[mlir][sparse] Adding standard pipeline for tests.

[wrengr]

Summary

Addresses https://bugs.llvm.org/show_bug.cgi?id=52409 aka https://github.com/llvm/llvm-project/issues/51751

[mehdi_amini]

Sweet! LGTM overall, but I'll let Aart have another look swell.

I also have one slight concerns about a global constructor for the pipeline registration.

mlir/lib/Dialect/SparseTensor/Transforms/SparseTensorPasses.cpp

142

Does this build? I would expect this to be caught as a global constructor which isn't allowed in the MLIR libraries.

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

10

It's not coming from you here, but using opaque integer in the options is really not ideal, I'd rather see enums! Do you mind filing a bug?

[wrengr]

mlir/lib/Dialect/SparseTensor/Transforms/SparseTensorPasses.cpp

142

It builds and works on the Google side; still waiting to see if the buildbot likes it.

If there's another/better approach I'm all ears. The documentation I could find was incomplete and I couldn't find any examples in the MLIR monorepo (except for debugging the pipeline functionality itself). The examples I could find were in tensorflow and ireee, but both were inconsistent about defining a top-level static variable vs wrapping it in a registration function. So yeah, I'm game to switch it to whatever the right thing is :)

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

10

Yeah, I'm not a fan of opaque numbers either, just didn't want to handle that in the same differential. Will file a bug for it. I did locate the documentation for using enums with the LLVM commandline framework, but I didn't see anything for tablegen or MLIR specifically. Do those exist, or should I just reuse the LLVM combinators directly?

[mehdi_amini]

mlir/lib/Dialect/SparseTensor/Transforms/SparseTensorPasses.cpp

142

You're right, this is terrible...
The pipeline registration is very old, and has no in-tree user. This is also somehow obsolete by the dynamic pipeline that we have now, maybe we should remove this pipeline registration mechanism entirely.

So I think we should try to use the dynamic pipeline mechanism here as well, do you want to sync offline on this? I can help you set this up here.

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

10

It just reuses LLVM functionality. Here is an example: https://github.com/llvm/llvm-project/blob/705d8c49f9be82415a9cdd793cda405333fba71e/mlir/include/mlir/Transforms/Passes.td#L138-L147

[aartbik]

Nice cleanup indeed! But indeed some fix is required, please work with Mehdi.

/var/lib/buildkite-agent/builds/llvm-project/mlir/lib/Dialect/SparseTensor/Transforms/SparseTensorPasses.cpp:144:5: error: declaration requires a global constructor [-Werror,-Wglobal-constructors]

registerSparseTensorPipeline("sparse-tensor-standard-pipeline",
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

1 error generated.

[aartbik]

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

10

I don't think anybody should be a fan of opaque numbers, but they sometimes are simply a convenient way to provide some "mechanisms" before a "policy" can be determined (used at other places too). There is a long term intention to keep these.

[aartbik]

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

10

I don't think anybody should be a fan of opaque numbers, but they sometimes are simply a convenient way to provide some "mechanisms" before a "policy" can be determined (used at other places too). There is a long term intention to keep these.

Oops, we need an edit button for this comments. This should obviously read: There is no long term intention to keep these.

[mehdi_amini]

I think we have a much deeper problem here actually, which is reflected in the BUILD file.

This introduces a lot of dependencies in the SparseTensor dialect directory to a lot of components which it shouldn't depend on (Seeing LinalgTransforms is already strange here).

What we're really missing here is very fundamental to the repository, we need to make progress on https://llvm.discourse.group/t/rfc-restructuring-of-the-mlir-repo/4927

[aartbik]

In D117919#3262952, @mehdi_amini wrote:

I think we have a much deeper problem here actually, which is reflected in the BUILD file.

Yeah, I noticed that. Any way to avoid that would be really great ;-)

[mehdi_amini]

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

10

, but they sometimes are simply a convenient way to provide some "mechanisms" before a "policy" can be determined

Feel like I am missing something: are they more convenient than an enum?
Unless you're doing arithmetic on these numbers, I would think that using None, DenseInnerLoop, and AnyStorageInnerLoop would just be as convenient to use?

[aartbik]

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

10

Feel like I am missing something: are they more convenient than an enum?

I am not trying to defend the indefensible here ;-) just tried to give some history of why these are here (we use proper enums "internally", but I at some point in the past I added a quick hack to control this through the command line with a pass "Option" and I could only find integral examples back then).

Perfectly okay with making this proper enums of course, although in the longer run we will need something like Nicolas' schedule language to control the transformations in an even better way, since they may vary over loops and the switch controls it globally now.

[wrengr]

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

10

Filed https://github.com/llvm/llvm-project/issues/53389

[wrengr]

Yeah, there's definitely a major issue re excessive dependencies; though I'm not sure https://llvm.discourse.group/t/rfc-restructuring-of-the-mlir-repo/4927 would fix it. Ultimately the tests do depend on all these different passes, however the rest of the SparseTensor library itself does not; so the fundamental problem is how to specify the test dependencies yet without over-constraining the library itself. To fix this we really need some sort of layering of the build dependencies, so that the pipeline can be broken out to group together with the tests themselves. (Which is afaict out of the scope of RFC-4927.) We could hack it by adding another layer of shell scripts between mlir-opt and filecheck/lilith, but (a) just using a string in the shell script loses track of the dependency information (same as the tests before this differential), and (b) layers of shell scripts are a maintenance nightmare. One possible way forward would be to restructure the pass/pipeline registration stuff to be late/lazy-binding such that: we can define the pipeline alongside the rest of the library, but those pass-registration calls are only validated when linking libraries together (ideally with a link-time error if things don't match up, rather than a run-time error). Not sure of the exact details nor if it runs counter to llvm/mlir ideology, but might be worth thinking about.

[wrengr]

Factoring out the pipeline into a separate library / build-target.

[wrengr]

I'm not entirely sure how to adjust mlir/include/mlir/Dialect/SparseTensor/Pipelines/CMakeLists.txt given that there's no tablegen file for this library, but we still want to have the pipeline show up in the generated documentation alongside the regular passes. Any pointers would be appreciated

[mehdi_amini]

LG, with a nit inline!

mlir/include/mlir/Dialect/SparseTensor/Pipelines/Passes.h

50

You're missing the declaration for buildSparseTensorPipeline ; I think the SparseTensorPipelineOptions won't be useful without it right ?

mlir/lib/Dialect/SparseTensor/Pipelines/CMakeLists.txt

16

If you're unsure that you got all the dependencies, building with -DBUILD_SHARED_LIBS=ON helps!

[aartbik]

Yeah, really looking good Wren!

mlir/lib/Dialect/SparseTensor/Pipelines/SparseTensorPipelines.cpp

56–57

I am not sure what the right phrasing is (since your English will be better than mine), but can we elaborate a bit here on that this runs all the passes required to invoke the "sparse compiler", i.e. map sparsity agnostic IR with sparse tensor type into concrete sparse code.

[aartbik]

mlir/lib/Dialect/SparseTensor/Pipelines/CMakeLists.txt

16

+1

I always run that build config after larger build-related changes just so I don't get unpleasant surprises later ;-)

[wrengr]

Fixing nits

[aartbik]

mlir/lib/Dialect/SparseTensor/Pipelines/SparseTensorPipelines.cpp

56–57

Yes, thanks! Last nit, can we call it a sparse compiler, perhaps in between quotes or some other stylistic marker. It is kind of my pet name for such a pipeline ;-)

So something like

"Pipeline of a `sparse compiler' that takes sparsity-agnostic ....

[wrengr]

fixing CMakeLists.txt

[wrengr]

Renaming sparse-tensor-standard-pipeline to sparse-compiler

[wrengr]

rebasing

[rriddle]

mlir/lib/Dialect/SparseTensor/Pipelines/SparseTensorPipelines.cpp

29

What operation is pm here expected to run on? It looks like this builder is relying on an implicitly nesting pass manager (e.g. createLinalgBufferizePass runs on a FuncOp, but createSparsificationPass runs on a ModuleOp). Pass pipelines should be explicit about the pass manager nesting, otherwise if implicit nesting isn't enabled (in many cases) the pipeline will fail to add to a pass manager (likely causing the problems in D118579).

[wrengr]

mlir/lib/Dialect/SparseTensor/Pipelines/SparseTensorPipelines.cpp

29

This should be fixed by D118658

mlir/include/mlir/Dialect/SparseTensor/Pipelines/Passes.h

+//===- Passes.h - Sparse tensor pipeline entry points -----------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This header file defines prototypes of all sparse tensor pipelines.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MLIR_DIALECT_SPARSETENSOR_PIPELINES_PASSES_H_
+#define MLIR_DIALECT_SPARSETENSOR_PIPELINES_PASSES_H_
+
+#include "mlir/Dialect/SparseTensor/Transforms/Passes.h"
+#include "mlir/Pass/PassOptions.h"
+
+using namespace mlir::detail;
+using namespace llvm::cl;
+
+namespace mlir {
+namespace sparse_tensor {
+
+/// Options for the "sparse-compiler" pipeline.  So far this contains
+/// only the same options as the sparsification pass, and must be kept
+/// in sync with the `SparseTensor/Transforms/Passes.td` file.  In the
+/// future this may be extended with options for other passes in the pipeline.
+struct SparseCompilerOptions
+    : public PassPipelineOptions<SparseCompilerOptions> {
+  PassOptions::Option<int32_t> parallelization{
+      *this, "parallelization-strategy",
+      desc("Set the parallelization strategy"), init(0)};
+  PassOptions::Option<int32_t> vectorization{
+      *this, "vectorization-strategy", desc("Set the vectorization strategy"),
+      init(0)};
+  PassOptions::Option<int32_t> vectorLength{
+      *this, "vl", desc("Set the vector length"), init(1)};
+  PassOptions::Option<bool> enableSIMDIndex32{
+      *this, "enable-simd-index32",
+      desc("Enable i32 indexing into vectors (for efficiency)"), init(false)};
+
+  /// Projects out the options for the sparsification pass.
+  SparsificationOptions sparsificationOptions() const {
+    return SparsificationOptions(sparseParallelizationStrategy(parallelization),
+                                 sparseVectorizationStrategy(vectorization),
+                                 vectorLength, enableSIMDIndex32);
+  }
+};
+
+//===----------------------------------------------------------------------===//
+// Building and Registering.
+//===----------------------------------------------------------------------===//
+
+/// Adds the "sparse-compiler" pipeline to the `OpPassManager`.  This
+/// is the standard pipeline for taking sparsity-agnostic IR using
+/// the sparse-tensor type and lowering it to LLVM IR with concrete
+/// representations and algorithms for sparse tensors.
+void buildSparseCompiler(OpPassManager &pm,
+                         const SparseCompilerOptions &options);
+
+/// Registers all pipelines for the `sparse_tensor` dialect.  At present,
+/// this includes only "sparse-compiler".
+void registerSparseTensorPipelines();
+
+} // namespace sparse_tensor
+} // namespace mlir
+
+#endif // MLIR_DIALECT_SPARSETENSOR_PIPELINES_PASSES_H_

mlir/include/mlir/Dialect/SparseTensor/Transforms/Passes.h

enum class SparseParallelizationStrategy {
   kNone,
   kDenseOuterLoop,
   kAnyStorageOuterLoop,
   kDenseAnyLoop,
   kAnyStorageAnyLoop
   // TODO: support reduction parallelization too?
 };
 
+/// Converts command-line parallelization flag to the strategy enum.
+SparseParallelizationStrategy sparseParallelizationStrategy(int32_t flag);
+
 /// Defines a vectorization strategy. Any inner loop is a candidate (full SIMD
 /// for parallel loops and horizontal SIMD for reduction loops). A loop is
 /// actually vectorized if (1) allowed by the strategy, and (2) the emitted
 /// code is an actual for-loop (and not a co-iterating while-loop).
 enum class SparseVectorizationStrategy {
   kNone,
   kDenseInnerLoop,
   kAnyStorageInnerLoop
 };
 
+/// Converts command-line vectorization flag to the strategy enum.
+SparseVectorizationStrategy sparseVectorizationStrategy(int32_t flag);
+
 /// Sparsification options.
 struct SparsificationOptions {
   SparsificationOptions(SparseParallelizationStrategy p,
                         SparseVectorizationStrategy v, unsigned vl, bool e)
       : parallelizationStrategy(p), vectorizationStrategy(v), vectorLength(vl),
         enableSIMDIndex32(e) {}
   SparsificationOptions()
       : SparsificationOptions(SparseParallelizationStrategy::kNone,
void populateSparsificationPatterns(
     RewritePatternSet &patterns,
     const SparsificationOptions &options = SparsificationOptions());
 
 /// Sets up sparse tensor conversion rules.
 void populateSparseTensorConversionPatterns(TypeConverter &typeConverter,
                                             RewritePatternSet &patterns);
 
 std::unique_ptr<Pass> createSparsificationPass();
+std::unique_ptr<Pass>
+createSparsificationPass(const SparsificationOptions &options);
 std::unique_ptr<Pass> createSparseTensorConversionPass();
 
 //===----------------------------------------------------------------------===//
 // Registration.
 //===----------------------------------------------------------------------===//
 
 /// Generate the code for registering passes.
 #define GEN_PASS_REGISTRATION
 #include "mlir/Dialect/SparseTensor/Transforms/Passes.h.inc"
 
 } // namespace mlir
 
 #endif // MLIR_DIALECT_SPARSETENSOR_TRANSFORMS_PASSES_H_

mlir/include/mlir/InitAllPasses.h

 #include "mlir/Dialect/GPU/Passes.h"
 #include "mlir/Dialect/LLVMIR/Transforms/Passes.h"
 #include "mlir/Dialect/Linalg/Passes.h"
 #include "mlir/Dialect/MemRef/Transforms/Passes.h"
 #include "mlir/Dialect/Quant/Passes.h"
 #include "mlir/Dialect/SCF/Passes.h"
 #include "mlir/Dialect/SPIRV/Transforms/Passes.h"
 #include "mlir/Dialect/Shape/Transforms/Passes.h"
+#include "mlir/Dialect/SparseTensor/Pipelines/Passes.h"
 #include "mlir/Dialect/SparseTensor/Transforms/Passes.h"
 #include "mlir/Dialect/StandardOps/Transforms/Passes.h"
 #include "mlir/Dialect/Tensor/Transforms/Passes.h"
 #include "mlir/Dialect/Tosa/Transforms/Passes.h"
 #include "mlir/Transforms/Passes.h"
 
 #include <cstdlib>
 
inline void registerAllPasses() {
   memref::registerMemRefPasses();
   quant::registerQuantPasses();
   registerSCFPasses();
   registerShapePasses();
   spirv::registerSPIRVPasses();
   registerStandardPasses();
   tensor::registerTensorPasses();
   tosa::registerTosaOptPasses();
+
+  // Dialect pipelines
+  sparse_tensor::registerSparseTensorPipelines();
 }
 
 } // namespace mlir
 
 #endif // MLIR_INITALLPASSES_H_

mlir/lib/Dialect/SparseTensor/CMakeLists.txt

 add_subdirectory(IR)
 add_subdirectory(Transforms)
+add_subdirectory(Pipelines)
 add_subdirectory(Utils)

mlir/lib/Dialect/SparseTensor/Pipelines/CMakeLists.txt

+add_mlir_dialect_library(MLIRSparseTensorPipelines
+  SparseTensorPipelines.cpp
+
+  ADDITIONAL_HEADER_DIRS
+  ${MLIR_MAIN_INCLUDE_DIR}/mlir/Dialect/SparseTensor
+
+  LINK_LIBS PUBLIC
+  MLIRAffineToStandard
+  MLIRBufferizationTransforms
+  MLIRLinalgTransforms
+  MLIRMathToLLVM
+  MLIRMemRefToLLVM
+  MLIRPass
+  MLIRReconcileUnrealizedCasts
+  MLIRSCFToStandard
+  MLIRSparseTensor
+  MLIRSparseTensorTransforms
+  MLIRStandardOpsTransforms
+  MLIRTensorTransforms
+  MLIRVectorToLLVM
+)

mlir/lib/Dialect/SparseTensor/Pipelines/SparseTensorPipelines.cpp

+//===- SparseTensorPipelines.cpp - Pipelines for sparse tensor code -------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include "mlir/Dialect/SparseTensor/Pipelines/Passes.h"
+
+#include "mlir/Conversion/Passes.h"
+#include "mlir/Dialect/Bufferization/Transforms/Passes.h"
+#include "mlir/Dialect/Linalg/Passes.h"
+#include "mlir/Dialect/SparseTensor/IR/SparseTensor.h"
+#include "mlir/Dialect/SparseTensor/Transforms/Passes.h"
+#include "mlir/Dialect/StandardOps/Transforms/Passes.h"
+#include "mlir/Dialect/Tensor/Transforms/Passes.h"
+#include "mlir/Pass/PassManager.h"
+
+using namespace mlir;
+using namespace mlir::sparse_tensor;
+
+//===----------------------------------------------------------------------===//
+// Pipeline implementation.
+//===----------------------------------------------------------------------===//
+
+void mlir::sparse_tensor::buildSparseCompiler(
+    OpPassManager &pm, const SparseCompilerOptions &options) {
+  pm.addPass(createSparsificationPass(options.sparsificationOptions()));
+  pm.addPass(createSparseTensorConversionPass());
+  pm.addPass(createLinalgBufferizePass());
+  pm.addPass(createConvertLinalgToLoopsPass());
+  pm.addPass(createConvertVectorToSCFPass());
+  pm.addPass(createLowerToCFGPass()); // --convert-scf-to-std
+  pm.addPass(createFuncBufferizePass());
+  pm.addPass(createTensorConstantBufferizePass());
+  pm.addPass(createTensorBufferizePass());
+  pm.addPass(createStdBufferizePass());
+  pm.addPass(mlir::bufferization::createFinalizingBufferizePass());
+  pm.addPass(createLowerAffinePass());
+  pm.addPass(createConvertVectorToLLVMPass());
+  pm.addPass(createMemRefToLLVMPass());
+  pm.addPass(createConvertMathToLLVMPass());
+  pm.addPass(createLowerToLLVMPass()); // --convert-std-to-llvm
+  pm.addPass(createReconcileUnrealizedCastsPass());
+}
+
+//===----------------------------------------------------------------------===//
+// Pipeline registration.
+//===----------------------------------------------------------------------===//
+
+void mlir::sparse_tensor::registerSparseTensorPipelines() {
+  PassPipelineRegistration<SparseCompilerOptions>(
+      "sparse-compiler",
+      "The standard pipeline for taking sparsity-agnostic IR using the"
+      " sparse-tensor type, and lowering it to LLVM IR with concrete"
+      " representations and algorithms for sparse tensors.",
+      buildSparseCompiler);
+}

mlir/lib/Dialect/SparseTensor/Transforms/SparseTensorPasses.cpp

 //===----------------------------------------------------------------------===//
 // Passes implementation.
 //===----------------------------------------------------------------------===//
 
 struct SparsificationPass : public SparsificationBase<SparsificationPass> {
 
   SparsificationPass() = default;
   SparsificationPass(const SparsificationPass &pass) = default;
-
-  /// Returns parallelization strategy given on command line.
-  SparseParallelizationStrategy parallelOption() {
-    switch (parallelization) {
-    default:
-      return SparseParallelizationStrategy::kNone;
-    case 1:
-      return SparseParallelizationStrategy::kDenseOuterLoop;
-    case 2:
-      return SparseParallelizationStrategy::kAnyStorageOuterLoop;
-    case 3:
-      return SparseParallelizationStrategy::kDenseAnyLoop;
-    case 4:
-      return SparseParallelizationStrategy::kAnyStorageAnyLoop;
-    }
-  }
-
-  /// Returns vectorization strategy given on command line.
-  SparseVectorizationStrategy vectorOption() {
-    switch (vectorization) {
-    default:
-      return SparseVectorizationStrategy::kNone;
-    case 1:
-      return SparseVectorizationStrategy::kDenseInnerLoop;
-    case 2:
-      return SparseVectorizationStrategy::kAnyStorageInnerLoop;
-    }
+  SparsificationPass(const SparsificationOptions &options) {
+    parallelization = static_cast<int32_t>(options.parallelizationStrategy);
+    vectorization = static_cast<int32_t>(options.vectorizationStrategy);
+    vectorLength = options.vectorLength;
+    enableSIMDIndex32 = options.enableSIMDIndex32;
   }
 
   void runOnOperation() override {
     auto *ctx = &getContext();
     RewritePatternSet patterns(ctx);
     // Translate strategy flags to strategy options.
-    SparsificationOptions options(parallelOption(), vectorOption(),
-                                  vectorLength, enableSIMDIndex32);
+    SparsificationOptions options(
+        sparseParallelizationStrategy(parallelization),
+        sparseVectorizationStrategy(vectorization), vectorLength,
+        enableSIMDIndex32);
     // Apply rewriting.
     populateSparsificationPatterns(patterns, options);
     vector::populateVectorToVectorCanonicalizationPatterns(patterns);
     (void)applyPatternsAndFoldGreedily(getOperation(), std::move(patterns));
   }
 };
 
 class SparseTensorTypeConverter : public TypeConverter {
void runOnOperation() override {
     if (failed(applyPartialConversion(getOperation(), target,
                                       std::move(patterns))))
       signalPassFailure();
   }
 };
 
 } // namespace
 
+SparseParallelizationStrategy
+mlir::sparseParallelizationStrategy(int32_t flag) {
+  switch (flag) {
+  default:
+    return SparseParallelizationStrategy::kNone;
+  case 1:
+    return SparseParallelizationStrategy::kDenseOuterLoop;
+  case 2:
+    return SparseParallelizationStrategy::kAnyStorageOuterLoop;
+  case 3:
+    return SparseParallelizationStrategy::kDenseAnyLoop;
+  case 4:
+    return SparseParallelizationStrategy::kAnyStorageAnyLoop;
+  }
+}
+
+SparseVectorizationStrategy mlir::sparseVectorizationStrategy(int32_t flag) {
+  switch (flag) {
+  default:
+    return SparseVectorizationStrategy::kNone;
+  case 1:
+    return SparseVectorizationStrategy::kDenseInnerLoop;
+  case 2:
+    return SparseVectorizationStrategy::kAnyStorageInnerLoop;
+  }
+}
+
 std::unique_ptr<Pass> mlir::createSparsificationPass() {
   return std::make_unique<SparsificationPass>();
 }
 
+std::unique_ptr<Pass>
+mlir::createSparsificationPass(const SparsificationOptions &options) {
+  return std::make_unique<SparsificationPass>(options);
+}
+
 std::unique_ptr<Pass> mlir::createSparseTensorConversionPass() {
   return std::make_unique<SparseTensorConversionPass>();
 }

mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output.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:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-std-to-llvm --reconcile-unrealized-casts | \
+// RUN: mlir-opt %s --sparse-compiler | \
 // RUN: TENSOR0="%mlir_integration_test_dir/data/test.mtx" \
 // RUN: TENSOR1="%mlir_integration_test_dir/data/zero.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_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-opt %s --sparse-compiler | \
 // 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 \
-// RUN:   --convert-std-to-llvm --reconcile-unrealized-casts | \
+// RUN: mlir-opt %s --sparse-compiler="vectorization-strategy=2 vl=2" | \
 // 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 = {

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_constant_to_sparse_tensor.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:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-std-to-llvm --reconcile-unrealized-casts | \
+// RUN: mlir-opt %s --sparse-compiler | \
 // 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
 
 #Tensor1  = #sparse_tensor.encoding<{
   dimLevelType = [ "compressed", "compressed"]
 }>

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_conversion.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:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-std-to-llvm --reconcile-unrealized-casts | \
+// RUN: mlir-opt %s --sparse-compiler | \
 // 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
 
 #Tensor1  = #sparse_tensor.encoding<{
   dimLevelType = [ "compressed", "compressed", "compressed" ],
   dimOrdering = affine_map<(i,j,k) -> (i,j,k)>

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

-// RUN: mlir-opt %s \
-// 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-opt %s --sparse-compiler | \
 // 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" ]
 }>

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

-// RUN: mlir-opt %s \
-// 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-opt %s --sparse-compiler | \
 // 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" ],
   pointerBitWidth = 8,

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_conversion_sparse2dense.mlir

-// RUN: mlir-opt %s \
-// 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 \
-// RUN:  -convert-vector-to-llvm -convert-memref-to-llvm -convert-std-to-llvm \
-// RUN:  -reconcile-unrealized-casts \
-// RUN:  | \
+// RUN: mlir-opt %s --sparse-compiler | \
 // RUN: mlir-cpu-runner \
 // RUN:  -e entry -entry-point-result=void \
 // RUN:  -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext \
 // RUN:  | \
 // RUN: FileCheck %s
 
 #Tensor1  = #sparse_tensor.encoding<{
   dimLevelType = [ "compressed", "compressed", "compressed" ],

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_filter_conv2d.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:   --sparse-compiler | \
 // 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:   --sparse-compiler="vectorization-strategy=2 vl=2" | \
 // 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 2D convolution with a sparse filter.

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 --lower-affine \
-// RUN:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-std-to-llvm --reconcile-unrealized-casts | \
+// RUN: mlir-opt %s --sparse-compiler | \
 // 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: mlir-opt %s --sparse-compiler="vectorization-strategy=2 vl=4" | \
 // 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_matmul.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 \
-// RUN:   --convert-std-to-llvm --reconcile-unrealized-casts | \
+// RUN:   --sparse-compiler | \
 // 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_matrix_ops.mlir

-// RUN: mlir-opt %s \
-// 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-opt %s --sparse-compiler | \
 // 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"]}>
 
 //

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_matvec.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:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-std-to-llvm --reconcile-unrealized-casts | \
+// RUN: mlir-opt %s --sparse-compiler | \
 // 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=16 enable-simd-index32" --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:   --sparse-compiler="vectorization-strategy=2 vl=16 enable-simd-index32" | \
 // 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_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 --lower-affine \
-// RUN:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-std-to-llvm --reconcile-unrealized-casts | \
+// RUN: mlir-opt %s --sparse-compiler | \
 // 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: mlir-opt %s --sparse-compiler="vectorization-strategy=2 vl=4" | \
 // 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_mult_elt.mlir

-// RUN: mlir-opt %s \
-// 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-opt %s --sparse-compiler | \
 // 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" ]
 }>

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

-// RUN: mlir-opt %s \
-// 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-opt %s --sparse-compiler | \
 // 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
 
 #SparseMatrix = #sparse_tensor.encoding<{
   dimLevelType = [ "compressed", "compressed" ]
 }>

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 --lower-affine \
-// RUN:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-std-to-llvm --reconcile-unrealized-casts | \
+// RUN: mlir-opt %s --sparse-compiler | \
 // 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: mlir-opt %s --sparse-compiler="vectorization-strategy=2 vl=4" | \
 // 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:   --sparse-compiler | \
 // 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:   --sparse-compiler="vectorization-strategy=2 vl=2" | \
 // 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

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_reductions.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-opt %s --sparse-compiler | \
 // 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=8" --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:   --sparse-compiler="vectorization-strategy=2 vl=8" | \
 // 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" ] }>
 #DV = #sparse_tensor.encoding<{ dimLevelType = [ "dense"      ] }>
 

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_sampled_matmul.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:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-std-to-llvm --reconcile-unrealized-casts | \
+// RUN: mlir-opt %s --sparse-compiler | \
 // 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 enable-simd-index32" --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:   --sparse-compiler="vectorization-strategy=2 vl=4 enable-simd-index32" | \
 // 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_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:   --sparse-compiler | \
 // 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=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:   --sparse-compiler="vectorization-strategy=2 vl=8" | \
 // 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
 
 #SM = #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed" ] }>
 
 #trait_sampled_dense_dense = {

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 --lower-affine \
-// RUN:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-std-to-llvm --reconcile-unrealized-casts | \
+// RUN: mlir-opt %s --sparse-compiler | \
 // 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-opt %s --sparse-compiler="vectorization-strategy=2 vl=4" | \
 // 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" ] }>
 
 #trait_scale = {

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 --lower-affine \
-// RUN:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-std-to-llvm --reconcile-unrealized-casts | \
+// RUN: mlir-opt %s --sparse-compiler | \
 // 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: mlir-opt %s --sparse-compiler="vectorization-strategy=2 vl=2" | \
 // 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_storage.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:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-std-to-llvm --reconcile-unrealized-casts | \
+// RUN: mlir-opt %s --sparse-compiler | \
 // 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
 
 //
 // Several common sparse storage schemes.
 //

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 --lower-affine \
-// RUN:   --convert-vector-to-llvm --convert-memref-to-llvm --convert-std-to-llvm --reconcile-unrealized-casts | \
+// RUN: mlir-opt %s --sparse-compiler | \
 // RUN: TENSOR0="%mlir_integration_test_dir/data/test_symmetric.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: mlir-opt %s --sparse-compiler="vectorization-strategy=2 vl=2" | \
 // RUN: TENSOR0="%mlir_integration_test_dir/data/test_symmetric.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_tensor_ops.mlir

-// RUN: mlir-opt %s \
-// 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-opt %s --sparse-compiler | \
 // 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
 
 #ST1 = #sparse_tensor.encoding<{dimLevelType = ["compressed", "compressed", "compressed"]}>
 #ST2 = #sparse_tensor.encoding<{dimLevelType = ["compressed", "compressed", "dense"]}>
 

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

-// RUN: mlir-opt %s \
-// 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-opt %s --sparse-compiler | \
 // 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
 
 #SparseVector = #sparse_tensor.encoding<{dimLevelType = ["compressed"]}>
 #DenseVector = #sparse_tensor.encoding<{dimLevelType = ["dense"]}>
 

utils/bazel/llvm-project-overlay/mlir/BUILD.bazel

deps = [
         ":StandardOpsTransforms",
         ":TensorDialect",
         ":Transforms",
         ":VectorOps",
         "//llvm:Support",
     ],
 )
 
+cc_library(
+    name = "SparseTensorPipelines",
+    srcs = glob(["lib/Dialect/SparseTensor/Pipelines/*.cpp"]),
+    hdrs = ["include/mlir/Dialect/SparseTensor/Pipelines/Passes.h"],
+    includes = ["include"],
+    deps = [
+        ":BufferizationTransforms",
+        ":ConversionPasses",
+        ":LinalgTransforms",
+        ":Pass",
+        ":SparseTensor",
+        ":SparseTensorTransforms",
+        ":StandardOpsTransforms",
+        ":TensorTransforms",
+    ],
+)
+
 td_library(
     name = "StdOpsTdFiles",
     srcs = [
         "include/mlir/Dialect/StandardOps/IR/Ops.td",
     ],
     includes = ["include"],
     deps = [
         ":CallInterfacesTdFiles",
deps = [
         ":SPIRVPassIncGen",
         ":SPIRVToLLVM",
         ":SPIRVTransforms",
         ":Shape",
         ":ShapeToStandard",
         ":ShapeTransforms",
         ":ShapeTransformsPassIncGen",
         ":SparseTensor",
+        ":SparseTensorPipelines",
         ":SparseTensorTransforms",
         ":StandardOps",
         ":StandardOpsTransforms",
         ":StandardOpsTransformsPassIncGen",
         ":StandardToLLVM",
         ":StandardToSPIRV",
         ":TensorDialect",
         ":TensorInferTypeOpInterfaceImpl",