sgl-project/sglang · error · ValueError

Only Swizzle<2,5,2> supported for 128B_BASE32B

Error message

Only Swizzle<2,5,2> supported for 128B_BASE32B

What it means

_layout_type only accepts Swizzle<2,5,2> for the num_base==5 case, which maps to LayoutType.SWIZZLE_128B_BASE32B. Any (B, S) other than (2, 2) with M==5 raises this ValueError.

Source

Thrown at python/sglang/kernels/ops/attention/flash_attn/cute/mma_sm100_desc.py:217

def _layout_type(swizzle: cute.Swizzle) -> LayoutType:
    B, M, S = swizzle.num_bits, swizzle.num_base, swizzle.num_shift

    if M == 4:  # Swizzle<*,4,3>
        if S != 3:
            raise ValueError("Unexpected swizzle shift – want S==3 for M==4")
        return {
            0: LayoutType.SWIZZLE_NONE,
            1: LayoutType.SWIZZLE_32B,
            2: LayoutType.SWIZZLE_64B,
            3: LayoutType.SWIZZLE_128B,
        }[
            B
        ]  # KeyError ⇒ invalid B→ raise
    if M == 5:  # Swizzle<2,5,2> (the only legal triple for M==5)
        if (B, S) != (2, 2):
            raise ValueError("Only Swizzle<2,5,2> supported for 128B_BASE32B")
        return LayoutType.SWIZZLE_128B_BASE32B

    # Any other (M,B,S) triple is not a UMMA-legal shared-memory layout
    raise ValueError("Unsupported swizzle triple for UMMA smem descriptor")


def make_smem_desc_base(
    layout: cute.Layout, swizzle: cute.Swizzle, major: Major
) -> int:
    """
    Convert a 2-D *shared-memory* Cute layout into the Blackwell 64-bit
    smem-descriptor, without the smem start address.
    layout must correspond to layout of an uint128 tensor.
    """
    # ------------------------------------------------------------------ meta
    layout_type = _layout_type(swizzle)  # resolve SWIZZLE_* family

    VERSION = 1  # bits 46–47

View on GitHub (pinned to 0132848349)

Solutions

1. Use exactly Swizzle<2,5,2> for the 128B_BASE32B layout.
2. Take the swizzle from a canonical cute atom (the layouts used by the shipped kernels) instead of constructing it manually.
3. If you don't need base32B, use a Swizzle<*,4,3> triple instead.

if swizzle.num_base == 5:
    assert (swizzle.num_bits, swizzle.num_shift) == (2, 2)

When it happens

Trigger: Calling make_smem_desc_base with a Swizzle whose num_base is 5 but whose num_bits/num_shift differ from 2/2 — e.g. Swizzle<3,5,2> or Swizzle<2,5,3>.

Common situations: Building 128B-base32B layouts manually with slightly-off parameters; mixing swizzle constants copied from a different atom.

Related errors

• Unexpected swizzle shift – want S==3 for M==4
• Unsupported swizzle triple for UMMA smem descriptor
• Not a canonical UMMA_MN Layout: Expected profile failure.
• SWIZZLE_128B_BASE32B is invalid for Major-K
• Not a canonical UMMA_K Layout: Expected profile failure.

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/5dd7992dfc253912. Report an issue: GitHub.