] constexpr auto next_multiple_of(const T a, const T b) noexcept -> T;
 ----
 
 Returns the smallest value `m` such that `m >= a` and `m % b == 0`.