assertion failure at mlir::triton::ColumnAction::apply in triton/lib/Tools/LinearLayout.cpp

### Describe the bug The following ttgir ```ttgir #blocked = #ttg.blocked<{sizePerThread = [4], threadsPerWarp = [32], warpsPerCTA = [4], order = [0]}> module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 4 : i32, ttg.target = "cuda:75", "ttg.threads-per-warp" = 32 : i32} { tt.func public @change_value(%X: !tt.ptr<f32> {tt.divisibility = 16 : i32}) attributes {noinline = false} { %c2_i32 = arith.constant 2 : i32 %cst = arith.constant dense<0.000000e+00> : tensor<2xf32, #blocked> %cst_0 = arith.constant dense<2.000000e+00> : tensor<2xf32, #blocked> %pid = tt.get_program_id x : i32 %block_start = arith.muli %pid, %c2_i32 : i32 %offsets = tt.make_range {end = 2 : i32, start = 0 : i32} : tensor<2xi32, #blocked> %offsets_1 = tt.splat %block_start : i32 -> tensor<2xi32, #blocked> %offsets_2 = arith.addi %offsets_1, %offsets : tensor<2xi32, #blocked> %0 = tt.splat %X : !tt.ptr<f32> -> tensor<2x!tt.ptr<f32>, #blocked> %1 = tt.addptr %0, %offsets_2 : tensor<2x!tt.ptr<f32>, #blocked>, tensor<2xi32, #blocked> %2 = tt.atomic_cas acquire, gpu, %1, %cst, %cst_0 : (tensor<2x!tt.ptr<f32>, #blocked>, tensor<2xf32, #blocked>, tensor<2xf32, #blocked>) -> tensor<2xf32, #blocked> tt.return } } ``` output: ``` root@d306a6352cb4:~# /workspace/triton/build/cmake.linux-x86_64-cpython-3.10/bin/triton-opt ./test.ttgir -convert-triton-gpu-to-llvm triton-opt: /workspace/triton/lib/Tools/LinearLayout.cpp:1331: llvm::SmallVector<mlir::Value> mlir::triton::ColumnAction::apply(mlir::ValueRange) const: Assertion `values.size() == (1 << inSizeLog2) && "Values have a different size than the ColumnAction"' failed. PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace and instructions to reproduce the bug. Stack dump: 0. Program arguments: /workspace/triton/build/cmake.linux-x86_64-cpython-3.10/bin/triton-opt ./test.ttgir -convert-triton-gpu-to-llvm Stack dump without symbol names (ensure you have llvm-symbolizer in your PATH or set the environment var `LLVM_SYMBOLIZER_PATH` to point to it): 0 triton-opt 0x000055e5a1459de7 1 triton-opt 0x000055e5a1456dce 2 triton-opt 0x000055e5a145acd5 3 libc.so.6 0x00007f55e0ab7520 4 libc.so.6 0x00007f55e0b0b9fc pthread_kill + 300 5 libc.so.6 0x00007f55e0ab7476 raise + 22 6 libc.so.6 0x00007f55e0a9d7f3 abort + 211 7 libc.so.6 0x00007f55e0a9d71b 8 libc.so.6 0x00007f55e0aaee96 9 triton-opt 0x000055e59961cf64 10 triton-opt 0x000055e5997fcb25 11 triton-opt 0x000055e5997fdffd 12 triton-opt 0x000055e599b397f6 13 triton-opt 0x000055e599b47cb5 14 triton-opt 0x000055e599b47e1f 15 triton-opt 0x000055e5a0786b54 16 triton-opt 0x000055e5a07e2382 17 triton-opt 0x000055e5a07dea54 18 triton-opt 0x000055e5a0788a09 19 triton-opt 0x000055e5a07881c5 20 triton-opt 0x000055e5a07895be 21 triton-opt 0x000055e5a0798858 22 triton-opt 0x000055e5a078f035 23 triton-opt 0x000055e5a078f10b 24 triton-opt 0x000055e599b57679 25 triton-opt 0x000055e599e75e3b 26 triton-opt 0x000055e599e76a39 27 triton-opt 0x000055e599e7dba2 28 triton-opt 0x000055e599e7d3c2 29 triton-opt 0x000055e599e67507 30 triton-opt 0x000055e599e6679e 31 triton-opt 0x000055e5a107c778 32 triton-opt 0x000055e599e5c632 33 triton-opt 0x000055e599e5ca2c 34 triton-opt 0x000055e599e5cc6d 35 triton-opt 0x000055e599359283 36 libc.so.6 0x00007f55e0a9ed90 37 libc.so.6 0x00007f55e0a9ee40 __libc_start_main + 128 38 triton-opt 0x000055e5993ddef5 Aborted (core dumped) ``` To reproduce: ``` triton-opt ./test.ttgir -convert-triton-gpu-to-llvm ``` ### Environment details Triton: main @ https://github.com/triton-lang/triton/commit/e3fe871f2f709d1f48344fbc06974877afbf7416 (2026-07-14), triton.version == 3.8.0 GPU: CUDA, compute capability 7.5 (Turing)