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 module

kernel_common

Shared kernel components for SM100 warp-specialized matmul kernels.

This module contains common components used by all SM100 matmul kernel variants:

• WarpRole: Warp specialization roles for 4-warp kernels (MMA, Load, Scheduler, Epilogue)
• WarpRole1D1D: Warp specialization roles for 3-warp kernels (MMA, Load, Epilogue)
• KernelContext: Common kernel state (election vars, CTA coords, masks)
• Barrier init helpers: compute_input_consumer_count, init_core_barriers, init_clc_barriers
• _Batched3DLayout / _to_batched_3d: Reshape 2D TileTensor to 3D (batch=1)

comptime values

MbarPtr

comptime MbarPtr = Pointer[SharedMemBarrier, MutUntrackedOrigin, address_space=AddressSpace.SHARED]

Structs

• ​KernelContext: Shared kernel state: election vars, CTA coords, multicast masks, pipeline states.
• ​WarpRole: Warp role identifiers for SM100 warp-specialized kernel.
• ​WarpRole1D1D: Warp role for 1D-1D kernels with warp specialization.

Functions

• ​compute_accum_barrier_counts: Compute accumulator pipeline barrier arrival counts.
• ​compute_clc_barrier_counts: Compute CLC barrier arrival counts.
• ​compute_input_consumer_count: Compute input pipeline barrier consumer count.
• ​compute_tma_tile_dims: Compute TMA tile dimensions (a_tile_dim0, b_tile_dim0, c_tile_dim0).
• ​init_clc_barriers: Initialize CLC full/empty barrier pairs.
• ​init_core_barriers: Initialize input, output, and TMEM deallocation barriers.