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 module
dispatch
Dispatch entry points for SM100 (B200+) structured matmul kernels.
Selects between GEMV, split-K GEMV, small-MN GEMMs, heuristic outlier configs, and the warp-specialized TMA/UMMA tile GEMM based on problem shape and dtype, falling back to vendor BLAS only for untuned or low-performance shapes.
comptime values
DISPATCH_HIT
comptime DISPATCH_HIT = 1
DISPATCH_MISS
comptime DISPATCH_MISS = 0
logger
comptime logger = Logger(stdout, prefix=String(""), source_location=False)
Functions
-
dispatch_gemv: Dispatch M=1 (or N=1) matmul to GEMV or SM100 GEMM based on (N, K). -
dispatch_sm100_batched_matmul: Dispatch batched matmul to SM100 kernel. -
heuristic_and_outliers_dispatch: Dispatches an SM100 matmul through the heuristic outlier config set. -
matmul_dispatch_sm100: Dispatches a 2D matmul to the appropriate SM100 (B200+) kernel. -
matmul_dispatch_sm100_bf16: Dispatches a bfloat16-input SM100 matmul to a tuned or heuristic config. -
matmul_dispatch_sm100_fp32: Dispatches a float32 SM100 matmul via the heuristic outlier dispatch. -
matmul_dispatch_sm100_fp8: Dispatches an FP8-input SM100 matmul to a tuned or heuristic config. -
select_and_launch_sm100_config: Selects and launches an SM100 matmul config for the given shape. -
sm100_heuristic_and_outliers_dispatch: Dispatches an SM100 matmul through the heuristic outlier config set. -
small_MN_gemms: Launches a small-MN GEMM via the configured split-K GEMV or MMA-CPasync kernel.
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!