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

Mojo struct

Depth512SM100Config

struct Depth512SM100Config[qkv_dtype: DType, *, rope_dtype_: Optional[DType] = None, scale_dtype_: Optional[DType] = None]

Fields

• ​MMA_M (Int):
• ​BM (Int):
• ​num_qk_stages (Int):
• ​split_o (Bool):
• ​v_cols_per_cta (Int):
• ​BN (Int):
• ​BK0 (Int):
• ​BK1 (Int):
• ​qk_depth (Int):
• ​ov_depth (Int):
• ​group (Int):
• ​num_q_heads (Int):
• ​num_kv_heads (Int):
• ​TMEM_O (Int):
• ​TMEM_O_hi (Int):
• ​TMEM_S_even (Int):
• ​TMEM_S_odd (Int):
• ​tmem_used (Int):
• ​fuse_gqa (Bool):
• ​num_kv_stages (Int):
• ​smem_used (Int):
• ​swizzle_mode (TensorMapSwizzle):
• ​p_buf_bytes (Int):

Implemented traits

AnyType, Copyable, Deinitable, ImplicitlyCopyable, Movable, RegisterPassable, TrivialRegisterPassable

comptime members

cta_group

comptime cta_group = Int(2)

mbar_size

comptime mbar_size = size_of[DType.int64]()

MMA_K

comptime MMA_K = Int(16) if qkv_dtype.is_half_float() else Int(32)

num_pv_stages

comptime num_pv_stages = Int(2)

num_threads

comptime num_threads = (Int(3) * _resolve_warpgroup_size())

qkv_dtype_size

comptime qkv_dtype_size = size_of[qkv_dtype]()

rope_dtype_size

comptime rope_dtype_size = size_of[(load_from_mem rope_dtype_.value())]() if rope_dtype_.__bool__() else Int(0)

scale_dtype_size

comptime scale_dtype_size = size_of[(load_from_mem scale_dtype_.value())]() if scale_dtype_.__bool__() else Int(0)

sm100_smem_carveout

comptime sm100_smem_carveout = (GPUInfo.from_family(AcceleratorArchitectureFamily(Int(32), Int(2048), Int(233472), Int(65536), Int(1024)), StringSpan("B200"), StringSpan("cuda"), StringSpan("blackwell"), SIMD(10), StringSpan("sm_100a"), Int(148)) - Int(1024))

sm100_tmem_cols

comptime sm100_tmem_cols = Int(512)

Methods

__init__

def __init__(*, num_q_heads: Int, group: Int, qk_depth: Int, ov_depth: Int, swizzle_mode: TensorMapSwizzle, page_size: Int) -> Self

BM_eff

def BM_eff(self) -> Int

Number of distinct sequence positions per CTA tile.

When fuse_gqa, each CTA tile covers BM // group seq positions × group heads = BM physical rows.

Returns:

Int

rope_depth

def rope_depth(self) -> Int

Returns:

Int

num_q

def num_q(self) -> Int

Returns:

Int

correction_smem_elements

def correction_smem_elements(self) -> Int

Returns:

Int

num_active_warps_per_group

def num_active_warps_per_group(self) -> Int

Returns:

Int

num_active_threads_per_group

def num_active_threads_per_group(self) -> Int

Returns:

Int

supported

def supported(self) -> Bool

Returns:

Bool

description

def description(self) -> String

Returns:

String