xai-org/x-algorithm · error · ValueError

Not a canonical UMMA_K Layout: Expected stride failure.

Error message

Not a canonical UMMA_K Layout: Expected stride failure.

What it means

In the K-major branch, stride_00 — the stride of the first mode of the M-divided (8,m) hierarchy — must equal swizzle_atom_mn_size (8 for 128B/BASE32B, 4 for 64B, 2 for 32B). This binds the layout's inner M repetition stride to the swizzle period; a mismatch means the declared swizzle does not describe how the layout repeats in M.

Source

Thrown at phoenix/xrex/cutedsl/ranker_fa4/mma_sm100_desc.py:270

        stride_10 = canonical_layout.stride[1][0]
        if stride_10 != swizzle_atom_mn_size:
            raise ValueError("Not a canonical UMMA_MN Layout: Expected stride failure.")
        stride_01, stride_11 = canonical_layout.stride[0][1], canonical_layout.stride[1][1]
        if layout_type is LayoutType.SWIZZLE_NONE:
            stride_byte_offset, leading_byte_offset = stride_01, stride_11
        else:
            stride_byte_offset, leading_byte_offset = stride_11, stride_01
    else:
        if layout_type == LayoutType.SWIZZLE_128B_BASE32B:
            raise ValueError("SWIZZLE_128B_BASE32B is invalid for Major-K")
        if not cute.size(layout.shape[0]) % 8 == 0:
            raise ValueError("Not a canonical UMMA_K Layout: Expected MN-size multiple of 8.")
        canonical_layout = cute.logical_divide(layout, (8, 2))
        if not cute.is_congruent(canonical_layout, ((1, 1), (1, 1))):
            raise ValueError("Not a canonical UMMA_K Layout: Expected profile failure.")
        stride_00 = canonical_layout.stride[0][0]
        if stride_00 != swizzle_atom_mn_size:
            raise ValueError("Not a canonical UMMA_K Layout: Expected stride failure.")
        stride_10 = canonical_layout.stride[1][0]
        if layout_type is not LayoutType.SWIZZLE_NONE and stride_10 != 1:
            raise ValueError("Not a canonical UMMA_K Layout: Expected stride failure.")
        stride_01 = canonical_layout.stride[0][1]
        stride_byte_offset, leading_byte_offset = stride_01, stride_10

    desc = 0
    desc |= (leading_byte_offset & 0x3FFF) << 16
    desc |= (stride_byte_offset & 0x3FFF) << 32
    desc |= (VERSION & 0x3) << 46
    desc |= (BASE_OFFSET & 0x7) << 49
    desc |= (LBO_MODE & 0x1) << 52
    desc |= (int(layout_type) & 0x7) << 61

    return desc & 0xFFFF_FFFF_FFFF_FFFF


def make_smem_desc_start_addr(start_addr: cute.Pointer) -> cutlass.Int32:

View on GitHub (pinned to 24c60942c5)

Solutions

1. Align the swizzle with the layout's M atom stride using the swizzle_atom_mn_size table (NONE:1, 32B:2, 64B:4, 128B:8)
2. Rebuild the layout via tile_to_shape from the same swizzle atom used in the descriptor call
3. Log canonical_layout.stride[0][0] to see the actual value and adjust one side

Example fix

# before
desc = make_smem_desc_base(layout_with_m_stride_8, cute.Swizzle(2,4,3), Major.K)  # 64B
# after
desc = make_smem_desc_base(layout_with_m_stride_4, cute.Swizzle(2,4,3), Major.K)

expected = {0:1,1:8,2:8,4:4,6:2}[layout_type]
assert cute.logical_divide(layout,(8,2)).stride[0][0] == expected

When it happens

Trigger: Passing a 64B swizzle (atom 4) with a layout tiled with M-stride 8, or any K-major layout whose stride_00 differs from the table entry for the swizzle's LayoutType.

Common situations: Changing dtype (fp16 -> fp8 changes the natural atom width) without changing the swizzle constant; reusing a descriptor helper with a layout built for a different swizzle atom.

Related errors

• Not a canonical UMMA_MN Layout: Expected stride failure.
• 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 MN-size multiple of

AI-assisted analysis of xai-org/x-algorithm@24c60942c5 (2026-08-28). Data as JSON: /api/errors/f2c78e02b2c0fb6f. Report an issue: GitHub.