Skip to main content

Mojo struct

KVCacheMHAOperand

@register_passable(trivial) struct KVCacheMHAOperand[cache_t: KVCacheT]

An implementation for mo.opaque KVCacheT arguments to MHA kernels.

We can eventually remove this trait and just add it as a sub-trait in the KVCacheT type, but we need to solve some cyclic dependencies first.

Fields

  • cache (cache_t):

Implemented traits

AnyType, Copyable, ImplicitlyCopyable, MHAOperand, Movable, UnknownDestructibility

Aliases

__copyinit__is_trivial

alias __copyinit__is_trivial = cache_t.__copyinit__is_trivial

__del__is_trivial

alias __del__is_trivial = cache_t.__del__is_trivial

__moveinit__is_trivial

alias __moveinit__is_trivial = cache_t.__moveinit__is_trivial

dtype

alias dtype = cache_t.dtype

Methods

__init__

__init__(cache: cache_t) -> Self

block_paged_ptr

block_paged_ptr[tile_size: Int](self, batch_idx: UInt32, start_tok_idx: UInt32, head_idx: UInt32, head_dim_idx: UInt32 = 0) -> UnsafePointer[Scalar[cache_t.dtype]]

Returns:

UnsafePointer

cache_length

cache_length(self, batch_idx: Int) -> Int

Returns:

Int

max_context_length

max_context_length(self) -> UInt32

Returns:

UInt32

row_idx

row_idx(self, batch_idx: UInt32, start_tok_idx: UInt32) -> UInt32

Returns the row idx when viewing the memory as a matrix.

Returns:

UInt32

col_idx

col_idx(self, head_idx: UInt32) -> UInt32

Returns the col idx when viewing the memory as a matrix.

Returns:

UInt32

create_tma_tile

create_tma_tile[tile_m: Int, tile_n: Int, swizzle_mode: TensorMapSwizzle, *, is_k_major: Bool](self, ctx: DeviceContext) -> TMATensorTile[cache_t.dtype, tile_layout_k_major[cache_t.dtype, tile_m, tile_n, swizzle_mode]() if is_k_major else tile_layout_mn_major[cache_t.dtype, tile_n, tile_m, swizzle_mode](), _tma_desc_tile_layout[cache_t.dtype, 2, IndexList[2, DType.int64](tile_m, tile_n, Tuple[]()), is_k_major, swizzle_mode](), is_k_major]

Creates a TMA tile for efficient GPU memory transfers.

Returns:

TMATensorTile

Was this page helpful?