Numerics

Concepts
concept	native::one_of
	N is one of the candidates
concept	native::not_one_of
	N is not one of the candidates

Classes
struct	native::fp16
struct	native::bf16

Typedefs
template<typename T>
using	native::int_t
template<typename T>
using	native::uint_t

Enumerations
enum class	native::CMPINT : std::size_t {   native::CMPINT::EQ = 0x0uz , native::CMPINT::LT = 0x1uz , native::CMPINT::LE = 0x2uz , native::CMPINT::FALSE = 0x3uz ,   native::CMPINT::NE = 0x4uz , native::CMPINT::NLT = 0x5uz , native::CMPINT::NLE = 0x6uz , native::CMPINT::TRUE = 0x7uz }
	Scalar integer-comparison predicates, encoded like the corresponding x86 immediates. More...
enum class	native::CMP : std::size_t {   native::CMP::EQ_OQ = 0x00uz , native::CMP::LT_OS = 0x01uz , native::CMP::LE_OS = 0x02uz , native::CMP::UNORD_Q = 0x03uz ,   native::CMP::NEQ_UQ = 0x04uz , native::CMP::NLT_US = 0x05uz , native::CMP::NLE_US = 0x06uz , native::CMP::ORD_Q = 0x07uz ,   native::CMP::EQ_UQ = 0x08uz , native::CMP::NGE_US = 0x09uz , native::CMP::NGT_US = 0x0Auz , native::CMP::FALSE_OQ = 0x0Buz ,   native::CMP::NEQ_OQ = 0x0Cuz , native::CMP::GE_OS = 0x0Duz , native::CMP::GT_OS = 0x0Euz , native::CMP::TRUE_UQ = 0x0Fuz ,   native::CMP::EQ_OS = 0x10uz , native::CMP::LT_OQ = 0x11uz , native::CMP::LE_OQ = 0x12uz , native::CMP::UNORD_S = 0x13uz ,   native::CMP::NEQ_US = 0x14uz , native::CMP::NLT_UQ = 0x15uz , native::CMP::NLE_UQ = 0x16uz , native::CMP::ORD_S = 0x17uz ,   native::CMP::EQ_US = 0x18uz , native::CMP::NGE_UQ = 0x19uz , native::CMP::NGT_UQ = 0x1Auz , native::CMP::FALSE_OS = 0x1Buz ,   native::CMP::NEQ_OS = 0x1Cuz , native::CMP::GE_OQ = 0x1Duz , native::CMP::GT_OQ = 0x1Euz , native::CMP::TRUE_US = 0x1Fuz }
	Scalar floating-comparison predicates, including ordered and unordered truth tables. More...

Functions
template<typename T>
constexpr bool	native::cmp_unord (T a, T b) noexcept(noexcept(isnan(a)||isnan(b)))
	Return true if either argument is NaN.
template<typename T>
constexpr bool	native::cmp_ord (T a, T b) noexcept(noexcept(!isnan(a) &&!isnan(b)))
	Return true if neither argument is NaN.
template<one_of_t< float, double > T>
constexpr T	native::scalef (T x, T y) noexcept
template<CMPINT imm8, typename T> requires (one_of_t<T, uint8_t, int8_t, uint16_t, int16_t, uint32_t, int32_t, uint64_t, int64_t> && (std::size_t(imm8) < 8uz))
constexpr bool	native::cmpint (T a, T b) noexcept
template<CMP imm8, typename T> requires (one_of_t<T, float, double> && (std::size_t(imm8) < 32uz))
constexpr bool	native::cmp (T a, T b) noexcept

Variables
constexpr std::size_t	native::max_fp_comparison_predicate

Detailed Description

Typedef Documentation

int_t

template<typename T>

using native::int_t

export

returns a signed integer type of the same size as T suitable for std::bit_cast

Definition at line 580 of file native.numerics.ccm.

uint_t

template<typename T>

using native::uint_t

export

returns an unsigned integer type of the same size as T suitable for std::bit_cast

Definition at line 586 of file native.numerics.ccm.

Enumeration Type Documentation

CMP

enum class native::CMP : std::size_t

exportstrong

Scalar floating-comparison predicates, including ordered and unordered truth tables.

Ordered predicates require neither operand to be NaN; unordered predicates include NaNs as indicated by their truth table. The Q/S suffixes retain the x86 encoding names, not a promise about hardware exception flags.

Enumerator
EQ_OQ	0x00uz	Equal (ordered, nonsignaling).
LT_OS	0x01uz	Less-than (ordered, signaling).
LE_OS	0x02uz	Less-than-or-equal (ordered, signaling).
UNORD_Q	0x03uz	Unordered (nonsignaling).
NEQ_UQ	0x04uz	Not-equal (unordered, nonsignaling).
NLT_US	0x05uz	Not-less-than (unordered, signaling).
NLE_US	0x06uz	Not-less-than-or-equal (unordered, signaling).
ORD_Q	0x07uz	Ordered (nonsignaling).
EQ_UQ	0x08uz	Equal (unordered, nonsignaling).
NGE_US	0x09uz	Not-greater-than-or-equal (unordered, signaling).
NGT_US	0x0Auz	Not-greater-than (unordered, signaling).
FALSE_OQ	0x0Buz	False (ordered, nonsignaling).
NEQ_OQ	0x0Cuz	Not-equal (ordered, nonsignaling).
GE_OS	0x0Duz	Greater-than-or-equal (ordered, signaling).
GT_OS	0x0Euz	Greater-than (ordered, signaling).
TRUE_UQ	0x0Fuz	True (unordered, nonsignaling).
EQ_OS	0x10uz	Equal (ordered, signaling).
LT_OQ	0x11uz	Less-than (ordered, nonsignaling).
LE_OQ	0x12uz	Less-than-or-equal (ordered, nonsignaling).
UNORD_S	0x13uz	Unordered (signaling).
NEQ_US	0x14uz	Not-equal (unordered, signaling).
NLT_UQ	0x15uz	Not-less-than (unordered, nonsignaling).
NLE_UQ	0x16uz	Not-less-than-or-equal (unordered, nonsignaling).
ORD_S	0x17uz	Ordered (signaling).
EQ_US	0x18uz	Equal (unordered, signaling).
NGE_UQ	0x19uz	Not-greater-than-or-equal (unordered, nonsignaling).
NGT_UQ	0x1Auz	Not-greater-than (unordered, nonsignaling).
FALSE_OS	0x1Buz	False (ordered, signaling).
NEQ_OS	0x1Cuz	Not-equal (ordered, signaling).
GE_OQ	0x1Duz	Greater-than-or-equal (ordered, nonsignaling).
GT_OQ	0x1Euz	Greater-than (ordered, nonsignaling).
TRUE_US	0x1Fuz	True (unordered, signaling).

Definition at line 734 of file native.numerics.ccm.

CMPINT

enum class native::CMPINT : std::size_t

exportstrong

Scalar integer-comparison predicates, encoded like the corresponding x86 immediates.

Select a truth table for cmpint(); comparison uses the signedness of its operands.

Enumerator
EQ	0x0uz	==
LT	0x1uz	<
LE	0x2uz	<=
FALSE	0x3uz	always false
NE	0x4uz	!=
NLT	0x5uz	>=
NLE	0x6uz	>
TRUE	0x7uz	always true

Definition at line 689 of file native.numerics.ccm.

Function Documentation

cmp()

template<CMP imm8, typename T>
requires (one_of_t<T, float, double> && (std::size_t(imm8) < 32uz))

bool native::cmp ( T a, T b )

inlineconstexprexportnoexcept

Evaluate the selected ordered/unordered truth table on two float or double values. The Q/S immediate variants have the same truth values; this helper does not emulate the x86 signaling versus nonsignaling exception behavior.

Definition at line 775 of file native.numerics.ccm.

Here is the caller graph for this function:

cmp_ord()

template<typename T>

bool native::cmp_ord ( T a, T b )

inlineconstexprexportnoexcept

Return true if neither argument is NaN.

This corresponds to checking if the two arguments are "ordered" with respect to the IEEE floating point comparison predicates

Definition at line 615 of file native.numerics.ccm.

Here is the caller graph for this function:

cmp_unord()

template<typename T>

bool native::cmp_unord ( T a, T b )

inlineconstexprexportnoexcept

Return true if either argument is NaN.

This corresponds to checking if the two arguments are "unordered" with respect to the IEEE floating point comparison predicates

Definition at line 598 of file native.numerics.ccm.

Here is the caller graph for this function:

cmpint()

template<CMPINT imm8, typename T>
requires (one_of_t<T, uint8_t, int8_t, uint16_t, int16_t, uint32_t, int32_t, uint64_t, int64_t> && (std::size_t(imm8) < 8uz))

bool native::cmpint ( T a, T b )

inlineconstexprexportnoexcept

Evaluate the selected integer truth table on two values of the same integer type. Signed and unsigned ordering follow T; the immediate selects one of eight predicates.

Definition at line 705 of file native.numerics.ccm.

Here is the caller graph for this function:

scalef()

template<one_of_t< float, double > T>

T native::scalef ( T x, T y )

inlineconstexprexportnoexcept

Scale x by 2 raised to the exponent obtained by truncating y toward zero.

Precondition

The truncated value of y is representable as int. Constant evaluation uses round-to-nearest, ties-to-even with gradual underflow. Runtime uses std::scalbn and follows the active rounding and denormal modes; it agrees with constant evaluation under the same floating-point assumptions. Signed zeros and overflow signs are preserved; NaN payloads are not promised. This scalar truncation contract differs from the vector scaleb API.

Definition at line 634 of file native.numerics.ccm.

Here is the caller graph for this function:

Variable Documentation

max_fp_comparison_predicate

std::size_t native::max_fp_comparison_predicate

inlineconstexprexport

Number of floating comparison immediates selected by the compile-time AVX-512 feature macro. This constant does not query the running CPU or admit any instruction set.

Definition at line 721 of file native.numerics.ccm.