[mlir][sparse] run more integration tests with and without SIMD

[wrengr]

A few questions (though none of them are blocking)

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_cast.mlir

16

Why the removal of enable-simd-index32? Doesn't seem to fit with the rest of the differential

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_flatten.mlir

5–19

Why the addition of --lower-affine? For correctness or performance?

6–18

Are we allowed to use shell for-loops here? If so, I think it'd be a lot clearer to do something like:

// RUN: for p in '' '="vectorization-strategy=2 vl=4"' ; do
// RUN: mlir-opt %s --sparsification${p} --sparse-tensor-conversion ...

rather than repeating everything with that one change. (That doesn't parse quite right due to how the runner quotes/evals things, but some variation of it ought to work I'd think.)

[aartbik]

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_flatten.mlir

5–19

For the vector case, it is needed for correctness, I add it to the scalar case for consistency.

6–18

Yeah, I was also thinking there should be better ways to do this. For now, I picked this since SIMD makes sense for some, but not all tests, so I added it where it makes sense. Also, at one point I want to move all those flags into one centralized play, so we do something like

mlir-opt ${SPARSE_FLAGS} ....

and that would encapsulate the SIMD/no-SIMD case. I will file a buganizer entry on that for follow-up!

[aartbik]

I filed https://bugs.llvm.org/show_bug.cgi?id=52409 for centralizing sparse compiler configs.

[aartbik]

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_cast.mlir

16

While here, I noticed that it does not makes sense, since we are not setting the bitwidth to anything < 64 in this test

[aartbik]

rebased with main

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_cast.mlir

 // RUN: mlir-opt %s \
 // RUN:   --sparsification --sparse-tensor-conversion \
 // RUN:   --convert-vector-to-scf --convert-scf-to-std \
 // RUN:   --func-bufferize --tensor-constant-bufferize --tensor-bufferize \
 // RUN:   --std-bufferize --finalizing-bufferize --lower-affine \
 // RUN:   --convert-vector-to-llvm --convert-memref-to-llvm \
 // RUN:   --convert-std-to-llvm --reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner \
 // RUN:  -e entry -entry-point-result=void  \
 // RUN:  -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
 //
 // Do the same run, but now with SIMDization as well. This should not change the outcome.
 //
 // RUN: mlir-opt %s \
-// RUN:   --sparsification="vectorization-strategy=2 vl=2 enable-simd-index32" --sparse-tensor-conversion \
+// RUN:   --sparsification="vectorization-strategy=2 vl=2" --sparse-tensor-conversion \
 // RUN:   --convert-vector-to-scf --convert-scf-to-std \
 // RUN:   --func-bufferize --tensor-constant-bufferize --tensor-bufferize \
 // RUN:   --std-bufferize --finalizing-bufferize --lower-affine \
 // RUN:   --convert-vector-to-llvm --convert-memref-to-llvm \
 // RUN:   --convert-std-to-llvm --reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner \
 // RUN:  -e entry -entry-point-result=void  \
 // RUN:  -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
-//
 
 #SV = #sparse_tensor.encoding<{ dimLevelType = [ "compressed" ] }>
 
 #trait_cast = {
   indexing_maps = [
     affine_map<(i) -> (i)>,  // A (in)
     affine_map<(i) -> (i)>   // X (out)
   ],

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_flatten.mlir

 // RUN: mlir-opt %s \
 // RUN:   --sparsification --sparse-tensor-conversion \
 // RUN:   --convert-vector-to-scf --convert-scf-to-std \
 // RUN:   --func-bufferize --tensor-constant-bufferize --tensor-bufferize \
-// RUN:   --std-bufferize --finalizing-bufferize  \
+// RUN:   --std-bufferize --finalizing-bufferize --lower-affine \
+// RUN:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-std-to-llvm --reconcile-unrealized-casts | \
+// RUN: TENSOR0="%mlir_integration_test_dir/data/test.tns" \
+// RUN: mlir-cpu-runner \
+// RUN:  -e entry -entry-point-result=void  \
+// RUN:  -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+//
+// Do the same run, but now with SIMDization as well. This should not change the outcome.
+//
+// RUN: mlir-opt %s \
+// RUN:   --sparsification="vectorization-strategy=2 vl=4" --sparse-tensor-conversion \
+// RUN:   --convert-vector-to-scf --convert-scf-to-std \
+// RUN:   --func-bufferize --tensor-constant-bufferize --tensor-bufferize \
+// RUN:   --std-bufferize --finalizing-bufferize --lower-affine \
 // RUN:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-std-to-llvm --reconcile-unrealized-casts | \
 // RUN: TENSOR0="%mlir_integration_test_dir/data/test.tns" \
 // RUN: mlir-cpu-runner \
 // RUN:  -e entry -entry-point-result=void  \
 // RUN:  -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
 
 !Filename = type !llvm.ptr<i8>

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_mttkrp.mlir

 // RUN: mlir-opt %s \
 // RUN:   --sparsification --sparse-tensor-conversion \
 // RUN:   --convert-vector-to-scf --convert-scf-to-std \
 // RUN:   --func-bufferize --tensor-constant-bufferize --tensor-bufferize \
-// RUN:   --std-bufferize --finalizing-bufferize  \
+// RUN:   --std-bufferize --finalizing-bufferize --lower-affine \
+// RUN:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-std-to-llvm --reconcile-unrealized-casts | \
+// RUN: TENSOR0="%mlir_integration_test_dir/data/mttkrp_b.tns" \
+// RUN: mlir-cpu-runner \
+// RUN:  -e entry -entry-point-result=void  \
+// RUN:  -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+//
+// Do the same run, but now with SIMDization as well. This should not change the outcome.
+//
+// RUN: mlir-opt %s \
+// RUN:   --sparsification="vectorization-strategy=2 vl=4" --sparse-tensor-conversion \
+// RUN:   --convert-vector-to-scf --convert-scf-to-std \
+// RUN:   --func-bufferize --tensor-constant-bufferize --tensor-bufferize \
+// RUN:   --std-bufferize --finalizing-bufferize --lower-affine \
 // RUN:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-std-to-llvm --reconcile-unrealized-casts | \
 // RUN: TENSOR0="%mlir_integration_test_dir/data/mttkrp_b.tns" \
 // RUN: mlir-cpu-runner \
 // RUN:  -e entry -entry-point-result=void  \
 // RUN:  -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
 
 !Filename = type !llvm.ptr<i8>

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_out_simple.mlir

 // RUN: mlir-opt %s \
 // RUN:   --sparsification --sparse-tensor-conversion \
 // RUN:   --convert-vector-to-scf --convert-scf-to-std \
 // RUN:   --func-bufferize --tensor-constant-bufferize --tensor-bufferize \
-// RUN:   --std-bufferize --finalizing-bufferize  \
+// RUN:   --std-bufferize --finalizing-bufferize --lower-affine \
+// RUN:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-std-to-llvm --reconcile-unrealized-casts | \
+// RUN: TENSOR0="%mlir_integration_test_dir/data/test.mtx" \
+// RUN: mlir-cpu-runner \
+// RUN:  -e entry -entry-point-result=void  \
+// RUN:  -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+//
+// Do the same run, but now with SIMDization as well. This should not change the outcome.
+//
+// RUN: mlir-opt %s \
+// RUN:   --sparsification="vectorization-strategy=2 vl=4" --sparse-tensor-conversion \
+// RUN:   --convert-vector-to-scf --convert-scf-to-std \
+// RUN:   --func-bufferize --tensor-constant-bufferize --tensor-bufferize \
+// RUN:   --std-bufferize --finalizing-bufferize --lower-affine \
 // RUN:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-std-to-llvm --reconcile-unrealized-casts | \
 // RUN: TENSOR0="%mlir_integration_test_dir/data/test.mtx" \
 // RUN: mlir-cpu-runner \
 // RUN:  -e entry -entry-point-result=void  \
 // RUN:  -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
 
 !Filename = type !llvm.ptr<i8>

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_quantized_matmul.mlir

 // RUN: mlir-opt %s \
 // RUN:   --linalg-generalize-named-ops --linalg-fuse-elementwise-ops \
 // RUN:   --sparsification --sparse-tensor-conversion \
 // RUN:   --convert-vector-to-scf --convert-scf-to-std \
 // RUN:   --func-bufferize --tensor-constant-bufferize --tensor-bufferize \
 // RUN:   --std-bufferize --finalizing-bufferize --lower-affine \
 // RUN:   --convert-vector-to-llvm --convert-memref-to-llvm \
 // RUN:   --convert-std-to-llvm --reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner \
 // RUN:  -e entry -entry-point-result=void  \
 // RUN:  -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
+//
+// Do the same run, but now with SIMDization as well. This should not change the outcome.
+//
+// RUN: mlir-opt %s \
+// RUN:   --linalg-generalize-named-ops --linalg-fuse-elementwise-ops \
+// RUN:   --sparsification="vectorization-strategy=2 vl=2" --sparse-tensor-conversion \
+// RUN:   --convert-vector-to-scf --convert-scf-to-std \
+// RUN:   --func-bufferize --tensor-constant-bufferize --tensor-bufferize \
+// RUN:   --std-bufferize --finalizing-bufferize --lower-affine \
+// RUN:   --convert-vector-to-llvm --convert-memref-to-llvm \
+// RUN:   --convert-std-to-llvm --reconcile-unrealized-casts | \
+// RUN: mlir-cpu-runner \
+// RUN:  -e entry -entry-point-result=void  \
+// RUN:  -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
 
 #DCSR = #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed" ] }>
 
 // An example of a quantized sparse matmul. With the zero offset for the
 // sparse input, the sparse compiler generates very efficient code for the
 //      x(i,j) += (ext(a(i,k)) - 2) * ext(b(k,j))
 // operation.
 module {

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_sampled_mm_fusion.mlir

 // RUN: mlir-opt %s \
 // RUN:   --linalg-generalize-named-ops --linalg-fuse-elementwise-ops \
 // RUN:   --sparsification --sparse-tensor-conversion \
 // RUN:   --linalg-bufferize --convert-linalg-to-loops \
 // RUN:   --convert-vector-to-scf --convert-scf-to-std \
 // RUN:   --func-bufferize --tensor-constant-bufferize --tensor-bufferize \
 // RUN:   --std-bufferize --finalizing-bufferize --lower-affine \
 // RUN:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-math-to-llvm \
 // RUN:   --convert-std-to-llvm --reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner \
 // RUN:  -e entry -entry-point-result=void  \
 // RUN:  -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
 //
-// Do the same run, but now with SIMDization as well.
-// This should not change the outcome.
+// Do the same run, but now with SIMDization as well. This should not change the outcome.
 //
 // RUN: mlir-opt %s \
 // RUN:   --linalg-generalize-named-ops --linalg-fuse-elementwise-ops \
 // RUN:   --sparsification="vectorization-strategy=2 vl=8" --sparse-tensor-conversion \
 // RUN:   --linalg-bufferize --convert-linalg-to-loops \
 // RUN:   --convert-vector-to-scf --convert-scf-to-std \
 // RUN:   --func-bufferize --tensor-constant-bufferize --tensor-bufferize \
 // RUN:   --std-bufferize --finalizing-bufferize --lower-affine \
 // RUN:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-math-to-llvm \
 // RUN:   --convert-std-to-llvm --reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner \
 // RUN:  -e entry -entry-point-result=void  \
 // RUN:  -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
-//
+
 // Interop between linalg/sparse leaves some issues to be revolved:
 // UNSUPPORTED: asan
 
-
 #SM = #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed" ] }>
 
 #trait_sampled_dense_dense = {
   indexing_maps = [
     affine_map<(i,j,k) -> (i,j)>,  // S
     affine_map<(i,j,k) -> (i,k)>,  // A
     affine_map<(i,j,k) -> (k,j)>,  // B
     affine_map<(i,j,k) -> (i,j)>   // X (out)

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_scale.mlir

 // RUN: mlir-opt %s \
 // RUN:   --sparsification --sparse-tensor-conversion \
 // RUN:   --convert-vector-to-scf --convert-scf-to-std \
 // RUN:   --func-bufferize --tensor-constant-bufferize --tensor-bufferize \
-// RUN:   --std-bufferize --finalizing-bufferize  \
+// RUN:   --std-bufferize --finalizing-bufferize --lower-affine \
+// RUN:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-std-to-llvm --reconcile-unrealized-casts | \
+// RUN: mlir-cpu-runner \
+// RUN:  -e entry -entry-point-result=void  \
+// RUN:  -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+//
+// Do the same run, but now with SIMDization as well. This should not change the outcome.
+//
+// RUN: mlir-opt %s \
+// RUN:   --sparsification="vectorization-strategy=2 vl=4" --sparse-tensor-conversion \
+// RUN:   --convert-vector-to-scf --convert-scf-to-std \
+// RUN:   --func-bufferize --tensor-constant-bufferize --tensor-bufferize \
+// RUN:   --std-bufferize --finalizing-bufferize --lower-affine \
 // RUN:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-std-to-llvm --reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner \
 // RUN:  -e entry -entry-point-result=void  \
 // RUN:  -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
 
 #CSR = #sparse_tensor.encoding<{ dimLevelType = [ "dense", "compressed" ] }>
 

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_spmm.mlir

 // RUN: mlir-opt %s \
 // RUN:   --sparsification --sparse-tensor-conversion \
 // RUN:   --convert-vector-to-scf --convert-scf-to-std \
 // RUN:   --func-bufferize --tensor-constant-bufferize --tensor-bufferize \
-// RUN:   --std-bufferize --finalizing-bufferize  \
+// RUN:   --std-bufferize --finalizing-bufferize --lower-affine \
+// RUN:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-std-to-llvm --reconcile-unrealized-casts | \
+// RUN: TENSOR0="%mlir_integration_test_dir/data/wide.mtx" \
+// RUN: mlir-cpu-runner \
+// RUN:  -e entry -entry-point-result=void  \
+// RUN:  -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+//
+// Do the same run, but now with SIMDization as well. This should not change the outcome.
+//
+// RUN: mlir-opt %s \
+// RUN:   --sparsification="vectorization-strategy=2 vl=2" --sparse-tensor-conversion \
+// RUN:   --convert-vector-to-scf --convert-scf-to-std \
+// RUN:   --func-bufferize --tensor-constant-bufferize --tensor-bufferize \
+// RUN:   --std-bufferize --finalizing-bufferize --lower-affine \
 // RUN:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-std-to-llvm --reconcile-unrealized-casts | \
 // RUN: TENSOR0="%mlir_integration_test_dir/data/wide.mtx" \
 // RUN: mlir-cpu-runner \
 // RUN:  -e entry -entry-point-result=void  \
 // RUN:  -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
 
 !Filename = type !llvm.ptr<i8>

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_sum.mlir

 // RUN: mlir-opt %s \
 // RUN:   --sparsification --sparse-tensor-conversion \
 // RUN:   --convert-vector-to-scf --convert-scf-to-std \
 // RUN:   --func-bufferize --tensor-constant-bufferize --tensor-bufferize \
-// RUN:   --std-bufferize --finalizing-bufferize  \
+// RUN:   --std-bufferize --finalizing-bufferize --lower-affine \
+// RUN:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-std-to-llvm --reconcile-unrealized-casts | \
+// RUN: TENSOR0="%mlir_integration_test_dir/data/test.mtx" \
+// RUN: mlir-cpu-runner \
+// RUN:  -e entry -entry-point-result=void  \
+// RUN:  -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+//
+// Do the same run, but now with SIMDization as well. This should not change the outcome.
+//
+// RUN: mlir-opt %s \
+// RUN:   --sparsification="vectorization-strategy=2 vl=2" --sparse-tensor-conversion \
+// RUN:   --convert-vector-to-scf --convert-scf-to-std \
+// RUN:   --func-bufferize --tensor-constant-bufferize --tensor-bufferize \
+// RUN:   --std-bufferize --finalizing-bufferize --lower-affine \
 // RUN:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-std-to-llvm --reconcile-unrealized-casts | \
 // RUN: TENSOR0="%mlir_integration_test_dir/data/test.mtx" \
 // RUN: mlir-cpu-runner \
 // RUN:  -e entry -entry-point-result=void  \
 // RUN:  -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
 
 !Filename = type !llvm.ptr<i8>