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

grouped_matmul_sm100_1d1d

Provides block-scaled grouped GEMM for 1D×1D scale layouts on SM100 (B200) GPUs.

`comptime` values

comptime WarpRole = WarpRole[False]

B200BlockScaledMatmulSmem: Holds the shared memory buffers and barriers for one block-scaled matmul tile.

blackwell_block_scaled_matmul_tma_umma_warp_specialized: Launch grouped block-scaled matmul kernel on SM100.
blackwell_block_scaled_tma_umma_warp_specialized_kernel:
consumer_main_loop: Consumes one pipeline stage and issues the block-scaled MMA into TMEM.
copy_accum_to_gmem: Copies one accumulator stage from TMEM to global memory via shared memory.
grouped_matmul_dynamic_scaled_nvfp4: Performs grouped matrix multiplication with NVFP4 quantization.
load_AB: Issues multicast TMA loads for A, B, and their scale factors into shared memory.
multi_stage_store_C: Drains one MMA output tile from TMEM into global memory in stages.