Remove RUN line duplication in mlir/test/Integration/Dialect/SparseTensor/CPU/

[banach-space]

When extending SparseCompiler integrations tests to run via SVE codegen (see https://reviews.llvm.org/D143514 and https://reviews.llvm.org/D121304), we effectively duplicated many of the RUN lines.

More specifically, if the MLIR_RUN_ARM_SVE_TESTS CMake flag is disabled, the following RUN lines are effectively identical (they verify similar code-path):

1. llvm-project/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_unary.mlir

   Lines 15 to 16 in 6a4e9cc

   	// REDEFINE: %{option} = "enable-runtime-library=false enable-buffer-initialization=true vl=2 reassociate-fp-reductions=true enable-index-optimizations=true"
   	// RUN: %{compile} | %{run}

2. llvm-project/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_unary.mlir

   Lines 20 to 27 in 4c72266

   	// REDEFINE: %{option} = "enable-runtime-library=false vl=4 enable-buffer-initialization=true reassociate-fp-reductions=true enable-index-optimizations=true enable-arm-sve=%ENABLE_VLA"
   	// REDEFINE: %{run} = %lli \
   	// REDEFINE: --entry-function=entry_lli \
   	// REDEFINE: --extra-module=%S/Inputs/main_for_lli.ll \
   	// REDEFINE: %VLA_ARCH_ATTR_OPTIONS \
   	// REDEFINE: --dlopen=%mlir_native_utils_lib_dir/libmlir_c_runner_utils%shlibext | \
   	// REDEFINE: FileCheck %s
   	// RUN: %{compile} | mlir-translate -mlir-to-llvmir | %{run}

We should avoid this. My suggestion would be to replace both RUN lines with RUN line 2. above. Additionally, we should make sure that there are public buildbots that

• set MLIR_RUN_ARM_SVE_TESTS to On (e.g. clang-aarch64-sve-vla
• set MLIR_RUN_ARM_SVE_TESTS to Off (that's the default, so any buildbot running MLIR integration tests should be fine, e.g. mlir-rocm-mi200

This way we will be making sure that all configurations are tested, but there is not code duplication in tests. WDYT?

CC @aartbik

-Andrzej

[llvmbot]

@llvm/issue-subscribers-mlir-sparse

[banach-space]

Apologies, I forgot to update this with the progress so far.

The most natural way to remove duplication would be to replace lli with mlir-cpu-runner and to fold the two RUN lines listed above into 1.

1. Patch to add -mattr and -march to -mlir-cpu-runner so that there's no need to use lli instead of mlir-cpu-runner: https://reviews.llvm.org/D146917 ✅
2. Patch to "fold" the duplicated RUN lines into one: https://reviews.llvm.org/D148005.

The 2nd patch is still under review.