[Mlir-commits] [mlir] [mlir][tosa] Lower ROW_GATHER to Linalg (PR #225417)

Ingo Müller llvmlistbot at llvm.org
Tue Oct 6 04:45:18 PDT 2026


ingomueller-net wrote:

This PR seems to break the tests. Weirdly, they didn't seem to be executed either before or after merging. This is the output of the [bazel CI job]/https://buildkite.com/llvm-project/upstream-bazel/builds/198359/list?sid=01a110e7-6063-4ac0-932a-302814800aad&tab=output), but the problem does not seem bazel-specific:

```
# | /tmp/bazel_build/sandbox/processwrapper-sandbox/392822/execroot/_main/bazel-out/k8-fastbuild/bin/external/+llvm_configure+llvm-project/mlir/test/Conversion/TosaToLinalg/tosa-to-linalg.mlir.test.runfiles/+llvm_configure+llvm-project/mlir/test/Conversion/TosaToLinalg/tosa-to-linalg.mlir:2022:12: error: CHECK: expected string not found in input
# |  // CHECK: %[[INDICES:.+]] = "tosa.const"()
# |            ^
# | <stdin>:2105:44: note: scanning from here
# |  func.func @row_gather_i64_boundary_indices(%arg0: tensor<2x5x2xi16>) -> tensor<2x4x2xi16> {
# |                                            ^
# | <stdin>:2105:44: note: pattern attempts to capture variables: "INDICES"
# |  func.func @row_gather_i64_boundary_indices(%arg0: tensor<2x5x2xi16>) -> tensor<2x4x2xi16> {
# |                                            ^
# | <stdin>:2106:2: note: possible intended match here
# |  %0 = tosa.const values(dense<[[0, 3], [3, 1]]> : tensor<2x2xi64>) : () -> tensor<2x2xi64>
# |  ^
```

https://github.com/llvm/llvm-project/pull/225417


More information about the Mlir-commits mailing list