[mlir] Start splitting the `tensor` dialect out of `std`.

[silvas]

Summary

This starts by moving std.extract_element to tensor.extract (this
mirrors the naming of vector.extract).

Curiously, std.extract_element supposedly works on vectors as well,
and this patch removes that functionality. I would tend to do that in
separate patch, but I couldn't find any downstream users relying on
this, and the fact that we have vector.extract made it seem safe
enough to lump in here.

This also sets up the tensor dialect as a dependency of the std
dialect, as some ops that currently live in std depend on
tensor.extract via their canonicalization patterns.

Part of RFC: https://llvm.discourse.group/t/rfc-split-the-tensor-dialect-from-std/2347/2

[bondhugula]

Nothing major, but I tend to prefer tensor.get_element to tensor.extract for two reasons: (1) extract in standard data structures/ADT literature also involves removing the element out of the structure (eg. extract_min/extract_max for heaps, priority queues), (2) extract is more suitable for containers but this is an immutable entity.

[nicolasvasilache]

Thanks!

I'm -1 on the proposed name change: LLVM has {insert|extract}{element|value}, op names are globally consistent with that terminology.

[bondhugula]

In D92991#2444781, @nicolasvasilache wrote:

Thanks!

I'm -1 on the proposed name change: LLVM has {insert|extract}{element|value}, op names are globally consistent with that terminology.

There are many things MLIR does different from LLVM - there is no need to be consistent with an LLVM naming if that naming is going to be poor for tensor types. There is no extraction happening here - for low-level things like vectors in the LLVM IR, one could make a case for extract though (because you could think of it as being extracted from a vector register).

[silvas]

I generally agree with @bondhugula on tensor.get_element being somewhat better. However, I think tensor.extract is the right thing for this patch because:

1. The original op used the term "extract", so, despite the inaccurate connotations that @bondhugula points out, this is the status quo and doesn't make things worse.
2. Consistency with vector.extract (which, as @nicolasvasilache mentions, is also consistent with LLVM's terminology). We should have a joint discussion about renaming both ops, separate from this patch, about that -- in fact, this patch unblocks that discussion from happening in the future!

[mehdi_amini]

extract and get_element seem both unambiguous to me in the context we're using them, but I'm biased by coming from LLVM :)
While MLIR is free to diverge, some consistency with LLVM for the same concepts seems valuable to preserve still.

[rriddle]

Thanks Sean! This looks great.

mlir/include/mlir/Dialect/Tensor/IR/TensorBase.td

2

Missing filename here.

45

nit: Remove the extra newline.

mlir/include/mlir/Dialect/Tensor/IR/TensorOps.td

2

Same here.

14

Where is this dependency coming from?

mlir/lib/Dialect/StandardOps/IR/Ops.cpp

157

I'm assuming this is just a staging thing?

[silvas]

Address River's comments.

mlir/include/mlir/Dialect/Tensor/IR/TensorOps.td

14

Stray copypasta. Removed.

mlir/lib/Dialect/StandardOps/IR/Ops.cpp

157

Canonicalization patterns for std create tensor ops, so we need the dependency, as in any normal case of a dialect's canonicalizations creating ops from another dialect. The idea is that the new tensor dialect is a more fundamental dialect that the old "grab bag of stuff" (i.e. today's std) dialect depends on in the usual fashion.

I believe that once we move enough ops, it can go away though. But the end-goal is of course to have no Standard dialect, so this entire file is staging in a sense.

[silvas]

Remove some more stray copypasta.

[silvas]

Partially revert https://reviews.llvm.org/D92980

As discussed with River (and described in the comment in FoldUtils.cpp), in
the interim state of splitting std, there is a need for the tensor
dialect to materialize constants with std.constant. Implementing the
TensorDialect::materializeConstant hook with std.constant isn't possible
because it would create a cyclic dependency with std.

[silvas]

Touch up a few more places that mention extract_element.

[silvas]

One last nit.

mlir/include/mlir/Dialect/CMakeLists.txt

 add_subdirectory(Affine)
 add_subdirectory(Async)
 add_subdirectory(AVX512)
 add_subdirectory(GPU)
 add_subdirectory(Linalg)
 add_subdirectory(LLVMIR)
 add_subdirectory(OpenACC)
 add_subdirectory(OpenMP)
 add_subdirectory(PDL)
 add_subdirectory(PDLInterp)
 add_subdirectory(Quant)
 add_subdirectory(SCF)
 add_subdirectory(Shape)
 add_subdirectory(SPIRV)
 add_subdirectory(StandardOps)
+add_subdirectory(Tensor)
 add_subdirectory(Tosa)
 add_subdirectory(Vector)

mlir/include/mlir/Dialect/Linalg/EDSC/FoldedIntrinsics.h

 //===- FoldedIntrinsics.h - MLIR EDSC Intrinsics for Linalg -----*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 #ifndef MLIR_DIALECT_LINALG_EDSC_FOLDEDINTRINSICS_H_
 #define MLIR_DIALECT_LINALG_EDSC_FOLDEDINTRINSICS_H_
 
 #include "mlir/Dialect/Linalg/EDSC/Builders.h"
 #include "mlir/Dialect/Linalg/EDSC/Intrinsics.h"
+#include "mlir/Dialect/Tensor/IR/Tensor.h"
 
 #include "mlir/Transforms/FoldUtils.h"
 
 namespace mlir {
 namespace edsc {
 namespace intrinsics {
 
 template <typename Op>

 using folded_std_addi = FoldedValueBuilder<AddIOp>;
 using folded_std_addf = FoldedValueBuilder<AddFOp>;
 using folded_std_alloc = FoldedValueBuilder<AllocOp>;
 using folded_std_constant = FoldedValueBuilder<ConstantOp>;
 using folded_std_constant_float = FoldedValueBuilder<ConstantFloatOp>;
 using folded_std_constant_index = FoldedValueBuilder<ConstantIndexOp>;
 using folded_std_constant_int = FoldedValueBuilder<ConstantIntOp>;
 using folded_std_dim = FoldedValueBuilder<DimOp>;
-using folded_std_extract_element = FoldedValueBuilder<ExtractElementOp>;
 using folded_std_index_cast = FoldedValueBuilder<IndexCastOp>;
 using folded_std_muli = FoldedValueBuilder<MulIOp>;
 using folded_std_mulf = FoldedValueBuilder<MulFOp>;
 using folded_std_memref_cast = FoldedValueBuilder<MemRefCastOp>;
 using folded_std_select = FoldedValueBuilder<SelectOp>;
 using folded_std_load = FoldedValueBuilder<LoadOp>;
 using folded_std_subi = FoldedValueBuilder<SubIOp>;
 using folded_std_sub_view = FoldedValueBuilder<SubViewOp>;
 using folded_std_tanh = FoldedValueBuilder<TanhOp>;
 using folded_std_tensor_load = FoldedValueBuilder<TensorLoadOp>;
 using folded_std_view = FoldedValueBuilder<ViewOp>;
 using folded_std_zero_extendi = FoldedValueBuilder<ZeroExtendIOp>;
 using folded_std_sign_extendi = FoldedValueBuilder<SignExtendIOp>;
+using folded_tensor_extract = FoldedValueBuilder<tensor::ExtractOp>;
 } // namespace intrinsics
 } // namespace edsc
 } // namespace mlir
 
 #endif // MLIR_DIALECT_LINALG_EDSC_FOLDEDINTRINSICS_H_

mlir/include/mlir/Dialect/StandardOps/EDSC/Intrinsics.h

 //===- Intrinsics.h - MLIR EDSC Intrinsics for StandardOps ------*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 #ifndef MLIR_DIALECT_STANDARDOPS_EDSC_INTRINSICS_H_
 #define MLIR_DIALECT_STANDARDOPS_EDSC_INTRINSICS_H_
 
 #include "mlir/Dialect/StandardOps/EDSC/Builders.h"
+#include "mlir/Dialect/Tensor/IR/Tensor.h"
 
 namespace mlir {
 namespace edsc {
 namespace intrinsics {
 
 using std_addi = ValueBuilder<AddIOp>;
 using std_addf = ValueBuilder<AddFOp>;
 using std_alloc = ValueBuilder<AllocOp>;
 using std_alloca = ValueBuilder<AllocaOp>;
 using std_call = OperationBuilder<CallOp>;
 using std_create_complex = ValueBuilder<CreateComplexOp>;
 using std_constant = ValueBuilder<ConstantOp>;
 using std_constant_float = ValueBuilder<ConstantFloatOp>;
 using std_constant_index = ValueBuilder<ConstantIndexOp>;
 using std_constant_int = ValueBuilder<ConstantIntOp>;
 using std_dealloc = OperationBuilder<DeallocOp>;
 using std_divis = ValueBuilder<SignedDivIOp>;
 using std_diviu = ValueBuilder<UnsignedDivIOp>;
 using std_dim = ValueBuilder<DimOp>;
-using std_extract_element = ValueBuilder<ExtractElementOp>;
 using std_fpext = ValueBuilder<FPExtOp>;
 using std_fptrunc = ValueBuilder<FPTruncOp>;
 using std_im = ValueBuilder<ImOp>;
 using std_index_cast = ValueBuilder<IndexCastOp>;
 using std_muli = ValueBuilder<MulIOp>;
 using std_mulf = ValueBuilder<MulFOp>;
 using std_memref_cast = ValueBuilder<MemRefCastOp>;
 using std_re = ValueBuilder<ReOp>;
 using std_ret = OperationBuilder<ReturnOp>;
 using std_rsqrt = ValueBuilder<RsqrtOp>;
 using std_select = ValueBuilder<SelectOp>;
 using std_load = ValueBuilder<LoadOp>;
 using std_splat = ValueBuilder<SplatOp>;
 using std_store = OperationBuilder<StoreOp>;
 using std_subf = ValueBuilder<SubFOp>;
 using std_subi = ValueBuilder<SubIOp>;
 using std_sub_view = ValueBuilder<SubViewOp>;
 using std_tanh = ValueBuilder<TanhOp>;
 using std_tensor_load = ValueBuilder<TensorLoadOp>;
 using std_tensor_store = OperationBuilder<TensorStoreOp>;
 using std_view = ValueBuilder<ViewOp>;
 using std_zero_extendi = ValueBuilder<ZeroExtendIOp>;
 using std_sign_extendi = ValueBuilder<SignExtendIOp>;
+using tensor_extract = ValueBuilder<tensor::ExtractOp>;
 
 /// Branches into `block` with `operands`.
 BranchOp std_br(Block *block, ValueRange operands);
 
 /// Branches into `trueBranch` with `trueOperands` if `cond` evaluates to `true`
 /// or to `falseBranch` and `falseOperand` if `cond` evaluates to `false`.
 CondBranchOp std_cond_br(Value cond, Block *trueBranch, ValueRange trueOperands,
                          Block *falseBranch, ValueRange falseOperands);

mlir/include/mlir/Dialect/StandardOps/IR/Ops.td

 // ExpOp
 //===----------------------------------------------------------------------===//
 
 def Exp2Op : FloatUnaryOp<"exp2"> {
   let summary = "base-2 exponential of the specified value";
 }
 
 //===----------------------------------------------------------------------===//
-// ExtractElementOp
-//===----------------------------------------------------------------------===//
-
-def ExtractElementOp : Std_Op<"extract_element",
-    [NoSideEffect,
-     TypesMatchWith<"result type matches element type of aggregate",
-                    "aggregate", "result",
-                    "$_self.cast<ShapedType>().getElementType()">]> {
-  let summary = "element extract operation";
-  let description = [{
-    The `extract_element` op reads a tensor or vector and returns one element
-    from it specified by an index list. The output of the 'extract_element' is a
-    new value with the same type as the elements of the tensor or vector. The
-    arity of indices matches the rank of the accessed value (i.e., if a tensor
-    is of rank 3, then 3 indices are required for the extract. The indices
-    should all be of `index` type.
-
-    Example:
-
-    ```mlir
-    %3 = extract_element %v[%1, %2] : vector<4x4xi32>
-    %4 = extract_element %t[%1, %2] : tensor<4x4xi32>
-    %5 = extract_element %ut[%1, %2] : tensor<*xi32>
-    ```
-  }];
-
-  let arguments = (ins AnyTypeOf<[AnyVector, AnyTensor]>:$aggregate,
-                       Variadic<Index>:$indices);
-  let results = (outs AnyType:$result);
-
-  let builders = [
-    OpBuilderDAG<(ins "Value":$aggregate, CArg<"ValueRange", "{}">:$indices), [{
-      auto resType = aggregate.getType().cast<ShapedType>()
-                                         .getElementType();
-      build($_builder, $_state, resType, aggregate, indices);
-    }]>];
-
-  let extraClassDeclaration = [{
-    Value getAggregate() { return getOperand(0); }
-
-    operand_range getIndices() {
-      return {operand_begin() + 1, operand_end()};
-    }
-  }];
-
-  let hasFolder = 1;
-
-  let assemblyFormat = [{
-    $aggregate `[` $indices `]` attr-dict `:` type($aggregate)
-  }];
-}
-
-//===----------------------------------------------------------------------===//
 // TensorFromElementsOp
 //===----------------------------------------------------------------------===//
 
 def TensorFromElementsOp : Std_Op<"tensor_from_elements", [
     NoSideEffect,
     TypesMatchWith<"operand types match result element type",
                    "result", "elements", "SmallVector<Type, 2>("
                    "$_self.cast<ShapedType>().getDimSize(0), "

mlir/include/mlir/Dialect/Tensor/CMakeLists.txt

+add_subdirectory(IR)
+add_subdirectory(Transforms)

mlir/include/mlir/Dialect/Tensor/IR/CMakeLists.txt

+add_mlir_dialect(TensorOps tensor)
+add_mlir_doc(TensorOps -gen-dialect-doc TensorOps Dialects/)

mlir/include/mlir/Dialect/Tensor/IR/Tensor.h

+//===- Tensor.h - Tensor dialect --------------------------------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MLIR_DIALECT_TENSOR_IR_TENSOR_H_
+#define MLIR_DIALECT_TENSOR_IR_TENSOR_H_
+
+#include "mlir/IR/BuiltinTypes.h"
+#include "mlir/IR/Dialect.h"
+#include "mlir/IR/OpDefinition.h"
+#include "mlir/IR/OpImplementation.h"
+#include "mlir/Interfaces/SideEffectInterfaces.h"
+
+//===----------------------------------------------------------------------===//
+// Tensor Dialect
+//===----------------------------------------------------------------------===//
+
+#include "mlir/Dialect/Tensor/IR/TensorOpsDialect.h.inc"
+
+//===----------------------------------------------------------------------===//
+// Tensor Dialect Operations
+//===----------------------------------------------------------------------===//
+
+#define GET_OP_CLASSES
+#include "mlir/Dialect/Tensor/IR/TensorOps.h.inc"
+
+#endif // MLIR_DIALECT_TENSOR_IR_TENSOR_H_

mlir/include/mlir/Dialect/Tensor/IR/TensorBase.td

+//===- TensorBase.td - Base definitions for tensor dialect -*- tablegen -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef TENSOR_BASE
+#define TENSOR_BASE
+
+include "mlir/IR/OpBase.td"
+
+def Tensor_Dialect : Dialect {
+  let name = "tensor";
+  let cppNamespace = "::mlir::tensor";
+  let description = [{
+    The `tensor` dialect is intended to hold core tensor creation and
+    manipulation ops, which are not strongly associated with any particular
+    other dialect or domain abstraction. The primary smoke test of this is ops
+    that make sense for any tensor element type.
+
+    We leave it to other dialects to hold the vast swath of possible
+    computations one might want to do on a tensor.
+
+    The `tensor` type is (for better or for worse) used to represent all kinds
+    of things, and supports an open-ended set of element types. Examples:
+
+    - representing large, dense aggregations of primitive types, suitable for
+      high-performance numerical computing.
+    - representing shapes in the `shape` dialect, which consist of small
+      1D tensors of `index` data type.
+    - representing aggregations of strings or “variant” types.
+    - representing large, sparse aggregations of primitive types, suitable
+      for high-performance numerical computing.
+
+    Thus, for the `tensor` dialect, we prefer for now to constrain the
+    scope as much as possible. The expectation is that at some point
+    in the future, the `tensor` dialect’s scope may be broadened through a
+    careful discussion of the tradeoffs.
+
+    The `tensor` type is actually a builtin type (it lives in the builtin
+    dialect), and does not live in this dialect.
+
+  }];
+}
+
+#endif // TENSOR_BASE

mlir/include/mlir/Dialect/Tensor/IR/TensorOps.td

+//===- TensorOps.td - Tensor op definitions ----------------*- tablegen -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef TENSOR_OPS
+#define TENSOR_OPS
+
+include "mlir/Dialect/Tensor/IR/TensorBase.td"
+include "mlir/Interfaces/SideEffectInterfaces.td"
+include "mlir/Interfaces/InferTypeOpInterface.td"
+include "mlir/Interfaces/ControlFlowInterfaces.td"
+include "mlir/IR/SymbolInterfaces.td"
+
+class Tensor_Op<string mnemonic, list<OpTrait> traits = []>
+    : Op<Tensor_Dialect, mnemonic, traits> {
+  let printer = [{ return ::print(p, *this); }];
+  let verifier = [{ return ::verify(*this); }];
+  let parser = [{ return ::parse$cppClass(parser, result); }];
+}
+
+//===----------------------------------------------------------------------===//
+// ExtractOp
+//===----------------------------------------------------------------------===//
+
+def Tensor_ExtractOp : Tensor_Op<"extract",
+    [NoSideEffect,
+     TypesMatchWith<"result type matches element type of tensor",
+                    "tensor", "result",
+                    "$_self.cast<ShapedType>().getElementType()">]> {
+  let summary = "element extraction operation";
+  let description = [{
+    The `tensor.extract` op reads a tensor and returns one
+    element from it specified by an index list. The output of the op is a
+    new value with the same type as the elements of the tensor. The
+    arity of indices must match the rank of the accessed value (i.e., if a
+    tensor is of rank 3, then 3 indices are required for the extract. The
+    indices should all be of `index` type.
+
+    Example:
+
+    ```mlir
+    %4 = tensor.extract %t[%1, %2] : tensor<4x4xi32>
+    %5 = tensor.extract %rt[%1, %2] : tensor<?x?xi32>
+    %6 = tensor.extract %ut[%1, %2] : tensor<*xi32>
+    ```
+  }];
+
+  let arguments = (ins AnyTensor:$tensor, Variadic<Index>:$indices);
+  let results = (outs AnyType:$result);
+  let assemblyFormat = "$tensor `[` $indices `]` attr-dict `:` type($tensor)";
+
+  let builders = [
+    OpBuilderDAG<(ins "Value":$tensor, CArg<"ValueRange", "{}">:$indices), [{
+      auto resType = tensor.getType().cast<ShapedType>().getElementType();
+      build($_builder, $_state, resType, tensor, indices);
+    }]>];
+
+  let hasFolder = 1;
+}
+
+#endif // TENSOR_OPS

mlir/include/mlir/Dialect/Tensor/Transforms/CMakeLists.txt

+set(LLVM_TARGET_DEFINITIONS Passes.td)
+mlir_tablegen(Passes.h.inc -gen-pass-decls -name Tensor)
+add_public_tablegen_target(MLIRTensorTransformsIncGen)
+
+add_mlir_doc(Passes -gen-pass-doc TensorPasses ./)

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

+//===- Passes.h - Pass Entrypoints ------------------------------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MLIR_DIALECT_TENSOR_TRANSFORMS_PASSES_H_
+#define MLIR_DIALECT_TENSOR_TRANSFORMS_PASSES_H_
+
+#include "mlir/Pass/Pass.h"
+#include "mlir/Transforms/Bufferize.h"
+
+namespace mlir {
+
+class OwningRewritePatternList;
+
+void populateTensorBufferizePatterns(MLIRContext *context,
+                                     BufferizeTypeConverter &typeConverter,
+                                     OwningRewritePatternList &patterns);
+
+/// Creates an instance of `tensor` dialect bufferization pass.
+std::unique_ptr<Pass> createTensorBufferizePass();
+
+//===----------------------------------------------------------------------===//
+// Registration
+//===----------------------------------------------------------------------===//
+
+namespace tensor {
+/// Generate the code for registering passes.
+#define GEN_PASS_REGISTRATION
+#include "mlir/Dialect/Tensor/Transforms/Passes.h.inc"
+} // namespace tensor
+
+} // end namespace mlir
+
+#endif // MLIR_DIALECT_TENSOR_TRANSFORMS_PASSES_H_

mlir/include/mlir/Dialect/Tensor/Transforms/Passes.td

+//===-- Passes.td - pass definition file -------------------*- tablegen -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MLIR_DIALECT_TENSOR_TRANSFORMS_PASSES
+#define MLIR_DIALECT_TENSOR_TRANSFORMS_PASSES
+
+include "mlir/Pass/PassBase.td"
+
+def TensorBufferize : FunctionPass<"tensor-bufferize"> {
+  let summary = "Bufferize the `tensor` dialect";
+  let constructor = "mlir::createTensorBufferizePass()";
+}
+
+#endif // MLIR_DIALECT_TENSOR_TRANSFORMS_PASSES

mlir/include/mlir/InitAllDialects.h

 #include "mlir/Dialect/PDL/IR/PDL.h"
 #include "mlir/Dialect/PDLInterp/IR/PDLInterp.h"
 #include "mlir/Dialect/Quant/QuantOps.h"
 #include "mlir/Dialect/SCF/SCF.h"
 #include "mlir/Dialect/SDBM/SDBMDialect.h"
 #include "mlir/Dialect/SPIRV/SPIRVDialect.h"
 #include "mlir/Dialect/Shape/IR/Shape.h"
 #include "mlir/Dialect/StandardOps/IR/Ops.h"
+#include "mlir/Dialect/Tensor/IR/Tensor.h"
 #include "mlir/Dialect/Tosa/IR/TosaOps.h"
 #include "mlir/Dialect/Vector/VectorOps.h"
 #include "mlir/IR/Dialect.h"
 
 namespace mlir {
 
 // Add all the MLIR dialects to the provided registry.
 inline void registerAllDialects(DialectRegistry &registry) {
registry.insert<acc::OpenACCDialect,
                   quant::QuantizationDialect,
                   spirv::SPIRVDialect,
                   StandardOpsDialect,
                   vector::VectorDialect,
                   NVVM::NVVMDialect,
                   ROCDL::ROCDLDialect,
                   SDBMDialect,
                   shape::ShapeDialect,
+                  tensor::TensorDialect,
                   tosa::TosaDialect>();
   // clang-format on
 }
 
 } // namespace mlir
 
 #endif // MLIR_INITALLDIALECTS_H_

mlir/include/mlir/InitAllPasses.h

 #include "mlir/Dialect/GPU/Passes.h"
 #include "mlir/Dialect/LLVMIR/Transforms/Passes.h"
 #include "mlir/Dialect/Linalg/Passes.h"
 #include "mlir/Dialect/Quant/Passes.h"
 #include "mlir/Dialect/SCF/Passes.h"
 #include "mlir/Dialect/SPIRV/Passes.h"
 #include "mlir/Dialect/Shape/Transforms/Passes.h"
 #include "mlir/Dialect/StandardOps/Transforms/Passes.h"
+#include "mlir/Dialect/Tensor/Transforms/Passes.h"
 #include "mlir/Dialect/Tosa/Transforms/Passes.h"
 #include "mlir/Transforms/Passes.h"
 
 #include <cstdlib>
 
 namespace mlir {
 
 // This function may be called to register the MLIR passes with the
inline void registerAllPasses() {
   registerGPUPasses();
   registerLinalgPasses();
   LLVM::registerLLVMPasses();
   quant::registerQuantPasses();
   registerSCFPasses();
   registerShapePasses();
   spirv::registerSPIRVPasses();
   registerStandardPasses();
+  tensor::registerTensorPasses();
   tosa::registerTosaOptPasses();
 }
 
 } // namespace mlir
 
 #endif // MLIR_INITALLPASSES_H_

mlir/lib/Conversion/ShapeToStandard/ConvertShapeConstraints.cpp

 //===- ConvertShapeConstraints.cpp - Conversion of shape constraints ------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
 #include "mlir/Conversion/ShapeToStandard/ShapeToStandard.h"
 
 #include "../PassDetail.h"
 #include "mlir/Dialect/SCF/SCF.h"
 #include "mlir/Dialect/Shape/IR/Shape.h"
 #include "mlir/Dialect/StandardOps/IR/Ops.h"
+#include "mlir/Dialect/Tensor/IR/Tensor.h"
 #include "mlir/IR/PatternMatch.h"
 #include "mlir/Pass/Pass.h"
 #include "mlir/Pass/PassRegistry.h"
 #include "mlir/Transforms/GreedyPatternRewriteDriver.h"
 
 using namespace mlir;
 
 namespace {
LogicalResult matchAndRewrite(shape::CstrBroadcastableOp op,
     // non-broadcast-compatible shapes.
     // Two extents are broadcast-compatible if
     // 1. they are both equal, or
     // 2. at least one of them is 1.
 
     rewriter.create<scf::ForOp>(
         loc, rankDiff, greaterRank, one, llvm::None,
         [&](OpBuilder &b, Location loc, Value iv, ValueRange) {
-          Value greaterRankOperandExtent = b.create<ExtractElementOp>(
+          Value greaterRankOperandExtent = b.create<tensor::ExtractOp>(
               loc, greaterRankOperand, ValueRange{iv});
           Value ivShifted = b.create<SubIOp>(loc, indexTy, iv, rankDiff);
-          Value lesserRankOperandExtent = b.create<ExtractElementOp>(
+          Value lesserRankOperandExtent = b.create<tensor::ExtractOp>(
               loc, lesserRankOperand, ValueRange{ivShifted});
 
           Value greaterRankOperandExtentIsOne = b.create<CmpIOp>(
               loc, CmpIPredicate::eq, greaterRankOperandExtent, one);
           Value lesserRankOperandExtentIsOne = b.create<CmpIOp>(
               loc, CmpIPredicate::eq, lesserRankOperandExtent, one);
           Value extentsAgree =
               b.create<CmpIOp>(loc, CmpIPredicate::eq, greaterRankOperandExtent,

mlir/lib/Conversion/ShapeToStandard/ShapeToStandard.cpp

 //===- ShapeToStandard.cpp - conversion from Shape to Standard dialect ----===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
 #include "mlir/Conversion/ShapeToStandard/ShapeToStandard.h"
 
 #include "../PassDetail.h"
 #include "mlir/Dialect/SCF/SCF.h"
 #include "mlir/Dialect/Shape/IR/Shape.h"
 #include "mlir/Dialect/StandardOps/IR/Ops.h"
+#include "mlir/Dialect/Tensor/IR/Tensor.h"
 #include "mlir/IR/BlockAndValueMapping.h"
 #include "mlir/Transforms/DialectConversion.h"
 
 using namespace mlir;
 using namespace mlir::shape;
 using namespace mlir::scf;
 
 /// Conversion patterns.
LogicalResult BroadcastOpConverter::matchAndRewrite(
   Value rankDiff =
       rewriter.create<SubIOp>(loc, indexTy, greaterRank, lesserRank);
   rewriter.replaceOpWithNewOp<DynamicTensorFromElementsOp>(
       op, getExtentTensorType(op.getContext()), ValueRange{greaterRank},
       [&](OpBuilder &b, Location loc, ValueRange args) {
         Value outputDimension = args[0];
         Value isUnchallengedDimension = b.create<CmpIOp>(
             loc, CmpIPredicate::ult, outputDimension, rankDiff);
-        Value greaterRankOperandExtent = b.create<ExtractElementOp>(
+        Value greaterRankOperandExtent = b.create<tensor::ExtractOp>(
             loc, greaterRankOperand, outputDimension);
         // The initial dimensions of the greater-rank operand are unchallenged,
         // so we can take them as-is. Otherwise, we need to do a comparison.
         // We need an actual branch here (instead of a select) because the
         // lesser-rank operand might be rank 0, so any extract_element would be
         // invalid.
         auto ifOp = b.create<IfOp>(
             loc, TypeRange{indexTy}, isUnchallengedDimension,
             [&](OpBuilder &b, Location loc) {
               b.create<scf::YieldOp>(loc, greaterRankOperandExtent);
             },
             [&](OpBuilder &b, Location loc) {
               // The broadcasting logic is:
               // - if one extent (here we arbitrarily choose the extent from
               // the greater-rank operand) is equal to 1, then take the extent
               // from the other operand
               // - otherwise, take the extent as-is.
               // Note that this logic remains correct in the presence of
               // dimensions of zero extent.
               Value lesserRankOperandDimension =
                   b.create<SubIOp>(loc, indexTy, outputDimension, rankDiff);
-              Value lesserRankOperandExtent = b.create<ExtractElementOp>(
+              Value lesserRankOperandExtent = b.create<tensor::ExtractOp>(
                   loc, lesserRankOperand,
                   ValueRange{lesserRankOperandDimension});
               Value greaterRankOperandExtentIsOne = b.create<CmpIOp>(
                   loc, CmpIPredicate::eq, greaterRankOperandExtent, one);
               Value broadcastedExtent = b.create<SelectOp>(
                   loc, greaterRankOperandExtentIsOne, lesserRankOperandExtent,
                   greaterRankOperandExtent);
               b.create<scf::YieldOp>(loc, broadcastedExtent);
LogicalResult IsBroadcastableOpConverter::matchAndRewrite(
   Type i1Ty = rewriter.getI1Type();
   Value init =
       rewriter.create<ConstantOp>(loc, i1Ty, rewriter.getBoolAttr(true));
 
   // Determine if all overlapping extents are broadcastable.
   auto reduceResult = rewriter.create<ForOp>(
       loc, rankDiff, greaterRank, one, ValueRange{init},
       [&](OpBuilder &b, Location loc, Value iv, ValueRange iterArgs) {
-        Value greaterRankOperandExtent =
-            b.create<ExtractElementOp>(loc, greaterRankOperand, ValueRange{iv});
+        Value greaterRankOperandExtent = b.create<tensor::ExtractOp>(
+            loc, greaterRankOperand, ValueRange{iv});
         Value greaterRankOperandExtentIsOne = b.create<CmpIOp>(
             loc, CmpIPredicate::eq, greaterRankOperandExtent, one);
         Value ivShifted = b.create<SubIOp>(loc, indexTy, iv, rankDiff);
-        Value lesserRankOperandExtent = b.create<ExtractElementOp>(
+        Value lesserRankOperandExtent = b.create<tensor::ExtractOp>(
             loc, lesserRankOperand, ValueRange{ivShifted});
         Value lesserRankOperandExtentIsOne = b.create<CmpIOp>(
             loc, CmpIPredicate::eq, lesserRankOperandExtent, one);
         Value extentsAreEqual =
             b.create<CmpIOp>(loc, CmpIPredicate::eq, greaterRankOperandExtent,
                              lesserRankOperandExtent);
         Value broadcastableExtents = b.create<AndOp>(
             loc, iterArgs[0],
LogicalResult GetExtentOpConverter::matchAndRewrite(
   if (auto shapeOfOp = op.shape().getDefiningOp<ShapeOfOp>()) {
     if (shapeOfOp.arg().getType().isa<ShapedType>()) {
       rewriter.replaceOpWithNewOp<DimOp>(op, shapeOfOp.arg(),
                                          transformed.dim());
       return success();
     }
   }
 
-  rewriter.replaceOpWithNewOp<ExtractElementOp>(op, rewriter.getIndexType(),
+  rewriter.replaceOpWithNewOp<tensor::ExtractOp>(op, rewriter.getIndexType(),
                                                  transformed.shape(),
                                                  ValueRange{transformed.dim()});
   return success();
 }
 
 namespace {
 class RankOpConverter : public OpConversionPattern<shape::RankOp> {
 public:
   using OpConversionPattern<shape::RankOp>::OpConversionPattern;
 
ReduceOpConverter::matchAndRewrite(shape::ReduceOp op, ArrayRef<Value> operands,
   Value zero = rewriter.create<ConstantIndexOp>(loc, 0);
   Value one = rewriter.create<ConstantIndexOp>(loc, 1);
   Type indexTy = rewriter.getIndexType();
   Value rank = rewriter.create<DimOp>(loc, indexTy, transformed.shape(), zero);
 
   auto loop = rewriter.create<scf::ForOp>(
       loc, zero, rank, one, op.initVals(),
       [&](OpBuilder &b, Location loc, Value iv, ValueRange args) {
-        Value extent = b.create<ExtractElementOp>(loc, transformed.shape(), iv);
+        Value extent =
+            b.create<tensor::ExtractOp>(loc, transformed.shape(), iv);
 
         SmallVector<Value, 2> mappedValues{iv, extent};
         mappedValues.append(args.begin(), args.end());
 
         BlockAndValueMapping mapping;
         Block *reduceBody = op.getBody();
         mapping.map(reduceBody->getArguments(), mappedValues);
         for (auto &nested : reduceBody->without_terminator())
rewriter.replaceOpWithNewOp<IfOp>(
       [&](OpBuilder &b, Location loc) {
         Value one = b.create<ConstantIndexOp>(loc, 1);
         Value init = b.create<ConstantOp>(loc, i1Ty, b.getBoolAttr(true));
         auto loop = b.create<scf::ForOp>(
             loc, zero, lhsRank, one, ValueRange{init},
             [&](OpBuilder &b, Location nestedLoc, Value iv, ValueRange args) {
               Value conj = args[0];
               Value lhsExtent =
-                  b.create<ExtractElementOp>(loc, transformed.lhs(), iv);
+                  b.create<tensor::ExtractOp>(loc, transformed.lhs(), iv);
               Value rhsExtent =
-                  b.create<ExtractElementOp>(loc, transformed.rhs(), iv);
+                  b.create<tensor::ExtractOp>(loc, transformed.rhs(), iv);
               Value eqExtent = b.create<CmpIOp>(loc, CmpIPredicate::eq,
                                                 lhsExtent, rhsExtent);
               Value conjNext = b.create<AndOp>(loc, conj, eqExtent);
               b.create<scf::YieldOp>(loc, ValueRange({conjNext}));
             });
         b.create<scf::YieldOp>(loc, loop.getResults());
       },
       [&](OpBuilder &b, Location loc) {
class ConvertShapeToStandardPass
   void runOnOperation() override;
 };
 } // namespace
 
 void ConvertShapeToStandardPass::runOnOperation() {
   // Setup target legality.
   MLIRContext &ctx = getContext();
   ConversionTarget target(ctx);
-  target.addLegalDialect<StandardOpsDialect, SCFDialect>();
+  target
+      .addLegalDialect<StandardOpsDialect, SCFDialect, tensor::TensorDialect>();
   target.addLegalOp<CstrRequireOp, FuncOp, ModuleOp, ModuleTerminatorOp>();
 
   // Setup conversion patterns.
   OwningRewritePatternList patterns;
   populateShapeToStandardConversionPatterns(patterns, &ctx);
 
   // Apply conversion.
   auto module = getOperation();

mlir/lib/Dialect/CMakeLists.txt

 add_subdirectory(Affine)
 add_subdirectory(Async)
 add_subdirectory(AVX512)
 add_subdirectory(GPU)
 add_subdirectory(Linalg)
 add_subdirectory(LLVMIR)
 add_subdirectory(OpenACC)
 add_subdirectory(OpenMP)
 add_subdirectory(PDL)
 add_subdirectory(PDLInterp)
 add_subdirectory(Quant)
 add_subdirectory(SCF)
 add_subdirectory(SDBM)
 add_subdirectory(Shape)
 add_subdirectory(SPIRV)
 add_subdirectory(StandardOps)
+add_subdirectory(Tensor)
 add_subdirectory(Tosa)
 add_subdirectory(Vector)
 
 set(LLVM_OPTIONAL_SOURCES
   Traits.cpp
 )
 
 add_mlir_library(MLIRDialect
   Traits.cpp
 
   ADDITIONAL_HEADER_DIRS
   ${MLIR_MAIN_INCLUDE_DIR}/mlir/Dialect
 
   LINK_LIBS PUBLIC
   MLIRIR
   )

mlir/lib/Dialect/StandardOps/CMakeLists.txt

add_mlir_dialect_library(MLIRStandard
   MLIRStandardOpsIncGen
 
   LINK_LIBS PUBLIC
   MLIRCallInterfaces
   MLIRControlFlowInterfaces
   MLIREDSC
   MLIRIR
   MLIRSideEffectInterfaces
+  MLIRTensor
   MLIRVectorInterfaces
   MLIRViewLikeInterface
   )
 
 add_subdirectory(Transforms)

mlir/lib/Dialect/StandardOps/IR/Ops.cpp

 //===- Ops.cpp - Standard MLIR Operations ---------------------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
 #include "mlir/Dialect/StandardOps/IR/Ops.h"
 
 #include "mlir/Dialect/CommonFolders.h"
+#include "mlir/Dialect/Tensor/IR/Tensor.h"
 #include "mlir/IR/AffineExpr.h"
 #include "mlir/IR/AffineMap.h"
 #include "mlir/IR/BlockAndValueMapping.h"
 #include "mlir/IR/Builders.h"
 #include "mlir/IR/BuiltinOps.h"
 #include "mlir/IR/BuiltinTypes.h"
 #include "mlir/IR/Matchers.h"
 #include "mlir/IR/OpImplementation.h"
static LogicalResult verifyCastOp(T op) {
   if (!T::areCastCompatible(opType, resType))
     return op.emitError("operand type ") << opType << " and result type "
                                          << resType << " are cast incompatible";
 
   return success();
 }
 
 void StandardOpsDialect::initialize() {
+  getContext()->loadDialect<tensor::TensorDialect>();
   addOperations<DmaStartOp, DmaWaitOp,
 #define GET_OP_LIST
 #include "mlir/Dialect/StandardOps/IR/Ops.cpp.inc"
                 >();
   addInterfaces<StdInlinerInterface>();
 }
 
 /// Materialize a single constant operation from a given attribute value with

 ///   ^bb0(%arg0: index):  // no predecessors
 ///   <computation>
 ///   yield %1 : index
 /// } : tensor<?xindex>
 /// %extracted_element = extract_element %tensor[%c0] : tensor<?xi32>
 ///
 /// to just <computation> with %arg0 replaced by %c0. We only do this if the
 /// dynamic_tensor_from_elements operation has no side-effects.
-struct ExtractElementFromDynamicTensorFromElements
-    : public OpRewritePattern<ExtractElementOp> {
-  using OpRewritePattern<ExtractElementOp>::OpRewritePattern;
+struct ExtractFromDynamicTensorFromElements
+    : public OpRewritePattern<tensor::ExtractOp> {
+  using OpRewritePattern<tensor::ExtractOp>::OpRewritePattern;
 
-  LogicalResult matchAndRewrite(ExtractElementOp extract,
+  LogicalResult matchAndRewrite(tensor::ExtractOp extract,
                                 PatternRewriter &rewriter) const final {
     auto tensorFromElements =
-        extract.aggregate().getDefiningOp<DynamicTensorFromElementsOp>();
+        extract.tensor().getDefiningOp<DynamicTensorFromElementsOp>();
     if (!tensorFromElements || !wouldOpBeTriviallyDead(tensorFromElements))
       return failure();
 
     BlockAndValueMapping mapping;
     Block *body = tensorFromElements.getBody();
     mapping.map(body->getArguments(), extract.indices());
     for (auto &op : body->without_terminator())
       rewriter.clone(op, mapping);
 
     auto yield = cast<YieldOp>(body->getTerminator());
 
     rewriter.replaceOp(extract, mapping.lookupOrDefault(yield.value()));
     return success();
   }
 };
 
 /// Canonicalizes the pattern of the form
 ///
 /// %val = tensor_cast %source : : tensor<?xi32> to tensor<2xi32>
-/// %extracted_element = extract_element %val[%c0] : tensor<2xi32>
+/// %extracted_element = tensor.extract %val[%c0] : tensor<2xi32>
 ///
 /// to
 ///
-/// %extracted_element = extract_element %source[%c0] : tensor<?xi32>
-struct ExtractElementFromTensorCast
-    : public OpRewritePattern<ExtractElementOp> {
-  using OpRewritePattern<ExtractElementOp>::OpRewritePattern;
+/// %extracted_element = tensor.extract %source[%c0] : tensor<?xi32>
+struct ExtractFromTensorCast : public OpRewritePattern<tensor::ExtractOp> {
+  using OpRewritePattern<tensor::ExtractOp>::OpRewritePattern;
 
-  LogicalResult matchAndRewrite(ExtractElementOp extract,
+  LogicalResult matchAndRewrite(tensor::ExtractOp extract,
                                 PatternRewriter &rewriter) const final {
-    auto tensorCast = extract.aggregate().getDefiningOp<TensorCastOp>();
+    auto tensorCast = extract.tensor().getDefiningOp<TensorCastOp>();
     if (!tensorCast)
       return failure();
 
-    rewriter.replaceOpWithNewOp<ExtractElementOp>(extract, tensorCast.source(),
-                                                  extract.getIndices());
+    rewriter.replaceOpWithNewOp<tensor::ExtractOp>(extract, tensorCast.source(),
+                                                   extract.indices());
     return success();
   }
 };
 
 } // namespace
 
 void DynamicTensorFromElementsOp::getCanonicalizationPatterns(
     OwningRewritePatternList &results, MLIRContext *context) {
-  results.insert<ExtractElementFromDynamicTensorFromElements,
-                 ExtractElementFromTensorCast, StaticDynamicTensorFromElements>(
-      context);
-}
-
-//===----------------------------------------------------------------------===//
-// ExtractElementOp
-//===----------------------------------------------------------------------===//
-
-static LogicalResult verify(ExtractElementOp op) {
-  // Verify the # indices match if we have a ranked type.
-  auto aggregateType = op.getAggregate().getType().cast<ShapedType>();
-  if (aggregateType.hasRank() &&
-      aggregateType.getRank() != op.getNumOperands() - 1)
-    return op.emitOpError("incorrect number of indices for extract_element");
-
-  return success();
-}
-
-OpFoldResult ExtractElementOp::fold(ArrayRef<Attribute> operands) {
-  assert(!operands.empty() && "extract_element takes at least one operand");
-
-  // The aggregate operand must be a known constant.
-  Attribute aggregate = operands.front();
-  if (!aggregate)
-    return {};
-
-  // If this is a splat elements attribute, simply return the value. All of the
-  // elements of a splat attribute are the same.
-  if (auto splatAggregate = aggregate.dyn_cast<SplatElementsAttr>())
-    return splatAggregate.getSplatValue();
-
-  // Otherwise, collect the constant indices into the aggregate.
-  SmallVector<uint64_t, 8> indices;
-  for (Attribute indice : llvm::drop_begin(operands, 1)) {
-    if (!indice || !indice.isa<IntegerAttr>())
-      return {};
-    indices.push_back(indice.cast<IntegerAttr>().getInt());
-  }
-
-  // If this is an elements attribute, query the value at the given indices.
-  auto elementsAttr = aggregate.dyn_cast<ElementsAttr>();
-  if (elementsAttr && elementsAttr.isValidIndex(indices))
-    return elementsAttr.getValue(indices);
-  return {};
+  // TODO: Move extract patterns to tensor::ExtractOp.
+  results.insert<ExtractFromDynamicTensorFromElements, ExtractFromTensorCast,
+                 StaticDynamicTensorFromElements>(context);
 }
 
 //===----------------------------------------------------------------------===//
 // TensorFromElementsOp
 //===----------------------------------------------------------------------===//
 
 void TensorFromElementsOp::build(OpBuilder &builder, OperationState &result,
                                  Type elementType, ValueRange elements) {

 
 // Canonicalizes the pattern of the form
 //
 // %tensor = "tensor_from_elements(%element) : (i32) -> tensor<1xi32>
 // %extracted_element = extract_element %tensor[%c0] : tensor<1xi32>
 //
 // to just %element.
 struct ExtractElementFromTensorFromElements
-    : public OpRewritePattern<ExtractElementOp> {
-  using OpRewritePattern<ExtractElementOp>::OpRewritePattern;
+    : public OpRewritePattern<tensor::ExtractOp> {
+  using OpRewritePattern<tensor::ExtractOp>::OpRewritePattern;
 
-  LogicalResult matchAndRewrite(ExtractElementOp extract,
+  LogicalResult matchAndRewrite(tensor::ExtractOp extract,
                                 PatternRewriter &rewriter) const final {
     if (extract.indices().size() != 1)
       return failure();
 
     auto tensorFromElements = dyn_cast_or_null<TensorFromElementsOp>(
-        extract.aggregate().getDefiningOp());
+        extract.tensor().getDefiningOp());
     if (tensorFromElements == nullptr)
       return failure();
 
     APInt index;
     if (!matchPattern(*extract.indices().begin(), m_ConstantInt(&index)))
       return failure();
     rewriter.replaceOp(extract,
                        tensorFromElements.getOperand(index.getZExtValue()));
struct LoadOfTensorToMemref : public OpRewritePattern<LoadOp> {
   using OpRewritePattern<LoadOp>::OpRewritePattern;
 
   LogicalResult matchAndRewrite(LoadOp load,
                                 PatternRewriter &rewriter) const override {
     auto tensorToMemref = load.memref().getDefiningOp<TensorToMemrefOp>();
     if (!tensorToMemref)
       return failure();
 
-    rewriter.replaceOpWithNewOp<ExtractElementOp>(load, tensorToMemref.tensor(),
-                                                  load.indices());
+    rewriter.replaceOpWithNewOp<tensor::ExtractOp>(
+        load, tensorToMemref.tensor(), load.indices());
     return success();
   }
 };
 } // end anonymous namespace.
 
 void LoadOp::getCanonicalizationPatterns(OwningRewritePatternList &results,
                                          MLIRContext *context) {
   results.insert<LoadOfTensorToMemref>(context);

mlir/lib/Dialect/StandardOps/Transforms/Bufferize.cpp

 //
 //===----------------------------------------------------------------------===//
 
 #include "mlir/Transforms/Bufferize.h"
 #include "PassDetail.h"
 #include "mlir/Dialect/SCF/SCF.h"
 #include "mlir/Dialect/StandardOps/IR/Ops.h"
 #include "mlir/Dialect/StandardOps/Transforms/Passes.h"
+#include "mlir/Dialect/Tensor/IR/Tensor.h"
 #include "mlir/IR/BlockAndValueMapping.h"
 #include "mlir/Transforms/DialectConversion.h"
 
 using namespace mlir;
 
 namespace {
 class BufferizeDimOp : public OpConversionPattern<DimOp> {
 public:
matchAndRewrite(DynamicTensorFromElementsOp op, ArrayRef<Value> operands,
 
     rewriter.replaceOp(op, {result});
     return success();
   }
 };
 } // namespace
 
 namespace {
-class BufferizeExtractElementOp : public OpConversionPattern<ExtractElementOp> {
-public:
-  using OpConversionPattern::OpConversionPattern;
-  LogicalResult
-  matchAndRewrite(ExtractElementOp op, ArrayRef<Value> operands,
-                  ConversionPatternRewriter &rewriter) const override {
-    ExtractElementOp::Adaptor adaptor(operands);
-    rewriter.replaceOpWithNewOp<LoadOp>(op, adaptor.aggregate(),
-                                        adaptor.indices());
-    return success();
-  }
-};
-} // namespace
-
-namespace {
 class BufferizeSelectOp : public OpConversionPattern<SelectOp> {
 public:
   using OpConversionPattern::OpConversionPattern;
   LogicalResult
   matchAndRewrite(SelectOp op, ArrayRef<Value> operands,
                   ConversionPatternRewriter &rewriter) const override {
     if (!op.condition().getType().isa<IntegerType>())
       return rewriter.notifyMatchFailure(op, "requires scalar condition");

 
 void mlir::populateStdBufferizePatterns(MLIRContext *context,
                                         BufferizeTypeConverter &typeConverter,
                                         OwningRewritePatternList &patterns) {
   patterns.insert<
       // clang-format off
       BufferizeDimOp,
       BufferizeDynamicTensorFromElementsOp,
-      BufferizeExtractElementOp,
       BufferizeSelectOp,
       BufferizeTensorCastOp,
       BufferizeTensorFromElementsOp
       // clang-format on
       >(typeConverter, context);
 }
 
 namespace {
 struct StdBufferizePass : public StdBufferizeBase<StdBufferizePass> {
   void runOnFunction() override {
     auto *context = &getContext();
     BufferizeTypeConverter typeConverter;
     OwningRewritePatternList patterns;
     ConversionTarget target(*context);
 
     target.addLegalDialect<StandardOpsDialect>();
     target.addLegalDialect<scf::SCFDialect>();
+    target.addLegalDialect<tensor::TensorDialect>();
 
     populateStdBufferizePatterns(context, typeConverter, patterns);
-    target.addIllegalOp<DynamicTensorFromElementsOp, ExtractElementOp,
-                        TensorCastOp, TensorFromElementsOp>();
+    target.addIllegalOp<DynamicTensorFromElementsOp, TensorCastOp,
+                        TensorFromElementsOp>();
     // We only bufferize the case of tensor selected type and scalar condition,
     // as that boils down to a select over memref descriptors (don't need to
     // touch the data).
     target.addDynamicallyLegalOp<SelectOp>([&](SelectOp op) {
       return typeConverter.isLegal(op.getType()) ||
              !op.condition().getType().isa<IntegerType>();
     });
     target.addDynamicallyLegalOp<DimOp>(

mlir/lib/Dialect/Tensor/CMakeLists.txt

+add_subdirectory(IR)
+add_subdirectory(Transforms)

mlir/lib/Dialect/Tensor/IR/CMakeLists.txt

+add_mlir_dialect_library(MLIRTensor
+  TensorDialect.cpp
+  TensorOps.cpp
+
+  ADDITIONAL_HEADER_DIRS
+  ${PROJECT_SOURCE_DIR}/include/mlir/Dialect/Tensor
+
+  DEPENDS
+  MLIRTensorOpsIncGen
+
+  LINK_COMPONENTS
+  Core
+
+  LINK_LIBS PUBLIC
+  MLIRIR
+  MLIRSupport
+  )

mlir/lib/Dialect/Tensor/IR/TensorDialect.cpp

+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include "mlir/Dialect/Tensor/IR/Tensor.h"
+#include "mlir/Transforms/InliningUtils.h"
+
+using namespace mlir;
+using namespace mlir::tensor;
+
+//===----------------------------------------------------------------------===//
+// TensorDialect Dialect Interfaces
+//===----------------------------------------------------------------------===//
+
+namespace {
+struct TensorInlinerInterface : public DialectInlinerInterface {
+  using DialectInlinerInterface::DialectInlinerInterface;
+  bool isLegalToInline(Region *dest, Region *src, bool wouldBeCloned,
+                       BlockAndValueMapping &valueMapping) const final {
+    return true;
+  }
+  bool isLegalToInline(Operation *, Region *, bool wouldBeCloned,
+                       BlockAndValueMapping &) const final {
+    return true;
+  }
+};
+} // end anonymous namespace
+
+void TensorDialect::initialize() {
+  addOperations<
+#define GET_OP_LIST
+#include "mlir/Dialect/Tensor/IR/TensorOps.cpp.inc"
+      >();
+  addInterfaces<TensorInlinerInterface>();
+}

mlir/lib/Dialect/Tensor/IR/TensorOps.cpp

+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include "mlir/Dialect/Shape/IR/Shape.h"
+#include "mlir/Dialect/Tensor/IR/Tensor.h"
+#include "mlir/IR/TypeUtilities.h"
+#include "llvm/ADT/STLExtras.h"
+
+using namespace mlir;
+using namespace mlir::tensor;
+
+//===----------------------------------------------------------------------===//
+// ExtractOp
+//===----------------------------------------------------------------------===//
+
+static LogicalResult verify(ExtractOp op) {
+  // Verify the # indices match if we have a ranked type.
+  if (auto tensorType = op.tensor().getType().dyn_cast<RankedTensorType>())
+    if (tensorType.getRank() != static_cast<int64_t>(op.indices().size()))
+      return op.emitOpError("incorrect number of indices for extract_element");
+
+  return success();
+}
+
+OpFoldResult ExtractOp::fold(ArrayRef<Attribute> operands) {
+  assert(!operands.empty() && "extract_element takes at least one operand");
+
+  // The tensor operand must be a known constant.
+  Attribute tensor = operands.front();
+  if (!tensor)
+    return {};
+
+  // If this is a splat elements attribute, simply return the value. All of the
+  // elements of a splat attribute are the same.
+  if (auto splatTensor = tensor.dyn_cast<SplatElementsAttr>())
+    return splatTensor.getSplatValue();
+
+  // Otherwise, collect the constant indices into the tensor.
+  SmallVector<uint64_t, 8> indices;
+  for (Attribute indice : llvm::drop_begin(operands, 1)) {
+    if (!indice || !indice.isa<IntegerAttr>())
+      return {};
+    indices.push_back(indice.cast<IntegerAttr>().getInt());
+  }
+
+  // If this is an elements attribute, query the value at the given indices.
+  auto elementsAttr = tensor.dyn_cast<ElementsAttr>();
+  if (elementsAttr && elementsAttr.isValidIndex(indices))
+    return elementsAttr.getValue(indices);
+  return {};
+}
+
+//===----------------------------------------------------------------------===//
+// TableGen'd op method definitions
+//===----------------------------------------------------------------------===//
+
+#define GET_OP_CLASSES
+#include "mlir/Dialect/Tensor/IR/TensorOps.cpp.inc"

mlir/lib/Dialect/Tensor/Transforms/Bufferize.cpp

+//===- Bufferize.cpp - Bufferization for `tensor` dialect ops -------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements bufferization of `tensor` dialect ops
+//
+//===----------------------------------------------------------------------===//
+
+#include "mlir/Transforms/Bufferize.h"
+#include "PassDetail.h"
+#include "mlir/Dialect/StandardOps/IR/Ops.h"
+#include "mlir/Dialect/Tensor/IR/Tensor.h"
+#include "mlir/Dialect/Tensor/Transforms/Passes.h"
+#include "mlir/Transforms/DialectConversion.h"
+
+using namespace mlir;
+
+namespace {
+class BufferizeExtractOp : public OpConversionPattern<tensor::ExtractOp> {
+public:
+  using OpConversionPattern::OpConversionPattern;
+  LogicalResult
+  matchAndRewrite(tensor::ExtractOp op, ArrayRef<Value> operands,
+                  ConversionPatternRewriter &rewriter) const override {
+    tensor::ExtractOp::Adaptor adaptor(operands);
+    rewriter.replaceOpWithNewOp<LoadOp>(op, adaptor.tensor(),
+                                        adaptor.indices());
+    return success();
+  }
+};
+} // namespace
+
+void mlir::populateTensorBufferizePatterns(
+    MLIRContext *context, BufferizeTypeConverter &typeConverter,
+    OwningRewritePatternList &patterns) {
+  patterns.insert<BufferizeExtractOp>(typeConverter, context);
+}
+
+namespace {
+struct TensorBufferizePass : public TensorBufferizeBase<TensorBufferizePass> {
+  void runOnFunction() override {
+    auto *context = &getContext();
+    BufferizeTypeConverter typeConverter;
+    OwningRewritePatternList patterns;
+    ConversionTarget target(*context);
+
+    populateTensorBufferizePatterns(context, typeConverter, patterns);
+    target.addIllegalOp<tensor::ExtractOp>();
+    target.addLegalDialect<StandardOpsDialect>();
+
+    if (failed(
+            applyPartialConversion(getFunction(), target, std::move(patterns))))
+      signalPassFailure();
+  }
+};
+} // namespace
+
+std::unique_ptr<Pass> mlir::createTensorBufferizePass() {
+  return std::make_unique<TensorBufferizePass>();
+}

mlir/lib/Dialect/Tensor/Transforms/CMakeLists.txt

+add_mlir_dialect_library(MLIRTensorTransforms
+  Bufferize.cpp
+
+  ADDITIONAL_HEADER_DIRS
+  ${MLIR_MAIN_INCLUDE_DIR}/mlir/Dialect/Tensor/Transforms
+
+  DEPENDS
+  MLIRTensorTransformsIncGen
+
+  LINK_LIBS PUBLIC
+  MLIRIR
+  MLIRPass
+  MLIRTensor
+  MLIRTransforms
+  )

mlir/lib/Dialect/Tensor/Transforms/PassDetail.h

+//===- PassDetail.h - GPU Pass class details --------------------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef DIALECT_TENSOR_TRANSFORMS_PASSDETAIL_H_
+#define DIALECT_TENSOR_TRANSFORMS_PASSDETAIL_H_
+
+#include "mlir/Pass/Pass.h"
+
+namespace mlir {
+
+#define GEN_PASS_CLASSES
+#include "mlir/Dialect/Tensor/Transforms/Passes.h.inc"
+
+} // end namespace mlir
+
+#endif // DIALECT_TENSOR_TRANSFORMS_PASSDETAIL_H_

mlir/test/Conversion/ShapeToStandard/convert-shape-constraints.mlir

 // CHECK:           %[[RHS_RANK:.*]] = dim %[[RHS]], %[[C0]] : tensor<?xindex>
 // CHECK:           %[[LHS_RANK_ULE:.*]] = cmpi "ule", %[[LHS_RANK]], %[[RHS_RANK]] : index
 // CHECK:           %[[LESSER_RANK:.*]] = select %[[LHS_RANK_ULE]], %[[LHS_RANK]], %[[RHS_RANK]] : index
 // CHECK:           %[[GREATER_RANK:.*]] = select %[[LHS_RANK_ULE]], %[[RHS_RANK]], %[[LHS_RANK]] : index
 // CHECK:           %[[LESSER_RANK_OPERAND:.*]] = select %[[LHS_RANK_ULE]], %[[LHS]], %[[RHS]] : tensor<?xindex>
 // CHECK:           %[[GREATER_RANK_OPERAND:.*]] = select %[[LHS_RANK_ULE]], %[[RHS]], %[[LHS]] : tensor<?xindex>
 // CHECK:           %[[RANK_DIFF:.*]] = subi %[[GREATER_RANK]], %[[LESSER_RANK]] : index
 // CHECK:           scf.for %[[IV:.*]] = %[[RANK_DIFF]] to %[[GREATER_RANK]] step %[[C1]] {
-// CHECK:             %[[GREATER_RANK_OPERAND_EXTENT:.*]] = extract_element %[[GREATER_RANK_OPERAND]][%[[IV]]] : tensor<?xindex>
+// CHECK:             %[[GREATER_RANK_OPERAND_EXTENT:.*]] = tensor.extract %[[GREATER_RANK_OPERAND]][%[[IV]]] : tensor<?xindex>
 // CHECK:             %[[IVSHIFTED:.*]] = subi %[[IV]], %[[RANK_DIFF]] : index
-// CHECK:             %[[LESSER_RANK_OPERAND_EXTENT:.*]] = extract_element %[[LESSER_RANK_OPERAND]][%[[IVSHIFTED]]] : tensor<?xindex>
+// CHECK:             %[[LESSER_RANK_OPERAND_EXTENT:.*]] = tensor.extract %[[LESSER_RANK_OPERAND]][%[[IVSHIFTED]]] : tensor<?xindex>
 // CHECK:             %[[GREATER_RANK_OPERAND_EXTENT_IS_ONE:.*]] = cmpi "eq", %[[GREATER_RANK_OPERAND_EXTENT]], %[[C1]] : index
 // CHECK:             %[[LESSER_RANK_OPERAND_EXTENT_IS_ONE:.*]] = cmpi "eq", %[[LESSER_RANK_OPERAND_EXTENT]], %[[C1]] : index
 // CHECK:             %[[EXTENTS_AGREE:.*]] = cmpi "eq", %[[GREATER_RANK_OPERAND_EXTENT]], %[[LESSER_RANK_OPERAND_EXTENT]] : index
 // CHECK:             %[[OR_TMP:.*]] = or %[[GREATER_RANK_OPERAND_EXTENT_IS_ONE]], %[[LESSER_RANK_OPERAND_EXTENT_IS_ONE]] : i1
 // CHECK:             %[[BROADCAST_IS_VALID:.*]] = or %[[EXTENTS_AGREE]], %[[OR_TMP]] : i1
 // CHECK:             assert %[[BROADCAST_IS_VALID]], "invalid broadcast"
 // CHECK:           }
 // CHECK:           return %[[RET]] : !shape.witness

mlir/test/Conversion/ShapeToStandard/shape-to-standard.mlir

func @get_extent_shape_of(%arg : tensor<2x3xf32>, %idx : index) -> index {
   // CHECK: return %[[RESULT]] : index
   %shape = shape.shape_of %arg : tensor<2x3xf32> -> tensor<?xindex>
   %result = shape.get_extent %shape, %idx : tensor<?xindex>, index -> index
   return %result : index
 }
 
 // -----
 
-// Express `get_extent` as `std.extract_element`.
+// Express `get_extent` as `std.tensor.extract`.
 // CHECK-LABEL: @get_extent_from_extent_tensor
 // CHECK-SAME: (%[[EXTENTS:.*]]: tensor<?xindex>, %[[IDX:.*]]: index) -> index
 func @get_extent_from_extent_tensor(%extents : tensor<?xindex>, %idx : index)
     -> index {
-  // CHECK: %[[RESULT:.*]] = extract_element %[[EXTENTS]][%[[IDX]]] : tensor<?xindex>
+  // CHECK: %[[RESULT:.*]] = tensor.extract %[[EXTENTS]][%[[IDX]]] : tensor<?xindex>
   // CHECK: return %[[RESULT]] : index
   %result = shape.get_extent %extents, %idx : tensor<?xindex>, index -> index
   return %result : index
 }
 
 // -----
 
 // Lower `const_shape` to `tensor_from_elements`.
func @shape_reduce(%shape : tensor<?xindex>) -> index {
   }
   return %num_elements : index
 }
 // CHECK-NEXT: %[[INIT:.*]] = constant 1 : index
 // CHECK-NEXT: %[[C0:.*]] = constant 0 : index
 // CHECK-NEXT: %[[C1:.*]] = constant 1 : index
 // CHECK-NEXT: %[[RANK:.*]] = dim %[[SHAPE]], %[[C0]] : tensor<?xindex>
 // CHECK-NEXT: %[[RESULT:.*]] = scf.for %[[I:.*]] = %[[C0]] to %[[RANK]] step %[[C1]] iter_args(%[[ACC:.*]] = %[[INIT]]) -> (index)
-// CHECK-NEXT:   %[[EXTENT:.*]] = extract_element %[[SHAPE]][%[[I]]]
+// CHECK-NEXT:   %[[EXTENT:.*]] = tensor.extract %[[SHAPE]][%[[I]]]
 // CHECK-NEXT:   %[[NEW_ACC:.*]] = muli %[[ACC]], %[[EXTENT]] : index
 // CHECK-NEXT:   scf.yield %[[NEW_ACC]] : index
 // CHECK-NEXT: }
 // CHECK-NEXT: return %[[RESULT]] : index
 
 // -----
 
 // Don't lower `shape_of` for result type of `shape.shape`.
func @shape_eq(%a : tensor<?xindex>, %b : tensor<?xindex>) -> i1 {
   // CHECK: %[[C0:.*]] = constant 0 : index
   // CHECK: %[[RANK_A:.*]] = dim %[[A]], %[[C0]] : tensor<?xindex>
   // CHECK: %[[RANK_B:.*]] = dim %[[B]], %[[C0]] : tensor<?xindex>
   // CHECK: %[[RANK_EQ:.*]] = cmpi "eq", %[[RANK_A]], %[[RANK_B]]
   // CHECK: %[[SHAPE_EQ:.*]] = scf.if %[[RANK_EQ]] -> (i1) {
   // CHECK:   %[[C1:.*]] = constant 1 : index
   // CHECK:   %[[INIT:.*]] = constant true
   // CHECK:   %[[SHAPE_EQ_INNER:.*]] = scf.for %[[I:.*]] = %[[C0]] to %[[RANK_A]] step %[[C1]] iter_args(%[[CONJ:.*]] = %[[INIT]]) -> (i1) {
-  // CHECK:     %[[EXTENT_A:.*]] = extract_element %[[A]][%[[I]]] : tensor<?xindex>
-  // CHECK:     %[[EXTENT_B:.*]] = extract_element %[[B]][%[[I]]] : tensor<?xindex>
+  // CHECK:     %[[EXTENT_A:.*]] = tensor.extract %[[A]][%[[I]]] : tensor<?xindex>
+  // CHECK:     %[[EXTENT_B:.*]] = tensor.extract %[[B]][%[[I]]] : tensor<?xindex>
   // CHECK:     %[[EXTENT_EQ:.*]] = cmpi "eq", %[[EXTENT_A]], %[[EXTENT_B]]
   // CHECK:     %[[CONJ_NEXT:.*]] = and %[[CONJ]], %[[EXTENT_EQ]]
   // CHECK:     scf.yield %[[CONJ_NEXT]] : i1
   // CHECK:   }
   // CHECK:   scf.yield %[[SHAPE_EQ_INNER]] : i1
   // CHECK: } else {
   // CHECK:   %[[SHAPE_EQ_INNER:.*]] = constant false
   // CHECK:   scf.yield %[[SHAPE_EQ_INNER]] : i1
func @broadcast_unknown_extents(%a : tensor<?xindex>, %b : tensor<?xindex>) {
   // CHECK:           %[[ERASED_LHS:.*]] = tensor_cast %[[LHS]] : tensor<?xindex> to tensor<?xindex>
   // CHECK:           %[[ERASED_RHS:.*]] = tensor_cast %[[RHS]] : tensor<?xindex> to tensor<?xindex>
   // CHECK:           %[[LESSER_RANK_OPERAND:.*]] = select %[[LHS_RANK_ULE]], %[[ERASED_LHS]], %[[ERASED_RHS]] : tensor<?xindex>
   // CHECK:           %[[GREATER_RANK_OPERAND:.*]] = select %[[LHS_RANK_ULE]], %[[ERASED_RHS]], %[[ERASED_LHS]] : tensor<?xindex>
   // CHECK:           %[[RANK_DIFF:.*]] = subi %[[GREATER_RANK]], %[[LESSER_RANK]] : index
   // CHECK:           %[[RESULT:.*]] = dynamic_tensor_from_elements %[[GREATER_RANK]] {
   // CHECK:           ^bb0(%[[OUTPUT_DIMENSION:.*]]: index):
   // CHECK:             %[[IS_UNCHALLENGED_DIMENSION:.*]] = cmpi "ult", %[[OUTPUT_DIMENSION]], %[[RANK_DIFF]] : index
-  // CHECK:             %[[GREATER_RANK_OPERAND_EXTENT:.*]] = extract_element %[[GREATER_RANK_OPERAND]][%[[OUTPUT_DIMENSION]]] : tensor<?xindex>
+  // CHECK:             %[[GREATER_RANK_OPERAND_EXTENT:.*]] = tensor.extract %[[GREATER_RANK_OPERAND]][%[[OUTPUT_DIMENSION]]] : tensor<?xindex>
   // CHECK:             %[[OUTPUT_EXTENT:.*]] = scf.if %[[IS_UNCHALLENGED_DIMENSION]] -> (index) {
   // CHECK:               scf.yield %[[GREATER_RANK_OPERAND_EXTENT]] : index
   // CHECK:             } else {
   // CHECK:               %[[LESSER_RANK_OPERAND_DIMENSION:.*]] = subi %[[OUTPUT_DIMENSION]], %[[RANK_DIFF]] : index
-  // CHECK:               %[[LESSER_RANK_OPERAND_EXTENT:.*]] = extract_element %[[LESSER_RANK_OPERAND]][%[[LESSER_RANK_OPERAND_DIMENSION]]] : tensor<?xindex>
+  // CHECK:               %[[LESSER_RANK_OPERAND_EXTENT:.*]] = tensor.extract %[[LESSER_RANK_OPERAND]][%[[LESSER_RANK_OPERAND_DIMENSION]]] : tensor<?xindex>
   // CHECK:               %[[GREATER_RANK_OPERAND_EXTENT_IS_ONE:.*]] = cmpi "eq", %[[GREATER_RANK_OPERAND_EXTENT]], %[[C1]] : index
   // CHECK:               %[[BROADCASTED_EXTENT:.*]] = select %[[GREATER_RANK_OPERAND_EXTENT_IS_ONE]], %[[LESSER_RANK_OPERAND_EXTENT]], %[[GREATER_RANK_OPERAND_EXTENT]] : index
   // CHECK:               scf.yield %[[BROADCASTED_EXTENT]] : index
   // CHECK:             }
   // CHECK:             yield %[[OUTPUT_EXTENT:.*]] : index
   // CHECK:           } : tensor<?xindex>
   // CHECK:           return
   // CHECK:         }
func @broadcast_known_different_extents(%a : tensor<2xindex>, %b : tensor<3xindex>) {
   // CHECK:           %[[ERASED_LHS:.*]] = tensor_cast %[[LHS]] : tensor<2xindex> to tensor<?xindex>
   // CHECK:           %[[ERASED_RHS:.*]] = tensor_cast %[[RHS]] : tensor<3xindex> to tensor<?xindex>
   // CHECK:           %[[LESSER_RANK_OPERAND:.*]] = select %[[LHS_RANK_ULE]], %[[ERASED_LHS]], %[[ERASED_RHS]] : tensor<?xindex>
   // CHECK:           %[[GREATER_RANK_OPERAND:.*]] = select %[[LHS_RANK_ULE]], %[[ERASED_RHS]], %[[ERASED_LHS]] : tensor<?xindex>
   // CHECK:           %[[RANK_DIFF:.*]] = subi %[[GREATER_RANK]], %[[LESSER_RANK]] : index
   // CHECK:           %[[RESULT:.*]] = dynamic_tensor_from_elements %[[GREATER_RANK]] {
   // CHECK:           ^bb0(%[[OUTPUT_DIMENSION:.*]]: index):
   // CHECK:             %[[IS_UNCHALLENGED_DIMENSION:.*]] = cmpi "ult", %[[OUTPUT_DIMENSION]], %[[RANK_DIFF]] : index
-  // CHECK:             %[[GREATER_RANK_OPERAND_EXTENT:.*]] = extract_element %[[GREATER_RANK_OPERAND]][%[[OUTPUT_DIMENSION]]] : tensor<?xindex>
+  // CHECK:             %[[GREATER_RANK_OPERAND_EXTENT:.*]] = tensor.extract %[[GREATER_RANK_OPERAND]][%[[OUTPUT_DIMENSION]]] : tensor<?xindex>
   // CHECK:             %[[OUTPUT_EXTENT:.*]] = scf.if %[[IS_UNCHALLENGED_DIMENSION]] -> (index) {
   // CHECK:               scf.yield %[[GREATER_RANK_OPERAND_EXTENT]] : index
   // CHECK:             } else {
   // CHECK:               %[[LESSER_RANK_OPERAND_DIMENSION:.*]] = subi %[[OUTPUT_DIMENSION]], %[[RANK_DIFF]] : index
-  // CHECK:               %[[LESSER_RANK_OPERAND_EXTENT:.*]] = extract_element %[[LESSER_RANK_OPERAND]][%[[LESSER_RANK_OPERAND_DIMENSION]]] : tensor<?xindex>
+  // CHECK:               %[[LESSER_RANK_OPERAND_EXTENT:.*]] = tensor.extract %[[LESSER_RANK_OPERAND]][%[[LESSER_RANK_OPERAND_DIMENSION]]] : tensor<?xindex>
   // CHECK:               %[[GREATER_RANK_OPERAND_EXTENT_IS_ONE:.*]] = cmpi "eq", %[[GREATER_RANK_OPERAND_EXTENT]], %[[C1]] : index
   // CHECK:               %[[BROADCASTED_EXTENT:.*]] = select %[[GREATER_RANK_OPERAND_EXTENT_IS_ONE]], %[[LESSER_RANK_OPERAND_EXTENT]], %[[GREATER_RANK_OPERAND_EXTENT]] : index
   // CHECK:               scf.yield %[[BROADCASTED_EXTENT]] : index
   // CHECK:             }
   // CHECK:             yield %[[OUTPUT_EXTENT:.*]] : index
   // CHECK:           } : tensor<?xindex>
   // CHECK:           return
   // CHECK:         }

 // CHECK:           %[[LARGER_RANK:.*]] = select %[[LHS_SMALLER]], %[[RHS_RANK]], %[[LHS_RANK]] : index
 // CHECK:           %[[RANK_ERASED_LHS:.*]] = tensor_cast %[[LHS]] : tensor<3xindex> to tensor<?xindex>
 // CHECK:           %[[RANK_ERASED_RHS:.*]] = tensor_cast %[[RHS]] : tensor<?xindex> to tensor<?xindex>
 // CHECK:           %[[SMALLER_SHAPE:.*]] = select %[[LHS_SMALLER]], %[[RANK_ERASED_LHS]], %[[RANK_ERASED_RHS]] : tensor<?xindex>
 // CHECK:           %[[LARGER_SHAPE:.*]] = select %[[LHS_SMALLER]], %[[RANK_ERASED_RHS]], %[[RANK_ERASED_LHS]] : tensor<?xindex>
 // CHECK:           %[[RANK_DIFF:.*]] = subi %[[LARGER_RANK]], %[[SMALLER_RANK]] : index
 // CHECK:           %[[TRUE:.*]] = constant true
 // CHECK:           %[[ALL_RESULT:.*]] = scf.for %[[I:.*]] = %[[RANK_DIFF]] to %[[LARGER_RANK]] step %[[C1]] iter_args(%[[ALL_SO_FAR:.*]] = %[[TRUE]]) -> (i1) {
-// CHECK:             %[[LARGER_EXTENT:.*]] = extract_element %[[LARGER_SHAPE]]{{\[}}%[[I]]] : tensor<?xindex>
+// CHECK:             %[[LARGER_EXTENT:.*]] = tensor.extract %[[LARGER_SHAPE]]{{\[}}%[[I]]] : tensor<?xindex>
 // CHECK:             %[[LARGER_EXTENT_IS_ONE:.*]] = cmpi "eq", %[[LARGER_EXTENT]], %[[C1]] : index
 // CHECK:             %[[SMALLER_EXTENT_INDEX:.*]] = subi %[[I]], %[[RANK_DIFF]] : index
-// CHECK:             %[[SMALLER_EXTENT:.*]] = extract_element %[[SMALLER_SHAPE]]{{\[}}%[[SMALLER_EXTENT_INDEX]]] : tensor<?xindex>
+// CHECK:             %[[SMALLER_EXTENT:.*]] = tensor.extract %[[SMALLER_SHAPE]]{{\[}}%[[SMALLER_EXTENT_INDEX]]] : tensor<?xindex>
 // CHECK:             %[[SMALLER_EXTENT_IS_ONE:.*]] = cmpi "eq", %[[SMALLER_EXTENT]], %[[C1]] : index
 // CHECK:             %[[EXTENTS_ARE_EQUAL:.*]] = cmpi "eq", %[[LARGER_EXTENT]], %[[SMALLER_EXTENT]] : index
 // CHECK:             %[[EITHER_EXTENT_IS_ONE:.*]] = or %[[LARGER_EXTENT_IS_ONE]], %[[SMALLER_EXTENT_IS_ONE]] : i1
 // CHECK:             %[[OR_EXTENTS_ARE_EQUAL:.*]] = or %[[EITHER_EXTENT_IS_ONE]], %[[EXTENTS_ARE_EQUAL]] : i1
 // CHECK:             %[[NEW_ALL_SO_FAR:.*]] = and %[[ALL_SO_FAR]], %[[OR_EXTENTS_ARE_EQUAL]] : i1
 // CHECK:             scf.yield %[[NEW_ALL_SO_FAR]] : i1
 // CHECK:           }
 // CHECK:           return %[[ALL_RESULT]] : i1

 // CHECK:           %[[LARGER_RANK:.*]] = select %[[LHS_SMALLER]], %[[RHS_RANK]], %[[LHS_RANK]] : index
 // CHECK:           %[[RANK_ERASED_LHS:.*]] = tensor_cast %[[LHS]] : tensor<?xindex> to tensor<?xindex>
 // CHECK:           %[[RANK_ERASED_RHS:.*]] = tensor_cast %[[RHS]] : tensor<?xindex> to tensor<?xindex>
 // CHECK:           %[[SMALLER_SHAPE:.*]] = select %[[LHS_SMALLER]], %[[RANK_ERASED_LHS]], %[[RANK_ERASED_RHS]] : tensor<?xindex>
 // CHECK:           %[[LARGER_SHAPE:.*]] = select %[[LHS_SMALLER]], %[[RANK_ERASED_RHS]], %[[RANK_ERASED_LHS]] : tensor<?xindex>
 // CHECK:           %[[RANK_DIFF:.*]] = subi %[[LARGER_RANK]], %[[SMALLER_RANK]] : index
 // CHECK:           %[[TRUE:.*]] = constant true
 // CHECK:           %[[ALL_RESULT:.*]] = scf.for %[[VAL_16:.*]] = %[[RANK_DIFF]] to %[[LARGER_RANK]] step %[[C1]] iter_args(%[[ALL_SO_FAR:.*]] = %[[TRUE]]) -> (i1) {
-// CHECK:             %[[LARGER_EXTENT:.*]] = extract_element %[[LARGER_SHAPE]]{{\[}}%[[VAL_16]]] : tensor<?xindex>
+// CHECK:             %[[LARGER_EXTENT:.*]] = tensor.extract %[[LARGER_SHAPE]]{{\[}}%[[VAL_16]]] : tensor<?xindex>
 // CHECK:             %[[LARGER_EXTENT_IS_ONE:.*]] = cmpi "eq", %[[LARGER_EXTENT]], %[[C1]] : index
 // CHECK:             %[[LHS_EXTENT_INDEX:.*]] = subi %[[VAL_16]], %[[RANK_DIFF]] : index
-// CHECK:             %[[SMALLER_EXTENT:.*]] = extract_element %[[SMALLER_SHAPE]]{{\[}}%[[LHS_EXTENT_INDEX]]] : tensor<?xindex>
+// CHECK:             %[[SMALLER_EXTENT:.*]] = tensor.extract %[[SMALLER_SHAPE]]{{\[}}%[[LHS_EXTENT_INDEX]]] : tensor<?xindex>
 // CHECK:             %[[SMALLER_EXTENT_IS_ONE:.*]] = cmpi "eq", %[[SMALLER_EXTENT]], %[[C1]] : index
 // CHECK:             %[[EXTENTS_ARE_EQUAL:.*]] = cmpi "eq", %[[LARGER_EXTENT]], %[[SMALLER_EXTENT]] : index
 // CHECK:             %[[EITHER_EXTENT_IS_ONE:.*]] = or %[[LARGER_EXTENT_IS_ONE]], %[[SMALLER_EXTENT_IS_ONE]] : i1
 // CHECK:             %[[OR_EXTENTS_ARE_EQUAL:.*]] = or %[[EITHER_EXTENT_IS_ONE]], %[[EXTENTS_ARE_EQUAL]] : i1
 // CHECK:             %[[NEW_ALL_SO_FAR:.*]] = and %[[ALL_SO_FAR]], %[[OR_EXTENTS_ARE_EQUAL]] : i1
 // CHECK:             scf.yield %[[NEW_ALL_SO_FAR]] : i1
 // CHECK:           }
 // CHECK:           %[[RESULT:.*]] = shape.cstr_require %[[ALL_RESULT]], "required broadcastable shapes"
 // CHECK:           return %[[RESULT]] : !shape.witness
 // CHECK:         }

mlir/test/Dialect/Linalg/fusion-tensor.mlir

func @scalar_indexed_generic_fusion
   %c0 = constant 0 : index  
   %cst = constant dense<1.000000e+00> : tensor<10xf32>
   %0 = linalg.indexed_generic
     {indexing_maps = [affine_map<() -> ()>, affine_map<() -> ()>],
      iterator_types = []}
     ins(%arg1 : tensor<i32>) {
     ^bb0(%arg2: i32):  // no predecessors
       %3 = index_cast %arg2 : i32 to index
-      %4 = extract_element %arg0[%3, %c0, %c0] : tensor<5x1x1xf32>
+      %4 = tensor.extract %arg0[%3, %c0, %c0] : tensor<5x1x1xf32>
       linalg.yield %4 : f32
     } -> tensor<f32>
   %1 = linalg.generic
    {indexing_maps = [affine_map<(d0) -> ()>, affine_map<(d0) -> (d0)>,
                      affine_map<(d0) -> (d0)>],
     iterator_types = ["parallel"]}
     ins(%0, %cst : tensor<f32>, tensor<10xf32>) {
     ^bb0(%arg2: f32, %arg3: f32):  // no predecessors
       %3 = mulf %arg2, %arg3 : f32
       linalg.yield %3 : f32
     } -> tensor<10xf32>
   return %1 : tensor<10xf32>
 }
 //   CHECK-DAG: #[[MAP0:.+]] = affine_map<(d0) -> ()>
 //   CHECK-DAG: #[[MAP1:.+]] = affine_map<(d0) -> (d0)>
 //       CHECK: func @scalar_indexed_generic_fusion
 //  CHECK-SAME:   %[[ARG0:[a-zA-Z0-9]+]]: tensor<5x1x1xf32>
 //  CHECK-SAME:   %[[ARG1:[a-zA-Z0-9]+]]: tensor<i32>
 //       CHECK:   %[[T0:.+]] = linalg.indexed_generic
 //  CHECK-SAME:     indexing_maps = [#[[MAP0]], #[[MAP1]]]
 //  CHECK-SAME:     iterator_types = ["parallel"]
 //  CHECK-SAME:     ins(%[[ARG1]] : tensor<i32>)
-//       CHECK:     extract_element %[[ARG0]]
+//       CHECK:     tensor.extract %[[ARG0]]
 //       CHECK:     linalg.yield
 //       CHECK   return %[[T0]]

mlir/test/Dialect/Standard/bufferize.mlir

func @dynamic_tensor_from_elements_static_and_dynamic(%arg0: index) -> tensor<16x?xindex> {
   %result = dynamic_tensor_from_elements %arg0 {
   ^bb0(%i: index, %j: index):
     %sum = addi %i, %j : index
     yield %sum : index
   } : tensor<16x?xindex>
   return %result : tensor<16x?xindex>
 }
 
-// CHECK-LABEL:   func @extract_element(
-// CHECK-SAME:                          %[[TENSOR:.*]]: tensor<?xf32>,
-// CHECK-SAME:                          %[[IDX:.*]]: index) -> f32 {
-// CHECK:           %[[MEMREF:.*]] = tensor_to_memref %[[TENSOR]] : memref<?xf32>
-// CHECK:           %[[RET:.*]] = load %[[MEMREF]][%[[IDX]]] : memref<?xf32>
-// CHECK:           return %[[RET]] : f32
-// CHECK:         }
-func @extract_element(%arg0: tensor<?xf32>, %arg1: index) -> f32 {
-  %0 = extract_element %arg0[%arg1] : tensor<?xf32>
-  return %0 : f32
-}
-
 // CHECK-LABEL:   func @select(
 // CHECK-SAME:                 %[[PRED:.*]]: i1,
 // CHECK-SAME:                 %[[TRUE_VAL:.*]]: tensor<f32>,
 // CHECK-SAME:                 %[[FALSE_VAL:.*]]: tensor<f32>) -> tensor<f32> {
 // CHECK:           %[[TRUE_VAL_MEMREF:.*]] = tensor_to_memref %[[TRUE_VAL]] : memref<f32>
 // CHECK:           %[[FALSE_VAL_MEMREF:.*]] = tensor_to_memref %[[FALSE_VAL]] : memref<f32>
 // CHECK:           %[[RET_MEMREF:.*]] = select %[[PRED]], %[[TRUE_VAL_MEMREF]], %[[FALSE_VAL_MEMREF]] : memref<f32>
 // CHECK:           %[[RET:.*]] = tensor_load %[[RET_MEMREF]] : memref<f32>

 func @tensor_from_elements(%arg0: index, %arg1: index) -> tensor<2xindex> {
   %0 = tensor_from_elements %arg0, %arg1 : tensor<2xindex>
   return %0 : tensor<2xindex>
 }
 
 // The dynamic_tensor_from_elements op clones each op in its body.
 // Make sure that regions nested within such ops are recursively converted.
 // CHECK-LABEL: func @recursively_convert_cloned_regions
-func @recursively_convert_cloned_regions(%arg0: tensor<?xindex>, %arg1: index, %arg2: i1) -> tensor<?xindex> {
+func @recursively_convert_cloned_regions(%arg0: tensor<*xf32>, %arg1: index, %arg2: i1) -> tensor<?xindex> {
   %tensor = dynamic_tensor_from_elements %arg1 {
   ^bb0(%iv: index):
     %48 = scf.if %arg2 -> (index) {
       scf.yield %iv : index
     } else {
-      // CHECK-NOT: extract_element
-      %50 = extract_element %arg0[%iv] : tensor<?xindex>
+      // CHECK-NOT: dim{{.*}}tensor
+      %50 = dim %arg0, %iv : tensor<*xf32>
       scf.yield %50 : index
     }
     yield %48 : index
   } : tensor<?xindex>
   return %tensor : tensor<?xindex>
 }

mlir/test/Dialect/Standard/canonicalize.mlir

 func @dim_of_tensor_load(%arg0: memref<?xf32>) -> index {
   %c0 = constant 0 : index
   %0 = tensor_load %arg0 : memref<?xf32>
   %1 = dim %0, %c0 : tensor<?xf32>
   return %1 : index
 }
 
 // Test case: Folding of load(tensor_to_memref(%v, %idxs))
-//            -> extract_element(%v, %idx)
+//            -> tensor.extract(%v, %idx)
 // CHECK-LABEL: func @load_from_tensor_to_memref(
 //  CHECK-SAME:     %[[IDX0:[0-9a-z]+]]: index, %[[IDX1:[0-9a-z]+]]: index
 //  CHECK-SAME:     %[[TENSOR:[0-9a-z]+]]: tensor<?x?xf32>
-//       CHECK:   %[[RES:.*]] = extract_element %[[TENSOR]][%[[IDX0]], %[[IDX1]]]
+//       CHECK:   %[[RES:.*]] = tensor.extract %[[TENSOR]][%[[IDX0]], %[[IDX1]]]
 //   CHECK-NOT:   load
 //       CHECK:   return %[[RES]] : f32
 func @load_from_tensor_to_memref(%arg0: index, %arg1: index, %arg2: tensor<?x?xf32>) -> f32 {
   %0 = tensor_to_memref %arg2 : memref<?x?xf32>
   %1 = load %0[%arg0, %arg1] : memref<?x?xf32>
   return %1 : f32
 }
 

mlir/test/Dialect/Tensor/bufferize.mlir

+// RUN: mlir-opt %s -tensor-bufferize | FileCheck %s
+
+// CHECK-LABEL:   func @extract(
+// CHECK-SAME:                  %[[TENSOR:.*]]: tensor<?xf32>,
+// CHECK-SAME:                  %[[IDX:.*]]: index) -> f32 {
+// CHECK:           %[[MEMREF:.*]] = tensor_to_memref %[[TENSOR]] : memref<?xf32>
+// CHECK:           %[[RET:.*]] = load %[[MEMREF]][%[[IDX]]] : memref<?xf32>
+// CHECK:           return %[[RET]] : f32
+// CHECK:         }
+func @extract(%arg0: tensor<?xf32>, %arg1: index) -> f32 {
+  %0 = tensor.extract %arg0[%arg1] : tensor<?xf32>
+  return %0 : f32
+}

mlir/test/Dialect/Tensor/canonicalize.mlir

+// RUN: mlir-opt %s -canonicalize | FileCheck %s
+
+// -----
+
+// CHECK-LABEL: func @fold_extract
+func @fold_extract(%arg0 : index) -> (f32, f16, f16, i32) {
+  %const_0 = constant 0 : index
+  %const_1 = constant 1 : index
+  %const_3 = constant 3 : index
+
+  // Fold an extract into a splat.
+  // CHECK-NEXT: [[C4:%.+]] = constant 4.{{0*}}e+00 : f32
+  %0 = constant dense<4.0> : tensor<4xf32>
+  %ext_1 = tensor.extract %0[%arg0] : tensor<4xf32>
+
+  // Fold an extract into a sparse with a sparse index.
+  // CHECK-NEXT: [[CM2:%.+]] = constant -2.{{0*}}e+00 : f16
+  %1 = constant sparse<[[0, 0, 0], [1, 1, 1]],  [-5.0, -2.0]> : tensor<4x4x4xf16>
+  %ext_2 = tensor.extract %1[%const_1, %const_1, %const_1] : tensor<4x4x4xf16>
+
+  // Fold an extract into a sparse with a non sparse index.
+  // CHECK-NEXT: [[C0:%.+]] = constant 0.{{0*}}e+00 : f16
+  %2 = constant sparse<[[1, 1, 1]],  [-2.0]> : tensor<1x1x1xf16>
+  %ext_3 = tensor.extract %2[%const_0, %const_0, %const_0] : tensor<1x1x1xf16>
+
+  // Fold an extract into a dense tensor.
+  // CHECK-NEXT: [[C64:%.+]] = constant 64 : i32
+  %3 = constant dense<[[[1, -2, 1, 36]], [[0, 2, -1, 64]]]> : tensor<2x1x4xi32>
+  %ext_4 = tensor.extract %3[%const_1, %const_0, %const_3] : tensor<2x1x4xi32>
+
+  // CHECK-NEXT: return [[C4]], [[CM2]], [[C0]], [[C64]]
+  return %ext_1, %ext_2, %ext_3, %ext_4 : f32, f16, f16, i32
+}

mlir/test/Dialect/Tensor/invalid.mlir

+// RUN: mlir-opt <%s -verify-diagnostics
+
+// -----
+
+func @extract_too_many_indices(%arg0: tensor<?xf32>) {
+  // expected-error@+1 {{incorrect number of indices for extract_element}}
+  %0 = tensor.extract %arg0[] : tensor<?xf32>
+  return
+}

mlir/test/Dialect/Tensor/ops.mlir

+// RUN: mlir-opt <%s | mlir-opt | FileCheck %s
+
+// CHECK-LABEL:   func @extract(
+// CHECK-SAME:                  %[[TENSOR:.*]]: tensor<?x?x?xf32>,
+// CHECK-SAME:                  %[[INDEX:.*]]: index) {
+func @extract(%arg0: tensor<?x?x?xf32>, %arg1: index) {
+  // CHECK: tensor.extract %[[TENSOR]][%[[INDEX]], %[[INDEX]], %[[INDEX]]] : tensor<?x?x?xf32>
+  %0 = tensor.extract %arg0[%arg1, %arg1, %arg1] : tensor<?x?x?xf32>
+  return
+}

mlir/test/IR/core-ops.mlir

func @calls(%arg0: i32) {
   %2 = call_indirect %f_0(%arg0) : (i32) -> i32
 
   // CHECK: %4 = call_indirect %f_0(%arg0) : (i32) -> i32
   %3 = "std.call_indirect"(%f_0, %arg0) : ((i32) -> i32, i32) -> i32
 
   return
 }
 
-// CHECK-LABEL: func @extract_element(%arg0: tensor<*xi32>, %arg1: tensor<4x4xf32>) -> i32 {
-func @extract_element(%arg0: tensor<*xi32>, %arg1 : tensor<4x4xf32>) -> i32 {
-  %c0 = "std.constant"() {value = 0: index} : () -> index
-
-  // CHECK: %0 = extract_element %arg0[%c0, %c0, %c0, %c0] : tensor<*xi32>
-  %0 = extract_element %arg0[%c0, %c0, %c0, %c0] : tensor<*xi32>
-
-  // CHECK: %1 = extract_element %arg1[%c0, %c0] : tensor<4x4xf32>
-  %1 = extract_element %arg1[%c0, %c0] : tensor<4x4xf32>
-
-  return %0 : i32
-}
-
 // CHECK-LABEL: func @tensor_from_elements() {
 func @tensor_from_elements() {
   %c0 = "std.constant"() {value = 0: index} : () -> index
   // CHECK: %0 = tensor_from_elements %c0 : tensor<1xindex>
   %0 = tensor_from_elements %c0 : tensor<1xindex>
 
   %c1 = "std.constant"() {value = 1: index} : () -> index
   // CHECK: %1 = tensor_from_elements %c0, %c1 : tensor<2xindex>
func @subtensor_insert(%t: tensor<8x16x4xf32>, %t2: tensor<16x32x8xf32>, %idx : index) {
   %2 = subtensor_insert %t into %t2[%c0, %idx, %c0][%idx, 4, %idx][%c1, 1, %c1]
     : tensor<8x16x4xf32> into tensor<16x32x8xf32>
 
   return
 }
 
 // CHECK-LABEL: func private @legacy_visibility_syntax
 func @legacy_visibility_syntax() attributes { sym_visibility = "private" }
-

mlir/test/IR/invalid-ops.mlir

 
 func @cmpf_canonical_type_mismatch(%a : f32, %b : f64) { // expected-note {{prior use here}}
   // expected-error@+1 {{use of value '%b' expects different type than prior uses}}
   %r = cmpf "oeq", %a, %b : f32
 }
 
 // -----
 
-func @extract_element_no_operands() {
-  // expected-error@+1 {{op expected 1 or more operands}}
-  %0 = "std.extract_element"() : () -> f32
-  return
-}
-
-// -----
-
-func @extract_element_no_indices(%v : vector<3xf32>) {
-  // expected-error@+1 {{incorrect number of indices for extract_element}}
-  %0 = "std.extract_element"(%v) : (vector<3xf32>) -> f32
-  return
-}
-
-// -----
-
-func @extract_element_invalid_index_type(%v : vector<3xf32>, %i : i32) {
-  // expected-error@+1 {{operand #1 must be index}}
-  %0 = "std.extract_element"(%v, %i) : (vector<3xf32>, i32) -> f32
-  return
-}
-
-// -----
-
-func @extract_element_element_result_type_mismatch(%v : vector<3xf32>, %i : index) {
-  // expected-error@+1 {{result type matches element type of aggregate}}
-  %0 = "std.extract_element"(%v, %i) : (vector<3xf32>, index) -> f64
-  return
-}
-
-// -----
-
-func @extract_element_vector_too_many_indices(%v : vector<3xf32>, %i : index) {
-  // expected-error@+1 {{incorrect number of indices for extract_element}}
-  %0 = "std.extract_element"(%v, %i, %i) : (vector<3xf32>, index, index) -> f32
-  return
-}
-
-// -----
-
-func @extract_element_tensor_too_many_indices(%t : tensor<2x3xf32>, %i : index) {
-  // expected-error@+1 {{incorrect number of indices for extract_element}}
-  %0 = "std.extract_element"(%t, %i, %i, %i) : (tensor<2x3xf32>, index, index, index) -> f32
-  return
-}
-
-// -----
-
-func @extract_element_tensor_too_few_indices(%t : tensor<2x3xf32>, %i : index) {
-  // expected-error@+1 {{incorrect number of indices for extract_element}}
-  %0 = "std.extract_element"(%t, %i) : (tensor<2x3xf32>, index) -> f32 return
-}
-
-// -----
-
 func @tensor_from_elements_wrong_result_type() {
   // expected-error@+2 {{'result' must be 1D tensor of any type values, but got 'tensor<*xi32>'}}
   %c0 = constant 0 : i32
   %0 = tensor_from_elements %c0 : tensor<*xi32>
   return
 }
 
 // -----

mlir/test/Transforms/canonicalize.mlir

func @memref_cast_folding_subview_static(%V: memref<16x16xf32>, %a: index, %b: index)
 
   // CHECK:  subview{{.*}}: memref<16x16xf32> to memref<3x4xf32, #[[$map0]]>
   // CHECK:  memref_cast{{.*}}: memref<3x4xf32, #[[$map0]]> to memref<3x4xf32, #[[$map1]]>
   return %1: memref<3x4xf32, offset:?, strides:[?, 1]>
 }
 
 // -----
 
-// CHECK-LABEL: func @extract_element_from_tensor_from_elements
-func @extract_element_from_tensor_from_elements(%element : index) -> index {
+// CHECK-LABEL: func @extract_from_tensor_from_elements
+func @extract_from_tensor_from_elements(%element : index) -> index {
   // CHECK-SAME: ([[ARG:%.*]]: index)
   %c0 = constant 0 : index
   %tensor = tensor_from_elements %element : tensor<1xindex>
-  %extracted_element = extract_element %tensor[%c0] : tensor<1xindex>
+  %extracted_element = tensor.extract %tensor[%c0] : tensor<1xindex>
   // CHECK: [[ARG]] : index
   return %extracted_element : index
 }
 
 // -----
 
-// CHECK-LABEL: func @extract_element_from_dynamic_tensor_from_elements
+// CHECK-LABEL: func @extract_from_dynamic_tensor_from_elements
 // CHECK-SAME: %[[IDX:.*]]: index, %[[TENSOR:.*]]: tensor<*xf32>
-func @extract_element_from_dynamic_tensor_from_elements(%idx: index, %tensor: tensor<*xf32>) -> index {
+func @extract_from_dynamic_tensor_from_elements(%idx: index, %tensor: tensor<*xf32>) -> index {
   %size = rank %tensor : tensor<*xf32>
   // CHECK-NEXT: %[[RES:.*]] = dim %[[TENSOR]], %[[IDX]]
   %0 = dynamic_tensor_from_elements %size {
     ^bb0(%arg0: index):
     %1 = dim %tensor, %arg0 : tensor<*xf32>
     yield %1 : index
   } : tensor<?xindex>
-  %1 = extract_element %0[%idx] : tensor<?xindex>
+  %1 = tensor.extract %0[%idx] : tensor<?xindex>
   // CHECK-NEXT: return %[[RES]]
   return %1 : index
 }
 
 // -----
 
-// CHECK-LABEL: func @extract_element_from_dynamic_tensor_from_elements_2d
+// CHECK-LABEL: func @extract_from_dynamic_tensor_from_elements_2d
 // CHECK-SAME: %[[IDX0:.*]]: index, %[[IDX1:.*]]: index, %[[TENSOR:.*]]: tensor<*xf32>
-func @extract_element_from_dynamic_tensor_from_elements_2d(%idx0: index, %idx1: index, %tensor: tensor<*xf32>) -> index {
+func @extract_from_dynamic_tensor_from_elements_2d(%idx0: index, %idx1: index, %tensor: tensor<*xf32>) -> index {
   %size = rank %tensor : tensor<*xf32>
   // CHECK-NEXT: %[[DIM0:.*]] = dim %[[TENSOR]], %[[IDX0]]
   // CHECK-NEXT: %[[DIM1:.*]] = dim %[[TENSOR]], %[[IDX1]]
   // CHECK-NEXT: %[[RES:.*]] = addi %[[DIM0]], %[[DIM1]]
   %0 = dynamic_tensor_from_elements %size, %size {
     ^bb0(%arg0: index, %arg1: index):
     %1 = dim %tensor, %arg0 : tensor<*xf32>
     %2 = dim %tensor, %arg1 : tensor<*xf32>
     %3 = addi %1, %2 : index
     yield %3 : index
   } : tensor<?x?xindex>
-  %4 = extract_element %0[%idx0, %idx1] : tensor<?x?xindex>
+  %4 = tensor.extract %0[%idx0, %idx1] : tensor<?x?xindex>
   // CHECK-NEXT: return %[[RES]]
   return %4 : index
 }
 
 // -----
 
-// CHECK-LABEL: func @extract_element_from_dynamic_tensor_from_elements_sideeffects
+// CHECK-LABEL: func @extract_from_dynamic_tensor_from_elements_sideeffects
 // CHECK-SAME: %[[IDX:.*]]: index
-func @extract_element_from_dynamic_tensor_from_elements_sideeffects(%idx: index, %tensor: tensor<*xf32>) -> index {
+func @extract_from_dynamic_tensor_from_elements_sideeffects(%idx: index, %tensor: tensor<*xf32>) -> index {
   %size = rank %tensor : tensor<*xf32>
   %mem = alloc(%size) : memref<?xindex>
   // CHECK: %[[DTENSOR:.*]] = dynamic_tensor_from_elements
   %0 = dynamic_tensor_from_elements %size {
     ^bb0(%arg0: index):
     %1 = dim %tensor, %arg0 : tensor<*xf32>
     store %1, %mem[%arg0] : memref<?xindex>
     yield %1 : index
   } : tensor<?xindex>
-  // CHECK: %[[RES:.*]] = extract_element %[[DTENSOR]][%[[IDX]]]
-  %1 = extract_element %0[%idx] : tensor<?xindex>
+  // CHECK: %[[RES:.*]] = tensor.extract %[[DTENSOR]][%[[IDX]]]
+  %1 = tensor.extract %0[%idx] : tensor<?xindex>
   // CHECK-NEXT: return %[[RES]]
   return %1 : index
 }
 
 // -----
 
 // CHECK-LABEL: @static_dynamic_tensor_from_elements
 // CHECK-SAME: %[[SIZE1:.*]]: index, %[[SIZE4:.*]]: index)
func @subtensor(%t: tensor<8x16x4xf32>, %arg0 : index, %arg1 : index)
   %2 = subtensor %1[%c0, %c0, %c0] [%c2, %arg0, %c2] [%c1, %c1, %c1]
     : tensor<?x?x?xf32> to tensor<?x?x?xf32>
 
   return %2 : tensor<?x?x?xf32>
 }
 
 // -----
 
-// CHECK-LABEL: func @extract_element_from_tensor_cast
+// CHECK-LABEL: func @extract_from_tensor_cast
 // CHECK-SAME: %[[TENSOR:.*]]: tensor<*xf32>
-func @extract_element_from_tensor_cast(%tensor: tensor<*xf32>) -> f32 {
+func @extract_from_tensor_cast(%tensor: tensor<*xf32>) -> f32 {
   // CHECK-NEXT: %[[C0:.*]] = constant 0 : index
   %c0 = constant 0 : index
   // CHECK-NOT: tensor_cast
   %casted = tensor_cast %tensor : tensor<*xf32> to tensor<?xf32>
-  // CHECK-NEXT: extract_element %[[TENSOR]][%[[C0]]]
-  %result = extract_element %casted[%c0] : tensor<?xf32>
+  // CHECK-NEXT: tensor.extract %[[TENSOR]][%[[C0]]]
+  %result = tensor.extract %casted[%c0] : tensor<?xf32>
   return %result : f32
 }

mlir/test/Transforms/constant-fold.mlir

func @cmpf_inf() -> (i1, i1, i1, i1, i1, i1, i1, i1, i1, i1, i1, i1, i1, i1, i1, i1) {
   %14 = cmpf "uno", %c42, %cpinf: f32
   // CHECK-SAME: [[T]]
   %15 = cmpf "true", %c42, %cpinf: f32
   return %0, %1, %2, %3, %4, %5, %6, %7, %8, %9, %10, %11, %12, %13, %14, %15 : i1, i1, i1, i1, i1, i1, i1, i1, i1, i1, i1, i1, i1, i1, i1, i1
 }
 
 // -----
 
-// CHECK-LABEL: func @fold_extract_element
-func @fold_extract_element(%arg0 : index) -> (f32, f16, f16, i32) {
-  %const_0 = constant 0 : index
-  %const_1 = constant 1 : index
-  %const_3 = constant 3 : index
-
-  // Fold an extract into a splat.
-  // CHECK-NEXT: [[C4:%.+]] = constant 4.{{0*}}e+00 : f32
-  %0 = constant dense<4.0> : tensor<4xf32>
-  %ext_1 = extract_element %0[%arg0] : tensor<4xf32>
-
-  // Fold an extract into a sparse with a sparse index.
-  // CHECK-NEXT: [[CM2:%.+]] = constant -2.{{0*}}e+00 : f16
-  %1 = constant sparse<[[0, 0, 0], [1, 1, 1]],  [-5.0, -2.0]> : vector<4x4x4xf16>
-  %ext_2 = extract_element %1[%const_1, %const_1, %const_1] : vector<4x4x4xf16>
-
-  // Fold an extract into a sparse with a non sparse index.
-  // CHECK-NEXT: [[C0:%.+]] = constant 0.{{0*}}e+00 : f16
-  %2 = constant sparse<[[1, 1, 1]],  [-2.0]> : vector<1x1x1xf16>
-  %ext_3 = extract_element %2[%const_0, %const_0, %const_0] : vector<1x1x1xf16>
-
-  // Fold an extract into a dense tensor.
-  // CHECK-NEXT: [[C64:%.+]] = constant 64 : i32
-  %3 = constant dense<[[[1, -2, 1, 36]], [[0, 2, -1, 64]]]> : tensor<2x1x4xi32>
-  %ext_4 = extract_element %3[%const_1, %const_0, %const_3] : tensor<2x1x4xi32>
-
-  // CHECK-NEXT: return [[C4]], [[CM2]], [[C0]], [[C64]]
-  return %ext_1, %ext_2, %ext_3, %ext_4 : f32, f16, f16, i32
-}
-
-// -----
-
 // CHECK-LABEL: func @fold_rank
 func @fold_rank() -> (index) {
   %const_0 = constant dense<[[[1, -2, 1, 36]], [[0, 2, -1, 64]]]> : tensor<2x1x4xi32>
 
   // Fold a rank into a constant
   // CHECK-NEXT: [[C3:%.+]] = constant 3 : index
   %rank_0 = rank %const_0 : tensor<2x1x4xi32>