IMPORTANT: To view this page as Markdown, append `.md` to the URL (e.g. /get-started.md). For the complete documentation index, see llms.txt.
Skip to main content
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 function

moe_source_scatter_quantize_supported

def moe_source_scatter_quantize_supported(num_cols: Int) -> Bool

Whether the source-major scatter+quantize kernel covers this launch.

The kernel walks contiguous rows and launches one num_cols // 8-thread CTA per source row, so the row width must be a whole number of 32-element MX blocks and the CTA must hold at least one 4-lane scale group and at most 1024 threads. The slot count is unconstrained: the grid is one CTA per source row, so the same kernel covers any fan-out.

Args:

  • ​num_cols (Int): Activation row width K.

Returns:

Bool: Whether _moe_source_scatter_quantize_mx_amd_kernel supports the launch.

Was this page helpful?