Mojo module

mla_decode_utils

comptime values

QOTMATile

comptime QOTMATile[dtype: DType, BM: Int, BK: Int, swizzle_mode: TensorMapSwizzle] = TMATensorTile[dtype, 2, IndexList(BM, BK, __list_literal__=NoneType(None)), _default_desc_shape[2, dtype, IndexList(BM, BK, __list_literal__=NoneType(None)), swizzle_mode]()]

Parameters

• ​dtype (DType):
• ​BM (Int):
• ​BK (Int):
• ​swizzle_mode (TensorMapSwizzle):

ScalesTMATile

comptime ScalesTMATile[BN: Int] = TMATensorTile[DType.float32, 2, IndexList(1, BN, __list_literal__=NoneType(None))]

Parameters

• ​BN (Int):

Structs

• ​DecodeCConsumer:
• ​DecodeCProducer:
• ​DecodeKVConsumer:
• ​DecodeKVProducer:
• ​DecodeOConsumer:
• ​DecodeOProducer:
• ​DecodeOutConsumer:
• ​DecodeOutProducer:
• ​DecodePConsumer:
• ​DecodePConsumerN:
• ​DecodePProducer:
• ​DecodePProducerN:
• ​DecodeSConsumer:
• ​DecodeSConsumerN:
• ​DecodeSM100MiscMBars:
• ​DecodeSM100PVSS:
• ​DecodeSM100PVSS_FP8:
• ​DecodeSM100QKTSS:
• ​DecodeSM100QKTSS_Content_FP8:
• ​DecodeSM100QKTSS_FP8:
• ​DecodeSM100QKTSS_Rope_BF16:
• ​DecodeSM100QKTTS:
• ​DecodeSProducer:
• ​DecodeSProducerN:
• ​KVCvt2MmaConsumer:
• ​KVCvt2MmaProducer:
• ​KVLoad2CvtConsumer:
• ​KVLoad2CvtProducer:
• ​KVPipelineGeneric: KVPipeline has num_kv_stages * num_qk_stages stages. num_kv_stages refers to how many K and V tiles we pipeline for performing the S = Q@K' and O += P@V MMAs. Each of these MMAs is broken up into num_qk_stages pipelined MMAs. We set step=False for all but the last MMA that completes the operation. An alternative implementation would separate the two, and potentially allow for more overall stages at the cost of slightly more bookkeeping.
• ​MLA_Decode_Pack:
• ​MLA_SM100_Decode_Common:
• ​MLA_SM100_Decode_Config:
• ​OffsetPosition:
• ​OutPipeline: OutPipeline has num_out_stages stages. num_out_stages refers to how many output stages we pipeline for performing the output store.

Functions

• ​build_mma_ss_ws:
• ​build_mma_ts_ws: Build PTX for TS (TMEM-A, SMEM-B) .ws MMA over multiple K-tiles.
• ​bulk_mma_ws:
• ​bulk_mma_ws_ts:
• ​clamped_index_coordinate:
• ​cvt_fp8x8_from_2xu32_to_bf16x8_packed_u32x4:
• ​e8m0_to_bf16_broadcast: Convert an e8m0 scale byte to a bf16 value broadcast into both halves of a uint32.
• ​hmul2_bf16x8_by_scalar: Multiply 8 packed bf16 values (in 4 uint32 registers) by a bf16x2 scalar broadcast.
• ​ld_shared_v4_u32:
• ​num_matrix_view_rows_decode: TileTensor overload of num_matrix_view_rows_decode.
• ​st_shared_v4_b32_at_bf16_elem_off:
• ​st_shared_v4_b32_at_fp8_elem_off:
• ​tma_tile_qo:
• ​tma_tile_scales: Create a TMA descriptor for per-token float32 scales.
• ​write_bf16x2_row_to_smem_chunked: Chunked write with optional scaling. Reduces register pressure.
• ​write_fp8_row_to_smem_chunked: Write float32 data to SMEM as FP8 with swizzle for SWIZZLE_64B.

Was this page helpful?

Thank you! We'll create more content like this.

Thank you for helping us improve!

😔 What went wrong?

Some code doesn’t work

It includes inaccurate information

It's missing information I need

It was difficult to understand

Other

Submit