[mlir] [integration-test] [VectorOps] Start an integration test directory for MLIR

[aartbik]

Summary

This CL introduces an integration test directory for MLIR in general, with
vector dialect integration tests in particular as a first working suite. To
run all the integration tests (and currently just the vector suite):

$ cmake --build . --target check-mlir-integration
[0/1] Running the MLIR integration tests
Testing Time: 0.24s
Passed: 22

The general call is to contribute to this integration test directory with more
tests and other suites, running end-to-end examples that may be too heavy for
the regular test directory, but should be tested occasionally to verify the
health of MLIR.

Background discussion at:
https://llvm.discourse.group/t/vectorops-rfc-add-suite-of-integration-tests-for-vector-dialect-operations/1213/

[aartbik]

Please have an extra careful look at the CMakeLists.txt and lit.cfg.py/lit.site.cfg.py.in files. I have no experience setting up a new testing directory inside the llvm repo, so most of this is a bit trail and error, and I welcome any constructive feedback on the setup.

[mehdi_amini]

Testing Time: 0.24s

As long as it is negligible, maybe we should enable the flag by default?

The separation unit-tests vs integration tests is nice so that we can still require bugfixes to come with a unit-tests.

[mehdi_amini]

LGTM overall (adding @stephenneuendorffer who spent time staring at CMake recently)

You likely need to disable these on windows (see mlir/test/mlir-cpu-runner/lit.local.cfg for an example) because of the lack of JIT support right now.

mlir/CMakeLists.txt

69

Actually I thought you'd introduce another CMakeOption MLIR_INCLUDE_INTEGRATION_TESTS to guard this (even if we enable it by default for now)

mlir/integration_test/CMakeLists.txt

4

I suspect this is a copy/paste from somewhere else?

[aartbik]

addressed first round of comments, disabled windows run

[aartbik]

In D81626#2086628, @mehdi_amini wrote:

You likely need to disable these on windows (see mlir/test/mlir-cpu-runner/lit.local.cfg for an example) because of the lack of JIT support right now.

Yes, sharp eye! All the window tests failed. Hopefull the lit.local.cfg in CPU avoids that now.

mlir/CMakeLists.txt

69

Actually, the target is defined as "check-integration-mlir" in the new CMakeList.txt I though this was just adding the rules if MLIR_INCLUDE_TESTS is set. That by itself is not a bad thing, right, or is it?

Or does this force the run? If so, what should this be?

mlir/integration_test/CMakeLists.txt

4

Yes, I had this as "look later if you need it" and forgot to remove.

[mehdi_amini]

mlir/CMakeLists.txt

69

when you do add_lit_testsuites(check-integration-mlir without passing the option EXCLUDE_FROM_CHECK_ALL, then it is added by default as a dependency to check.

Here seems the most natural branching point to me (MLIR_INCLUDE_TESTS / MLIR_INCLUDE_INTEGRATION_TESTS)

[stephenneuendorffer]

LGTM. My main comment is that I don't know if I see the value of *LOTS* of integration tests. I'm used to using tests like this for design drivers, but it seems like you've tried to get some coverage of the vector dialect here?

mlir/integration_test/CMakeLists.txt

23

"check-mlir-integration" is more natural to me, since it exists under mlir/ Or do you expect that "check-integration" will be introduced as a dependence in other projects?

mlir/integration_test/Dialect/Vector/CPU/test-shuffle.mlir

7–25

Seems pretty simple for an integration test?

[stephenneuendorffer]

mlir/integration_test/CMakeLists.txt

25

I think "check-integration-mlir' here needs to be 'MLIR-INTEGRATION' matching the string in lit.cfg.py

mlir/integration_test/lit.cfg.py

19

This string should be referred to in add_lit_testsuites()

[aartbik]

mlir/integration_test/CMakeLists.txt

23

No, I just picked one. I changed it into what you suggested.

25

Done. Note that the MLIR_INTEGRATION_TEST_DEPS was inspired by the existing code as well, but looks undefined? Shouldn't this be MLIR_INTEGRATION_TEST_DEPENDS? I have changed it for now, please let me know if this is right or wrong?

mlir/integration_test/Dialect/Vector/CPU/test-shuffle.mlir

7–25

The initial suite is aimed at having one integration test per vector dialect operation that runs end-to-end from mlir file to llvm ir running in JIT. They are not necessarily exhaustive unit tests for the operation, and besides testing, also serve to supplement the documentation with real working examples. Of course, I expect the tests to grow organically. Also, by moving this to the public repo, our hope is to see lots of external contributions as well :-)

[aartbik]

addressed second round of comments

[aartbik]

renamed substitution variable to reflect "integration" location better

[aartbik]

Any more feedback? Can I proceed with the new framework?

[stephenneuendorffer]

Thanks for this!

[nicolasvasilache]

Thanks Aart!

[aartbik]

rebased

mlir/CMakeLists.txt

 set(MLIR_CUDA_RUNNER_ENABLED 0 CACHE BOOL "Enable building the mlir CUDA runner")
 set(MLIR_ROCM_RUNNER_ENABLED 0 CACHE BOOL "Enable building the mlir ROCm runner")
 set(MLIR_VULKAN_RUNNER_ENABLED 0 CACHE BOOL "Enable building the mlir Vulkan runner")
 
 option(MLIR_INCLUDE_TESTS
        "Generate build targets for the MLIR unit tests."
        ${LLVM_INCLUDE_TESTS})
 
+option(MLIR_INCLUDE_INTEGRATION_TESTS
+       "Generate build targets for the MLIR integration tests."
+       ${LLVM_INCLUDE_TESTS})
+
 include_directories( "include")
 include_directories( ${MLIR_INCLUDE_DIR})
 
 # Adding tools/mlir-tblgen here as calling add_tablegen sets some variables like
 # MLIR_TABLEGEN_EXE in PARENT_SCOPE which gets lost if that folder is included
 # from another directory like tools
 add_subdirectory(tools/mlir-tblgen)
 
 add_subdirectory(include/mlir)
 add_subdirectory(lib)
 if (MLIR_INCLUDE_TESTS)
   add_definitions(-DMLIR_INCLUDE_TESTS)
   add_subdirectory(unittests)
   add_subdirectory(test)
 endif()
+if (MLIR_INCLUDE_INTEGRATION_TESTS)
+  add_definitions(-DMLIR_INCLUDE_INTEGRATION_TESTS)
+  add_subdirectory(integration_test)
+endif()
 # Tools needs to come late to ensure that MLIR_ALL_LIBS is populated.
 # Generally things after this point may depend on MLIR_ALL_LIBS or libMLIR.so.
 add_subdirectory(tools)
 
 if( LLVM_INCLUDE_EXAMPLES )
   add_subdirectory(examples)
 endif()
 

mlir/integration_test/CMakeLists.txt

+# Passed to lit.site.cfg.py.in to set up the path where to find the libraries.
+set(MLIR_INTEGRATION_TEST_DIR ${CMAKE_LIBRARY_OUTPUT_DIRECTORY})
+
+set(MLIR_INTEGRATION_TEST_DEPENDS
+  FileCheck count not
+  mlir-cpu-runner
+  mlir-opt
+  mlir_runner_utils
+  mlir_c_runner_utils
+  )
+
+configure_lit_site_cfg(
+  ${CMAKE_CURRENT_SOURCE_DIR}/lit.site.cfg.py.in
+  ${CMAKE_CURRENT_BINARY_DIR}/lit.site.cfg.py
+  ${CMAKE_CURRENT_SOURCE_DIR}/lit.cfg.py
+  )
+
+add_lit_testsuite(check-mlir-integration "Running the MLIR integration tests"
+  ${CMAKE_CURRENT_BINARY_DIR}
+  DEPENDS ${MLIR_INTEGRATION_TEST_DEPENDS}
+  )
+set_target_properties(check-mlir-integration PROPERTIES FOLDER "integration_test")
+
+add_lit_testsuites(MLIR_INTEGRATION ${CMAKE_CURRENT_SOURCE_DIR}
+  DEPENDS ${MLIR_INTEGRATION_TEST_DEPENDS}
+  )

mlir/integration_test/Dialect/Vector/CPU/lit.local.cfg

+import sys
+
+# No JIT on win32.
+if sys.platform == 'win32':
+    config.unsupported = True

mlir/integration_test/Dialect/Vector/CPU/test-broadcast.mlir

+// RUN: mlir-opt %s -convert-scf-to-std -convert-vector-to-llvm -convert-std-to-llvm | \
+// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
+// RUN:   -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+
+func @entry() {
+  %i = constant 2147483647: i32
+  %l = constant 9223372036854775807 : i64
+
+  %f0 = constant 0.0: f32
+  %f1 = constant 1.0: f32
+  %f2 = constant 2.0: f32
+  %f3 = constant 3.0: f32
+  %f4 = constant 4.0: f32
+  %f5 = constant 5.0: f32
+
+  // Test simple broadcasts.
+  %vi = vector.broadcast %i : i32 to vector<2xi32>
+  %vl = vector.broadcast %l : i64 to vector<2xi64>
+  %vf = vector.broadcast %f1 : f32 to vector<2x2x2xf32>
+  vector.print %vi : vector<2xi32>
+  vector.print %vl : vector<2xi64>
+  vector.print %vf : vector<2x2x2xf32>
+  // CHECK: ( 2147483647, 2147483647 )
+  // CHECK: ( 9223372036854775807, 9223372036854775807 )
+  // CHECK: ( ( ( 1, 1 ), ( 1, 1 ) ), ( ( 1, 1 ), ( 1, 1 ) ) )
+
+  // Test "duplication" in leading dimensions.
+  %v0 = vector.broadcast %f1 : f32 to vector<4xf32>
+  %v1 = vector.insert %f2, %v0[1] : f32 into vector<4xf32>
+  %v2 = vector.insert %f3, %v1[2] : f32 into vector<4xf32>
+  %v3 = vector.insert %f4, %v2[3] : f32 into vector<4xf32>
+  %v4 = vector.broadcast %v3 : vector<4xf32> to vector<3x4xf32>
+  %v5 = vector.broadcast %v3 : vector<4xf32> to vector<2x2x4xf32>
+  vector.print %v3 : vector<4xf32>
+  vector.print %v4 : vector<3x4xf32>
+  vector.print %v5 : vector<2x2x4xf32>
+  // CHECK: ( 1, 2, 3, 4 )
+  // CHECK: ( ( 1, 2, 3, 4 ), ( 1, 2, 3, 4 ), ( 1, 2, 3, 4 ) )
+  // CHECK: ( ( ( 1, 2, 3, 4 ), ( 1, 2, 3, 4 ) ), ( ( 1, 2, 3, 4 ), ( 1, 2, 3, 4 ) ) )
+
+  // Test straightforward "stretch" on a 1-D "scalar".
+  %x = vector.broadcast %f5 : f32 to vector<1xf32>
+  %y = vector.broadcast %x  : vector<1xf32> to vector<8xf32>
+  vector.print %y : vector<8xf32>
+  // CHECK : ( 5, 5, 5, 5, 5, 5, 5, 5 )
+
+  // Test "stretch" in leading dimension.
+  %s = vector.broadcast %v3 : vector<4xf32> to vector<1x4xf32>
+  %t = vector.broadcast %s  : vector<1x4xf32> to vector<3x4xf32>
+  vector.print %s : vector<1x4xf32>
+  vector.print %t : vector<3x4xf32>
+  // CHECK: ( ( 1, 2, 3, 4 ) )
+  // CHECK: ( ( 1, 2, 3, 4 ), ( 1, 2, 3, 4 ), ( 1, 2, 3, 4 ) )
+
+  // Test "stretch" in trailing dimension.
+  %a0 = vector.broadcast %f1 : f32 to vector<3x1xf32>
+  %a1 = vector.insert %f2, %a0[1, 0] : f32 into vector<3x1xf32>
+  %a2 = vector.insert %f3, %a1[2, 0] : f32 into vector<3x1xf32>
+  %a3 = vector.broadcast %a2 : vector<3x1xf32> to vector<3x4xf32>
+  vector.print %a2 : vector<3x1xf32>
+  vector.print %a3 : vector<3x4xf32>
+  // CHECK: ( ( 1 ), ( 2 ), ( 3 ) )
+  // CHECK: ( ( 1, 1, 1, 1 ), ( 2, 2, 2, 2 ), ( 3, 3, 3, 3 ) )
+
+  // Test "stretch" in middle dimension.
+  %m0 = vector.broadcast %f0 : f32 to vector<3x1x2xf32>
+  %m1 = vector.insert %f1, %m0[0, 0, 1] : f32 into vector<3x1x2xf32>
+  %m2 = vector.insert %f2, %m1[1, 0, 0] : f32 into vector<3x1x2xf32>
+  %m3 = vector.insert %f3, %m2[1, 0, 1] : f32 into vector<3x1x2xf32>
+  %m4 = vector.insert %f4, %m3[2, 0, 0] : f32 into vector<3x1x2xf32>
+  %m5 = vector.insert %f5, %m4[2, 0, 1] : f32 into vector<3x1x2xf32>
+  %m6 = vector.broadcast %m5 : vector<3x1x2xf32> to vector<3x4x2xf32>
+  vector.print %m5 : vector<3x1x2xf32>
+  vector.print %m6 : vector<3x4x2xf32>
+  // CHECK: ( ( ( 0, 1 ) ), ( ( 2, 3 ) ), ( ( 4, 5 ) ) )
+  // CHECK: ( ( ( 0, 1 ), ( 0, 1 ), ( 0, 1 ), ( 0, 1 ) ),
+  // CHECK-SAME: ( ( 2, 3 ), ( 2, 3 ), ( 2, 3 ), ( 2, 3 ) ),
+  // CHECK-SAME: ( ( 4, 5 ), ( 4, 5 ), ( 4, 5 ), ( 4, 5 ) ) )
+
+  return
+}

mlir/integration_test/Dialect/Vector/CPU/test-constant-mask.mlir

+// RUN: mlir-opt %s -convert-scf-to-std -convert-vector-to-llvm -convert-std-to-llvm | \
+// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
+// RUN:   -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+
+func @entry() {
+  %0 = vector.constant_mask [4] : vector<8xi1>
+  vector.print %0 : vector<8xi1>
+  // CHECK: ( 1, 1, 1, 1, 0, 0, 0, 0 )
+
+  %1 = vector.constant_mask [1, 3] : vector<4x4xi1>
+  vector.print %1 : vector<4x4xi1>
+  // CHECK: ( ( 1, 1, 1, 0 ), ( 0, 0, 0, 0 ), ( 0, 0, 0, 0 ), ( 0, 0, 0, 0 ) )
+
+  %2 = vector.constant_mask [2, 2] : vector<4x4xi1>
+  vector.print %2 : vector<4x4xi1>
+  // CHECK: ( ( 1, 1, 0, 0 ), ( 1, 1, 0, 0 ), ( 0, 0, 0, 0 ), ( 0, 0, 0, 0 ) )
+
+  %3 = vector.constant_mask [2, 4] : vector<4x4xi1>
+  vector.print %3 : vector<4x4xi1>
+  // CHECK: ( ( 1, 1, 1, 1 ), ( 1, 1, 1, 1 ), ( 0, 0, 0, 0 ), ( 0, 0, 0, 0 ) )
+
+  %4 = vector.constant_mask [3, 1] : vector<4x4xi1>
+  vector.print %4 : vector<4x4xi1>
+  // CHECK: ( ( 1, 0, 0, 0 ), ( 1, 0, 0, 0 ), ( 1, 0, 0, 0 ), ( 0, 0, 0, 0 ) )
+
+  %5 = vector.constant_mask [3, 2] : vector<4x4xi1>
+  vector.print %5 : vector<4x4xi1>
+  // CHECK: ( ( 1, 1, 0, 0 ), ( 1, 1, 0, 0 ), ( 1, 1, 0, 0 ), ( 0, 0, 0, 0 ) )
+
+  %6 = vector.constant_mask [4, 3] : vector<4x4xi1>
+  vector.print %6 : vector<4x4xi1>
+  // CHECK: ( ( 1, 1, 1, 0 ), ( 1, 1, 1, 0 ), ( 1, 1, 1, 0 ), ( 1, 1, 1, 0 ) )
+
+  %7 = vector.constant_mask [4, 4] : vector<4x4xi1>
+  vector.print %7 : vector<4x4xi1>
+  // CHECK: ( ( 1, 1, 1, 1 ), ( 1, 1, 1, 1 ), ( 1, 1, 1, 1 ), ( 1, 1, 1, 1 ) )
+
+  %8 = vector.constant_mask [1, 2, 3] : vector<2x3x4xi1>
+  vector.print %8 : vector<2x3x4xi1>
+  // CHECK: ( ( ( 1, 1, 1, 0 ), ( 1, 1, 1, 0 ), ( 0, 0, 0, 0 ) ), ( ( 0, 0, 0, 0 ), ( 0, 0, 0, 0 ), ( 0, 0, 0, 0 ) ) )
+
+  %9 = vector.constant_mask [2, 2, 3] : vector<2x3x4xi1>
+  vector.print %9 : vector<2x3x4xi1>
+  // CHECK: ( ( ( 1, 1, 1, 0 ), ( 1, 1, 1, 0 ), ( 0, 0, 0, 0 ) ), ( ( 1, 1, 1, 0 ), ( 1, 1, 1, 0 ), ( 0, 0, 0, 0 ) ) )
+
+  return
+}
+

mlir/integration_test/Dialect/Vector/CPU/test-contraction.mlir

+// RUN: mlir-opt %s -convert-scf-to-std -convert-vector-to-llvm -convert-std-to-llvm | \
+// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
+// RUN:   -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+
+#dotp_accesses = [
+  affine_map<(i) -> (i)>,
+  affine_map<(i) -> (i)>,
+  affine_map<(i) -> ()>
+]
+#dotp_trait = {
+  indexing_maps = #dotp_accesses,
+  iterator_types = ["reduction"]
+}
+
+#matvec_accesses = [
+  affine_map<(i, j) -> (i, j)>,
+  affine_map<(i, j) -> (j)>,
+  affine_map<(i, j) -> (i)>
+]
+#matvec_trait = {
+  indexing_maps = #matvec_accesses,
+  iterator_types = ["parallel", "reduction"]
+}
+
+#mattransvec_accesses = [
+  affine_map<(i, j) -> (j, i)>,
+  affine_map<(i, j) -> (j)>,
+  affine_map<(i, j) -> (i)>
+]
+#mattransvec_trait = {
+  indexing_maps = #mattransvec_accesses,
+  iterator_types = ["parallel", "reduction"]
+}
+
+#matmat_accesses = [
+  affine_map<(i, j, k) -> (i, k)>,
+  affine_map<(i, j, k) -> (k, j)>,
+  affine_map<(i, j, k) -> (i, j)>
+]
+#matmat_trait = {
+  indexing_maps = #matmat_accesses,
+  iterator_types = ["parallel", "parallel", "reduction"]
+}
+
+#mattransmat_accesses = [
+  affine_map<(i, j, k) -> (k, i)>,
+  affine_map<(i, j, k) -> (k, j)>,
+  affine_map<(i, j, k) -> (i, j)>
+]
+#mattransmat_trait = {
+  indexing_maps = #mattransmat_accesses,
+  iterator_types = ["parallel", "parallel", "reduction"]
+}
+
+#matmattrans_accesses = [
+  affine_map<(i, j, k) -> (i, k)>,
+  affine_map<(i, j, k) -> (j, k)>,
+  affine_map<(i, j, k) -> (i, j)>
+]
+#matmattrans_trait = {
+  indexing_maps = #matmattrans_accesses,
+  iterator_types = ["parallel", "parallel", "reduction"]
+}
+
+#mattransmattrans_accesses = [
+  affine_map<(i, j, k) -> (k, i)>,
+  affine_map<(i, j, k) -> (j, k)>,
+  affine_map<(i, j, k) -> (i, j)>
+]
+#mattransmattrans_trait = {
+  indexing_maps = #mattransmattrans_accesses,
+  iterator_types = ["parallel", "parallel", "reduction"]
+}
+
+#matmat_then_trans_accesses = [
+  affine_map<(i, j, k) -> (i, k)>,
+  affine_map<(i, j, k) -> (k, j)>,
+  affine_map<(i, j, k) -> (j, i)>
+]
+#matmat_then_trans_trait = {
+  indexing_maps = #matmat_then_trans_accesses,
+  iterator_types = ["parallel", "parallel", "reduction"]
+}
+
+#contract2d_accesses = [
+  affine_map<(i, j) -> (i, j)>,
+  affine_map<(i, j) -> (i, j)>,
+  affine_map<(i, j) -> ()>
+]
+#contract2d_trait = {
+  indexing_maps = #contract2d_accesses,
+  iterator_types = ["reduction", "reduction"]
+}
+
+#contract2d_alt_accesses = [
+  affine_map<(i, j) -> (j, i)>,
+  affine_map<(i, j) -> (j, i)>,
+  affine_map<(i, j) -> ()>
+]
+#contract2d_alt_trait = {
+  indexing_maps = #contract2d_alt_accesses,
+  iterator_types = ["reduction", "reduction"]
+}
+
+#contract2d_trans_accesses = [
+  affine_map<(i, j) -> (i, j)>,
+  affine_map<(i, j) -> (j, i)>,
+  affine_map<(i, j) -> ()>
+]
+#contract2d_trans_trait = {
+  indexing_maps = #contract2d_trans_accesses,
+  iterator_types = ["reduction", "reduction"]
+}
+
+#contract2d_trans_alt_accesses = [
+  affine_map<(i, j) -> (j, i)>,
+  affine_map<(i, j) -> (i, j)>,
+  affine_map<(i, j) -> ()>
+]
+#contract2d_trans_alt_trait = {
+  indexing_maps = #contract2d_trans_alt_accesses,
+  iterator_types = ["reduction", "reduction"]
+}
+
+#column_major_matmat_accesses = [
+  affine_map<(i, j, k) -> (k, j)>,
+  affine_map<(i, j, k) -> (i, k)>,
+  affine_map<(i, j, k) -> (j, i)>
+]
+#column_major_matmat_trait = {
+  indexing_maps = #column_major_matmat_accesses,
+  iterator_types = ["parallel", "parallel", "reduction"]
+}
+
+func @entry() {
+  %f0 = constant 0.0: f32
+  %f1 = constant 1.0: f32
+  %f2 = constant 2.0: f32
+  %f3 = constant 3.0: f32
+  %f4 = constant 4.0: f32
+  %f5 = constant 5.0: f32
+  %f6 = constant 6.0: f32
+  %f7 = constant 7.0: f32
+  %f8 = constant 8.0: f32
+
+  // Zero vectors.
+  %z1 = vector.broadcast %f0 : f32 to vector<2xf32>
+  %z2 = vector.broadcast %f0 : f32 to vector<2x2xf32>
+  %z3 = vector.broadcast %f0 : f32 to vector<3x4xf32>
+
+  // Construct test vectors.
+  %0 = vector.broadcast %f1 : f32 to vector<2xf32>
+  %a = vector.insert %f2, %0[1] : f32 into vector<2xf32>
+  %1 = vector.broadcast %f3 : f32 to vector<2xf32>
+  %b = vector.insert %f4, %1[1] : f32 into vector<2xf32>
+  %2 = vector.broadcast %f5 : f32 to vector<2xf32>
+  %c = vector.insert %f6, %2[1] : f32 into vector<2xf32>
+  %3 = vector.broadcast %f7 : f32 to vector<2xf32>
+  %d = vector.insert %f8, %3[1] : f32 into vector<2xf32>
+
+  vector.print %a : vector<2xf32>
+  vector.print %b : vector<2xf32>
+  vector.print %c : vector<2xf32>
+  vector.print %d : vector<2xf32>
+  //
+  // test vectors:
+  //
+  // CHECK: ( 1, 2 )
+  // CHECK: ( 3, 4 )
+  // CHECK: ( 5, 6 )
+  // CHECK: ( 7, 8 )
+
+  // Construct test matrices.
+  %4 = vector.broadcast %f0 : f32 to vector<2x2xf32>
+  %5 = vector.insert %a, %4[0] : vector<2xf32> into vector<2x2xf32>
+  %A = vector.insert %b, %5[1] : vector<2xf32> into vector<2x2xf32>
+  %6 = vector.broadcast %f0 : f32 to vector<2x2xf32>
+  %7 = vector.insert %c, %6[0] : vector<2xf32> into vector<2x2xf32>
+  %B = vector.insert %d, %7[1] : vector<2xf32> into vector<2x2xf32>
+  %8 = vector.broadcast %f0 : f32 to vector<3x2xf32>
+  %9 = vector.insert %a, %8[0] : vector<2xf32> into vector<3x2xf32>
+  %10 = vector.insert %b, %9[1] : vector<2xf32> into vector<3x2xf32>
+  %C = vector.insert %c, %10[2] : vector<2xf32> into vector<3x2xf32>
+  %11 = vector.tuple %A, %B : vector<2x2xf32>, vector<2x2xf32>
+  %D = vector.insert_slices %11, [2, 2], [1, 1]
+    : tuple<vector<2x2xf32>, vector<2x2xf32>> into vector<2x4xf32>
+
+  vector.print %A : vector<2x2xf32>
+  vector.print %B : vector<2x2xf32>
+  vector.print %C : vector<3x2xf32>
+  vector.print %D : vector<2x4xf32>
+  //
+  // test matrices:
+  //
+  // CHECK: ( ( 1, 2 ), ( 3, 4 ) )
+  // CHECK: ( ( 5, 6 ), ( 7, 8 ) )
+  // CHECK: ( ( 1, 2 ), ( 3, 4 ), ( 5, 6 ) )
+  // CHECK: ( ( 1, 2, 5, 6 ), ( 3, 4, 7, 8 ) )
+
+  // Contraction: dot-product a x b
+  %dp1 = vector.contract #dotp_trait %a, %b, %f0
+    : vector<2xf32>, vector<2xf32> into f32
+  %dp2 = vector.contract #dotp_trait %a, %b, %f1
+    : vector<2xf32>, vector<2xf32> into f32
+
+  vector.print %dp1 : f32
+  vector.print %dp2 : f32
+  //
+  // dot products:
+  //
+  // CHECK: 11
+  // CHECK: 12
+
+  // Contraction: matrix-vector A x c
+  %mv1 = vector.contract #matvec_trait %A, %c, %z1
+    : vector<2x2xf32>, vector<2xf32> into vector<2xf32>
+  %mv2 = vector.contract #matvec_trait %A, %c, %a
+    : vector<2x2xf32>, vector<2xf32> into vector<2xf32>
+
+  vector.print %mv1 : vector<2xf32>
+  vector.print %mv2 : vector<2xf32>
+  //
+  // matrix x vector:
+  //
+  // CHECK: ( 17, 39 )
+  // CHECK: ( 18, 41 )
+
+  // Contraction: matrix-trans-vector A^T x c
+  %mv3 = vector.contract #mattransvec_trait %A, %c, %z1
+    : vector<2x2xf32>, vector<2xf32> into vector<2xf32>
+  %mv4 = vector.contract #mattransvec_trait %A, %c, %a
+    : vector<2x2xf32>, vector<2xf32> into vector<2xf32>
+
+  vector.print %mv3 : vector<2xf32>
+  vector.print %mv4 : vector<2xf32>
+  //
+  // matrix x vector:
+  //
+  // CHECK: ( 23, 34 )
+  // CHECK: ( 24, 36 )
+
+  // Contraction: matrix-matrix A x B
+  %mm1 = vector.contract #matmat_trait %A, %B, %z2
+    : vector<2x2xf32>, vector<2x2xf32> into vector<2x2xf32>
+  %mm2 = vector.contract #matmat_trait %A, %B, %A
+    : vector<2x2xf32>, vector<2x2xf32> into vector<2x2xf32>
+
+  vector.print %mm1 : vector<2x2xf32>
+  vector.print %mm2 : vector<2x2xf32>
+  //
+  // matrix x matrix:
+  //
+  // CHECK: ( ( 19, 22 ), ( 43, 50 ) )
+  // CHECK: ( ( 20, 24 ), ( 46, 54 ) )
+
+  // Contraction: matrix-matrix A x B where A, B, C have column-major layout.
+  // ( 1 * 5 + 3 * 6 = 23, 2 * 5 + 4 * 6 = 34)
+  // ( 1 * 7 + 3 * 8 = 31, 2 * 7 + 4 * 8 = 46)
+  // +
+  // ( ( 1, 2 ), ( 3, 4 ) )
+  %llvm_matrix_column_major_mm0 =
+    vector.contract #column_major_matmat_trait %A, %B, %z2
+      : vector<2x2xf32>, vector<2x2xf32> into vector<2x2xf32>
+  %llvm_matrix_column_major_mm1 =
+    vector.contract #column_major_matmat_trait %A, %B, %A
+      : vector<2x2xf32>, vector<2x2xf32> into vector<2x2xf32>
+
+  vector.print %llvm_matrix_column_major_mm0 : vector<2x2xf32>
+  vector.print %llvm_matrix_column_major_mm1 : vector<2x2xf32>
+  //
+  // matrix x matrix:
+  //
+  // CHECK: ( ( 23, 31 ), ( 34, 46 ) )
+  // CHECK: ( ( 24, 33 ), ( 37, 50 ) )
+
+  // Contraction: matrix-trans-matrix A^T x B
+  %mm3 = vector.contract #mattransmat_trait %A, %B, %z2
+    : vector<2x2xf32>, vector<2x2xf32> into vector<2x2xf32>
+  %mm4 = vector.contract #mattransmat_trait %A, %B, %A
+    : vector<2x2xf32>, vector<2x2xf32> into vector<2x2xf32>
+
+  vector.print %mm3 : vector<2x2xf32>
+  vector.print %mm4 : vector<2x2xf32>
+  //
+  // matrix x matrix:
+  //
+  // CHECK: ( ( 26, 30 ), ( 38, 44 ) )
+  // CHECK: ( ( 27, 32 ), ( 41, 48 ) )
+
+  // Contraction: matrix-matrix-trans A x B^T
+  %mm5 = vector.contract #matmattrans_trait %A, %B, %z2
+    : vector<2x2xf32>, vector<2x2xf32> into vector<2x2xf32>
+  %mm6 = vector.contract #matmattrans_trait %A, %B, %A
+    : vector<2x2xf32>, vector<2x2xf32> into vector<2x2xf32>
+
+  vector.print %mm5 : vector<2x2xf32>
+  vector.print %mm6 : vector<2x2xf32>
+  //
+  // matrix x matrix:
+  //
+  // CHECK: ( ( 17, 23 ), ( 39, 53 ) )
+  // CHECK: ( ( 18, 25 ), ( 42, 57 ) )
+
+  // Contraction: matrix-trans-matrix-trans A^T x B^T
+  %mm7 = vector.contract #mattransmattrans_trait %A, %B, %z2
+    : vector<2x2xf32>, vector<2x2xf32> into vector<2x2xf32>
+  %mm8 = vector.contract #mattransmattrans_trait %A, %B, %A
+    : vector<2x2xf32>, vector<2x2xf32> into vector<2x2xf32>
+
+  vector.print %mm7 : vector<2x2xf32>
+  vector.print %mm8 : vector<2x2xf32>
+  //
+  // matrix x matrix:
+  //
+  // CHECK: ( ( 23, 31 ), ( 34, 46 ) )
+  // CHECK: ( ( 24, 33 ), ( 37, 50 ) )
+
+  // Contraction: matrix-matrix-then-trans (A x B)^T
+  %mm9 = vector.contract #matmat_then_trans_trait %A, %B, %z2
+    : vector<2x2xf32>, vector<2x2xf32> into vector<2x2xf32>
+  %mm10 = vector.contract #matmat_then_trans_trait %A, %B, %A
+    : vector<2x2xf32>, vector<2x2xf32> into vector<2x2xf32>
+
+  vector.print %mm9 : vector<2x2xf32>
+  vector.print %mm10 : vector<2x2xf32>
+  //
+  // matrix x matrix:
+  //
+  // CHECK: ( ( 19, 43 ), ( 22, 50 ) )
+  // CHECK: ( ( 20, 45 ), ( 25, 54 ) )
+
+  // Contraction: matrix-matrix C x D
+  %mm11 = vector.contract #matmat_trait %C, %D, %z3
+    : vector<3x2xf32>, vector<2x4xf32> into vector<3x4xf32>
+  %mm12 = vector.contract #matmat_trait %C, %D, %mm11
+    : vector<3x2xf32>, vector<2x4xf32> into vector<3x4xf32>
+
+  vector.print %mm11 : vector<3x4xf32>
+  vector.print %mm12 : vector<3x4xf32>
+  // CHECK: ( ( 7, 10, 19, 22 ), ( 15, 22, 43, 50 ), ( 23, 34, 67, 78 ) )
+  // CHECK: ( ( 14, 20, 38, 44 ), ( 30, 44, 86, 100 ), ( 46, 68, 134, 156 ) )
+
+  // Contractions in 2D.
+  %c1 = vector.contract #contract2d_trait %A, %B, %f0
+    : vector<2x2xf32>, vector<2x2xf32> into f32
+  %c2 = vector.contract #contract2d_trait %A, %B, %f1
+    : vector<2x2xf32>, vector<2x2xf32> into f32
+  %c3 = vector.contract #contract2d_alt_trait %A, %B, %f0
+    : vector<2x2xf32>, vector<2x2xf32> into f32
+  %c4 = vector.contract #contract2d_alt_trait %A, %B, %f1
+    : vector<2x2xf32>, vector<2x2xf32> into f32
+  %c5 = vector.contract #contract2d_trans_trait %A, %B, %f0
+    : vector<2x2xf32>, vector<2x2xf32> into f32
+  %c6 = vector.contract #contract2d_trans_trait %A, %B, %f1
+    : vector<2x2xf32>, vector<2x2xf32> into f32
+  %c7 = vector.contract #contract2d_trans_alt_trait %A, %B, %f0
+    : vector<2x2xf32>, vector<2x2xf32> into f32
+  %c8 = vector.contract #contract2d_trans_alt_trait %A, %B, %f1
+    : vector<2x2xf32>, vector<2x2xf32> into f32
+
+  vector.print %c1 : f32
+  vector.print %c2 : f32
+  vector.print %c3 : f32
+  vector.print %c4 : f32
+  vector.print %c5 : f32
+  vector.print %c6 : f32
+  vector.print %c7 : f32
+  vector.print %c8 : f32
+  //
+  // 2D contractions:
+  //
+  // CHECK: 70
+  // CHECK: 71
+  // CHECK: 70
+  // CHECK: 71
+  // CHECK: 69
+  // CHECK: 70
+  // CHECK: 69
+  // CHECK: 70
+
+  return
+}

mlir/integration_test/Dialect/Vector/CPU/test-extract-slices.mlir

+// RUN: mlir-opt %s -convert-scf-to-std -convert-vector-to-llvm -convert-std-to-llvm | \
+// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
+// RUN:   -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+
+func @entry() {
+  %f0 = constant 0.0: f32
+  %f1 = constant 1.0: f32
+  %f2 = constant 2.0: f32
+  %f3 = constant 3.0: f32
+  %f4 = constant 4.0: f32
+  %f5 = constant 5.0: f32
+  %f6 = constant 6.0: f32
+  %f7 = constant 7.0: f32
+  %f8 = constant 8.0: f32
+  %f9 = constant 9.0: f32
+  %f10 = constant 10.0: f32
+  %f11 = constant 11.0: f32
+  %f12 = constant 12.0: f32
+  %f13 = constant 13.0: f32
+  %f14 = constant 14.0: f32
+  %f15 = constant 15.0: f32
+
+  %a0 = vector.broadcast %f0 : f32 to vector<4x4xf32>
+  %a1 = vector.insert %f0, %a0[0, 0] : f32 into vector<4x4xf32>
+  %a2 = vector.insert %f1, %a1[0, 1] : f32 into vector<4x4xf32>
+  %a3 = vector.insert %f2, %a2[0, 2] : f32 into vector<4x4xf32>
+  %a4 = vector.insert %f3, %a3[0, 3] : f32 into vector<4x4xf32>
+  %a5 = vector.insert %f4, %a4[1, 0] : f32 into vector<4x4xf32>
+  %a6 = vector.insert %f5, %a5[1, 1] : f32 into vector<4x4xf32>
+  %a7 = vector.insert %f6, %a6[1, 2] : f32 into vector<4x4xf32>
+  %a8 = vector.insert %f7, %a7[1, 3] : f32 into vector<4x4xf32>
+  %a9 = vector.insert %f8, %a8[2, 0] : f32 into vector<4x4xf32>
+  %a10 = vector.insert %f9, %a9[2, 1] : f32 into vector<4x4xf32>
+  %a11 = vector.insert %f10, %a10[2, 2] : f32 into vector<4x4xf32>
+  %a12 = vector.insert %f11, %a11[2, 3] : f32 into vector<4x4xf32>
+  %a13 = vector.insert %f12, %a12[3, 0] : f32 into vector<4x4xf32>
+  %a14 = vector.insert %f13, %a13[3, 1] : f32 into vector<4x4xf32>
+  %a15 = vector.insert %f14, %a14[3, 2] : f32 into vector<4x4xf32>
+  %a16 = vector.insert %f15, %a15[3, 3] : f32 into vector<4x4xf32>
+
+  vector.print %a16 : vector<4x4xf32>
+  //
+  // test matrix:
+  //
+  // CHECK: ( ( 0, 1, 2, 3 ), ( 4, 5, 6, 7 ), ( 8, 9, 10, 11 ), ( 12, 13, 14, 15 ) )
+
+  // Tile 4x4 with 3x3 as follows:
+  //
+  //   +--------+--+
+  //   +0   1  2| 3|
+  //   |4   5  6| 7|
+  //   |8   9 10|11|
+  //   +--------+--+
+  //   |12 13 14|15|
+  //   +--------+--+
+  //
+  %es = vector.extract_slices %a16, [3, 3], [1, 1] :
+     vector<4x4xf32> into tuple<vector<3x3xf32>, vector<3x1xf32>, vector<1x3xf32>, vector<1x1xf32>>
+
+  %0 = vector.tuple_get %es, 0 : tuple<vector<3x3xf32>, vector<3x1xf32>, vector<1x3xf32>, vector<1x1xf32>>
+  %1 = vector.tuple_get %es, 1 : tuple<vector<3x3xf32>, vector<3x1xf32>, vector<1x3xf32>, vector<1x1xf32>>
+  %2 = vector.tuple_get %es, 2 : tuple<vector<3x3xf32>, vector<3x1xf32>, vector<1x3xf32>, vector<1x1xf32>>
+  %3 = vector.tuple_get %es, 3 : tuple<vector<3x3xf32>, vector<3x1xf32>, vector<1x3xf32>, vector<1x1xf32>>
+
+  vector.print %0 : vector<3x3xf32>
+  vector.print %1 : vector<3x1xf32>
+  vector.print %2 : vector<1x3xf32>
+  vector.print %3 : vector<1x1xf32>
+  //
+  // extract slices:
+  //
+  // CHECK: ( ( 0, 1, 2 ), ( 4, 5, 6 ), ( 8, 9, 10 ) )
+  // CHECK: ( ( 3 ), ( 7 ), ( 11 ) )
+  // CHECK: ( ( 12, 13, 14 ) )
+  // CHECK: ( ( 15 ) )
+
+  return
+}

mlir/integration_test/Dialect/Vector/CPU/test-extract-strided-slice.mlir

+// RUN: mlir-opt %s -convert-scf-to-std -convert-vector-to-llvm -convert-std-to-llvm | \
+// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
+// RUN:   -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+
+func @entry() {
+  %f0 = constant 0.0: f32
+  %f1 = constant 1.0: f32
+  %f2 = constant 2.0: f32
+  %f3 = constant 3.0: f32
+  %f4 = constant 4.0: f32
+  %v1 = vector.broadcast %f1 : f32 to vector<8xf32>
+  %v2 = vector.broadcast %f2 : f32 to vector<8xf32>
+  %v3 = vector.broadcast %f3 : f32 to vector<8xf32>
+  %v4 = vector.broadcast %f4 : f32 to vector<8xf32>
+
+  %a0 = vector.broadcast %f0 : f32 to vector<4x4x8xf32>
+  %a1 = vector.insert %v1, %a0[1, 1] : vector<8xf32> into vector<4x4x8xf32>
+  %a2 = vector.insert %v2, %a1[1, 2] : vector<8xf32> into vector<4x4x8xf32>
+  %a3 = vector.insert %v3, %a2[2, 1] : vector<8xf32> into vector<4x4x8xf32>
+  %a4 = vector.insert %v4, %a3[2, 2] : vector<8xf32> into vector<4x4x8xf32>
+
+  %ss = vector.extract_strided_slice %a4 {offsets = [1, 1], sizes = [2, 2], strides = [1, 1]} : vector<4x4x8xf32> to vector<2x2x8xf32>
+
+  vector.print %ss : vector<2x2x8xf32>
+  //
+  // extract strided slice:
+  //
+  // CHECK: ( ( ( 1, 1, 1, 1, 1, 1, 1, 1 ), ( 2, 2, 2, 2, 2, 2, 2, 2 ) ), ( ( 3, 3, 3, 3, 3, 3, 3, 3 ), ( 4, 4, 4, 4, 4, 4, 4, 4 ) ) )
+
+  return
+}

mlir/integration_test/Dialect/Vector/CPU/test-flat-transpose-col.mlir

+// RUN: mlir-opt %s -convert-scf-to-std -convert-vector-to-llvm -convert-std-to-llvm | \
+// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
+// RUN:   -lower-matrix-intrinsics -matrix-allow-contract -matrix-default-layout=column-major \
+// RUN:   -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+
+func @entry() {
+  %f0 = constant 0.0: f64
+  %f1 = constant 1.0: f64
+  %f2 = constant 2.0: f64
+  %f3 = constant 3.0: f64
+  %f4 = constant 4.0: f64
+  %f5 = constant 5.0: f64
+  %f6 = constant 6.0: f64
+  %f7 = constant 7.0: f64
+
+  // Construct test vectors.
+  %0 = vector.broadcast %f0 : f64 to vector<4xf64>
+  %1 = vector.insert %f1, %0[1] : f64 into vector<4xf64>
+  %2 = vector.insert %f2, %1[2] : f64 into vector<4xf64>
+  %a = vector.insert %f3, %2[3] : f64 into vector<4xf64>
+  %3 = vector.broadcast %f4 : f64 to vector<4xf64>
+  %4 = vector.insert %f5, %3[1] : f64 into vector<4xf64>
+  %5 = vector.insert %f6, %4[2] : f64 into vector<4xf64>
+  %b = vector.insert %f7, %5[3] : f64 into vector<4xf64>
+  %6 = vector.broadcast %f0 : f64 to vector<6xf64>
+  %7 = vector.insert %f1, %6[1] : f64 into vector<6xf64>
+  %8 = vector.insert %f2, %7[2] : f64 into vector<6xf64>
+  %9 = vector.insert %f3, %8[3] : f64 into vector<6xf64>
+  %10 = vector.insert %f4, %9[4] : f64 into vector<6xf64>
+  %c = vector.insert %f5, %10[5] : f64 into vector<6xf64>
+
+  vector.print %a : vector<4xf64>
+  vector.print %b : vector<4xf64>
+  vector.print %c : vector<6xf64>
+  //
+  // Test vectors:
+  //
+  // CHECK: ( 0, 1, 2, 3 )
+  // CHECK: ( 4, 5, 6, 7 )
+  // CHECK: ( 0, 1, 2, 3, 4, 5 )
+
+  // Performs matrix transpositions interpreting the vectors as
+  // flattened column-major 2-D matrices.
+  //
+  // ( 0, 2 )       ( 0, 1 )   | /|
+  // ( 1, 3 )    -> ( 2, 3 )   |/ | column-major!
+  //
+  // ( 4, 6 )       ( 4, 5 )
+  // ( 5, 7 )    -> ( 6, 7 )
+  //
+  // ( 0, 2, 4 )    ( 0, 1 )
+  // ( 1, 3, 5 ) -> ( 2, 3 )
+  //                ( 4, 5 )
+  //
+  // ( 0, 3 )        ( 0, 1, 2 )
+  // ( 1, 4 )    ->  ( 3, 4, 5 )
+  // ( 2, 5 )
+  //
+  %d = vector.flat_transpose %a { rows = 2: i32, columns = 2: i32 } : vector<4xf64> -> vector<4xf64>
+  %e = vector.flat_transpose %b { rows = 2: i32, columns = 2: i32 } : vector<4xf64> -> vector<4xf64>
+  %f = vector.flat_transpose %c { rows = 2: i32, columns = 3: i32 } : vector<6xf64> -> vector<6xf64>
+  %g = vector.flat_transpose %c { rows = 3: i32, columns = 2: i32 } : vector<6xf64> -> vector<6xf64>
+
+  vector.print %d : vector<4xf64>
+  vector.print %e : vector<4xf64>
+  vector.print %f : vector<6xf64>
+  vector.print %g : vector<6xf64>
+  //
+  // Transposed results:
+  //
+  // CHECK: ( 0, 2, 1, 3 )
+  // CHECK: ( 4, 6, 5, 7 )
+  // CHECK: ( 0, 2, 4, 1, 3, 5 )
+  // CHECK: ( 0, 3, 1, 4, 2, 5 )
+
+  return
+}

mlir/integration_test/Dialect/Vector/CPU/test-flat-transpose-row.mlir

+// RUN: mlir-opt %s -convert-scf-to-std -convert-vector-to-llvm -convert-std-to-llvm | \
+// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
+// RUN:   -lower-matrix-intrinsics -matrix-allow-contract -matrix-default-layout=row-major \
+// RUN:   -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+
+func @entry() {
+  %f0 = constant 0.0: f64
+  %f1 = constant 1.0: f64
+  %f2 = constant 2.0: f64
+  %f3 = constant 3.0: f64
+  %f4 = constant 4.0: f64
+  %f5 = constant 5.0: f64
+  %f6 = constant 6.0: f64
+  %f7 = constant 7.0: f64
+
+  // Construct test vectors.
+  %0 = vector.broadcast %f0 : f64 to vector<4xf64>
+  %1 = vector.insert %f1, %0[1] : f64 into vector<4xf64>
+  %2 = vector.insert %f2, %1[2] : f64 into vector<4xf64>
+  %a = vector.insert %f3, %2[3] : f64 into vector<4xf64>
+  %3 = vector.broadcast %f4 : f64 to vector<4xf64>
+  %4 = vector.insert %f5, %3[1] : f64 into vector<4xf64>
+  %5 = vector.insert %f6, %4[2] : f64 into vector<4xf64>
+  %b = vector.insert %f7, %5[3] : f64 into vector<4xf64>
+  %6 = vector.broadcast %f0 : f64 to vector<6xf64>
+  %7 = vector.insert %f1, %6[1] : f64 into vector<6xf64>
+  %8 = vector.insert %f2, %7[2] : f64 into vector<6xf64>
+  %9 = vector.insert %f3, %8[3] : f64 into vector<6xf64>
+  %10 = vector.insert %f4, %9[4] : f64 into vector<6xf64>
+  %c = vector.insert %f5, %10[5] : f64 into vector<6xf64>
+
+  vector.print %a : vector<4xf64>
+  vector.print %b : vector<4xf64>
+  vector.print %c : vector<6xf64>
+  //
+  // Test vectors:
+  //
+  // CHECK: ( 0, 1, 2, 3 )
+  // CHECK: ( 4, 5, 6, 7 )
+  // CHECK: ( 0, 1, 2, 3, 4, 5 )
+
+  // Performs matrix transpositions interpreting the vectors as
+  // flattened row-major 2-D matrices.
+  //
+  // ( 0, 1 )       ( 0, 2 )
+  // ( 2, 3 )    -> ( 1, 3 )
+  //
+  // ( 4, 5 )       ( 4, 6 )
+  // ( 6, 7 )    -> ( 5, 7 )
+  //
+  // ( 0, 1, 2 )    ( 0, 3 )
+  // ( 3, 4, 5 ) -> ( 1, 4 )
+  //                ( 2, 5 )
+  //
+  // ( 0, 1 )        ( 0, 2, 4 )
+  // ( 2, 3 )    ->  ( 1, 3, 5 )
+  // ( 4, 5 )
+  //
+  %d = vector.flat_transpose %a { rows = 2: i32, columns = 2: i32 } : vector<4xf64> -> vector<4xf64>
+  %e = vector.flat_transpose %b { rows = 2: i32, columns = 2: i32 } : vector<4xf64> -> vector<4xf64>
+  %f = vector.flat_transpose %c { rows = 2: i32, columns = 3: i32 } : vector<6xf64> -> vector<6xf64>
+  %g = vector.flat_transpose %c { rows = 3: i32, columns = 2: i32 } : vector<6xf64> -> vector<6xf64>
+
+  vector.print %d : vector<4xf64>
+  vector.print %e : vector<4xf64>
+  vector.print %f : vector<6xf64>
+  vector.print %g : vector<6xf64>
+  //
+  // Transposed results:
+  //
+  // CHECK: ( 0, 2, 1, 3 )
+  // CHECK: ( 4, 6, 5, 7 )
+  // CHECK: ( 0, 3, 1, 4, 2, 5 )
+  // CHECK: ( 0, 2, 4, 1, 3, 5 )
+
+  return
+}

mlir/integration_test/Dialect/Vector/CPU/test-fma.mlir

+// RUN: mlir-opt %s -convert-scf-to-std -convert-vector-to-llvm -convert-std-to-llvm | \
+// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
+// RUN:   -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+
+func @entry() {
+  %f1 = constant 1.0: f32
+  %f3 = constant 3.0: f32
+  %f7 = constant 7.0: f32
+  %v1 = vector.broadcast %f1 : f32 to vector<8xf32>
+  %v3 = vector.broadcast %f3 : f32 to vector<8xf32>
+  %v7 = vector.broadcast %f7 : f32 to vector<8xf32>
+  vector.print %v1 : vector<8xf32>
+  vector.print %v3 : vector<8xf32>
+  vector.print %v7 : vector<8xf32>
+  //
+  // test vectors:
+  //
+  // CHECK: ( 1, 1, 1, 1, 1, 1, 1, 1 )
+  // CHECK: ( 3, 3, 3, 3, 3, 3, 3, 3 )
+  // CHECK: ( 7, 7, 7, 7, 7, 7, 7, 7 )
+
+  %v = vector.fma %v3, %v7, %v1: vector<8xf32>
+  vector.print %v : vector<8xf32>
+  // CHECK: ( 22, 22, 22, 22, 22, 22, 22, 22 )
+
+  return
+}

mlir/integration_test/Dialect/Vector/CPU/test-insert-slices.mlir

+// RUN: mlir-opt %s -convert-scf-to-std -convert-vector-to-llvm -convert-std-to-llvm | \
+// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
+// RUN:   -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+
+func @entry() {
+  %f0 = constant 0.0: f32
+  %f1 = constant 1.0: f32
+  %f2 = constant 2.0: f32
+  %f3 = constant 3.0: f32
+  %f4 = constant 4.0: f32
+  %f5 = constant 5.0: f32
+  %f6 = constant 6.0: f32
+  %f7 = constant 7.0: f32
+  %f8 = constant 8.0: f32
+  %f9 = constant 9.0: f32
+  %f10 = constant 10.0: f32
+  %f11 = constant 11.0: f32
+  %f12 = constant 12.0: f32
+  %f13 = constant 13.0: f32
+  %f14 = constant 14.0: f32
+  %f15 = constant 15.0: f32
+
+  %a0 = vector.broadcast %f0 : f32 to vector<3x3xf32>
+  %a1 = vector.insert %f0, %a0[0, 0] : f32 into vector<3x3xf32>
+  %a2 = vector.insert %f1, %a1[0, 1] : f32 into vector<3x3xf32>
+  %a3 = vector.insert %f2, %a2[0, 2] : f32 into vector<3x3xf32>
+  %a4 = vector.insert %f4, %a3[1, 0] : f32 into vector<3x3xf32>
+  %a5 = vector.insert %f5, %a4[1, 1] : f32 into vector<3x3xf32>
+  %a6 = vector.insert %f6, %a5[1, 2] : f32 into vector<3x3xf32>
+  %a7 = vector.insert %f8, %a6[2, 0] : f32 into vector<3x3xf32>
+  %a8 = vector.insert %f9, %a7[2, 1] : f32 into vector<3x3xf32>
+  %a9 = vector.insert %f10, %a8[2, 2] : f32 into vector<3x3xf32>
+
+  %b0 = vector.broadcast %f0 : f32 to vector<3x1xf32>
+  %b1 = vector.insert %f3, %b0[0, 0] : f32 into vector<3x1xf32>
+  %b2 = vector.insert %f7, %b1[1, 0] : f32 into vector<3x1xf32>
+  %b3 = vector.insert %f11, %b2[2, 0] : f32 into vector<3x1xf32>
+
+  %c0 = vector.broadcast %f0 : f32 to vector<1x3xf32>
+  %c1 = vector.insert %f12, %c0[0, 0] : f32 into vector<1x3xf32>
+  %c2 = vector.insert %f13, %c1[0, 1] : f32 into vector<1x3xf32>
+  %c3 = vector.insert %f14, %c2[0, 2] : f32 into vector<1x3xf32>
+
+  %d0 = vector.broadcast %f0 : f32 to vector<1x1xf32>
+  %d1 = vector.insert %f15, %d0[0, 0] : f32 into vector<1x1xf32>
+
+  vector.print %a9 : vector<3x3xf32>
+  vector.print %b3 : vector<3x1xf32>
+  vector.print %c3 : vector<1x3xf32>
+  vector.print %d1 : vector<1x1xf32>
+  //
+  // input slices:
+  //
+  // CHECK: ( ( 0, 1, 2 ), ( 4, 5, 6 ), ( 8, 9, 10 ) )
+  // CHECK: ( ( 3 ), ( 7 ), ( 11 ) )
+  // CHECK: ( ( 12, 13, 14 ) )
+  // CHECK: ( ( 15 ) )
+
+  %vt = vector.tuple %a9, %b3, %c3, %d1 :
+     vector<3x3xf32>, vector<3x1xf32>, vector<1x3xf32>, vector<1x1xf32>
+  %is = vector.insert_slices %vt, [3, 3], [1, 1] :
+     tuple<vector<3x3xf32>, vector<3x1xf32>, vector<1x3xf32>, vector<1x1xf32>> into vector<4x4xf32>
+
+  vector.print %is : vector<4x4xf32>
+  //
+  // insert slices:
+  //
+  // CHECK: ( ( 0, 1, 2, 3 ), ( 4, 5, 6, 7 ), ( 8, 9, 10, 11 ), ( 12, 13, 14, 15 ) )
+
+  return
+}

mlir/integration_test/Dialect/Vector/CPU/test-insert-strided-slice.mlir

+// RUN: mlir-opt %s -convert-scf-to-std -convert-vector-to-llvm -convert-std-to-llvm | \
+// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
+// RUN:   -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+
+func @entry() {
+  %f1 = constant 1.0: f32
+  %f2 = constant 2.0: f32
+  %f3 = constant 3.0: f32
+  %f4 = constant 4.0: f32
+  %v1 = vector.broadcast %f1 : f32 to vector<4xf32>
+  %v2 = vector.broadcast %f2 : f32 to vector<3xf32>
+  %v3 = vector.broadcast %f3 : f32 to vector<4x4xf32>
+  %v4 = vector.broadcast %f4 : f32 to vector<1xf32>
+
+  %s1 = vector.insert_strided_slice %v1, %v3 {offsets = [2, 0], strides = [1]} : vector<4xf32> into vector<4x4xf32>
+  %s2 = vector.insert_strided_slice %v2, %s1 {offsets = [1, 1], strides = [1]} : vector<3xf32> into vector<4x4xf32>
+  %s3 = vector.insert_strided_slice %v2, %s2 {offsets = [0, 0], strides = [1]} : vector<3xf32> into vector<4x4xf32>
+  %s4 = vector.insert_strided_slice %v4, %s3 {offsets = [3, 3], strides = [1]} : vector<1xf32> into vector<4x4xf32>
+
+  vector.print %v3 : vector<4x4xf32>
+  vector.print %s1 : vector<4x4xf32>
+  vector.print %s2 : vector<4x4xf32>
+  vector.print %s3 : vector<4x4xf32>
+  vector.print %s4 : vector<4x4xf32>
+  //
+  // insert strided slice:
+  //
+  // CHECK: ( ( 3, 3, 3, 3 ), ( 3, 3, 3, 3 ), ( 3, 3, 3, 3 ), ( 3, 3, 3, 3 ) )
+  // CHECK: ( ( 3, 3, 3, 3 ), ( 3, 3, 3, 3 ), ( 1, 1, 1, 1 ), ( 3, 3, 3, 3 ) )
+  // CHECK: ( ( 3, 3, 3, 3 ), ( 3, 2, 2, 2 ), ( 1, 1, 1, 1 ), ( 3, 3, 3, 3 ) )
+  // CHECK: ( ( 2, 2, 2, 3 ), ( 3, 2, 2, 2 ), ( 1, 1, 1, 1 ), ( 3, 3, 3, 3 ) )
+  // CHECK: ( ( 2, 2, 2, 3 ), ( 3, 2, 2, 2 ), ( 1, 1, 1, 1 ), ( 3, 3, 3, 4 ) )
+
+  return
+}

mlir/integration_test/Dialect/Vector/CPU/test-matrix-multiply-col.mlir

+// RUN: mlir-opt %s -convert-scf-to-std -convert-vector-to-llvm -convert-std-to-llvm | \
+// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
+// RUN:   -lower-matrix-intrinsics -matrix-allow-contract -matrix-default-layout=column-major \
+// RUN:   -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+
+func @entry() {
+  %f0 = constant 0.0: f64
+  %f1 = constant 1.0: f64
+  %f2 = constant 2.0: f64
+  %f3 = constant 3.0: f64
+  %f4 = constant 4.0: f64
+  %f5 = constant 5.0: f64
+  %f6 = constant 6.0: f64
+  %f7 = constant 7.0: f64
+
+  // Construct test vectors.
+  %0 = vector.broadcast %f0 : f64 to vector<4xf64>
+  %1 = vector.insert %f1, %0[1] : f64 into vector<4xf64>
+  %2 = vector.insert %f2, %1[2] : f64 into vector<4xf64>
+  %a = vector.insert %f3, %2[3] : f64 into vector<4xf64>
+  %3 = vector.broadcast %f4 : f64 to vector<4xf64>
+  %4 = vector.insert %f5, %3[1] : f64 into vector<4xf64>
+  %5 = vector.insert %f6, %4[2] : f64 into vector<4xf64>
+  %b = vector.insert %f7, %5[3] : f64 into vector<4xf64>
+
+  vector.print %a : vector<4xf64>
+  vector.print %b : vector<4xf64>
+  //
+  // test vectors:
+  //
+  // CHECK: ( 0, 1, 2, 3 )
+  // CHECK: ( 4, 5, 6, 7 )
+
+  // Performs matrix x matrix, interpreting the vectors as
+  // flattened column-major 2-D matrices.
+  //
+  // ( 0, 2 )     (4, 6)     ( 10, 14 )      | /|
+  //           x          =                  |/ | column-major!
+  // ( 1, 3 )     (5, 7)     ( 19, 27 )
+  //
+  %c = vector.matrix_multiply %a, %b
+      { lhs_rows = 2: i32, lhs_columns = 2: i32 , rhs_columns = 2: i32 }
+      : (vector<4xf64>, vector<4xf64>) -> vector<4xf64>
+
+  vector.print %c : vector<4xf64>
+  //
+  // matrix x matrix:
+  //
+  // CHECK: ( 10, 19, 14, 27 )
+
+  return
+}

mlir/integration_test/Dialect/Vector/CPU/test-matrix-multiply-row.mlir

+// RUN: mlir-opt %s -convert-scf-to-std -convert-vector-to-llvm -convert-std-to-llvm | \
+// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
+// RUN:   -lower-matrix-intrinsics -matrix-allow-contract -matrix-default-layout=row-major \
+// RUN:   -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+
+func @entry() {
+  %f0 = constant 0.0: f64
+  %f1 = constant 1.0: f64
+  %f2 = constant 2.0: f64
+  %f3 = constant 3.0: f64
+  %f4 = constant 4.0: f64
+  %f5 = constant 5.0: f64
+  %f6 = constant 6.0: f64
+  %f7 = constant 7.0: f64
+
+  // Construct test vectors.
+  %0 = vector.broadcast %f0 : f64 to vector<4xf64>
+  %1 = vector.insert %f1, %0[1] : f64 into vector<4xf64>
+  %2 = vector.insert %f2, %1[2] : f64 into vector<4xf64>
+  %a = vector.insert %f3, %2[3] : f64 into vector<4xf64>
+  %3 = vector.broadcast %f4 : f64 to vector<4xf64>
+  %4 = vector.insert %f5, %3[1] : f64 into vector<4xf64>
+  %5 = vector.insert %f6, %4[2] : f64 into vector<4xf64>
+  %b = vector.insert %f7, %5[3] : f64 into vector<4xf64>
+
+  vector.print %a : vector<4xf64>
+  vector.print %b : vector<4xf64>
+  //
+  // test vectors:
+  //
+  // CHECK: ( 0, 1, 2, 3 )
+  // CHECK: ( 4, 5, 6, 7 )
+
+  // Performs matrix x matrix, interpreting the vectors as
+  // flattened row-major 2-D matrices.
+  //
+  // ( 0, 1 )     (4, 5)     (  6,  7 )
+  //           x          =
+  // ( 2, 3 )     (6, 7)     ( 26, 31 )
+  //
+  %c = vector.matrix_multiply %a, %b
+      { lhs_rows = 2: i32, lhs_columns = 2: i32 , rhs_columns = 2: i32 }
+      : (vector<4xf64>, vector<4xf64>) -> vector<4xf64>
+
+  vector.print %c : vector<4xf64>
+  //
+  // matrix x matrix:
+  //
+  // CHECK: ( 6, 7, 26, 31 )
+
+  return
+}

mlir/integration_test/Dialect/Vector/CPU/test-outerproduct.mlir

+// RUN: mlir-opt %s -convert-scf-to-std -convert-vector-to-llvm -convert-std-to-llvm | \
+// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
+// RUN:   -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+
+!vector_type_A = type vector<8xf32>
+!vector_type_B = type vector<8xf32>
+!vector_type_C = type vector<8x8xf32>
+
+!vector_type_X = type vector<2xf32>
+!vector_type_Y = type vector<3xf32>
+!vector_type_Z = type vector<2x3xf32>
+
+func @vector_outerproduct_splat_8x8(%fa: f32, %fb: f32, %fc: f32) -> !vector_type_C {
+  %a = splat %fa: !vector_type_A
+  %b = splat %fb: !vector_type_B
+  %c = splat %fc: !vector_type_C
+  %d = vector.outerproduct %a, %b, %c : !vector_type_A, !vector_type_B
+  return %d: !vector_type_C
+}
+
+func @vector_outerproduct_vec_2x3(%x : !vector_type_X,
+                                  %y : !vector_type_Y) -> !vector_type_Z {
+  %o = vector.outerproduct %x, %y : !vector_type_X, !vector_type_Y
+  return %o: !vector_type_Z
+}
+
+func @vector_outerproduct_vec_2x3_acc(%x : !vector_type_X,
+                                      %y : !vector_type_Y,
+                                      %z : !vector_type_Z) -> !vector_type_Z {
+  %o = vector.outerproduct %x, %y, %z : !vector_type_X, !vector_type_Y
+  return %o: !vector_type_Z
+}
+
+func @entry() {
+  %f1 = constant 1.0: f32
+  %f2 = constant 2.0: f32
+  %f3 = constant 3.0: f32
+  %f4 = constant 4.0: f32
+  %f5 = constant 5.0: f32
+  %f10 = constant 10.0: f32
+
+  // Simple case, splat scalars into vectors, then take outer product.
+  %v = call @vector_outerproduct_splat_8x8(%f1, %f2, %f10)
+      : (f32, f32, f32) -> (!vector_type_C)
+  vector.print %v : !vector_type_C
+  //
+  // outer product 8x8:
+  //
+  // CHECK-COUNT-8: ( 12, 12, 12, 12, 12, 12, 12, 12 )
+
+  // Direct outerproduct on vectors with different size.
+  %0 = vector.broadcast %f1 : f32 to !vector_type_X
+  %x = vector.insert %f2, %0[1] : f32 into !vector_type_X
+  %1 = vector.broadcast %f3 : f32 to !vector_type_Y
+  %2 = vector.insert %f4, %1[1] : f32 into !vector_type_Y
+  %y = vector.insert %f5, %2[2] : f32 into !vector_type_Y
+
+  %p = call @vector_outerproduct_vec_2x3(%x, %y)
+      : (!vector_type_X, !vector_type_Y) -> (!vector_type_Z)
+  vector.print %p : !vector_type_Z
+  //
+  // outer product 2x3:
+  //
+  // CHECK: ( ( 3, 4, 5 ), ( 6, 8, 10 ) )
+
+  %q = call @vector_outerproduct_vec_2x3_acc(%x, %y, %p)
+      : (!vector_type_X, !vector_type_Y, !vector_type_Z) -> (!vector_type_Z)
+  vector.print %q : !vector_type_Z
+  //
+  // outer product 2x3:
+  //
+  // CHECK: ( ( 6, 8, 10 ), ( 12, 16, 20 ) )
+
+  return
+}

mlir/integration_test/Dialect/Vector/CPU/test-reductions-f32.mlir

+// RUN: mlir-opt %s -convert-scf-to-std -convert-vector-to-llvm -convert-std-to-llvm | \
+// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
+// RUN:   -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+
+func @entry() {
+  // Construct test vector.
+  %f1 = constant 1.5: f32
+  %f2 = constant 2.0: f32
+  %f3 = constant 3.0: f32
+  %f4 = constant 4.0: f32
+  %f5 = constant 5.0: f32
+  %f6 = constant -1.0: f32
+  %f7 = constant -2.0: f32
+  %f8 = constant -4.0: f32
+  %f9 = constant -0.25: f32
+  %f10 = constant -16.0: f32
+  %v0 = vector.broadcast %f1 : f32 to vector<10xf32>
+  %v1 = vector.insert %f2, %v0[1] : f32 into vector<10xf32>
+  %v2 = vector.insert %f3, %v1[2] : f32 into vector<10xf32>
+  %v3 = vector.insert %f4, %v2[3] : f32 into vector<10xf32>
+  %v4 = vector.insert %f5, %v3[4] : f32 into vector<10xf32>
+  %v5 = vector.insert %f6, %v4[5] : f32 into vector<10xf32>
+  %v6 = vector.insert %f7, %v5[6] : f32 into vector<10xf32>
+  %v7 = vector.insert %f8, %v6[7] : f32 into vector<10xf32>
+  %v8 = vector.insert %f9, %v7[8] : f32 into vector<10xf32>
+  %v9 = vector.insert %f10, %v8[9] : f32 into vector<10xf32>
+  vector.print %v9 : vector<10xf32>
+  //
+  // test vector:
+  //
+  // CHECK: ( 1.5, 2, 3, 4, 5, -1, -2, -4, -0.25, -16 )
+
+  // Various vector reductions. Not full functional unit tests, but
+  // a simple integration test to see if the code runs end-to-end.
+  %0 = vector.reduction "add", %v9 : vector<10xf32> into f32
+  vector.print %0 : f32
+  // CHECK: -7.75
+  %1 = vector.reduction "mul", %v9 : vector<10xf32> into f32
+  vector.print %1 : f32
+  // CHECK: -5760
+  %2 = vector.reduction "min", %v9 : vector<10xf32> into f32
+  vector.print %2 : f32
+  // CHECK: -16
+  %3 = vector.reduction "max", %v9 : vector<10xf32> into f32
+  vector.print %3 : f32
+  // CHECK: 5
+
+  return
+}

mlir/integration_test/Dialect/Vector/CPU/test-reductions-f64.mlir

+// RUN: mlir-opt %s -convert-scf-to-std -convert-vector-to-llvm -convert-std-to-llvm | \
+// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
+// RUN:   -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+
+func @entry() {
+  // Construct test vector.
+  %f1 = constant 1.5: f64
+  %f2 = constant 2.0: f64
+  %f3 = constant 3.0: f64
+  %f4 = constant 4.0: f64
+  %f5 = constant 5.0: f64
+  %f6 = constant -1.0: f64
+  %f7 = constant -2.0: f64
+  %f8 = constant -4.0: f64
+  %f9 = constant -0.25: f64
+  %f10 = constant -16.0: f64
+  %v0 = vector.broadcast %f1 : f64 to vector<10xf64>
+  %v1 = vector.insert %f2, %v0[1] : f64 into vector<10xf64>
+  %v2 = vector.insert %f3, %v1[2] : f64 into vector<10xf64>
+  %v3 = vector.insert %f4, %v2[3] : f64 into vector<10xf64>
+  %v4 = vector.insert %f5, %v3[4] : f64 into vector<10xf64>
+  %v5 = vector.insert %f6, %v4[5] : f64 into vector<10xf64>
+  %v6 = vector.insert %f7, %v5[6] : f64 into vector<10xf64>
+  %v7 = vector.insert %f8, %v6[7] : f64 into vector<10xf64>
+  %v8 = vector.insert %f9, %v7[8] : f64 into vector<10xf64>
+  %v9 = vector.insert %f10, %v8[9] : f64 into vector<10xf64>
+  vector.print %v9 : vector<10xf64>
+  //
+  // test vector:
+  //
+  // CHECK: ( 1.5, 2, 3, 4, 5, -1, -2, -4, -0.25, -16 )
+
+  // Various vector reductions. Not full functional unit tests, but
+  // a simple integration test to see if the code runs end-to-end.
+  %0 = vector.reduction "add", %v9 : vector<10xf64> into f64
+  vector.print %0 : f64
+  // CHECK: -7.75
+  %1 = vector.reduction "mul", %v9 : vector<10xf64> into f64
+  vector.print %1 : f64
+  // CHECK: -5760
+  %2 = vector.reduction "min", %v9 : vector<10xf64> into f64
+  vector.print %2 : f64
+  // CHECK: -16
+  %3 = vector.reduction "max", %v9 : vector<10xf64> into f64
+  vector.print %3 : f64
+  // CHECK: 5
+
+  return
+}

mlir/integration_test/Dialect/Vector/CPU/test-reductions-i32.mlir

+// RUN: mlir-opt %s -convert-scf-to-std -convert-vector-to-llvm -convert-std-to-llvm | \
+// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
+// RUN:   -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+
+func @entry() {
+  // Construct test vector.
+  %i1 = constant 1: i32
+  %i2 = constant 2: i32
+  %i3 = constant 3: i32
+  %i4 = constant 4: i32
+  %i5 = constant 5: i32
+  %i6 = constant -1: i32
+  %i7 = constant -2: i32
+  %i8 = constant -4: i32
+  %i9 = constant -80: i32
+  %i10 = constant -16: i32
+  %v0 = vector.broadcast %i1 : i32 to vector<10xi32>
+  %v1 = vector.insert %i2, %v0[1] : i32 into vector<10xi32>
+  %v2 = vector.insert %i3, %v1[2] : i32 into vector<10xi32>
+  %v3 = vector.insert %i4, %v2[3] : i32 into vector<10xi32>
+  %v4 = vector.insert %i5, %v3[4] : i32 into vector<10xi32>
+  %v5 = vector.insert %i6, %v4[5] : i32 into vector<10xi32>
+  %v6 = vector.insert %i7, %v5[6] : i32 into vector<10xi32>
+  %v7 = vector.insert %i8, %v6[7] : i32 into vector<10xi32>
+  %v8 = vector.insert %i9, %v7[8] : i32 into vector<10xi32>
+  %v9 = vector.insert %i10, %v8[9] : i32 into vector<10xi32>
+  vector.print %v9 : vector<10xi32>
+  //
+  // test vector:
+  //
+  // CHECK: ( 1, 2, 3, 4, 5, -1, -2, -4, -80, -16 )
+
+  // Various vector reductions. Not full functional unit tests, but
+  // a simple integration test to see if the code runs end-to-end.
+  %0 = vector.reduction "add", %v9 : vector<10xi32> into i32
+  vector.print %0 : i32
+  // CHECK: -88
+  %1 = vector.reduction "mul", %v9 : vector<10xi32> into i32
+  vector.print %1 : i32
+  // CHECK: -1228800
+  %2 = vector.reduction "min", %v9 : vector<10xi32> into i32
+  vector.print %2 : i32
+  // CHECK: -80
+  %3 = vector.reduction "max", %v9 : vector<10xi32> into i32
+  vector.print %3 : i32
+  // CHECK: 5
+  %4 = vector.reduction "and", %v9 : vector<10xi32> into i32
+  vector.print %4 : i32
+  // CHECK: 0
+  %5 = vector.reduction "or", %v9 : vector<10xi32> into i32
+  vector.print %5 : i32
+  // CHECK: -1
+  %6 = vector.reduction "xor", %v9 : vector<10xi32> into i32
+  vector.print %6 : i32
+  // CHECK: -68
+
+  return
+}

mlir/integration_test/Dialect/Vector/CPU/test-reductions-i64.mlir

+// RUN: mlir-opt %s -convert-scf-to-std -convert-vector-to-llvm -convert-std-to-llvm | \
+// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
+// RUN:   -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+
+func @entry() {
+  // Construct test vector.
+  %i1 = constant 1: i64
+  %i2 = constant 2: i64
+  %i3 = constant 3: i64
+  %i4 = constant 4: i64
+  %i5 = constant 5: i64
+  %i6 = constant -1: i64
+  %i7 = constant -2: i64
+  %i8 = constant -4: i64
+  %i9 = constant -80: i64
+  %i10 = constant -16: i64
+  %v0 = vector.broadcast %i1 : i64 to vector<10xi64>
+  %v1 = vector.insert %i2, %v0[1] : i64 into vector<10xi64>
+  %v2 = vector.insert %i3, %v1[2] : i64 into vector<10xi64>
+  %v3 = vector.insert %i4, %v2[3] : i64 into vector<10xi64>
+  %v4 = vector.insert %i5, %v3[4] : i64 into vector<10xi64>
+  %v5 = vector.insert %i6, %v4[5] : i64 into vector<10xi64>
+  %v6 = vector.insert %i7, %v5[6] : i64 into vector<10xi64>
+  %v7 = vector.insert %i8, %v6[7] : i64 into vector<10xi64>
+  %v8 = vector.insert %i9, %v7[8] : i64 into vector<10xi64>
+  %v9 = vector.insert %i10, %v8[9] : i64 into vector<10xi64>
+  vector.print %v9 : vector<10xi64>
+  //
+  // test vector:
+  //
+  // CHECK: ( 1, 2, 3, 4, 5, -1, -2, -4, -80, -16 )
+
+  // Various vector reductions. Not full functional unit tests, but
+  // a simple integration test to see if the code runs end-to-end.
+  %0 = vector.reduction "add", %v9 : vector<10xi64> into i64
+  vector.print %0 : i64
+  // CHECK: -88
+  %1 = vector.reduction "mul", %v9 : vector<10xi64> into i64
+  vector.print %1 : i64
+  // CHECK: -1228800
+  %2 = vector.reduction "min", %v9 : vector<10xi64> into i64
+  vector.print %2 : i64
+  // CHECK: -80
+  %3 = vector.reduction "max", %v9 : vector<10xi64> into i64
+  vector.print %3 : i64
+  // CHECK: 5
+  %4 = vector.reduction "and", %v9 : vector<10xi64> into i64
+  vector.print %4 : i64
+  // CHECK: 0
+  %5 = vector.reduction "or", %v9 : vector<10xi64> into i64
+  vector.print %5 : i64
+  // CHECK: -1
+  %6 = vector.reduction "xor", %v9 : vector<10xi64> into i64
+  vector.print %6 : i64
+  // CHECK: -68
+
+  return
+}

mlir/integration_test/Dialect/Vector/CPU/test-shape-cast.mlir

+// RUN: mlir-opt %s -convert-scf-to-std -convert-vector-to-llvm -convert-std-to-llvm | \
+// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
+// RUN:   -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+
+func @entry() {
+  %f1 = constant 1.0: f32
+  %f2 = constant 2.0: f32
+  %f3 = constant 3.0: f32
+  %f4 = constant 4.0: f32
+  %f5 = constant 5.0: f32
+  %f6 = constant 6.0: f32
+
+  // Construct test vector.
+  %0 = vector.broadcast %f1 : f32 to vector<3x2xf32>
+  %1 = vector.insert %f2, %0[0, 1] : f32 into vector<3x2xf32>
+  %2 = vector.insert %f3, %1[1, 0] : f32 into vector<3x2xf32>
+  %3 = vector.insert %f4, %2[1, 1] : f32 into vector<3x2xf32>
+  %4 = vector.insert %f5, %3[2, 0] : f32 into vector<3x2xf32>
+  %x = vector.insert %f6, %4[2, 1] : f32 into vector<3x2xf32>
+  vector.print %x : vector<3x2xf32>
+  // CHECK:  ( ( 1, 2 ), ( 3, 4 ), ( 5, 6 ) )
+
+  // Reshapes.
+  %a = vector.shape_cast %x : vector<3x2xf32> to vector<3x2xf32>
+  %b = vector.shape_cast %x : vector<3x2xf32> to vector<2x3xf32>
+  %c = vector.shape_cast %x : vector<3x2xf32> to vector<6xf32>
+  %d = vector.shape_cast %c : vector<6xf32> to vector<2x3xf32>
+  %e = vector.shape_cast %c : vector<6xf32> to vector<3x2xf32>
+
+  // Reshaped vectors:
+  // CHECK:  ( ( 1, 2 ), ( 3, 4 ), ( 5, 6 ) )
+  // CHECK: ( ( 1, 2, 3 ), ( 4, 5, 6 ) )
+  // CHECK: ( 1, 2, 3, 4, 5, 6 )
+  // CHECK: ( ( 1, 2, 3 ), ( 4, 5, 6 ) )
+  // CHECK: ( ( 1, 2 ), ( 3, 4 ), ( 5, 6 ) )
+  vector.print %a : vector<3x2xf32>
+  vector.print %b : vector<2x3xf32>
+  vector.print %c : vector<6xf32>
+  vector.print %d : vector<2x3xf32>
+  vector.print %e : vector<3x2xf32>
+
+  return
+}

mlir/integration_test/Dialect/Vector/CPU/test-shuffle.mlir

+// RUN: mlir-opt %s -convert-scf-to-std -convert-vector-to-llvm -convert-std-to-llvm | \
+// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
+// RUN:   -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+
+func @entry() {
+  %f1 = constant 1.0: f32
+  %f2 = constant 2.0: f32
+  %v1 = vector.broadcast %f1 : f32 to vector<2x4xf32>
+  %v2 = vector.broadcast %f2 : f32 to vector<2x4xf32>
+  vector.print %v1 : vector<2x4xf32>
+  vector.print %v2 : vector<2x4xf32>
+  //
+  // test vectors:
+  //
+  // CHECK: ( ( 1, 1, 1, 1 ), ( 1, 1, 1, 1 ) )
+  // CHECK: ( ( 2, 2, 2, 2 ), ( 2, 2, 2, 2 ) )
+
+  %v3 = vector.shuffle %v1, %v2 [3, 1, 2] : vector<2x4xf32>, vector<2x4xf32>
+  vector.print %v3 : vector<3x4xf32>
+  // CHECK: ( ( 2, 2, 2, 2 ), ( 1, 1, 1, 1 ), ( 2, 2, 2, 2 ) )
+
+  return
+}

mlir/integration_test/Dialect/Vector/CPU/test-transfer-read.mlir

+// RUN: mlir-opt %s -convert-scf-to-std -convert-vector-to-llvm -convert-std-to-llvm | \
+// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
+// RUN:   -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+
+func @transfer_read_1d(%A : memref<?xf32>, %base: index) {
+  %fm42 = constant -42.0: f32
+  %f = vector.transfer_read %A[%base], %fm42
+      {permutation_map = affine_map<(d0) -> (d0)>} :
+    memref<?xf32>, vector<13xf32>
+  vector.print %f: vector<13xf32>
+  return
+}
+
+func @transfer_write_1d(%A : memref<?xf32>, %base: index) {
+  %f0 = constant 0.0 : f32
+  %vf0 = splat %f0 : vector<4xf32>
+  vector.transfer_write %vf0, %A[%base]
+      {permutation_map = affine_map<(d0) -> (d0)>} :
+    vector<4xf32>, memref<?xf32>
+  return
+}
+
+func @entry() {
+  %c0 = constant 0: index
+  %c1 = constant 1: index
+  %c2 = constant 2: index
+  %c3 = constant 3: index
+  %c4 = constant 4: index
+  %c5 = constant 5: index
+  %A = alloc(%c5) : memref<?xf32>
+  scf.for %i = %c0 to %c5 step %c1 {
+    %i32 = index_cast %i : index to i32
+    %fi = sitofp %i32 : i32 to f32
+    store %fi, %A[%i] : memref<?xf32>
+  }
+  // On input, memory contains [[ 0, 1, 2, 3, 4, xxx garbage xxx ]]
+  // Read shifted by 2 and pad with -42:
+  //   ( 2, 3, 4, -42, ..., -42)
+  call @transfer_read_1d(%A, %c2) : (memref<?xf32>, index) -> ()
+  // Write into memory shifted by 3
+  //   memory contains [[ 0, 1, 2, 0, 0, xxx garbage xxx ]]
+  call @transfer_write_1d(%A, %c3) : (memref<?xf32>, index) -> ()
+  // Read shifted by 0 and pad with -42:
+  //   ( 0, 1, 2, 0, 0, -42, ..., -42)
+  call @transfer_read_1d(%A, %c0) : (memref<?xf32>, index) -> ()
+  return
+}
+
+// CHECK: ( 2, 3, 4, -42, -42, -42, -42, -42, -42, -42, -42, -42, -42 )
+// CHECK: ( 0, 1, 2, 0, 0, -42, -42, -42, -42, -42, -42, -42, -42 )

mlir/integration_test/Dialect/Vector/CPU/test-transfer-write.mlir

+// RUN: mlir-opt %s -convert-scf-to-std -convert-vector-to-llvm -convert-std-to-llvm | \
+// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
+// RUN:   -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+
+func @transfer_write16_1d(%A : memref<?xf32>, %base: index) {
+  %f = constant 16.0 : f32
+  %v = splat %f : vector<16xf32>
+  vector.transfer_write %v, %A[%base]
+    {permutation_map = affine_map<(d0) -> (d0)>}
+    : vector<16xf32>, memref<?xf32>
+  return
+}
+
+func @transfer_write13_1d(%A : memref<?xf32>, %base: index) {
+  %f = constant 13.0 : f32
+  %v = splat %f : vector<13xf32>
+  vector.transfer_write %v, %A[%base]
+    {permutation_map = affine_map<(d0) -> (d0)>}
+    : vector<13xf32>, memref<?xf32>
+  return
+}
+
+func @transfer_write17_1d(%A : memref<?xf32>, %base: index) {
+  %f = constant 17.0 : f32
+  %v = splat %f : vector<17xf32>
+  vector.transfer_write %v, %A[%base]
+    {permutation_map = affine_map<(d0) -> (d0)>}
+    : vector<17xf32>, memref<?xf32>
+  return
+}
+
+func @transfer_read_1d(%A : memref<?xf32>) -> vector<32xf32> {
+  %z = constant 0: index
+  %f = constant 0.0: f32
+  %r = vector.transfer_read %A[%z], %f
+    {permutation_map = affine_map<(d0) -> (d0)>}
+    : memref<?xf32>, vector<32xf32>
+  return %r : vector<32xf32>
+}
+
+func @entry() {
+  %c0 = constant 0: index
+  %c1 = constant 1: index
+  %c32 = constant 32: index
+  %A = alloc(%c32) {alignment=64} : memref<?xf32>
+  scf.for %i = %c0 to %c32 step %c1 {
+    %f = constant 0.0: f32
+    store %f, %A[%i] : memref<?xf32>
+  }
+
+  // On input, memory contains all zeros.
+  %0 = call @transfer_read_1d(%A) : (memref<?xf32>) -> (vector<32xf32>)
+  vector.print %0 : vector<32xf32>
+
+  // Overwrite with 16 values of 16 at base 4.
+  %c4 = constant 4: index
+  call @transfer_write16_1d(%A, %c4) : (memref<?xf32>, index) -> ()
+  %1 = call @transfer_read_1d(%A) : (memref<?xf32>) -> (vector<32xf32>)
+  vector.print %1 : vector<32xf32>
+
+  // Overwrite with 13 values of 13 at base 3.
+  %c3 = constant 3: index
+  call @transfer_write13_1d(%A, %c3) : (memref<?xf32>, index) -> ()
+  %2 = call @transfer_read_1d(%A) : (memref<?xf32>) -> (vector<32xf32>)
+  vector.print %2 : vector<32xf32>
+
+  // Overwrite with 17 values of 17 at base 7.
+  %c7 = constant 7: index
+  call @transfer_write17_1d(%A, %c3) : (memref<?xf32>, index) -> ()
+  %3 = call @transfer_read_1d(%A) : (memref<?xf32>) -> (vector<32xf32>)
+  vector.print %3 : vector<32xf32>
+
+  // Overwrite with 13 values of 13 at base 8.
+  %c8 = constant 8: index
+  call @transfer_write13_1d(%A, %c8) : (memref<?xf32>, index) -> ()
+  %4 = call @transfer_read_1d(%A) : (memref<?xf32>) -> (vector<32xf32>)
+  vector.print %4 : vector<32xf32>
+
+  // Overwrite with 17 values of 17 at base 14.
+  %c14 = constant 14: index
+  call @transfer_write17_1d(%A, %c14) : (memref<?xf32>, index) -> ()
+  %5 = call @transfer_read_1d(%A) : (memref<?xf32>) -> (vector<32xf32>)
+  vector.print %5 : vector<32xf32>
+
+  // Overwrite with 13 values of 13 at base 19.
+  %c19 = constant 19: index
+  call @transfer_write13_1d(%A, %c19) : (memref<?xf32>, index) -> ()
+  %6 = call @transfer_read_1d(%A) : (memref<?xf32>) -> (vector<32xf32>)
+  vector.print %6 : vector<32xf32>
+
+  return
+}
+
+// CHECK: ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
+// CHECK: ( 0, 0, 0, 0, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
+// CHECK: ( 0, 0, 0, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 16, 16, 16, 16, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
+// CHECK: ( 0, 0, 0, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
+// CHECK: ( 0, 0, 0, 17, 17, 17, 17, 17, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
+// CHECK: ( 0, 0, 0, 17, 17, 17, 17, 17, 13, 13, 13, 13, 13, 13, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 0 )
+// CHECK: ( 0, 0, 0, 17, 17, 17, 17, 17, 13, 13, 13, 13, 13, 13, 17, 17, 17, 17, 17, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13 )

mlir/integration_test/Dialect/Vector/CPU/test-transpose.mlir

+// RUN: mlir-opt %s -convert-scf-to-std -convert-vector-to-llvm -convert-std-to-llvm | \
+// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
+// RUN:   -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
+// RUN: FileCheck %s
+
+func @entry() {
+  %f0 = constant 0.0: f32
+  %f1 = constant 1.0: f32
+  %f2 = constant 2.0: f32
+  %f3 = constant 3.0: f32
+  %f4 = constant 4.0: f32
+  %f5 = constant 5.0: f32
+  %f6 = constant 6.0: f32
+  %f7 = constant 7.0: f32
+  %f8 = constant 8.0: f32
+
+  // Construct test vectors and matrices.
+  %0 = vector.broadcast %f1 : f32 to vector<2xf32>
+  %a = vector.insert %f2, %0[1] : f32 into vector<2xf32>
+  %1 = vector.broadcast %f3 : f32 to vector<2xf32>
+  %b = vector.insert %f4, %1[1] : f32 into vector<2xf32>
+  %2 = vector.broadcast %f5 : f32 to vector<2xf32>
+  %c = vector.insert %f6, %2[1] : f32 into vector<2xf32>
+  %3 = vector.broadcast %f7 : f32 to vector<2xf32>
+  %d = vector.insert %f8, %3[1] : f32 into vector<2xf32>
+  %4 = vector.broadcast %f0 : f32 to vector<2x2xf32>
+  %5 = vector.insert %a, %4[0] : vector<2xf32> into vector<2x2xf32>
+  %A = vector.insert %b, %5[1] : vector<2xf32> into vector<2x2xf32>
+  %6 = vector.broadcast %f0 : f32 to vector<2x2xf32>
+  %7 = vector.insert %c, %6[0] : vector<2xf32> into vector<2x2xf32>
+  %B = vector.insert %d, %7[1] : vector<2xf32> into vector<2x2xf32>
+  %8 = vector.broadcast %f0 : f32 to vector<3x2xf32>
+  %9 = vector.insert %a, %8[0] : vector<2xf32> into vector<3x2xf32>
+  %10 = vector.insert %b, %9[1] : vector<2xf32> into vector<3x2xf32>
+  %C = vector.insert %c, %10[2] : vector<2xf32> into vector<3x2xf32>
+  %11 = vector.tuple %A, %B : vector<2x2xf32>, vector<2x2xf32>
+  %D = vector.insert_slices %11, [2, 2], [1, 1]
+    : tuple<vector<2x2xf32>, vector<2x2xf32>> into vector<2x4xf32>
+
+  vector.print %A : vector<2x2xf32>
+  vector.print %B : vector<2x2xf32>
+  vector.print %C : vector<3x2xf32>
+  vector.print %D : vector<2x4xf32>
+  //
+  // test matrices:
+  //
+  // CHECK: ( ( 1, 2 ), ( 3, 4 ) )
+  // CHECK: ( ( 5, 6 ), ( 7, 8 ) )
+  // CHECK: ( ( 1, 2 ), ( 3, 4 ), ( 5, 6 ) )
+  // CHECK: ( ( 1, 2, 5, 6 ), ( 3, 4, 7, 8 ) )
+
+  %tA = vector.transpose %A, [1, 0] : vector<2x2xf32> to vector<2x2xf32>
+  %tB = vector.transpose %B, [1, 0] : vector<2x2xf32> to vector<2x2xf32>
+  %tC = vector.transpose %C, [1, 0] : vector<3x2xf32> to vector<2x3xf32>
+  %tD = vector.transpose %D, [1, 0] : vector<2x4xf32> to vector<4x2xf32>
+
+  vector.print %tA : vector<2x2xf32>
+  vector.print %tB : vector<2x2xf32>
+  vector.print %tC : vector<2x3xf32>
+  vector.print %tD : vector<4x2xf32>
+  //
+  // transposed matrices:
+  //
+  // CHECK: ( ( 1, 3 ), ( 2, 4 ) )
+  // CHECK: ( ( 5, 7 ), ( 6, 8 ) )
+  // CHECK: ( ( 1, 3, 5 ), ( 2, 4, 6 ) )
+  // CHECK: ( ( 1, 3 ), ( 2, 4 ), ( 5, 7 ), ( 6, 8 ) )
+
+  %idD = vector.transpose %D, [0, 1] : vector<2x4xf32> to vector<2x4xf32>
+  %ttD = vector.transpose %tD, [1, 0] : vector<4x2xf32> to vector<2x4xf32>
+
+  vector.print %idD : vector<2x4xf32>
+  vector.print %ttD : vector<2x4xf32>
+  //
+  // back to original after transpose matrices:
+  //
+  // CHECK: ( ( 1, 2, 5, 6 ), ( 3, 4, 7, 8 ) )
+  // CHECK: ( ( 1, 2, 5, 6 ), ( 3, 4, 7, 8 ) )
+
+  // Construct test tensor.
+  %p = vector.broadcast %f1 : f32 to vector<2x2x2xf32>
+  %q = vector.insert %f2, %p[0, 0, 1] : f32 into vector<2x2x2xf32>
+  %r = vector.insert %f3, %q[0, 1, 0] : f32 into vector<2x2x2xf32>
+  %s = vector.insert %f4, %r[0, 1, 1] : f32 into vector<2x2x2xf32>
+  %t = vector.insert %f5, %s[1, 0, 0] : f32 into vector<2x2x2xf32>
+  %u = vector.insert %f6, %t[1, 0, 1] : f32 into vector<2x2x2xf32>
+  %v = vector.insert %f7, %u[1, 1, 0] : f32 into vector<2x2x2xf32>
+  %w = vector.insert %f8, %v[1, 1, 1] : f32 into vector<2x2x2xf32>
+
+  vector.print %w : vector<2x2x2xf32>
+  //
+  // test tensors:
+  //
+  // CHECK: ( ( ( 1, 2 ), ( 3, 4 ) ), ( ( 5, 6 ), ( 7, 8 ) ) )
+
+  %tP = vector.transpose %w, [0, 1, 2] : vector<2x2x2xf32> to vector<2x2x2xf32>
+  %tQ = vector.transpose %w, [0, 2, 1] : vector<2x2x2xf32> to vector<2x2x2xf32>
+  %tR = vector.transpose %w, [1, 0, 2] : vector<2x2x2xf32> to vector<2x2x2xf32>
+  %tS = vector.transpose %w, [2, 0, 1] : vector<2x2x2xf32> to vector<2x2x2xf32>
+  %tT = vector.transpose %w, [1, 2, 0] : vector<2x2x2xf32> to vector<2x2x2xf32>
+  %tU = vector.transpose %w, [2, 1, 0] : vector<2x2x2xf32> to vector<2x2x2xf32>
+
+  vector.print %tP : vector<2x2x2xf32>
+  vector.print %tQ : vector<2x2x2xf32>
+  vector.print %tR : vector<2x2x2xf32>
+  vector.print %tS : vector<2x2x2xf32>
+  vector.print %tT : vector<2x2x2xf32>
+  vector.print %tU : vector<2x2x2xf32>
+  //
+  // transposed tensors:
+  //
+  // CHECK: ( ( ( 1, 2 ), ( 3, 4 ) ), ( ( 5, 6 ), ( 7, 8 ) ) )
+  // CHECK: ( ( ( 1, 3 ), ( 2, 4 ) ), ( ( 5, 7 ), ( 6, 8 ) ) )
+  // CHECK: ( ( ( 1, 2 ), ( 5, 6 ) ), ( ( 3, 4 ), ( 7, 8 ) ) )
+  // CHECK: ( ( ( 1, 3 ), ( 5, 7 ) ), ( ( 2, 4 ), ( 6, 8 ) ) )
+  // CHECK: ( ( ( 1, 5 ), ( 2, 6 ) ), ( ( 3, 7 ), ( 4, 8 ) ) )
+  // CHECK: ( ( ( 1, 5 ), ( 3, 7 ) ), ( ( 2, 6 ), ( 4, 8 ) ) )
+
+  return
+}

mlir/integration_test/lit.cfg.py

+# -*- Python -*-
+
+import os
+import platform
+import re
+import subprocess
+import tempfile
+
+import lit.formats
+import lit.util
+
+from lit.llvm import llvm_config
+from lit.llvm.subst import ToolSubst
+
+# Configuration file for the 'lit' integration test runner.
+
+# name: The name of this integration test suite.
+config.name = 'MLIR_INTEGRATION'
+
+config.test_format = lit.formats.ShTest(not llvm_config.use_lit_shell)
+
+# suffixes: A list of file extensions to treat as integration test files.
+config.suffixes = ['.mlir']
+
+# test_source_root: The root path where integration tests are located.
+config.test_source_root = os.path.dirname(__file__)
+
+# test_exec_root: The root path where integration tests should be run.
+config.test_exec_root = os.path.join(config.mlir_obj_root, 'integration_test')
+
+config.substitutions.append(('%PATH%', config.environment['PATH']))
+config.substitutions.append(('%shlibext', config.llvm_shlib_ext))
+config.substitutions.append(('%mlir_src_root', config.mlir_src_root))
+
+llvm_config.with_system_environment(['HOME', 'INCLUDE', 'LIB', 'TMP', 'TEMP'])
+
+llvm_config.use_default_substitutions()
+
+# excludes: A list of directories to exclude from the integraiton testsuite.
+config.excludes = ['CMakeLists.txt', 'README.txt', 'LICENSE.txt']
+
+# Tweak the PATH to include the tools dir.
+llvm_config.with_environment('PATH', config.llvm_tools_dir, append_path=True)
+tool_dirs = [config.mlir_tools_dir, config.llvm_tools_dir]
+tools = [
+    'mlir-opt',
+    'mlir-cpu-runner',
+]
+
+# The following tools are optional.
+tools.extend([
+    ToolSubst(
+        '%mlir_integration_test_dir',
+        config.mlir_integration_test_dir,
+        unresolved='ignore'),
+])
+
+llvm_config.add_tool_substitutions(tools, tool_dirs)

mlir/integration_test/lit.site.cfg.py.in

+@LIT_SITE_CFG_IN_HEADER@
+
+import sys
+
+config.host_triple = "@LLVM_HOST_TRIPLE@"
+config.target_triple = "@TARGET_TRIPLE@"
+config.llvm_src_root = "@LLVM_SOURCE_DIR@"
+config.llvm_obj_root = "@LLVM_BINARY_DIR@"
+config.llvm_tools_dir = "@LLVM_TOOLS_DIR@"
+config.llvm_lib_dir = "@LLVM_LIBRARY_DIR@"
+config.llvm_shlib_dir = "@SHLIBDIR@"
+config.llvm_shlib_ext = "@SHLIBEXT@"
+config.llvm_exe_ext = "@EXEEXT@"
+config.lit_tools_dir = "@LLVM_LIT_TOOLS_DIR@"
+config.python_executable = "@PYTHON_EXECUTABLE@"
+config.gold_executable = "@GOLD_EXECUTABLE@"
+config.ld64_executable = "@LD64_EXECUTABLE@"
+config.enable_shared = @ENABLE_SHARED@
+config.enable_assertions = @ENABLE_ASSERTIONS@
+config.targets_to_build = "@TARGETS_TO_BUILD@"
+config.native_target = "@LLVM_NATIVE_ARCH@"
+config.llvm_bindings = "@LLVM_BINDINGS@".split(' ')
+config.host_os = "@HOST_OS@"
+config.host_cc = "@HOST_CC@"
+config.host_cxx = "@HOST_CXX@"
+config.host_cmake = "@CMAKE_COMMAND@"
+# Note: ldflags can contain double-quoted paths, so must use single quotes here.
+config.host_ldflags = '@HOST_LDFLAGS@'
+config.llvm_use_sanitizer = "@LLVM_USE_SANITIZER@"
+config.llvm_host_triple = '@LLVM_HOST_TRIPLE@'
+config.host_arch = "@HOST_ARCH@"
+config.mlir_src_root = "@MLIR_SOURCE_DIR@"
+config.mlir_obj_root = "@MLIR_BINARY_DIR@"
+config.mlir_tools_dir = "@MLIR_TOOLS_DIR@"
+config.mlir_integration_test_dir = "@MLIR_INTEGRATION_TEST_DIR@"
+
+# Support substitution of the tools_dir with user parameters. This is
+# used when we can't determine the tool dir at configuration time.
+try:
+    config.llvm_tools_dir = config.llvm_tools_dir % lit_config.params
+    config.llvm_shlib_dir = config.llvm_shlib_dir % lit_config.params
+except KeyError:
+    e = sys.exc_info()[1]
+    key, = e.args
+    lit_config.fatal("unable to find %r parameter, use '--param=%s=VALUE'" % (key,key))
+
+import lit.llvm
+lit.llvm.initialize(lit_config, config)
+
+# Let the main config do the real work.
+lit_config.load_config(config, "@MLIR_SOURCE_DIR@/integration_test/lit.cfg.py")