Mojo struct
DecodeSM100PVSS_FP8
struct DecodeSM100PVSS_FP8[operand_type: DType, accum_type: DType, *, config: MLA_SM100_Decode_Config]
Implemented traits
AnyType,
Copyable,
ImplicitlyCopyable,
ImplicitlyDestructible,
Movable,
RegisterPassable,
TrivialRegisterPassable
comptime members
BK
comptime BK = config.BK1
BM
comptime BM = config.BM
BN
comptime BN = DecodeSM100PVSS_FP8[operand_type, accum_type, config=config].MMA_N
MMA_K
comptime MMA_K = 32
MMA_M
comptime MMA_M = config.MMA_M
MMA_N
comptime MMA_N = config.MMA_PV_N
num_k_mmas
comptime num_k_mmas = (DecodeSM100PVSS_FP8[operand_type, accum_type, config=config].BK // 32)
operand_size
comptime operand_size = size_of[operand_type]()
UMMAPVSS
comptime UMMAPVSS = UMMAInsDescriptor.create[accum_type, operand_type, operand_type, Index[Int, Int, dtype=DType.uint32](VariadicPack(DecodeSM100PVSS_FP8[operand_type, accum_type, config=config].MMA_M, DecodeSM100PVSS_FP8[operand_type, accum_type, config=config].MMA_N)), transpose_b=False]()
Methods
descriptor_v_block
static descriptor_v_block(kv_smem: UnsafePointer[Scalar[operand_type], MutAnyOrigin, address_space=AddressSpace.SHARED]) -> MMASmemDescriptorPair
Returns:
descriptor_p_block
static descriptor_p_block(p_smem: UnsafePointer[Scalar[operand_type], MutAnyOrigin, address_space=AddressSpace.SHARED]) -> MMASmemDescriptorPair
Returns:
mma
static mma[*, stage_idx: Int = 0](a: MMASmemDescriptorPair, b: MMASmemDescriptorPair, c: UInt32, *, c_scale: UInt32, elect: Int32)
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!