For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /max/get-started.md).

Mojo struct

GEMVAlgorithm

struct GEMVAlgorithm

Enumerates the GEMV kernel algorithm variants used by the GPU dispatcher.

Each variant targets a distinct operand shape, memory access pattern, or hardware capability. The dispatcher in gemv_gpu selects among these based on runtime shape information and target architecture.

Implemented traits

AnyType, Copyable, Equatable, Hashable, ImplicitlyCopyable, ImplicitlyDeletable, Movable, RegisterPassable, TrivialRegisterPassable, Writable

`comptime` members

`GEMM_MMA_CPASYNC`

comptime GEMM_MMA_CPASYNC = GEMVAlgorithm(Int(6))

`GEMV_KERNEL`

comptime GEMV_KERNEL = GEMVAlgorithm(Int(0))

`GEMV_KERNEL_VECTOR`

comptime GEMV_KERNEL_VECTOR = GEMVAlgorithm(Int(1))

`GEMV_SPLIT_K`

comptime GEMV_SPLIT_K = GEMVAlgorithm(Int(2))

`GEVM_KERNEL`

comptime GEVM_KERNEL = GEMVAlgorithm(Int(4))

`GEVM_KERNEL_VECTOR`

comptime GEVM_KERNEL_VECTOR = GEMVAlgorithm(Int(3))

`MATMUL_NAIVE`

comptime MATMUL_NAIVE = GEMVAlgorithm(Int(5))

Methods

`eq`

def __eq__(self, other: Self) -> Bool

Returns:

Bool

`ne`

def __ne__(self, other: Self) -> Bool

Returns:

Bool

`is`

def __is__(self, other: Self) -> Bool

Returns:

Bool

`isnot`

def __isnot__(self, other: Self) -> Bool

Returns:

Bool

`int`

def __int__(self) -> Int

Returns:

Int

`hash`

def __hash__(self) -> Int

Returns:

Int

`write_to`

def write_to(self, mut writer: T)

Implemented traits​

comptime members​

GEMM_MMA_CPASYNC​

GEMV_KERNEL​

GEMV_KERNEL_VECTOR​

GEMV_SPLIT_K​

GEVM_KERNEL​

GEVM_KERNEL_VECTOR​

MATMUL_NAIVE​

Methods​

__eq__​

__ne__​

__is__​

__isnot__​

__int__​

__hash__​

write_to​

Implemented traits

`comptime` members

`GEMM_MMA_CPASYNC`

`GEMV_KERNEL`

`GEMV_KERNEL_VECTOR`

`GEMV_SPLIT_K`

`GEVM_KERNEL`

`GEVM_KERNEL_VECTOR`

`MATMUL_NAIVE`

Methods

`eq`

`ne`

`is`

`isnot`

`int`

`hash`

`write_to`