[codegen] `NestedLayoutAttr::reshape` changes thread ownership for non-contiguous shape casts

[snarang181]

Summary

Vector layout propagation reshapes a nested_layout through
vector.shape_cast even when one thread's row-major source elements become a
non-contiguous target set. Vector distribution then performs only a local
shape cast and stores a contiguous tile, silently misplacing eight of sixteen elements.

Reproducer

#l = #iree_vector_ext.nested_layout<
 subgroup_tile=[1,1],batch_tile=[1,1],outer_tile=[1,1],
 thread_tile=[2,2],element_tile=[2,2],
 subgroup_strides=[0,0],thread_strides=[1,2]>
#t = #iree_codegen.translation_info<
 pipeline=#iree_codegen.no_pipeline workgroup_size=[4,1,1] subgroup_size=4>
func.func @case(%src: memref<4x4xf32>,%dst: memref<2x8xf32>)
 attributes {translation_info=#t} {
 %c0=arith.constant 0:index
 %p=arith.constant 0.0:f32
 %v=vector.transfer_read %src[%c0,%c0],%p {in_bounds=[true,true]} : memref<4x4xf32>,vector<4x4xf32>
 %x=iree_vector_ext.to_layout %v to layout(#l) : vector<4x4xf32>
 %r=vector.shape_cast %x : vector<4x4xf32> to vector<2x8xf32>
 vector.transfer_write %r,%dst[%c0,%c0] {in_bounds=[true,true]} : vector<2x8xf32>,memref<2x8xf32>
 return
}

Verified with iree-opt at a5ca3e2beaa8 (LLVM 24.0.0git, optimized with assertions):

$ iree-opt '--pass-pipeline=builtin.module(func.func(iree-llvmgpu-vector-distribute))' repro.mlir | grep -E 'affine.linearize_index|vector.shape_cast' | head -7
    %0 = affine.linearize_index disjoint [%thread_id_z, %thread_id_y, %thread_id_x] by (1, 1, 4) : index
    %3 = affine.linearize_index disjoint [%1#2, %c0, %c0, %2#2, %c0] by (1, 1, 1, 2, 2) : index
    %4 = affine.linearize_index disjoint [%1#1, %c0, %c0, %2#1, %c0] by (1, 1, 1, 2, 2) : index
    %7 = vector.shape_cast %6 : vector<1x1x1x1x2x2xf32> to vector<1x1x1x1x1x4xf32>
      %12 = affine.linearize_index disjoint [%8#2, %c0, %c0, %9#2, %c0] by (1, 1, 1, 2, 1) : index
      %13 = affine.linearize_index disjoint [%8#1, %c0, %c0, %9#1, %c0] by (1, 1, 1, 2, 4) : index

Expected thread 2 target columns are {2,3,6,7}; the inferred layout assigns
{4,5,6,7}. For input 0..15, rows become [0,1,4,5,2,3,6,7] and
[8,9,12,13,10,11,14,15] instead of row-major order.

Agent Analysis

At a5ca3e2beaa8, compiler/src/iree/compiler/Codegen/Dialect/VectorExt/IR/ VectorExtAttrs.cpp:105 implements NestedLayoutAttr::reshape; forward
propagation calls it from VectorLayoutAnalysis.cpp:319 without checking
ownership contiguity. Decline the reshape or insert an explicit redistribution.

AI tool use disclosure

Adding this to comply with the project's
AI tool use policy

Assisted-by: Claude Code

Reproducer and root-cause analysis were prepared with LLM assistance and verified manually on the referenced commit.

[github-actions]

Hello @snarang181 👋

Thank you for submitting an issue to IREE! It looks like it's your first one.

The volume of issues is huge, and triage times may vary. If you need quicker assistance on the issue, consider posting it on IREE Discord and providing short context there.
Feel free to identify the most frequent contributors to the corresponding part of the codebase via Git history/CODEOWNERS and (gently!) tag them below.

---

Action required: acknowledge IREE project policies

IREE is a Linux Foundation project. All participants are expected to follow the LF Projects Code of Conduct.

Please note that all contributions to IREE, including GitHub issues, must follow our IREE AI Tool Use Policy. This means:

• Issue reporters must fully understand the primary issue and the impact of the problem.
  For bug reports, specifics of the use-case, workflow and/or crash scenario usually suffice.
  For improvement suggestions, this usually implies deeper understanding of suboptimal code, IR design choices, assumed performance tradeoffs, etc.
• Contributors must write the summary of the issue themselves, so that the exact impact and scenario is clear to other participants.
• Issue authors are expected to review and understand all LLM/AI-assisted segments of the report, if any. E.g.: CLI commands and IR reproducers, model/framework details, lowering pipeline details, assumptions about problematic behaviors of IREE components.
• Substantial use of LLM/generative AI tools must be disclaimed in the header or footer of the issue description.

Before the issue can be triaged, we kindly ask you to reply to this message and confirm that you understand and accept the proposed policy. As required, please adjust your initial issue description.

---

If you have any questions, feel free to leave a comment here, or ask away on IREE Discord.

Thank you,
The IREE Community

[snarang181]

Acknowledged. I understand and accept the IREE AI Tool Use Policy and the LF Code of Conduct. The reproducer and analysis were verified by hand on the stated commit; I've added a note on tool assistance to the description.