Create the MLIR Reduce framework

[msifontes]

Summary

Create the framework and testing environment for MLIR Reduce - a tool
with the objective to reduce large test cases into smaller ones while
preserving their interesting behavior.

Implement the framework to parse the command line arguments, parse the
input MLIR test case into a module and call reduction passes on the MLIR module.

Implement the Tester class which allows the different reduction passes to test the
interesting behavior of the generated reduced variants of the test case and keep track
of the most reduced generated variant.

[msifontes]

#Updating D82803: Create the MLIR Reduce framework

Uncomment the canonicalization pass of the MLIR module.

[msifontes]

Updating D82803: Create the MLIR Reduce framework

Change variable names to comply with camelCase formatting.

[rriddle]

Thanks Mauricio, added some initial comments.

mlir/tools/mlir-reduce/Tester.cpp

17 ↗

(On Diff #274236)

Please avoid using namespace llvm whenever you can.

28 ↗

(On Diff #274236)

nit: Spell out auto here.

mlir/tools/mlir-reduce/Tester.h

1 ↗

(On Diff #274236)

nit: Remove this newline.

22 ↗

(On Diff #274236)

Please trim all of the headers here that you can.

28 ↗

(On Diff #274236)

Please do not having using statements like this in a header file.

31 ↗

(On Diff #274236)

Please properly wrap this in the mlir namespace.

31 ↗

(On Diff #274236)

Can you please add proper documentation for this class?

32 ↗

(On Diff #274236)

nit: Remove this newline.

34 ↗

(On Diff #274236)

nit: Use ArrayRef here instead of std::vector.

47 ↗

(On Diff #274236)

ArrayRef here as well.

48 ↗

(On Diff #274236)

Seems like you want an OwningModuleRef here and not a ModuleOp.

mlir/tools/mlir-reduce/mlir-reduce.cpp

25

nit: Could you please try to trim some of these headers?

42

Same here, please remove the using llvm.

58

nit: Please use static if the function is meant to be internal.

61

nit: Swap these two conditions and return early:

if (!...)
  return;

73

nit: Remove this newline.

75

nit: Same here.

79

nit: You don't seem to be doing anything with this pass manager.

[mehdi_amini]

If the tools is intended to be extended, I suspect that classes like class Tester should be in a library (and so the header be under the include folder).
The tools folder is only for the generated binary, I'd follow what we do with mlir-opt as a pattern here.

[msifontes]

Updating D82803: Create the MLIR Reduce framework

Apply changes specified in review

• Trim headers
• Replace std:::vector use with llvm::RefArray
• Apply formatting changes
• Add Tester class documentation
• Erase llvm namespace uses

[stephenneuendorffer]

mlir/tools/mlir-reduce/CMakeLists.txt

42–46

DEPENDS is unnecessary. All target_link_libraries are implicitly dependencies.

[msifontes]

Updating D82803: Create the MLIR Reduce framework

Delete unnecessary dependency statement in the CMakeLists.txt file

[tpopp]

mlir/test/CMakeLists.txt

45

Alphabetical order. This is weird because of the dashes and underscores, but in this case, it goes 3 lines up.

mlir/test/mlir-reduce/testcase.mlir

2

%p/testcase.mlir can just be %s

mlir/tools/CMakeLists.txt

8

nit: alphabetic order here also, although the previous line isn't sorted either.

mlir/tools/mlir-reduce/CMakeLists.txt

2

Is this supposed to be here?

25

Alphabetize. Maybe it's not expected with LLVM, but it's a good habit to get into.

mlir/tools/mlir-reduce/mlir-reduce.cpp

49

If possible, using this enum and helpers would be nice. http://google3/third_party/llvm/llvm-project/mlir/include/mlir/Support/LogicalResult.h?l=19

54

parseSourceFile might return a nullptr.

[msifontes]

Applied Review changes

• Alphabetize dependencies
• Replace filename with %s in testcase.mlir
• Use LogicalResult enum for error reporting
• Add check for nullptr parsing result

[mehdi_amini]

In D82803#2121353, @mehdi_amini wrote:

If the tools is intended to be extended, I suspect that classes like class Tester should be in a library (and so the header be under the include folder).
The tools folder is only for the generated binary, I'd follow what we do with mlir-opt as a pattern here.

Ping here, in case you missed it?

[msifontes]

In D82803#2123056, @mehdi_amini wrote:

In D82803#2121353, @mehdi_amini wrote:

If the tools is intended to be extended, I suspect that classes like class Tester should be in a library (and so the header be under the include folder).
The tools folder is only for the generated binary, I'd follow what we do with mlir-opt as a pattern here.

Ping here, in case you missed it?

Thanks Mehdi, I am currently looking at mlir-opt and making the change

[msifontes]

Updating D82803: Create the MLIR Reduce framework

Move Tester.h to include/Support directory

[jpienaar]

In D82803#2123102, @msifontes wrote:

In D82803#2123056, @mehdi_amini wrote:

In D82803#2121353, @mehdi_amini wrote:

If the tools is intended to be extended, I suspect that classes like class Tester should be in a library (and so the header be under the include folder).
The tools folder is only for the generated binary, I'd follow what we do with mlir-opt as a pattern here.

Ping here, in case you missed it?

Thanks Mehdi, I am currently looking at mlir-opt and making the change

I think the proposal is to have a include/mlir/Reducer to mirror TableGen side - at least for me include/mlir/Support is a bit broad. But wait for others comments before moving, I'm not sure if this is what Mehdi ment.

[jpienaar]

Looks like a good start :)

mlir/include/mlir/Support/Tester.h

15 ↗

(On Diff #274511)

You'll need to update this post the move.

37 ↗

(On Diff #274511)

Could we use something other than int and also document what the returns status signifies?

43 ↗

(On Diff #274511)

setTestCase to me would just mean to update whatever is being tested, if this is meant to only be called to update the most reduced, then the function name doesn't convey that to me.

mlir/tools/mlir-reduce/Tester.cpp

44 ↗

(On Diff #274511)

It seems like either you can return (although I don't know what int return signifies here :) ) or use llvm::report_fatal_error here.

mlir/tools/mlir-reduce/mlir-reduce.cpp

36

<input file> to match mlir-opt ?

40

Interestingness ? (or just Testing script as I don't think it is overloaded here and there will only be one).

44

Same as above

47

Output filename for reduced test case ?

51

Nit: How about loadModule?

53

Why do we care? E.g., if it is in a .txt file and still parses it seems fine.

64

I'd just for the load in the main function and then have the processing/reduction in a next. I don't think much is gained in terms of reuse or readabability from loadAndProcessMLIR vs a loadModule and reduceReproducer calls in main.

70

Should we just remove these for now? Seems like nops as passmanager isn't run).

92

Just use report_fatal_errors here for simplicity as you are in the main binary file (removes need for trailing newline too, yes we don't do that consistently in mlir-opt ...)

[mehdi_amini]

I think the proposal is to have a include/mlir/Reducer to mirror TableGen side - at least for me include/mlir/Support is a bit broad. But wait for others comments before moving, I'm not sure if this is what Mehdi ment.

Spot on, thanks!

[msifontes]

Updating D82803: Create the MLIR Reduce framework

Apply requested changes

• Use llvm::report_fatal_error for error reporting
• Use LogicalResult value to indicate the result of the interestingness script
• Change testCase name to mostReduced to improve readability
• Remove loadAndProcessMLIR method in favor of moving the canonicalization pass to a reducer call

[msifontes]

Updating D82803: Create the MLIR Reduce framework

Create include/mlir/Reducer directory and move Tester.h to it

[msifontes]

Modify testing script so that it is not OS specific

[msifontes]

Normalize path separators

[msifontes]

Remove mlir-opt from testing script

Remove mlir-opt call from the testing script to debug the Windows unit test that is currently failing.

[rriddle]

mlir/include/mlir/Reducer/Tester.h

29

nit: Please use /// for top-level comments.

35

ArrayRef is a value-type and const by construction, so you can drop the explicit const &. Also please drop the llvm:: from ArrayRef, it is re-exported in the mlir namespace.

44

nit: Drop the std::move, it doesn't do anything differently from a normal assignment. Operations are non-owning smart pointers, there is no need to move them.

mlir/tools/mlir-reduce/Tester.cpp

34 ↗

(On Diff #274879)

Why is the testCast pushed into the argument list twice?

44 ↗

(On Diff #274879)

nit: return success(result)

mlir/tools/mlir-reduce/mlir-reduce.cpp

53

parseSourceFile already verifies the module, so you can remove this.

87

nit: Please use proper punctuation in comments. (In this case, add a period)

[msifontes]

Add windows specific testing script and test

[msifontes]

Apply requested changes

• Remove module verification
• Remove llvm namespace use and explicit const declaration
• Remove testCase argument duplicate
• Drop move use
• Add proper punctuation to comments

[mehdi_amini]

The implementation mlir/tools/mlir-reduce/Tester.cpp does not seems in the right place: it should be somewhere under lib/

[msifontes]

In D82803#2126922, @mehdi_amini wrote:

The implementation mlir/tools/mlir-reduce/Tester.cpp does not seems in the right place: it should be somewhere under lib/

Understood, working on it

[mehdi_amini]

mlir/tools/mlir-reduce/mlir-reduce.cpp

90

I don't understand the API contract with the tester right now: why do we provide the original source file here?

[msifontes]

Move Tester.cpp to lib/Reducer

[msifontes]

Remove STATIC library declaration

[msifontes]

mlir/tools/mlir-reduce/mlir-reduce.cpp

90

The Tester's run method runs the interestingess script on a given specified file containing a mlir test case variant. This method will be continually invoked every time a transformation is performed on a temporary file containing the new reduced variant. We provide the input file without modification in this call to make sure that it exhibits the interesting behavior. If the initial file does not exhibit the behavior, then no reductions can be performed since any variant won't exhibit the behavior either.

[mehdi_amini]

mlir/tools/mlir-reduce/mlir-reduce.cpp

90

I still don’t understand this: the file was already loaded and you have a module in memory in the tester, why isn’t the tester simply processing the module instead?
Going through file seems inefficient and unnecessary here.

[msifontes]

mlir/tools/mlir-reduce/mlir-reduce.cpp

90

The module is imported into memory to perform the transformations and keep track of the most reduced version. The tester object is the argument to be passed to each reduction pass. The current design uses a user defined testing script which runs with a mlir test case file as an argument to test for the interesting behavior. It would make more sense to run this initial check before setting up the Tester so I can go and make that change. Please let me know if this addresses your concerns or if you know how we could run the interestingness scripts directly on the modules.

[mehdi_amini]

mlir/tools/mlir-reduce/mlir-reduce.cpp

90

The current design uses a user defined testing script which runs with a mlir test case file as an argument to test for the interesting behavior.

Ok, that makes sense that you need to use a file to call the user-provided script, but this does not require to pass a file to the run method as you already have the current most reduced module available, and this is what the test is supposed to use anyway I believe.

[msifontes]

mlir/tools/mlir-reduce/mlir-reduce.cpp

90

The main idea behind having the file as the argument is that the testing script needs a mlir file to interact with as it doesn't have access to the in memory module. I have worked on a utility that could replace this run method: it receives a module as an argument, it then prints the module to a temporary file and runs the testing script using that temporary file as argument. It is important to point out that a module/file is still needed as an argument to the run method to specify which module is being tested. The test is not always run on the most reduced test case. For example, multiple variants can be created at any given point in a pass and those need to be tested and sorted before deciding which is the shortest successful one that will replace the most reduced module.

[mehdi_amini]

mlir/tools/mlir-reduce/mlir-reduce.cpp

90

OK, that makes sense, but I think you need *a lot* more description on the Tester class to capture its intended usage, lifetime, etc.

[msifontes]

Add Tester class documentation

[msifontes]

mlir/tools/mlir-reduce/mlir-reduce.cpp

90

I have added documentation that I believe clears up the ambiguity in the use of the Tester class. Let me know if there is something that is still not clear form the description.

[mehdi_amini]

mlir/tools/mlir-reduce/mlir-reduce.cpp

90

Looks good for now, thanks for your patience in the back and forth!

[jpienaar]

Nice, looking good

mlir/lib/Reducer/Tester.cpp

35

I dislike using LogicalResult for cases which are just queries. E.g., here failure doesn't mean something exceptional or that this function didn't work/failed. If we aren't returning for failures here (e.g., return failure if result < 0), then it seems like this is a query function:

bool isInteresting(StringRef testCase)

then you also get something like

while (tester.isInteresting(test)) tester.reduce(test)

and then don't have to explain what failure means here as the logic is captured in the name.

49

I think we can set this to not generate the debug dump here (as this is due to a bug in the interestingness test so the stack trace reported from here wouldn't be as useful).

mlir/test/mlir-reduce/testcase.mlir

4

Can you have both in the same file? I would have expected the two requires to both apply and so nothing runs.

[msifontes]

Apply requested changes

• Change name of run method to isInteresting
• Change return type of run method to bool
• Remove debug dump generation for interestingness error
• Split test case files for OS specific tests

[msifontes]

This comment was removed by msifontes.

[jpienaar]

Looks good thanks

mlir/include/mlir/Reducer/Tester.h

12

Nit: it feels as if parts of this is very specific to class methods below and more useful at that point (e.g., here I don't know it has a run method and perhaps I shouldn't care yet vs knowing it has a way to test interestingness).

Also run method is no more.

[msifontes]

Fix Tester.cpp dependency issue

[msifontes]

Modify Tester class documentation

[rriddle]

mlir/lib/Reducer/Tester.cpp

35

This is marked as done, but I don't this pattern used anywhere? Are you missing an upload?

[msifontes]

mlir/lib/Reducer/Tester.cpp

35

I call this method in line 89 of mlir-reduce.cpp, is this what you are referring to?

[msifontes]

Remove the windows test case

In order to isolate the build failure cause due to the permission denied when running the testing script.

[msifontes]

Remove windows specific test case

[msifontes]

Add windows specific testcase

[mehdi_amini]

reverted in e10e034f4bb7d3c639cc65a8d60147ff1e330e56

[msifontes]

Remove unsupported windows tests

[mehdi_amini]

mlir/test/mlir-reduce/testcase-linux.mlir

14 ↗

(On Diff #276263)

Please terminate files with a new line.

Also what is linux specific to all this?

[msifontes]

mlir/test/mlir-reduce/testcase-linux.mlir

14 ↗

(On Diff #276263)

The execution of the shell script was only working on linux. I tried using a .bat script for windows but the build test was having trouble running the script due to permissions. I will continue to look into how we can test on windows.

mlir/include/mlir/Reducer/Tester.h

+//===- Tester.h -------------------------------------------------*- 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 file defines the Tester class used in the MLIR Reduce tool.
+//
+// A Tester object is passed as an argument to the reduction passes and it is
+// used to keep track of the state of the reduction throughout the multiple
+// passes. The testScript and testScriptArgs specify the filename of the
+// interestingness script and its respective arguments. The mostReduced module
+// is used to keep track of the most reduced variant of the test case generated
+// up to any given point in time.
+//
+// The run method allows the different reduction passes to test the interesting
+// behavior of a generated variant. If a generated variant is found to be
+// successful and shorter than the mostReduced module, the mostReduced module
+// must be updated with the new variant.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MLIR_REDUCER_TESTER_H
+#define MLIR_REDUCER_TESTER_H
+
+#include <vector>
+
+#include "mlir/IR/Module.h"
+#include "llvm/ADT/SmallString.h"
+#include "llvm/Support/Error.h"
+#include "llvm/Support/FileSystem.h"
+#include "llvm/Support/Program.h"
+
+namespace mlir {
+
+/// This class is used to keep track of the state of the reduction. It contains
+/// a method to run the interestingness testing script on MLIR test case files
+/// and provides functionality to track the most reduced test case.
+class Tester {
+public:
+  Tester(StringRef testScript, ArrayRef<std::string> testScriptArgs);
+
+  /// Runs the interestingness testing script on a MLIR test case file.
+  LogicalResult run(StringRef testCase);
+
+  /// Returns the most reduced MLIR test case module.
+  ModuleOp getMostReduced() const { return mostReduced; }
+
+  /// Updates the most reduced MLIR test case module.
+  void setMostReduced(ModuleOp t) { mostReduced = t; }
+
+private:
+  StringRef testScript;
+  ArrayRef<std::string> testScriptArgs;
+  ModuleOp mostReduced;
+};
+
+} // end namespace mlir
+
+#endif
 No newline at end of file

mlir/lib/CMakeLists.txt

 # Enable errors for any global constructors.
 add_flag_if_supported("-Werror=global-constructors" WERROR_GLOBAL_CONSTRUCTOR)
 
 add_subdirectory(Analysis)
 add_subdirectory(Conversion)
 add_subdirectory(Dialect)
 add_subdirectory(EDSC)
 add_subdirectory(ExecutionEngine)
 add_subdirectory(IR)
 add_subdirectory(Interfaces)
 add_subdirectory(Parser)
 add_subdirectory(Pass)
+add_subdirectory(Reducer)
 add_subdirectory(Support)
 add_subdirectory(TableGen)
 add_subdirectory(Target)
 add_subdirectory(Transforms)
-add_subdirectory(Translation)
+add_subdirectory(Translation)
 No newline at end of file

mlir/lib/Reducer/CMakeLists.txt

+llvm_add_library(MLIRReduce
+  Tester.cpp
+)
+
+mlir_check_all_link_libraries(MLIRReduce)
+
+add_mlir_library_install(MLIRReduce)
 No newline at end of file

mlir/lib/Reducer/Tester.cpp

+//===- Tester.cpp ---------------------------------------------------------===//
+//
+// 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 file defines the Tester class used in the MLIR Reduce tool.
+//
+// A Tester object is passed as an argument to the reduction passes and it is
+// used to keep track of the state of the reduction throughout the multiple
+// passes. The testScript and testScriptArgs specify the filename of the
+// interestingness script and its respective arguments. The mostReduced module
+// is used to keep track of the most reduced variant of the test case generated
+// up to any given point in time.
+//
+// The run method allows the different reduction passes to test the interesting
+// behavior of a generated variant. If a generated variant is found to be
+// successful and shorter than the mostReduced module, the mostReduced module
+// must be updated with the new variant.
+//
+//===----------------------------------------------------------------------===//
+
+#include "mlir/Reducer/Tester.h"
+
+using namespace mlir;
+
+Tester::Tester(StringRef scriptName, ArrayRef<std::string> scriptArgs)
+    : testScript(scriptName), testScriptArgs(scriptArgs) {}
+
+/// Runs the interestingness testing script on a MLIR test case file
+/// Returns success() if the interesting behavior is present in the test case or
+/// failure() otherwise
+LogicalResult Tester::run(StringRef testCase) {
+
+  std::vector<StringRef> testerArgs;
+  testerArgs.push_back(testCase);
+
+  for (const std::string &arg : testScriptArgs)
+    testerArgs.push_back(arg);
+
+  std::string errMsg;
+  int result = llvm::sys::ExecuteAndWait(
+      testScript, testerArgs, /*Env=*/None, /*Redirects=*/None,
+      /*SecondsToWait=*/0, /*MemoryLimit=*/0, &errMsg);
+
+  if (result < 0)
+    llvm::report_fatal_error("Error running interestingness test: " + errMsg);
+
+  if (!result)
+    return failure();
+
+  return success(result);
+}

mlir/test/CMakeLists.txt

 
 set(MLIR_TEST_DEPENDS
   FileCheck count not
   MLIRUnitTests
   mlir-cpu-runner
   mlir-edsc-builder-api-test
   mlir-linalg-ods-gen
   mlir-opt
+  mlir-reduce
   mlir-sdbm-api-test
   mlir-tblgen
   mlir-translate
   mlir_test_cblas
   mlir_test_cblas_interface
   mlir_runner_utils
   mlir_c_runner_utils
   )
 
 if(LLVM_BUILD_EXAMPLES)
   list(APPEND MLIR_TEST_DEPENDS
     toyc-ch1

mlir/test/mlir-reduce/test.bat

+#Replicate bug
+
+#Interesting behavior
+exit 1
 No newline at end of file

mlir/test/mlir-reduce/test.sh

+#!/bin/bash
+
+#Replicate bug
+
+#Interesting behavior
+exit 1
 No newline at end of file

mlir/test/mlir-reduce/testcase.mlir

+// REQUIRES: linux
+// RUN: mlir-reduce %s -test %S/test.sh
+// REQUIRES: windows
+// RUN: mlir-reduce %s -test %S/test.bat
+
+func @simple1(%arg0: i1, %arg1: memref<2xf32>, %arg2: memref<2xf32>) {
+  cond_br %arg0, ^bb1, ^bb2
+^bb1:
+  br ^bb3(%arg1 : memref<2xf32>)
+^bb2:
+  %0 = alloc() : memref<2xf32>
+  br ^bb3(%0 : memref<2xf32>)
+^bb3(%1: memref<2xf32>):
+  return
+}
+
+// -----

mlir/tools/CMakeLists.txt

 add_subdirectory(mlir-cuda-runner)
 add_subdirectory(mlir-cpu-runner)
 add_subdirectory(mlir-linalg-ods-gen)
 add_subdirectory(mlir-opt)
+add_subdirectory(mlir-reduce)
 add_subdirectory(mlir-rocm-runner)
-add_subdirectory(mlir-translate)
-add_subdirectory(mlir-vulkan-runner)
 add_subdirectory(mlir-shlib)
+add_subdirectory(mlir-translate)
+add_subdirectory(mlir-vulkan-runner)
 No newline at end of file

mlir/tools/mlir-reduce/CMakeLists.txt

+get_property(dialect_libs GLOBAL PROPERTY MLIR_DIALECT_LIBS)
+get_property(conversion_libs GLOBAL PROPERTY MLIR_CONVERSION_LIBS)
+set(LLVM_LINK_COMPONENTS
+  AllTargetsAsmParsers
+  AllTargetsCodeGens
+  AllTargetsDescs
+  AllTargetsInfos
+  AsmParser
+  Core
+  IRReader
+  Support
+  Target
+  TransformUtils
+  )
+
+set(LIBS
+  ${dialect_libs}
+  ${conversion_libs}
+  ${test_libs}
+  MLIRAnalysis
+  MLIRDialect
+  MLIREDSC
+  MLIRIR
+  MLIRLoopAnalysis
+  MLIROptLib
+  MLIRParser
+  MLIRPass
+  MLIRReduce
+  MLIRSupport
+  MLIRTransforms
+  MLIRTransformUtils
+  )
+
+add_llvm_tool(mlir-reduce
+  mlir-reduce.cpp
+  )
+
+target_link_libraries(mlir-reduce PRIVATE ${LIBS})
+llvm_update_compile_flags(mlir-reduce)
+
+mlir_check_all_link_libraries(mlir-reduce)
 No newline at end of file

mlir/tools/mlir-reduce/mlir-reduce.cpp

+//===- mlir-reduce.cpp - The MLIR reducer ---------------------------------===//
+//
+// 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 file implements the general framework of the MLIR reducer tool. It
+// parses the command line arguments, parses the initial MLIR test case and sets
+// up the testing environment. It  outputs the most reduced test case variant
+// after executing the reduction passes.
+//
+//===----------------------------------------------------------------------===//
+
+#include <vector>
+
+#include "mlir/InitAllDialects.h"
+#include "mlir/Parser.h"
+#include "mlir/Pass/Pass.h"
+#include "mlir/Pass/PassManager.h"
+#include "mlir/Reducer/Tester.h"
+#include "mlir/Support/FileUtilities.h"
+#include "mlir/Support/LogicalResult.h"
+#include "mlir/Transforms/Passes.h"
+#include "llvm/Support/InitLLVM.h"
+#include "llvm/Support/ToolOutputFile.h"
+
+using namespace mlir;
+
+static llvm::cl::opt<std::string> inputFilename(llvm::cl::Positional,
+                                                llvm::cl::Required,
+                                                llvm::cl::desc("<input file>"));
+
+static llvm::cl::opt<std::string>
+    testFilename("test", llvm::cl::Required, llvm::cl::desc("Testing script"));
+
+static llvm::cl::list<std::string>
+    testArguments("test-args", llvm::cl::ZeroOrMore,
+                  llvm::cl::desc("Testing script arguments"));
+
+static llvm::cl::opt<std::string>
+    outputFilename("o",
+                   llvm::cl::desc("Output filename for the reduced test case"),
+                   llvm::cl::init("-"));
+
+// Parse and verify the input MLIR file.
+static LogicalResult loadModule(MLIRContext &context, OwningModuleRef &module,
+                                StringRef inputFilename) {
+  module = parseSourceFile(inputFilename, &context);
+  if (!module)
+    return failure();
+
+  return success();
+}
+
+int main(int argc, char **argv) {
+
+  llvm::InitLLVM y(argc, argv);
+
+  registerAllDialects();
+  registerMLIRContextCLOptions();
+  registerPassManagerCLOptions();
+
+  llvm::cl::ParseCommandLineOptions(argc, argv,
+                                    "MLIR test case reduction tool.\n");
+
+  std::string errorMessage;
+
+  auto testscript = openInputFile(testFilename, &errorMessage);
+  if (!testscript)
+    llvm::report_fatal_error(errorMessage);
+
+  auto output = openOutputFile(outputFilename, &errorMessage);
+  if (!output)
+    llvm::report_fatal_error(errorMessage);
+
+  mlir::MLIRContext context;
+  mlir::OwningModuleRef moduleRef;
+  context.allowUnregisteredDialects(true);
+
+  if (failed(loadModule(context, moduleRef, inputFilename)))
+    llvm::report_fatal_error("Input test case can't be parsed");
+
+  // Initialize test environment.
+  Tester test(testFilename, testArguments);
+  test.setMostReduced(moduleRef.get());
+
+  if (failed(test.run(inputFilename)))
+    llvm::report_fatal_error(
+        "Input test case does not exhibit interesting behavior");
+
+  moduleRef.get().print(output->os());
+  output->keep();
+
+  return 0;
+}
 No newline at end of file