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 function

mha_sm100_dispatch

def mha_sm100_dispatch[q_type: DType, KVType: MHAOperand, MaskType: MHAMask, output_type: DType, MaxPromptLenType: OptionallyStaticInt, //, config: MHAConfig[config.dtype], group: Int, ragged: Bool, sink: Bool, _is_cache_length_accurate: Bool, pair_cta: Bool](output: DeviceBuffer[output_type], q_arg: Pointer[Scalar[q_type], _safe=False], k: KVType, v: KVType, num_rows_q: Int, mask: MaskType, valid_length: Pointer[UInt32, _safe=False], max_prompt_len_arg: MaxPromptLenType, max_cache_valid_length_arg: Int, scale: Float32, kv_input_row_offsets: OptionalReg[TileTensor[DType.uint32, Layout[*?, *?], ImmutAnyOrigin]], batch_size_arg: Int, ctx: DeviceContext, sink_weights: OptionalReg[TileTensor[q_type, Layout[*?, *?], ImmutAnyOrigin]])

Dispatches the SM100 FA4 flash-attention kernel for a prefill workload.

Selects between the 1Q split-K and 2Q FA4 configurations based on a occupancy and prompt-length heuristic, constructs the Q/K/V/O TMA tile descriptors and transient tile scheduler, threads optional ragged valid-length, KV-row-offset, and sink-attention arguments through to the compiled kernel, and enqueues the launch onto the supplied device context.

Parameters:

q_type (DType): Element type of the query tensor (inferred).
KVType (MHAOperand): Key/value operand descriptor with dtype and page size (inferred).
MaskType (MHAMask): Attention mask scheme applied to the Q@K' scores (inferred).
output_type (DType): Element type of the attention output buffer (inferred).
MaxPromptLenType (OptionallyStaticInt): Optionally-static type encoding the maximum prompt length (inferred).
config (MHAConfig[config.dtype]): MHA configuration supplying dtype, head count, depth, and swizzle mode.
group (Int): Number of query heads per KV head (GQA group size).
ragged (Bool): Whether to dispatch the variable-length valid-length path.
sink (Bool): Whether to thread sink-attention weights into the kernel.
_is_cache_length_accurate (Bool): Whether the supplied cache length is accurate, threaded to the compiled kernel.
pair_cta (Bool): Whether to launch the pair-CTA configuration with a cluster size of 2.

Args:

output (DeviceBuffer[output_type]): Device buffer that receives the attention output rows.
q_arg (Pointer[Scalar[q_type], _safe=False]): Pointer to the query tensor data.
k (KVType): Key operand descriptor.
v (KVType): Value operand descriptor.
num_rows_q (Int): Number of query rows to attend over.
mask (MaskType): Attention mask applied to the Q@K' scores.
valid_length (Pointer[UInt32, _safe=False]): Per-row valid KV length pointer, used when ragged is set.
max_prompt_len_arg (MaxPromptLenType): Maximum prompt length, optionally static.
max_cache_valid_length_arg (Int): Maximum valid KV cache length across the batch.
scale (Float32): Scalar applied to the Q@K' product before softmax.
kv_input_row_offsets (OptionalReg[TileTensor[DType.uint32, Layout[*?, *?], ImmutAnyOrigin]]): Optional per-row KV input offsets for ragged layouts.
batch_size_arg (Int): Number of sequences in the batch.
ctx (DeviceContext): Device context used to build TMA descriptors and enqueue the kernel.
sink_weights (OptionalReg[TileTensor[q_type, Layout[*?, *?], ImmutAnyOrigin]]): Optional sink-attention weights used when sink is set.