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

msa_sm100_accum

MSA-private copy of the SM100 tcgen05 accumulator / operand machinery.

Duplicate of the dense SM100 MHA accumulator types from nn.attention.gpu.nvidia.sm100.mha_1q, lifted here so the block-major forward (msa_2q.mojo) can carry the num_m_mmas == 2 ping-pong without dragging the dense MHA / MLA decode kernels through their full test matrix.

Structs

• ​MMAOperandOffsetFn:
• ​MSASM100TensorAccumulatorSS:
• ​MSASM100TensorAccumulatorTS:
• ​RegisterAccumulatorDescription:
• ​RegisterAccumulatorLayout:
• ​TMemAccumulator:
• ​TMemOperand:
• ​UMMADescriptorSS:
• ​UMMADescriptorTS:

Traits

• ​AccumulatorTile:
• ​DescriptorPair:
• ​DescriptorPairTS:
• ​WriteableMMAOperandDescriptor:

Functions

• ​local_tensor_type:

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