Enum pjrt_sys::protos::xla::precision_config::Algorithm

#[repr(i32)]
pub enum Algorithm {
    AlgUnset = 0,
    AlgDotAnyF8AnyF8F32 = 1,
    AlgDotAnyF8AnyF8F32FastAccum = 2,
    AlgDotF16F16F16 = 3,
    AlgDotF16F16F32 = 4,
    AlgDotBf16Bf16Bf16 = 5,
    AlgDotBf16Bf16F32 = 6,
    AlgDotBf16Bf16F32X3 = 7,
    AlgDotBf16Bf16F32X6 = 8,
    AlgDotTf32Tf32F32 = 9,
    AlgDotTf32Tf32F32X3 = 10,
    AlgDotF32F32F32 = 11,
    AlgDotF64F64F64 = 12,
}

The algorithm used to evaluate the instruction.

The naming convention for the dot instruction is ALG_DOT_{A_TYPE}{B_TYPE}{ACCUM_TYPE}[_X{NUM_OPS}] where A_TYPE, B_TYPE and ACCUM_TYPE correspond to the types in the “primitive dot operations” (such as TensorCore operations) and NUM_OPS is the number of such operations used per “primitive tile”. When the NUM_OPS field is skipped, it is assumed to be 1. The types mentioned in the name are independent of the storage types.

In general ATYPE and BTYPE are the precisions that the LHS and RHS of the operation are rounded to and ACCUMTYPE is the accumulation type. If a backend does not support the given algorithm, an error is raised. The Algorithm enum is intended to eventually replace the Precision enum.

Variants

AlgUnset = 0

If the algorithm is ALG_UNSET, we will decide the algorithm based on the operand_precision values (for now).

AlgDotAnyF8AnyF8F32 = 1

The storage type can be any 8-bit floating point type.

AlgDotAnyF8AnyF8F32FastAccum = 2

The storage type can be any 8-bit floating point type. Intermediate results will not periodically be promoted to a higher precision. This corresponds to CUBLASLT_MATMUL_DESC_FAST_ACCUM. Triton’s maxNumImpreciseAcc=32 setting may be similar.

AlgDotF16F16F16 = 3

AlgDotF16F16F32 = 4

AlgDotBf16Bf16Bf16 = 5

AlgDotBf16Bf16F32 = 6

AlgDotBf16Bf16F32X3 = 7

An algorithm which uses 3 BF16_BF16_F32 matmuls to achieve better precision.

AlgDotBf16Bf16F32X6 = 8

An algorithm which uses 6 BF16_BF16_F32 matmuls to achieve better precision (similar to F32).

AlgDotTf32Tf32F32 = 9

AlgDotTf32Tf32F32X3 = 10

An algorithm which uses 3 TF32_TF32_F32 matmuls to achieve better precision (similar to F32).

AlgDotF32F32F32 = 11

AlgDotF64F64F64 = 12

Implementations

impl Algorithm

pub fn is_valid(value: i32) -> bool

Returns true if value is a variant of Algorithm.

pub fn from_i32(value: i32) -> Option<Algorithm>

👎Deprecated: Use the TryFrom<i32> implementation instead

Converts an i32 to a Algorithm, or None if value is not a valid variant.

impl Algorithm

pub fn as_str_name(&self) -> &'static str

String value of the enum field names used in the ProtoBuf definition.

The values are not transformed in any way and thus are considered stable (if the ProtoBuf definition does not change) and safe for programmatic use.

pub fn from_str_name(value: &str) -> Option<Self>

Creates an enum from field names used in the ProtoBuf definition.

Trait Implementations

impl Clone for Algorithm

fn clone(&self) -> Algorithm

Returns a copy of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for Algorithm

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for Algorithm

fn default() -> Algorithm

Returns the “default value” for a type.

impl From<Algorithm> for i32

fn from(value: Algorithm) -> i32

Converts to this type from the input type.

impl Hash for Algorithm

fn hash<__H: Hasher>(&self, state: &mut __H)

Feeds this value into the given Hasher.

fn hash_slice<H>(data: &[Self], state: &mut H)

where H: Hasher, Self: Sized,

Feeds a slice of this type into the given Hasher.

impl Ord for Algorithm

fn cmp(&self, other: &Algorithm) -> Ordering

This method returns an Ordering between self and other.

fn max(self, other: Self) -> Self

where Self: Sized,

Compares and returns the maximum of two values.

fn min(self, other: Self) -> Self

where Self: Sized,

Compares and returns the minimum of two values.

fn clamp(self, min: Self, max: Self) -> Self

where Self: Sized + PartialOrd,

Restrict a value to a certain interval.

impl PartialEq for Algorithm

fn eq(&self, other: &Algorithm) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl PartialOrd for Algorithm

fn partial_cmp(&self, other: &Algorithm) -> Option<Ordering>

This method returns an ordering between self and other values if one exists.

fn lt(&self, other: &Rhs) -> bool

Tests less than (for self and other) and is used by the < operator.

fn le(&self, other: &Rhs) -> bool

Tests less than or equal to (for self and other) and is used by the <= operator.

fn gt(&self, other: &Rhs) -> bool

Tests greater than (for self and other) and is used by the > operator.

fn ge(&self, other: &Rhs) -> bool

Tests greater than or equal to (for self and other) and is used by the >= operator.

impl TryFrom<i32> for Algorithm

type Error = UnknownEnumValue

The type returned in the event of a conversion error.

fn try_from(value: i32) -> Result<Algorithm, UnknownEnumValue>

Performs the conversion.

impl Copy for Algorithm

impl Eq for Algorithm

impl StructuralPartialEq for Algorithm