C++ · Scalar Types

Scalar Types

#include <numkong/types.hpp>#include <numkong/cast.hpp> 8.0.0-rc.1 · b1f866ab

The scalar wrappers in include/numkong/types.hpp are storage-first types. They encode raw layout, default output types, and the kernel function pointer signatures for each family.

TypeLayoutBytesRangeInfNaN
nk_f16_t1+5+102±65504yesyes
nk_bf16_t1+8+72±3.4×10³⁸yesyes
nk_e4m3_t1+4+31±448noyes
nk_e5m2_t1+5+21±57344yesyes
nk_e2m3_t1+2+31±7.5nono
nk_e3m2_t1+3+21±28nono
nk_e2m1x2_t2 × (1+2+1)1±6 per nibblenono
nk_u1x8_t8 packed bits10 or 1 per bitn/an/a
nk_u4x2_t2 × 4-bit10…15 per nibblen/an/a
nk_i4x2_t2 × 4-bit signed1−8…7 per nibblen/an/a

The layout column shows sign, exponent, and mantissa bit counts for floating-point types. For nk_f16_t, 1+5+10 means one sign bit, five exponent bits, and ten mantissa bits, totaling 16 bits stored in 2 bytes. For nk_bf16_t, the wider exponent field (8 bits) gives the same dynamic range as IEEE 754 single precision but with reduced mantissa precision. The Float8 types nk_e4m3_t and nk_e5m2_t follow the OFP8 specification. The narrower nk_e2m3_t and nk_e3m2_t types are MX-compatible micro-floats, and nk_e2m1x2_t packs two FP4 values of the MXFP4 and NVFP4 formats. Sub-byte types nk_e2m1x2_t, nk_u1x8_t, nk_u4x2_t, and nk_i4x2_t pack multiple logical values into a single byte. Element 0 sits in the high nibble or the most significant bit, and every dimension count must be a multiple of the values per byte.

Default promotions are encoded on the type. For example, f32_t::dot_result_t is wider than f32_t. i8_t::dot_result_t is i32_t. u1x8_t::dot_result_t is u32_t.

The higher-level templates use result_type_ = typename in_type_::dot_result_t and similar defaults. The fast typed overloads are constrained so that overriding the result type away from the native policy can disable the specialized path and fall back to the more generic one.

When __cpp_lib_format >= 202110L for the C++23 <format> header support, all NumKong scalar types provide std::formatter specializations with similar format specs to the traditional float. For the Float16 type, the output for nk::f16_t::from_f32(3.14f) will look like:

Format specOutput exampleDescription
{}3.140625Clean float value
{:#}3.140625 [0x4248]Annotated with hex bits
{:.2f}3.14Precision forwarded to float formatter
{:x}4248Raw hex bits
{:#x}0x4248Hex with prefix
{:X}4248Uppercase hex
{:b}0100001001001000Binary bits
{:#b}0b0100001001001000Binary with prefix

API

Types

f16c_t32-bit half-precision complex number, composed of two f16_t.
types.hpp:2122 ↗
struct f16c_t

Layout: {real: f16_t, imag: f16_t}. Kernel outputs widen to f32c; math runs via f32 upcast.

bf16c_t32-bit BFloat16 complex number, composed of two bf16_t.
types.hpp:2218 ↗
struct bf16c_t

Layout: {real: bf16_t, imag: bf16_t}. Kernel outputs widen to f32c; math runs via f32 upcast.

e2m3_tFloat6 E2M3FN: 1 sign + 2 exponent, bias=1, + 3 mantissa bits, with 2 bits of padding.
types.hpp:2826 ↗
struct e2m3_t

Range: [-7.5, +7.5], stored byte-aligned as 0b00SEEMMM with the upper 2 bits as padding. Format: OCP Microscaling FN with no Inf/NaN, so all 64 bit patterns are valid numbers. Codes: 48 normal, 14 subnormal with exp=0 and mant!=0, and 2 zeros, ±0. Unit range: only 18 of 64 values, 28.1%, fall in [-1, +1] — 72% of codes represent |x| > 1. Subnormals: span [±0.125, ±0.875] as 0.mmm × 2^(1-bias). Dot products: exact in integers, as every value × 16 is an integer in [-120, +120], fitting i16 products and i32 sums without rounding. Promotion: lossless to E4M3, rebiasing exponents by +6 for normals or normalizing subnormals.

See also

e3m2_tFloat6 E3M2FN: 1 sign + 3 exponent, bias=3, + 2 mantissa bits, with 2 bits of padding.
types.hpp:3034 ↗
struct e3m2_t

Range: [-28, +28], stored byte-aligned as 0b00SEEEMM with the upper 2 bits as padding. Format: OCP Microscaling FN with no Inf/NaN, so all 64 bit patterns are valid numbers. Codes: 56 normal, 6 subnormal with exp=0 and mant!=0, and 2 zeros, ±0. Unit range: 26 of 64 values, 40.6%, fall in [-1, +1]. Subnormals: span [±0.0625, ±0.1875] as 0.mm × 2^(1-bias). Dot products: exact in integers, as every value × 4 is an integer in [-28, +28], fitting i16 products and i32 sums without rounding. Promotion: lossless to E5M2, rebiasing exponents by +12 for normals or normalizing subnormals.

See also

ue8m0_tUnsigned 8-bit power-of-two scale, OCP MX v1.0, biased exponent only: no sign or mantissa.
types.hpp:3234 ↗
struct ue8m0_t

Encoding: v=0 → zero; v=0xFF → NaN-block sentinel; otherwise 2^(v − 127). Used as the per-block scale byte for MXFP4 / MXFP6 / MXFP8 / MXINT8.

ue4m3_tUnsigned 8-bit E4M3 scale, NVFP4: sign-bit forced to 0, otherwise identical to E4M3.
types.hpp:3301 ↗
struct ue4m3_t

Range: [0, +448]. Paired with an f32 tensor_scale multiplier in block-scaled NVFP4, block=16.

f118c_tExtended-precision complex number using f118_t components.
types.hpp:4303 ↗
struct f118c_t

Provides ~91 bits of precision for complex arithmetic, useful for:

  • High-precision reference computations in tests
  • Complex dot products and correlations
  • Avoiding Boost.Multiprecision dependency

Format: real + i*imag where real and imag are f118_t (double-double).

i8_tSigned 8-bit integer wrapper.
types.hpp:4387 ↗
struct i8_t

Range: [−128, +127]. Exact integer arithmetic, wrapping on overflow. Dot products accumulated into i32 via VNNI / VPDPBUSD / SDOT.

u8_tUnsigned 8-bit integer wrapper.
types.hpp:4572 ↗
struct u8_t

Range: [0, 255]. Exact integer arithmetic, wrapping on overflow. Dot products accumulated into i32 via VNNI / VPDPBUSD / UDOT.

i32_tSigned 32-bit integer wrapper.
types.hpp:4737 ↗
struct i32_t

Range: [−2³¹, +2³¹−1]. Exact integer arithmetic, wrapping on overflow.

u32_tUnsigned 32-bit integer wrapper.
types.hpp:4885 ↗
struct u32_t

Range: [0, 2³²−1]. Exact integer arithmetic, wrapping on overflow.

i64_tSigned 64-bit integer wrapper.
types.hpp:5020 ↗
struct i64_t

Range: [−2⁶³, +2⁶³−1]. Exact integer arithmetic, wrapping on overflow.

u64_tUnsigned 64-bit integer wrapper.
types.hpp:5169 ↗
struct u64_t

Range: [0, 2⁶⁴−1]. Exact integer arithmetic, wrapping on overflow.

i16_tSigned 16-bit integer wrapper.
types.hpp:5302 ↗
struct i16_t

Range: [−32768, +32767]. Exact integer arithmetic, wrapping on overflow.

u16_tUnsigned 16-bit integer wrapper.
types.hpp:5445 ↗
struct u16_t

Range: [0, 65535]. Exact integer arithmetic, wrapping on overflow.

u1x8_tPacked 8-bit bit-vector, 8 booleans in one byte.
types.hpp:5830 ↗
struct u1x8_t

Layout: 8 bits packed into one byte, MSB = dimension 0. Used for Hamming distance and Jaccard similarity via popcount.

i4x2_tPacked 4-bit signed integer pair, 2 × i4 in one byte.
types.hpp:5955 ↗
struct i4x2_t

Layout: [high nibble : low nibble]. Elements sign-extended to i8 for arithmetic. Range per element: [−8, +7].

u4x2_tPacked 4-bit unsigned integer pair, 2 × u4 in one byte.
types.hpp:6085 ↗
struct u4x2_t

Layout: [high nibble : low nibble]. Elements zero-extended to u8 for arithmetic. Range per element: [0, 15].

nvfp4_tNVIDIA NVFP4 block value: 16 E2M1 nibbles + 1 UE4M3 scale byte per value.
types.hpp:6288 ↗
struct nvfp4_t

Per-tensor f32 tensor_scale multiplier lives on the enclosing tensor, not per block, and is passed explicitly to decode_to and encode_from. sizeof(nvfp4_t) == 9, and one value represents 16 logical elements — nk::tensor<nvfp4_t>(n) allocates ⌈n/16⌉ × 9 bytes.

mxfp4_tOCP MXFP4 block value: 32 E2M1 nibbles + 1 UE8M0 pow-2 scale byte per value.
types.hpp:6349 ↗
struct mxfp4_t
mxfp6e2m3_tOCP MXFP6 E2M3 block value: 32 E2M3 elements + 1 UE8M0 scale byte per value.
types.hpp:6408 ↗
struct mxfp6e2m3_t
mxfp6e3m2_tOCP MXFP6 E3M2 block value: 32 E3M2 elements + 1 UE8M0 scale byte per value.
types.hpp:6459 ↗
struct mxfp6e3m2_t
mxfp8e4m3_tOCP MXFP8 E4M3 block value: 32 E4M3 elements + 1 UE8M0 scale byte per value.
types.hpp:6510 ↗
struct mxfp8e4m3_t
mxfp8e5m2_tOCP MXFP8 E5M2 block value: 32 E5M2 elements + 1 UE8M0 scale byte per value.
types.hpp:6569 ↗
struct mxfp8e5m2_t
mxint8_tOCP MXINT8 block value: 32 signed i8 elements + 1 UE8M0 scale byte per value.
types.hpp:6628 ↗
struct mxint8_t
numeric_limits_ofWhat std::numeric_limits reports for a real NumKong scalar: every member of the primary template, read off the scalar's own traits.
types.hpp:6739 ↗
template <typename scalar_type_>
struct numeric_limits_of

Complex and packed scalars get none. The decimal bounds take 30103 / 100000 for log10(2), truncating toward zero as the standard rounds them.

Fields

is_specializedbool
is_signedbool
is_integerbool
is_exactbool
has_infinitybool
has_quiet_NaNbool
has_signaling_NaNbool
round_stylestd::float_round_style
is_iec559bool
is_boundedbool
is_modulobool
digitsint
digits10int
max_digits10int
radixint
min_exponentint
min_exponent10int
max_exponentint
max_exponent10int
trapsbool
tinyness_beforebool
word_typeExtract the word type from a value type.
types.hpp:6886 ↗
template <typename value_type_,
          typename = void>
struct word_type

For complex types, f32c_t, f64c_t, and so on, returns the component type, f32_t, f64_t. For all other types, returns the type itself.

Structbf16_t16-bit brain floating-point wrapper with f32-compatible exponent range.6
types.hpp:1832 ↗
struct bf16_t

Layout: sign(1) + exponent(8) + mantissa(7), bias=127. Same exponent range as f32, ~38 orders of magnitude, but only 7 mantissa bits versus 23. Range: [-3.39e38, +3.39e38], epsilon at 1.0: ~7.81e-3, subnormal min: ~9.18e-41. 32,514 of 65,280 finite values, 49.8%, fall in [-1, +1]. All math runs via f32 upcast/downcast.

bf16_t
types.hpp:1938 ↗
constexpr bf16_t() noexcept;
bf16_t(float v) noexcept;
bf16_t(double v) noexcept;
template <std::integral integral_type_>
bf16_t(integral_type_ v) noexcept;
*× 8
types.hpp:2010 ↗
bool operator==(bf16_t o) const noexcept;
Symbolo
operator==bf16_t
operator==e4m3_t
operator==e5m2_t
operator==f16_t
operator==f32_t
operator==f32c_t
operator==f64_t
operator==f64c_t
orderTotal ordering via nk_bf16_order_best.
types.hpp:2030 ↗
int order(bf16_t o) const noexcept;
total_*_cmp× 7Alias for order(), Rust-style.
types.hpp:2033 ↗
int total__cmp(bf16_t o) const noexcept;
Symbolo
total_cmpbf16_t
total_cmpe4m3_t
total_cmpe5m2_t
total_cmpf118_t
total_cmpf16_t
total_cmpf32_t
total_cmpf64_t
saturating_*_add× 6Saturating addition: clamps to finite range on overflow.
types.hpp:2101 ↗
bf16_t saturating__add(bf16_t o) const noexcept;
Symbolo
saturating_addbf16_t
saturating_adde4m3_t
saturating_adde5m2_t
saturating_addf16_t
saturating_addf32_t
saturating_addf64_t
saturating_*_sub× 6Saturating subtraction: clamps to finite range on overflow.
types.hpp:2109 ↗
bf16_t saturating__sub(bf16_t o) const noexcept;
Symbolo
saturating_subbf16_t
saturating_sube4m3_t
saturating_sube5m2_t
saturating_subf16_t
saturating_subf32_t
saturating_subf64_t
Structe2m1x2_tPacked 4-bit E2M1 micro-float pair, 2 × e2m1 in one byte.4
types.hpp:6206 ↗
struct e2m1x2_t

Layout: [high nibble : low nibble]. Each nibble: sign(1) + exponent(2) + mantissa(1), bias=1. Eight magnitudes {0, 0.5, 1.0, 1.5, 2.0, 3.0, 4.0, 6.0} × 2 signs. Range: ±6.0, no Inf/NaN. Element of MXFP4, block=32, UE8M0 scales, and NVFP4, block=16, UE4M3 scales.

nibble_to_f32Decodes one nibble, low 4 bits, of sign, 2 exponent and 1 mantissa bits.
types.hpp:6267 ↗
float nibble_to_f32(nk_u8_t nibble) noexcept;
e2m1x2_t
types.hpp:6255 ↗
constexpr e2m1x2_t() noexcept;
constexpr e2m1x2_t(raw_t v) noexcept;
firstThe first element, stored in the high nibble.
types.hpp:6273 ↗
float first() const noexcept;
secondThe second element, stored in the low nibble.
types.hpp:6276 ↗
float second() const noexcept;
Structe4m3_tFP8 E4M3 wrapper: 8-bit float with 4-bit exponent, 3-bit mantissa.2
types.hpp:2318 ↗
struct e4m3_t

Layout: sign(1) + exponent(4) + mantissa(3), bias=7. Range: ±448, epsilon at 1.0: 0.125, subnormal min: ~1.95e-3. No infinity representation; only 2 NaN codes, 0x7F and 0xFF. 114 of 254 finite values, 44.9%, fall in [-1, +1]. Dot products can be computed exactly via integer decomposition: mantissa products (8+m_a)*(8+m_b) are accumulated into 29 bins indexed by exponent sum (e_a+e_b). All math runs via f32 upcast/downcast.

e4m3_t
types.hpp:2409 ↗
constexpr e4m3_t() noexcept;
e4m3_t(float v) noexcept;
e4m3_t(double v) noexcept;
template <std::integral integral_type_>
e4m3_t(integral_type_ v) noexcept;
orderTotal ordering via nk_e4m3_order_best.
types.hpp:2499 ↗
int order(e4m3_t o) const noexcept;
Structe5m2_tFP8 E5M2 wrapper: 8-bit float with 5-bit exponent, 2-bit mantissa.2
types.hpp:2577 ↗
struct e5m2_t

Layout: sign(1) + exponent(5) + mantissa(2), bias=15. Same exponent range as f16, but only 2 mantissa bits. Range: ±57344, epsilon at 1.0: 0.25, subnormal min: ~1.53e-5. Supports infinity and NaN, IEEE-like, exp=31. 122 of 248 finite values, 49.2%, fall in [-1, +1]. The huge dynamic range, ratio ~3.7e9, makes f32 accumulation vulnerable to catastrophic cancellation when large and small products coexist. Dot products can be computed exactly via integer decomposition: mantissa products (4+m_a)*(4+m_b) accumulate into 63 bins indexed by exponent sum (e_a+e_b). All math runs via f32 upcast/downcast.

e5m2_t
types.hpp:2653 ↗
constexpr e5m2_t() noexcept;
e5m2_t(float v) noexcept;
e5m2_t(double v) noexcept;
template <std::integral integral_type_>
e5m2_t(integral_type_ v) noexcept;
orderTotal ordering via nk_e5m2_order_best.
types.hpp:2745 ↗
int order(e5m2_t o) const noexcept;
Structf118_tHardware-friendly "double-double" arithmetic with ~106-bit mantissa.68
types.hpp:3404 ↗
struct f118_t

Uses Knuth two-sum + FMA for error-free transformations. Provides ~106 bits of mantissa precision using two doubles, at roughly 11x the cost of native double arithmetic, being roughly 8x more efficient than Boost.Multiprecision types.

Speed comparison, relative to double:

TypeSpeedMantissaNotes
double1.0x53-bitHardware
long double1.5x64-bitx87 hardware
f118_t11x~106-bitSoftware, FMA-based
__float12888x113-bitlibquadmath
boost::float12891x113-bitWrapper around __float128
boost::cpp_bin_float_quad200x113-bitPure C++ (slowest!)
boost::cpp_bin_float_50237x~166-bit50 decimal digits

Measured precision vs __float128, 1M random samples, full double-double input:

OperationMax Rel ErrPrecisionNotes
+, -~2e-28~91 bitsTwo-Sum algorithm
*2.5e-32~104 bitsFMA-based Two-Prod
/2.4e-32~105 bits3-iteration Newton-Raphson
sqrt6.8e-32~103 bitsNewton-Raphson refinement
cbrt2.4e-31~101 bitsNewton-Raphson refinement
exp2.2e-31~101 bitsArgument reduction + Taylor
log2.3e-32~105 bitsNewton-Raphson refinement
sin6.8e-32~103 bits|x|<10, quad-double π/2 + Taylor
cos7.8e-32~103 bits|x|<10, quad-double π/2 + Taylor
tan9.4e-32~103 bitssin/cos ratio
sinh1.4e-31~102 bitsTaylor for |x|<1, optimized exp
cosh1.5e-31~102 bitsVia exp()
tanh1.0e-31~102 bitssinh/cosh ratio
pow1.0e-30~99 bitsVia exp(y·log(x))
f118_tDefault constructor, initializes to zero.
types.hpp:3439 ↗
constexpr f118_t() noexcept;
constexpr f118_t(double h, double l) noexcept;

Construct from high and low double components.

constexpr f118_t(double v) noexcept;

Construct from single double.

constexpr f118_t(int v) noexcept;

Construct from int, resolving ambiguity with the double,double constructor.

constexpr f118_t(unsigned int v) noexcept;

Construct from unsigned int.

constexpr f118_t(std::int64_t v) noexcept;

Construct from int64_t, capturing precision beyond double's 53-bit mantissa.

constexpr f118_t(std::uint64_t v) noexcept;

Construct from uint64_t, capturing precision beyond double's 53-bit mantissa.

operator+Addition with ~91 bits precision, max rel err 2.3e-28 vs __float128.
types.hpp:3474 ↗
f118_t operator+(f118_t const & o) const noexcept;
operator+=In-place addition.
types.hpp:3481 ↗
f118_t & operator+=(f118_t const & o) noexcept;
operator-=In-place subtraction.
types.hpp:3484 ↗
f118_t & operator-=(f118_t const & o) noexcept;
operator*=In-place multiplication.
types.hpp:3487 ↗
f118_t & operator*=(f118_t const & o) noexcept;
operator/=In-place division.
types.hpp:3490 ↗
f118_t & operator/=(f118_t const & o) noexcept;
operator-Subtraction with ~93 bits precision, max rel err 7.4e-29 vs __float128.
types.hpp:3493 ↗
f118_t operator-(f118_t const & o) const noexcept;
f118_t operator-() const noexcept;

Unary minus.

operator*Multiplication with ~104 bits precision, max rel err 2.5e-32 vs __float128, ~105 bits vs Boost.
types.hpp:3501 ↗
f118_t operator*(f118_t const & o) const noexcept;
operator/Division with ~105 bits precision, max rel err 2.4e-32 vs __float128, ~106 bits vs Boost.
types.hpp:3511 ↗
f118_t operator/(f118_t const & o) const noexcept;
saturating_addSaturating addition - clamps to max finite value instead of infinity.
types.hpp:3523 ↗
f118_t saturating_add(f118_t o) const noexcept;
saturating_subSaturating subtraction - clamps to max finite value instead of infinity.
types.hpp:3530 ↗
f118_t saturating_sub(f118_t o) const noexcept;
saturating_mulSaturating multiplication - forwards to operator*, no saturation semantics for double-double.
types.hpp:3538 ↗
f118_t saturating_mul(f118_t o) const noexcept;
operator==Exact equality, both high_ and low_ must match.
types.hpp:3541 ↗
bool operator==(f118_t const & o) const noexcept;
operator!=Exact inequality.
types.hpp:3544 ↗
bool operator!=(f118_t const & o) const noexcept;
operator<Less-than comparison, lexicographic on high_, then low_.
types.hpp:3547 ↗
bool operator<(f118_t const & o) const noexcept;
operator>Greater-than comparison.
types.hpp:3552 ↗
bool operator>(f118_t const & o) const noexcept;
operator<=Less-than-or-equal comparison.
types.hpp:3555 ↗
bool operator<=(f118_t const & o) const noexcept;
operator>=Greater-than-or-equal comparison.
types.hpp:3558 ↗
bool operator>=(f118_t const & o) const noexcept;
operator doubleConvert to double, losing ~53 bits of precision.
types.hpp:3561 ↗
constexpr operator double() const noexcept;
*× 3Convert to any numeric type.
types.hpp:3569 ↗
template <typename target_type_>
target_type_ to() const noexcept;
sqrtSquare root with ~103 bits precision, max rel err 6.8e-32 vs __float128, roughly ~103 bits vs Boost.
types.hpp:3575 ↗
f118_t sqrt() const noexcept;
rsqrtReciprocal square root, 1/sqrt.
types.hpp:3585 ↗
f118_t rsqrt() const noexcept;
expExponential with ~101 bits precision, max rel err 2.2e-31 vs __float128, roughly ~102 bits vs Boost.
types.hpp:3589 ↗
f118_t exp() const noexcept;
logNatural logarithm with ~105 bits precision, max rel err 2.2e-32 vs __float128, ~105 bits vs Boost.
types.hpp:3614 ↗
f118_t log() const noexcept;
sinSine with ~103 bits precision (max rel err: 6.8e-32 vs __float128) for |x| < 10.
types.hpp:3653 ↗
f118_t sin() const noexcept;

Notes

Quad-double π/2 with staged error-free reduction; precision degrades for |x| > 1000.

cosCosine with ~103 bits precision (max rel err: 7.8e-32 vs __float128) for |x| < 10.
types.hpp:3674 ↗
f118_t cos() const noexcept;

Notes

Quad-double π/2 with staged error-free reduction; precision degrades for |x| > 1000.

absAbsolute value.
types.hpp:3692 ↗
f118_t abs() const noexcept;
is_nanReturns true if either component is NaN.
types.hpp:3698 ↗
bool is_nan() const noexcept;
is_infiniteReturns true if the value is positive or negative infinity.
types.hpp:3701 ↗
bool is_infinite() const noexcept;
is_finiteReturns true if the value is neither infinite nor NaN.
types.hpp:3708 ↗
bool is_finite() const noexcept;
is_sign_positiveReturns true if the sign bit is positive, +0 included.
types.hpp:3714 ↗
bool is_sign_positive() const noexcept;
is_sign_negativeReturns true if the sign bit is negative, -0 included.
types.hpp:3717 ↗
bool is_sign_negative() const noexcept;
floorLargest integer less than or equal to self.
types.hpp:3720 ↗
f118_t floor() const noexcept;
ceilSmallest integer greater than or equal to self.
types.hpp:3727 ↗
f118_t ceil() const noexcept;
roundNearest integer, rounding half away from zero.
types.hpp:3734 ↗
f118_t round() const noexcept;
truncInteger part, truncated toward zero.
types.hpp:3741 ↗
f118_t trunc() const noexcept;
fractFractional part, self - trunc(self).
types.hpp:3748 ↗
f118_t fract() const noexcept;
minReturns the minimum of self and other.
types.hpp:3751 ↗
f118_t min(f118_t o) const noexcept;
maxReturns the maximum of self and other.
types.hpp:3754 ↗
f118_t max(f118_t o) const noexcept;
clampClamps self to the range [lower, upper].
types.hpp:3757 ↗
f118_t clamp(f118_t lower, f118_t upper) const noexcept;
orderTotal ordering: -NaN < -Inf < ... < -0 < +0 < ... < +Inf < +NaN.
types.hpp:3760 ↗
int order(f118_t o) const noexcept;

Returns -1, 0, or 1.

signumReturns -1, 0, or 1 based on sign.
types.hpp:3782 ↗
f118_t signum() const noexcept;
copysignReturns value with magnitude of self and sign of sign.
types.hpp:3790 ↗
f118_t copysign(f118_t sign) const noexcept;
recipReturns 1 / self.
types.hpp:3798 ↗
f118_t recip() const noexcept;
powfxʸ with ~99 bits precision, max rel err 1.0e-30 vs __float128.
types.hpp:3801 ↗
f118_t powf(f118_t y) const noexcept;
powiInteger power via binary exponentiation.
types.hpp:3808 ↗
f118_t powi(int exponent) const noexcept;
cbrtCube root with ~101 bits precision, max rel err 2.4e-31 vs __float128.
types.hpp:3822 ↗
f118_t cbrt() const noexcept;
exp22ˣ.
types.hpp:3832 ↗
f118_t exp2() const noexcept;
log2Base-2 logarithm.
types.hpp:3839 ↗
f118_t log2() const noexcept;
log10Base-10 logarithm.
types.hpp:3846 ↗
f118_t log10() const noexcept;
exp_m1eˣ − 1, accurate for small x.
types.hpp:3853 ↗
f118_t exp_m1() const noexcept;
ln_1pln(1 + x), accurate for small x.
types.hpp:3868 ↗
f118_t ln_1p() const noexcept;
tanTangent with ~103 bits precision, max rel err 9.4e-32 vs __float128.
types.hpp:3883 ↗
f118_t tan() const noexcept;
asinArcsine, inverse sine.
types.hpp:3886 ↗
f118_t asin() const noexcept;
acosArccosine, inverse cosine.
types.hpp:3894 ↗
f118_t acos() const noexcept;
atanArctangent, inverse tangent.
types.hpp:3902 ↗
f118_t atan() const noexcept;
atan2Four-quadrant arctangent: atan2(y, x) where this = y.
types.hpp:3958 ↗
f118_t atan2(f118_t x) const noexcept;
sin_cosComputes both sin(x) and cos(x), returning them in an array.
types.hpp:3975 ↗
void sin_cos(f118_t & out_sin, f118_t & out_cos) const noexcept;
sinhHyperbolic sine with ~102 bits precision, max rel err ~1.4e-31 vs __float128.
types.hpp:3981 ↗
f118_t sinh() const noexcept;
coshHyperbolic cosine with ~102 bits precision, max rel err 1.5e-31 vs __float128.
types.hpp:4004 ↗
f118_t cosh() const noexcept;
tanhHyperbolic tangent with ~102 bits precision, max rel err ~1e-31 vs __float128.
types.hpp:4010 ↗
f118_t tanh() const noexcept;
asinhInverse hyperbolic sine: ln(x + √(x² + 1)).
types.hpp:4017 ↗
f118_t asinh() const noexcept;
acoshInverse hyperbolic cosine: ln(x + √(x² − 1)).
types.hpp:4022 ↗
f118_t acosh() const noexcept;
atanhInverse hyperbolic tangent: ½ · ln((1+x)/(1−x)).
types.hpp:4028 ↗
f118_t atanh() const noexcept;
hypot√(x² + y²) without overflow.
types.hpp:4034 ↗
f118_t hypot(f118_t y) const noexcept;
fmaFused multiply-add: self * a + b.
types.hpp:4043 ↗
f118_t fma(f118_t a, f118_t b) const noexcept;
to_radiansConvert degrees to radians.
types.hpp:4046 ↗
f118_t to_radians() const noexcept;
to_degreesConvert radians to degrees.
types.hpp:4053 ↗
f118_t to_degrees() const noexcept;
Structf16_t16-bit half-precision IEEE 754 floating-point wrapper.2
types.hpp:1544 ↗
struct f16_t

Layout: sign(1) + exponent(5) + mantissa(10), bias=15. Range: [-65504, +65504], epsilon at 1.0: ~9.77e-4, subnormal min: ~5.96e-8. 30,722 of 63,488 finite values, 48.4%, fall in [-1, +1]. All math runs via f32 upcast/downcast.

f16_t
types.hpp:1649 ↗
constexpr f16_t() noexcept;
f16_t(float v) noexcept;
f16_t(double v) noexcept;
template <std::integral integral_type_>
f16_t(integral_type_ v) noexcept;
orderTotal ordering via nk_f16_order_best.
types.hpp:1743 ↗
int order(f16_t o) const noexcept;
Structf32_t32-bit single-precision IEEE 754 floating-point wrapper.3
types.hpp:463 ↗
struct f32_t

Layout: sign(1) + exponent(8) + mantissa(23), bias=127. Range: ±3.40×10³⁸, epsilon at 1.0 ≈ 1.19×10⁻⁷, subnormal min ≈ 1.40×10⁻⁴⁵.

Notes

Only bit-manipulation and pure-arithmetic functions are constexpr. STL cmath functions become constexpr in C++26.

f32_t
types.hpp:564 ↗
constexpr f32_t() noexcept;
constexpr f32_t(float v) noexcept;
constexpr f32_t(double v) noexcept;
template <std::integral integral_type_>
constexpr f32_t(integral_type_ v) noexcept;
*× 2Total ordering: -NaN < -Inf < ... < +Inf < +NaN.
types.hpp:643 ↗
int order(f32_t o) const noexcept;
Symbolo
orderf32_t
orderf64_t
saturating_*_mul× 2Saturating multiplication: clamps to finite range on overflow.
types.hpp:741 ↗
f32_t saturating__mul(f32_t o) const noexcept;
Symbolo
saturating_mulf32_t
saturating_mulf64_t
Structf32c_t64-bit single-precision complex number, composed of two f32_t.24
types.hpp:1051 ↗
struct f32c_t

Layout: {real: f32_t, imag: f32_t}. Uses composition, not inheritance. Supports complex arithmetic, conjugate, norm, abs, arg, and full transcendentals.

f32c_t
types.hpp:1088 ↗
constexpr f32c_t() noexcept;
constexpr f32c_t(f32_t r) noexcept;
constexpr f32c_t(float r) noexcept;
constexpr f32c_t(f32_t r, f32_t i) noexcept;
constexpr f32c_t(float r, float i) noexcept;
constexpr f32c_t(raw_t r) noexcept;
*× 2Complex conjugate: a - bi.
types.hpp:1159 ↗
f32c_t conj() const noexcept;
*× 2Squared magnitude, a² + b².
types.hpp:1162 ↗
f32_t norm() const noexcept;
*× 2Magnitude, √(a² + b²).
types.hpp:1165 ↗
f32_t abs() const noexcept;
*× 2Phase angle: atan2(b, a).
types.hpp:1168 ↗
f32_t arg() const noexcept;
*× 2Complex exponential, e^(a+bi) = e^a(cos(b) + i · sin(b)).
types.hpp:1171 ↗
f32c_t exp() const noexcept;
*× 2Complex natural logarithm, ln|z| + i · arg(z).
types.hpp:1177 ↗
f32c_t log() const noexcept;
*× 2Complex base-10 logarithm.
types.hpp:1180 ↗
f32c_t log10() const noexcept;
*× 2Complex square root, principal value.
types.hpp:1186 ↗
f32c_t sqrt() const noexcept;
*× 2Complex power, z^w = e^(w · ln(z)).
types.hpp:1194 ↗
f32c_t pow(f32c_t w) const noexcept;
Symbolw
powf32c_t
powf64c_t
*× 2Real power: z^x.
types.hpp:1197 ↗
f32c_t powf(f32_t x) const noexcept;
Symbolx
powff32_t
powff64_t
*× 2Complex sine: sin(z) = (eⁱᶻ − e⁻ⁱᶻ) / 2i.
types.hpp:1206 ↗
f32c_t sin() const noexcept;
*× 2Complex cosine: cos(z) = (eⁱᶻ + e⁻ⁱᶻ) / 2.
types.hpp:1212 ↗
f32c_t cos() const noexcept;
*× 2Complex tangent: tan(z) = sin(z) / cos(z).
types.hpp:1218 ↗
f32c_t tan() const noexcept;
*× 2Complex arcsine, asin(z) = -i · ln(iz + √(1 - z²)).
types.hpp:1221 ↗
f32c_t asin() const noexcept;
*× 2Complex arccosine: acos(z) = π/2 - asin(z).
types.hpp:1230 ↗
f32c_t acos() const noexcept;
*× 2Complex arctangent, atan(z) = (i/2) · ln((i+z)/(i-z)).
types.hpp:1237 ↗
f32c_t atan() const noexcept;
*× 2Complex hyperbolic sine: sinh(z) = (eᶻ − e⁻ᶻ) / 2.
types.hpp:1247 ↗
f32c_t sinh() const noexcept;
*× 2Complex hyperbolic cosine: cosh(z) = (eᶻ + e⁻ᶻ) / 2.
types.hpp:1253 ↗
f32c_t cosh() const noexcept;
*× 2Complex hyperbolic tangent: tanh(z) = sinh(z) / cosh(z).
types.hpp:1259 ↗
f32c_t tanh() const noexcept;
*× 2Complex inverse hyperbolic sine, asinh(z) = ln(z + √(z² + 1)).
types.hpp:1262 ↗
f32c_t asinh() const noexcept;
*× 2Complex inverse hyperbolic cosine, acosh(z) = ln(z + √(z² - 1)).
types.hpp:1268 ↗
f32c_t acosh() const noexcept;
*× 2Complex inverse hyperbolic tangent, atanh(z) = (1/2) · ln((1+z)/(1-z)).
types.hpp:1274 ↗
f32c_t atanh() const noexcept;
*× 2Reciprocal: 1 / z.
types.hpp:1282 ↗
f32c_t recip() const noexcept;
Structf64_t64-bit double-precision IEEE 754 floating-point wrapper.1
types.hpp:762 ↗
struct f64_t

Layout: sign(1) + exponent(11) + mantissa(52), bias=1023. Range: ±1.80×10³⁰⁸, epsilon at 1.0 ≈ 2.22×10⁻¹⁶, subnormal min ≈ 4.94×10⁻³²⁴.

Notes

Only bit-manipulation and pure-arithmetic functions are constexpr. STL cmath functions become constexpr in C++26.

f64_tUpcast from any nk numeric wrapper with operator float(), such as f16_t, bf16_t, e4m3_t, i8_t.
types.hpp:851 ↗
constexpr f64_t() noexcept;
constexpr f64_t(double v) noexcept;
constexpr f64_t(float v) noexcept;
template <std::integral integral_type_>
constexpr f64_t(integral_type_ v) noexcept;
template <typename nk_type_>
requires (std::is_class_v<nk_type_> && std::is_convertible_v<nk_type_, float>)
f64_t(nk_type_ v) noexcept;

Upcast from any nk numeric wrapper with operator float(), such as f16_t, bf16_t, e4m3_t, i8_t.

Structf64c_t128-bit double-precision complex number, composed of two f64_t.1
types.hpp:1291 ↗
struct f64c_t

Layout: {real: f64_t, imag: f64_t}. Uses composition, not inheritance. Supports complex arithmetic, conjugate, norm, abs, arg, and full transcendentals.

f64c_tUpcast from any nk complex wrapper with .real() / .imag(), such as f16c_t, bf16c_t, f32c_t.
types.hpp:1328 ↗
constexpr f64c_t() noexcept;
constexpr f64c_t(f64_t r) noexcept;
constexpr f64c_t(double r) noexcept;
constexpr f64c_t(f64_t r, f64_t i) noexcept;
constexpr f64c_t(double r, double i) noexcept;
constexpr f64c_t(raw_t r) noexcept;
template <typename nk_type_>
requires (!std::is_same_v<nk_type_, f64c_t> && !std::is_same_v<nk_type_, nk_f64c_t> && requires(nk_type_ v) { { static_cast<double>(v.real()) }; { static_cast<double>(v.imag()) }; })
f64c_t(nk_type_ v) noexcept;

Upcast from any nk complex wrapper with .real() / .imag(), such as f16c_t, bf16c_t, f32c_t.

Functions

castElementwise type-cast from one numeric type to another.
cast.hpp:32 ↗
template <numeric_dtype from_type_, numeric_dtype to_type_>
status_t cast(
    from_type_ const *from,
    std::size_t n,
    to_type_ *to,
    nk_capability_t capabilities = default_capabilities(),
    nk_stream_t stream = nullptr) noexcept;
template <numeric_dtype from_type_, numeric_dtype to_type_, vector_of<from_type_> from_vector_type_, mutable_vector_of<to_type_> to_vector_type_>
status_t cast(
    from_vector_type_ const & from,
    to_vector_type_ && to,
    nk_capability_t capabilities = default_capabilities(),
    nk_stream_t stream = nullptr) noexcept;

Elementwise type-cast of one run into another of equal dimensions; unexpected_dimensions_k when they differ or either run is strided or ends mid-value.

template <typename format_, std::size_t max_rank_>
status_t cast(
    tensor_view<f32_t, max_rank_> from,
    scaled_tensor_span<format_, max_rank_> to,
    nk_capability_t capabilities = default_capabilities(),
    nk_stream_t stream = nullptr) noexcept;

Encode (quantize) a dense f32 tensor into a preallocated block-scaled tensor.

template <typename format_>
status_t cast(
    vector_view<f32_t> from,
    scaled_tensor_span<format_> to,
    nk_capability_t capabilities = default_capabilities(),
    nk_stream_t stream = nullptr) noexcept;

Encode, or quantize, a dense f32 vector into a preallocated single-row block-scaled tensor.

template <typename format_, std::size_t max_rank_>
status_t cast(
    scaled_tensor_view<format_, max_rank_> from,
    vector_span<f32_t> to,
    nk_capability_t capabilities = default_capabilities(),
    nk_stream_t stream = nullptr) noexcept;

Decode, or dequantize, one block-scaled row into a dense f32 vector.

template <typename format_, std::size_t max_rank_>
status_t cast(
    scaled_tensor_view<format_, max_rank_> from,
    tensor_span<f32_t, max_rank_> to,
    nk_capability_t capabilities = default_capabilities(),
    nk_stream_t stream = nullptr) noexcept;

Decode (dequantize) a block-scaled tensor into a dense f32 tensor.

template <typename from_format_, typename to_format_, std::size_t max_rank_>
status_t cast(
    scaled_tensor_view<from_format_, max_rank_> from,
    scaled_tensor_span<to_format_, max_rank_> to,
    nk_capability_t capabilities = default_capabilities(),
    nk_stream_t stream = nullptr) noexcept;

Transcode one block-scaled tensor into another (e.g.

Derives the per-tensor scale (NVFP4) over the whole tensor in one contiguous pass and writes it back through the destination's scale slot. The source must be contiguous (the scale is a whole-tensor reduction). Allocate the destination with scaled_tensor<format_>::uninitialized(...), which reports the allocation's own status.

Parameters

frominInput array of n elements.
ninCounts dimensions, a multiple of the values per byte.
tooutOutput array of n elements.
capabilitiesinCapabilities to pick from.
streaminNull on the CPU, or the stream of the device capabilities describes.
is_numeric_dtypeDetect NumKong wrapper types with required static members.
types.hpp:366 ↗
template <typename scalar_type_>
bool is_numeric_dtype() noexcept;
is_integral_dtypeCheck if a type is an integer type.
types.hpp:372 ↗
template <typename scalar_type_>
bool is_integral_dtype() noexcept;
is_complex_dtypeCheck if a type is a complex type - STL or NumKong.
types.hpp:390 ↗
template <typename scalar_type_>
bool is_complex_dtype() noexcept;
is_signed_dtypeCheck if a type is an complex type - STL or NumKong.
types.hpp:397 ↗
template <typename scalar_type_>
bool is_signed_dtype() noexcept;
finite_maxGet the maximum representable value for a type.
types.hpp:6723 ↗
template <typename scalar_type_>
scalar_type_ finite_max() noexcept;
finite_minGet the lowest representable value for a type.
types.hpp:6730 ↗
template <typename scalar_type_>
scalar_type_ finite_min() noexcept;
is_nanWhether a value is NaN, never true for a type without one.
types.hpp:6828 ↗
template <numeric_dtype scalar_type_>
bool is_nan(scalar_type_ value) noexcept;
bits_per_valueBits per value.
types.hpp:6835 ↗
template <typename scalar_type_>
unsigned bits_per_value() noexcept;

For complex types matches value size.

bits_per_dimensionBits per value.
types.hpp:6842 ↗
template <typename scalar_type_>
unsigned bits_per_dimension() noexcept;

For complex types matches value size.

dimensions_per_valueDimensions per container value.
types.hpp:6857 ↗
template <typename scalar_type_>
unsigned dimensions_per_value() noexcept;

For normal types = 1, for sub-byte packed types > 1.

Block-scaled values, MX and NVFP4, carry a per-block scale byte inside the struct, so bits_per_value / bits_per_dimension would over-count by the scale overhead — NVFP4 is a 9-byte struct holding 16 four-bit dims, so the ratio yields 18, not 16. Every packed/block type exposes the authoritative logical-dimension count via elements(); prefer it when available and fall back to the bit ratio, correct at 1, for plain scalar types.

divide_round_upCeiling division, compile-time divisor: (n + divisor - 1) / divisor.
types.hpp:6965 ↗
template <std::size_t divisor_>
std::size_t divide_round_up(std::size_t n) noexcept;
std::size_t divide_round_up(
    std::size_t n,
    std::size_t divisor) noexcept;

Ceiling division, runtime divisor: (n + divisor - 1) / divisor.

round_up_to_multipleRound up to next multiple: ((n + multiple - 1) / multiple) * multiple.
types.hpp:6976 ↗
template <std::size_t multiple_>
std::size_t round_up_to_multiple(std::size_t n);
fcmaFMA helper template for baseline conjugate complex dot-product implementations.
types.hpp:6997 ↗
template <typename in_type_, typename accumulator_type_>
requires (dimensions_per_value<in_type_>() == 1)
accumulator_type_ fcma(
    in_type_ a,
    in_type_ b,
    accumulator_type_ acc) noexcept;
saturating_addFree-standing saturating addition for baseline implementations.
types.hpp:7015 ↗
template <typename accumulator_type_, typename in_type_>
requires (std::is_same<accumulator_type_, in_type_>::value || dimensions_per_value<in_type_>() == 1)
accumulator_type_ saturating_add(
    accumulator_type_ a,
    in_type_ b) noexcept;
template <typename accumulator_type_>
accumulator_type_ saturating_add(
    accumulator_type_ a,
    i4x2_t b) noexcept;

Saturating addition specialization for i4x2_t, signed 4-bit packed pairs.

template <typename accumulator_type_>
accumulator_type_ saturating_add(
    accumulator_type_ a,
    u4x2_t b) noexcept;

Saturating addition specialization for u4x2_t, unsigned 4-bit packed pairs.

saturating_mulFree-standing saturating multiplication for baseline implementations.
types.hpp:7022 ↗
template <typename accumulator_type_, typename in_type_>
requires (std::is_same<accumulator_type_, in_type_>::value || dimensions_per_value<in_type_>() == 1)
accumulator_type_ saturating_mul(
    accumulator_type_ a,
    in_type_ b) noexcept;
saturating_fmaFree-standing saturating FMA, a*b + c, for baseline implementations.
types.hpp:7042 ↗
template <typename in_type_, typename accumulator_type_>
requires fma_operands<in_type_, accumulator_type_>
accumulator_type_ saturating_fma(
    in_type_ a,
    in_type_ b,
    accumulator_type_ acc) noexcept;
template <typename in_type_, typename accumulator_type_>
requires signed_into_unsigned_fma_operands<in_type_, accumulator_type_>
accumulator_type_ saturating_fma(
    in_type_ a,
    in_type_ b,
    accumulator_type_ acc) noexcept;

Saturating FMA of signed operands into an unsigned integer accumulator, multiplied in i64_t first.

template <typename accumulator_type_>
requires (!is_integral_dtype<accumulator_type_>())
accumulator_type_ saturating_fma(
    i4x2_t a,
    i4x2_t b,
    accumulator_type_ acc) noexcept;

Saturating FMA specialization for i4x2_t, signed 4-bit packed pairs.

template <typename accumulator_type_>
requires (!is_integral_dtype<accumulator_type_>())
accumulator_type_ saturating_fma(
    u4x2_t a,
    u4x2_t b,
    accumulator_type_ acc) noexcept;

Saturating FMA specialization for u4x2_t, unsigned 4-bit packed pairs.

count_differencesCount element-level differences.
types.hpp:7126 ↗
template <typename scalar_type_>
unsigned count_differences(scalar_type_ a, scalar_type_ b) noexcept;
unsigned count_differences(u1x8_t a, u1x8_t b) noexcept;

Count bit-level differences for u1x8_t, 8 packed bits.

For scalar types, returns 1 if different, 0 if equal.

count_intersectionCount element-level intersection.
types.hpp:7135 ↗
template <typename scalar_type_>
unsigned count_intersection(scalar_type_ a, scalar_type_ b) noexcept;
unsigned count_intersection(u1x8_t a, u1x8_t b) noexcept;

Count bit-level intersection for u1x8_t, 8 packed bits.

For scalar types, returns 1 if equal, 0 if different.

count_unionCount element-level union.
types.hpp:7141 ↗
template <typename scalar_type_>
unsigned count_union(scalar_type_ a, scalar_type_ b) noexcept;
unsigned count_union(u1x8_t a, u1x8_t b) noexcept;

Count bit-level union for u1x8_t, 8 packed bits.

For scalar types, always returns 1, one element per pair.

TypesMember types319
bf16_t::raw_tnk_bf16_t
bf16_t::uint_tnk_u16_t
bf16_t::component_tbf16_t
bf16_t::scale_tnk_f32_t
bf16_t::sparse_dot_index_tu16_t
bf16_t::dot_result_tf32_t
bf16_t::attention_result_tf32_t
bf16_t::sqeuclidean_result_tf32_t
bf16_t::euclidean_result_tf32_t
bf16_t::angular_result_tf32_t
bf16_t::curved_result_tf32_t
bf16_t::probability_result_tf32_t
bf16_t::reduce_moments_sum_tf32_t
bf16_t::reduce_moments_sumsq_tf32_t
bf16_t::reduce_minmax_value_tbf16_t
bf16_t::maxsim_result_tf32_t
bf16_t::dot_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f32_t *, nk_stream_t)
bf16_t::angular_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f32_t *, nk_stream_t)
bf16_t::euclidean_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f32_t *, nk_stream_t)
bf16_t::sqeuclidean_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f32_t *, nk_stream_t)
bf16_t::curved_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, raw_t const *, nk_size_t, nk_f32_t *, nk_stream_t)
bf16_t::probability_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f32_t *, nk_stream_t)
bf16_t::sparse_dot_kernel_tnk_status_t(*)(nk_u16_t const *, nk_u16_t const *, raw_t const *, raw_t const *, nk_size_t, nk_size_t, nk_f32_t *, nk_stream_t)
bf16_t::scale_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, scale_t const *, scale_t const *, raw_t *, nk_stream_t)
bf16_t::sum_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, raw_t *, nk_stream_t)
bf16_t::blend_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, scale_t const *, scale_t const *, raw_t *, nk_stream_t)
bf16_t::fma_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, raw_t const *, nk_size_t, scale_t const *, scale_t const *, raw_t *, nk_stream_t)
bf16_t::swiglu_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, raw_t *, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_f32_t, nk_f32_t, nk_stream_t)
bf16_t::attention_rope_kernel_tnk_status_t(*)(raw_t const *, nk_f32_t const *, nk_f32_t const *, raw_t *, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_stream_t)
bf16_t::reduce_moments_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_f32_t *, nk_f32_t *, nk_stream_t)
bf16_t::reduce_minmax_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, raw_t *, nk_size_t *, raw_t *, nk_size_t *, nk_stream_t)
bf16_t::rmsnorm_kernel_tnk_status_t(*)(raw_t const *, nk_f32_t const *, raw_t *, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_f32_t, nk_stream_t)
bf16_t::dots_pack_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t *)
bf16_t::dots_pack_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_size_t, void *, nk_size_t, nk_size_t, nk_stream_t)
bf16_t::dots_packed_kernel_tnk_status_t(*)(raw_t const *, void const *, nk_f32_t *, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t, void *, nk_size_t, nk_stream_t)
bf16_t::dots_packed_workspace_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t, nk_size_t *)
bf16_t::dots_symmetric_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_size_t, nk_f32_t *, nk_size_t, nk_size_t, nk_size_t, void *, nk_size_t, nk_stream_t)
bf16_t::dots_symmetric_workspace_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t *)
bf16_t::angulars_packed_kernel_tdots_packed_kernel_t
bf16_t::euclideans_packed_kernel_tdots_packed_kernel_t
bf16_t::angulars_symmetric_kernel_tdots_symmetric_kernel_t
bf16_t::euclideans_symmetric_kernel_tdots_symmetric_kernel_t
bf16_t::maxsim_pack_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_size_t, void *, nk_stream_t)
bf16_t::maxsim_packed_kernel_tnk_status_t(*)(void const *, void const *, nk_size_t, nk_size_t, nk_size_t, nk_f32_t *, nk_stream_t)
bf16_t::attention_pack_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t *)
bf16_t::attention_pack_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_size_t, nk_u32_t const *, nk_u32_t const *, nk_size_t, nk_size_t, nk_size_t, void *, nk_size_t, nk_size_t, nk_stream_t)
bf16_t::attention_packed_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t, nk_u32_t const *, nk_size_t, nk_f32_t, nk_size_t, nk_size_t, raw_t const *, nk_size_t, void const *, nk_f32_t *, nk_size_t, nk_f32_t *, nk_size_t, nk_size_t, void *, nk_size_t, nk_stream_t)
bf16_t::attention_packed_gradients_kernel_tnk_status_t(*)( nk_size_t, nk_size_t, nk_size_t, nk_u32_t const *, nk_size_t, nk_f32_t, nk_size_t, nk_size_t, raw_t const *, nk_size_t, void const *, nk_f32_t const *, nk_f32_t const *, nk_size_t, nk_f32_t const *, nk_f32_t *, nk_size_t, nk_f32_t *, nk_f32_t *, nk_size_t, nk_size_t, nk_size_t, void *, nk_size_t, nk_stream_t)
e2m1x2_t::raw_tnk_e2m1x2_t
e2m1x2_t::component_tf32_t
e2m1x2_t::dot_result_tf32_t
e2m1x2_t::euclidean_result_tf32_t
e2m1x2_t::angular_result_tf32_t
e2m1x2_t::reduce_moments_sum_tf32_t
e2m1x2_t::reduce_moments_sumsq_tf32_t
e2m1x2_t::dot_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f32_t *, nk_stream_t)
e2m1x2_t::reduce_moments_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_f32_t *, nk_f32_t *, nk_stream_t)
e2m1x2_t::dots_pack_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t *)
e2m1x2_t::dots_pack_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_size_t, void *, nk_size_t, nk_size_t, nk_stream_t)
e2m1x2_t::dots_packed_kernel_tnk_status_t(*)(raw_t const *, void const *, nk_f32_t *, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t, void *, nk_size_t, nk_stream_t)
e2m1x2_t::dots_packed_workspace_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t, nk_size_t *)
e2m1x2_t::dots_symmetric_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_size_t, nk_f32_t *, nk_size_t, nk_size_t, nk_size_t, void *, nk_size_t, nk_stream_t)
e2m1x2_t::dots_symmetric_workspace_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t *)
e2m1x2_t::angulars_packed_kernel_tdots_packed_kernel_t
e2m1x2_t::euclideans_packed_kernel_tdots_packed_kernel_t
e2m1x2_t::angulars_symmetric_kernel_tdots_symmetric_kernel_t
e2m1x2_t::euclideans_symmetric_kernel_tdots_symmetric_kernel_t
e4m3_t::raw_tnk_e4m3_t
e4m3_t::uint_tnk_u8_t
e4m3_t::component_te4m3_t
e4m3_t::scale_tnk_f32_t
e4m3_t::dot_result_tf32_t
e4m3_t::attention_result_tf32_t
e4m3_t::sqeuclidean_result_tf32_t
e4m3_t::euclidean_result_tf32_t
e4m3_t::angular_result_tf32_t
e4m3_t::reduce_moments_sum_tf32_t
e4m3_t::reduce_moments_sumsq_tf32_t
e4m3_t::reduce_minmax_value_te4m3_t
e4m3_t::dot_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f32_t *, nk_stream_t)
e4m3_t::angular_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f32_t *, nk_stream_t)
e4m3_t::euclidean_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f32_t *, nk_stream_t)
e4m3_t::sqeuclidean_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f32_t *, nk_stream_t)
e4m3_t::scale_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, scale_t const *, scale_t const *, raw_t *, nk_stream_t)
e4m3_t::sum_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, raw_t *, nk_stream_t)
e4m3_t::blend_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, scale_t const *, scale_t const *, raw_t *, nk_stream_t)
e4m3_t::fma_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, raw_t const *, nk_size_t, scale_t const *, scale_t const *, raw_t *, nk_stream_t)
e4m3_t::swiglu_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, raw_t *, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_f32_t, nk_f32_t, nk_stream_t)
e4m3_t::attention_rope_kernel_tnk_status_t(*)(raw_t const *, nk_f32_t const *, nk_f32_t const *, raw_t *, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_stream_t)
e4m3_t::reduce_moments_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_f32_t *, nk_f32_t *, nk_stream_t)
e4m3_t::reduce_minmax_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, raw_t *, nk_size_t *, raw_t *, nk_size_t *, nk_stream_t)
e4m3_t::rmsnorm_kernel_tnk_status_t(*)(raw_t const *, nk_f32_t const *, raw_t *, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_f32_t, nk_stream_t)
e4m3_t::dots_pack_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t *)
e4m3_t::dots_pack_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_size_t, void *, nk_size_t, nk_size_t, nk_stream_t)
e4m3_t::dots_packed_kernel_tnk_status_t(*)(raw_t const *, void const *, nk_f32_t *, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t, void *, nk_size_t, nk_stream_t)
e4m3_t::dots_packed_workspace_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t, nk_size_t *)
e4m3_t::dots_symmetric_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_size_t, nk_f32_t *, nk_size_t, nk_size_t, nk_size_t, void *, nk_size_t, nk_stream_t)
e4m3_t::dots_symmetric_workspace_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t *)
e4m3_t::angulars_packed_kernel_tdots_packed_kernel_t
e4m3_t::euclideans_packed_kernel_tdots_packed_kernel_t
e4m3_t::angulars_symmetric_kernel_tdots_symmetric_kernel_t
e4m3_t::euclideans_symmetric_kernel_tdots_symmetric_kernel_t
e4m3_t::maxsim_pack_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_size_t, void *, nk_stream_t)
e4m3_t::attention_pack_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t *)
e4m3_t::attention_pack_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_size_t, nk_u32_t const *, nk_u32_t const *, nk_size_t, nk_size_t, nk_size_t, void *, nk_size_t, nk_size_t, nk_stream_t)
e4m3_t::attention_packed_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t, nk_u32_t const *, nk_size_t, nk_f32_t, nk_size_t, nk_size_t, raw_t const *, nk_size_t, void const *, nk_f32_t *, nk_size_t, nk_f32_t *, nk_size_t, nk_size_t, void *, nk_size_t, nk_stream_t)
e5m2_t::raw_tnk_e5m2_t
e5m2_t::uint_tnk_u8_t
e5m2_t::component_te5m2_t
e5m2_t::scale_tnk_f32_t
e5m2_t::dot_result_tf32_t
e5m2_t::sqeuclidean_result_tf32_t
e5m2_t::euclidean_result_tf32_t
e5m2_t::angular_result_tf32_t
e5m2_t::reduce_moments_sum_tf32_t
e5m2_t::reduce_moments_sumsq_tf32_t
e5m2_t::reduce_minmax_value_te5m2_t
e5m2_t::dot_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f32_t *, nk_stream_t)
e5m2_t::angular_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f32_t *, nk_stream_t)
e5m2_t::euclidean_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f32_t *, nk_stream_t)
e5m2_t::sqeuclidean_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f32_t *, nk_stream_t)
e5m2_t::scale_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, scale_t const *, scale_t const *, raw_t *, nk_stream_t)
e5m2_t::sum_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, raw_t *, nk_stream_t)
e5m2_t::blend_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, scale_t const *, scale_t const *, raw_t *, nk_stream_t)
e5m2_t::fma_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, raw_t const *, nk_size_t, scale_t const *, scale_t const *, raw_t *, nk_stream_t)
e5m2_t::reduce_moments_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_f32_t *, nk_f32_t *, nk_stream_t)
e5m2_t::reduce_minmax_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, raw_t *, nk_size_t *, raw_t *, nk_size_t *, nk_stream_t)
e5m2_t::dots_pack_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t *)
e5m2_t::dots_pack_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_size_t, void *, nk_size_t, nk_size_t, nk_stream_t)
e5m2_t::dots_packed_kernel_tnk_status_t(*)(raw_t const *, void const *, nk_f32_t *, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t, void *, nk_size_t, nk_stream_t)
e5m2_t::dots_packed_workspace_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t, nk_size_t *)
e5m2_t::dots_symmetric_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_size_t, nk_f32_t *, nk_size_t, nk_size_t, nk_size_t, void *, nk_size_t, nk_stream_t)
e5m2_t::dots_symmetric_workspace_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t *)
e5m2_t::angulars_packed_kernel_tdots_packed_kernel_t
e5m2_t::euclideans_packed_kernel_tdots_packed_kernel_t
e5m2_t::angulars_symmetric_kernel_tdots_symmetric_kernel_t
e5m2_t::euclideans_symmetric_kernel_tdots_symmetric_kernel_t
e5m2_t::maxsim_pack_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_size_t, void *, nk_stream_t)
f118_t::component_tf118_t
f16_t::raw_tnk_f16_t
f16_t::uint_tnk_u16_t
f16_t::component_tf16_t
f16_t::scale_tnk_f32_t
f16_t::dot_result_tf32_t
f16_t::sqeuclidean_result_tf32_t
f16_t::euclidean_result_tf32_t
f16_t::angular_result_tf32_t
f16_t::curved_result_tf32_t
f16_t::probability_result_tf32_t
f16_t::reduce_moments_sum_tf32_t
f16_t::reduce_moments_sumsq_tf32_t
f16_t::reduce_minmax_value_tf16_t
f16_t::shapes_distance_tf32_t
f16_t::attention_result_tf32_t
f16_t::maxsim_result_tf32_t
f16_t::dot_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f32_t *, nk_stream_t)
f16_t::angular_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f32_t *, nk_stream_t)
f16_t::euclidean_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f32_t *, nk_stream_t)
f16_t::sqeuclidean_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f32_t *, nk_stream_t)
f16_t::curved_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, raw_t const *, nk_size_t, nk_f32_t *, nk_stream_t)
f16_t::probability_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f32_t *, nk_stream_t)
f16_t::shapes_pack_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t *)
f16_t::shapes_pack_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, void *, nk_stream_t)
f16_t::shapes_packed_shape_kernel_tnk_status_t(*)(void const *, nk_size_t *, nk_f32_t *, nk_f32_t *, nk_f32_t *, nk_f32_t *, nk_f32_t *, nk_stream_t)
f16_t::shapes_unpack_kernel_tnk_status_t(*)(void const *, raw_t *, nk_stream_t)
f16_t::shapes_workspace_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t *)
f16_t::chamfers_packed_kernel_tnk_status_t(*)(void const *const *, nk_size_t, void const *const *, nk_size_t, nk_f32_t *, nk_size_t, nk_f32_t *, nk_f32_t const *, void *, nk_size_t, nk_stream_t)
f16_t::hausdorffs_packed_kernel_tnk_status_t(*)(void const *const *, nk_size_t, void const *const *, nk_size_t, nk_f32_t *, nk_size_t, nk_f32_t *, nk_f32_t const *, nk_stream_t)
f16_t::wassersteinprocrustes_packed_kernel_tnk_status_t(*)(void const *const *, nk_size_t, void const *const *, nk_size_t, nk_f32_t *, nk_size_t, nk_f32_t *, void *, nk_size_t, nk_stream_t)
f16_t::gromovwassersteins_packed_kernel_twassersteinprocrustes_packed_kernel_t
f16_t::rmsds_packed_kernel_tnk_status_t(*)(void const *const *, nk_size_t, void const *const *, nk_size_t, nk_f32_t *, nk_size_t, nk_stream_t)
f16_t::scale_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, scale_t const *, scale_t const *, raw_t *, nk_stream_t)
f16_t::sum_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, raw_t *, nk_stream_t)
f16_t::blend_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, scale_t const *, scale_t const *, raw_t *, nk_stream_t)
f16_t::fma_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, raw_t const *, nk_size_t, scale_t const *, scale_t const *, raw_t *, nk_stream_t)
f16_t::trigonometry_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, raw_t *, nk_stream_t)
f16_t::reduce_moments_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_f32_t *, nk_f32_t *, nk_stream_t)
f16_t::reduce_minmax_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, raw_t *, nk_size_t *, raw_t *, nk_size_t *, nk_stream_t)
f16_t::dots_pack_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t *)
f16_t::dots_pack_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_size_t, void *, nk_size_t, nk_size_t, nk_stream_t)
f16_t::dots_packed_kernel_tnk_status_t(*)(raw_t const *, void const *, nk_f32_t *, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t, void *, nk_size_t, nk_stream_t)
f16_t::dots_packed_workspace_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t, nk_size_t *)
f16_t::dots_symmetric_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_size_t, nk_f32_t *, nk_size_t, nk_size_t, nk_size_t, void *, nk_size_t, nk_stream_t)
f16_t::dots_symmetric_workspace_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t *)
f16_t::angulars_packed_kernel_tdots_packed_kernel_t
f16_t::euclideans_packed_kernel_tdots_packed_kernel_t
f16_t::angulars_symmetric_kernel_tdots_symmetric_kernel_t
f16_t::euclideans_symmetric_kernel_tdots_symmetric_kernel_t
f16_t::maxsim_pack_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_size_t, void *, nk_stream_t)
f16_t::maxsim_packed_kernel_tnk_status_t(*)(void const *, void const *, nk_size_t, nk_size_t, nk_size_t, nk_f32_t *, nk_stream_t)
f32_t::raw_tnk_f32_t
f32_t::uint_tnk_u32_t
f32_t::component_tf32_t
f32_t::scale_tnk_f32_t
f32_t::sparse_dot_index_tu32_t
f32_t::dot_result_tf64_t
f32_t::sqeuclidean_result_tf64_t
f32_t::euclidean_result_tf64_t
f32_t::angular_result_tf64_t
f32_t::curved_result_tf64_t
f32_t::probability_result_tf64_t
f32_t::reduce_moments_sum_tf64_t
f32_t::reduce_moments_sumsq_tf64_t
f32_t::reduce_minmax_value_tf32_t
f32_t::shapes_distance_tf64_t
f32_t::maxsim_result_tf64_t
f32_t::dot_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f64_t *, nk_stream_t)
f32_t::angular_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f64_t *, nk_stream_t)
f32_t::euclidean_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f64_t *, nk_stream_t)
f32_t::sqeuclidean_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f64_t *, nk_stream_t)
f32_t::curved_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, raw_t const *, nk_size_t, nk_f64_t *, nk_stream_t)
f32_t::probability_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f64_t *, nk_stream_t)
f32_t::shapes_pack_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t *)
f32_t::shapes_pack_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, void *, nk_stream_t)
f32_t::shapes_packed_shape_kernel_tnk_status_t(*)(void const *, nk_size_t *, nk_f64_t *, nk_f64_t *, nk_f64_t *, nk_f64_t *, nk_f64_t *, nk_stream_t)
f32_t::shapes_unpack_kernel_tnk_status_t(*)(void const *, raw_t *, nk_stream_t)
f32_t::shapes_workspace_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t *)
f32_t::chamfers_packed_kernel_tnk_status_t(*)(void const *const *, nk_size_t, void const *const *, nk_size_t, nk_f64_t *, nk_size_t, nk_f32_t *, nk_f64_t const *, void *, nk_size_t, nk_stream_t)
f32_t::hausdorffs_packed_kernel_tnk_status_t(*)(void const *const *, nk_size_t, void const *const *, nk_size_t, nk_f64_t *, nk_size_t, nk_f32_t *, nk_f64_t const *, nk_stream_t)
f32_t::wassersteinprocrustes_packed_kernel_tnk_status_t(*)(void const *const *, nk_size_t, void const *const *, nk_size_t, nk_f64_t *, nk_size_t, nk_f32_t *, void *, nk_size_t, nk_stream_t)
f32_t::gromovwassersteins_packed_kernel_twassersteinprocrustes_packed_kernel_t
f32_t::rmsds_packed_kernel_tnk_status_t(*)(void const *const *, nk_size_t, void const *const *, nk_size_t, nk_f64_t *, nk_size_t, nk_stream_t)
f32_t::sparse_dot_kernel_tnk_status_t(*)(nk_u32_t const *, nk_u32_t const *, raw_t const *, raw_t const *, nk_size_t, nk_size_t, nk_f64_t *, nk_stream_t)
f32_t::scale_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, scale_t const *, scale_t const *, raw_t *, nk_stream_t)
f32_t::sum_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, raw_t *, nk_stream_t)
f32_t::blend_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, scale_t const *, scale_t const *, raw_t *, nk_stream_t)
f32_t::fma_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, raw_t const *, nk_size_t, scale_t const *, scale_t const *, raw_t *, nk_stream_t)
f32_t::swiglu_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, raw_t *, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_f32_t, nk_f32_t, nk_stream_t)
f32_t::trigonometry_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, raw_t *, nk_stream_t)
f32_t::attention_rope_kernel_tnk_status_t(*)(raw_t const *, nk_f32_t const *, nk_f32_t const *, raw_t *, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_stream_t)
f32_t::reduce_moments_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_f64_t *, nk_f64_t *, nk_stream_t)
f32_t::reduce_minmax_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, raw_t *, nk_size_t *, raw_t *, nk_size_t *, nk_stream_t)
f32_t::rmsnorm_kernel_tnk_status_t(*)(raw_t const *, nk_f32_t const *, raw_t *, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_f32_t, nk_stream_t)
f32_t::dots_pack_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t *)
f32_t::dots_pack_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_size_t, void *, nk_size_t, nk_size_t, nk_stream_t)
f32_t::dots_packed_kernel_tnk_status_t(*)(raw_t const *, void const *, nk_f64_t *, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t, void *, nk_size_t, nk_stream_t)
f32_t::dots_packed_workspace_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t, nk_size_t *)
f32_t::dots_symmetric_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_size_t, nk_f64_t *, nk_size_t, nk_size_t, nk_size_t, void *, nk_size_t, nk_stream_t)
f32_t::dots_symmetric_workspace_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t *)
f32_t::angulars_packed_kernel_tdots_packed_kernel_t
f32_t::euclideans_packed_kernel_tdots_packed_kernel_t
f32_t::angulars_symmetric_kernel_tdots_symmetric_kernel_t
f32_t::euclideans_symmetric_kernel_tdots_symmetric_kernel_t
f32_t::maxsim_pack_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_size_t, void *, nk_stream_t)
f32_t::maxsim_packed_kernel_tnk_status_t(*)(void const *, void const *, nk_size_t, nk_size_t, nk_size_t, nk_f64_t *, nk_stream_t)
f32c_t::component_tf32_t
f32c_t::raw_tnk_f32c_t
f32c_t::dot_result_tf64c_t
f32c_t::vdot_result_tf64c_t
f32c_t::curved_result_tf64c_t
f32c_t::scale_tnk_f32c_t
f32c_t::dot_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f64c_t *, nk_stream_t)
f32c_t::vdot_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f64c_t *, nk_stream_t)
f32c_t::curved_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, raw_t const *, nk_size_t, nk_f64c_t *, nk_stream_t)
f32c_t::scale_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, scale_t const *, scale_t const *, raw_t *, nk_stream_t)
f32c_t::sum_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, raw_t *, nk_stream_t)
f32c_t::blend_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, scale_t const *, scale_t const *, raw_t *, nk_stream_t)
f32c_t::fma_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, raw_t const *, nk_size_t, scale_t const *, scale_t const *, raw_t *, nk_stream_t)
f64_t::raw_tnk_f64_t
f64_t::uint_tnk_u64_t
f64_t::component_tf64_t
f64_t::scale_tnk_f64_t
f64_t::dot_result_tf64_t
f64_t::sqeuclidean_result_tf64_t
f64_t::euclidean_result_tf64_t
f64_t::angular_result_tf64_t
f64_t::curved_result_tf64_t
f64_t::probability_result_tf64_t
f64_t::reduce_moments_sum_tf64_t
f64_t::reduce_moments_sumsq_tf64_t
f64_t::reduce_minmax_value_tf64_t
f64_t::shapes_distance_tf64_t
f64_t::dot_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f64_t *, nk_stream_t)
f64_t::angular_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f64_t *, nk_stream_t)
f64_t::euclidean_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f64_t *, nk_stream_t)
f64_t::sqeuclidean_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f64_t *, nk_stream_t)
f64_t::curved_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, raw_t const *, nk_size_t, raw_t *, nk_stream_t)
f64_t::probability_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f64_t *, nk_stream_t)
f64_t::scale_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, scale_t const *, scale_t const *, raw_t *, nk_stream_t)
f64_t::sum_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, raw_t *, nk_stream_t)
f64_t::blend_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, scale_t const *, scale_t const *, raw_t *, nk_stream_t)
f64_t::fma_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, raw_t const *, nk_size_t, scale_t const *, scale_t const *, raw_t *, nk_stream_t)
f64_t::trigonometry_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, raw_t *, nk_stream_t)
f64_t::reduce_moments_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_f64_t *, nk_f64_t *, nk_stream_t)
f64_t::reduce_minmax_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, raw_t *, nk_size_t *, raw_t *, nk_size_t *, nk_stream_t)
f64_t::dots_pack_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t *)
f64_t::dots_pack_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_size_t, void *, nk_size_t, nk_size_t, nk_stream_t)
f64_t::dots_packed_kernel_tnk_status_t(*)(raw_t const *, void const *, nk_f64_t *, nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t, void *, nk_size_t, nk_stream_t)
f64_t::dots_packed_workspace_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t, nk_size_t *)
f64_t::dots_symmetric_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_size_t, nk_f64_t *, nk_size_t, nk_size_t, nk_size_t, void *, nk_size_t, nk_stream_t)
f64_t::dots_symmetric_workspace_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t, nk_size_t, nk_size_t *)
f64_t::angulars_packed_kernel_tdots_packed_kernel_t
f64_t::euclideans_packed_kernel_tdots_packed_kernel_t
f64_t::angulars_symmetric_kernel_tdots_symmetric_kernel_t
f64_t::euclideans_symmetric_kernel_tdots_symmetric_kernel_t
f64_t::maxsim_pack_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, nk_size_t, nk_size_t, void *, nk_stream_t)
f64_t::shapes_pack_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t *)
f64_t::shapes_pack_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, void *, nk_stream_t)
f64_t::shapes_packed_shape_kernel_tnk_status_t(*)(void const *, nk_size_t *, nk_f64_t *, nk_f64_t *, nk_f64_t *, nk_f64_t *, nk_f64_t *, nk_stream_t)
f64_t::shapes_unpack_kernel_tnk_status_t(*)(void const *, raw_t *, nk_stream_t)
f64_t::shapes_workspace_size_kernel_tnk_status_t(*)(nk_size_t, nk_size_t, nk_size_t *)
f64_t::chamfers_packed_kernel_tnk_status_t(*)(void const *const *, nk_size_t, void const *const *, nk_size_t, nk_f64_t *, nk_size_t, nk_f32_t *, nk_f64_t const *, void *, nk_size_t, nk_stream_t)
f64_t::hausdorffs_packed_kernel_tnk_status_t(*)(void const *const *, nk_size_t, void const *const *, nk_size_t, nk_f64_t *, nk_size_t, nk_f32_t *, nk_f64_t const *, nk_stream_t)
f64_t::wassersteinprocrustes_packed_kernel_tnk_status_t(*)(void const *const *, nk_size_t, void const *const *, nk_size_t, nk_f64_t *, nk_size_t, nk_f32_t *, void *, nk_size_t, nk_stream_t)
f64_t::gromovwassersteins_packed_kernel_twassersteinprocrustes_packed_kernel_t
f64_t::rmsds_packed_kernel_tnk_status_t(*)(void const *const *, nk_size_t, void const *const *, nk_size_t, nk_f64_t *, nk_size_t, nk_stream_t)
f64c_t::component_tf64_t
f64c_t::raw_tnk_f64c_t
f64c_t::dot_result_tf64c_t
f64c_t::vdot_result_tf64c_t
f64c_t::curved_result_tf64c_t
f64c_t::scale_tnk_f64c_t
f64c_t::dot_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f64c_t *, nk_stream_t)
f64c_t::vdot_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, nk_f64c_t *, nk_stream_t)
f64c_t::curved_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, raw_t const *, nk_size_t, raw_t *, nk_stream_t)
f64c_t::scale_kernel_tnk_status_t(*)(raw_t const *, nk_size_t, scale_t const *, scale_t const *, raw_t *, nk_stream_t)
f64c_t::sum_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, raw_t *, nk_stream_t)
f64c_t::blend_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, nk_size_t, scale_t const *, scale_t const *, raw_t *, nk_stream_t)
f64c_t::fma_kernel_tnk_status_t(*)(raw_t const *, raw_t const *, raw_t const *, nk_size_t, scale_t const *, scale_t const *, raw_t *, nk_stream_t)