Timing the MLIR test suite (latest git) on a fast 32-core system shows the results pasted below. The sparse tensor integration tests completely dominate the profile in a way disproportionate to their footprint in the repo. @aartbik: can these tests be minimized/trimmed/reduced in a way they take about 1/4th of the time that they currently take? They are otherwise slowing down things for everyone – more energy, more CPU cycles, longer waits, etc. I personally always keep integration tests on for better testing/coverage – typically, for landing commits as well as for most development. So, turning those off isn’t a solution.

`LIT_OPTS="--time-tests" ninja check-mlir`

```
Slowest Tests:
--------------------------------------------------------------------------
3.68s: MLIR :: Integration/Dialect/SparseTensor/CPU/sparse_binary.mlir
3.20s: MLIR :: Integration/Dialect/SparseTensor/CPU/sparse_conversion_sparse2dense.mlir
3.18s: MLIR :: Integration/Dialect/SparseTensor/CPU/sparse_conv_3d_ndhwc_dhwcf.mlir
3.04s: MLIR :: Integration/Dialect/SparseTensor/CPU/sparse_conversion.mlir
2.99s: MLIR :: Integration/Dialect/SparseTensor/CPU/sparse_conv_3d.mlir
2.90s: MLIR :: Integration/Dialect/SparseTensor/CPU/concatenate_dim_0_permute.mlir
2.86s: MLIR :: Integration/Dialect/SparseTensor/CPU/concatenate_dim_1_permute.mlir
2.85s: MLIR :: Integration/Dialect/SparseTensor/CPU/sparse_conv_2d.mlir
2.85s: MLIR :: Integration/Dialect/SparseTensor/CPU/sparse_collapse_shape.mlir
2.82s: MLIR :: Integration/Dialect/SparseTensor/CPU/sparse_matmul.mlir
2.64s: MLIR :: Integration/Dialect/SparseTensor/CPU/sparse_conversion_ptr.mlir
2.59s: MLIR :: Integration/Dialect/SparseTensor/CPU/sparse_conv_2d_nhwc_hwcf.mlir
2.48s: MLIR :: Integration/Dialect/SparseTensor/CPU/sparse_conversion_sparse2sparse.mlir
2.34s: MLIR :: Integration/Dialect/SparseTensor/CPU/sparse_storage.mlir
2.13s: MLIR :: Integration/Dialect/SparseTensor/CPU/sparse_flatten.mlir
1.95s: MLIR :: Integration/Dialect/SparseTensor/CPU/sparse_expand_shape.mlir
1.69s: MLIR :: Integration/Dialect/SparseTensor/CPU/sparse_codegen_foreach.mlir
1.65s: MLIR :: Integration/Dialect/Async/CPU/microbench-scf-async-parallel-for.mlir
1.63s: MLIR :: Integration/Dialect/SparseTensor/CPU/sparse_conversion_dyn.mlir
1.61s: MLIR :: Integration/Dialect/SparseTensor/CPU/sparse_conv_1d_nwc_wcf.mlir
Tests Times:
--------------------------------------------------------------------------
[ Range ] :: [ Percentage ] :: [ Count ]
--------------------------------------------------------------------------
[3.60s,3.80s) :: [ ] :: [ 1/1696]
[3.40s,3.60s) :: [ ] :: [ 0/1696]
[3.20s,3.40s) :: [ ] :: [ 2/1696]
[3.00s,3.20s) :: [ ] :: [ 2/1696]
[2.80s,3.00s) :: [ ] :: [ 5/1696]
[2.60s,2.80s) :: [ ] :: [ 2/1696]
[2.40s,2.60s) :: [ ] :: [ 2/1696]
[2.20s,2.40s) :: [ ] :: [ 0/1696]
[2.00s,2.20s) :: [ ] :: [ 2/1696]
[1.80s,2.00s) :: [ ] :: [ 0/1696]
[1.60s,1.80s) :: [ ] :: [ 4/1696]
[1.40s,1.60s) :: [ ] :: [ 6/1696]
[1.20s,1.40s) :: [ ] :: [ 4/1696]
[1.00s,1.20s) :: [ ] :: [ 11/1696]
[0.80s,1.00s) :: [ ] :: [ 15/1696]
[0.60s,0.80s) :: [ ] :: [ 12/1696]
[0.40s,0.60s) :: [ ] :: [ 12/1696]
[0.20s,0.40s) :: [ ] :: [ 41/1696]
[0.00s,0.20s) :: [************************************* ] :: [1575/1696]
--------------------------------------------------------------------------
********************
Failed Tests (1):
MLIR :: Integration/Dialect/Vector/CPU/test-contraction.mlir
Testing Time: 3.76s
Unsupported: 139
Passed : 1963
Failed : 1
```

(On a relatively minor note, the naming of these test files deviates from the style. Nearly all of MLIR and LLVM uses hyphens for test cases.)