Invalid IR produced for MetalLayoutAttr

[brnorris03]

Describe the bug

The #ttcore.metal_layout attribute printer outputs an empty memory_layout field, producing l1,, which fails to parse back.

Impact

• Cannot incrementally run pipeline stages using intermediate files
• Debugging requires running the full pipeline rather than individual passes

Workaround

Use the full pipeline (--ttir-to-ttmetal-pipeline) instead of piping intermediate outputs between frontend/middleend/backend stages.

Steps to reproduce the issue

Reproducer 1: Simple add operation

Create /tmp/simple_add.mlir:

func.func @add(%arg0: tensor<32x32xf32>, %arg1: tensor<32x32xf32>) -> tensor<32x32xf32> {
  %0 = ttir.empty() : tensor<32x32xf32>
  %1 = "ttir.add"(%arg0, %arg1, %0) : (tensor<32x32xf32>, tensor<32x32xf32>, tensor<32x32xf32>) -> tensor<32x32xf32>
  return %1 : tensor<32x32xf32>
}

Run:

# Run frontend pipeline and save output
build/bin/ttmlir-opt --ttir-to-ttmetal-fe-pipeline="system-desc-path=" \
  /tmp/simple_add.mlir -o /tmp/simple_add_fe.mlir

# Attempt to parse the output (FAILS)
build/bin/ttmlir-opt /tmp/simple_add_fe.mlir

Reproducer 2: Reduction test

# Run frontend pipeline and save output
build/bin/ttmlir-opt --ttir-to-ttmetal-fe-pipeline="system-desc-path=" \
  test/ttmlir/Dialect/D2M/Transforms/dst_graph_coloring/debug_reduction_dst.mlir \
  -o /tmp/reduction_fe.mlir

# Attempt to parse the output (FAILS)
build/bin/ttmlir-opt /tmp/reduction_fe.mlir

Error Message

/tmp/reduction_fe.mlir:1:156: error: expected keyword for TTCore Tensor Memory Layout
#layout = #ttcore.metal_layout<logical_shape = 256x256, dim_alignments = 32x32, collapsed_intervals = dense<[[0, 1], [1, 2]]> : tensor<2x2xi64>, undef, l1,, index_map = map(0)>
                                                                                                                                                           ^
/tmp/reduction_fe.mlir:1:156: error: failed to parse TTCore_MetalLayoutAttr parameter 'memory_layout' which is to be a `TensorMemoryLayout`

Analysis

The printed output shows l1,, (note the double comma) indicating the memory_layout field is empty/unset. The parser expects a keyword for TensorMemoryLayout but finds nothing.

Expected behavior

Either:

1. The printer should output a valid default value when memory_layout is unset
2. The attribute should be defined to always have a valid memory_layout
3. The parser should handle the case when memory_layout is empty

Please complete the following environment information

reproducible anywhere