IMPORTANT: To view this page as Markdown, append `.md` to the URL (e.g. /get-started.md). For the complete documentation index, see llms.txt.
Skip to main content
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 function

test_matmul_sm90_swapAB_comparison_v2

def test_matmul_sm90_swapAB_comparison_v2[MType: CoordLike, NType: CoordLike, KType: CoordLike, ComputeFnType: ElementwiseComputeFn = inflated$def[comptime[dtype: DType, width: SIMDLength, *, alignment: Int] #kgen.func.symbol<@linalg::@utils::@"identity_compute_fn[::DType,::SIMDLength,::SIMD[DType.int, 1],KGENParamList[::Intable & ::Defaultable & ::RegisterPassable & ::TrivialRegisterPassable & ::Writable & ::AnyType & ::Copyable & ::ImplicitlyCopyable & ::Deinitable & ::Movable & ::CoordLike],::TypeList[::Intable & ::Defaultable & ::RegisterPassable & ::TrivialRegisterPassable & ::Writable & ::AnyType & ::Copyable & ::ImplicitlyCopyable & ::Deinitable & ::Movable & ::CoordLike, $3]](::Coord[$3, $4],::SIMD[$0, $1])"<:!lit.struct<@std::@builtin::@dtype::@DType> dtype, :!lit.struct<@std::@builtin::@simd_length::@SIMDLength> width, :!lit.struct<@std::@simd::@SIMD<:!lit.struct<@std::@builtin::@dtype::@DType> {:dtype index}, :!lit.struct<@std::@builtin::@simd_length::@SIMDLength> {1}>> alignment, :param_list<trait<@std::@builtin::@int::@Intable, @std::@builtin::@value::@Defaultable, @std::@builtin::@value::@RegisterPassable, @std::@builtin::@value::@TrivialRegisterPassable, @std::@format::@__init__::@Writable, @std::@traits::@anytype::@AnyType, @std::@traits::@copyable::@Copyable, @std::@traits::@copyable::@ImplicitlyCopyable, @std::@traits::@deinitable::@Deinitable, @std::@traits::@movable::@Movable, @std::@utils::@coord::@CoordLike>> *?, :!lit.struct<@std::@builtin::@variadics::@TypeList<:meta<!lit.trait<@std::@traits::@anytype::@AnyType>> trait<@std::@builtin::@int::@Intable, @std::@builtin::@value::@Defaultable, @std::@builtin::@value::@RegisterPassable, @std::@builtin::@value::@TrivialRegisterPassable, @std::@format::@__init__::@Writable, @std::@traits::@anytype::@AnyType, @std::@traits::@copyable::@Copyable, @std::@traits::@copyable::@ImplicitlyCopyable, @std::@traits::@deinitable::@Deinitable, @std::@traits::@movable::@Movable, @std::@utils::@coord::@CoordLike>, :param_list<trait<@std::@builtin::@int::@Intable, @std::@builtin::@value::@Defaultable, @std::@builtin::@value::@RegisterPassable, @std::@builtin::@value::@TrivialRegisterPassable, @std::@format::@__init__::@Writable, @std::@traits::@anytype::@AnyType, @std::@traits::@copyable::@Copyable, @std::@traits::@copyable::@ImplicitlyCopyable, @std::@traits::@deinitable::@Deinitable, @std::@traits::@movable::@Movable, @std::@utils::@coord::@CoordLike>> *?>> *?>>], //, a_type: DType, b_type: DType, c_type: DType, BM: Int, BN: Int, BK: Int, MMA_M: Int, MMA_N: Int, MMA_K: Int, num_pipeline_stages: Int, num_consumer: Int, k_group_size: Int = Int(1), num_k_partitions: Int = Int(1), partitioned_multicast: Bool = False, BM_SWAPAB: Int = BM, BN_SWAPAB: Int = BN, BK_SWAPAB: Int = BK, MMA_M_SWAPAB: Int = MMA_M, MMA_N_SWAPAB: Int = MMA_N, MMA_K_SWAPAB: Int = MMA_K, num_pipeline_stages_swapAB: Int = num_pipeline_stages, num_consumer_swapAB: Int = num_consumer, k_group_size_swapAB: Int = k_group_size, num_k_partitions_swapAB: Int = num_k_partitions, partitioned_multicast_swapAB: Bool = partitioned_multicast, use_vendor_reference: Bool = False, default_epilogue: Bool = False, has_compute_fn: Bool = False](ctx: DeviceContext, m: MType, n: NType, k: KType, compute_fn: ComputeFnType = no_compute_fn)

Compare matmul results between normal execution and swapAB execution.

This version accepts config parameters directly as compile-time values and builds configs internally.

Both compute: C[M,N] = A[M,K] @ B[N,K]^T swapAB internally swaps A/B and transposes C on store, but result should match.

Parameters:

  • ​MType (CoordLike): The type of the M dimension.
  • ​NType (CoordLike): The type of the N dimension.
  • ​KType (CoordLike): The type of the K dimension.
  • ​ComputeFnType (ElementwiseComputeFn): Type of compute_fn (inferred).
  • ​a_type (DType): Data type of matrix A.
  • ​b_type (DType): Data type of matrix B.
  • ​c_type (DType): Data type of output matrix C.
  • ​BM (Int): Block tile M dimension for normal kernel.
  • ​BN (Int): Block tile N dimension for normal kernel.
  • ​BK (Int): Block tile K dimension for normal kernel.
  • ​MMA_M (Int): MMA M dimension for normal kernel.
  • ​MMA_N (Int): MMA N dimension for normal kernel.
  • ​MMA_K (Int): MMA K dimension for normal kernel.
  • ​num_pipeline_stages (Int): Number of pipeline stages for normal kernel.
  • ​num_consumer (Int): Number of consumers for normal kernel.
  • ​k_group_size (Int): K group size for normal kernel.
  • ​num_k_partitions (Int): Number of K partitions for normal kernel.
  • ​partitioned_multicast (Bool): Partitioned multicast for normal kernel.
  • ​BM_SWAPAB (Int): Block tile M dimension for swapAB kernel.
  • ​BN_SWAPAB (Int): Block tile N dimension for swapAB kernel.
  • ​BK_SWAPAB (Int): Block tile K dimension for swapAB kernel.
  • ​MMA_M_SWAPAB (Int): MMA M dimension for swapAB kernel.
  • ​MMA_N_SWAPAB (Int): MMA N dimension for swapAB kernel.
  • ​MMA_K_SWAPAB (Int): MMA K dimension for swapAB kernel.
  • ​num_pipeline_stages_swapAB (Int): Number of pipeline stages for swapAB kernel.
  • ​num_consumer_swapAB (Int): Number of consumers for swapAB kernel.
  • ​k_group_size_swapAB (Int): K group size for swapAB kernel.
  • ​num_k_partitions_swapAB (Int): Number of K partitions for swapAB kernel.
  • ​partitioned_multicast_swapAB (Bool): Partitioned multicast for swapAB kernel.
  • ​use_vendor_reference (Bool): If True, use vendor matmul (cuBLAS) as reference instead of normal kernel.
  • ​default_epilogue (Bool): If True, store the output through an epilogue closure instead of the kernel's own store.
  • ​has_compute_fn (Bool): If True, the kernels store compute_fn(idx, a @ b); takes precedence over default_epilogue.

Args:

  • ​ctx (DeviceContext): The device context.
  • ​m (MType): The M dimension (can be static or dynamic).
  • ​n (NType): The N dimension (can be static or dynamic).
  • ​k (KType): The K dimension (can be static or dynamic).
  • ​compute_fn (ComputeFnType): Maps each output index and value to the value to store when has_compute_fn is set.

Was this page helpful?