[mlir][VectorOps] Use SCF for vector.print and allow scalable vectors

[benmxwl-arm]

Summary

This patch splits the lowering of vector.print into first converting
an n-D print into a loop of scalar prints of the elements, then a second
pass that converts those scalar prints into the runtime calls. The
former is done in VectorToSCF and the latter in VectorToLLVM.

The main reason for this is to allow printing scalable vector types,
which are not possible to fully unroll at compile time, though this
also avoids fully unrolling very large vectors.

To allow VectorToSCF to add the necessary punctuation between vectors
and elements, a "punctuation" attribute has been added to vector.print.
This abstracts calling the runtime functions such as printNewline(),
without leaking the LLVM details into the higher abstraction levels.
For example:

vector.print <comma>

lowers to

llvm.call @printComma() : () -> ()

The output format and runtime functions remain the same, which avoids
the need to alter a large number of tests (aside from the pipelines).

[benmxwl-arm]

Rebase, correct comments, and update some missed tests.

[kuhar]

I can see one test failure:

FAIL: MLIR :: Integration/Dialect/LLVMIR/CPU/X86/test-inline-asm-vector.mlir (210 of 2225)
******************** TEST 'MLIR :: Integration/Dialect/LLVMIR/CPU/X86/test-inline-asm-vector.mlir' FAILED ********************
Script:
--
: 'RUN: at line 1';   /llvm/relass/bin/mlir-opt /llvm/llvm-project/mlir/test/Integration/Dialect/LLVMIR/CPU/X86/test-inline-asm-vector.mlir -convert-vector-to-llvm |   /llvm/relass/bin/mlir-cpu-runner -e entry_point_with_all_constants -entry-point-result=void    -shared-libs=/llvm/relass/lib/libmlir_c_runner_utils.so
--
Exit Code: 1

Command Output (stderr):
--
loc("<stdin>":4:5): error: Dialect `vector' not found for custom op 'vector.print' 
could not parse the input IR

[c-rhodes]

Thanks for patch the Ben, this will improve tests like the one added in D155839 and mlir/test/Integration/Dialect/Vector/CPU/ArmSME/vector-load-store.mlir that print scalable vectors. I've left a few minor comments but otherwise LGTM, aside from the test failure @kuhar mentioned. Please allow time for others to review.

mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp

1420–1429

nit: splits

mlir/lib/Conversion/VectorToSCF/VectorToSCF.cpp

716

nit: a comment here like "shape cast to rank 1" would be helpful

mlir/test/Conversion/VectorToSCF/vector-to-scf.mlir

566

nit: actual names like IS_NOT_LAST_ITER would be helpful in the tests

[benmxwl-arm]

• Fix test-inline-asm-vector.mlir (thanks for the pointer)
• Always extend odd integer sizes (non-pow2 or < i8) to avoid backend issues
• Use actual names in tests
• Fix nits

[awarzynski]

Thanks for working on this @benmxwl-arm, this is a very nice and a very welcome improvement for scalable vectors!

a "punctuation" attribute has been added to vector.print.
This abstracts calling the runtime functions such as printNewline(),
without leaking the LLVM details into the higher abstraction levels.
For example:

vector.print <comma>

lowers to

lvm.call @printComma() : () -> ()

While this avoids leaking the LLVM details higher up, it requires vector.printto evolve into a bit of a "Swiss knife" that can print Vector(s) and other things too :) It would be good to get an opinion from vector.print veterans - perhaps @aartbik or @rriddle ?

mlir/include/mlir/Dialect/Vector/IR/VectorOps.td

2483

Why Optional? So that vector.print can print punctuation without anything else?

mlir/lib/Conversion/VectorToSCF/VectorToSCF.cpp

675

Why not llvm.call @printComma() : () -> ()?

mlir/test/Conversion/VectorToSCF/vector-to-scf.mlir

623

How about 2d scalable vectors?

mlir/test/Integration/Dialect/LLVMIR/CPU/X86/test-inline-asm-vector.mlir

6

Which conversion fails if you keep this as llvm.func?

mlir/test/Integration/Dialect/Vector/CPU/test-compress.mlir

1

This is a bit unrelated. We should either switch all tests to -test-lower-to-llvm or none. IMO, it would be safer to keep the original pipeline. Switching to -test-lower-to-llvm could be proposed separately.

[benmxwl-arm]

While this avoids leaking the LLVM details higher up, it requires vector.printto evolve into a bit of a "Swiss knife" that can print Vector(s) and other things too :) It would be good to get an opinion from vector.print veterans - perhaps aartbik or rriddle ?

Note that vector.print already is a bit of a "swiss knife" as it supports printing scalars already (and is used for non-vector things), so it does not seem like a huge stretch :)

mlir/include/mlir/Dialect/Vector/IR/VectorOps.td

2483

Yes, it has to be able be used alone for the VectorToSCF pass, see the test outputs :)

mlir/lib/Conversion/VectorToSCF/VectorToSCF.cpp

675

Because that'd be introducing LLVM specifics before lowering to LLVM, which does not seem right (and is not done elsewhere).

mlir/test/Conversion/VectorToSCF/vector-to-scf.mlir

623

I could add a test for the SCF output here (which I believe is valid), though currently >= 2D scalables can't be lowered any further.

mlir/test/Integration/Dialect/LLVMIR/CPU/X86/test-inline-asm-vector.mlir

6

I may be lowering it incorrectly, but it gets stuck with some index types and cf.brs that fail to lower to LLVM following something like:
-convert-vector-to-scf -convert-scf-to-cf -convert-cf-to-llvm -convert-vector-to-llvm -convert-arith-to-llvm -reconcile-unrealized-casts

mlir/test/Integration/Dialect/Vector/CPU/test-compress.mlir

1

It's mainly because these tests use memrefs now, and hand-rolling the pipeline is quite unwieldy (and the original can't be kept anyway).

[benmxwl-arm]

Note that if I removed the punctuation attribute entire lowering would need to be done in VectorToSCF directly calling the runtime functions, since individual calls to vector.print would add newlines after each element (currently disabled via the punctuation attribute), which would be incorrect when not printing a scalar.

[aartbik]

Thanks for adding this. I am quite happy that this (1) support scalable vectors now and (2) the loops instead of unrolling will actually avoid some of the code size issues we were seeing.
I would like to see a bit more explanation in the vector op documentation itself, please.

mlir/include/mlir/Dialect/Vector/IR/VectorOps.td

2483

It feels like the documentation needs to be updated a bit more on this. We now, if I understand correctly, have the "pure vector print that still implies punctuation" and then
after lowering, we get into the semantics of printing the punctuation through the attribute and scalars. I understand why you picked this (not having the llvm dep early), but it is something that could use some explanation

2497–2499

can we add another example with punctuation also now?

2500–2501

broken is a bit ambigous, can we use something like decomposed or so?

mlir/include/mlir/IR/BuiltinTypes.td

378 ↗

(On Diff #545643)

this feels like it should be in a separate CL

mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp

1419–1420

Now that you are here, this is no longer proof-of-concept, but simply "Lowering implementation ...."

mlir/test/Integration/Dialect/Vector/CPU/test-transfer-read-1d.mlir

1 ↗

(On Diff #545643)

this seems to repeat the pass?

[benmxwl-arm]

• Update vector.print documentation
• Remove changes to BuiltinTypes.td
• Add 2D scalable test case in vector-to-scf.mlir
• Address miscellaneous comments :)

[aartbik]

mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp

1441

since you are not using the result of the cast ,just use isa?

[awarzynski]

mlir/test/Conversion/VectorToSCF/vector-to-scf.mlir

623

To me that's a good reason to completely disallow 2D scalables. If we are not going to use it, it's effectively dead code 🤔 . This way we are much clearer about what's supported and what isn't.

[benmxwl-arm]

mlir/test/Conversion/VectorToSCF/vector-to-scf.mlir

623

It does not require extra code to support here (it comes as a part of already supporting n-D vectors + scalable vectors). It's just that currently 2D+ scalable vectors can't be lowered to LLVM, so the lowering actually fails before any specific to vector.print (e.g. at the arith.constant dense<0.0> : vector<[4]x[4]xf32> in the test case).

[awarzynski]

LGTM, thanks! (% request for a few more comments)

mlir/lib/Conversion/VectorToSCF/VectorToSCF.cpp

682–683

I am realising that "dynamic" may mean different thing depending on the context. I usually think of "dynamically" vs "statically" shaped tensors/memrefs. But that's not what you had in mind, is it? Could you elaborate a bit? In particular, what the actual limitation is? With e.g. references to vector.extract vs vector.extractelement?

You may also want to refer to:

• https://discourse.llvm.org/t/rfc-psa-remove-vector-extractelement-and-vector-insertelement-ops-in-favor-of-vector-extract-and-vector-insert-ops/, and
• https://reviews.llvm.org/D155034.

mlir/test/Conversion/VectorToSCF/vector-to-scf.mlir

623

This is a good point, but I would still be hesitant to allow vector.print %arg0 : vector<[4]x[4]xf32> anywhere. We know that it will never be lowered to anything useful (well, there's nothing on the horizon). In cases like this I try to follow the principle of least surprises :)

Having said that, I appreciate that it's a bit weird to impose such limitations at the Vector dialect level. Please go with whichever approach you prefer.

[benmxwl-arm]

Thanks for the reviews and approvals, however, I've decided to rework the patch a little.
Rather than have the spill to a memref, I now do a simple reshape of n-D vectors (which then allows indexing with SSA values via vector.extractelement).
This explicitly does not support 2D scalable vectors, but as those can't be lowered currently this is no loss.

The main advantage of this is it keeps the lowering pipeline simple, so I've now removed the few -test-lower-to-llvms I added.

This did however uncover a bug in the vector.extract folds, so this patch now depends on D157003.

[benmxwl-arm]

• Fix an accidental use-after-free
• Rebase on main

[awarzynski]

Thanks for the reviews and approvals, however, I've decided to rework the patch a little.

Fantastic refactor, thank you Ben! This simplifies your original implementation and makes it less intrusive (thinking about the tests). I've left one small comment, but otherwise re-LGTM!

Btw, as this latest update doesn't really affect the overall direction/logic (which has already been approved by 3 reviewers), I think that it's fine to land it without waiting for further re-approvals ;-)

mlir/lib/Conversion/VectorToSCF/VectorToSCF.cpp

721–728

I would bump this to the very top. In general, it's good to have this sort of assumptions documented and enforced quite early on (as opposed mixed with regular code). Not always possible, but should be fine in this case.

[benmxwl-arm]

Reverted due to test failures in the MLIR python bindings: https://lab.llvm.org/buildbot/#/builders/220/builds/25791
Will address those before re-landing.

[benmxwl-arm]

Relanded after updating the few test failures I could find (Python bindings and a few CUDA/GPU tests), I've tried to go over all the tests I can find (though I am mindful there's hardware tested I can't run).

[mehdi_amini]

Reverted because of the broken test, you tried to fix it as a follow up but that didn't work: it actually made it worse. The fix wasn't correct anyway, the problem is a parsing error where vector.print <open> wouldn't parse back.

$ echo 'vector.print <open>' | bin/mlir-opt
<stdin>:1:13: error: expected operation name in quotes
vector.print <open>
            ^

[mehdi_amini]

mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp

1436–1437

There isn't necessarily a parent module op I believe.

[benmxwl-arm]

Sorry for the inconvenience and thanks for reverting (sorry I missed the continued failures).

I see the issue with the ambiguous assembly format and have a fix for that.
I'm just now trying to ensure I can at least run the pipeline for gpu-to-cubin locally to verify it will build.

[awarzynski]

Hi @mehdi_amini,

In D156519#4575277, @mehdi_amini wrote:

Reverted because of the broken test

Sorry about the inconvenience.

We are actually struggling to reproduce the failure in gpu-to-cubin.mlir. These CUDA integration tests use pipelines (e.g. gpu-to-cubin) that just won't build without CUDA runtime/drivers (which we don't have). Perhaps we're missing something obvious, but so far this is proving quite tricky.

Do you know anyone able to and willing to share an mlir-print-ir-after-all dump with us? :) Perhaps there are some instructions somewhere on how to run these tests using free tools? Just to clarify - we only want to/need to compile these tests. We don't need to run them (i.e. with mlir-cpu-runner).

@benmxwl-arm will shortly update this PR with a fix for the ASM format, but "gpu-to-cubin.mlir" is likely to continue failing. I'm not sure how to resolve that without some external help.

[benmxwl-arm]

• Check parent module exists (required for print, note that getting the parent this way is not changed from the previous implementation)
• Ensure assembly format is round-trippable
• Best guess fix for gpu-to-cubin.mlir (no luck testing this locally)

[mehdi_amini]

We are actually struggling to reproduce the failure in gpu-to-cubin.mlir. These CUDA integration tests use pipelines (e.g. gpu-to-cubin) that just won't build without CUDA runtime/drivers (which we don't have). Perhaps we're missing something obvious, but so far this is proving quite tricky.

You can't run them, however you should be able to compile everything (assuming you enabled the NVPTX target).
You can just copy the test and invoke it from the regular test locations (the Integration Cuda test will be filtered if you don't have Cuda) or just call mlir-opt directly, for example there shouldn't be any issue doing:

bin/mlir-opt /mlir/test/Integration/GPU/CUDA/gpu-to-cubin.mlir  \
| bin/mlir-opt -pass-pipeline='builtin.module(gpu.module(strip-debuginfo,convert-gpu-to-nvvm,gpu-to-cubin))' \
| bin/mlir-opt -gpu-to-llvm

(this the test RUN description)

[mehdi_amini]

Patched this revision and ran locally:

PASS: MLIR :: Integration/GPU/CUDA/gpu-to-cubin.mlir (166 of 2066)

Just try to push at a time you can keep an eye on the bot if you wanna be safe!

[mehdi_amini]

Sorry for the inconvenience and thanks for reverting (sorry I missed the continued failures).

It's very easy to miss, because the bot won't send a notification when it is already broken, only when it goes from green to red.

[benmxwl-arm]

Thanks a lot for the help, and checking that test! I'll definitely be keeping a close eye this time.

The extra complication with the gpu-to-cubin pass is that it's gated under MLIR_ENABLE_CUDA_RUNNER and links with the CUDA libraries, which we didn't have.

It's very easy to miss, because the bot won't send a notification when it is already broken, only when it goes from green to red.

See that now 😅, I was expecting a notification.

mlir/include/mlir/Dialect/Vector/IR/VectorOps.td

def Vector_TransposeOp :
   let assemblyFormat = [{
     $vector `,` $transp attr-dict `:` type($vector) `to` type($result)
   }];
   let hasCanonicalizer = 1;
   let hasFolder = 1;
   let hasVerifier = 1;
 }
 
+def PrintPunctuation : I32EnumAttr<"PrintPunctuation",
+                                  "Punctuation for separating vectors or vector elements", [
+  I32EnumAttrCase<"NoPunctuation", 0, "no_punctuation">,
+  I32EnumAttrCase<"NewLine", 1, "newline">,
+  I32EnumAttrCase<"Comma", 2, "comma">,
+  I32EnumAttrCase<"Open", 3, "open">,
+  I32EnumAttrCase<"Close", 4, "close">
+]> {
+  let cppNamespace = "::mlir::vector";
+  let genSpecializedAttr = 0;
+}
+
+def Vector_PrintPunctuation : EnumAttr<Vector_Dialect, PrintPunctuation, "punctuation"> {
+  let assemblyFormat = "`<` $value `>`";
+}
+
 def Vector_PrintOp :
   Vector_Op<"print", []>,
-  Arguments<(ins Type<Or<[
+  Arguments<(ins Optional<Type<Or<[
     AnyVectorOfAnyRank.predicate,
     AnyInteger.predicate, Index.predicate, AnyFloat.predicate
-  ]>>:$source)> {
+  ]>>>:$source, DefaultValuedAttr<Vector_PrintPunctuation,
+                      "::mlir::vector::PrintPunctuation::NewLine">:$punctuation)
+  > {
   let summary = "print operation (for testing and debugging)";
   let description = [{
-    Prints the source vector (or scalar) to stdout in human readable
-    format (for testing and debugging). No return value.
+    Prints the source vector (or scalar) to stdout in a human-readable format
+    (for testing and debugging). No return value.
 
     Example:
 
     ```mlir
-    %0 = arith.constant 0.0 : f32
-    %1 = vector.broadcast %0 : f32 to vector<4xf32>
-    vector.print %1 : vector<4xf32>
+    %v = arith.constant dense<0.0> : vector<4xf32>
+    vector.print %v : vector<4xf32>
+    ```
 
-    when lowered to LLVM, the vector print is unrolled into
-    elementary printing method calls that at runtime will yield
+    When lowered to LLVM, the vector print is decomposed into elementary
+    printing method calls that at runtime will yield:
 
+    ```
     ( 0.0, 0.0, 0.0, 0.0 )
+    ```
 
-    on stdout when linked with a small runtime support library,
-    which only needs to provide a few printing methods (single
-    value for all data types, opening/closing bracket, comma,
-    newline).
+    This is printed to stdout via a small runtime support library, which only
+    needs to provide a few printing methods (single value for all data
+    types, opening/closing bracket, comma, newline).
+
+    By default `vector.print` adds a newline after the vector, but this can be
+    controlled by the `punctuation` attribute. For example, to print a comma
+    after instead do:
+
+    ```mlir
+    vector.print %v : vector<4xf32> #vector.punctuation<comma>
+    ```
+
+    Note that it is possible to use the punctuation attribute alone. The
+    following will print a single newline:
+
+    ```mlir
+    vector.print #vector.punctuation<newline>
     ```
   }];
   let extraClassDeclaration = [{
     Type getPrintType() {
       return getSource().getType();
     }
   }];
-  let assemblyFormat = "$source attr-dict `:` type($source)";
+  let builders = [
+    OpBuilder<(ins "PrintPunctuation":$punctuation), [{
+      build($_builder, $_state, {}, punctuation);
+    }]>,
+  ];
+
+  let assemblyFormat = "($source^ `:` type($source))? ($punctuation^)? attr-dict";
 }
 
 //===----------------------------------------------------------------------===//
 // Ops used for supporting progressive lowering and conversion type changes.
 // The Ops are typically not used directly by higher level dialects, but are
 // used by intra-dialect rewriting rules to bring vector operations closer
 // to the hardware ISA.
 //===----------------------------------------------------------------------===//

mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp

 #include "mlir/Target/LLVMIR/TypeToLLVM.h"
 #include "mlir/Transforms/DialectConversion.h"
 #include "llvm/Support/Casting.h"
 #include <optional>
 
 using namespace mlir;
 using namespace mlir::vector;
 
-// Helper to reduce vector type by one rank at front.
-static VectorType reducedVectorTypeFront(VectorType tp) {
-  assert((tp.getRank() > 1) && "unlowerable vector type");
-  return VectorType::get(tp.getShape().drop_front(), tp.getElementType(),
-                         tp.getScalableDims().drop_front());
-}
-
 // Helper to reduce vector type by *all* but one rank at back.
 static VectorType reducedVectorTypeBack(VectorType tp) {
   assert((tp.getRank() > 1) && "unlowerable vector type");
   return VectorType::get(tp.getShape().take_back(), tp.getElementType(),
                          tp.getScalableDims().take_back());
 }
 
 // Helper that picks the proper sequence for inserting.

 
 private:
   const bool force32BitVectorIndices;
 };
 
 class VectorPrintOpConversion : public ConvertOpToLLVMPattern<vector::PrintOp> {
 public:
   using ConvertOpToLLVMPattern<vector::PrintOp>::ConvertOpToLLVMPattern;
 
-  // Proof-of-concept lowering implementation that relies on a small
-  // runtime support library, which only needs to provide a few
-  // printing methods (single value for all data types, opening/closing
-  // bracket, comma, newline). The lowering fully unrolls a vector
-  // in terms of these elementary printing operations. The advantage
-  // of this approach is that the library can remain unaware of all
-  // low-level implementation details of vectors while still supporting
-  // output of any shaped and dimensioned vector. Due to full unrolling,
-  // this approach is less suited for very large vectors though.
+  // Lowering implementation that relies on a small runtime support library,
+  // which only needs to provide a few printing methods (single value for all
+  // data types, opening/closing bracket, comma, newline). The lowering splits
+  // the vector into elementary printing operations. The advantage of this
+  // approach is that the library can remain unaware of all low-level
+  // implementation details of vectors while still supporting output of any
+  // shaped and dimensioned vector.
+  //
+  // Note: This lowering only handles scalars, n-D vectors are broken into
+  // printing scalars in loops in VectorToSCF.
   //
   // TODO: rely solely on libc in future? something else?
   //
   LogicalResult
   matchAndRewrite(vector::PrintOp printOp, OpAdaptor adaptor,
                   ConversionPatternRewriter &rewriter) const override {
+    auto parent = printOp->getParentOfType<ModuleOp>();
+    auto loc = printOp->getLoc();
+
+    if (auto value = adaptor.getSource()) {
       Type printType = printOp.getPrintType();
+      if (dyn_cast<VectorType>(printType)) {
+        // Vectors should be broken into elementary print ops in VectorToSCF.
+        return failure();
+      }
+      if (failed(emitScalarPrint(rewriter, parent, loc, printType, value)))
+        return failure();
+    }
 
+    auto punct = printOp.getPunctuation();
+    if (punct != PrintPunctuation::NoPunctuation) {
+      emitCall(rewriter, printOp->getLoc(), [&] {
+        switch (punct) {
+        case PrintPunctuation::Close:
+          return LLVM::lookupOrCreatePrintCloseFn(parent);
+        case PrintPunctuation::Open:
+          return LLVM::lookupOrCreatePrintOpenFn(parent);
+        case PrintPunctuation::Comma:
+          return LLVM::lookupOrCreatePrintCommaFn(parent);
+        case PrintPunctuation::NewLine:
+          return LLVM::lookupOrCreatePrintNewlineFn(parent);
+        default:
+          llvm_unreachable("unexpected punctuation");
+        }
+      }());
+    }
+
+    rewriter.eraseOp(printOp);
+    return success();
+  }
+
+private:
+  enum class PrintConversion {
+    // clang-format off
+    None,
+    ZeroExt64,
+    SignExt64,
+    Bitcast16
+    // clang-format on
+  };
+
+  LogicalResult emitScalarPrint(ConversionPatternRewriter &rewriter,
+                                ModuleOp parent, Location loc, Type printType,
+                                Value value) const {
     if (typeConverter->convertType(printType) == nullptr)
       return failure();
 
     // Make sure element type has runtime support.
     PrintConversion conversion = PrintConversion::None;
-    VectorType vectorType = dyn_cast<VectorType>(printType);
-    Type eltType = vectorType ? vectorType.getElementType() : printType;
-    auto parent = printOp->getParentOfType<ModuleOp>();
     Operation *printer;
-    if (eltType.isF32()) {
+    if (printType.isF32()) {
       printer = LLVM::lookupOrCreatePrintF32Fn(parent);
-    } else if (eltType.isF64()) {
+    } else if (printType.isF64()) {
       printer = LLVM::lookupOrCreatePrintF64Fn(parent);
-    } else if (eltType.isF16()) {
+    } else if (printType.isF16()) {
       conversion = PrintConversion::Bitcast16; // bits!
       printer = LLVM::lookupOrCreatePrintF16Fn(parent);
-    } else if (eltType.isBF16()) {
+    } else if (printType.isBF16()) {
       conversion = PrintConversion::Bitcast16; // bits!
       printer = LLVM::lookupOrCreatePrintBF16Fn(parent);
-    } else if (eltType.isIndex()) {
+    } else if (printType.isIndex()) {
       printer = LLVM::lookupOrCreatePrintU64Fn(parent);
-    } else if (auto intTy = dyn_cast<IntegerType>(eltType)) {
+    } else if (auto intTy = dyn_cast<IntegerType>(printType)) {
       // Integers need a zero or sign extension on the operand
       // (depending on the source type) as well as a signed or
       // unsigned print method. Up to 64-bit is supported.
       unsigned width = intTy.getWidth();
       if (intTy.isUnsigned()) {
         if (width <= 64) {
           if (width < 64)
             conversion = PrintConversion::ZeroExt64;
if (printType.isF32()) {
         } else {
           return failure();
         }
       }
     } else {
       return failure();
     }
 
-    // Unroll vector into elementary print calls.
-    int64_t rank = vectorType ? vectorType.getRank() : 0;
-    Type type = vectorType ? vectorType : eltType;
-    emitRanks(rewriter, printOp, adaptor.getSource(), type, printer, rank,
-              conversion);
-    emitCall(rewriter, printOp->getLoc(),
-             LLVM::lookupOrCreatePrintNewlineFn(parent));
-    rewriter.eraseOp(printOp);
-    return success();
-  }
-
-private:
-  enum class PrintConversion {
-    // clang-format off
-    None,
-    ZeroExt64,
-    SignExt64,
-    Bitcast16
-    // clang-format on
-  };
-
-  void emitRanks(ConversionPatternRewriter &rewriter, Operation *op,
-                 Value value, Type type, Operation *printer, int64_t rank,
-                 PrintConversion conversion) const {
-    VectorType vectorType = dyn_cast<VectorType>(type);
-    Location loc = op->getLoc();
-    if (!vectorType) {
-      assert(rank == 0 && "The scalar case expects rank == 0");
-      switch (conversion) {
-      case PrintConversion::ZeroExt64:
-        value = rewriter.create<arith::ExtUIOp>(
-            loc, IntegerType::get(rewriter.getContext(), 64), value);
-        break;
-      case PrintConversion::SignExt64:
-        value = rewriter.create<arith::ExtSIOp>(
-            loc, IntegerType::get(rewriter.getContext(), 64), value);
-        break;
-      case PrintConversion::Bitcast16:
-        value = rewriter.create<LLVM::BitcastOp>(
-            loc, IntegerType::get(rewriter.getContext(), 16), value);
-        break;
-      case PrintConversion::None:
-        break;
-      }
-      emitCall(rewriter, loc, printer, value);
-      return;
-    }
-
-    auto parent = op->getParentOfType<ModuleOp>();
-    emitCall(rewriter, loc, LLVM::lookupOrCreatePrintOpenFn(parent));
-    Operation *printComma = LLVM::lookupOrCreatePrintCommaFn(parent);
-
-    if (rank <= 1) {
-      auto reducedType = vectorType.getElementType();
-      auto llvmType = typeConverter->convertType(reducedType);
-      int64_t dim = rank == 0 ? 1 : vectorType.getDimSize(0);
-      for (int64_t d = 0; d < dim; ++d) {
-        Value nestedVal = extractOne(rewriter, *getTypeConverter(), loc, value,
-                                     llvmType, /*rank=*/0, /*pos=*/d);
-        emitRanks(rewriter, op, nestedVal, reducedType, printer, /*rank=*/0,
-                  conversion);
-        if (d != dim - 1)
-          emitCall(rewriter, loc, printComma);
-      }
-      emitCall(rewriter, loc, LLVM::lookupOrCreatePrintCloseFn(parent));
-      return;
-    }
-
-    int64_t dim = vectorType.getDimSize(0);
-    for (int64_t d = 0; d < dim; ++d) {
-      auto reducedType = reducedVectorTypeFront(vectorType);
-      auto llvmType = typeConverter->convertType(reducedType);
-      Value nestedVal = extractOne(rewriter, *getTypeConverter(), loc, value,
-                                   llvmType, rank, d);
-      emitRanks(rewriter, op, nestedVal, reducedType, printer, rank - 1,
-                conversion);
-      if (d != dim - 1)
-        emitCall(rewriter, loc, printComma);
-    }
-    emitCall(rewriter, loc, LLVM::lookupOrCreatePrintCloseFn(parent));
+    return success();
   }
 
   // Helper to emit a call.
   static void emitCall(ConversionPatternRewriter &rewriter, Location loc,
                        Operation *ref, ValueRange params = ValueRange()) {
     rewriter.create<LLVM::CallOp>(loc, TypeRange(), SymbolRefAttr::get(ref),
                                   params);
   }

mlir/lib/Conversion/VectorToSCF/VectorToSCF.cpp

if (xferOp.getMask()) {
         xferOp.getMaskMutable().assign(buffers.maskBuffer);
       });
     }
 
     return success();
   }
 };
 
+/// Decompose a n-D PrintOp into a loop of elementary/scalar prints. This allows
+/// printing both scalable and fixed size vectors.
+///
+/// E.g.:
+/// ```
+/// vector.print %v : vector<[4]xi32>
+/// ```
+/// is rewritten to:
+/// ```
+/// %c0 = arith.constant 0 : index
+/// %c4 = arith.constant 4 : index
+/// %c1 = arith.constant 1 : index
+/// %vscale = vector.vscale
+/// %length = arith.muli %vscale, %c4 : index
+/// %lastIndex = arith.subi %length, %c1 : index
+/// vector.print <open>
+/// scf.for %i = %c0 to %length step %c1 {
+///   %el = vector.extractelement %v[%i : index] : vector<[4]xi32>
+///   vector.print %el : i32 <no_punctuation>
+///   %notLastIndex = arith.cmpi ult, %i, %lastIndex : index
+///   scf.if %notLastIndex {
+///     vector.print <comma>
+///   }
+/// }
+/// vector.print <close>
+/// vector.print
+/// ```
+///
+/// Note: A temporary buffer is allocated for rank > 1 vectors to allow dynamic
+/// indexing of elements.
+struct DecomposePrintOpConversion : public VectorToSCFPattern<vector::PrintOp> {
+  using VectorToSCFPattern<vector::PrintOp>::VectorToSCFPattern;
+  LogicalResult matchAndRewrite(vector::PrintOp printOp,
+                                PatternRewriter &rewriter) const override {
+    if (!printOp.getSource())
+      return failure();
+
+    VectorType vectorType = dyn_cast<VectorType>(printOp.getPrintType());
+    if (!vectorType)
+      return failure();
+
+    auto loc = printOp.getLoc();
+    auto value = printOp.getSource();
+
+    if (auto intTy = dyn_cast<IntegerType>(vectorType.getElementType())) {
+      // Oddly sized integers are (somewhat) buggy on a lot of backends, so to
+      // avoid issues extend them to a more standard size.
+      // https://github.com/llvm/llvm-project/issues/30613
+      auto width = intTy.getWidth();
+      auto legalWidth = llvm::NextPowerOf2(std::max(8u, width) - 1);
+      auto legalIntTy = IntegerType::get(rewriter.getContext(), legalWidth,
+                                         intTy.getSignedness());
+      // arith can only take signless integers, so we must cast back and forth.
+      auto signlessSourceVectorType =
+          vectorType.cloneWith({}, getIntTypeWithSignlessSemantics(intTy));
+      auto signlessTargetVectorType =
+          vectorType.cloneWith({}, getIntTypeWithSignlessSemantics(legalIntTy));
+      auto targetVectorType = vectorType.cloneWith({}, legalIntTy);
+      value = rewriter.create<vector::BitCastOp>(loc, signlessSourceVectorType,
+                                                 value);
+      if (width == 1 || intTy.isUnsigned())
+        value = rewriter.create<arith::ExtUIOp>(loc, signlessTargetVectorType,
+                                                value);
+      else
+        value = rewriter.create<arith::ExtSIOp>(loc, signlessTargetVectorType,
+                                                value);
+      value = rewriter.create<vector::BitCastOp>(loc, targetVectorType, value);
+      vectorType = targetVectorType;
+    }
+
+    Value alloc;
+    if (vectorType.getRank() > 1) {
+      // Spill vector to allow dynamically indexing elements.
+      alloc = vectorToMemref(rewriter, loc, value, vectorType);
+    } else if (vectorType.getRank() == 0) {
+      // Shape cast rank 0 vectors to rank 1.
+      vectorType = VectorType::get({1}, vectorType.getElementType());
+      value = rewriter.create<vector::ShapeCastOp>(loc, vectorType, value);
+    }
+
+    vector::PrintOp firstClose;
+    SmallVector<Value, 8> loopIndices;
+    auto shape = vectorType.getShape();
+    auto scalableDimensions = vectorType.getScalableDims();
+    for (unsigned d = 0; d < shape.size(); d++) {
+      // Setup loop bounds and step.
+      Value lowerBound = rewriter.create<arith::ConstantIndexOp>(loc, 0);
+      Value upperBound = rewriter.create<arith::ConstantIndexOp>(loc, shape[d]);
+      Value step = rewriter.create<arith::ConstantIndexOp>(loc, 1);
+      if (scalableDimensions[d]) {
+        auto vscale = rewriter.create<vector::VectorScaleOp>(
+            loc, rewriter.getIndexType());
+        upperBound = rewriter.create<arith::MulIOp>(loc, upperBound, vscale);
+      }
+      auto lastIndex = rewriter.create<arith::SubIOp>(loc, upperBound, step);
+
+      // Create a loop to print the elements surrounded by parentheses.
+      rewriter.create<vector::PrintOp>(loc, vector::PrintPunctuation::Open);
+      auto loop =
+          rewriter.create<scf::ForOp>(loc, lowerBound, upperBound, step);
+      auto printClose = rewriter.create<vector::PrintOp>(
+          loc, vector::PrintPunctuation::Close);
+      if (!firstClose)
+        firstClose = printClose;
+
+      auto loopIdx = loop.getInductionVar();
+      loopIndices.push_back(loopIdx);
+
+      // Print a comma after all but the last element.
+      rewriter.setInsertionPointToStart(loop.getBody());
+      auto notLastIndex = rewriter.create<arith::CmpIOp>(
+          loc, arith::CmpIPredicate::ult, loopIdx, lastIndex);
+      rewriter.create<scf::IfOp>(loc, notLastIndex,
+                                 [&](OpBuilder &builder, Location loc) {
+                                   builder.create<vector::PrintOp>(
+                                       loc, vector::PrintPunctuation::Comma);
+                                   builder.create<scf::YieldOp>(loc);
+                                 });
+
+      rewriter.setInsertionPointToStart(loop.getBody());
+    }
+
+    // Print the scalar elements in the inner most loop.
+    auto element = [&]() -> Value {
+      if (alloc)
+        return rewriter.create<memref::LoadOp>(loc, alloc, loopIndices);
+      return rewriter.create<vector::ExtractElementOp>(loc, value,
+                                                       loopIndices.front());
+    }();
+    rewriter.create<vector::PrintOp>(loc, element,
+                                     vector::PrintPunctuation::NoPunctuation);
+
+    rewriter.setInsertionPointAfter(firstClose);
+    rewriter.create<vector::PrintOp>(loc, printOp.getPunctuation());
+    rewriter.eraseOp(printOp);
+    return success();
+  }
+
+  static Value vectorToMemref(OpBuilder &builder, Location loc, Value value,
+                              VectorType type) {
+    SmallVector<int64_t, 8> memrefShape;
+    SmallVector<Value, 8> zeroIndices;
+    SmallVector<Value, 8> memrefDynamicDimensions;
+    auto shape = type.getShape();
+    auto scalableDimensions = type.getScalableDims();
+    auto zeroIndex = builder.create<arith::ConstantIndexOp>(loc, 0);
+    for (unsigned d = 0; d < shape.size(); d++) {
+      if (scalableDimensions[d]) {
+        memrefShape.push_back(ShapedType::kDynamic);
+        auto vscale =
+            builder.create<vector::VectorScaleOp>(loc, builder.getIndexType());
+        auto size = builder.create<arith::ConstantIndexOp>(loc, shape[d]);
+        auto scaledSize = builder.create<arith::MulIOp>(loc, size, vscale);
+        memrefDynamicDimensions.push_back(scaledSize);
+      } else {
+        memrefShape.push_back(shape[d]);
+      }
+      zeroIndices.push_back(zeroIndex);
+    }
+    auto memrefType = MemRefType::get(memrefShape, type.getElementType());
+    auto alloc = builder.create<memref::AllocaOp>(loc, memrefType,
+                                                  memrefDynamicDimensions);
+    builder.create<vector::TransferWriteOp>(loc, value, alloc,
+                                            ValueRange(zeroIndices));
+    return alloc;
+  }
+
+  static IntegerType getIntTypeWithSignlessSemantics(IntegerType intTy) {
+    return IntegerType::get(intTy.getContext(), intTy.getWidth(),
+                            IntegerType::Signless);
+  };
+};
+
 /// Progressive lowering of vector transfer ops: Unpack one dimension.
 ///
 /// 1. Unpack one dimension from the current buffer type and cast the buffer
 ///    to that new type. E.g.:
 ///    ```
 ///    %vec = memref.load %0[%1] : memref<5xvector<4x3xf32>>
 ///    vector.transfer_write %vec ...
 ///    ```
patterns.add<lowering_n_d::PrepareTransferReadConversion,
         patterns.getContext(), options);
   }
 
   if (options.targetRank == 1) {
     patterns.add<lowering_1_d::TransferOp1dConversion<TransferReadOp>,
                  lowering_1_d::TransferOp1dConversion<TransferWriteOp>>(
         patterns.getContext(), options);
   }
+  patterns.add<lowering_n_d::DecomposePrintOpConversion>(patterns.getContext(),
+                                                         options);
 }
 
 namespace {
 
 struct ConvertVectorToSCFPass
     : public impl::ConvertVectorToSCFBase<ConvertVectorToSCFPass> {
   ConvertVectorToSCFPass() = default;
   ConvertVectorToSCFPass(const VectorTransferToSCFOptions &options) {

mlir/test/Conversion/VectorToLLVM/vector-to-llvm.mlir

 }
 // CHECK-LABEL: @vector_print_scalar_f64(
 // CHECK-SAME: %[[A:.*]]: f64)
 //       CHECK:    llvm.call @printF64(%[[A]]) : (f64) -> ()
 //       CHECK:    llvm.call @printNewline() : () -> ()
 
 // -----
 
-func.func @vector_print_vector_0d(%arg0: vector<f32>) {
-  vector.print %arg0 : vector<f32>
-  return
-}
-// CHECK-LABEL: @vector_print_vector_0d(
-// CHECK-SAME: %[[A:.*]]: vector<f32>)
-//       CHECK: %[[T0:.*]] = builtin.unrealized_conversion_cast %[[A]] : vector<f32> to vector<1xf32>
-//       CHECK: llvm.call @printOpen() : () -> ()
-//       CHECK: %[[T1:.*]] = llvm.mlir.constant(0 : index) : i64
-//       CHECK: %[[T2:.*]] = llvm.extractelement %[[T0]][%[[T1]] : i64] : vector<1xf32>
-//       CHECK: llvm.call @printF32(%[[T2]]) : (f32) -> ()
-//       CHECK: llvm.call @printClose() : () -> ()
-//       CHECK: llvm.call @printNewline() : () -> ()
-//       CHECK: return
-
-// -----
-
-func.func @vector_print_vector(%arg0: vector<2x2xf32>) {
-  vector.print %arg0 : vector<2x2xf32>
-  return
-}
-// CHECK-LABEL: @vector_print_vector(
-// CHECK-SAME: %[[A:.*]]: vector<2x2xf32>)
-//       CHECK:    %[[VAL_1:.*]] = builtin.unrealized_conversion_cast %[[A]] : vector<2x2xf32> to !llvm.array<2 x vector<2xf32>>
-//       CHECK:    llvm.call @printOpen() : () -> ()
-//       CHECK:    %[[x0:.*]] = llvm.extractvalue %[[VAL_1]][0] : !llvm.array<2 x vector<2xf32>>
-//       CHECK:    llvm.call @printOpen() : () -> ()
-//       CHECK:    %[[x1:.*]] = llvm.mlir.constant(0 : index) : i64
-//       CHECK:    %[[x2:.*]] = llvm.extractelement %[[x0]][%[[x1]] : i64] : vector<2xf32>
-//       CHECK:    llvm.call @printF32(%[[x2]]) : (f32) -> ()
-//       CHECK:    llvm.call @printComma() : () -> ()
-//       CHECK:    %[[x3:.*]] = llvm.mlir.constant(1 : index) : i64
-//       CHECK:    %[[x4:.*]] = llvm.extractelement %[[x0]][%[[x3]] : i64] : vector<2xf32>
-//       CHECK:    llvm.call @printF32(%[[x4]]) : (f32) -> ()
-//       CHECK:    llvm.call @printClose() : () -> ()
-//       CHECK:    llvm.call @printComma() : () -> ()
-//       CHECK:    %[[x5:.*]] = llvm.extractvalue %[[VAL_1]][1] : !llvm.array<2 x vector<2xf32>>
-//       CHECK:    llvm.call @printOpen() : () -> ()
-//       CHECK:    %[[x6:.*]] = llvm.mlir.constant(0 : index) : i64
-//       CHECK:    %[[x7:.*]] = llvm.extractelement %[[x5]][%[[x6]] : i64] : vector<2xf32>
-//       CHECK:    llvm.call @printF32(%[[x7]]) : (f32) -> ()
-//       CHECK:    llvm.call @printComma() : () -> ()
-//       CHECK:    %[[x8:.*]] = llvm.mlir.constant(1 : index) : i64
-//       CHECK:    %[[x9:.*]] = llvm.extractelement %[[x5]][%[[x8]] : i64] : vector<2xf32>
-//       CHECK:    llvm.call @printF32(%[[x9]]) : (f32) -> ()
-//       CHECK:    llvm.call @printClose() : () -> ()
-//       CHECK:    llvm.call @printClose() : () -> ()
-//       CHECK:    llvm.call @printNewline() : () -> ()
-
-// -----
-
 func.func @extract_strided_slice1(%arg0: vector<4xf32>) -> vector<2xf32> {
   %0 = vector.extract_strided_slice %arg0 {offsets = [2], sizes = [2], strides = [1]} : vector<4xf32> to vector<2xf32>
   return %0 : vector<2xf32>
 }
 // CHECK-LABEL: @extract_strided_slice1(
 //  CHECK-SAME:    %[[A:.*]]: vector<4xf32>)
 //       CHECK:    %[[T0:.*]] = llvm.shufflevector %[[A]], %[[A]] [2, 3] : vector<4xf32>
 //       CHECK:    return %[[T0]] : vector<2xf32>

mlir/test/Conversion/VectorToSCF/vector-to-scf.mlir

 // CHECK:           scf.for %[[IDX:.*]] = %[[C_0]] to %[[UB]] step %[[STEP]] {
 // CHECK:             %[[MASK_VAL:.*]] = vector.extractelement %[[MASK_VEC]][%[[IDX]] : index] : vector<[16]xi1>
 // CHECK:             scf.if %[[MASK_VAL]] {
 // CHECK:               %[[VAL_TO_STORE:.*]] = vector.extractelement %{{.*}}[%[[IDX]] : index] : vector<[16]xf32>
 // CHECK:               memref.store %[[VAL_TO_STORE]], %[[ARG_0]][%[[IDX]]] : memref<?xf32, strided<[?], offset: ?>>
 // CHECK:             } else {
 // CHECK:             }
 // CHECK:           }
+
+// -----
+
+func.func @vector_print_vector_0d(%arg0: vector<f32>) {
+  vector.print %arg0 : vector<f32>
+  return
+}
+// CHECK-LABEL:   func.func @vector_print_vector_0d(
+// CHECK-SAME:                                      %[[RANK_0_VEC:.*]]: vector<f32>) {
+// CHECK:           %[[C0:.*]] = arith.constant 0 : index
+// CHECK:           %[[C1:.*]] = arith.constant 1 : index
+// CHECK:           %[[RANK_1_VEC:.*]] = vector.shape_cast %[[RANK_0_VEC]] : vector<f32> to vector<1xf32>
+// CHECK:           vector.print <open>
+// CHECK:           scf.for %[[IDX:.*]] = %[[C0]] to %[[C1]] step %[[C1]] {
+// CHECK:             %[[EL:.*]] = vector.extractelement %[[RANK_1_VEC]]{{\[}}%[[IDX]] : index] : vector<1xf32>
+// CHECK:             vector.print %[[EL]] : f32 <no_punctuation>
+// CHECK:             %[[IS_NOT_LAST:.*]] = arith.cmpi ult, %[[IDX]], %[[C0]] : index
+// CHECK:             scf.if %[[IS_NOT_LAST]] {
+// CHECK:               vector.print <comma>
+// CHECK:             }
+// CHECK:           }
+// CHECK:           vector.print <close>
+// CHECK:           vector.print
+// CHECK:           return
+// CHECK:         }
+
+// -----
+
+func.func @vector_print_vector(%arg0: vector<2x2xf32>) {
+  vector.print %arg0 : vector<2x2xf32>
+  return
+}
+// CHECK-LABEL:   func.func @vector_print_vector(
+// CHECK-SAME:                                   %[[VEC_2D:.*]]: vector<2x2xf32>) {
+// CHECK:           %[[C0:.*]] = arith.constant 0 : index
+// CHECK:           %[[C2:.*]] = arith.constant 2 : index
+// CHECK:           %[[C1:.*]] = arith.constant 1 : index
+// CHECK:           %[[ALLOCA_VEC:.*]] = memref.alloca() : memref<vector<2x2xf32>>
+// CHECK:           %[[ALLOCA:.*]] = memref.alloca() : memref<2x2xf32>
+// CHECK:           memref.store %[[VEC_2D]], %[[ALLOCA_VEC]][] : memref<vector<2x2xf32>>
+// CHECK:           %[[ALLOCA_TYPECAST:.*]] = vector.type_cast %[[ALLOCA_VEC]] : memref<vector<2x2xf32>> to memref<2xvector<2xf32>>
+// CHECK:           scf.for %[[COPY_IDX:.*]] = %[[C0]] to %[[C2]] step %[[C1]] {
+// CHECK:             %[[VEC_SLICE:.*]] = memref.load %[[ALLOCA_TYPECAST]]{{\[}}%[[COPY_IDX]]] : memref<2xvector<2xf32>>
+// CHECK:             vector.transfer_write %[[VEC_SLICE]], %[[ALLOCA]]{{\[}}%[[COPY_IDX]], %[[C0]]] {in_bounds = [true]} : vector<2xf32>, memref<2x2xf32>
+// CHECK:           }
+// CHECK:           vector.print <open>
+// CHECK:           scf.for %[[I:.*]] = %[[C0]] to %[[C2]] step %[[C1]] {
+// CHECK:             vector.print <open>
+// CHECK:             scf.for %[[J:.*]] = %[[C0]] to %[[C2]] step %[[C1]] {
+// CHECK:               %[[EL:.*]] = memref.load %[[ALLOCA]]{{\[}}%[[I]], %[[J]]] : memref<2x2xf32>
+// CHECK:               vector.print %[[EL]] : f32 <no_punctuation>
+// CHECK:               %[[IS_NOT_LAST_INNER:.*]] = arith.cmpi ult, %[[J]], %[[C1]] : index
+// CHECK:               scf.if %[[IS_NOT_LAST_INNER]] {
+// CHECK:                 vector.print <comma>
+// CHECK:               }
+// CHECK:             }
+// CHECK:             vector.print <close>
+// CHECK:             %[[IS_NOT_LAST_OUTER:.*]] = arith.cmpi ult, %[[I]], %[[C1]] : index
+// CHECK:             scf.if %[[IS_NOT_LAST_OUTER]] {
+// CHECK:               vector.print <comma>
+// CHECK:             }
+// CHECK:           }
+// CHECK:           vector.print <close>
+// CHECK:           vector.print
+// CHECK:           return
+// CHECK:         }
+
+// -----
+
+func.func @vector_print_scalable_vector(%arg0: vector<[4]xi32>) {
+  vector.print %arg0 : vector<[4]xi32>
+  return
+}
+// CHECK-LABEL:   func.func @vector_print_scalable_vector(
+// CHECK-SAME:                                            %[[SCALABLE_VEC:.*]]: vector<[4]xi32>) {
+// CHECK:           %[[C0:.*]] = arith.constant 0 : index
+// CHECK:           %[[C4:.*]] = arith.constant 4 : index
+// CHECK:           %[[C1:.*]] = arith.constant 1 : index
+// CHECK:           %[[VSCALE:.*]] = vector.vscale
+// CHECK:           %[[UPPER_BOUND:.*]] = arith.muli %[[VSCALE]], %[[C4]] : index
+// CHECK:           %[[LAST_IDX:.*]] = arith.subi %[[UPPER_BOUND]], %[[C1]] : index
+// CHECK:           vector.print <open>
+// CHECK:           scf.for %[[IDX:.*]] = %[[C0]] to %[[UPPER_BOUND]] step %[[C1]] {
+// CHECK:             %[[EL:.*]] = vector.extractelement %[[SCALABLE_VEC]]{{\[}}%[[IDX]] : index] : vector<[4]xi32>
+// CHECK:             vector.print %[[EL]] : i32 <no_punctuation>
+// CHECK:             %[[IS_NOT_LAST:.*]] = arith.cmpi ult, %[[IDX]], %[[LAST_IDX]] : index
+// CHECK:             scf.if %[[IS_NOT_LAST]] {
+// CHECK:               vector.print <comma>
+// CHECK:             }
+// CHECK:           }
+// CHECK:           vector.print <close>
+// CHECK:           vector.print
+// CHECK:           return
+// CHECK:         }
+
+// -----
+
+func.func @vector_print_2d_scalable_vector(%arg0: vector<[4]x[4]xf32>) {
+  vector.print %arg0 : vector<[4]x[4]xf32>
+  return
+}
+// CHECK-LABEL:   func.func @vector_print_2d_scalable_vector(
+// CHECK-SAME:                                               %[[SCALABLE_VEC_2D:.*]]: vector<[4]x[4]xf32>) {
+// CHECK:           %[[C0:.*]] = arith.constant 0 : index
+// CHECK:           %[[C4:.*]] = arith.constant 4 : index
+// CHECK:           %[[C1:.*]] = arith.constant 1 : index
+// CHECK:           %[[ALLOCA_VEC:.*]] = memref.alloca() : memref<vector<[4]x[4]xf32>>
+// CHECK:           %[[VSCALE_0:.*]] = vector.vscale
+// CHECK:           %[[UPPER_BOUND_0:.*]] = arith.muli %[[VSCALE_0]], %[[C4]] : index
+// CHECK:           %[[VSCALE_1:.*]] = vector.vscale
+// CHECK:           %[[UPPER_BOUND_1:.*]] = arith.muli %[[VSCALE_1]], %[[C4]] : index
+// CHECK:           %[[ALLOCA:.*]] = memref.alloca(%[[UPPER_BOUND_0]], %[[UPPER_BOUND_1]]) : memref<?x?xf32>
+// CHECK:           memref.store %[[SCALABLE_VEC_2D]], %[[ALLOCA_VEC]][] : memref<vector<[4]x[4]xf32>>
+// CHECK:           %[[VAL_10:.*]] = vector.type_cast %[[ALLOCA_VEC]] : memref<vector<[4]x[4]xf32>> to memref<4xvector<4xf32>>
+// CHECK:           scf.for %[[II:.*]] = %[[C0]] to %[[C4]] step %[[C1]] {
+// CHECK:             %[[NOT_END:.*]] = arith.cmpi sgt, %[[UPPER_BOUND_0]], %[[II]] : index
+// CHECK:             scf.if %[[NOT_END]] {
+// CHECK:               %[[VECTOR_SLICE:.*]] = memref.load %[[VAL_10]]{{\[}}%[[II]]] : memref<4xvector<4xf32>>
+// CHECK:               vector.transfer_write %[[VECTOR_SLICE]], %[[ALLOCA]]{{\[}}%[[II]], %[[C0]]] : vector<4xf32>, memref<?x?xf32>
+// CHECK:             } else {
+// CHECK:             }
+// CHECK:           }
+// CHECK:           %[[VSCALE_2:.*]] = vector.vscale
+// CHECK:           %[[UPPER_BOUND_3:.*]] = arith.muli %[[VSCALE_2]], %[[C4]] : index
+// CHECK:           %[[LAST_INDEX_0:.*]] = arith.subi %[[UPPER_BOUND_3]], %[[C1]] : index
+// CHECK:           vector.print <open>
+// CHECK:           scf.for %[[OUTER_IDX:.*]] = %[[C0]] to %[[UPPER_BOUND_3]] step %[[C1]] {
+// CHECK:             %[[VSCALE_3:.*]] = vector.vscale
+// CHECK:             %[[UPPER_BOUND_4:.*]] = arith.muli %[[VSCALE_3]], %[[C4]] : index
+// CHECK:             %[[LAST_INDEX_1:.*]] = arith.subi %[[UPPER_BOUND_4]], %[[C1]] : index
+// CHECK:             vector.print <open>
+// CHECK:             scf.for %[[INNER_IDX:.*]] = %[[C0]] to %[[UPPER_BOUND_4]] step %[[C1]] {
+// CHECK:               %[[EL:.*]] = memref.load %[[ALLOCA]]{{\[}}%[[OUTER_IDX]], %[[INNER_IDX]]] : memref<?x?xf32>
+// CHECK:               vector.print %[[EL]] : f32 <no_punctuation>
+// CHECK:               %[[IS_NOT_LAST_0:.*]] = arith.cmpi ult, %[[INNER_IDX]], %[[LAST_INDEX_1]] : index
+// CHECK:               scf.if %[[IS_NOT_LAST_0]] {
+// CHECK:                 vector.print <comma>
+// CHECK:               }
+// CHECK:             }
+// CHECK:             vector.print <close>
+// CHECK:             %[[IS_NOT_LAST_1:.*]] = arith.cmpi ult, %[[OUTER_IDX]], %[[LAST_INDEX_0]] : index
+// CHECK:             scf.if %[[IS_NOT_LAST_1]] {
+// CHECK:               vector.print <comma>
+// CHECK:             }
+// CHECK:           }
+// CHECK:           vector.print <close>
+// CHECK:           vector.print
+// CHECK:           return
+// CHECK:         }

mlir/test/Integration/Dialect/Arith/CPU/test-wide-int-emulation-compare-results-i16.mlir

 // Check that the wide integer emulation produces the same result as wide
 // calculations. Emulate i16 ops with i8 ops.
 
 // RUN: mlir-opt %s --test-arith-emulate-wide-int="widest-int-supported=8" \
-// RUN:             --convert-scf-to-cf --convert-cf-to-llvm --convert-vector-to-llvm \
-// RUN:             --convert-func-to-llvm --convert-arith-to-llvm | \
+// RUN:             --convert-vector-to-scf --convert-scf-to-cf --convert-cf-to-llvm \
+// RUN:             --convert-vector-to-llvm --convert-func-to-llvm --convert-arith-to-llvm \
+// RUN:             --reconcile-unrealized-casts | \
 // RUN:   mlir-cpu-runner -e entry -entry-point-result=void \
 // RUN:      --shared-libs="%mlir_c_runner_utils,%mlir_runner_utils" | \
 // RUN:   FileCheck %s
 
 // CHECK-NOT: Mismatch
 
 //===----------------------------------------------------------------------===//
 // Common Utility Functions

mlir/test/Integration/Dialect/Arith/CPU/test-wide-int-emulation-constants-i16.mlir

 // Check that the wide integer constant emulation produces the same result as wide
 // constants and that printing works. Emulate i16 ops with i8 ops.
 
 // RUN: mlir-opt %s --test-arith-emulate-wide-int="widest-int-supported=8" \
-// RUN:             --convert-scf-to-cf --convert-cf-to-llvm --convert-vector-to-llvm \
-// RUN:             --convert-func-to-llvm --convert-arith-to-llvm | \
+// RUN:             --convert-vector-to-scf --convert-scf-to-cf --convert-cf-to-llvm --convert-vector-to-llvm \
+// RUN:             --convert-func-to-llvm --convert-arith-to-llvm --reconcile-unrealized-casts | \
 // RUN:   mlir-cpu-runner -e entry -entry-point-result=void \
 // RUN:                   --shared-libs=%mlir_c_runner_utils | \
 // RUN:   FileCheck %s --match-full-lines --check-prefix=EMULATED
 
 func.func @entry() {
   %cst0 = arith.constant 0 : i16
   func.call @emulate_constant(%cst0) : (i16) -> ()
   func.call @foo(%cst0) : (i16) -> ()

mlir/test/Integration/Dialect/LLVMIR/CPU/X86/test-inline-asm-vector.mlir

-// RUN: mlir-opt %s -convert-vector-to-llvm |  \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts |  \
 // RUN: mlir-cpu-runner -e entry_point_with_all_constants -entry-point-result=void \
 // RUN:   -shared-libs=%mlir_c_runner_utils
 
 module {
-  llvm.func @function_to_run(%a: vector<8xf32>, %b: vector<8xf32>)  {
+  func.func @function_to_run(%a: vector<8xf32>, %b: vector<8xf32>)  {
     // CHECK: ( 8, 10, 12, 14, 16, 18, 20, 22 )
     %r0 = llvm.inline_asm asm_dialect = intel
         "vaddps $0, $1, $2", "=x,x,x" %a, %b:
       (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
     vector.print %r0: vector<8xf32>
 
     // vblendps implemented with inline_asm.
     // CHECK: ( 0, 1, 10, 11, 4, 5, 14, 15 )
func.func @function_to_run(%a: vector<8xf32>, %b: vector<8xf32>)  {
     vector.print %r3: vector<8xf32>
 
     // vblendps 0x33 via vector.shuffle (emulates clang intrinsics impl)
     // CHECK: ( 8, 9, 2, 3, 12, 13, 6, 7 )
     %r4 = vector.shuffle %a, %b[8, 9, 2, 3, 12, 13, 6, 7]
       : vector<8xf32>, vector<8xf32>
     vector.print %r4: vector<8xf32>
 
-    llvm.return
+    return
   }
 
   // Solely exists to prevent inlining and get the expected assembly.
-  llvm.func @entry_point(%a: vector<8xf32>, %b: vector<8xf32>)  {
-    llvm.call @function_to_run(%a, %b) : (vector<8xf32>, vector<8xf32>) -> ()
-    llvm.return
+  func.func @entry_point(%a: vector<8xf32>, %b: vector<8xf32>)  {
+    func.call @function_to_run(%a, %b) : (vector<8xf32>, vector<8xf32>) -> ()
+    return
   }
 
-  llvm.func @entry_point_with_all_constants()  {
+  func.func @entry_point_with_all_constants()  {
     %a = llvm.mlir.constant(dense<[0.0, 1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0]>
       : vector<8xf32>) : vector<8xf32>
     %b = llvm.mlir.constant(dense<[8.0, 9.0, 10.0, 11.0, 12.0, 13.0, 14.0, 15.0]>
       : vector<8xf32>) : vector<8xf32>
-    llvm.call @function_to_run(%a, %b) : (vector<8xf32>, vector<8xf32>) -> ()
-    llvm.return
+    func.call @function_to_run(%a, %b) : (vector<8xf32>, vector<8xf32>) -> ()
+    return
   }
 }

mlir/test/Integration/Dialect/LLVMIR/CPU/test-vp-intrinsic.mlir

-// RUN: mlir-opt %s -convert-vector-to-llvm -finalize-memref-to-llvm \
-// RUN:             -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-cf-to-llvm \
+// RUN: -convert-vector-to-llvm -convert-index-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm \
+// RUN: -reconcile-unrealized-casts | \
 // RUN: mlir-translate -mlir-to-llvmir | \
 // RUN: %lli --entry-function=entry \
 // RUN:      --dlopen=%mlir_native_utils_lib_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
 
 // %mlir_native_utils_lib_dir is incorrect on Windows
 // UNSUPPORTED: system-windows
 

mlir/test/Integration/Dialect/Vector/CPU/ArmSVE/test-sve.mlir

-// RUN: mlir-opt %s -lower-affine -convert-scf-to-cf -convert-vector-to-llvm="enable-arm-sve" -finalize-memref-to-llvm -convert-func-to-llvm -convert-arith-to-llvm -canonicalize | \
+// RUN: mlir-opt %s -lower-affine -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm="enable-arm-sve" -finalize-memref-to-llvm -convert-func-to-llvm -convert-arith-to-llvm -canonicalize | \
 // RUN: %mcr_aarch64_cmd -e=entry -entry-point-result=void --march=aarch64 --mattr="+sve" -shared-libs=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
 
 // Note: To run this test, your CPU must support SVE
 
 // VLA memcopy
 func.func @kernel_copy(%src : memref<?xi64>, %dst : memref<?xi64>, %size : index) {
   %c0 = arith.constant 0 : index

mlir/test/Integration/Dialect/Vector/CPU/X86Vector/test-dot.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm="enable-x86vector" -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm="enable-x86vector" -convert-func-to-llvm -reconcile-unrealized-casts | \
 // RUN: mlir-translate --mlir-to-llvmir | \
 // RUN: %lli --entry-function=entry --mattr="avx" --dlopen=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() -> i32 {
   %i0 = arith.constant 0 : i32
   %i4 = arith.constant 4 : i32
 

mlir/test/Integration/Dialect/Vector/CPU/X86Vector/test-mask-compress.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm="enable-x86vector" -convert-func-to-llvm -reconcile-unrealized-casts  | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm="enable-x86vector" -convert-func-to-llvm -reconcile-unrealized-casts  | \
 // RUN: mlir-translate  --mlir-to-llvmir | \
 // RUN: %lli --entry-function=entry --mattr="avx512bw" --dlopen=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() -> i32 {
   %i0 = arith.constant 0 : i32
 
   %a = arith.constant dense<[1., 0., 0., 2., 4., 3., 5., 7., 8., 1., 5., 5., 3., 1., 0., 7.]> : vector<16xf32>

mlir/test/Integration/Dialect/Vector/CPU/X86Vector/test-rsqrt.mlir

-// RUN: mlir-opt %s -convert-vector-to-llvm="enable-x86vector" -convert-func-to-llvm | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm="enable-x86vector" -convert-func-to-llvm -reconcile-unrealized-casts  | \
 // RUN: mlir-translate --mlir-to-llvmir | \
 // RUN: %lli --entry-function=entry --mattr="avx" --dlopen=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() -> i32 {
   %i0 = arith.constant 0 : i32
 
   %v = arith.constant dense<[0.125, 0.25, 0.5, 1.0, 2.0, 4.0, 8.0, 16.0]> : vector<8xf32>
   %r = x86vector.avx.rsqrt %v : vector<8xf32>
   // `rsqrt` may produce slightly different results on Intel and AMD machines: accept both results here.
   // CHECK: {{( 2.82[0-9]*, 1.99[0-9]*, 1.41[0-9]*, 0.99[0-9]*, 0.70[0-9]*, 0.49[0-9]*, 0.35[0-9]*, 0.24[0-9]* )}}
   vector.print %r : vector<8xf32>
 
   return %i0 : i32
 }

mlir/test/Integration/Dialect/Vector/CPU/X86Vector/test-vp2intersect-i32.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm="enable-x86vector" -convert-func-to-llvm -reconcile-unrealized-casts  | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm="enable-x86vector" -convert-func-to-llvm -reconcile-unrealized-casts  | \
 // RUN: mlir-translate  --mlir-to-llvmir | \
 // RUN: %lli --entry-function=entry --mattr="avx512bw,avx512vp2intersect" --dlopen=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 // Note: To run this test, your CPU must support AVX512 vp2intersect.
 
 func.func @entry() -> i32 {
   %i0 = arith.constant 0 : i32

mlir/test/Integration/Dialect/Vector/CPU/test-0-d-vectors.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -test-lower-to-llvm | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @extract_element_0d(%a: vector<f32>) {
   %1 = vector.extractelement %a[] : vector<f32>
   // CHECK: 42
   vector.print %1: f32

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

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -test-lower-to-llvm | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   %i = arith.constant 2147483647: i32
   %l = arith.constant 9223372036854775807 : i64
 

mlir/test/Integration/Dialect/Vector/CPU/test-compress.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -test-lower-to-llvm | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @compress16(%base: memref<?xf32>,
                  %mask: vector<16xi1>, %value: vector<16xf32>) {
   %c0 = arith.constant 0: index
   vector.compressstore %base[%c0], %mask, %value

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

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -test-lower-to-llvm | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   %0 = vector.constant_mask [4] : vector<8xi1>
   vector.print %0 : vector<8xi1>
   // CHECK: ( 1, 1, 1, 1, 0, 0, 0, 0 )

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

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -test-lower-to-llvm | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 #dotp_accesses = [
   affine_map<(i) -> (i)>,
   affine_map<(i) -> (i)>,
   affine_map<(i) -> ()>

mlir/test/Integration/Dialect/Vector/CPU/test-create-mask-v4i1.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -test-lower-to-llvm | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 // NOTE: This is similar to test-create-mask.mlir, but with a different length,
 //       because the v4i1 vector specifically exposed bugs in the LLVM backend.
 
 func.func @entry() {

mlir/test/Integration/Dialect/Vector/CPU/test-create-mask.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -test-lower-to-llvm | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   %cneg1 = arith.constant -1 : index
   %c0 = arith.constant 0 : index
   %c1 = arith.constant 1 : index

mlir/test/Integration/Dialect/Vector/CPU/test-expand.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @expand16(%base: memref<?xf32>,
                %mask: vector<16xi1>,
                %pass_thru: vector<16xf32>) -> vector<16xf32> {
   %c0 = arith.constant 0: index

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

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -test-lower-to-llvm | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   %f0 = arith.constant 0.0: f32
   %f1 = arith.constant 1.0: f32
   %f2 = arith.constant 2.0: f32

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

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -O0 -enable-matrix -matrix-allow-contract -matrix-default-layout=column-major \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   %f0 = arith.constant 0.0: f64
   %f1 = arith.constant 1.0: f64

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

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -O0 -enable-matrix -matrix-allow-contract -matrix-default-layout=row-major \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   %f0 = arith.constant 0.0: f64
   %f1 = arith.constant 1.0: f64

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

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   %f1 = arith.constant 1.0: f32
   %f3 = arith.constant 3.0: f32
   %f7 = arith.constant 7.0: f32

mlir/test/Integration/Dialect/Vector/CPU/test-gather.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @gather8(%base: memref<?xf32>, %indices: vector<8xi32>,
               %mask: vector<8xi1>, %pass_thru: vector<8xf32>) -> vector<8xf32> {
   %c0 = arith.constant 0: index
   %g = vector.gather %base[%c0][%indices], %mask, %pass_thru

mlir/test/Integration/Dialect/Vector/CPU/test-index-vectors.mlir

-// RUN: mlir-opt %s -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -test-lower-to-llvm | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   %c0 = arith.constant dense<[0, 1, 2, 3]>: vector<4xindex>
   %c1 = arith.constant dense<[0, 1]>: vector<2xindex>
   %c2 = arith.constant 2 : index

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

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -test-lower-to-llvm | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   %f1 = arith.constant 1.0: f32
   %f2 = arith.constant 2.0: f32
   %f3 = arith.constant 3.0: f32

mlir/test/Integration/Dialect/Vector/CPU/test-maskedload.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @maskedload16(%base: memref<?xf32>, %mask: vector<16xi1>,
                    %pass_thru: vector<16xf32>) -> vector<16xf32> {
   %c0 = arith.constant 0: index
   %ld = vector.maskedload %base[%c0], %mask, %pass_thru

mlir/test/Integration/Dialect/Vector/CPU/test-maskedstore.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @maskedstore16(%base: memref<?xf32>,
                     %mask: vector<16xi1>, %value: vector<16xf32>) {
   %c0 = arith.constant 0: index
   vector.maskedstore %base[%c0], %mask, %value

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

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -O0 -enable-matrix -matrix-allow-contract -matrix-default-layout=column-major \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   %f0 = arith.constant 0.0: f64
   %f1 = arith.constant 1.0: f64

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

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -O0 -enable-matrix -matrix-allow-contract -matrix-default-layout=row-major \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   %f0 = arith.constant 0.0: f64
   %f1 = arith.constant 1.0: f64

mlir/test/Integration/Dialect/Vector/CPU/test-outerproduct-f32.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -test-lower-to-llvm | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 !vector_type_A = vector<8xf32>
 !vector_type_B = vector<8xf32>
 !vector_type_C = vector<8x8xf32>
 

mlir/test/Integration/Dialect/Vector/CPU/test-outerproduct-i64.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -test-lower-to-llvm | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 !vector_type_A = vector<8xi64>
 !vector_type_B = vector<8xi64>
 !vector_type_C = vector<8x8xi64>
 

mlir/test/Integration/Dialect/Vector/CPU/test-print-fp.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 //
 // Test various floating-point types.
 //
 func.func @entry() {

mlir/test/Integration/Dialect/Vector/CPU/test-print-int.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 //
 // Test various signless, signed, unsigned integer types.
 //
 func.func @entry() {

mlir/test/Integration/Dialect/Vector/CPU/test-realloc.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts |\
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts |\
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void \
 // RUN:   -shared-libs=%mlir_c_runner_utils
-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -finalize-memref-to-llvm='use-aligned-alloc=1' -convert-func-to-llvm -arith-expand -reconcile-unrealized-casts |\
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -finalize-memref-to-llvm='use-aligned-alloc=1' -convert-func-to-llvm -arith-expand -reconcile-unrealized-casts |\
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void \
 // RUN:   -shared-libs=%mlir_c_runner_utils | FileCheck %s
 
 // FIXME: Windows does not have aligned_alloc
 // UNSUPPORTED: system-windows
 
 func.func @entry() {
   // Set up memory.

mlir/test/Integration/Dialect/Vector/CPU/test-reductions-f32-reassoc.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf \
 // RUN:             -convert-vector-to-llvm='reassociate-fp-reductions' \
 // RUN:             -convert-func-to-llvm -reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   // Construct test vector, numerically very stable.

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

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   // Construct test vector.
   %f1 = arith.constant 1.5: f32
   %f2 = arith.constant 2.0: f32

mlir/test/Integration/Dialect/Vector/CPU/test-reductions-f64-reassoc.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf \
 // RUN:             -convert-vector-to-llvm='reassociate-fp-reductions' \
 // RUN:             -convert-func-to-llvm -reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   // Construct test vector, numerically very stable.

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

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   // Construct test vector.
   %f1 = arith.constant 1.5: f64
   %f2 = arith.constant 2.0: f64

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

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   // Construct test vector.
   %i1 = arith.constant 1: i32
   %i2 = arith.constant 2: i32

mlir/test/Integration/Dialect/Vector/CPU/test-reductions-i4.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   %v = arith.constant dense<[-8, -7, -6, -5, -4, -3, -2, -1, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]> : vector<24xi4>
   vector.print %v : vector<24xi4>
   //

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

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   // Construct test vector.
   %i1 = arith.constant 1: i64
   %i2 = arith.constant 2: i64

mlir/test/Integration/Dialect/Vector/CPU/test-reductions-si4.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   %v0 = arith.constant dense<[-8, -7, -6, -5, -4, -3, -2, -1, 0, 1, 2, 3, 4, 5, 6, 7]> : vector<16xi4>
   %v = vector.bitcast %v0 : vector<16xi4> to vector<16xsi4>
   vector.print %v : vector<16xsi4>

mlir/test/Integration/Dialect/Vector/CPU/test-reductions-ui4.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   %v0 = arith.constant dense<[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]> : vector<16xi4>
   %v = vector.bitcast %v0 : vector<16xi4> to vector<16xui4>
   vector.print %v : vector<16xui4>

mlir/test/Integration/Dialect/Vector/CPU/test-scan.mlir

-// RUN: mlir-opt %s -test-vector-scan-lowering -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -test-vector-scan-lowering -test-lower-to-llvm| \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   %f1 = arith.constant 1.0: f32
   %f2 = arith.constant 2.0: f32
   %f3 = arith.constant 3.0: f32

mlir/test/Integration/Dialect/Vector/CPU/test-scatter.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @scatter8(%base: memref<?xf32>,
                %indices: vector<8xi32>,
                %mask: vector<8xi1>, %value: vector<8xf32>) {
   %c0 = arith.constant 0: index

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

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -test-lower-to-llvm | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   %f1 = arith.constant 1.0: f32
   %f2 = arith.constant 2.0: f32
   %f3 = arith.constant 3.0: f32

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

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -test-lower-to-llvm | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   %f1 = arith.constant 1.0: f32
   %f2 = arith.constant 2.0: f32
   %v1 = vector.broadcast %f1 : f32 to vector<2x4xf32>

mlir/test/Integration/Dialect/Vector/CPU/test-shuffle16x16.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf \
+// RUN: mlir-opt %s  \
 // RUN:   -test-transform-dialect-interpreter \
-// RUN:   -test-transform-dialect-erase-schedule \
-// RUN:   -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN:   -test-transform-dialect-erase-schedule -test-lower-to-llvm | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   %in = arith.constant dense<[[0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0, 13.0, 14.0, 15.0], [16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 25.0, 26.0, 27.0, 28.0, 29.0, 30.0, 31.0], [32.0, 33.0, 34.0, 35.0, 36.0, 37.0, 38.0, 39.0, 40.0, 41.0, 42.0, 43.0, 44.0, 45.0, 46.0, 47.0], [48.0, 49.0, 50.0, 51.0, 52.0, 53.0, 54.0, 55.0, 56.0, 57.0, 58.0, 59.0, 60.0, 61.0, 62.0, 63.0], [64.0, 65.0, 66.0, 67.0, 68.0, 69.0, 70.0, 71.0, 72.0, 73.0, 74.0, 75.0, 76.0, 77.0, 78.0, 79.0], [80.0, 81.0, 82.0, 83.0, 84.0, 85.0, 86.0, 87.0, 88.0, 89.0, 90.0, 91.0, 92.0, 93.0, 94.0, 95.0], [96.0, 97.0, 98.0, 99.0, 100.0, 101.0, 102.0, 103.0, 104.0, 105.0, 106.0, 107.0, 108.0, 109.0, 110.0, 111.0], [112.0, 113.0, 114.0, 115.0, 116.0, 117.0, 118.0, 119.0, 120.0, 121.0, 122.0, 123.0, 124.0, 125.0, 126.0, 127.0], [128.0, 129.0, 130.0, 131.0, 132.0, 133.0, 134.0, 135.0, 136.0, 137.0, 138.0, 139.0, 140.0, 141.0, 142.0, 143.0], [144.0, 145.0, 146.0, 147.0, 148.0, 149.0, 150.0, 151.0, 152.0, 153.0, 154.0, 155.0, 156.0, 157.0, 158.0, 159.0], [160.0, 161.0, 162.0, 163.0, 164.0, 165.0, 166.0, 167.0, 168.0, 169.0, 170.0, 171.0, 172.0, 173.0, 174.0, 175.0], [176.0, 177.0, 178.0, 179.0, 180.0, 181.0, 182.0, 183.0, 184.0, 185.0, 186.0, 187.0, 188.0, 189.0, 190.0, 191.0], [192.0, 193.0, 194.0, 195.0, 196.0, 197.0, 198.0, 199.0, 200.0, 201.0, 202.0, 203.0, 204.0, 205.0, 206.0, 207.0], [208.0, 209.0, 210.0, 211.0, 212.0, 213.0, 214.0, 215.0, 216.0, 217.0, 218.0, 219.0, 220.0, 221.0, 222.0, 223.0], [224.0, 225.0, 226.0, 227.0, 228.0, 229.0, 230.0, 231.0, 232.0, 233.0, 234.0, 235.0, 236.0, 237.0, 238.0, 239.0], [240.0, 241.0, 242.0, 243.0, 244.0, 245.0, 246.0, 247.0, 248.0, 249.0, 250.0, 251.0, 252.0, 253.0, 254.0, 255.0]]> : vector<16x16xf32>
   %0 = vector.transpose %in, [1, 0] : vector<16x16xf32> to vector<16x16xf32>
   vector.print %0 : vector<16x16xf32>

mlir/test/Integration/Dialect/Vector/CPU/test-sparse-dot-matvec.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 // Illustrates an 8x8 Sparse Matrix x Vector implemented with only operations
 // of the vector dialect (and some std/scf). Essentially, this example performs
 // the following multiplication:
 //

mlir/test/Integration/Dialect/Vector/CPU/test-sparse-saxpy-jagged-matvec.mlir

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -finalize-memref-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 // Illustrates an 8x8 Sparse Matrix x Vector implemented with only operations
 // of the vector dialect (and some std/scf). Essentially, this example performs
 // the following multiplication:
 //

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

-// RUN: mlir-opt %s -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
+// RUN: mlir-opt %s -test-lower-to-llvm | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_c_runner_utils | \
 // RUN: FileCheck %s
 
 func.func @entry() {
   %f0 = arith.constant 0.0: f32
   %f1 = arith.constant 1.0: f32
   %f2 = arith.constant 2.0: f32

mlir/test/Integration/Dialect/Vector/GPU/CUDA/test-reduction-distribute.mlir

 // RUN: mlir-opt %s -test-vector-warp-distribute="hoist-uniform distribute-transfer-write propagate-distribution" -canonicalize |\
 // RUN: mlir-opt -test-vector-warp-distribute=rewrite-warp-ops-to-scf-if |\
-// RUN: mlir-opt  -lower-affine -convert-scf-to-cf -convert-vector-to-llvm \
+// RUN: mlir-opt -lower-affine -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm \
 // RUN:  -convert-arith-to-llvm -gpu-kernel-outlining |\
 // RUN: mlir-opt -pass-pipeline='builtin.module(gpu.module(strip-debuginfo,convert-gpu-to-nvvm,reconcile-unrealized-casts,gpu-to-cubin))' |\
 // RUN: mlir-opt -gpu-to-llvm -reconcile-unrealized-casts |\
 // RUN: mlir-cpu-runner -e main -entry-point-result=void \
 // RUN:   -shared-libs=%mlir_cuda_runtime \
 // RUN:   -shared-libs=%mlir_c_runner_utils \
 // RUN:   -shared-libs=%mlir_runner_utils | \
 // RUN: FileCheck %s

mlir/test/Integration/Dialect/Vector/GPU/CUDA/test-warp-distribute.mlir

 // Run the test cases without distributing ops to test default lowering. Run
 // everything on the same thread.
 // RUN: mlir-opt %s -test-vector-warp-distribute=rewrite-warp-ops-to-scf-if -canonicalize | \
-// RUN: mlir-opt -convert-scf-to-cf -convert-cf-to-llvm -convert-vector-to-llvm -convert-arith-to-llvm \
+// RUN: mlir-opt -convert-vector-to-scf -convert-scf-to-cf -convert-cf-to-llvm -convert-vector-to-llvm -convert-arith-to-llvm \
 // RUN:  -gpu-kernel-outlining |\
 // RUN: mlir-opt -pass-pipeline='builtin.module(gpu.module(strip-debuginfo,convert-gpu-to-nvvm,reconcile-unrealized-casts,gpu-to-cubin))' |\
 // RUN: mlir-opt -gpu-to-llvm -reconcile-unrealized-casts |\
 // RUN: mlir-cpu-runner -e main -entry-point-result=void \
 // RUN:   -shared-libs=%mlir_cuda_runtime \
 // RUN:   -shared-libs=%mlir_c_runner_utils \
 // RUN:   -shared-libs=%mlir_runner_utils | \
 // RUN: FileCheck %s
 
 // Run the same test cases with distribution and propagation.
 // RUN: mlir-opt %s  -test-vector-warp-distribute="hoist-uniform distribute-transfer-write" \
 // RUN:   -test-vector-warp-distribute=rewrite-warp-ops-to-scf-if -canonicalize | \
-// RUN: mlir-opt -convert-scf-to-cf -convert-cf-to-llvm -convert-vector-to-llvm -convert-arith-to-llvm \
+// RUN: mlir-opt -convert-vector-to-scf -convert-scf-to-cf -convert-cf-to-llvm -convert-vector-to-llvm -convert-arith-to-llvm \
 // RUN:  -gpu-kernel-outlining |\
 // RUN: mlir-opt -pass-pipeline='builtin.module(gpu.module(strip-debuginfo,convert-gpu-to-nvvm,reconcile-unrealized-casts,gpu-to-cubin))' |\
 // RUN: mlir-opt -gpu-to-llvm -reconcile-unrealized-casts |\
 // RUN: mlir-cpu-runner -e main -entry-point-result=void \
 // RUN:   -shared-libs=%mlir_cuda_runtime \
 // RUN:   -shared-libs=%mlir_c_runner_utils \
 // RUN:   -shared-libs=%mlir_runner_utils | \
 // RUN: FileCheck %s
 
 // RUN: mlir-opt %s  -test-vector-warp-distribute="hoist-uniform distribute-transfer-write propagate-distribution" \
 // RUN:   -test-vector-warp-distribute=rewrite-warp-ops-to-scf-if -canonicalize | \
-// RUN: mlir-opt -convert-scf-to-cf -convert-cf-to-llvm -convert-vector-to-llvm -convert-arith-to-llvm \
+// RUN: mlir-opt -convert-vector-to-scf -convert-scf-to-cf -convert-cf-to-llvm -convert-vector-to-llvm -convert-arith-to-llvm \
 // RUN:  -gpu-kernel-outlining |\
 // RUN: mlir-opt -pass-pipeline='builtin.module(gpu.module(strip-debuginfo,convert-gpu-to-nvvm,reconcile-unrealized-casts,gpu-to-cubin))' |\
 // RUN: mlir-opt -gpu-to-llvm -reconcile-unrealized-casts |\
 // RUN: mlir-cpu-runner -e main -entry-point-result=void \
 // RUN:   -shared-libs=%mlir_cuda_runtime \
 // RUN:   -shared-libs=%mlir_c_runner_utils \
 // RUN:   -shared-libs=%mlir_runner_utils | \
 // RUN: FileCheck %s

mlir/test/mlir-cpu-runner/math-polynomial-approx.mlir

-// RUN:   mlir-opt %s -pass-pipeline="builtin.module(func.func(test-math-polynomial-approximation,convert-arith-to-llvm),convert-vector-to-llvm,func.func(convert-math-to-llvm),convert-func-to-llvm,reconcile-unrealized-casts)" \
+// RUN:   mlir-opt %s -pass-pipeline="builtin.module(func.func(test-math-polynomial-approximation,convert-arith-to-llvm),convert-vector-to-scf,convert-scf-to-cf,convert-cf-to-llvm,convert-vector-to-llvm,func.func(convert-math-to-llvm),convert-func-to-llvm,reconcile-unrealized-casts)" \
 // RUN: | mlir-cpu-runner                                                      \
 // RUN:     -e main -entry-point-result=void -O0                               \
 // RUN:     -shared-libs=%mlir_c_runner_utils  \
 // RUN:     -shared-libs=%mlir_runner_utils    \
 // RUN: | FileCheck %s
 
 // -------------------------------------------------------------------------- //
 // Tanh.

mlir/test/mlir-cpu-runner/test-expand-math-approx.mlir

-// RUN:   mlir-opt %s -pass-pipeline="builtin.module(func.func(test-expand-math,convert-arith-to-llvm),convert-vector-to-llvm,func.func(convert-math-to-llvm),convert-func-to-llvm,reconcile-unrealized-casts)" \
+// RUN:   mlir-opt %s -pass-pipeline="builtin.module(func.func(test-expand-math,convert-arith-to-llvm),convert-vector-to-scf,convert-scf-to-cf,convert-cf-to-llvm,convert-vector-to-llvm,func.func(convert-math-to-llvm),convert-func-to-llvm,reconcile-unrealized-casts)" \
 // RUN: | mlir-cpu-runner                                                      \
 // RUN:     -e main -entry-point-result=void -O0                               \
 // RUN:     -shared-libs=%mlir_c_runner_utils  \
 // RUN:     -shared-libs=%mlir_runner_utils    \
 // RUN: | FileCheck %s
 
 // -------------------------------------------------------------------------- //
 // exp2f.