|
MayaFlux 1.0.0
Digital-First Multimedia Processing Framework
|
Element type as kind, bit width and lane count. More...
#include <TensorDType.hpp>
Public Member Functions | |
| constexpr size_t | element_bits () const noexcept |
| Bits of one element, lanes included. | |
| constexpr size_t | element_bytes () const noexcept |
| Bytes of one element, zero for a sub-byte type. | |
| std::optional< GpuDataFormat > | gpu_format () const noexcept |
| GpuDataFormat holding exactly this element, when one exists. | |
| template<TensorElement T> | |
| constexpr bool | is () const noexcept |
| Whether this type is exactly the C++ element type T. | |
| constexpr bool | is_sub_byte () const noexcept |
| Whether one element is not a whole number of bytes. | |
| std::string | name () const |
| Short readable name such as f32, bf16, i64, u8. | |
| constexpr bool | operator== (const TensorDType &) const noexcept=default |
| std::optional< size_t > | storage_bytes (uint64_t count) const noexcept |
Packed bytes needed for count elements. | |
Static Public Member Functions | |
| static constexpr TensorDType | bfloat16 () noexcept |
| Brain floating point, 16 bits. | |
| static constexpr TensorDType | float16 () noexcept |
| IEEE half precision. | |
| template<TensorElement T> | |
| static constexpr TensorDType | of () noexcept |
| Type for a C++ scalar or complex element. | |
Public Attributes | |
| uint8_t | bits { 32 } |
| TensorDTypeKind | kind { TensorDTypeKind::FLOAT } |
| uint16_t | lanes { 1 } |
Element type as kind, bit width and lane count.
Mirrors the DLPack data type, so half, bfloat16, float8, int4 and every integer width are first class and nothing is widened. Sub-byte types are packed; their storage is rounded up to whole bytes for the tensor.
Usage:
Definition at line 64 of file TensorDType.hpp.