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

grouped_matmul_amd_kernel_launcher

def grouped_matmul_amd_kernel_launcher[c_type: DType, a_type: DType, b_type: DType, LayoutC: TensorLayout, LayoutA: TensorLayout, LayoutB: TensorLayout, AOffsetsLayout: TensorLayout, ExpertIdsLayout: TensorLayout, transpose_b: Bool, config: MatmulConfig[a_type, b_type, c_type, transpose_b], elementwise_lambda_fn: Optional[def[dtype: DType, width: SIMDLength, *, alignment: Int = Int(1)](IndexList[Int(2)], SIMD[dtype, width]) capturing thin -> None] = None](c_tensor: TileTensor[c_type, LayoutC, MutAnyOrigin], a_tensor: TileTensor[a_type, LayoutA, MutAnyOrigin], b_tensor: TileTensor[b_type, LayoutB, MutAnyOrigin], a_offsets: TileTensor[DType.uint32, AOffsetsLayout, ImmUnsafeAnyOrigin], expert_ids: TileTensor[DType.int32, ExpertIdsLayout, ImmUnsafeAnyOrigin], num_active_experts: Int)

Computes the AMD GPU grouped matmul by dispatching per-expert tiles through AMDMatmul, with separate zero-fill handling for inactive (expert_id == -1) blocks.

For active experts, delegates the per-tile matmul (and optional elementwise epilogue) to AMDMatmul. For inactive experts, zeroes the output row range and invokes the epilogue with zero values so that LoRA-style inactive blocks still produce a defined output.