Skip to main content

/

DocsCode

Mojo struct

BlackwellBlockwiseFP8MatmulKernel

struct BlackwellBlockwiseFP8MatmulKernel[a_type: DType, b_type: DType, c_type: DType, a_scales_type: DType, b_scales_type: DType, b_scales_layout: TensorLayout, transpose_b: Bool, config: MatmulConfig[a_type, b_type, c_type, transpose_b], cluster_shape: StaticTuple[Int32, 3] = StaticTuple(1)]

Blockwise FP8 matmul kernel with register-based accumulation.

This kernel implements per-K-iteration scaling in CUDA cores:

  1. Load warp: TMA loads A, B, A-scales to SMEM
  2. MMA warp: Standard MMA (partial to TMEM)
  3. Epilogue warp: TMEM read → scale → register accumulate → output

Implemented traits​

AnyType, ImplicitlyDestructible

comptime members​

__del__is_trivial​

comptime __del__is_trivial = True

a_expected_bytes​

comptime a_expected_bytes = (BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].a_smem_layout.size() * size_of[a_type]())

a_scales_expected_bytes​

comptime a_scales_expected_bytes = (BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].a_scales_smem_layout.size() * size_of[a_scales_type]())

a_scales_smem_layout​

comptime a_scales_smem_layout = Layout.row_major(1, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].BM)

a_smem_layout​

comptime a_smem_layout = tile_layout_k_major[a_type, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].BM, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].BK, config.a_swizzle]()

a_swizzle_elems​

comptime a_swizzle_elems = (config.a_swizzle.bytes() // size_of[a_type]())

a_tile_dim0​

comptime a_tile_dim0 = (BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].BM // BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].CLUSTER_N)

a_tma_load_size​

comptime a_tma_load_size = (BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].a_tile_dim0 * BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].a_swizzle_elems)

a_tma_rows​

comptime a_tma_rows = BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].a_tile_dim0

accum_dims​

comptime accum_dims = get_accumulator_dims[c_smem_dim1=BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].OutputN, block_tile_shape=config.block_tile_shape, mma_shape=config.mma_shape, cta_group=BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].cta_group]()

accum_pipeline_consumer_arv_count​

comptime accum_pipeline_consumer_arv_count = (BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].cta_group * BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].EPILOGUE_THREADS)

accum_pipeline_producer_arv_count​

comptime accum_pipeline_producer_arv_count = 1

accum_type​

comptime accum_type = DType.float32

AccumTensor​

comptime AccumTensor = TmemTensor[DType.float32, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].tmem_accum_layout, cta_group=BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].cta_group]

Accumulator​

comptime Accumulator = BlockwiseFP8Accumulator[DType.float32, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].accum_dims.__getitem__[Int](0), BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].accum_dims.__getitem__[Int](1), BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].is_lower_required, config.block_tile_shape, config.mma_shape, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].CLUSTER_SIZE]

ADescLayout​

comptime ADescLayout = Layout[ComptimeInt[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].a_tile_dim0], ComptimeInt[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].a_swizzle_elems], ComptimeInt[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].a_swizzle_elems], ComptimeInt[1]]

AScalesLayout​

comptime AScalesLayout = Layout[ComptimeInt[1], ComptimeInt[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].BM], ComptimeInt[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].BM], ComptimeInt[1]]

AScalesTmaOp​

comptime AScalesTmaOp = BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].AScalesTmaTile.InnerType

AScalesTmaTile​

comptime AScalesTmaTile = TMATile[a_scales_type, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].AScalesLayout, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].AScalesLayout]

ATileLayout​

comptime ATileLayout = Layout[ComptimeInt[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].a_tile_dim0], ComptimeInt[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].BK], ComptimeInt[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].BK], ComptimeInt[1]]

ATmaOp​

comptime ATmaOp = BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].ATmaTile.InnerType

ATmaTile​

comptime ATmaTile = TMATile[a_type, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].ATileLayout, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].ADescLayout]

b_expected_bytes​

comptime b_expected_bytes = (BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].b_smem_layout.size() * size_of[b_type]())

b_smem_layout​

comptime b_smem_layout = tile_layout_k_major[b_type, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].BN, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].BK, config.b_swizzle]() if transpose_b else tile_layout_mn_major[b_type, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].BN, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].BK, config.b_swizzle]()

b_swizzle_elems​

comptime b_swizzle_elems = (config.b_swizzle.bytes() // size_of[b_type]())

b_tile_dim0​

comptime b_tile_dim0 = (BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].BN // (BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].CLUSTER_M // BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].cta_group))

b_tma_load_size​

comptime b_tma_load_size = (BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].b_tile_dim0 * BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].b_swizzle_elems)

b_tma_rows​

comptime b_tma_rows = BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].b_tile_dim0

BDescLayout​

comptime BDescLayout = Layout[ComptimeInt[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].b_tile_dim0], ComptimeInt[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].b_swizzle_elems], ComptimeInt[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].b_swizzle_elems], ComptimeInt[1]]

BK​

comptime BK = config.block_tile_shape.__getitem__[Int](2)

BM​

comptime BM = config.block_tile_shape.__getitem__[Int](0)

BN​

comptime BN = config.block_tile_shape.__getitem__[Int](1)

BScalesTile​

comptime BScalesTile = TileTensor[b_scales_type, b_scales_layout, ImmutAnyOrigin]

BTileLayout​

comptime BTileLayout = Layout[ComptimeInt[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].b_tile_dim0], ComptimeInt[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].BK], ComptimeInt[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].BK], ComptimeInt[1]]

BTmaOp​

comptime BTmaOp = BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].BTmaTile.InnerType

BTmaTile​

comptime BTmaTile = TMATile[b_type, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].BTileLayout, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].BDescLayout]

c_smem_layout​

comptime c_smem_layout = Layout.row_major(BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].OutputM, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].OutputN)

c_swizzle_elems​

comptime c_swizzle_elems = (config.c_swizzle.bytes() // size_of[c_type]())

c_tile_dim0​

comptime c_tile_dim0 = BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].OutputM if (BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].MMA_M == 256) if (BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].MMA_M == 256)._mlir_value else (BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].cta_group == 1) else 64

CDescLayout​

comptime CDescLayout = Layout[ComptimeInt[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].c_tile_dim0], ComptimeInt[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].c_swizzle_elems], ComptimeInt[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].c_swizzle_elems], ComptimeInt[1]]

clc_consumer_arv_count​

comptime clc_consumer_arv_count = (BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].SCHEDULER_THREADS + (BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].CLUSTER_SIZE * ((BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].TMA_LOAD_THREADS + BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].MMA_THREADS) + BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].EPILOGUE_THREADS)))

clc_producer_arv_count​

comptime clc_producer_arv_count = 1

clc_throttle_consumer_arv_count​

comptime clc_throttle_consumer_arv_count = BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].SCHEDULER_THREADS

clc_throttle_producer_arv_count​

comptime clc_throttle_producer_arv_count = BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].TMA_LOAD_THREADS

CLUSTER_M​

comptime CLUSTER_M = config.cluster_shape.__getitem__[Int](0)

CLUSTER_N​

comptime CLUSTER_N = config.cluster_shape.__getitem__[Int](1)

CLUSTER_SIZE​

comptime CLUSTER_SIZE = (BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].CLUSTER_M * BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].CLUSTER_N)

Context​

comptime Context = KernelContext[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].num_clc_pipeline_stages, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].cta_group, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].CLUSTER_M, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].CLUSTER_N]

cta_group​

comptime cta_group = config.cta_group

CTileLayout​

comptime CTileLayout = Layout[ComptimeInt[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].c_tile_dim0], ComptimeInt[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].OutputN], ComptimeInt[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].OutputN], ComptimeInt[1]]

CTmaOp​

comptime CTmaOp = BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].CTmaTile.InnerType

CTmaTile​

comptime CTmaTile = TMATile[c_type, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].CTileLayout, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].CDescLayout]

EPILOGUE_THREADS​

comptime EPILOGUE_THREADS = (4 * WARP_SIZE)

EpilogueCtx​

comptime EpilogueCtx = EpilogueWarpContext[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].num_accum_pipeline_stages, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].stage_stride_cols, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].cta_group, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].MMA_THREADS, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].EPILOGUE_THREADS]

EpilogueHandle​

comptime EpilogueHandle = EpilogueWarp[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].num_accum_pipeline_stages, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].stage_stride_cols, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].cta_group, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].MMA_THREADS, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].EPILOGUE_THREADS]

input_expected_bytes​

comptime input_expected_bytes = (BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].cta_group * ((BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].a_expected_bytes + BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].b_expected_bytes) + BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].a_scales_expected_bytes))

InputTilePipeline​

comptime InputTilePipeline = InputTilePipeline[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].TilePayload, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].SmemType.num_group_pipeline_stages, config.k_group_size]

is_lower_required​

comptime is_lower_required = is_lower_fragment_required[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].cta_group, config.block_tile_shape]()

max_tmem_cols​

comptime max_tmem_cols = 512

MMA_K​

comptime MMA_K = config.mma_shape.__getitem__[Int](2)

MMA_M​

comptime MMA_M = config.mma_shape.__getitem__[Int](0)

MMA_N​

comptime MMA_N = config.mma_shape.__getitem__[Int](1)

MMA_THREADS​

comptime MMA_THREADS = WARP_SIZE

MmaCtx​

comptime MmaCtx = MmaWarpContext[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].num_accum_pipeline_stages, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].stage_stride_cols, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].cta_group, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].MMA_THREADS, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].EPILOGUE_THREADS]

MmaHandle​

comptime MmaHandle = MmaWarp[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].num_accum_pipeline_stages, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].stage_stride_cols, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].cta_group, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].MMA_THREADS, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].EPILOGUE_THREADS]

MmaOp​

comptime MmaOp = MmaOpSM100_SS[c_type, a_type, b_type, config.block_tile_shape, config.mma_shape, cta_group=BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].cta_group, cluster_shape=config.cluster_shape, a_swizzle=config.a_swizzle, b_swizzle=config.b_swizzle, transpose_b=transpose_b]

num_accum_pipeline_stages​

comptime num_accum_pipeline_stages = config.num_accum_pipeline_stages

num_clc_pipeline_stages​

comptime num_clc_pipeline_stages = config.num_clc_pipeline_stages

num_group_pipeline_stages​

comptime num_group_pipeline_stages = (BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].num_pipeline_stages // config)

num_output_stages​

comptime num_output_stages = config.num_output_stages

num_output_warps​

comptime num_output_warps = 4

num_pipeline_stages​

comptime num_pipeline_stages = config.num_pipeline_stages

NUM_THREADS​

comptime NUM_THREADS = (((BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].SCHEDULER_THREADS + BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].TMA_LOAD_THREADS) + BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].MMA_THREADS) + BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].EPILOGUE_THREADS)

NUM_TMEM_COLS​

comptime NUM_TMEM_COLS = 512

OutputM​

comptime OutputM = config.output_tile_shape.__getitem__[Int](0)

OutputN​

comptime OutputN = config.output_tile_shape.__getitem__[Int](1)

OutputPipeline​

comptime OutputPipeline = OutputTilePipeline[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].num_accum_pipeline_stages, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].stage_stride_cols, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].cta_group]

Scheduler​

comptime Scheduler = TileScheduler[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].num_clc_pipeline_stages, Index[dtype=DType.uint32](config.cluster_shape.__getitem__[Int](0), config.cluster_shape.__getitem__[Int](1), config.cluster_shape.__getitem__[Int](2)), config.raster_order, config.block_swizzle_size]

SCHEDULER_THREADS​

comptime SCHEDULER_THREADS = WARP_SIZE

SmemType​

comptime SmemType = BlockwiseFP8Smem[a_type, b_type, c_type, a_scales_type, transpose_b, config=config]

stage_stride_cols​

comptime stage_stride_cols = (512 // config)

TilePayload​

comptime TilePayload = BlockwiseFP8TilePayload[a_type, b_type, a_scales_type, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].SmemType.BM, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].SmemType.BK, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].SmemType.BN, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].SmemType.BK, 1, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].SmemType.BM, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].SmemType.num_pipeline_stages]

TileWriterType​

comptime TileWriterType = BlockwiseFP8TileWriter[c_type, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].OutputM, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].OutputN, DType.float32, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].accum_dims.__getitem__[Int](0), BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].accum_dims.__getitem__[Int](1), block_tile_shape=config.block_tile_shape, mma_shape=config.mma_shape, is_lower_frag_required=BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].is_lower_required, cta_group=BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].cta_group, num_output_stages=BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].num_output_stages, num_output_warps=BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].num_output_warps, c_swizzle=config.c_swizzle]

TMA_LOAD_THREADS​

comptime TMA_LOAD_THREADS = WARP_SIZE

Tmem​

comptime Tmem = TmemAllocation[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].cta_group]

tmem_accum_layout​

comptime tmem_accum_layout = Layout.row_major(BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].MMA_M, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].MMA_N)

TmemDealloc​

comptime TmemDealloc = TmemDeallocBarrier[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].cta_group]

Methods​

load_input_tiles​

static load_input_tiles[a_tma_origin: ImmutOrigin, b_tma_origin: ImmutOrigin, a_scales_tma_origin: ImmutOrigin, tiles_origin: MutOrigin, //](a_loader: TileLoader[a_tma_origin, a_type, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].ATileLayout, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].ADescLayout, cta_group=BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].cta_group], b_loader: TileLoader[b_tma_origin, b_type, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].BTileLayout, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].BDescLayout, cta_group=BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].cta_group], a_scales_loader: ScalesLoader[a_scales_tma_origin, a_scales_type, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].AScalesLayout, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].AScalesLayout, cta_group=BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].cta_group], tiles: InputProducerStage[tiles_origin, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].TilePayload, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].SmemType.num_group_pipeline_stages, config.k_group_size], peer_cta_coord: Tuple[UInt, UInt, UInt], work_tile_coord: Tuple[UInt, UInt], iter_idx: Scalar[DType.uint], elect_one_cta: Bool)

Load A, B, and A-scales tiles using TMA.

Args:

  • ​a_loader (TileLoader): TileLoader for A matrix.
  • ​b_loader (TileLoader): TileLoader for B matrix.
  • ​a_scales_loader (ScalesLoader): ScalesLoader for A-scales.
  • ​tiles (InputProducerStage): InputProducerStage context with encapsulated tile access.
  • ​peer_cta_coord (Tuple): Peer CTA coordinates for multicast.
  • ​work_tile_coord (Tuple): Current work tile M/N coordinates.
  • ​iter_idx (Scalar): K iteration index.
  • ​elect_one_cta (Bool): Whether this is the elected CTA in the cluster.

mma​

static mma[tiles_origin: MutOrigin, //](tiles: InputConsumerStage[tiles_origin, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].TilePayload, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].SmemType.num_group_pipeline_stages, config.k_group_size], mma_op: MmaOpSM100_SS[c_type, a_type, b_type, config.block_tile_shape, config.mma_shape, cta_group=BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].cta_group, cluster_shape=config.cluster_shape, a_swizzle=config.a_swizzle, b_swizzle=config.b_swizzle, transpose_b=transpose_b], accum_tensor: TmemTensor[DType.float32, BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].tmem_accum_layout, cta_group=BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].cta_group])

Execute standard MMA operations (partial results to TMEM).

For blockwise FP8, each K iteration writes a fresh partial to TMEM. The epilogue accumulates across K in registers, not TMEM. Therefore init_c is always True (unlike standard matmul).

Args:

  • ​tiles (InputConsumerStage): Input consumer stage with A, B, A-scales tiles.
  • ​mma_op (MmaOpSM100_SS): The MMA operator.
  • ​accum_tensor (TmemTensor): Typed TMEM tensor view for the accumulator stage.

validate_config​

static validate_config()

Validate configuration constraints at compile time.

run​

static run(a_tma_op: TMATensorTile[a_type, _to_legacy_layout[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].ATileLayout](), _to_legacy_layout[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].ADescLayout]()], b_tma_op: TMATensorTile[b_type, _to_legacy_layout[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].BTileLayout](), _to_legacy_layout[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].BDescLayout]()], c_tma_op: TMATensorTile[c_type, _to_legacy_layout[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].CTileLayout](), _to_legacy_layout[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].CDescLayout]()], a_scales_tma_op: TMATensorTile[a_scales_type, _to_legacy_layout[BlackwellBlockwiseFP8MatmulKernel[a_type, b_type, c_type, a_scales_type, b_scales_type, b_scales_layout, transpose_b, config, cluster_shape].AScalesLayout]()], cluster_dim: StaticTuple[Int32, 3], num_iters: Scalar[DType.uint], b_scales: TileTensor[b_scales_type, b_scales_layout, ImmutAnyOrigin], problem_shape: StaticTuple[Int32, 3])

Kernel entry point for blockwise FP8 matmul.

View source

View source

Was this page helpful?

Models

Cookbook

Puzzles

Blog

Community

Contact