Skip to main content

Mojo struct

QKMMAOp

struct QKMMAOp[dtype: DType, accum_dtype: DType, config: MLASparseConfig[config.qkv_dtype]]

Implemented traits​

AnyType, ImplicitlyDestructible

comptime members​

NUM_TS_STAGES​

comptime NUM_TS_STAGES = 3

SSMMAType​

comptime SSMMAType = SM100TensorAccumulatorSS[dtype, accum_dtype, config.num_q_heads, 128, 192, cta_group=2]

TSMMAType​

comptime TSMMAType = SM100TensorAccumulatorTS[dtype, accum_dtype, config.num_q_heads, 128, 384, cta_group=2, num_stages=3]

Methods​

smem_descriptor_q​

static smem_descriptor_q(q_smem: UnsafePointer[Scalar[dtype], address_space=AddressSpace.SHARED]) -> MMASmemDescriptorPair

Returns:

MMASmemDescriptorPair

tmem_descriptor_q​

static tmem_descriptor_q(q_smem: UnsafePointer[Scalar[dtype], address_space=AddressSpace.SHARED]) -> MMASmemDescriptorPair

Returns:

MMASmemDescriptorPair

descriptor_k_p0​

static descriptor_k_p0(k_smem: UnsafePointer[Scalar[dtype], address_space=AddressSpace.SHARED]) -> MMASmemDescriptorPair

Returns:

MMASmemDescriptorPair

descriptor_k_p1​

static descriptor_k_p1(k_smem: UnsafePointer[Scalar[dtype], address_space=AddressSpace.SHARED]) -> MMASmemDescriptorPair

Returns:

MMASmemDescriptorPair