[MLIR] Update Vector To LLVM conversion to be aware of assume_alignment

[stephenneuendorffer]

Summary

vector.transfer_read and vector.transfer_write operations are converted
to llvm intrinsics with specific alignment information, however there
doesn't seem to be a way in llvm to take information from llvm.assume
intrinsics and change this alignment information. In any
event, due the to the structure of the llvm.assume instrinsic, applying
this information at the llvm level is more cumbersome. Instead, let's
generate the masked vector load and store instrinsic with the right
alignment information from MLIR in the first place. Since
we're bothering to do this, lets just emit the proper alignment for
loads, stores, scatter, and gather ops too.

[nicolasvasilache]

mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp

124

Can we fold this into a:

LogicalResult getMemRefAlignment(LLVMTypeConverter &typeConverter,
                                 Value memRef, unsigned &align) {
  MemRefType memrefType = memRef.getType().cast<MemRefType>();
  
  // old code

  // your code
}

?

131

align = std::max(align, op.alignment()) (with a cast if needed) ?

[aartbik]

mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp

124

+1 with probably a name to reflect that

getMemRefAlignmentAndUpdate()

or something like that

[stephenneuendorffer]

In incorporating your feedback, I decided to emit the proper alignment other loads and stores too. Thoughts?

[nicolasvasilache]

In D100444#2767792, @stephenneuendorffer wrote:

In incorporating your feedback, I decided to emit the proper alignment other loads and stores too. Thoughts?

Looks good!

Now I'm wondering whether we should go for the LICM instead?

mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp

124

If no assumed alignment

->

Return the minimal alignment value that satisfies all the AssumeAlignment uses of `value`.
If no such uses exist, return 0.

131

Thanks for re-spelling.
Now that I see it in this form, should this be licm instead of max ?

[nicolasvasilache]

And by licm I mean lcm :p ..

mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp

131

And by licm I mean lcm :p ..

[stephenneuendorffer]

mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp

131

Probably.. I guess I was implicitly assuming everything was a power of 2.

[stephenneuendorffer]

mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp

131

Sure enough : 'error: 'memref.assume_alignment' op alignment must be power of 2'

[nicolasvasilache]

thanks!

[ThomasRaoux]

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

1555

How can we know that this store is aligned on 32B without knowing anything about %i and %j.
I'm running into a bug because the memref alignment is propagated to load/store associated without considering the indices used for indexing.
I don't believe we can infer any kind of alignment without having alignment information about the indices.
@stephenneuendorffer, could you clarify how you expect this to work?

[nicolasvasilache]

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

1555

This one is clearly not aligned on 32B but it should be aligned on:

lcm(annotation + offset_alignment, most_minor_dim_alignment) == lcm(32 + 0 * 4, 100 * 4) == 16B

There seems indeed to be a small bug somewhere but I would prefer we fix rather than flatly revert.

[mehdi_amini]

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

1555

Why? Revert is conservative and cheap: it is trivial to reapply the patch with the fix without pressure.

[nicolasvasilache]

Fair enough, my bias was against losing the contribution; didn't think about it your way, thanks!

[ThomasRaoux]

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

1555

Yes Stephen reverted it as a quick solution as this was causing miscompile on IREE side.
@nicolasvasilache, I don't understand how we can prove that this is aligned on 16B. If %j = 1 and %i = 0 for instance we would be accessing 4B from the base and since the base is 32B aligned we would only be able to get 4B alignment.
What am I missing?

[nicolasvasilache]

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

1555

You're right, I was very focused on the memref type itself and did not look deeper at the indexing..

What I meant to say is that the alignment for each start of every row of memref is already degraded wrt the base pointer.

More precisely:

@%memref[0, 0] is aligned at 32B
@%memref[1, 0] is aligned at 16B but not 32B
@%memref[2, 0] is aligned at 32B
...

the indexing within the row can indeed be any 4B aligned.

But even worse, even @%memref[1, 32] is aligned at 16B and not 32B.

This is one of the important things to grasp when talking about memref and alignment: a memref is not a pointer.
An n-D memref is closer in principle to a:

(n-1)-D array of pointers to 1-D row

(even if they are not materialized)

Please ignore my earlier comment on the revert and sorry about the noise 😛

mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp

 #include "mlir/Conversion/StandardToLLVM/ConvertStandardToLLVM.h"
 #include "mlir/Conversion/StandardToLLVM/ConvertStandardToLLVMPass.h"
 #include "mlir/Dialect/LLVMIR/FunctionCallUtils.h"
 #include "mlir/Dialect/LLVMIR/LLVMDialect.h"
 #include "mlir/Dialect/MemRef/IR/MemRef.h"
 #include "mlir/Dialect/StandardOps/IR/Ops.h"
 #include "mlir/Dialect/Vector/VectorOps.h"
 #include "mlir/IR/BuiltinTypes.h"
+#include "mlir/Support/MathExtras.h"
 #include "mlir/Target/LLVMIR/TypeTranslation.h"
 #include "mlir/Transforms/DialectConversion.h"
 
 using namespace mlir;
 using namespace mlir::vector;
 
 // Helper to reduce vector type by one rank at front.
 static VectorType reducedVectorTypeFront(VectorType tp) {
LogicalResult getMemRefAlignment(LLVMTypeConverter &typeConverter,
   // TODO: this should use the MLIR data layout when it becomes available and
   // stop depending on translation.
   llvm::LLVMContext llvmContext;
   align = LLVM::TypeToLLVMIRTranslator(llvmContext)
               .getPreferredAlignment(elementTy, typeConverter.getDataLayout());
   return success();
 }
 
+// Return the minimal alignment value that satisfies all the AssumeAlignment
+// uses of `value`. If no such uses exist, return 1.
+static unsigned getAssumedAlignment(Value value) {
+  unsigned align = 1;
+  for (auto &u : value.getUses()) {
+    Operation *owner = u.getOwner();
+    if (auto op = dyn_cast<memref::AssumeAlignmentOp>(owner))
+      align = mlir::lcm(align, op.alignment());
+  }
+  return align;
+}
+// Helper that returns data layout alignment of a memref associated with a
+// transfer op, including additional information from assume_alignment calls
+// on the source of the transfer
+LogicalResult getTransferOpAlignment(LLVMTypeConverter &typeConverter,
+                                     VectorTransferOpInterface xfer,
+                                     unsigned &align) {
+  if (failed(getMemRefAlignment(
+          typeConverter, xfer.getShapedType().cast<MemRefType>(), align)))
+    return failure();
+  align = std::max(align, getAssumedAlignment(xfer.source()));
+  return success();
+}
+
+// Helper that returns data layout alignment of a memref associated with a
+// load, store, scatter, or gather op, including additional information from
+// assume_alignment calls on the source of the transfer
+template <class OpAdaptor>
+LogicalResult getMemRefOpAlignment(LLVMTypeConverter &typeConverter,
+                                   OpAdaptor op, unsigned &align) {
+  if (failed(getMemRefAlignment(typeConverter, op.getMemRefType(), align)))
+    return failure();
+  align = std::max(align, getAssumedAlignment(op.base()));
+  return success();
+}
+
 // Add an index vector component to a base pointer. This almost always succeeds
 // unless the last stride is non-unit or the memory space is not zero.
 static LogicalResult getIndexedPtrs(ConversionPatternRewriter &rewriter,
                                     Location loc, Value memref, Value base,
                                     Value index, MemRefType memRefType,
                                     VectorType vType, Value &ptrs) {
   int64_t offset;
   SmallVector<int64_t, 4> strides;

 }
 
 static LogicalResult
 replaceTransferOpWithLoadOrStore(ConversionPatternRewriter &rewriter,
                                  LLVMTypeConverter &typeConverter, Location loc,
                                  TransferReadOp xferOp,
                                  ArrayRef<Value> operands, Value dataPtr) {
   unsigned align;
-  if (failed(getMemRefAlignment(
-          typeConverter, xferOp.getShapedType().cast<MemRefType>(), align)))
+  if (failed(getTransferOpAlignment(typeConverter, xferOp, align)))
     return failure();
   rewriter.replaceOpWithNewOp<LLVM::LoadOp>(xferOp, dataPtr, align);
   return success();
 }
 
 static LogicalResult
 replaceTransferOpWithMasked(ConversionPatternRewriter &rewriter,
                             LLVMTypeConverter &typeConverter, Location loc,
                             TransferReadOp xferOp, ArrayRef<Value> operands,
                             Value dataPtr, Value mask) {
   Type vecTy = typeConverter.convertType(xferOp.getVectorType());
   if (!vecTy)
     return failure();
 
   auto adaptor = TransferReadOpAdaptor(operands, xferOp->getAttrDictionary());
   Value fill = rewriter.create<SplatOp>(loc, vecTy, adaptor.padding());
 
   unsigned align;
-  if (failed(getMemRefAlignment(
-          typeConverter, xferOp.getShapedType().cast<MemRefType>(), align)))
+  if (failed(getTransferOpAlignment(typeConverter, xferOp, align)))
     return failure();
-
   rewriter.replaceOpWithNewOp<LLVM::MaskedLoadOp>(
       xferOp, vecTy, dataPtr, mask, ValueRange{fill},
       rewriter.getI32IntegerAttr(align));
   return success();
 }
 
 static LogicalResult
 replaceTransferOpWithLoadOrStore(ConversionPatternRewriter &rewriter,
                                  LLVMTypeConverter &typeConverter, Location loc,
                                  TransferWriteOp xferOp,
                                  ArrayRef<Value> operands, Value dataPtr) {
   unsigned align;
-  if (failed(getMemRefAlignment(
-          typeConverter, xferOp.getShapedType().cast<MemRefType>(), align)))
+  if (failed(getTransferOpAlignment(typeConverter, xferOp, align)))
     return failure();
   auto adaptor = TransferWriteOpAdaptor(operands, xferOp->getAttrDictionary());
   rewriter.replaceOpWithNewOp<LLVM::StoreOp>(xferOp, adaptor.vector(), dataPtr,
                                              align);
   return success();
 }
 
 static LogicalResult
 replaceTransferOpWithMasked(ConversionPatternRewriter &rewriter,
                             LLVMTypeConverter &typeConverter, Location loc,
                             TransferWriteOp xferOp, ArrayRef<Value> operands,
                             Value dataPtr, Value mask) {
   unsigned align;
-  if (failed(getMemRefAlignment(
-          typeConverter, xferOp.getShapedType().cast<MemRefType>(), align)))
+  if (failed(getTransferOpAlignment(typeConverter, xferOp, align)))
     return failure();
 
   auto adaptor = TransferWriteOpAdaptor(operands, xferOp->getAttrDictionary());
   rewriter.replaceOpWithNewOp<LLVM::MaskedStoreOp>(
       xferOp, adaptor.vector(), dataPtr, mask,
       rewriter.getI32IntegerAttr(align));
   return success();
 }
if (vectorTy.getRank() > 1)
       return failure();
 
     auto loc = loadOrStoreOp->getLoc();
     auto adaptor = LoadOrStoreOpAdaptor(operands);
     MemRefType memRefTy = loadOrStoreOp.getMemRefType();
 
     // Resolve alignment.
     unsigned align;
-    if (failed(getMemRefAlignment(*this->getTypeConverter(), memRefTy, align)))
+    if (failed(getMemRefOpAlignment(*this->getTypeConverter(), loadOrStoreOp,
+                                    align)))
       return failure();
 
     // Resolve address.
     auto vtype = this->typeConverter->convertType(loadOrStoreOp.getVectorType())
                      .template cast<VectorType>();
     Value dataPtr = this->getStridedElementPtr(loc, memRefTy, adaptor.base(),
                                                adaptor.indices(), rewriter);
     Value ptr = castDataPtr(rewriter, loc, dataPtr, memRefTy, vtype);
public:
   matchAndRewrite(vector::GatherOp gather, ArrayRef<Value> operands,
                   ConversionPatternRewriter &rewriter) const override {
     auto loc = gather->getLoc();
     auto adaptor = vector::GatherOpAdaptor(operands);
     MemRefType memRefType = gather.getMemRefType();
 
     // Resolve alignment.
     unsigned align;
-    if (failed(getMemRefAlignment(*getTypeConverter(), memRefType, align)))
+    if (failed(getMemRefOpAlignment(*getTypeConverter(), gather, align)))
       return failure();
 
     // Resolve address.
     Value ptrs;
     VectorType vType = gather.getVectorType();
     Value ptr = getStridedElementPtr(loc, memRefType, adaptor.base(),
                                      adaptor.indices(), rewriter);
     if (failed(getIndexedPtrs(rewriter, loc, adaptor.base(), ptr,
public:
   matchAndRewrite(vector::ScatterOp scatter, ArrayRef<Value> operands,
                   ConversionPatternRewriter &rewriter) const override {
     auto loc = scatter->getLoc();
     auto adaptor = vector::ScatterOpAdaptor(operands);
     MemRefType memRefType = scatter.getMemRefType();
 
     // Resolve alignment.
     unsigned align;
-    if (failed(getMemRefAlignment(*getTypeConverter(), memRefType, align)))
+    if (failed(getMemRefOpAlignment(*getTypeConverter(), scatter, align)))
       return failure();
 
     // Resolve address.
     Value ptrs;
     VectorType vType = scatter.getVectorType();
     Value ptr = getStridedElementPtr(loc, memRefType, adaptor.base(),
                                      adaptor.indices(), rewriter);
     if (failed(getIndexedPtrs(rewriter, loc, adaptor.base(), ptr,

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

 //       CHECK: %[[loaded:.*]] = llvm.intr.masked.load %{{.*}}, %{{.*}}, %{{.*}} {alignment = 8 : i32} :
 //  CHECK-SAME: (!llvm.ptr<vector<17xi64>>, vector<17xi1>, vector<17xi64>) -> vector<17xi64>
 
 //       CHECK: llvm.intr.masked.store %[[loaded]], %{{.*}}, %{{.*}} {alignment = 8 : i32} :
 //  CHECK-SAME: vector<17xi64>, vector<17xi1> into !llvm.ptr<vector<17xi64>>
 
 // -----
 
+func @transfer_read_1d_aligned(%A : memref<?xf32>, %base: index) -> vector<17xf32> {
+  memref.assume_alignment %A, 32 : memref<?xf32>
+  %f7 = constant 7.0: f32
+  %f = vector.transfer_read %A[%base], %f7
+      {permutation_map = affine_map<(d0) -> (d0)>} :
+    memref<?xf32>, vector<17xf32>
+  vector.transfer_write %f, %A[%base]
+      {permutation_map = affine_map<(d0) -> (d0)>} :
+    vector<17xf32>, memref<?xf32>
+  return %f: vector<17xf32>
+}
+//       CHECK: llvm.intr.masked.load
+//  CHECK-SAME: {alignment = 32 : i32}
+//  CHECK-SAME: (!llvm.ptr<vector<17xf32>>, vector<17xi1>, vector<17xf32>) -> vector<17xf32>
+//       CHECK: llvm.intr.masked.store
+//  CHECK-SAME: {alignment = 32 : i32}
+//  CHECK-SAME: vector<17xf32>, vector<17xi1> into !llvm.ptr<vector<17xf32>>
+
+// -----
+
 func @transfer_read_2d_to_1d(%A : memref<?x?xf32>, %base0: index, %base1: index) -> vector<17xf32> {
   %f7 = constant 7.0: f32
   %f = vector.transfer_read %A[%base0, %base1], %f7
       {permutation_map = affine_map<(d0, d1) -> (d1)>} :
     memref<?x?xf32>, vector<17xf32>
   return %f: vector<17xf32>
 }
 // CHECK-LABEL: func @transfer_read_2d_to_1d

 }
 // CHECK-LABEL: func @vector_load_op_index
 // CHECK: %[[T0:.*]] = llvm.load %{{.*}} {alignment = 8 : i64} : !llvm.ptr<vector<8xi64>>
 // CHECK: %[[T1:.*]] = llvm.mlir.cast %[[T0]] : vector<8xi64> to vector<8xindex>
 // CHECK: return %[[T1]] : vector<8xindex>
 
 // -----
 
+func @vector_load_op_aligned(%memref : memref<200x100xf32>, %i : index, %j : index) -> vector<8xf32> {
+  memref.assume_alignment %memref, 32 : memref<200x100xf32>
+  %0 = vector.load %memref[%i, %j] : memref<200x100xf32>, vector<8xf32>
+  return %0 : vector<8xf32>
+}
+
+// CHECK-LABEL: func @vector_load_op_aligned
+// CHECK: %[[c100:.*]] = llvm.mlir.constant(100 : index) : i64
+// CHECK: %[[mul:.*]] = llvm.mul %{{.*}}, %[[c100]]  : i64
+// CHECK: %[[add:.*]] = llvm.add %[[mul]], %{{.*}}  : i64
+// CHECK: %[[gep:.*]] = llvm.getelementptr %{{.*}}[%[[add]]] : (!llvm.ptr<f32>, i64) -> !llvm.ptr<f32>
+// CHECK: %[[bcast:.*]] = llvm.bitcast %[[gep]] : !llvm.ptr<f32> to !llvm.ptr<vector<8xf32>>
+// CHECK: llvm.load %[[bcast]] {alignment = 32 : i64} : !llvm.ptr<vector<8xf32>>
+
+// -----
+
 func @vector_store_op(%memref : memref<200x100xf32>, %i : index, %j : index) {
   %val = constant dense<11.0> : vector<4xf32>
   vector.store %val, %memref[%i, %j] : memref<200x100xf32>, vector<4xf32>
   return
 }
 
 // CHECK-LABEL: func @vector_store_op
 // CHECK: %[[c100:.*]] = llvm.mlir.constant(100 : index) : i64
func @vector_store_op_index(%memref : memref<200x100xindex>, %i : index, %j : index) {
   vector.store %val, %memref[%i, %j] : memref<200x100xindex>, vector<4xindex>
   return
 }
 // CHECK-LABEL: func @vector_store_op_index
 // CHECK: llvm.store %{{.*}}, %{{.*}} {alignment = 8 : i64} : !llvm.ptr<vector<4xi64>>
 
 // -----
 
+func @vector_store_op_aligned(%memref : memref<200x100xf32>, %i : index, %j : index) {
+  memref.assume_alignment %memref, 32 : memref<200x100xf32>
+  %val = constant dense<11.0> : vector<4xf32>
+  vector.store %val, %memref[%i, %j] : memref<200x100xf32>, vector<4xf32>
+  return
+}
+
+// CHECK-LABEL: func @vector_store_op_aligned
+// CHECK: %[[c100:.*]] = llvm.mlir.constant(100 : index) : i64
+// CHECK: %[[mul:.*]] = llvm.mul %{{.*}}, %[[c100]]  : i64
+// CHECK: %[[add:.*]] = llvm.add %[[mul]], %{{.*}}  : i64
+// CHECK: %[[gep:.*]] = llvm.getelementptr %{{.*}}[%[[add]]] : (!llvm.ptr<f32>, i64) -> !llvm.ptr<f32>
+// CHECK: %[[bcast:.*]] = llvm.bitcast %[[gep]] : !llvm.ptr<f32> to !llvm.ptr<vector<4xf32>>
+// CHECK: llvm.store %{{.*}}, %[[bcast]] {alignment = 32 : i64} : !llvm.ptr<vector<4xf32>>
+
+// -----
+
 func @masked_load_op(%arg0: memref<?xf32>, %arg1: vector<16xi1>, %arg2: vector<16xf32>) -> vector<16xf32> {
   %c0 = constant 0: index
   %0 = vector.maskedload %arg0[%c0], %arg1, %arg2 : memref<?xf32>, vector<16xi1>, vector<16xf32> into vector<16xf32>
   return %0 : vector<16xf32>
 }
 
 // CHECK-LABEL: func @masked_load_op
 // CHECK: %[[CO:.*]] = constant 0 : index

 
 // CHECK-LABEL: func @gather_op_index
 // CHECK: %[[P:.*]] = llvm.getelementptr %{{.*}}[%{{.*}}] : (!llvm.ptr<i64>, vector<3xi64>) -> !llvm.vec<3 x ptr<i64>>
 // CHECK: %[[G:.*]] = llvm.intr.masked.gather %{{.*}}, %{{.*}}, %{{.*}} {alignment = 8 : i32} : (!llvm.vec<3 x ptr<i64>>, vector<3xi1>, vector<3xi64>) -> vector<3xi64>
 // CHECK: %{{.*}} = llvm.mlir.cast %[[G]] : vector<3xi64> to vector<3xindex>
 
 // -----
 
+func @gather_op_aligned(%arg0: memref<?xf32>, %arg1: vector<3xi32>, %arg2: vector<3xi1>, %arg3: vector<3xf32>) -> vector<3xf32> {
+  memref.assume_alignment %arg0, 32 : memref<?xf32>
+  %0 = constant 0: index
+  %1 = vector.gather %arg0[%0][%arg1], %arg2, %arg3 : memref<?xf32>, vector<3xi32>, vector<3xi1>, vector<3xf32> into vector<3xf32>
+  return %1 : vector<3xf32>
+}
+
+// CHECK-LABEL: func @gather_op_aligned
+// CHECK: %[[P:.*]] = llvm.getelementptr %{{.*}}[%{{.*}}] : (!llvm.ptr<f32>, vector<3xi32>) -> !llvm.vec<3 x ptr<f32>>
+// CHECK: %[[G:.*]] = llvm.intr.masked.gather %[[P]], %{{.*}}, %{{.*}} {alignment = 32 : i32} : (!llvm.vec<3 x ptr<f32>>, vector<3xi1>, vector<3xf32>) -> vector<3xf32>
+// CHECK: return %[[G]] : vector<3xf32>
+
+// -----
+
 func @gather_2d_op(%arg0: memref<4x4xf32>, %arg1: vector<4xi32>, %arg2: vector<4xi1>, %arg3: vector<4xf32>) -> vector<4xf32> {
   %0 = constant 3 : index
   %1 = vector.gather %arg0[%0, %0][%arg1], %arg2, %arg3 : memref<4x4xf32>, vector<4xi32>, vector<4xi1>, vector<4xf32> into vector<4xf32>
   return %1 : vector<4xf32>
 }
 
 // CHECK-LABEL: func @gather_2d_op
 // CHECK: %[[B:.*]] = llvm.getelementptr %{{.*}}[%{{.*}}] : (!llvm.ptr<f32>, i64) -> !llvm.ptr<f32>

 }
 
 // CHECK-LABEL: func @scatter_op_index
 // CHECK: %[[P:.*]] = llvm.getelementptr %{{.*}}[%{{.*}}] : (!llvm.ptr<i64>, vector<3xi64>) -> !llvm.vec<3 x ptr<i64>>
 // CHECK: llvm.intr.masked.scatter %{{.*}}, %[[P]], %{{.*}} {alignment = 8 : i32} : vector<3xi64>, vector<3xi1> into !llvm.vec<3 x ptr<i64>>
 
 // -----
 
+func @scatter_op_aligned(%arg0: memref<?xf32>, %arg1: vector<3xi32>, %arg2: vector<3xi1>, %arg3: vector<3xf32>) {
+  memref.assume_alignment %arg0, 32 : memref<?xf32>
+  %0 = constant 0: index
+  vector.scatter %arg0[%0][%arg1], %arg2, %arg3 : memref<?xf32>, vector<3xi32>, vector<3xi1>, vector<3xf32>
+  return
+}
+
+// CHECK-LABEL: func @scatter_op_aligned
+// CHECK: %[[P:.*]] = llvm.getelementptr %{{.*}}[%{{.*}}] : (!llvm.ptr<f32>, vector<3xi32>) -> !llvm.vec<3 x ptr<f32>>
+// CHECK: llvm.intr.masked.scatter %{{.*}}, %[[P]], %{{.*}} {alignment = 32 : i32} : vector<3xf32>, vector<3xi1> into !llvm.vec<3 x ptr<f32>>
+
+// -----
+
 func @scatter_2d_op(%arg0: memref<4x4xf32>, %arg1: vector<4xi32>, %arg2: vector<4xi1>, %arg3: vector<4xf32>) {
   %0 = constant 3 : index
   vector.scatter %arg0[%0, %0][%arg1], %arg2, %arg3 : memref<4x4xf32>, vector<4xi32>, vector<4xi1>, vector<4xf32>
   return
 }
 
 // CHECK-LABEL: func @scatter_2d_op
 // CHECK: %[[B:.*]] = llvm.getelementptr %{{.*}}[%{{.*}}] : (!llvm.ptr<f32>, i64) -> !llvm.ptr<f32>