[mlir][bufferization] Move ModuleBufferization to bufferization dialect

[springerm]

Summary

• Move Module Bufferization to the bufferization dialect. The implementation is split into OneShotModuleBufferize.cpp and FuncBufferizableOpInterfaceImpl.cpp, so that the external model implementation can be easily moved to the func dialect in the future.
• Split and clean up test cases. A few test cases are still remaining in Linalg and will be updated separately.
• linalg.inplaceable is renamed to bufferization.writable to accurately reflect its current usage.
• Attributes and their verifiers are moved from the Linalg dialect to the Bufferization dialect.
• Expand documentation.
• Add a new flag to One-Shot Bufferize to allow for function boundary bufferization.

Depends On D122228

[springerm]

rebase

[springerm]

fix test case

[springerm]

update

[springerm]

update

[springerm]

update

[springerm]

rebase

[springerm]

rebase

[springerm]

rebase

mlir/include/mlir/Dialect/Bufferization/IR/BufferizableOpInterface.td

bool bufferizesToAliasOnly(OpOperand &opOperand,
                                const AnalysisState &state) {
       auto bufferizableOp =
           cast<BufferizableOpInterface>(getOperation());
       return !bufferizableOp.bufferizesToMemoryRead(opOperand, state)
           && !bufferizableOp.bufferizesToMemoryWrite(opOperand, state)
           && !bufferizableOp.getAliasingOpResult(opOperand, state).empty();
     }
 
-    // TODO: The following two attributes should belong to the tensor dialect.
-    // The corresponding verifier should also be in the tensor dialect.
+    // TODO: This attribute is deprecated. Use `bufferization.writable` or add
+    // a new attribute in a different dialect.
     /// Attribute name used to mark region arguments that can be bufferized
     /// in-place during one-shot bufferization.
     constexpr const static ::llvm::StringLiteral
         kInplaceableAttrName = "linalg.inplaceable";
-
-    /// Attribute name used to mark the bufferization layout for region
-    /// arguments during one-shot bufferization.
-    constexpr const static ::llvm::StringLiteral
-      kBufferLayoutAttrName = "linalg.buffer_layout";
   }];
 }
 
 #endif  // BUFFERIZABLE_OP_INTERFACE

mlir/include/mlir/Dialect/Bufferization/IR/BufferizationBase.td

let description = [{
     specific to the bufferization passes.
 
     Overview of the bufferization infrastructure and important conceptual
     details related to using the MLIR dialect conversion infrastructure can be
     found in [bufferization](/docs/Bufferization/) and [buffer
     deallocation](/docs/BufferDeallocationInternals/).
   }];
   let dependentDialects = ["memref::MemRefDialect", "tensor::TensorDialect"];
+
+  let extraClassDeclaration = [{
+    /// An attribute that can override writability of buffers of tensor function
+    /// arguments during One-Shot Module Bufferize.
+    constexpr const static ::llvm::StringLiteral
+        kWritableAttrName = "bufferization.writable";
+
+    /// Attribute name used to mark the bufferization layout for region
+    /// arguments during One-Shot Module Bufferize.
+    constexpr const static ::llvm::StringLiteral
+        kBufferLayoutAttrName = "bufferization.buffer_layout";
+  }];
+  let hasOperationAttrVerify = 1;
 }
 
 #endif // BUFFERIZATION_BASE

mlir/include/mlir/Dialect/Bufferization/Transforms/FuncBufferizableOpInterfaceImpl.h

+//===- BufferizableOpInterfaceImpl.h - Impl. of BufferizableOpInterface ---===//
+//
+// 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_BUFFERIZATION_TRANSFORMS_FUNCBUFFERIZABLEOPINTERFACEIMPL_H
+#define MLIR_BUFFERIZATION_TRANSFORMS_FUNCBUFFERIZABLEOPINTERFACEIMPL_H
+
+#include "mlir/Dialect/Bufferization/IR/BufferizableOpInterface.h"
+
+namespace mlir {
+class DialectRegistry;
+
+namespace func {
+class FuncOp;
+} // namespace func
+
+namespace bufferization {
+namespace func_ext {
+/// The state of analysis of a FuncOp.
+enum class FuncOpAnalysisState { NotAnalyzed, InProgress, Analyzed };
+
+using func::FuncOp;
+
+/// Extra analysis state that is required for bufferization of function
+/// boundaries.
+struct FuncAnalysisState : public DialectAnalysisState {
+  /// A mapping of ReturnOp OpOperand indices to equivalent FuncOp BBArg
+  /// indices.
+  DenseMap<FuncOp, DenseMap<int64_t, int64_t>> equivalentFuncArgs;
+
+  /// A mapping of ReturnOp OpOperand indices to aliasing FuncOp BBArg indices.
+  DenseMap<FuncOp, DenseMap<int64_t, SmallVector<int64_t>>> aliasingFuncArgs;
+
+  /// A mapping of FuncOp BBArg indices to aliasing ReturnOp OpOperand indices.
+  DenseMap<FuncOp, DenseMap<int64_t, SmallVector<int64_t>>> aliasingReturnVals;
+
+  /// A set of all read BlockArguments of FuncOps.
+  DenseMap<FuncOp, DenseSet<int64_t>> readBbArgs;
+
+  /// A set of all written-to BlockArguments of FuncOps.
+  DenseMap<FuncOp, DenseSet<int64_t>> writtenBbArgs;
+
+  /// Keep track of which FuncOps are fully analyzed or currently being
+  /// analyzed.
+  DenseMap<FuncOp, FuncOpAnalysisState> analyzedFuncOps;
+};
+
+void registerBufferizableOpInterfaceExternalModels(DialectRegistry &registry);
+} // namespace func_ext
+} // namespace bufferization
+} // namespace mlir
+
+#endif // MLIR_BUFFERIZATION_TRANSFORMS_FUNCBUFFERIZABLEOPINTERFACEIMPL_H

mlir/include/mlir/Dialect/Bufferization/Transforms/OneShotModuleBufferize.h

+//===- OneShotModuleBufferize.h - Bufferization across Func. Boundaries ---===//
+//
+// 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_BUFFERIZATION_TRANSFORMS_ONESHOTMODULEBUFFERIZE_H
+#define MLIR_DIALECT_BUFFERIZATION_TRANSFORMS_ONESHOTMODULEBUFFERIZE_H
+
+namespace mlir {
+
+struct LogicalResult;
+class ModuleOp;
+
+namespace bufferization {
+struct OneShotBufferizationOptions;
+
+/// Run One-Shot Module Bufferization on the given module. Performs a simple
+/// function call analysis to determine which function arguments are
+/// inplaceable. Then analyzes and bufferizes FuncOps one-by-one with One-Shot
+/// Bufferize.
+LogicalResult
+runOneShotModuleBufferize(ModuleOp moduleOp,
+                          bufferization::OneShotBufferizationOptions options);
+
+} // namespace bufferization
+} // namespace mlir
+
+#endif // MLIR_DIALECT_BUFFERIZATION_TRANSFORMS_ONESHOTMODULEBUFFERIZE_H

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

let description = [{
     filter and `allow-unknown-ops` is not turned on, bufferization would fail
     when encountering an op that is not included in the filter (even if it is
     bufferizable).
 
     For testing/debugging purposes, `test-analysis-only=1 print-conflicts=1`
     prints analysis results and explains why an OpOperand was decided to
     bufferize out-of-place. This is useful for understanding why One-Shot
     Bufferize chose to insert a certain buffer copy.
+
+    `bufferize-function-boundaries` is an experimental flag for bufferizing
+    `FuncOp`, `ReturnOp` and `CallOp`. This feature is still under development
+    and supports only simple cases at the moment. In particular:
+
+    * Recursive or circular function call graphs are not supported.
+    * If a newly allocated buffer is returned from a function (with
+      `allow-return-allocs`), the buffer will never be deallocated and leak.
+      Such IR needs special handling, e.g., allocation hoisting or reference
+      counting.
+    * External functions (without bodies) that return a tensor are not
+      supported.
+    * Function with multiple blocks or multiple ReturnOps are not supported.
+
+    One-Shot Bufferize implements the following contract around function calls:
+    The buffer of function arguments is always writable (unless annotated with
+    `bufferization.writable = false`). A buffer copy may be inserted at the call
+    site where necessary. Alias sets and equivalence info is propagated through
+    function calls. Whenever a function is bufferized, all other functions that
+    are being called were already analyzed and bufferized, so exact alias and
+    equivalence information is available. This is why recursive function calls
+    are not yet supported.
+
+    One-Shot Bufferize gathers additional information during the analysis phase
+    when function boundary bufferization is activated. E.g., whether a function
+    argument is read/written and which returned values are aliasing/equivalent.
+    For debugging purposes, such information can be printed with
+    `test-analysis-only`.
   }];
   let options = [
     Option<"allowReturnAllocs", "allow-return-allocs", "bool",
             /*default=*/"false",
            "Allows returning/yielding new allocations from a block.">,
     Option<"allowUnknownOps", "allow-unknown-ops", "bool",
            /*default=*/"false",
            "Allows unknown (not bufferizable) ops in the input IR.">,
     Option<"analysisFuzzerSeed", "analysis-fuzzer-seed", "unsigned",
            /*default=*/"0",
            "Test only: Analyze ops in random order with a given seed (fuzzer)">,
+    Option<"bufferizeFunctionBoundaries", "bufferize-function-boundaries",
+           "bool", /*default=*/"0",
+           "Bufferize function boundaries (experimental).">,
     Option<"createDeallocs", "create-deallocs", "bool", /*default=*/"true",
            "Specify if buffers should be deallocated. For compatibility with "
            "core bufferization passes.">,
     ListOption<"dialectFilter", "dialect-filter", "std::string",
                "Restrict bufferization to ops from these dialects.",
                "llvm::cl::MiscFlags::CommaSeparated">,
     Option<"fullyDynamicLayoutMaps", "fully-dynamic-layout-maps", "bool",
            /*default=*/"true",

mlir/include/mlir/Dialect/Linalg/ComprehensiveBufferize/CMakeLists.txt

-# no targets defined here
-

mlir/include/mlir/Dialect/Linalg/ComprehensiveBufferize/ModuleBufferization.h

-//===- ModuleBufferization.h - Bufferization across Func. Boundaries ------===//
-//
-// 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_COMPREHENSIVEBUFFERIZE_MODULEBUFFERIZATION_H
-#define MLIR_DIALECT_LINALG_COMPREHENSIVEBUFFERIZE_MODULEBUFFERIZATION_H
-
-#include <memory>
-
-namespace mlir {
-
-class DialectRegistry;
-struct LogicalResult;
-class ModuleOp;
-
-namespace bufferization {
-struct OneShotBufferizationOptions;
-} // namespace bufferization
-
-namespace linalg {
-namespace comprehensive_bufferize {
-
-/// Run Module Bufferization on the given module. Performs a simple function
-/// call analysis to determine which function arguments are inplaceable. Then
-/// analyzes and bufferizes FuncOps one-by-one with One-Shot Bufferize.
-LogicalResult
-runModuleBufferize(ModuleOp moduleOp,
-                   bufferization::OneShotBufferizationOptions options);
-
-namespace std_ext {
-
-void registerModuleBufferizationExternalModels(DialectRegistry &registry);
-
-} // namespace std_ext
-} // namespace comprehensive_bufferize
-} // namespace linalg
-} // namespace mlir
-
-#endif // MLIR_DIALECT_LINALG_COMPREHENSIVEBUFFERIZE_MODULEBUFFERIZATION_H

mlir/include/mlir/InitAllDialects.h

 #include "mlir/Dialect/AMX/AMXDialect.h"
 #include "mlir/Dialect/Affine/IR/AffineOps.h"
 #include "mlir/Dialect/Arithmetic/IR/Arithmetic.h"
 #include "mlir/Dialect/Arithmetic/Transforms/BufferizableOpInterfaceImpl.h"
 #include "mlir/Dialect/ArmNeon/ArmNeonDialect.h"
 #include "mlir/Dialect/ArmSVE/ArmSVEDialect.h"
 #include "mlir/Dialect/Async/IR/Async.h"
 #include "mlir/Dialect/Bufferization/IR/Bufferization.h"
+#include "mlir/Dialect/Bufferization/Transforms/FuncBufferizableOpInterfaceImpl.h"
 #include "mlir/Dialect/Complex/IR/Complex.h"
 #include "mlir/Dialect/ControlFlow/IR/ControlFlow.h"
 #include "mlir/Dialect/DLTI/DLTI.h"
 #include "mlir/Dialect/EmitC/IR/EmitC.h"
 #include "mlir/Dialect/Func/IR/FuncOps.h"
 #include "mlir/Dialect/GPU/GPUDialect.h"
 #include "mlir/Dialect/LLVMIR/LLVMDialect.h"
 #include "mlir/Dialect/LLVMIR/NVVMDialect.h"
 #include "mlir/Dialect/LLVMIR/ROCDLDialect.h"
 #include "mlir/Dialect/Linalg/IR/Linalg.h"
 #include "mlir/Dialect/Linalg/Transforms/BufferizableOpInterfaceImpl.h"
 #include "mlir/Dialect/Math/IR/Math.h"
 #include "mlir/Dialect/MemRef/IR/MemRef.h"
 #include "mlir/Dialect/OpenACC/OpenACC.h"
 #include "mlir/Dialect/OpenMP/OpenMPDialect.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/BufferizableOpInterfaceImpl.h"
 #include "mlir/Dialect/SCF/SCF.h"
 #include "mlir/Dialect/SPIRV/IR/SPIRVDialect.h"
 #include "mlir/Dialect/Shape/IR/Shape.h"
+#include "mlir/Dialect/Shape/Transforms/BufferizableOpInterfaceImpl.h"
 #include "mlir/Dialect/SparseTensor/IR/SparseTensor.h"
 #include "mlir/Dialect/Tensor/IR/Tensor.h"
 #include "mlir/Dialect/Tensor/IR/TensorInferTypeOpInterfaceImpl.h"
 #include "mlir/Dialect/Tensor/IR/TensorTilingInterfaceImpl.h"
 #include "mlir/Dialect/Tensor/Transforms/BufferizableOpInterfaceImpl.h"
 #include "mlir/Dialect/Tosa/IR/TosaOps.h"
 #include "mlir/Dialect/Vector/IR/VectorOps.h"
 #include "mlir/Dialect/Vector/Transforms/BufferizableOpInterfaceImpl.h"
registry.insert<acc::OpenACCDialect,
                   ROCDL::ROCDLDialect,
                   shape::ShapeDialect,
                   sparse_tensor::SparseTensorDialect,
                   tensor::TensorDialect,
                   tosa::TosaDialect,
                   x86vector::X86VectorDialect>();
   // clang-format on
   arith::registerBufferizableOpInterfaceExternalModels(registry);
+  bufferization::func_ext::registerBufferizableOpInterfaceExternalModels(
+      registry);
   linalg::registerBufferizableOpInterfaceExternalModels(registry);
   scf::registerBufferizableOpInterfaceExternalModels(registry);
+  shape::registerBufferizableOpInterfaceExternalModels(registry);
   tensor::registerBufferizableOpInterfaceExternalModels(registry);
   tensor::registerInferTypeOpInterfaceExternalModels(registry);
   tensor::registerTilingOpInterfaceExternalModels(registry);
   vector::registerBufferizableOpInterfaceExternalModels(registry);
 }
 
 /// Append all the MLIR dialects to the registry contained in the given context.
 inline void registerAllDialects(MLIRContext &context) {
   DialectRegistry registry;
   registerAllDialects(registry);
   context.appendDialectRegistry(registry);
 }
 
 } // namespace mlir
 
 #endif // MLIR_INITALLDIALECTS_H_

mlir/lib/Dialect/Bufferization/IR/BufferizableOpInterface.cpp

 
 #define DEBUG_TYPE "bufferizable-op-interface"
 #define DBGS() (llvm::dbgs() << '[' << DEBUG_TYPE << "] ")
 #define LDBG(X) LLVM_DEBUG(DBGS() << (X))
 
 using namespace mlir;
 using namespace bufferization;
 
-/// Attribute name used to mark the bufferization layout for region
-/// arguments during linalg comprehensive bufferization.
-constexpr const ::llvm::StringLiteral
-    bufferization::BufferizableOpInterface::kBufferLayoutAttrName;
-
 /// Attribute name used to mark region arguments that can be bufferized
 /// in-place during linalg comprehensive bufferization.
 constexpr const ::llvm::StringLiteral
     bufferization::BufferizableOpInterface::kInplaceableAttrName;
 
 /// Attribute name used to mark allocs that are created by the bufferization.
 static const char *kBufferAllocationAttr = "bufferization.allocation";
 

mlir/lib/Dialect/Bufferization/IR/BufferizationDialect.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/Bufferization/IR/Bufferization.h"
+#include "mlir/IR/FunctionInterfaces.h"
 #include "mlir/Transforms/InliningUtils.h"
 
 using namespace mlir;
 using namespace mlir::bufferization;
 
 #include "mlir/Dialect/Bufferization/IR/BufferizationOpsDialect.cpp.inc"
 
+/// Attribute name used to mark function arguments who's buffers can be written
+/// to during One-Shot Module Bufferize.
+constexpr const ::llvm::StringLiteral BufferizationDialect::kWritableAttrName;
+
+/// Attribute name used to mark the bufferization layout for region arguments
+/// during One-Shot Module Bufferize.
+constexpr const ::llvm::StringLiteral
+    BufferizationDialect::kBufferLayoutAttrName;
+
 //===----------------------------------------------------------------------===//
 // Bufferization Dialect Interfaces
 //===----------------------------------------------------------------------===//
 
 namespace {
 struct BufferizationInlinerInterface : public DialectInlinerInterface {
   using DialectInlinerInterface::DialectInlinerInterface;
 

 
 void mlir::bufferization::BufferizationDialect::initialize() {
   addOperations<
 #define GET_OP_LIST
 #include "mlir/Dialect/Bufferization/IR/BufferizationOps.cpp.inc"
       >();
   addInterfaces<BufferizationInlinerInterface>();
 }
+
+LogicalResult
+BufferizationDialect::verifyOperationAttribute(Operation *op,
+                                               NamedAttribute attr) {
+  using bufferization::BufferizableOpInterface;
+
+  if (attr.getName() == kWritableAttrName) {
+    if (!attr.getValue().isa<BoolAttr>()) {
+      return op->emitError() << "'" << kWritableAttrName
+                             << "' is expected to be a boolean attribute";
+    }
+    if (!isa<FunctionOpInterface>(op))
+      return op->emitError() << "expected " << attr.getName()
+                             << " to be used on function-like operations";
+    return success();
+  }
+  if (attr.getName() == kBufferLayoutAttrName) {
+    if (!attr.getValue().isa<AffineMapAttr>()) {
+      return op->emitError() << "'" << kBufferLayoutAttrName
+                             << "' is expected to be a affine map attribute";
+    }
+    if (!isa<FunctionOpInterface>(op))
+      return op->emitError() << "expected " << attr.getName()
+                             << " to be used on function-like operations";
+    return success();
+  }
+
+  return op->emitError() << "attribute '" << attr.getName()
+                         << "' not supported by the bufferization dialect";
+}

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

 //===- Bufferize.cpp - Bufferization utilities ----------------------------===//
 //
 // 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 "PassDetail.h"
 
 #include "mlir/Dialect/Bufferization/IR/BufferizableOpInterface.h"
 #include "mlir/Dialect/Bufferization/IR/Bufferization.h"
 #include "mlir/Dialect/Bufferization/Transforms/Bufferize.h"
 #include "mlir/Dialect/Bufferization/Transforms/OneShotAnalysis.h"
+#include "mlir/Dialect/Bufferization/Transforms/OneShotModuleBufferize.h"
 #include "mlir/Dialect/Bufferization/Transforms/Passes.h"
 #include "mlir/Dialect/Func/IR/FuncOps.h"
 #include "mlir/IR/Operation.h"
 #include "mlir/Pass/PassManager.h"
 #include "mlir/Transforms/GreedyPatternRewriteDriver.h"
 #include "mlir/Transforms/Passes.h"
 
 using namespace mlir;
if (!options) {
       opt.createDeallocs = createDeallocs;
       opt.fullyDynamicLayoutMaps = fullyDynamicLayoutMaps;
       opt.printConflicts = printConflicts;
       opt.testAnalysisOnly = testAnalysisOnly;
 
       BufferizationOptions::OpFilterEntry::FilterFn filterFn =
           [&](Operation *op) {
             // Disallow non-func dialect ops. I.e., no ops related to function
-            // calls.
-            if (isa<func::FuncDialect>(op->getDialect()))
+            // calls. (Unless explicitly activated.)
+            bool isFuncBoundaryOp =
+                isa_and_nonnull<func::FuncDialect>(op->getDialect());
+            if (!this->bufferizeFunctionBoundaries && isFuncBoundaryOp)
               return false;
             // Filter may be specified via options.
             if (this->dialectFilter.hasValue())
               return llvm::find(this->dialectFilter,
                                 op->getDialect()->getNamespace()) !=
                      this->dialectFilter.end();
             // No filter specified: All other ops are allowed.
             return true;
           };
       opt.allowOperationInFilter(filterFn);
     } else {
       opt = *options;
     }
 
     ModuleOp moduleOp = getOperation();
+    if (bufferizeFunctionBoundaries) {
+      if (failed(runOneShotModuleBufferize(moduleOp, opt))) {
+        signalPassFailure();
+        return;
+      }
+    } else {
       if (failed(runOneShotBufferize(moduleOp, opt))) {
         signalPassFailure();
         return;
       }
+    }
 
     if (opt.testAnalysisOnly)
       return;
 
     OpPassManager cleanupPipeline("builtin.module");
     cleanupPipeline.addPass(createCanonicalizerPass());
     cleanupPipeline.addPass(createCSEPass());
     cleanupPipeline.addPass(createLoopInvariantCodeMotionPass());

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

 add_mlir_dialect_library(MLIRBufferizationTransforms
   Bufferize.cpp
   BufferDeallocation.cpp
   BufferOptimizations.cpp
   BufferResultsToOutParams.cpp
   BufferUtils.cpp
+  FuncBufferizableOpInterfaceImpl.cpp
   OneShotAnalysis.cpp
+  OneShotModuleBufferize.cpp
 
   ADDITIONAL_HEADER_DIRS
   ${MLIR_MAIN_INCLUDE_DIR}/mlir/Dialect/Bufferization
 
   DEPENDS
   MLIRBufferizationPassIncGen
 
   LINK_LIBS PUBLIC

mlir/lib/Dialect/Bufferization/Transforms/FuncBufferizableOpInterfaceImpl.cpp

+//===- BufferizableOpInterfaceImpl.cpp - Impl. of BufferizableOpInterface -===//
+//
+// 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/Bufferization/Transforms/FuncBufferizableOpInterfaceImpl.h"
+#include "mlir/Dialect/Bufferization/IR/BufferizableOpInterface.h"
+#include "mlir/Dialect/Bufferization/IR/Bufferization.h"
+#include "mlir/Dialect/Bufferization/Transforms/OneShotAnalysis.h"
+#include "mlir/Dialect/Func/IR/FuncOps.h"
+#include "mlir/Dialect/MemRef/IR/MemRef.h"
+#include "mlir/IR/Dialect.h"
+#include "mlir/IR/Operation.h"
+
+namespace mlir {
+namespace bufferization {
+namespace func_ext {
+
+/// Return the unique ReturnOp that terminates `funcOp`.
+/// Return nullptr if there is no such unique ReturnOp.
+static func::ReturnOp getAssumedUniqueReturnOp(FuncOp funcOp) {
+  func::ReturnOp returnOp;
+  for (Block &b : funcOp.getBody()) {
+    if (auto candidateOp = dyn_cast<func::ReturnOp>(b.getTerminator())) {
+      if (returnOp)
+        return nullptr;
+      returnOp = candidateOp;
+    }
+  }
+  return returnOp;
+}
+
+/// Return the index-th bufferized function argument type. This assumes that the
+/// specified argument is a tensor. If the tensor is ranked, a layout map may be
+/// specified by the user. If no layout map is specified, a fully dynamic map is
+/// used.
+static BaseMemRefType
+getBufferizedFunctionArgType(FuncOp funcOp, int64_t index,
+                             const BufferizationOptions &options) {
+  auto tensorType =
+      funcOp.getFunctionType().getInput(index).dyn_cast<TensorType>();
+  assert(tensorType && "expected TensorType");
+  BaseMemRefType memrefType = getMemRefType(tensorType, options);
+
+  auto layoutAttr = funcOp.getArgAttrOfType<AffineMapAttr>(
+      index, BufferizationDialect::kBufferLayoutAttrName);
+  if (!layoutAttr)
+    return memrefType;
+
+  auto rankedMemrefType = memrefType.dyn_cast<MemRefType>();
+  assert(rankedMemrefType && "buffer layout not supported on unranked tensors");
+  return MemRefType::get(
+      rankedMemrefType.getShape(), rankedMemrefType.getElementType(),
+      layoutAttr.getValue(), rankedMemrefType.getMemorySpaceAsInt());
+}
+
+/// Return the FuncOp called by `callOp`.
+static FuncOp getCalledFunction(CallOpInterface callOp) {
+  SymbolRefAttr sym = callOp.getCallableForCallee().dyn_cast<SymbolRefAttr>();
+  if (!sym)
+    return nullptr;
+  return dyn_cast_or_null<FuncOp>(
+      SymbolTable::lookupNearestSymbolFrom(callOp, sym));
+}
+
+/// Get FuncAnalysisState.
+static const FuncAnalysisState &
+getFuncAnalysisState(const AnalysisState &state) {
+  Optional<const FuncAnalysisState *> maybeState =
+      state.getDialectState<FuncAnalysisState>(
+          func::FuncDialect::getDialectNamespace());
+  assert(maybeState.hasValue() && "FuncAnalysisState does not exist");
+  return **maybeState;
+}
+
+/// Return the state (phase) of analysis of the FuncOp.
+static FuncOpAnalysisState getFuncOpAnalysisState(const AnalysisState &state,
+                                                  FuncOp funcOp) {
+  const FuncAnalysisState &funcState = getFuncAnalysisState(state);
+  auto it = funcState.analyzedFuncOps.find(funcOp);
+  if (it == funcState.analyzedFuncOps.end())
+    return FuncOpAnalysisState::NotAnalyzed;
+  return it->second;
+}
+
+/// Return the index of the bbArg in the given FuncOp that is equivalent to the
+/// specified return value (if any).
+static Optional<int64_t> getEquivalentFuncArgIdx(FuncOp funcOp,
+                                                 const FuncAnalysisState &state,
+                                                 int64_t returnValIdx) {
+  auto funcOpIt = state.equivalentFuncArgs.find(funcOp);
+  if (funcOpIt == state.equivalentFuncArgs.end())
+    // No equivalence info stores for funcOp.
+    return None;
+
+  auto retValIt = funcOpIt->getSecond().find(returnValIdx);
+  if (retValIt == funcOpIt->getSecond().end())
+    // Return value has no equivalent bbArg.
+    return None;
+
+  return retValIt->getSecond();
+}
+
+/// If `value` is a memref::CastOp, return its source. Otherwise, return
+/// `value` directly.
+static Value getNonCastedValue(Value value) {
+  while (auto castOp = value.getDefiningOp<memref::CastOp>())
+    value = castOp.source();
+  return value;
+}
+
+struct CallOpInterface
+    : public BufferizableOpInterface::ExternalModel<CallOpInterface,
+                                                    func::CallOp> {
+  bool bufferizesToMemoryRead(Operation *op, OpOperand &opOperand,
+                              const AnalysisState &state) const {
+    func::CallOp callOp = cast<func::CallOp>(op);
+    FuncOp funcOp = getCalledFunction(callOp);
+    assert(funcOp && "expected CallOp to a FuncOp");
+
+    const FuncAnalysisState &funcState = getFuncAnalysisState(state);
+    if (getFuncOpAnalysisState(state, funcOp) != FuncOpAnalysisState::Analyzed)
+      // FuncOp not analyzed yet. Assume that OpOperand is read.
+      return true;
+
+    auto it = funcState.readBbArgs.find(funcOp);
+    assert(it != funcState.readBbArgs.end() &&
+           "expected analysis info for analyzed FuncOps");
+    return it->second.contains(opOperand.getOperandNumber());
+  }
+
+  bool bufferizesToMemoryWrite(Operation *op, OpOperand &opOperand,
+                               const AnalysisState &state) const {
+    func::CallOp callOp = cast<func::CallOp>(op);
+    FuncOp funcOp = getCalledFunction(callOp);
+    assert(funcOp && "expected CallOp to a FuncOp");
+
+    const FuncAnalysisState &funcState = getFuncAnalysisState(state);
+    if (getFuncOpAnalysisState(state, funcOp) != FuncOpAnalysisState::Analyzed)
+      // FuncOp not analyzed yet. Assume that OpOperand is written.
+      return true;
+
+    auto it = funcState.writtenBbArgs.find(funcOp);
+    assert(it != funcState.writtenBbArgs.end() &&
+           "expected analysis info for analyzed FuncOps");
+    return it->second.contains(opOperand.getOperandNumber());
+  }
+
+  SmallVector<OpResult> getAliasingOpResult(Operation *op, OpOperand &opOperand,
+                                            const AnalysisState &state) const {
+    func::CallOp callOp = cast<func::CallOp>(op);
+    FuncOp funcOp = getCalledFunction(callOp);
+    assert(funcOp && "expected CallOp to a FuncOp");
+    const FuncAnalysisState &funcState = getFuncAnalysisState(state);
+    if (getFuncOpAnalysisState(state, funcOp) !=
+        FuncOpAnalysisState::Analyzed) {
+      // FuncOp not analyzed yet. Any OpResult may be aliasing.
+      SmallVector<OpResult> result;
+      for (OpResult opResult : op->getOpResults())
+        if (opResult.getType().isa<TensorType>())
+          result.push_back(opResult);
+      return result;
+    }
+
+    // Get aliasing results from state.
+    auto mapping = funcState.aliasingReturnVals.find(funcOp);
+    assert(mapping != funcState.aliasingReturnVals.end() &&
+           "expected analysis info analyzed FuncOps");
+    auto aliasingReturnVals =
+        mapping->second.find(opOperand.getOperandNumber());
+    if (aliasingReturnVals == mapping->second.end())
+      return {};
+
+    SmallVector<OpResult> result;
+    for (int64_t resultIdx : aliasingReturnVals->second)
+      result.push_back(callOp->getOpResult(resultIdx));
+    return result;
+  }
+
+  SmallVector<OpOperand *>
+  getAliasingOpOperand(Operation *op, OpResult opResult,
+                       const AnalysisState &state) const {
+    func::CallOp callOp = cast<func::CallOp>(op);
+    FuncOp funcOp = getCalledFunction(callOp);
+    assert(funcOp && "expected CallOp to a FuncOp");
+    const FuncAnalysisState &funcState = getFuncAnalysisState(state);
+    if (getFuncOpAnalysisState(state, funcOp) !=
+        FuncOpAnalysisState::Analyzed) {
+      // FuncOp not analyzed yet. Any OpOperand may be aliasing.
+      SmallVector<OpOperand *> result;
+      for (OpOperand &opOperand : op->getOpOperands())
+        if (opOperand.get().getType().isa<TensorType>())
+          result.push_back(&opOperand);
+      return result;
+    }
+
+    // Get aliasing bbArgs from state.
+    auto mapping = funcState.aliasingFuncArgs.find(funcOp);
+    assert(mapping != funcState.aliasingFuncArgs.end() &&
+           "expected analysis info analyzed FuncOps");
+    auto aliasingFuncArgs = mapping->second.find(opResult.getResultNumber());
+    if (aliasingFuncArgs == mapping->second.end())
+      return {};
+
+    SmallVector<OpOperand *> result;
+    for (int64_t bbArgIdx : aliasingFuncArgs->second)
+      result.push_back(&callOp->getOpOperand(bbArgIdx));
+    return result;
+  }
+
+  BufferRelation bufferRelation(Operation *op, OpResult opResult,
+                                const AnalysisState &state) const {
+    return BufferRelation::Equivalent;
+  }
+
+  /// All function arguments are writable. It is the responsibility of the
+  /// CallOp to insert buffer copies where necessary.
+  LogicalResult bufferize(Operation *op, RewriterBase &rewriter,
+                          BufferizationState &state) const {
+    func::CallOp callOp = cast<func::CallOp>(op);
+    unsigned numResults = callOp.getNumResults();
+    unsigned numOperands = callOp->getNumOperands();
+    FuncOp funcOp = getCalledFunction(callOp);
+    assert(funcOp && "expected CallOp to a FuncOp");
+    FunctionType funcType = funcOp.getFunctionType();
+    const FuncAnalysisState &moduleState =
+        getFuncAnalysisState(state.getAnalysisState());
+    const OneShotBufferizationOptions &options =
+        static_cast<const OneShotBufferizationOptions &>(state.getOptions());
+
+    // Result types of the bufferized CallOp.
+    SmallVector<Type> resultTypes;
+    // Replacement values for the existing CallOp. These are usually the results
+    // of the bufferized CallOp, unless a tensor result folds onto an operand.
+    SmallVector<Value> replacementValues(numResults, Value());
+    // For non-tensor results: A mapping from return val indices of the old
+    // CallOp to return val indices of the bufferized CallOp.
+    SmallVector<Optional<unsigned>> retValMapping(numResults, None);
+    // Operands of the bufferized CallOp.
+    SmallVector<Value> newOperands(numOperands, Value());
+
+    // Based on previously gathered equivalence information, we know if a
+    // tensor result folds onto an operand. These are the only tensor value
+    // results that are supported at the moment.
+    //
+    // For tensors return values that do not fold onto an operand, additional
+    // work is needed (TODO) to either:
+    // * hoist a result into an inplaceable operand or
+    // * devise a better representation to truly return a buffer.
+    //
+    // Note: If a function has no body, no equivalence information is
+    // available. Consequently, a tensor return value cannot be proven to fold
+    // onto a FuncOp bbArg, so calls to such functions are not bufferizable at
+    // the moment.
+
+    // 1. Compute the result types of the new CallOp. Tensor results that are
+    // equivalent to a FuncOp bbArg are no longer returned.
+    for (const auto &it : llvm::enumerate(callOp.getResultTypes())) {
+      unsigned returnValIdx = it.index();
+      Type returnType = it.value();
+      if (!returnType.isa<TensorType>()) {
+        // Non-tensor values are returned.
+        retValMapping[returnValIdx] = resultTypes.size();
+        resultTypes.push_back(returnType);
+        continue;
+      }
+
+      if (Optional<int64_t> bbArgIdx =
+              getEquivalentFuncArgIdx(funcOp, moduleState, returnValIdx)) {
+        // Return operands that are equivalent to some bbArg, are not
+        // returned.
+        FailureOr<Value> bufferOrFailure =
+            state.getBuffer(rewriter, callOp->getOpOperand(*bbArgIdx));
+        if (failed(bufferOrFailure))
+          return failure();
+        replacementValues[returnValIdx] = *bufferOrFailure;
+        newOperands[*bbArgIdx] = *bufferOrFailure;
+        continue;
+      }
+
+      if (!options.allowReturnAllocs)
+        return callOp->emitError(
+            "call to FuncOp that returns non-equivalent tensors not supported");
+
+      // Returning a memref. This memref is not equivalent to any bbArg. It is
+      // likely a newly allocated buffer. We may want to hoist such allocations
+      // to the call site in the future.
+      retValMapping[returnValIdx] = resultTypes.size();
+      resultTypes.push_back(funcType.getResult(resultTypes.size()));
+    }
+
+    // 2. Rewrite tensor operands as memrefs based on `bufferizedFuncType`.
+    for (OpOperand &opOperand : callOp->getOpOperands()) {
+      unsigned idx = opOperand.getOperandNumber();
+      Value tensorOperand = opOperand.get();
+
+      // Non-tensor operands are just copied.
+      if (!tensorOperand.getType().isa<TensorType>()) {
+        newOperands[idx] = tensorOperand;
+        continue;
+      }
+
+      // Retrieve buffers for tensor operands. Tensor operand buffers, who's
+      // corresponding FuncOp bbArgs are equivalent to a returned tensor, were
+      // already stored in `newOperands` during Step 1.
+      Value buffer = newOperands[idx];
+      if (!buffer) {
+        FailureOr<Value> bufferOrFailure = state.getBuffer(rewriter, opOperand);
+        if (failed(bufferOrFailure))
+          return failure();
+        buffer = *bufferOrFailure;
+      }
+
+      // Caller / callee type mismatch is handled with a CastOp.
+      auto memRefType = funcType.getInput(idx);
+      // Since we don't yet have a clear layout story, to_memref may
+      // conservatively turn tensors into more dynamic memref than necessary.
+      // If the memref type of the callee fails, introduce an extra memref.cast
+      // that will either canonicalize away or fail compilation until we can do
+      // something better.
+      if (buffer.getType() != memRefType) {
+        assert(
+            memref::CastOp::areCastCompatible(buffer.getType(), memRefType) &&
+            "CallOp::bufferize: cast incompatible");
+        Value castBuffer = rewriter.create<memref::CastOp>(callOp.getLoc(),
+                                                           memRefType, buffer);
+        buffer = castBuffer;
+      }
+      newOperands[idx] = buffer;
+    }
+
+    // 3. Create the new CallOp.
+    Operation *newCallOp = rewriter.create<func::CallOp>(
+        callOp.getLoc(), funcOp.getSymName(), resultTypes, newOperands);
+    newCallOp->setAttrs(callOp->getAttrs());
+    // Get replacement values for non-tensor / non-equivalent results.
+    for (unsigned i = 0; i < replacementValues.size(); ++i) {
+      if (replacementValues[i])
+        continue;
+      replacementValues[i] = newCallOp->getResult(*retValMapping[i]);
+    }
+
+    // 4. Replace the old op with the new op.
+    replaceOpWithBufferizedValues(rewriter, callOp, replacementValues);
+
+    return success();
+  }
+};
+
+struct ReturnOpInterface
+    : public BufferizableOpInterface::ExternalModel<ReturnOpInterface,
+                                                    func::ReturnOp> {
+  bool bufferizesToMemoryRead(Operation *op, OpOperand &opOperand,
+                              const AnalysisState &state) const {
+    return true;
+  }
+
+  bool bufferizesToMemoryWrite(Operation *op, OpOperand &opOperand,
+                               const AnalysisState &state) const {
+    return false;
+  }
+
+  SmallVector<OpResult> getAliasingOpResult(Operation *op, OpOperand &opOperand,
+                                            const AnalysisState &state) const {
+    return {};
+  }
+
+  LogicalResult bufferize(Operation *op, RewriterBase &rewriter,
+                          BufferizationState &state) const {
+#ifndef NDEBUG
+    auto returnOp = cast<func::ReturnOp>(op);
+    assert(isa<FuncOp>(returnOp->getParentOp()) &&
+           "only support FuncOp parent for ReturnOp");
+#endif // NDEBUG
+
+    // ReturnOps are bufferized as part of FuncOps.
+    return failure();
+  }
+};
+
+struct FuncOpInterface
+    : public BufferizableOpInterface::ExternalModel<FuncOpInterface, FuncOp> {
+  LogicalResult bufferize(Operation *op, RewriterBase &rewriter,
+                          BufferizationState &state) const {
+    // Rewrite function bbArgs and return values into buffer form (using the
+    // canonical memref layout for now).
+    //
+    // This relies on a buffer equivalence analysis of each return operand. When
+    // a result buffer is equivalent to a function bbArg, it is dropped from the
+    // return values and becomes inplaceable at all callers.
+    //
+    // All function bbArgs are writable unless they are explicitly marked as
+    // read-only. Callers must insert copies when needed.
+    //
+    // Note: Returning a memref is possible, but corresponding CallOp
+    // bufferizations fail unless `allowReturnAllocs`.
+    auto funcOp = cast<FuncOp>(op);
+    FunctionType funcType = funcOp.getFunctionType();
+    const FuncAnalysisState &moduleState =
+        getFuncAnalysisState(state.getAnalysisState());
+    const BufferizationOptions &options = state.getOptions();
+
+    // Construct the bufferized function type.
+    SmallVector<Type> argTypes;
+    for (const auto &it : llvm::enumerate(funcType.getInputs())) {
+      Type argType = it.value();
+      if (auto tensorType = argType.dyn_cast<TensorType>()) {
+        argTypes.push_back(
+            getBufferizedFunctionArgType(funcOp, it.index(), options));
+        continue;
+      }
+      argTypes.push_back(argType);
+    }
+
+    // Bodiless functions are assumed opaque and we cannot know the
+    // bufferization contract they want to enforce. As a consequence, only
+    // support functions that don't return any tensors atm.
+    if (funcOp.getBody().empty()) {
+      SmallVector<Type> retTypes;
+      for (Type resultType : funcType.getResults()) {
+        if (resultType.isa<TensorType>())
+          return funcOp->emitError() << "cannot bufferize bodiless function "
+                                     << "that returns a tensor";
+        retTypes.push_back(resultType);
+      }
+      funcOp.setType(FunctionType::get(op->getContext(), argTypes, retTypes));
+      return success();
+    }
+
+    // TODO: Support functions with multiple returns.
+    func::ReturnOp returnOp = getAssumedUniqueReturnOp(funcOp);
+    assert(returnOp && "expected func with single return op");
+
+    // 1. Rewrite the bbArgs. Turn every tensor bbArg into a memref bbArg.
+    Block &frontBlock = funcOp.getBody().front();
+    for (BlockArgument &bbArg : frontBlock.getArguments()) {
+      auto tensorType = bbArg.getType().dyn_cast<TensorType>();
+      // Non-tensor types stay the same.
+      if (!tensorType)
+        continue;
+
+      // Collect all uses of the bbArg.
+      SmallVector<OpOperand *> bbArgUses;
+      for (OpOperand &use : bbArg.getUses())
+        bbArgUses.push_back(&use);
+
+      // Change the bbArg type to memref.
+      Type memrefType =
+          getBufferizedFunctionArgType(funcOp, bbArg.getArgNumber(), options);
+      bbArg.setType(memrefType);
+
+      // Replace all uses of the original tensor bbArg.
+      rewriter.setInsertionPointToStart(&frontBlock);
+      if (!bbArgUses.empty()) {
+        // Insert to_tensor because the remaining function body has not been
+        // bufferized yet.
+        Value toTensorOp =
+            rewriter.create<bufferization::ToTensorOp>(funcOp.getLoc(), bbArg);
+        for (OpOperand *use : bbArgUses)
+          use->set(toTensorOp);
+      }
+    }
+
+    // 2. For each result, keep track of which inplace argument it reuses.
+    SmallVector<Value> returnValues;
+    for (OpOperand &returnOperand : returnOp->getOpOperands()) {
+      Value returnVal = returnOperand.get();
+
+      // If not a tensor type just forward it.
+      if (!returnVal.getType().isa<RankedTensorType>()) {
+        returnValues.push_back(returnVal);
+        continue;
+      }
+
+      // If return operand is equivalent to some bbArg, no need to return it.
+      if (Optional<int64_t> equivBbArgIdx = getEquivalentFuncArgIdx(
+              funcOp, moduleState, returnOperand.getOperandNumber())) {
+        rewriter.setInsertionPoint(returnOp);
+        Location loc = returnOp.getLoc();
+        Value toMemrefOp = rewriter.create<bufferization::ToMemrefOp>(
+            loc, getMemRefType(returnVal.getType().cast<TensorType>(), options),
+            returnVal);
+        BlockArgument equivBbArg = funcOp.getArgument(*equivBbArgIdx);
+        // Note: This copy will fold away. It must be inserted here to ensure
+        // that `returnVal` still has at least one use and does not fold away.
+        if (failed(
+                createMemCpy(rewriter, loc, toMemrefOp, equivBbArg, options)))
+          return funcOp->emitError("could not generate copy for bbArg");
+        continue;
+      }
+
+      // Cast values at the call site if necessary.
+      returnValues.push_back(
+          getNonCastedValue(*state.getBuffer(rewriter, returnOperand)));
+    }
+
+    // 3. Rewrite the terminator without the in-place bufferizable values.
+    returnOp.operandsMutable().assign(returnValues);
+
+    // 4. Rewrite the FuncOp type to buffer form.
+    funcOp.setType(FunctionType::get(op->getContext(), argTypes,
+                                     ValueRange(returnValues).getTypes()));
+
+    return success();
+  }
+
+  /// Return `true` if the given function argument is writable.
+  bool isWritable(Operation *op, Value value,
+                  const AnalysisState &state) const {
+    auto funcOp = cast<FuncOp>(op);
+    BlockArgument bbArg = value.dyn_cast<BlockArgument>();
+    assert(bbArg && "expected BlockArgument");
+
+    // "bufferization.writable" overrides other writability decisions. This is
+    // currently used for testing only.
+    if (BoolAttr writable = funcOp.getArgAttrOfType<BoolAttr>(
+            bbArg.getArgNumber(), BufferizationDialect::kWritableAttrName))
+      return writable.getValue();
+
+    // All function arguments are writable by default.
+    return true;
+  }
+
+  bool isAllocationHoistingBarrier(Operation *op) const { return true; }
+};
+
+} // namespace func_ext
+} // namespace bufferization
+} // namespace mlir
+
+void mlir::bufferization::func_ext::
+    registerBufferizableOpInterfaceExternalModels(DialectRegistry &registry) {
+  registry.addExtension(+[](MLIRContext *ctx, func::FuncDialect *dialect) {
+    func::CallOp::attachInterface<func_ext::CallOpInterface>(*ctx);
+    func::FuncOp::attachInterface<func_ext::FuncOpInterface>(*ctx);
+    func::ReturnOp::attachInterface<func_ext::ReturnOpInterface>(*ctx);
+  });
+}

mlir/lib/Dialect/Bufferization/Transforms/OneShotModuleBufferize.cpp

+//===- ModuleBufferization.cpp - Bufferization across Func. Boundaries ----===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+//
+// Module Bufferization is an extension of One-Shot Bufferize that
+// bufferizes function boundaries. It provides `BufferizableOpInterface`
+// implementations for FuncOp, CallOp and ReturnOp.
+//
+// Module Bufferization is run via `runOneShotModuleBufferize(ModuleOp, ...)`.
+// This function analyzes the given module and determines the order of analysis
+// and bufferization: Functions that are called are processed before their
+// respective callers.
+//
+// After analyzing a FuncOp, additional information about its bbArgs is
+// gathered through PostAnalysisStepFns and stored in
+// `FuncAnalysisState`.
+//
+// * `aliasingFuncOpBBArgsAnalysis` determines the equivalent/aliasing bbArgs
+// for
+//   each tensor return value (if any).
+// * `funcOpBbArgReadWriteAnalysis` determines whether or not a tensor bbArg is
+//   read/written.
+//
+// Only tensors that are equivalent to some FuncOp bbArg may be returned.
+// Bufferization currently fails if other tensors (in particular tensors that
+// bufferize out-of-place and result in a new buffer allocation) are returned.
+// In the future, such allocations could be hoisted to the caller.
+//
+// Example: `foo` fails bufferization because %0 is not equivalent to any bbArg.
+// ```
+// func @foo() -> tensor<?xf32> {
+//   %0 = linalg.init_tensor [...] : tensor<?xf32>
+//   return %0 : tensor<?xf32>
+// }
+// ```
+//
+// Module Bufferization implements the following calling convention.
+//
+// * In the absence of conflicts within a FuncOp, the FuncOp's bbArgs may always
+//   be written to in-place.
+// * If a tensor operand of a CallOp is read after the CallOp, the operand of
+//   the CallOp must bufferize out-of-place.
+//
+// Example: The tensor.insert op bufferizes in-place because it is allowed to
+// modify the buffer of `%t1` directly. The CallOp in `caller` must bufferize
+// out-of-place because `%t0` is modified by the callee but read by the
+// tensor.extract op. The analysis of CallOps decides whether an OpOperand must
+// bufferize out-of-place based on results of `funcOpBbArgReadWriteAnalysis`.
+// ```
+// func @callee(%t1 : tensor<?xf32>) -> tensor<?xf32> {
+//   %f = ... : f32
+//   %0 = tensor.insert %f into %t1[...] : tensor<?xf32>
+//   return %0 : tensor<?xf32>
+// }
+//
+// func @caller() -> () {
+//   %t0 = ... : tensor<?xf32>
+//   %1 = call @callee(%t0) : (tensor<?xf32>) -> (tensor<?xf32>)
+//   %2 = tensor.extract %1[...]  : tensor<?xf32>
+// }
+// ```
+//
+// Note: If a function is external, `funcOpBbArgReadWriteAnalysis` cannot
+// analyze the function body. In such a case, the CallOp analysis conservatively
+// assumes that each tensor OpOperand is both read and written.
+//
+// TODO: Add FuncOp attributes so that bbArgs of external FuncOps can be marked
+// as "not reading" and/or "not writing".
+
+#include "mlir/Dialect/Bufferization/Transforms/OneShotModuleBufferize.h"
+
+#include "mlir/Dialect/Bufferization/IR/BufferizableOpInterface.h"
+#include "mlir/Dialect/Bufferization/IR/Bufferization.h"
+#include "mlir/Dialect/Bufferization/Transforms/Bufferize.h"
+#include "mlir/Dialect/Bufferization/Transforms/FuncBufferizableOpInterfaceImpl.h"
+#include "mlir/Dialect/Bufferization/Transforms/OneShotAnalysis.h"
+#include "mlir/Dialect/Func/IR/FuncOps.h"
+#include "mlir/Dialect/MemRef/IR/MemRef.h"
+#include "mlir/IR/Operation.h"
+
+using namespace mlir;
+using namespace mlir::bufferization;
+using namespace mlir::bufferization::func_ext;
+
+/// A mapping of FuncOps to their callers.
+using FuncCallerMap = DenseMap<FuncOp, DenseSet<Operation *>>;
+
+/// Get FuncAnalysisState.
+static const FuncAnalysisState &
+getFuncAnalysisState(const AnalysisState &state) {
+  Optional<const FuncAnalysisState *> maybeState =
+      state.getDialectState<FuncAnalysisState>(
+          func::FuncDialect::getDialectNamespace());
+  assert(maybeState.hasValue() && "FuncAnalysisState does not exist");
+  return **maybeState;
+}
+
+/// Get or create FuncAnalysisState.
+static FuncAnalysisState &getFuncAnalysisState(AnalysisState &state) {
+  return state.getOrCreateDialectState<FuncAnalysisState>(
+      func::FuncDialect::getDialectNamespace());
+}
+
+/// Return the state (phase) of analysis of the FuncOp.
+static FuncOpAnalysisState getFuncOpAnalysisState(const AnalysisState &state,
+                                                  FuncOp funcOp) {
+  const FuncAnalysisState &funcState = getFuncAnalysisState(state);
+  auto it = funcState.analyzedFuncOps.find(funcOp);
+  if (it == funcState.analyzedFuncOps.end())
+    return FuncOpAnalysisState::NotAnalyzed;
+  return it->second;
+}
+
+/// Return the unique ReturnOp that terminates `funcOp`.
+/// Return nullptr if there is no such unique ReturnOp.
+static func::ReturnOp getAssumedUniqueReturnOp(FuncOp funcOp) {
+  func::ReturnOp returnOp;
+  for (Block &b : funcOp.getBody()) {
+    if (auto candidateOp = dyn_cast<func::ReturnOp>(b.getTerminator())) {
+      if (returnOp)
+        return nullptr;
+      returnOp = candidateOp;
+    }
+  }
+  return returnOp;
+}
+
+namespace {
+
+/// Annotate IR with the results of the analysis. For testing purposes only.
+static void annotateEquivalentReturnBbArg(OpOperand &returnVal,
+                                          BlockArgument bbArg) {
+  const char *kEquivalentArgsAttr = "__equivalent_func_args__";
+  Operation *op = returnVal.getOwner();
+
+  SmallVector<int64_t> equivBbArgs;
+  if (op->hasAttr(kEquivalentArgsAttr)) {
+    auto attr = op->getAttr(kEquivalentArgsAttr).cast<ArrayAttr>();
+    equivBbArgs = llvm::to_vector<4>(llvm::map_range(attr, [](Attribute a) {
+      return a.cast<IntegerAttr>().getValue().getSExtValue();
+    }));
+  } else {
+    equivBbArgs.append(op->getNumOperands(), -1);
+  }
+  equivBbArgs[returnVal.getOperandNumber()] = bbArg.getArgNumber();
+
+  OpBuilder b(op->getContext());
+  op->setAttr(kEquivalentArgsAttr, b.getI64ArrayAttr(equivBbArgs));
+}
+
+/// Store function BlockArguments that are equivalent to/aliasing a returned
+/// value in FuncAnalysisState.
+static LogicalResult
+aliasingFuncOpBBArgsAnalysis(Operation *op, AnalysisState &state,
+                             BufferizationAliasInfo &aliasInfo,
+                             SmallVector<Operation *> &newOps) {
+  FuncAnalysisState &funcState = getFuncAnalysisState(state);
+
+  // Support only single return-terminated block in the function.
+  auto funcOp = cast<FuncOp>(op);
+  func::ReturnOp returnOp = getAssumedUniqueReturnOp(funcOp);
+  assert(returnOp && "expected func with single return op");
+
+  // Initialize data structure.
+  auto createdEquiv = funcState.equivalentFuncArgs.try_emplace(
+      funcOp, DenseMap<int64_t, int64_t>());
+  auto createdAliasingOperands = funcState.aliasingFuncArgs.try_emplace(
+      funcOp, DenseMap<int64_t, SmallVector<int64_t>>());
+  auto createdAliasingResults = funcState.aliasingReturnVals.try_emplace(
+      funcOp, DenseMap<int64_t, SmallVector<int64_t>>());
+  (void)createdEquiv;
+  (void)createdAliasingOperands;
+  (void)createdAliasingResults;
+#ifndef NDEBUG
+  assert(createdEquiv.second && "equivalence info exists already");
+  assert(createdAliasingOperands.second && "aliasing info exists already");
+  assert(createdAliasingResults.second && "aliasing info exists already");
+#endif // NDEBUG
+
+  for (OpOperand &returnVal : returnOp->getOpOperands())
+    if (returnVal.get().getType().isa<RankedTensorType>())
+      for (BlockArgument bbArg : funcOp.getArguments())
+        if (bbArg.getType().isa<RankedTensorType>()) {
+          int64_t returnIdx = returnVal.getOperandNumber();
+          int64_t bbArgIdx = bbArg.getArgNumber();
+          if (aliasInfo.areEquivalentBufferizedValues(returnVal.get(), bbArg)) {
+            funcState.equivalentFuncArgs[funcOp][returnIdx] = bbArgIdx;
+            if (state.getOptions().testAnalysisOnly)
+              annotateEquivalentReturnBbArg(returnVal, bbArg);
+          }
+          if (aliasInfo.areAliasingBufferizedValues(returnVal.get(), bbArg)) {
+            funcState.aliasingFuncArgs[funcOp][returnIdx].push_back(bbArgIdx);
+            funcState.aliasingReturnVals[funcOp][bbArgIdx].push_back(returnIdx);
+          }
+        }
+
+  return success();
+}
+
+/// Return true if the buffer of the given tensor value is written to. Must not
+/// be called for values inside not yet analyzed functions. (Post-analysis
+/// steps do not have to be run yet, i.e., "in progress" is also OK.)
+static bool isValueWritten(Value value, const AnalysisState &state,
+                           const BufferizationAliasInfo &aliasInfo) {
+#ifndef NDEBUG
+  assert(value.getType().isa<TensorType>() && "expected TensorType");
+  FuncOp funcOp;
+  if (auto bbArg = value.dyn_cast<BlockArgument>()) {
+    Operation *owner = bbArg.getOwner()->getParentOp();
+    funcOp = isa<FuncOp>(owner) ? cast<FuncOp>(owner)
+                                : owner->getParentOfType<FuncOp>();
+  } else {
+    funcOp = value.getDefiningOp()->getParentOfType<FuncOp>();
+  }
+  assert(getFuncOpAnalysisState(state, funcOp) !=
+             FuncOpAnalysisState::NotAnalyzed &&
+         "FuncOp must be fully analyzed or analysis in progress");
+#endif // NDEBUG
+
+  bool isWritten = false;
+  aliasInfo.applyOnAliases(value, [&](Value val) {
+    for (OpOperand &use : val.getUses())
+      if (state.isInPlace(use) && state.bufferizesToMemoryWrite(use))
+        isWritten = true;
+  });
+  return isWritten;
+}
+
+static void annotateFuncArgAccess(FuncOp funcOp, BlockArgument bbArg,
+                                  bool isRead, bool isWritten) {
+  OpBuilder b(funcOp.getContext());
+  Attribute accessType;
+  if (isRead && isWritten) {
+    accessType = b.getStringAttr("read-write");
+  } else if (isRead) {
+    accessType = b.getStringAttr("read");
+  } else if (isWritten) {
+    accessType = b.getStringAttr("write");
+  } else {
+    accessType = b.getStringAttr("none");
+  }
+  funcOp.setArgAttr(bbArg.getArgNumber(), "bufferization.access", accessType);
+}
+
+/// Determine which FuncOp bbArgs are read and which are written. If this
+/// PostAnalysisStepFn is run on a function with unknown ops, it will
+/// conservatively assume that such ops bufferize to a read + write.
+static LogicalResult
+funcOpBbArgReadWriteAnalysis(Operation *op, AnalysisState &state,
+                             BufferizationAliasInfo &aliasInfo,
+                             SmallVector<Operation *> &newOps) {
+  FuncAnalysisState &funcState = getFuncAnalysisState(state);
+  auto funcOp = cast<FuncOp>(op);
+
+  // Initialize data structure.
+  auto createdRead =
+      funcState.readBbArgs.try_emplace(funcOp, DenseSet<int64_t>());
+  auto createdWritten =
+      funcState.writtenBbArgs.try_emplace(funcOp, DenseSet<int64_t>());
+  (void)createdRead;
+  (void)createdWritten;
+#ifndef NDEBUG
+  assert(createdRead.second && "bbarg access info exists already");
+  assert(createdWritten.second && "bbarg access info exists already");
+#endif // NDEBUG
+
+  // If the function has no body, conservatively assume that all args are
+  // read + written.
+  if (funcOp.getBody().empty()) {
+    for (BlockArgument bbArg : funcOp.getArguments()) {
+      funcState.readBbArgs[funcOp].insert(bbArg.getArgNumber());
+      funcState.writtenBbArgs[funcOp].insert(bbArg.getArgNumber());
+    }
+
+    return success();
+  }
+
+  for (BlockArgument bbArg : funcOp.getArguments()) {
+    if (!bbArg.getType().isa<TensorType>())
+      continue;
+    bool isRead = state.isValueRead(bbArg);
+    bool isWritten = isValueWritten(bbArg, state, aliasInfo);
+    if (state.getOptions().testAnalysisOnly)
+      annotateFuncArgAccess(funcOp, bbArg, isRead, isWritten);
+    if (isRead)
+      funcState.readBbArgs[funcOp].insert(bbArg.getArgNumber());
+    if (isWritten)
+      funcState.writtenBbArgs[funcOp].insert(bbArg.getArgNumber());
+  }
+
+  return success();
+}
+} // namespace
+
+/// Remove bufferization attributes on FuncOp arguments.
+static void removeBufferizationAttributes(BlockArgument bbArg) {
+  auto funcOp = cast<FuncOp>(bbArg.getOwner()->getParentOp());
+  funcOp.removeArgAttr(bbArg.getArgNumber(),
+                       BufferizationDialect::kBufferLayoutAttrName);
+  funcOp.removeArgAttr(bbArg.getArgNumber(),
+                       BufferizationDialect::kWritableAttrName);
+}
+
+/// Return the FuncOp called by `callOp`.
+static FuncOp getCalledFunction(CallOpInterface callOp) {
+  SymbolRefAttr sym = callOp.getCallableForCallee().dyn_cast<SymbolRefAttr>();
+  if (!sym)
+    return nullptr;
+  return dyn_cast_or_null<FuncOp>(
+      SymbolTable::lookupNearestSymbolFrom(callOp, sym));
+}
+
+/// Gather equivalence info of CallOps.
+/// Note: This only adds new equivalence info if the called function was already
+/// analyzed.
+// TODO: This does not handle cyclic function call graphs etc.
+static void equivalenceAnalysis(FuncOp funcOp,
+                                BufferizationAliasInfo &aliasInfo,
+                                FuncAnalysisState &funcState) {
+  funcOp->walk([&](func::CallOp callOp) {
+    FuncOp calledFunction = getCalledFunction(callOp);
+    assert(calledFunction && "could not retrieved called FuncOp");
+
+    // No equivalence info available for the called function.
+    if (!funcState.equivalentFuncArgs.count(calledFunction))
+      return WalkResult::skip();
+
+    for (auto it : funcState.equivalentFuncArgs[calledFunction]) {
+      int64_t returnIdx = it.first;
+      int64_t bbargIdx = it.second;
+      Value returnVal = callOp.getResult(returnIdx);
+      Value argVal = callOp->getOperand(bbargIdx);
+      aliasInfo.unionEquivalenceClasses(returnVal, argVal);
+    }
+
+    return WalkResult::advance();
+  });
+}
+
+/// Store all functions of the `moduleOp` in `orderedFuncOps`, sorted by
+/// callee-caller order (i.e. callees without callers first).
+/// Store the map of FuncOp to all its callers in `callerMap`.
+/// Return `failure()` if a cycle of calls is detected or if we are unable to
+/// retrieve the called FuncOp from any CallOpInterface.
+static LogicalResult
+getFuncOpsOrderedByCalls(ModuleOp moduleOp,
+                         SmallVectorImpl<FuncOp> &orderedFuncOps,
+                         FuncCallerMap &callerMap) {
+  // For each FuncOp, the set of functions called by it (i.e. the union of
+  // symbols of all nested CallOpInterfaceOp).
+  DenseMap<FuncOp, DenseSet<FuncOp>> calledBy;
+  // For each FuncOp, the number of CallOpInterface it contains.
+  DenseMap<FuncOp, unsigned> numberCallOpsContainedInFuncOp;
+  WalkResult res = moduleOp.walk([&](FuncOp funcOp) -> WalkResult {
+    if (!funcOp.getBody().empty()) {
+      func::ReturnOp returnOp = getAssumedUniqueReturnOp(funcOp);
+      if (!returnOp)
+        return funcOp->emitError()
+               << "cannot bufferize a FuncOp with tensors and "
+                  "without a unique ReturnOp";
+    }
+
+    numberCallOpsContainedInFuncOp[funcOp] = 0;
+    return funcOp.walk([&](CallOpInterface callOp) -> WalkResult {
+      // Only support CallOp for now.
+      if (!isa<func::CallOp>(callOp.getOperation()))
+        return callOp->emitError() << "expected a CallOp";
+      FuncOp calledFunction = getCalledFunction(callOp);
+      assert(calledFunction && "could not retrieved called FuncOp");
+      auto it = callerMap.try_emplace(calledFunction, DenseSet<Operation *>{});
+      it.first->getSecond().insert(callOp);
+      if (calledBy[calledFunction].count(funcOp) == 0) {
+        calledBy[calledFunction].insert(funcOp);
+        numberCallOpsContainedInFuncOp[funcOp]++;
+      }
+      return WalkResult::advance();
+    });
+  });
+  if (res.wasInterrupted())
+    return failure();
+  // Iteratively remove function operation that do not call any of the
+  // functions remaining in the callCounter map and add them to the worklist.
+  while (!numberCallOpsContainedInFuncOp.empty()) {
+    auto it = llvm::find_if(numberCallOpsContainedInFuncOp,
+                            [](auto entry) { return entry.getSecond() == 0; });
+    if (it == numberCallOpsContainedInFuncOp.end())
+      return moduleOp.emitOpError(
+          "expected callgraph to be free of circular dependencies.");
+    orderedFuncOps.push_back(it->getFirst());
+    for (auto callee : calledBy[it->getFirst()])
+      numberCallOpsContainedInFuncOp[callee]--;
+    numberCallOpsContainedInFuncOp.erase(it);
+  }
+  return success();
+}
+
+static void foreachCaller(const FuncCallerMap &callerMap, FuncOp callee,
+                          llvm::function_ref<void(Operation *)> doit) {
+  auto itCallers = callerMap.find(callee);
+  if (itCallers == callerMap.end())
+    return;
+  for (Operation *caller : itCallers->second)
+    doit(caller);
+}
+
+/// Set the attribute that triggers inplace bufferization on a FuncOp argument
+/// `bbArg`.
+static void setInPlaceFuncArgument(BlockArgument bbArg, bool inPlace) {
+  auto funcOp = cast<FuncOp>(bbArg.getOwner()->getParentOp());
+  funcOp.setArgAttr(bbArg.getArgNumber(),
+                    BufferizableOpInterface::kInplaceableAttrName,
+                    BoolAttr::get(bbArg.getContext(), inPlace));
+}
+
+/// Annotate the IR with the result of the analysis. For testing/debugging only.
+static void annotateOpsWithBufferizationMarkers(FuncOp funcOp,
+                                                const AnalysisState &state) {
+  auto bufferizableOp = cast<BufferizableOpInterface>(funcOp.getOperation());
+  for (BlockArgument bbArg : funcOp.getArguments())
+    if (bbArg.getType().isa<TensorType>())
+      setInPlaceFuncArgument(bbArg, bufferizableOp.isWritable(bbArg, state));
+}
+
+/// Fold return values that are memref casts.
+static void foldMemRefCasts(FuncOp funcOp) {
+  if (funcOp.getBody().empty())
+    return;
+
+  func::ReturnOp returnOp = getAssumedUniqueReturnOp(funcOp);
+  SmallVector<Type> resultTypes;
+
+  for (OpOperand &operand : returnOp->getOpOperands()) {
+    if (auto castOp = operand.get().getDefiningOp<memref::CastOp>()) {
+      operand.set(castOp.source());
+      resultTypes.push_back(castOp.source().getType());
+    } else {
+      resultTypes.push_back(operand.get().getType());
+    }
+  }
+
+  auto newFuncType = FunctionType::get(
+      funcOp.getContext(), funcOp.getFunctionType().getInputs(), resultTypes);
+  funcOp.setType(newFuncType);
+}
+
+LogicalResult mlir::bufferization::runOneShotModuleBufferize(
+    ModuleOp moduleOp, OneShotBufferizationOptions options) {
+  IRRewriter rewriter(moduleOp.getContext());
+  OneShotAnalysisState analysisState(moduleOp, options);
+  BufferizationState bufferizationState(analysisState);
+  FuncAnalysisState &funcState = getFuncAnalysisState(analysisState);
+  BufferizationAliasInfo &aliasInfo = analysisState.getAliasInfo();
+
+  // A list of functions in the order in which they are analyzed + bufferized.
+  SmallVector<FuncOp> orderedFuncOps;
+
+  // A mapping of FuncOps to their callers.
+  FuncCallerMap callerMap;
+
+  if (failed(getFuncOpsOrderedByCalls(moduleOp, orderedFuncOps, callerMap)))
+    return failure();
+
+  // Collect bbArg/return value information after the analysis.
+  options.addPostAnalysisStep(aliasingFuncOpBBArgsAnalysis);
+  options.addPostAnalysisStep(funcOpBbArgReadWriteAnalysis);
+
+  // Analyze ops.
+  for (FuncOp funcOp : orderedFuncOps) {
+    // No body => no analysis.
+    if (funcOp.getBody().empty())
+      continue;
+
+    // Now analyzing function.
+    funcState.analyzedFuncOps[funcOp] = FuncOpAnalysisState::InProgress;
+
+    // Gather equivalence info for CallOps.
+    equivalenceAnalysis(funcOp, aliasInfo, funcState);
+
+    // Analyze funcOp.
+    if (failed(analyzeOp(funcOp, analysisState)))
+      return failure();
+
+    // Mark op as fully analyzed.
+    funcState.analyzedFuncOps[funcOp] = FuncOpAnalysisState::Analyzed;
+
+    // Add annotations to function arguments.
+    if (options.testAnalysisOnly)
+      annotateOpsWithBufferizationMarkers(funcOp, analysisState);
+  }
+
+  if (options.testAnalysisOnly)
+    return success();
+
+  // Bufferize functions.
+  for (FuncOp funcOp : orderedFuncOps) {
+    // Note: It would be good to apply cleanups here but we cannot as aliasInfo
+    // would be invalidated.
+    if (failed(bufferizeOp(funcOp, bufferizationState)))
+      return failure();
+    foldMemRefCasts(funcOp);
+  }
+
+  // Check result.
+  for (FuncOp funcOp : orderedFuncOps) {
+    if (!options.allowReturnAllocs &&
+        llvm::any_of(funcOp.getFunctionType().getResults(), [](Type t) {
+          return t.isa<MemRefType, UnrankedMemRefType>();
+        })) {
+      funcOp->emitError("memref return type is unsupported");
+      return failure();
+    }
+  }
+
+  // Finalize all buffers.
+  if (failed(finalizeBuffers(moduleOp, options)))
+    return failure();
+
+  // Post-pass cleanup of function argument attributes.
+  moduleOp.walk([&](FuncOp op) {
+    for (BlockArgument bbArg : op.getArguments())
+      removeBufferizationAttributes(bbArg);
+  });
+
+  return success();
+}

mlir/lib/Dialect/Linalg/ComprehensiveBufferize/CMakeLists.txt

-add_mlir_dialect_library(MLIRModuleBufferization
-  ModuleBufferization.cpp
-
-  LINK_LIBS PUBLIC
-  MLIRBufferization
-  MLIRBufferizationTransforms
-  MLIRFunc
-  MLIRFuncTransforms
-  MLIRIR
-  MLIRMemRef
-)

mlir/lib/Dialect/Linalg/ComprehensiveBufferize/ModuleBufferization.cpp

-//===- ModuleBufferization.cpp - Bufferization across Func. Boundaries ----===//
-//
-// 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
-//
-//===----------------------------------------------------------------------===//
-//
-// Module Bufferization is an extension of One-Shot Bufferize that
-// bufferizes function boundaries. It provides `BufferizableOpInterface`
-// implementations for FuncOp, CallOp and ReturnOp.
-//
-// Module Bufferization is run via `runModuleBufferize(ModuleOp, ...)`. This
-// function analyzes the given module and determines the order of analysis and
-// bufferization: Functions that are called are processed before their
-// respective callers.
-//
-// After analyzing a FuncOp, additional information about its bbArgs is
-// gathered through PostAnalysisStepFns and stored in
-// `FuncAnalysisState`.
-//
-// * `aliasingFuncOpBBArgsAnalysis` determines the equivalent/aliasing bbArgs
-// for
-//   each tensor return value (if any).
-// * `funcOpBbArgReadWriteAnalysis` determines whether or not a tensor bbArg is
-//   read/written.
-//
-// Only tensors that are equivalent to some FuncOp bbArg may be returned.
-// Bufferization currently fails if other tensors (in particular tensors that
-// bufferize out-of-place and result in a new buffer allocation) are returned.
-// In the future, such allocations could be hoisted to the caller.
-//
-// Example: `foo` fails bufferization because %0 is not equivalent to any bbArg.
-// ```
-// func @foo() -> tensor<?xf32> {
-//   %0 = linalg.init_tensor [...] : tensor<?xf32>
-//   return %0 : tensor<?xf32>
-// }
-// ```
-//
-// Module Bufferization implements the following calling convention.
-//
-// * In the absence of conflicts within a FuncOp, the FuncOp's bbArgs may always
-//   be written to in-place.
-// * If a tensor operand of a CallOp is read after the CallOp, the operand of
-//   the CallOp must bufferize out-of-place.
-//
-// Example: The tensor.insert op bufferizes in-place because it is allowed to
-// modify the buffer of `%t1` directly. The CallOp in `caller` must bufferize
-// out-of-place because `%t0` is modified by the callee but read by the
-// tensor.extract op. The analysis of CallOps decides whether an OpOperand must
-// bufferize out-of-place based on results of `funcOpBbArgReadWriteAnalysis`.
-// ```
-// func @callee(%t1 : tensor<?xf32>) -> tensor<?xf32> {
-//   %f = ... : f32
-//   %0 = tensor.insert %f into %t1[...] : tensor<?xf32>
-//   return %0 : tensor<?xf32>
-// }
-//
-// func @caller() -> () {
-//   %t0 = ... : tensor<?xf32>
-//   %1 = call @callee(%t0) : (tensor<?xf32>) -> (tensor<?xf32>)
-//   %2 = tensor.extract %1[...]  : tensor<?xf32>
-// }
-// ```
-//
-// Note: If a function is external, `funcOpBbArgReadWriteAnalysis` cannot
-// analyze the function body. In such a case, the CallOp analysis conservatively
-// assumes that each tensor OpOperand is both read and written.
-//
-// TODO: Add FuncOp attributes so that bbArgs of external FuncOps can be marked
-// as "not reading" and/or "not writing".
-
-#include "mlir/Dialect/Linalg/ComprehensiveBufferize/ModuleBufferization.h"
-
-#include "mlir/Dialect/Bufferization/IR/BufferizableOpInterface.h"
-#include "mlir/Dialect/Bufferization/IR/Bufferization.h"
-#include "mlir/Dialect/Bufferization/Transforms/Bufferize.h"
-#include "mlir/Dialect/Bufferization/Transforms/OneShotAnalysis.h"
-#include "mlir/Dialect/Func/IR/FuncOps.h"
-#include "mlir/Dialect/MemRef/IR/MemRef.h"
-#include "mlir/IR/Operation.h"
-
-using namespace mlir;
-using namespace linalg;
-using namespace tensor;
-using namespace comprehensive_bufferize;
-using namespace mlir::bufferization;
-
-/// A mapping of FuncOps to their callers.
-using FuncCallerMap = DenseMap<FuncOp, DenseSet<Operation *>>;
-
-namespace {
-/// The state of analysis of a FuncOp.
-enum class FuncOpAnalysisState { NotAnalyzed, InProgress, Analyzed };
-
-/// Extra analysis state that is required for bufferization of function
-/// boundaries.
-struct FuncAnalysisState : public DialectAnalysisState {
-  /// A mapping of ReturnOp OpOperand indices to equivalent FuncOp BBArg
-  /// indices.
-  DenseMap<FuncOp, DenseMap<int64_t, int64_t>> equivalentFuncArgs;
-
-  /// A mapping of ReturnOp OpOperand indices to aliasing FuncOp BBArg indices.
-  DenseMap<FuncOp, DenseMap<int64_t, SmallVector<int64_t>>> aliasingFuncArgs;
-
-  /// A mapping of FuncOp BBArg indices to aliasing ReturnOp OpOperand indices.
-  DenseMap<FuncOp, DenseMap<int64_t, SmallVector<int64_t>>> aliasingReturnVals;
-
-  /// A set of all read BlockArguments of FuncOps.
-  DenseMap<FuncOp, DenseSet<int64_t>> readBbArgs;
-
-  /// A set of all written-to BlockArguments of FuncOps.
-  DenseMap<FuncOp, DenseSet<int64_t>> writtenBbArgs;
-
-  /// Keep track of which FuncOps are fully analyzed or currently being
-  /// analyzed.
-  DenseMap<FuncOp, FuncOpAnalysisState> analyzedFuncOps;
-};
-} // namespace
-
-/// Get FuncAnalysisState.
-static const FuncAnalysisState &
-getFuncAnalysisState(const AnalysisState &state) {
-  Optional<const FuncAnalysisState *> maybeState =
-      state.getDialectState<FuncAnalysisState>(
-          func::FuncDialect::getDialectNamespace());
-  assert(maybeState.hasValue() && "FuncAnalysisState does not exist");
-  return **maybeState;
-}
-
-/// Get or create FuncAnalysisState.
-static FuncAnalysisState &getFuncAnalysisState(AnalysisState &state) {
-  return state.getOrCreateDialectState<FuncAnalysisState>(
-      func::FuncDialect::getDialectNamespace());
-}
-
-/// Return the state (phase) of analysis of the FuncOp.
-static FuncOpAnalysisState getFuncOpAnalysisState(const AnalysisState &state,
-                                                  FuncOp funcOp) {
-  const FuncAnalysisState &moduleState = getFuncAnalysisState(state);
-  auto it = moduleState.analyzedFuncOps.find(funcOp);
-  if (it == moduleState.analyzedFuncOps.end())
-    return FuncOpAnalysisState::NotAnalyzed;
-  return it->second;
-}
-
-/// Return the unique ReturnOp that terminates `funcOp`.
-/// Return nullptr if there is no such unique ReturnOp.
-static func::ReturnOp getAssumedUniqueReturnOp(FuncOp funcOp) {
-  func::ReturnOp returnOp;
-  for (Block &b : funcOp.getBody()) {
-    if (auto candidateOp = dyn_cast<func::ReturnOp>(b.getTerminator())) {
-      if (returnOp)
-        return nullptr;
-      returnOp = candidateOp;
-    }
-  }
-  return returnOp;
-}
-
-namespace {
-
-/// Annotate IR with the results of the analysis. For testing purposes only.
-static void annotateEquivalentReturnBbArg(OpOperand &returnVal,
-                                          BlockArgument bbArg) {
-  const char *kEquivalentArgsAttr = "__equivalent_func_args__";
-  Operation *op = returnVal.getOwner();
-
-  SmallVector<int64_t> equivBbArgs;
-  if (op->hasAttr(kEquivalentArgsAttr)) {
-    auto attr = op->getAttr(kEquivalentArgsAttr).cast<ArrayAttr>();
-    equivBbArgs = llvm::to_vector<4>(llvm::map_range(attr, [](Attribute a) {
-      return a.cast<IntegerAttr>().getValue().getSExtValue();
-    }));
-  } else {
-    equivBbArgs.append(op->getNumOperands(), -1);
-  }
-  equivBbArgs[returnVal.getOperandNumber()] = bbArg.getArgNumber();
-
-  OpBuilder b(op->getContext());
-  op->setAttr(kEquivalentArgsAttr, b.getI64ArrayAttr(equivBbArgs));
-}
-
-/// Store function BlockArguments that are equivalent to/aliasing a returned
-/// value in FuncAnalysisState.
-static LogicalResult
-aliasingFuncOpBBArgsAnalysis(Operation *op, AnalysisState &state,
-                             BufferizationAliasInfo &aliasInfo,
-                             SmallVector<Operation *> &newOps) {
-  FuncAnalysisState &moduleState = getFuncAnalysisState(state);
-
-  // Support only single return-terminated block in the function.
-  auto funcOp = cast<FuncOp>(op);
-  func::ReturnOp returnOp = getAssumedUniqueReturnOp(funcOp);
-  assert(returnOp && "expected func with single return op");
-
-  // Initialize data structure.
-  auto createdEquiv = moduleState.equivalentFuncArgs.try_emplace(
-      funcOp, DenseMap<int64_t, int64_t>());
-  auto createdAliasingOperands = moduleState.aliasingFuncArgs.try_emplace(
-      funcOp, DenseMap<int64_t, SmallVector<int64_t>>());
-  auto createdAliasingResults = moduleState.aliasingReturnVals.try_emplace(
-      funcOp, DenseMap<int64_t, SmallVector<int64_t>>());
-  (void)createdEquiv;
-  (void)createdAliasingOperands;
-  (void)createdAliasingResults;
-#ifndef NDEBUG
-  assert(createdEquiv.second && "equivalence info exists already");
-  assert(createdAliasingOperands.second && "aliasing info exists already");
-  assert(createdAliasingResults.second && "aliasing info exists already");
-#endif // NDEBUG
-
-  for (OpOperand &returnVal : returnOp->getOpOperands())
-    if (returnVal.get().getType().isa<RankedTensorType>())
-      for (BlockArgument bbArg : funcOp.getArguments())
-        if (bbArg.getType().isa<RankedTensorType>()) {
-          int64_t returnIdx = returnVal.getOperandNumber();
-          int64_t bbArgIdx = bbArg.getArgNumber();
-          if (aliasInfo.areEquivalentBufferizedValues(returnVal.get(), bbArg)) {
-            moduleState.equivalentFuncArgs[funcOp][returnIdx] = bbArgIdx;
-            if (state.getOptions().testAnalysisOnly)
-              annotateEquivalentReturnBbArg(returnVal, bbArg);
-          }
-          if (aliasInfo.areAliasingBufferizedValues(returnVal.get(), bbArg)) {
-            moduleState.aliasingFuncArgs[funcOp][returnIdx].push_back(bbArgIdx);
-            moduleState.aliasingReturnVals[funcOp][bbArgIdx].push_back(
-                returnIdx);
-          }
-        }
-
-  return success();
-}
-
-/// Return true if the buffer of the given tensor value is written to. Must not
-/// be called for values inside not yet analyzed functions. (Post-analysis
-/// steps do not have to be run yet, i.e., "in progress" is also OK.)
-static bool isValueWritten(Value value, const AnalysisState &state,
-                           const BufferizationAliasInfo &aliasInfo) {
-#ifndef NDEBUG
-  assert(value.getType().isa<TensorType>() && "expected TensorType");
-  FuncOp funcOp;
-  if (auto bbArg = value.dyn_cast<BlockArgument>()) {
-    Operation *owner = bbArg.getOwner()->getParentOp();
-    funcOp = isa<FuncOp>(owner) ? cast<FuncOp>(owner)
-                                : owner->getParentOfType<FuncOp>();
-  } else {
-    funcOp = value.getDefiningOp()->getParentOfType<FuncOp>();
-  }
-  assert(getFuncOpAnalysisState(state, funcOp) !=
-             FuncOpAnalysisState::NotAnalyzed &&
-         "FuncOp must be fully analyzed or analysis in progress");
-#endif // NDEBUG
-
-  bool isWritten = false;
-  aliasInfo.applyOnAliases(value, [&](Value val) {
-    for (OpOperand &use : val.getUses())
-      if (state.isInPlace(use) && state.bufferizesToMemoryWrite(use))
-        isWritten = true;
-  });
-  return isWritten;
-}
-
-static void annotateFuncArgAccess(FuncOp funcOp, BlockArgument bbArg,
-                                  bool isRead, bool isWritten) {
-  OpBuilder b(funcOp.getContext());
-  Attribute accessType;
-  if (isRead && isWritten) {
-    accessType = b.getStringAttr("read-write");
-  } else if (isRead) {
-    accessType = b.getStringAttr("read");
-  } else if (isWritten) {
-    accessType = b.getStringAttr("write");
-  } else {
-    accessType = b.getStringAttr("none");
-  }
-  funcOp.setArgAttr(bbArg.getArgNumber(), "bufferization.access", accessType);
-}
-
-/// Determine which FuncOp bbArgs are read and which are written. If this
-/// PostAnalysisStepFn is run on a function with unknown ops, it will
-/// conservatively assume that such ops bufferize to a read + write.
-static LogicalResult
-funcOpBbArgReadWriteAnalysis(Operation *op, AnalysisState &state,
-                             BufferizationAliasInfo &aliasInfo,
-                             SmallVector<Operation *> &newOps) {
-  FuncAnalysisState &moduleState = getFuncAnalysisState(state);
-  auto funcOp = cast<FuncOp>(op);
-
-  // Initialize data structure.
-  auto createdRead =
-      moduleState.readBbArgs.try_emplace(funcOp, DenseSet<int64_t>());
-  auto createdWritten =
-      moduleState.writtenBbArgs.try_emplace(funcOp, DenseSet<int64_t>());
-  (void)createdRead;
-  (void)createdWritten;
-#ifndef NDEBUG
-  assert(createdRead.second && "bbarg access info exists already");
-  assert(createdWritten.second && "bbarg access info exists already");
-#endif // NDEBUG
-
-  // If the function has no body, conservatively assume that all args are
-  // read + written.
-  if (funcOp.getBody().empty()) {
-    for (BlockArgument bbArg : funcOp.getArguments()) {
-      moduleState.readBbArgs[funcOp].insert(bbArg.getArgNumber());
-      moduleState.writtenBbArgs[funcOp].insert(bbArg.getArgNumber());
-    }
-
-    return success();
-  }
-
-  for (BlockArgument bbArg : funcOp.getArguments()) {
-    if (!bbArg.getType().isa<TensorType>())
-      continue;
-    bool isRead = state.isValueRead(bbArg);
-    bool isWritten = isValueWritten(bbArg, state, aliasInfo);
-    if (state.getOptions().testAnalysisOnly)
-      annotateFuncArgAccess(funcOp, bbArg, isRead, isWritten);
-    if (isRead)
-      moduleState.readBbArgs[funcOp].insert(bbArg.getArgNumber());
-    if (isWritten)
-      moduleState.writtenBbArgs[funcOp].insert(bbArg.getArgNumber());
-  }
-
-  return success();
-}
-} // namespace
-
-/// If `value` is a memref::CastOp, return its source. Otherwise, return
-/// `value` directly.
-static Value getNonCastedValue(Value value) {
-  while (auto castOp = value.getDefiningOp<memref::CastOp>())
-    value = castOp.source();
-  return value;
-}
-
-/// Remove the attribute that triggers inplace bufferization on a FuncOp
-/// argument `bbArg`.
-static void removeBufferizationFuncArguments(BlockArgument bbArg) {
-  auto funcOp = cast<FuncOp>(bbArg.getOwner()->getParentOp());
-  funcOp.removeArgAttr(bbArg.getArgNumber(),
-                       BufferizableOpInterface::kBufferLayoutAttrName);
-  funcOp.removeArgAttr(bbArg.getArgNumber(),
-                       BufferizableOpInterface::kInplaceableAttrName);
-}
-
-/// Return the FuncOp called by `callOp`.
-static FuncOp getCalledFunction(CallOpInterface callOp) {
-  SymbolRefAttr sym = callOp.getCallableForCallee().dyn_cast<SymbolRefAttr>();
-  if (!sym)
-    return nullptr;
-  return dyn_cast_or_null<FuncOp>(
-      SymbolTable::lookupNearestSymbolFrom(callOp, sym));
-}
-
-/// Return the index-th bufferized function argument type. This assumes that the
-/// specified argument is a tensor. If the tensor is ranked, a layout map may be
-/// specified by the user. If no layout map is specified, a fully dynamic map is
-/// used.
-static BaseMemRefType
-getBufferizedFunctionArgType(FuncOp funcOp, int64_t index,
-                             const BufferizationOptions &options) {
-  auto tensorType =
-      funcOp.getFunctionType().getInput(index).dyn_cast<TensorType>();
-  assert(tensorType && "expected TensorType");
-  BaseMemRefType memrefType = getMemRefType(tensorType, options);
-
-  auto layoutAttr = funcOp.getArgAttrOfType<AffineMapAttr>(
-      index, BufferizableOpInterface::kBufferLayoutAttrName);
-  if (!layoutAttr)
-    return memrefType;
-
-  auto rankedMemrefType = memrefType.dyn_cast<MemRefType>();
-  assert(rankedMemrefType && "buffer layout not supported on unranked tensors");
-  return MemRefType::get(
-      rankedMemrefType.getShape(), rankedMemrefType.getElementType(),
-      layoutAttr.getValue(), rankedMemrefType.getMemorySpaceAsInt());
-}
-
-/// Gather equivalence info of CallOps.
-/// Note: This only adds new equivalence info if the called function was already
-/// analyzed.
-// TODO: This does not handle cyclic function call graphs etc.
-static void equivalenceAnalysis(FuncOp funcOp,
-                                BufferizationAliasInfo &aliasInfo,
-                                FuncAnalysisState &moduleState) {
-  funcOp->walk([&](func::CallOp callOp) {
-    FuncOp calledFunction = getCalledFunction(callOp);
-    assert(calledFunction && "could not retrieved called FuncOp");
-
-    // No equivalence info available for the called function.
-    if (!moduleState.equivalentFuncArgs.count(calledFunction))
-      return WalkResult::skip();
-
-    for (auto it : moduleState.equivalentFuncArgs[calledFunction]) {
-      int64_t returnIdx = it.first;
-      int64_t bbargIdx = it.second;
-      Value returnVal = callOp.getResult(returnIdx);
-      Value argVal = callOp->getOperand(bbargIdx);
-      aliasInfo.unionEquivalenceClasses(returnVal, argVal);
-    }
-
-    return WalkResult::advance();
-  });
-}
-
-/// Return the index of the bbArg in the given FuncOp that is equivalent to the
-/// specified return value (if any).
-static Optional<int64_t> getEquivalentFuncArgIdx(FuncOp funcOp,
-                                                 const FuncAnalysisState &state,
-                                                 int64_t returnValIdx) {
-  auto funcOpIt = state.equivalentFuncArgs.find(funcOp);
-  if (funcOpIt == state.equivalentFuncArgs.end())
-    // No equivalence info stores for funcOp.
-    return None;
-
-  auto retValIt = funcOpIt->getSecond().find(returnValIdx);
-  if (retValIt == funcOpIt->getSecond().end())
-    // Return value has no equivalent bbArg.
-    return None;
-
-  return retValIt->getSecond();
-}
-
-/// Store all functions of the `moduleOp` in `orderedFuncOps`, sorted by
-/// callee-caller order (i.e. callees without callers first).
-/// Store the map of FuncOp to all its callers in `callerMap`.
-/// Return `failure()` if a cycle of calls is detected or if we are unable to
-/// retrieve the called FuncOp from any CallOpInterface.
-static LogicalResult
-getFuncOpsOrderedByCalls(ModuleOp moduleOp,
-                         SmallVectorImpl<FuncOp> &orderedFuncOps,
-                         FuncCallerMap &callerMap) {
-  // For each FuncOp, the set of functions called by it (i.e. the union of
-  // symbols of all nested CallOpInterfaceOp).
-  DenseMap<FuncOp, DenseSet<FuncOp>> calledBy;
-  // For each FuncOp, the number of CallOpInterface it contains.
-  DenseMap<FuncOp, unsigned> numberCallOpsContainedInFuncOp;
-  WalkResult res = moduleOp.walk([&](FuncOp funcOp) -> WalkResult {
-    if (!funcOp.getBody().empty()) {
-      func::ReturnOp returnOp = getAssumedUniqueReturnOp(funcOp);
-      if (!returnOp)
-        return funcOp->emitError()
-               << "cannot bufferize a FuncOp with tensors and "
-                  "without a unique ReturnOp";
-    }
-
-    numberCallOpsContainedInFuncOp[funcOp] = 0;
-    return funcOp.walk([&](CallOpInterface callOp) -> WalkResult {
-      // Only support CallOp for now.
-      if (!isa<func::CallOp>(callOp.getOperation()))
-        return callOp->emitError() << "expected a CallOp";
-      FuncOp calledFunction = getCalledFunction(callOp);
-      assert(calledFunction && "could not retrieved called FuncOp");
-      auto it = callerMap.try_emplace(calledFunction, DenseSet<Operation *>{});
-      it.first->getSecond().insert(callOp);
-      if (calledBy[calledFunction].count(funcOp) == 0) {
-        calledBy[calledFunction].insert(funcOp);
-        numberCallOpsContainedInFuncOp[funcOp]++;
-      }
-      return WalkResult::advance();
-    });
-  });
-  if (res.wasInterrupted())
-    return failure();
-  // Iteratively remove function operation that do not call any of the
-  // functions remaining in the callCounter map and add them to the worklist.
-  while (!numberCallOpsContainedInFuncOp.empty()) {
-    auto it = llvm::find_if(numberCallOpsContainedInFuncOp,
-                            [](auto entry) { return entry.getSecond() == 0; });
-    if (it == numberCallOpsContainedInFuncOp.end())
-      return moduleOp.emitOpError(
-          "expected callgraph to be free of circular dependencies.");
-    orderedFuncOps.push_back(it->getFirst());
-    for (auto callee : calledBy[it->getFirst()])
-      numberCallOpsContainedInFuncOp[callee]--;
-    numberCallOpsContainedInFuncOp.erase(it);
-  }
-  return success();
-}
-
-static void foreachCaller(const FuncCallerMap &callerMap, FuncOp callee,
-                          llvm::function_ref<void(Operation *)> doit) {
-  auto itCallers = callerMap.find(callee);
-  if (itCallers == callerMap.end())
-    return;
-  for (Operation *caller : itCallers->second)
-    doit(caller);
-}
-
-namespace mlir {
-namespace linalg {
-namespace comprehensive_bufferize {
-namespace std_ext {
-
-struct CallOpInterface
-    : public BufferizableOpInterface::ExternalModel<CallOpInterface,
-                                                    func::CallOp> {
-  bool bufferizesToMemoryRead(Operation *op, OpOperand &opOperand,
-                              const AnalysisState &state) const {
-    func::CallOp callOp = cast<func::CallOp>(op);
-    FuncOp funcOp = getCalledFunction(callOp);
-    assert(funcOp && "expected CallOp to a FuncOp");
-
-    const FuncAnalysisState &moduleState = getFuncAnalysisState(state);
-    if (getFuncOpAnalysisState(state, funcOp) != FuncOpAnalysisState::Analyzed)
-      // FuncOp not analyzed yet. Assume that OpOperand is read.
-      return true;
-
-    auto it = moduleState.readBbArgs.find(funcOp);
-    assert(it != moduleState.readBbArgs.end() &&
-           "expected analysis info for analyzed FuncOps");
-    return it->second.contains(opOperand.getOperandNumber());
-  }
-
-  bool bufferizesToMemoryWrite(Operation *op, OpOperand &opOperand,
-                               const AnalysisState &state) const {
-    func::CallOp callOp = cast<func::CallOp>(op);
-    FuncOp funcOp = getCalledFunction(callOp);
-    assert(funcOp && "expected CallOp to a FuncOp");
-
-    const FuncAnalysisState &moduleState = getFuncAnalysisState(state);
-    if (getFuncOpAnalysisState(state, funcOp) != FuncOpAnalysisState::Analyzed)
-      // FuncOp not analyzed yet. Assume that OpOperand is written.
-      return true;
-
-    auto it = moduleState.writtenBbArgs.find(funcOp);
-    assert(it != moduleState.writtenBbArgs.end() &&
-           "expected analysis info for analyzed FuncOps");
-    return it->second.contains(opOperand.getOperandNumber());
-  }
-
-  SmallVector<OpResult> getAliasingOpResult(Operation *op, OpOperand &opOperand,
-                                            const AnalysisState &state) const {
-    func::CallOp callOp = cast<func::CallOp>(op);
-    FuncOp funcOp = getCalledFunction(callOp);
-    assert(funcOp && "expected CallOp to a FuncOp");
-    const FuncAnalysisState &moduleState = getFuncAnalysisState(state);
-    if (getFuncOpAnalysisState(state, funcOp) !=
-        FuncOpAnalysisState::Analyzed) {
-      // FuncOp not analyzed yet. Any OpResult may be aliasing.
-      SmallVector<OpResult> result;
-      for (OpResult opResult : op->getOpResults())
-        if (opResult.getType().isa<TensorType>())
-          result.push_back(opResult);
-      return result;
-    }
-
-    // Get aliasing results from state.
-    auto mapping = moduleState.aliasingReturnVals.find(funcOp);
-    assert(mapping != moduleState.aliasingReturnVals.end() &&
-           "expected analysis info analyzed FuncOps");
-    auto aliasingReturnVals =
-        mapping->second.find(opOperand.getOperandNumber());
-    if (aliasingReturnVals == mapping->second.end())
-      return {};
-
-    SmallVector<OpResult> result;
-    for (int64_t resultIdx : aliasingReturnVals->second)
-      result.push_back(callOp->getOpResult(resultIdx));
-    return result;
-  }
-
-  SmallVector<OpOperand *>
-  getAliasingOpOperand(Operation *op, OpResult opResult,
-                       const AnalysisState &state) const {
-    func::CallOp callOp = cast<func::CallOp>(op);
-    FuncOp funcOp = getCalledFunction(callOp);
-    assert(funcOp && "expected CallOp to a FuncOp");
-    const FuncAnalysisState &moduleState = getFuncAnalysisState(state);
-    if (getFuncOpAnalysisState(state, funcOp) !=
-        FuncOpAnalysisState::Analyzed) {
-      // FuncOp not analyzed yet. Any OpOperand may be aliasing.
-      SmallVector<OpOperand *> result;
-      for (OpOperand &opOperand : op->getOpOperands())
-        if (opOperand.get().getType().isa<TensorType>())
-          result.push_back(&opOperand);
-      return result;
-    }
-
-    // Get aliasing bbArgs from state.
-    auto mapping = moduleState.aliasingFuncArgs.find(funcOp);
-    assert(mapping != moduleState.aliasingFuncArgs.end() &&
-           "expected analysis info analyzed FuncOps");
-    auto aliasingFuncArgs = mapping->second.find(opResult.getResultNumber());
-    if (aliasingFuncArgs == mapping->second.end())
-      return {};
-
-    SmallVector<OpOperand *> result;
-    for (int64_t bbArgIdx : aliasingFuncArgs->second)
-      result.push_back(&callOp->getOpOperand(bbArgIdx));
-    return result;
-  }
-
-  BufferRelation bufferRelation(Operation *op, OpResult opResult,
-                                const AnalysisState &state) const {
-    return BufferRelation::Equivalent;
-  }
-
-  /// All function arguments are writable. It is the responsibility of the
-  /// CallOp to insert buffer copies where necessary.
-  LogicalResult bufferize(Operation *op, RewriterBase &rewriter,
-                          BufferizationState &state) const {
-    func::CallOp callOp = cast<func::CallOp>(op);
-    unsigned numResults = callOp.getNumResults();
-    unsigned numOperands = callOp->getNumOperands();
-    FuncOp funcOp = getCalledFunction(callOp);
-    assert(funcOp && "expected CallOp to a FuncOp");
-    const FuncAnalysisState &moduleState =
-        getFuncAnalysisState(state.getAnalysisState());
-    const OneShotBufferizationOptions &options =
-        static_cast<const OneShotBufferizationOptions &>(state.getOptions());
-
-    // Result types of the bufferized CallOp.
-    SmallVector<Type> resultTypes;
-    // Replacement values for the existing CallOp. These are usually the results
-    // of the bufferized CallOp, unless a tensor result folds onto an operand.
-    SmallVector<Value> replacementValues(numResults, Value());
-    // For non-tensor results: A mapping from return val indices of the old
-    // CallOp to return val indices of the bufferized CallOp.
-    SmallVector<Optional<unsigned>> retValMapping(numResults, None);
-    // Operands of the bufferized CallOp.
-    SmallVector<Value> newOperands(numOperands, Value());
-
-    // Based on previously gathered equivalence information, we know if a
-    // tensor result folds onto an operand. These are the only tensor value
-    // results that are supported at the moment.
-    //
-    // For tensors return values that do not fold onto an operand, additional
-    // work is needed (TODO) to either:
-    // * hoist a result into an inplaceable operand or
-    // * devise a better representation to truly return a buffer.
-    //
-    // Note: If a function has no body, no equivalence information is
-    // available. Consequently, a tensor return value cannot be proven to fold
-    // onto a FuncOp bbArg, so calls to such functions are not bufferizable at
-    // the moment.
-
-    // 1. Compute the result types of the new CallOp. Tensor results that are
-    // equivalent to a FuncOp bbArg are no longer returned.
-    for (const auto &it : llvm::enumerate(callOp.getResultTypes())) {
-      unsigned returnValIdx = it.index();
-      Type returnType = it.value();
-      if (!returnType.isa<TensorType>()) {
-        // Non-tensor values are returned.
-        retValMapping[returnValIdx] = resultTypes.size();
-        resultTypes.push_back(returnType);
-        continue;
-      }
-
-      if (Optional<int64_t> bbArgIdx =
-              getEquivalentFuncArgIdx(funcOp, moduleState, returnValIdx)) {
-        // Return operands that are equivalent to some bbArg, are not
-        // returned.
-        FailureOr<Value> bufferOrFailure =
-            state.getBuffer(rewriter, callOp->getOpOperand(*bbArgIdx));
-        if (failed(bufferOrFailure))
-          return failure();
-        replacementValues[returnValIdx] = *bufferOrFailure;
-        newOperands[*bbArgIdx] = *bufferOrFailure;
-        continue;
-      }
-
-      if (!options.allowReturnAllocs)
-        return callOp->emitError(
-            "call to FuncOp that returns non-equivalent tensors not supported");
-
-      // Returning a memref. This memref is not equivalent to any bbArg. It is
-      // likely a newly allocated buffer. We may want to hoist such allocations
-      // to the call site in the future.
-      retValMapping[returnValIdx] = resultTypes.size();
-      resultTypes.push_back(
-          funcOp.getFunctionType().getResult(resultTypes.size()));
-    }
-
-    // 2. Compute bufferized FunctionType.
-    FunctionType bufferizedFuncType = funcOp.getFunctionType();
-
-    // 3. Rewrite tensor operands as memrefs based on `bufferizedFuncType`.
-    for (OpOperand &opOperand : callOp->getOpOperands()) {
-      unsigned idx = opOperand.getOperandNumber();
-      Value tensorOperand = opOperand.get();
-
-      // Non-tensor operands are just copied.
-      if (!tensorOperand.getType().isa<TensorType>()) {
-        newOperands[idx] = tensorOperand;
-        continue;
-      }
-
-      // Retrieve buffers for tensor operands. Tensor operand buffers, who's
-      // corresponding FuncOp bbArgs are equivalent to a returned tensor, were
-      // already stored in `newOperands` during Step 1.
-      Value buffer = newOperands[idx];
-      if (!buffer) {
-        FailureOr<Value> bufferOrFailure = state.getBuffer(rewriter, opOperand);
-        if (failed(bufferOrFailure))
-          return failure();
-        buffer = *bufferOrFailure;
-      }
-
-      // Caller / callee type mismatch is handled with a CastOp.
-      auto memRefType = bufferizedFuncType.getInput(idx);
-      // Since we don't yet have a clear layout story, to_memref may
-      // conservatively turn tensors into more dynamic memref than necessary.
-      // If the memref type of the callee fails, introduce an extra memref.cast
-      // that will either canonicalize away or fail compilation until we can do
-      // something better.
-      if (buffer.getType() != memRefType) {
-        assert(
-            memref::CastOp::areCastCompatible(buffer.getType(), memRefType) &&
-            "CallOp::bufferize: cast incompatible");
-        Value castBuffer = rewriter.create<memref::CastOp>(callOp.getLoc(),
-                                                           memRefType, buffer);
-        buffer = castBuffer;
-      }
-      newOperands[idx] = buffer;
-    }
-
-    // 4. Create the new CallOp.
-    Operation *newCallOp = rewriter.create<func::CallOp>(
-        callOp.getLoc(), funcOp.getSymName(), resultTypes, newOperands);
-    newCallOp->setAttrs(callOp->getAttrs());
-    // Get replacement values for non-tensor / non-equivalent results.
-    for (unsigned i = 0; i < replacementValues.size(); ++i) {
-      if (replacementValues[i])
-        continue;
-      replacementValues[i] = newCallOp->getResult(*retValMapping[i]);
-    }
-
-    // 5. Replace the old op with the new op.
-    replaceOpWithBufferizedValues(rewriter, callOp, replacementValues);
-
-    return success();
-  }
-};
-
-struct ReturnOpInterface
-    : public BufferizableOpInterface::ExternalModel<ReturnOpInterface,
-                                                    func::ReturnOp> {
-  bool bufferizesToMemoryRead(Operation *op, OpOperand &opOperand,
-                              const AnalysisState &state) const {
-    return true;
-  }
-
-  bool bufferizesToMemoryWrite(Operation *op, OpOperand &opOperand,
-                               const AnalysisState &state) const {
-    return false;
-  }
-
-  SmallVector<OpResult> getAliasingOpResult(Operation *op, OpOperand &opOperand,
-                                            const AnalysisState &state) const {
-    return {};
-  }
-
-  LogicalResult bufferize(Operation *op, RewriterBase &rewriter,
-                          BufferizationState &state) const {
-#ifndef NDEBUG
-    auto returnOp = cast<func::ReturnOp>(op);
-    assert(isa<FuncOp>(returnOp->getParentOp()) &&
-           "only support FuncOp parent for ReturnOp");
-#endif // NDEBUG
-
-    // ReturnOps are bufferized as part of FuncOps.
-    return failure();
-  }
-};
-
-struct FuncOpInterface
-    : public BufferizableOpInterface::ExternalModel<FuncOpInterface, FuncOp> {
-  LogicalResult bufferize(Operation *op, RewriterBase &rewriter,
-                          BufferizationState &state) const {
-    // Rewrite function bbArgs and return values into buffer form (using the
-    // canonical memref layout for now).
-    //
-    // This relies on a buffer equivalence analysis of each return operand. When
-    // a result buffer is equivalent to a function bbArg, it is dropped from the
-    // return values and becomes inplaceable at all callers.
-    //
-    // All function bbArgs are writable unless they are explicitly marked as
-    // read-only. Callers must insert copies when needed.
-    //
-    // Note: Returning a memref is possible, but corresponding CallOp
-    // bufferizations fail unless `allowReturnAllocs`.
-    auto funcOp = cast<FuncOp>(op);
-    FunctionType funcType = funcOp.getFunctionType();
-    const FuncAnalysisState &moduleState =
-        getFuncAnalysisState(state.getAnalysisState());
-    const BufferizationOptions &options = state.getOptions();
-
-    // Construct the bufferized function type.
-    SmallVector<Type> argTypes;
-    for (const auto &it : llvm::enumerate(funcType.getInputs())) {
-      Type argType = it.value();
-      if (auto tensorType = argType.dyn_cast<TensorType>()) {
-        argTypes.push_back(
-            getBufferizedFunctionArgType(funcOp, it.index(), options));
-        continue;
-      }
-      argTypes.push_back(argType);
-    }
-
-    // Bodiless functions are assumed opaque and we cannot know the
-    // bufferization contract they want to enforce. As a consequence, only
-    // support functions that don't return any tensors atm.
-    if (funcOp.getBody().empty()) {
-      SmallVector<Type> retTypes;
-      for (Type resultType : funcType.getResults()) {
-        if (resultType.isa<TensorType>())
-          return funcOp->emitError() << "cannot bufferize bodiless function "
-                                     << "that returns a tensor";
-        retTypes.push_back(resultType);
-      }
-      funcOp.setType(FunctionType::get(op->getContext(), argTypes, retTypes));
-      return success();
-    }
-
-    // TODO: Support functions with multiple returns.
-    func::ReturnOp returnOp = getAssumedUniqueReturnOp(funcOp);
-    assert(returnOp && "expected func with single return op");
-
-    // 1. Rewrite the bbArgs. Turn every tensor bbArg into a memref bbArg.
-    Block &frontBlock = funcOp.getBody().front();
-    for (BlockArgument &bbArg : frontBlock.getArguments()) {
-      auto tensorType = bbArg.getType().dyn_cast<TensorType>();
-      // Non-tensor types stay the same.
-      if (!tensorType)
-        continue;
-
-      // Collect all uses of the bbArg.
-      SmallVector<OpOperand *> bbArgUses;
-      for (OpOperand &use : bbArg.getUses())
-        bbArgUses.push_back(&use);
-
-      // Change the bbArg type to memref.
-      Type memrefType =
-          getBufferizedFunctionArgType(funcOp, bbArg.getArgNumber(), options);
-      bbArg.setType(memrefType);
-
-      // Replace all uses of the original tensor bbArg.
-      rewriter.setInsertionPointToStart(&frontBlock);
-      if (!bbArgUses.empty()) {
-        // Insert to_tensor because the remaining function body has not been
-        // bufferized yet.
-        Value toTensorOp =
-            rewriter.create<bufferization::ToTensorOp>(funcOp.getLoc(), bbArg);
-        for (OpOperand *use : bbArgUses)
-          use->set(toTensorOp);
-      }
-    }
-
-    // 2. For each result, keep track of which inplace argument it reuses.
-    SmallVector<Value> returnValues;
-    for (OpOperand &returnOperand : returnOp->getOpOperands()) {
-      Value returnVal = returnOperand.get();
-
-      // If not a tensor type just forward it.
-      if (!returnVal.getType().isa<RankedTensorType>()) {
-        returnValues.push_back(returnVal);
-        continue;
-      }
-
-      // If return operand is equivalent to some bbArg, no need to return it.
-      if (Optional<int64_t> equivBbArgIdx = getEquivalentFuncArgIdx(
-              funcOp, moduleState, returnOperand.getOperandNumber())) {
-        rewriter.setInsertionPoint(returnOp);
-        Location loc = returnOp.getLoc();
-        Value toMemrefOp = rewriter.create<bufferization::ToMemrefOp>(
-            loc, getMemRefType(returnVal.getType().cast<TensorType>(), options),
-            returnVal);
-        BlockArgument equivBbArg = funcOp.getArgument(*equivBbArgIdx);
-        // Note: This copy will fold away. It must be inserted here to ensure
-        // that `returnVal` still has at least one use and does not fold away.
-        if (failed(
-                createMemCpy(rewriter, loc, toMemrefOp, equivBbArg, options)))
-          return funcOp->emitError("could not generate copy for bbArg");
-        continue;
-      }
-
-      // Cast values at the call site if necessary.
-      returnValues.push_back(
-          getNonCastedValue(*state.getBuffer(rewriter, returnOperand)));
-    }
-
-    // 3. Rewrite the terminator without the in-place bufferizable values.
-    returnOp.operandsMutable().assign(returnValues);
-
-    // 4. Rewrite the FuncOp type to buffer form.
-    funcOp.setType(FunctionType::get(op->getContext(), argTypes,
-                                     ValueRange(returnValues).getTypes()));
-
-    return success();
-  }
-
-  /// Return `true` if the given function argument is writable.
-  bool isWritable(Operation *op, Value value,
-                  const AnalysisState &state) const {
-    auto funcOp = cast<FuncOp>(op);
-    BlockArgument bbArg = value.dyn_cast<BlockArgument>();
-    assert(bbArg && "expected BlockArgument");
-
-    // "linalg.inplaceable" overrides other writability decisions. This is
-    // currently used for testing only.
-    if (BoolAttr inplaceAttr = funcOp.getArgAttrOfType<BoolAttr>(
-            bbArg.getArgNumber(),
-            BufferizableOpInterface::kInplaceableAttrName))
-      return inplaceAttr.getValue();
-
-    // All function arguments are writable by default.
-    return true;
-  }
-
-  bool isAllocationHoistingBarrier(Operation *op) const { return true; }
-};
-
-} // namespace std_ext
-} // namespace comprehensive_bufferize
-} // namespace linalg
-} // namespace mlir
-
-void mlir::linalg::comprehensive_bufferize::std_ext::
-    registerModuleBufferizationExternalModels(DialectRegistry &registry) {
-  registry.addExtension(+[](MLIRContext *ctx, func::FuncDialect *dialect) {
-    func::CallOp::attachInterface<std_ext::CallOpInterface>(*ctx);
-    func::ReturnOp::attachInterface<std_ext::ReturnOpInterface>(*ctx);
-    func::FuncOp::attachInterface<std_ext::FuncOpInterface>(*ctx);
-  });
-}
-
-/// Set the attribute that triggers inplace bufferization on a FuncOp argument
-/// `bbArg`.
-static void setInPlaceFuncArgument(BlockArgument bbArg, bool inPlace) {
-  auto funcOp = cast<FuncOp>(bbArg.getOwner()->getParentOp());
-  funcOp.setArgAttr(bbArg.getArgNumber(),
-                    BufferizableOpInterface::kInplaceableAttrName,
-                    BoolAttr::get(bbArg.getContext(), inPlace));
-}
-
-/// Annotate the IR with the result of the analysis. For testing/debugging only.
-static void annotateOpsWithBufferizationMarkers(FuncOp funcOp,
-                                                const AnalysisState &state) {
-  auto bufferizableOp = cast<BufferizableOpInterface>(funcOp.getOperation());
-  for (BlockArgument bbArg : funcOp.getArguments())
-    if (bbArg.getType().isa<TensorType>())
-      setInPlaceFuncArgument(bbArg, bufferizableOp.isWritable(bbArg, state));
-}
-
-/// Fold return values that are memref casts.
-static void foldMemRefCasts(FuncOp funcOp) {
-  if (funcOp.getBody().empty())
-    return;
-
-  func::ReturnOp returnOp = getAssumedUniqueReturnOp(funcOp);
-  SmallVector<Type> resultTypes;
-
-  for (OpOperand &operand : returnOp->getOpOperands()) {
-    if (auto castOp = operand.get().getDefiningOp<memref::CastOp>()) {
-      operand.set(castOp.source());
-      resultTypes.push_back(castOp.source().getType());
-    } else {
-      resultTypes.push_back(operand.get().getType());
-    }
-  }
-
-  auto newFuncType = FunctionType::get(
-      funcOp.getContext(), funcOp.getFunctionType().getInputs(), resultTypes);
-  funcOp.setType(newFuncType);
-}
-
-LogicalResult mlir::linalg::comprehensive_bufferize::runModuleBufferize(
-    ModuleOp moduleOp, OneShotBufferizationOptions options) {
-  IRRewriter rewriter(moduleOp.getContext());
-  OneShotAnalysisState analysisState(moduleOp, options);
-  BufferizationState bufferizationState(analysisState);
-  FuncAnalysisState &moduleState = getFuncAnalysisState(analysisState);
-  BufferizationAliasInfo &aliasInfo = analysisState.getAliasInfo();
-
-  // A list of functions in the order in which they are analyzed + bufferized.
-  SmallVector<FuncOp> orderedFuncOps;
-
-  // A mapping of FuncOps to their callers.
-  FuncCallerMap callerMap;
-
-  if (failed(getFuncOpsOrderedByCalls(moduleOp, orderedFuncOps, callerMap)))
-    return failure();
-
-  // Collect bbArg/return value information after the analysis.
-  options.addPostAnalysisStep(aliasingFuncOpBBArgsAnalysis);
-  options.addPostAnalysisStep(funcOpBbArgReadWriteAnalysis);
-
-  // Analyze ops.
-  for (FuncOp funcOp : orderedFuncOps) {
-    // No body => no analysis.
-    if (funcOp.getBody().empty())
-      continue;
-
-    // Now analyzing function.
-    moduleState.analyzedFuncOps[funcOp] = FuncOpAnalysisState::InProgress;
-
-    // Gather equivalence info for CallOps.
-    equivalenceAnalysis(funcOp, aliasInfo, moduleState);
-
-    // Analyze funcOp.
-    if (failed(analyzeOp(funcOp, analysisState)))
-      return failure();
-
-    // Mark op as fully analyzed.
-    moduleState.analyzedFuncOps[funcOp] = FuncOpAnalysisState::Analyzed;
-
-    // Add annotations to function arguments.
-    if (options.testAnalysisOnly)
-      annotateOpsWithBufferizationMarkers(funcOp, analysisState);
-  }
-
-  if (options.testAnalysisOnly)
-    return success();
-
-  // Bufferize functions.
-  for (FuncOp funcOp : orderedFuncOps) {
-    // Note: It would be good to apply cleanups here but we cannot as aliasInfo
-    // would be invalidated.
-    if (failed(bufferizeOp(funcOp, bufferizationState)))
-      return failure();
-    foldMemRefCasts(funcOp);
-  }
-
-  // Check result.
-  for (FuncOp funcOp : orderedFuncOps) {
-    if (!options.allowReturnAllocs &&
-        llvm::any_of(funcOp.getFunctionType().getResults(), [](Type t) {
-          return t.isa<MemRefType, UnrankedMemRefType>();
-        })) {
-      funcOp->emitError("memref return type is unsupported");
-      return failure();
-    }
-  }
-
-  // Finalize all buffers.
-  if (failed(finalizeBuffers(moduleOp, options)))
-    return failure();
-
-  // Post-pass cleanup of inplaceable and buffer_layout attributes.
-  moduleOp.walk([&](FuncOp op) {
-    for (BlockArgument bbArg : op.getArguments())
-      removeBufferizationFuncArguments(bbArg);
-  });
-
-  return success();
-}

mlir/lib/Dialect/Linalg/IR/LinalgDialect.cpp

if (!attr.getValue().isa<BoolAttr>()) {
              << "'" << BufferizableOpInterface::kInplaceableAttrName
              << "' is expected to be a boolean attribute";
     }
     if (!isa<FunctionOpInterface>(op))
       return op->emitError() << "expected " << attr.getName()
                              << " to be used on function-like operations";
     return success();
   }
-  if (attr.getName() == BufferizableOpInterface::kBufferLayoutAttrName) {
-    if (!attr.getValue().isa<AffineMapAttr>()) {
-      return op->emitError()
-             << "'" << BufferizableOpInterface::kBufferLayoutAttrName
-             << "' is expected to be a affine map attribute";
-    }
-    if (!isa<FunctionOpInterface>(op))
-      return op->emitError() << "expected " << attr.getName()
-                             << " to be used on function-like operations";
-    return success();
-  }
   if (attr.getName() == LinalgDialect::kMemoizedIndexingMapsAttrName)
     return success();
   return op->emitError() << "attribute '" << attr.getName()
                          << "' not supported by the linalg dialect";
 }
 
 #include "mlir/Dialect/Linalg/IR/LinalgOpsEnums.cpp.inc"
 
 #define GET_ATTRDEF_CLASSES
 #include "mlir/Dialect/Linalg/IR/LinalgOpsAttrDefs.cpp.inc"
 
 #include "mlir/Dialect/Linalg/IR/LinalgOpsDialect.cpp.inc"

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

add_mlir_dialect_library(MLIRLinalgTransforms
 
   LINK_LIBS PUBLIC
   MLIRAffine
   MLIRAffineUtils
   MLIRAnalysis
   MLIRArithmetic
   MLIRArithmeticTransforms
   MLIRBufferization
+  MLIRBufferizationTransforms
   MLIRComplex
   MLIRFunc
   MLIRFuncToLLVM
   MLIRFuncTransforms
   MLIRInferTypeOpInterface
   MLIRIR
   MLIRMemRef
   MLIRLinalg
   MLIRLinalgAnalysis
   MLIRLinalgUtils
-  MLIRModuleBufferization
   MLIRSCF
   MLIRSCFTransforms
   MLIRSCFUtils
   MLIRPass
   MLIRSparseTensor
   MLIRTensor
   MLIRTensorTilingInterfaceImpl
   MLIRTensorTransforms
   MLIRTransforms
   MLIRTransformUtils
   MLIRVector
   MLIRVectorTransforms
   MLIRVectorUtils
   MLIRX86VectorTransforms
   MLIRVectorToSCF
 )

mlir/lib/Dialect/Linalg/Transforms/ComprehensiveBufferizePass.cpp

 //===- ComprehensiveBufferize.cpp - Single pass bufferization -------------===//
 //
 // 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 "PassDetail.h"
 
 #include "mlir/Dialect/Arithmetic/Transforms/BufferizableOpInterfaceImpl.h"
 #include "mlir/Dialect/Bufferization/IR/BufferizableOpInterface.h"
 #include "mlir/Dialect/Bufferization/IR/Bufferization.h"
+#include "mlir/Dialect/Bufferization/Transforms/FuncBufferizableOpInterfaceImpl.h"
 #include "mlir/Dialect/Bufferization/Transforms/OneShotAnalysis.h"
+#include "mlir/Dialect/Bufferization/Transforms/OneShotModuleBufferize.h"
 #include "mlir/Dialect/Bufferization/Transforms/Passes.h"
 #include "mlir/Dialect/Func/IR/FuncOps.h"
-#include "mlir/Dialect/Linalg/ComprehensiveBufferize/ModuleBufferization.h"
 #include "mlir/Dialect/Linalg/Passes.h"
 #include "mlir/Dialect/Linalg/Transforms/BufferizableOpInterfaceImpl.h"
 #include "mlir/Dialect/SCF/BufferizableOpInterfaceImpl.h"
 #include "mlir/Dialect/Tensor/Transforms/BufferizableOpInterfaceImpl.h"
 #include "mlir/Dialect/Vector/Transforms/BufferizableOpInterfaceImpl.h"
 #include "mlir/Pass/Pass.h"
 #include "mlir/Pass/PassManager.h"
 #include "mlir/Transforms/GreedyPatternRewriteDriver.h"
 #include "mlir/Transforms/Passes.h"
 
 using namespace mlir;
 using namespace mlir::bufferization;
 using namespace mlir::linalg;
-using namespace mlir::linalg::comprehensive_bufferize;
 
 namespace {
 struct LinalgComprehensiveModuleBufferize
     : public LinalgComprehensiveModuleBufferizeBase<
           LinalgComprehensiveModuleBufferize> {
   LinalgComprehensiveModuleBufferize() = default;
 
   LinalgComprehensiveModuleBufferize(
registry
         .insert<bufferization::BufferizationDialect, linalg::LinalgDialect,
                 memref::MemRefDialect, tensor::TensorDialect,
                 vector::VectorDialect, scf::SCFDialect,
                 arith::ArithmeticDialect, func::FuncDialect, AffineDialect>();
     arith::registerBufferizableOpInterfaceExternalModels(registry);
     bufferization::registerAllocationOpInterfaceExternalModels(registry);
     linalg::registerBufferizableOpInterfaceExternalModels(registry);
     scf::registerBufferizableOpInterfaceExternalModels(registry);
-    std_ext::registerModuleBufferizationExternalModels(registry);
+    func_ext::registerBufferizableOpInterfaceExternalModels(registry);
     tensor::registerBufferizableOpInterfaceExternalModels(registry);
     vector::registerBufferizableOpInterfaceExternalModels(registry);
   }
 
 private:
   llvm::Optional<OneShotBufferizationOptions> options;
 };
 } // namespace
if (!options) {
     }
   } else {
     opt = *options;
   }
 
   ModuleOp moduleOp = getOperation();
   applyEnablingTransformations(moduleOp);
 
-  if (failed(runModuleBufferize(moduleOp, opt))) {
+  if (failed(runOneShotModuleBufferize(moduleOp, opt))) {
     signalPassFailure();
     return;
   }
 
   if (opt.testAnalysisOnly)
     return;
 
   OpPassManager cleanupPipeline("builtin.module");

mlir/test/Dialect/Bufferization/Transforms/one-shot-bufferize-partial.mlir

func @simple_tensor_test(%t1 : tensor<?xf32>, %f : f32) -> tensor<?xf32> {
   %0 = tensor.insert %f into %t1[%c0] : tensor<?xf32>
   // CHECK-TENSOR: return %[[casted_tensor]]
   return %0 : tensor<?xf32>
 }
 
 // -----
 
 // CHECK-SCF-LABEL: func @simple_scf_if(
-//  CHECK-SCF-SAME:     %[[t1:.*]]: tensor<?xf32> {linalg.inplaceable = true}, %[[c:.*]]: i1, %[[pos:.*]]: index
-func @simple_scf_if(%t1: tensor<?xf32> {linalg.inplaceable = true}, %c: i1, %pos: index, %f: f32)
+//  CHECK-SCF-SAME:     %[[t1:.*]]: tensor<?xf32> {bufferization.writable = true}, %[[c:.*]]: i1, %[[pos:.*]]: index
+func @simple_scf_if(%t1: tensor<?xf32> {bufferization.writable = true}, %c: i1, %pos: index, %f: f32)
     -> (tensor<?xf32>, index) {
   // CHECK-SCF: %[[r:.*]] = scf.if %[[c]] -> (memref<?xf32, #{{.*}}>) {
   %r1, %r2 = scf.if %c -> (tensor<?xf32>, index) {
     // CHECK-SCF: %[[t1_memref:.*]] = bufferization.to_memref %[[t1]]
     // CHECK-SCF: scf.yield %[[t1_memref]]
     scf.yield %t1, %pos : tensor<?xf32>, index
   // CHECK-SCF: } else {
   } else {

mlir/test/Dialect/Bufferization/Transforms/one-shot-module-bufferize-allow-return-allocs.mlir

-// RUN: mlir-opt %s -linalg-comprehensive-module-bufferize=allow-return-allocs -split-input-file | FileCheck %s
+// RUN: mlir-opt %s -one-shot-bufferize="bufferize-function-boundaries=1 allow-return-allocs" -split-input-file | FileCheck %s
 
 // Run fuzzer with different seeds.
-// RUN: mlir-opt %s -linalg-comprehensive-module-bufferize="allow-return-allocs test-analysis-only analysis-fuzzer-seed=23" -split-input-file -o /dev/null
-// RUN: mlir-opt %s -linalg-comprehensive-module-bufferize="allow-return-allocs test-analysis-only analysis-fuzzer-seed=59" -split-input-file -o /dev/null
-// RUN: mlir-opt %s -linalg-comprehensive-module-bufferize="allow-return-allocs test-analysis-only analysis-fuzzer-seed=91" -split-input-file -o /dev/null
+// RUN: mlir-opt %s -one-shot-bufferize="bufferize-function-boundaries=1 allow-return-allocs test-analysis-only analysis-fuzzer-seed=23" -split-input-file -o /dev/null
+// RUN: mlir-opt %s -one-shot-bufferize="bufferize-function-boundaries=1 allow-return-allocs test-analysis-only analysis-fuzzer-seed=59" -split-input-file -o /dev/null
+// RUN: mlir-opt %s -one-shot-bufferize="bufferize-function-boundaries=1 allow-return-allocs test-analysis-only analysis-fuzzer-seed=91" -split-input-file -o /dev/null
 
 // Test bufferization using memref types that have no layout map.
-// RUN: mlir-opt %s -linalg-comprehensive-module-bufferize="allow-return-allocs fully-dynamic-layout-maps=0" -split-input-file -o /dev/null
+// RUN: mlir-opt %s -one-shot-bufferize="bufferize-function-boundaries=1 allow-return-allocs fully-dynamic-layout-maps=0" -split-input-file -o /dev/null
 
 // Make sure that the returned buffer is not deallocated.
 // TODO: Such buffers currently leak. We need buffer hoisting / ref counting for
 // this in the future.
 
 // CHECK-LABEL: func @create_tensor() -> memref<10xf32> {
 //       CHECK:   %[[alloc:.*]] = memref.alloc
 //       CHECK:   return %[[alloc]]

mlir/test/Dialect/Bufferization/Transforms/one-shot-module-bufferize-analysis.mlir

-// RUN: mlir-opt %s -linalg-comprehensive-module-bufferize="test-analysis-only allow-return-allocs" -split-input-file | FileCheck %s
+// RUN: mlir-opt %s -one-shot-bufferize="bufferize-function-boundaries test-analysis-only allow-return-allocs" -split-input-file | FileCheck %s
 
 // Run fuzzer with different seeds.
-// RUN: mlir-opt %s -linalg-comprehensive-module-bufferize="test-analysis-only allow-return-allocs analysis-fuzzer-seed=23" -split-input-file -o /dev/null
-// RUN: mlir-opt %s -linalg-comprehensive-module-bufferize="test-analysis-only allow-return-allocs analysis-fuzzer-seed=59" -split-input-file -o /dev/null
-// RUN: mlir-opt %s -linalg-comprehensive-module-bufferize="test-analysis-only allow-return-allocs analysis-fuzzer-seed=91" -split-input-file -o /dev/null
+// RUN: mlir-opt %s -one-shot-bufferize="bufferize-function-boundaries test-analysis-only allow-return-allocs analysis-fuzzer-seed=23" -split-input-file -o /dev/null
+// RUN: mlir-opt %s -one-shot-bufferize="bufferize-function-boundaries test-analysis-only allow-return-allocs analysis-fuzzer-seed=59" -split-input-file -o /dev/null
+// RUN: mlir-opt %s -one-shot-bufferize="bufferize-function-boundaries test-analysis-only allow-return-allocs analysis-fuzzer-seed=91" -split-input-file -o /dev/null
+
+// TODO: Extract op-specific test cases and move them to their respective
+// dialects.
 
 //===----------------------------------------------------------------------===//
 // Simple cases
 //===----------------------------------------------------------------------===//
 
 // -----
 
 // CHECK-LABEL: func @extract_slice_fun(
-func @extract_slice_fun(%A : tensor<?xf32> {linalg.inplaceable = false},
+func @extract_slice_fun(%A : tensor<?xf32> {bufferization.writable = false},
 //  CHECK-SAME:          bufferization.access = "read"
-                        %B : tensor<?xf32> {linalg.inplaceable = true})
+                        %B : tensor<?xf32> {bufferization.writable = true})
 //  CHECK-SAME:         bufferization.access = "read"
   -> (tensor<4xf32>, tensor<8xf32>)
 {
   // tensor.extract_slice is not used in a write, it is not compelled to
   // bufferize out of place. Let callers decide whether they want to create
   // aliasing subviews at all call sites or whether they allocate.
   // This is true irrespective of whether the function argument is inplaceable.
   //     CHECK: tensor.extract_slice
   // CHECK-SAME: {__inplace_operands_attr__ = ["true"]}
   %r0 = tensor.extract_slice %A[0][4][1] : tensor<?xf32> to tensor<4xf32>
 
   //     CHECK: tensor.extract_slice
   // CHECK-SAME: {__inplace_operands_attr__ = ["true"]}
   %r1 = tensor.extract_slice %B[0][8][1] : tensor<?xf32> to tensor<8xf32>
 
   return %r0, %r1: tensor<4xf32>, tensor<8xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @insert_slice_fun(
-func @insert_slice_fun(%A : tensor<?xf32> {linalg.inplaceable = false},
+func @insert_slice_fun(%A : tensor<?xf32> {bufferization.writable = false},
 //  CHECK-SAME:        bufferization.access = "read"
-                       %B : tensor<?xf32> {linalg.inplaceable = true},
+                       %B : tensor<?xf32> {bufferization.writable = true},
 //  CHECK-SAME:        bufferization.access = "read-write"
-                       %C : tensor<4xf32> {linalg.inplaceable = false})
+                       %C : tensor<4xf32> {bufferization.writable = false})
 //  CHECK-SAME:        bufferization.access = "read"
   -> (tensor<?xf32>, tensor<?xf32>)
 {
   // must bufferize out of place.
   //      CHECK: tensor.insert_slice
   // CHECK-SAME: {__inplace_operands_attr__ = ["true", "false"]}
   %r0 = tensor.insert_slice %C into %A[0][4][1] : tensor<4xf32> into tensor<?xf32>
 
   // bufferizes inplace.
   //      CHECK: tensor.insert_slice
   // CHECK-SAME: {__inplace_operands_attr__ = ["true", "true"]}
   %r1 = tensor.insert_slice %C into %B[0][4][1] : tensor<4xf32> into tensor<?xf32>
 
   //      CHECK: return
   // CHECK-SAME: __equivalent_func_args__ = [-1, 1]
   return %r0, %r1: tensor<?xf32>, tensor<?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @conflict_on_B(
-func @conflict_on_B(%A : tensor<4x4xf32> {linalg.inplaceable = true},
+func @conflict_on_B(%A : tensor<4x4xf32> {bufferization.writable = true},
 //  CHECK-SAME:     bufferization.access = "read"
-                    %B : tensor<4x4xf32> {linalg.inplaceable = true})
+                    %B : tensor<4x4xf32> {bufferization.writable = true})
 //  CHECK-SAME:     bufferization.access = "read-write"
   -> (tensor<4x4xf32>, tensor<4x4xf32>, tensor<4x4xf32>)
 {
   // matmul output operand interferes with input operand.
   //     CHECK: linalg.matmul
   // CHECK-SAME: {__inplace_operands_attr__ = ["true", "true", "false"]}
   %C = linalg.matmul  ins(%A, %B: tensor<4x4xf32>, tensor<4x4xf32>)
                      outs(%B: tensor<4x4xf32>)

 //===----------------------------------------------------------------------===//
 // Length-1 producer-consumer cases.
 //===----------------------------------------------------------------------===//
 
 // -----
 
 // CHECK-LABEL: func @extract_slice_extract_slice(
 func @extract_slice_extract_slice(
-    %A : tensor<?xf32> {linalg.inplaceable = true},
+    %A : tensor<?xf32> {bufferization.writable = true},
 //  CHECK-SAME:         bufferization.access = "read"
-    %B : tensor<?xf32> {linalg.inplaceable = false})
+    %B : tensor<?xf32> {bufferization.writable = false})
 //  CHECK-SAME:         bufferization.access = "read"
   -> (tensor<2xf32>, tensor<2xf32>)
 {
   // tensor.extract_slice is not used in a write, it is not compelled to
   // bufferize out of place. Let callers decide whether they want to create
   // aliasing subviews at all call sites or whether they allocate.
   // This is true irrespective of whether the function argument is inplaceable.
   // CHECK: {__inplace_operands_attr__ = ["true"]}
//  CHECK-SAME:         bufferization.access = "read"
 
   return %r1, %r3: tensor<2xf32>, tensor<2xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @insert_slice_insert_slice(
 func @insert_slice_insert_slice(
-    %A : tensor<?xf32> {linalg.inplaceable = true},
+    %A : tensor<?xf32> {bufferization.writable = true},
 //  CHECK-SAME:         bufferization.access = "read-write"
-    %A2 : tensor<4xf32> {linalg.inplaceable = true},
+    %A2 : tensor<4xf32> {bufferization.writable = true},
 //  CHECK-SAME:          bufferization.access = "read-write"
-    %A3 : tensor<2xf32> {linalg.inplaceable = true},
+    %A3 : tensor<2xf32> {bufferization.writable = true},
 //  CHECK-SAME:          bufferization.access = "read"
-    %B : tensor<?xf32> {linalg.inplaceable = false},
+    %B : tensor<?xf32> {bufferization.writable = false},
 //  CHECK-SAME:         bufferization.access = "read"
-    %B2 : tensor<4xf32> {linalg.inplaceable = false},
+    %B2 : tensor<4xf32> {bufferization.writable = false},
 //  CHECK-SAME:          bufferization.access = "read"
-    %B3 : tensor<2xf32> {linalg.inplaceable = false})
+    %B3 : tensor<2xf32> {bufferization.writable = false})
 //  CHECK-SAME:          bufferization.access = "read"
   -> (tensor<?xf32>, tensor<?xf32>)
 {
   // CHECK: {__inplace_operands_attr__ = ["true", "true"]}
   %r0 = tensor.insert_slice %A3 into %A2[0][2][1] : tensor<2xf32> into tensor<4xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["true", "true"]}
   %r1 = tensor.insert_slice %r0 into %A[0][4][1] : tensor<4xf32> into tensor<?xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["true", "false"]}
   %r2 = tensor.insert_slice %B3 into %B2[0][2][1] : tensor<2xf32> into tensor<4xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["true", "false"]}
   %r3 = tensor.insert_slice %r2 into %B[0][4][1] : tensor<4xf32> into tensor<?xf32>
 
   //      CHECK: return
   // CHECK-SAME: __equivalent_func_args__ = [0, -1]
   return %r1, %r3: tensor<?xf32>, tensor<?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @extract_slice_nonmatching_insert_slice
 func @extract_slice_nonmatching_insert_slice(
-    %A : tensor<?xf32> {linalg.inplaceable = true},
-    %B : tensor<?xf32> {linalg.inplaceable = false},
+    %A : tensor<?xf32> {bufferization.writable = true},
+    %B : tensor<?xf32> {bufferization.writable = false},
     %idx: index)
   -> (tensor<?xf32>, tensor<?xf32>)
 {
   // %r1 bufferizes inplace because %A is inplaceable.
   // %r0 is an overlapping tensor.extract_slice that does not match, it must be
   // out of place.
   //      CHECK: tensor.extract_slice
   // CHECK-SAME: {__inplace_operands_attr__ = ["false"]}
func @extract_slice_nonmatching_insert_slice(
   // CHECK-SAME: __equivalent_func_args__ = [0, -1]
   return %r1, %r3: tensor<?xf32>, tensor<?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @extract_slice_matching_insert_slice
 func @extract_slice_matching_insert_slice(
-    %A : tensor<?xf32> {linalg.inplaceable = true},
-    %B : tensor<?xf32> {linalg.inplaceable = false})
+    %A : tensor<?xf32> {bufferization.writable = true},
+    %B : tensor<?xf32> {bufferization.writable = false})
   -> (tensor<?xf32>, tensor<?xf32>)
 {
   // %r1 bufferizes inplace because %A is inplaceable.
   // %r0 is a tensor.extract_slice that matches, it can also be bufferized
   // inplace.
   //      CHECK: tensor.extract_slice
   // CHECK-SAME: {__inplace_operands_attr__ = ["true"]}
   %r0 = tensor.extract_slice %A[0][4][1] : tensor<?xf32> to tensor<4xf32>
func @extract_slice_matching_insert_slice(
   // CHECK-SAME: __equivalent_func_args__ = [0, -1]
   return %r1, %r3: tensor<?xf32>, tensor<?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: @read_of_matching_insert_slice_source
 func @read_of_matching_insert_slice_source(
-    %A : tensor<?xf32> {linalg.inplaceable = true},
+    %A : tensor<?xf32> {bufferization.writable = true},
     %idx : index,
     %idx2 : index)
   -> (tensor<?xf32>, vector<5xf32>)
 {
   %cst = arith.constant 0.0 : f32
   %cst2 = arith.constant 1.0 : f32
 
   //      CHECK: tensor.extract_slice
func @read_of_matching_insert_slice_source(
   // CHECK-SAME: __equivalent_func_args__ = [0, -1]
   return %2, %3 : tensor<?xf32>, vector<5xf32>
 }
 
 // -----
 
 // CHECK-LABEL: @read_of_matching_insert_slice_source_interleaved
 func @read_of_matching_insert_slice_source_interleaved(
-    %A : tensor<?xf32> {linalg.inplaceable = true},
+    %A : tensor<?xf32> {bufferization.writable = true},
     %idx : index,
     %idx2 : index,
     %idx3 : index)
   -> (tensor<?xf32>, vector<5xf32>)
 {
   %cst = arith.constant 0.0 : f32
   %cst2 = arith.constant 1.0 : f32
 
func @read_of_matching_insert_slice_source_interleaved(
   // CHECK-SAME: __equivalent_func_args__ = [0, -1]
   return %6, %3 : tensor<?xf32>, vector<5xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @extract_slice_linalg_readonly_use
 func @extract_slice_linalg_readonly_use(
-    %A : tensor<?x?xf32> {linalg.inplaceable = false},
-    %B : tensor<4x4xf32> {linalg.inplaceable = false},
-    %C : tensor<4x4xf32> {linalg.inplaceable = true})
+    %A : tensor<?x?xf32> {bufferization.writable = false},
+    %B : tensor<4x4xf32> {bufferization.writable = false},
+    %C : tensor<4x4xf32> {bufferization.writable = true})
   ->  (tensor<4x4xf32>, tensor<4x4xf32>)
 {
   // tensor.extract_slice is only used as a read, no interference irrespective
   // of user's inplace status.
   //     CHECK: tensor.extract_slice
   // CHECK-SAME: {__inplace_operands_attr__ = ["true"]}
   %sA = tensor.extract_slice %A[0, 0][4, 4][1, 1] : tensor<?x?xf32> to tensor<4x4xf32>
 
func @extract_slice_linalg_readonly_use(
   // CHECK-SAME: __equivalent_func_args__ = [-1, 2]
   return %D, %E: tensor<4x4xf32>, tensor<4x4xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @extract_slice_to_linalg_write_use
 func @extract_slice_to_linalg_write_use(
-    %A : tensor<4x4xf32> {linalg.inplaceable = false},
-    %B : tensor<?x?xf32> {linalg.inplaceable = false},
-    %C : tensor<?x?xf32> {linalg.inplaceable = true})
+    %A : tensor<4x4xf32> {bufferization.writable = false},
+    %B : tensor<?x?xf32> {bufferization.writable = false},
+    %C : tensor<?x?xf32> {bufferization.writable = true})
   ->  (tensor<4x4xf32>, tensor<4x4xf32>)
 {
   // Step 4. %sB forward propagates to a write in %D but it is not inplace.
   // So this is only ever read and can bufferize inplace.
   //     CHECK: tensor.extract_slice
   // CHECK-SAME: {__inplace_operands_attr__ = ["true"]}
   %sB = tensor.extract_slice %B[0, 0][4, 4][1, 1] : tensor<?x?xf32> to tensor<4x4xf32>
 

 // -----
 
 // CHECK-LABEL: func @insert_slice_double_extract_slice
 func @insert_slice_double_extract_slice(
     %s1: index,
     %s2: index,
     %s3: index,
     %s4: index,
-    %A: tensor<8x6xf32> {linalg.inplaceable = false},
-    %B: tensor<6x6xf32> {linalg.inplaceable = false},
-    %C: tensor<30x20xf32> {linalg.inplaceable = true})
+    %A: tensor<8x6xf32> {bufferization.writable = false},
+    %B: tensor<6x6xf32> {bufferization.writable = false},
+    %C: tensor<30x20xf32> {bufferization.writable = true})
   -> tensor<30x20xf32>
 {
   //      CHECK: tensor.extract_slice
   // CHECK-SAME: {__inplace_operands_attr__ = ["true", "none", "none", "none", "none"]}
   %15 = tensor.extract_slice %C[%s3, %s4] [%s1, %s2] [1, 1] : tensor<30x20xf32> to tensor<?x?xf32>
 
   //      CHECK: linalg.matmul
   // CHECK-SAME: {__inplace_operands_attr__ = ["true", "true", "true"]}

 //===----------------------------------------------------------------------===//
 // Transitive cases
 //===----------------------------------------------------------------------===//
 
 // -----
 
 // CHECK-LABEL: func @extract_slice_to_linalg_write_use
 func @extract_slice_to_linalg_write_use(
-    %A : tensor<4x4xf32> {linalg.inplaceable = false},
-    %B : tensor<?x?xf32> {linalg.inplaceable = false},
-    %C : tensor<?x?xf32> {linalg.inplaceable = true})
+    %A : tensor<4x4xf32> {bufferization.writable = false},
+    %B : tensor<?x?xf32> {bufferization.writable = false},
+    %C : tensor<?x?xf32> {bufferization.writable = true})
   ->  (tensor<4x4xf32>, tensor<4x4xf32>)
 {
   // Step 4. %sB forward propagates to an inplace write in %D.
   // %sB backward propagates to %B which is not inplaceable.
   // As a consequence this is bufferized out of place.
   //     CHECK: tensor.extract_slice
   // CHECK-SAME: {__inplace_operands_attr__ = ["false"]}
   %sB = tensor.extract_slice %B[0, 0][4, 4][1, 1] : tensor<?x?xf32> to tensor<4x4xf32>
func @extract_slice_to_linalg_write_use(
 
   return %D, %E: tensor<4x4xf32>, tensor<4x4xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @nested_extract_slice_and_insert
 func @nested_extract_slice_and_insert(
-    %A : tensor<?x?xf32> {linalg.inplaceable = false},
-    %B : tensor<?x?xf32> {linalg.inplaceable = true},
-    %C : tensor<?x?xf32> {linalg.inplaceable = true},
+    %A : tensor<?x?xf32> {bufferization.writable = false},
+    %B : tensor<?x?xf32> {bufferization.writable = true},
+    %C : tensor<?x?xf32> {bufferization.writable = true},
     %idx : index,
     %sz1 : index,
     %sz2 : index)
   ->  (tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>)
 {
   %f0 = arith.constant 0.0 : f32
 
   // 2-level matching tensor.extract_slice / tensor.insert_slice into non

 //===----------------------------------------------------------------------===//
 // Simple loop cases
 //===----------------------------------------------------------------------===//
 
 // -----
 
 // CHECK-LABEL: func @scf_for_yield_only
 func @scf_for_yield_only(
-    %A : tensor<?xf32> {linalg.inplaceable = false},
-    %B : tensor<?xf32> {linalg.inplaceable = true},
+    %A : tensor<?xf32> {bufferization.writable = false},
+    %B : tensor<?xf32> {bufferization.writable = true},
     %lb : index,
     %ub : index,
     %step : index)
   -> (tensor<?xf32>, tensor<?xf32>)
 {
   //      CHECK: scf.for
   // CHECK-NEXT: scf.yield
   // CHECK-SAME: {__inplace_operands_attr__ = ["true"]}
func @scf_for_yield_only(
   // CHECK-SAME: __equivalent_func_args__ = [-1, 1]
   return %r0, %r1: tensor<?xf32>, tensor<?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @scf_for_with_tensor.insert_slice
 func @scf_for_with_tensor.insert_slice(
-    %A : tensor<?xf32> {linalg.inplaceable = false},
-    %B : tensor<?xf32> {linalg.inplaceable = true},
-    %C : tensor<4xf32> {linalg.inplaceable = false},
+    %A : tensor<?xf32> {bufferization.writable = false},
+    %B : tensor<?xf32> {bufferization.writable = true},
+    %C : tensor<4xf32> {bufferization.writable = false},
     %lb : index,
     %ub : index,
     %step : index)
   -> (tensor<?xf32>, tensor<?xf32>)
 {
   //      CHECK: scf.for
   // scf.for bbArgs are always inplaceable seen from ops inside the body:
   //   1. Either the matching tensor is not inplaceable and an alloc occurs

 }
 
 // -----
 
 func private @some_use(tensor<?xf32>) -> ()
 
 // CHECK-LABEL: func @scf_for_deps
 func @scf_for_deps(
-    %A : tensor<?xf32> {linalg.inplaceable = true},
-    %B : tensor<?xf32> {linalg.inplaceable = true},
+    %A : tensor<?xf32> {bufferization.writable = true},
+    %B : tensor<?xf32> {bufferization.writable = true},
     %lb : index,
     %ub : index,
     %step : index)
   -> (tensor<?xf32>)
 {
   // %r0 must be out of place because one use of %t in the subsequent production
   // of %r1 is read.
   //      CHECK: scf.for

 
 //===----------------------------------------------------------------------===//
 // Cross function boundary cases.
 //===----------------------------------------------------------------------===//
 
 func private @foo(tensor<64xf32>)
 
 // CHECK-LABEL: dependence_through_call
-func @dependence_through_call(%I : tensor<64xf32> {linalg.inplaceable = true}) {
+func @dependence_through_call(%I : tensor<64xf32> {bufferization.writable = true}) {
   %f1 = arith.constant 1.000000e+00 : f32
   %f2 = arith.constant 2.000000e+00 : f32
 
   // 2. %B already bufferizes inplace, %A would alias and have a different
   // value. The calls to `foo` are determined to read conservatively, so %A
   // cannot bufferize inplace.
   //     CHECK: fill
   // CHECK-SAME: {__inplace_operands_attr__ = ["none", "false"]}

 func private @foo(tensor<64xf32>)
 
 func private @bar(%A : tensor<64xf32>) {
   call @foo(%A) : (tensor<64xf32>) -> ()
   return
 }
 
 func @read_dependence_through_scf_and_call(
-    %I : tensor<64xf32> {linalg.inplaceable = true},
-    %I2 : tensor<64xf32> {linalg.inplaceable = true}) {
+    %I : tensor<64xf32> {bufferization.writable = true},
+    %I2 : tensor<64xf32> {bufferization.writable = true}) {
   %c0 = arith.constant 0 : index
   %c1 = arith.constant 1 : index
   %c10 = arith.constant 10 : index
   %f1 = arith.constant 1.000000e+00 : f32
   %f2 = arith.constant 2.000000e+00 : f32
 
   // 5. %B bufferizes inplace, %A would alias and have a different value.
   // The calls to `foo` are determined to read conservatively, so %A cannot
func @write_into_constant_via_alias(%v : vector<5xi32>,
   //      CHECK: vector.transfer_write
   // CHECK-SAME: {__inplace_operands_attr__ = ["none", "true", "none"]}
   %r = vector.transfer_write %v, %b[%s3] : vector<5xi32>, tensor<?xi32>
   return %r : tensor<?xi32>
 }
 
 // -----
 
-func.func @matmul_on_tensors(
-    %arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+func @matmul_on_tensors(
+    %arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, bufferization.writable = false},
+    %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, bufferization.writable = false},
+    %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, bufferization.writable = true})
     -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst_0 = arith.constant 0.000000e+00 : f32
   %cst_1 = arith.constant 1.000000e+00 : f32
 
   %7 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
func @matmul_on_tensors(
 
   //      CHECK: return
   // CHECK-SAME: __equivalent_func_args__ = [2]
   return %r : tensor<256x256xf32>
 }
 
 // -----
 
-func.func @matmul_on_tensors(
-    %arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+func @matmul_on_tensors(
+    %arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, bufferization.writable = false},
+    %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, bufferization.writable = false},
+    %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, bufferization.writable = true})
     -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst_0 = arith.constant 0.000000e+00 : f32
   %cst_1 = arith.constant 1.000000e+00 : f32
 
   %7 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 

 // Chain of tensor.insert_slice is better traversed in reverse order without
 // prioritizing  the tensor.insert_slice ops.
 //===----------------------------------------------------------------------===//
 
 // CHECK-LABEL: func @insert_slice_chain(
 func @insert_slice_chain(
     %v1: vector<32x90xf32>,
     %v2: vector<30x90xf32>,
-    %arg0: tensor<62x126xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg0: tensor<62x126xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, bufferization.writable = false},
 // CHECK-SAME: bufferization.access = "none"
-    %arg1: tensor<126x90xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg1: tensor<126x90xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, bufferization.writable = false},
 // CHECK-SAME: bufferization.access = "none"
-    %arg2: tensor<62x90xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+    %arg2: tensor<62x90xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, bufferization.writable = true})
 // CHECK-SAME: bufferization.access = "write"
   -> tensor<62x90xf32> attributes {passthrough = [["target-cpu", "skylake-avx512"], ["prefer-vector-width", "512"]]}
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
 
   //      CHECK: linalg.fill
   // CHECK-SAME: {__inplace_operands_attr__ = ["none", "true"]

 // -----
 
 //===----------------------------------------------------------------------===//
 // Insert point issue cases.
 //===----------------------------------------------------------------------===//
 
 // Only test IR validity wrt dominance.
 // CHECK-LABEL: func @ip
-func @ip(%t: tensor<10x20xf32> {linalg.inplaceable = true},
+func @ip(%t: tensor<10x20xf32> {bufferization.writable = true},
          %x: index, %y: index, %v: vector<5x6xf32>)
   -> tensor<10x20xf32>
 {
   %c0 = arith.constant 0 : index
   %c256 = arith.constant 256 : index
   %c257 = arith.constant 257 : index
   %r = scf.for %arg0 = %c0 to %c257 step %c256 iter_args(%arg1 = %t) -> (tensor<10x20xf32>) {
     %t1 = tensor.extract_slice %arg1[%x, 0] [5, %y] [1, 1] : tensor<10x20xf32> to tensor<5x?xf32>

 ]
 #trait = {
   indexing_maps = #accesses,
   iterator_types = ["parallel"]
 }
 
 // CHECK-LABEL: func @linalg_op_same_out_tensors(
 func @linalg_op_same_out_tensors(
-    %t1: tensor<?xf32> {linalg.inplaceable = true},
+    %t1: tensor<?xf32> {bufferization.writable = true},
 // CHECK-SAME:          bufferization.access = "read"
-    %t2: tensor<?xf32> {linalg.inplaceable = true})
+    %t2: tensor<?xf32> {bufferization.writable = true})
 // CHECK-SAME:          bufferization.access = "write"
   -> (tensor<?xf32>, tensor<?xf32>){
 
   //      CHECK: linalg.generic
   // CHECK-SAME: {__inplace_operands_attr__ = ["true", "true", "false"]
   %o:2 = linalg.generic #trait ins(%t1 : tensor<?xf32>)
                                outs (%t2, %t2 : tensor<?xf32>, tensor<?xf32>) {
       ^bb(%0: f32, %1: f32, %2 : f32) :

 ]
 #trait = {
   indexing_maps = #accesses,
   iterator_types = ["parallel"]
 }
 
 // CHECK-LABEL: func @linalg_op_same_out_tensors_2(
 func @linalg_op_same_out_tensors_2(
-    %t1: tensor<?xf32> {linalg.inplaceable = true},
+    %t1: tensor<?xf32> {bufferization.writable = true},
 // CHECK-SAME:          bufferization.access = "read"
-    %t2: tensor<?xf32> {linalg.inplaceable = true})
+    %t2: tensor<?xf32> {bufferization.writable = true})
 // CHECK-SAME:          bufferization.access = "write"
         -> (tensor<?xf32>, tensor<?xf32>, tensor<?xf32>){
 
   //      CHECK: linalg.generic
   // CHECK-SAME: {__inplace_operands_attr__ = ["true", "true", "false", "false"]
   %o:3 = linalg.generic #trait
           ins(%t1 : tensor<?xf32>)
           outs (%t2, %t2, %t2 : tensor<?xf32>, tensor<?xf32>, tensor<?xf32>) {
       ^bb(%0: f32, %1: f32, %2 : f32, %3 : f32) :
         linalg.yield %0, %0, %0 : f32, f32, f32
     } -> (tensor<?xf32>, tensor<?xf32>, tensor<?xf32>)
 
   //      CHECK: return
   // CHECK-SAME: __equivalent_func_args__ = [1, -1, -1]
   return %o#0, %o#1, %o#2 : tensor<?xf32>, tensor<?xf32>, tensor<?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @double_insert_slice_into_alias
 func @double_insert_slice_into_alias(
     %v1: vector<32x90xf32>,
     %v2: vector<30x90xf32>,
-    %arg2: tensor<62x90xf32> {linalg.inplaceable = true},
+    %arg2: tensor<62x90xf32> {bufferization.writable = true},
     %s1: index, %s2: index, %s3: index, %s4: index)
   -> (tensor<62x90xf32>, tensor<?x?xf32>)
 {
   %c0 = arith.constant 0 : index
 
   // Cannot bufferize inplace this extract_slice because both operand and result
   // are modified and returned separately.
   //      CHECK: tensor.extract_slice
func @double_insert_slice_into_alias(
   // CHECK-SAME: __equivalent_func_args__ = [2, -1]
   return %8, %15 : tensor<62x90xf32>, tensor<?x?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @interleaved_extract_insert_slice_chain_1
 func @interleaved_extract_insert_slice_chain_1(
-    %arg2: tensor<62x90xf32> {linalg.inplaceable = true})
+    %arg2: tensor<62x90xf32> {bufferization.writable = true})
   -> (tensor<62x90xf32>)
 {
   //      CHECK: tensor.extract_slice
   // CHECK-SAME: {__inplace_operands_attr__ = ["true"]
   %2 = tensor.extract_slice %arg2[0, 0] [32, 90] [1, 1] : tensor<62x90xf32> to tensor<32x90xf32>
 
   // TODO: This should bufferize inplace once we have a proper range analysis.
   //      CHECK: tensor.extract_slice
func @interleaved_extract_insert_slice_chain_1(
   // CHECK-SAME: __equivalent_func_args__ = [0]
   return %15 : tensor<62x90xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @interleaved_extract_insert_slice_chain_2
 func @interleaved_extract_insert_slice_chain_2(
-    %arg2: tensor<62x90xf32> {linalg.inplaceable = true})
+    %arg2: tensor<62x90xf32> {bufferization.writable = true})
   -> (tensor<62x90xf32>)
 {
   //      CHECK: tensor.extract_slice
   // CHECK-SAME: {__inplace_operands_attr__ = ["true"]
   %2 = tensor.extract_slice %arg2[0, 0] [32, 90] [1, 1] : tensor<62x90xf32> to tensor<32x90xf32>
 
   // The slices are overlapping, so this can never bufferize inplace.
   //      CHECK: tensor.extract_slice
func @interleaved_extract_insert_slice_chain_2(
   // CHECK-SAME: __equivalent_func_args__ = [0]
   return %15 : tensor<62x90xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @extract_once_insert_twice
 func @extract_once_insert_twice(
-    %arg2: tensor<62x90xf32> {linalg.inplaceable = true})
+    %arg2: tensor<62x90xf32> {bufferization.writable = true})
   -> (tensor<62x90xf32>)
 {
   //      CHECK: tensor.extract_slice
   // CHECK-SAME: {__inplace_operands_attr__ = ["false"]
   %2 = tensor.extract_slice %arg2[0, 0] [32, 90] [1, 1] : tensor<62x90xf32> to tensor<32x90xf32>
 
   //      CHECK: tensor.insert_slice
   // CHECK-SAME: {__inplace_operands_attr__ = ["true", "true"]
#accesses = [
   affine_map<(i) -> (i)>
 ]
 #trait = {
   indexing_maps = #accesses,
   iterator_types = ["parallel"]
 }
 
 // CHECK-LABEL: func @reading_scf_for
-func @reading_scf_for(%t1: tensor<?xf32> {linalg.inplaceable = true},
+func @reading_scf_for(%t1: tensor<?xf32> {bufferization.writable = true},
                       %s: index, %v: vector<5xf32>) -> (tensor<?xf32>, vector<5xf32>) {
 
   %c0 = arith.constant 0 : index
   %c1 = arith.constant 1 : index
   %cst = arith.constant 0.0 : f32
 
   // Write to %t1.
   // CHECK:      vector.transfer_write
#accesses = [
   affine_map<(i) -> (i)>
 ]
 #trait = {
   indexing_maps = #accesses,
   iterator_types = ["parallel"]
 }
 
 // CHECK-LABEL: func @non_reading_scf_for
-func @non_reading_scf_for(%t1: tensor<?xf32> {linalg.inplaceable = true},
+func @non_reading_scf_for(%t1: tensor<?xf32> {bufferization.writable = true},
                           %s: index, %v: vector<5xf32>) -> (tensor<?xf32>, vector<5xf32>) {
 
   %c0 = arith.constant 0 : index
   %c1 = arith.constant 1 : index
   %cst = arith.constant 0.0 : f32
 
   // Write to %t1.
   // CHECK:      vector.transfer_write

 // -----
 
 //===----------------------------------------------------------------------===//
 // scf.if cases
 //===----------------------------------------------------------------------===//
 
 // This example passes analysis, but it fails when bufferizing.
 // CHECK-LABEL: func @scf_if_inplace1
-func @scf_if_inplace1(%t1: tensor<?xf32> {linalg.inplaceable = true},
-                      %t2: tensor<?xf32> {linalg.inplaceable = true},
+func @scf_if_inplace1(%t1: tensor<?xf32> {bufferization.writable = true},
+                      %t2: tensor<?xf32> {bufferization.writable = true},
                       %cond: i1) -> tensor<?xf32> {
   %r = scf.if %cond -> (tensor<?xf32>) {
     // CHECK:      scf.yield
     // CHECK-SAME: {__inplace_operands_attr__ = ["true"]}
     scf.yield %t1 : tensor<?xf32>
   } else {
     // CHECK:      scf.yield
     // CHECK-SAME: {__inplace_operands_attr__ = ["true"]}
     scf.yield %t2 : tensor<?xf32>
   }
   return %r : tensor<?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @scf_if_inplace2
-func @scf_if_inplace2(%t1: tensor<?xf32> {linalg.inplaceable = true},
+func @scf_if_inplace2(%t1: tensor<?xf32> {bufferization.writable = true},
                       %v: vector<5xf32>, %idx: index,
                       %cond: i1) -> tensor<?xf32> {
   %r = scf.if %cond -> (tensor<?xf32>) {
     // CHECK:      scf.yield
     // CHECK-SAME: {__inplace_operands_attr__ = ["true"]}
     scf.yield %t1 : tensor<?xf32>
   } else {
     //      CHECK: vector.transfer_write
     // CHECK-SAME: {__inplace_operands_attr__ = ["none", "true", "none"]
     %t2 = vector.transfer_write %v, %t1[%idx] : vector<5xf32>, tensor<?xf32>
     scf.yield %t2 : tensor<?xf32>
   }
   //      CHECK: return
   // CHECK-SAME: __equivalent_func_args__ = [0]
   return %r : tensor<?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @scf_if_inplace3
-func @scf_if_inplace3(%t1: tensor<?xf32> {linalg.inplaceable = true},
+func @scf_if_inplace3(%t1: tensor<?xf32> {bufferization.writable = true},
                       %v1: vector<5xf32>, %v2: vector<5xf32>, %idx: index,
                       %cond: i1) -> tensor<?xf32> {
   //      CHECK: tensor.extract_slice
   // CHECK-SAME: {__inplace_operands_attr__ = ["true", "none", "none"]
   %e = tensor.extract_slice %t1[%idx][%idx][1] : tensor<?xf32> to tensor<?xf32>
   %r = scf.if %cond -> (tensor<?xf32>) {
     //      CHECK: vector.transfer_write
     // CHECK-SAME: {__inplace_operands_attr__ = ["none", "true", "none"]
%r = scf.if %cond -> (tensor<?xf32>) {
     scf.yield %t3 : tensor<?xf32>
   }
   return %r : tensor<?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @scf_if_in_place4
-func @scf_if_in_place4(%t1: tensor<?xf32> {linalg.inplaceable = true},
+func @scf_if_in_place4(%t1: tensor<?xf32> {bufferization.writable = true},
                        %v: vector<5xf32>, %idx: index,
                        %cond: i1, %cond2: i1) -> (tensor<?xf32>, vector<10xf32>) {
   %cst = arith.constant 0.0 : f32
   %r = scf.if %cond -> (tensor<?xf32>) {
     //      CHECK: scf.yield
     // CHECK-SAME: {__inplace_operands_attr__ = ["true"]}
     scf.yield %t1 : tensor<?xf32>
   } else {
func @scf_if_in_place4(%t1: tensor<?xf32> {bufferization.writable = true},
   //      CHECK: return
   // CHECK-SAME: __equivalent_func_args__ = [0, -1]
   return %r_alias, %v2 : tensor<?xf32>, vector<10xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @scf_if_inplace5
-func @scf_if_inplace5(%t1: tensor<?xf32> {linalg.inplaceable = true},
+func @scf_if_inplace5(%t1: tensor<?xf32> {bufferization.writable = true},
                       %idx: index, %cond: i1) -> tensor<?xf32> {
   %r = scf.if %cond -> (tensor<?xf32>) {
     //      CHECK: tensor.extract_slice
     // CHECK-SAME: {__inplace_operands_attr__ = ["true", "none", "none"]
     %e = tensor.extract_slice %t1[%idx][%idx][1] : tensor<?xf32> to tensor<?xf32>
     //      CHECK: scf.yield
     // CHECK-SAME: {__inplace_operands_attr__ = ["true"]}
     scf.yield %e : tensor<?xf32>
func @scf_if_inplace5(%t1: tensor<?xf32> {bufferization.writable = true},
   //      CHECK: return
   // CHECK-SAME: __equivalent_func_args__ = [0]
   return %r2 : tensor<?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @scf_if_inplace6
-func @scf_if_inplace6(%t1: tensor<?xf32> {linalg.inplaceable = true},
+func @scf_if_inplace6(%t1: tensor<?xf32> {bufferization.writable = true},
                       %v1: vector<5xf32>, %v2: vector<5xf32>,
                       %v3: vector<5xf32>, %idx: index,
                       %cond: i1, %cond2: i1) -> tensor<?xf32> {
   // Test nested scf.if ops.
   %r = scf.if %cond -> (tensor<?xf32>) {
     %t2 = scf.if %cond2 -> (tensor<?xf32>) {
       //      CHECK: vector.transfer_write
       // CHECK-SAME: {__inplace_operands_attr__ = ["none", "true", "none"]
func @scf_if_inplace6(%t1: tensor<?xf32> {bufferization.writable = true},
   //      CHECK: return
   // CHECK-SAME: __equivalent_func_args__ = [0]
   return %r : tensor<?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @scf_if_inplace7
-func @scf_if_inplace7(%t1: tensor<?xf32> {linalg.inplaceable = true},
+func @scf_if_inplace7(%t1: tensor<?xf32> {bufferization.writable = true},
                       %v1: vector<5xf32>, %v2: vector<5xf32>, %idx: index,
                       %idx2: index, %cond: i1) -> (tensor<?xf32>, vector<5xf32>) {
   %cst = arith.constant 0.0 : f32
   %r, %v_r2 = scf.if %cond -> (tensor<?xf32>, vector<5xf32>) {
     //      CHECK: vector.transfer_write
     // CHECK-SAME: {__inplace_operands_attr__ = ["none", "true", "none"]
     %t2 = vector.transfer_write %v1, %t1[%idx] : vector<5xf32>, tensor<?xf32>
     //      CHECK: scf.yield
%r, %v_r2 = scf.if %cond -> (tensor<?xf32>, vector<5xf32>) {
     scf.yield %t3, %v_r : tensor<?xf32>, vector<5xf32>
   }
   return %r, %v_r2 : tensor<?xf32>, vector<5xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @scf_if_out_of_place1a
-func @scf_if_out_of_place1a(%t1: tensor<?xf32> {linalg.inplaceable = true},
+func @scf_if_out_of_place1a(%t1: tensor<?xf32> {bufferization.writable = true},
                             %idx: index, %idx2: index,
                             %cond: i1) -> tensor<?xf32> {
   %r = scf.if %cond -> (tensor<?xf32>) {
     //      CHECK: tensor.extract_slice
     // CHECK-SAME: {__inplace_operands_attr__ = ["true", "none", "none"]
     %e = tensor.extract_slice %t1[%idx][%idx][1] : tensor<?xf32> to tensor<?xf32>
     //      CHECK: scf.yield
     // CHECK-SAME: {__inplace_operands_attr__ = ["true"]}
func @scf_if_out_of_place1a(%t1: tensor<?xf32> {bufferization.writable = true},
   // CHECK-SAME: {__inplace_operands_attr__ = ["true", "false", "none", "none"]
   %r2 = tensor.insert_slice %r into %t1[%idx2][%idx2][1] : tensor<?xf32> into tensor<?xf32>
   return %r2 : tensor<?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @scf_if_out_of_place1b
-func @scf_if_out_of_place1b(%t1: tensor<?xf32> {linalg.inplaceable = true},
+func @scf_if_out_of_place1b(%t1: tensor<?xf32> {bufferization.writable = true},
                             %idx: index, %idx2: index, %idx3: index,
                             %cond: i1) -> tensor<?xf32> {
   %r = scf.if %cond -> (tensor<?xf32>) {
     //      CHECK: tensor.extract_slice
     // CHECK-SAME: {__inplace_operands_attr__ = ["false", "none", "none"]
     %e = tensor.extract_slice %t1[%idx][%idx][1] : tensor<?xf32> to tensor<?xf32>
     //      CHECK: scf.yield
     // CHECK-SAME: {__inplace_operands_attr__ = ["true"]}
func @scf_if_out_of_place1b(%t1: tensor<?xf32> {bufferization.writable = true},
   //      CHECK: return
   // CHECK-SAME: __equivalent_func_args__ = [0]
   return %r2 : tensor<?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @scf_if_out_of_place1c
-func @scf_if_out_of_place1c(%t1: tensor<?xf32> {linalg.inplaceable = true},
+func @scf_if_out_of_place1c(%t1: tensor<?xf32> {bufferization.writable = true},
                             %idx: index, %idx2: index, %cond: i1) -> tensor<?xf32> {
   %r = scf.if %cond -> (tensor<?xf32>) {
     //      CHECK: tensor.extract_slice
     // CHECK-SAME: {__inplace_operands_attr__ = ["false", "none", "none"]
     %e = tensor.extract_slice %t1[%idx][%idx][1] : tensor<?xf32> to tensor<?xf32>
     //      CHECK: scf.yield
     // CHECK-SAME: {__inplace_operands_attr__ = ["true"]}
     scf.yield %e : tensor<?xf32>
func @scf_if_out_of_place1c(%t1: tensor<?xf32> {bufferization.writable = true},
   //      CHECK: return
   // CHECK-SAME: __equivalent_func_args__ = [0]
   return %r2 : tensor<?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @scf_if_out_of_place2
-func @scf_if_out_of_place2(%t1: tensor<?xf32> {linalg.inplaceable = true},
+func @scf_if_out_of_place2(%t1: tensor<?xf32> {bufferization.writable = true},
                            %v: vector<5xf32>, %idx: index,
                            %cond: i1) -> (tensor<?xf32>, vector<10xf32>) {
   %cst = arith.constant 0.0 : f32
   %r = scf.if %cond -> (tensor<?xf32>) {
     scf.yield %t1 : tensor<?xf32>
   } else {
     //      CHECK: vector.transfer_write
     // CHECK-SAME: {__inplace_operands_attr__ = ["none", "false", "none"]
     %t2 = vector.transfer_write %v, %t1[%idx] : vector<5xf32>, tensor<?xf32>
     //      CHECK: scf.yield
     // CHECK-SAME: {__inplace_operands_attr__ = ["true"]}
     scf.yield %t2 : tensor<?xf32>
   }
 
   // Read the old value of %t1. Forces the transfer_write to bufferize
   // out-of-place.
   %v2 = vector.transfer_read %t1[%idx], %cst : tensor<?xf32>, vector<10xf32>
   return %r, %v2 : tensor<?xf32>, vector<10xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @scf_if_out_of_place3
-func @scf_if_out_of_place3(%t1: tensor<?xf32> {linalg.inplaceable = true},
+func @scf_if_out_of_place3(%t1: tensor<?xf32> {bufferization.writable = true},
                            %v: vector<5xf32>, %idx: index,
                            %cond: i1, %cond2: i1) -> (tensor<?xf32>, vector<10xf32>) {
   %cst = arith.constant 0.0 : f32
   %r = scf.if %cond -> (tensor<?xf32>) {
     scf.yield %t1 : tensor<?xf32>
   } else {
     //      CHECK: vector.transfer_write
     // CHECK-SAME: {__inplace_operands_attr__ = ["none", "false", "none"]
func @scf_if_out_of_place3(%t1: tensor<?xf32> {bufferization.writable = true},
   }
   %v2 = vector.transfer_read %t1_alias[%idx], %cst : tensor<?xf32>, vector<10xf32>
   return %r, %v2 : tensor<?xf32>, vector<10xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @some_use
-func @some_use(%A : tensor<?xf32> {linalg.inplaceable = true},
+func @some_use(%A : tensor<?xf32> {bufferization.writable = true},
                %v : vector<5xf32>) -> (tensor<?xf32>) {
   %idx = arith.constant 0 : index
   //      CHECK: vector.transfer_write
   // CHECK-SAME: {__inplace_operands_attr__ = ["none", "true", "none"]
   %0 = vector.transfer_write %v, %A[%idx] : vector<5xf32>, tensor<?xf32>
   return %0 : tensor<?xf32>
 }
 
 
 // CHECK-LABEL: func @main_func
-func @main_func(%A : tensor<?xf32> {linalg.inplaceable = true},
+func @main_func(%A : tensor<?xf32> {bufferization.writable = true},
                 %v : vector<5xf32>) -> (tensor<?xf32>) {
   //      CHECK: call
   // CHECK-SAME: {__inplace_operands_attr__ = ["true", "none"]
   %0 = call @some_use(%A, %v) : (tensor<?xf32>, vector<5xf32>) -> (tensor<?xf32>)
   return %0 : tensor<?xf32>
 }
 
 // -----
func @to_tensor_op_not_writable(%m: memref<?xf32>, %v:  vector<5xf32>,
   %cst = arith.constant 0.0 : f32
   %r = vector.transfer_read %w[%idx2], %cst : tensor<?xf32>, vector<10xf32>
   return %r : vector<10xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @to_memref_op_is_reading
-func @to_memref_op_is_reading(%t1: tensor<?xf32> {linalg.inplaceable = true},
+func @to_memref_op_is_reading(%t1: tensor<?xf32> {bufferization.writable = true},
                               %idx1: index, %idx2: index, %idx3: index,
                               %v1: vector<5xf32>)
     -> (vector<5xf32>, vector<5xf32>) {
   // Write + read to/from tensor.
   //      CHECK: vector.transfer_write
   // CHECK-SAME: {__inplace_operands_attr__ = ["none", "false", "none"]
   %1 = vector.transfer_write %v1, %t1[%idx2] : vector<5xf32>, tensor<?xf32>
   %cst = arith.constant 0.0 : f32
func @equivalent_func_arg_2(%c0: index, %c10: index, %c1: index, %t0: tensor<?xf32>) -> tensor<?xf32> {
   return %1: tensor<?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @write_after_select_read_one
 //  CHECK-SAME:     %[[t1:.*]]: tensor<?xf32> {{.*}}, %[[t2:.*]]: tensor<?xf32>
 func @write_after_select_read_one(
-    %t1 : tensor<?xf32> {linalg.inplaceable = true},
-    %t2 : tensor<?xf32> {linalg.inplaceable = true},
+    %t1 : tensor<?xf32> {bufferization.writable = true},
+    %t2 : tensor<?xf32> {bufferization.writable = true},
     %c : i1)
   -> (f32, tensor<?xf32>)
 {
   %cst = arith.constant 0.0 : f32
   %idx = arith.constant 0 : index
 
   //      CHECK: arith.select %{{.*}}, %[[t1]], %[[t2]]
   // CHECK-SAME:   {__inplace_operands_attr__ = ["none", "false", "true"]}
   %s = arith.select %c, %t1, %t2 : tensor<?xf32>
   //      CHECK: tensor.insert
   // CHECK-SAME:   {__inplace_operands_attr__ = ["none", "true", "none"]}
   %w = tensor.insert %cst into %s[%idx] : tensor<?xf32>
   //      CHECK: tensor.extract
   // CHECK-SAME:   {__inplace_operands_attr__ = ["true", "none"]}
   %f = tensor.extract %t1[%idx] : tensor<?xf32>
 
   return %f, %w : f32, tensor<?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @write_after_select_read_both
 //  CHECK-SAME:     %[[t1:.*]]: tensor<?xf32> {{.*}}, %[[t2:.*]]: tensor<?xf32>
 func @write_after_select_read_both(
-    %t1 : tensor<?xf32> {linalg.inplaceable = true},
-    %t2 : tensor<?xf32> {linalg.inplaceable = true},
+    %t1 : tensor<?xf32> {bufferization.writable = true},
+    %t2 : tensor<?xf32> {bufferization.writable = true},
     %c : i1)
   -> (f32, f32, tensor<?xf32>)
 {
   %cst = arith.constant 0.0 : f32
   %idx = arith.constant 0 : index
 
   //      CHECK: arith.select %{{.*}}, %[[t1]], %[[t2]]
   // CHECK-SAME:   {__inplace_operands_attr__ = ["none", "false", "false"]}
func @write_after_select_read_both(
   return %f, %f2, %w : f32, f32, tensor<?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @write_after_select_no_conflict
 //  CHECK-SAME:     %[[t1:.*]]: tensor<?xf32> {{.*}}, %[[t2:.*]]: tensor<?xf32>
 func @write_after_select_no_conflict(
-    %t1 : tensor<?xf32> {linalg.inplaceable = true},
-    %t2 : tensor<?xf32> {linalg.inplaceable = true},
+    %t1 : tensor<?xf32> {bufferization.writable = true},
+    %t2 : tensor<?xf32> {bufferization.writable = true},
     %c : i1)
   -> (f32, tensor<?xf32>)
 {
   %cst = arith.constant 0.0 : f32
   %idx = arith.constant 0 : index
 
   //      CHECK: arith.select %{{.*}}, %[[t1]], %[[t2]]
   // CHECK-SAME:   {__inplace_operands_attr__ = ["none", "true", "true"]}

mlir/test/Dialect/Bufferization/Transforms/one-shot-module-bufferize-invalid.mlir

-// RUN: mlir-opt %s -allow-unregistered-dialect -linalg-comprehensive-module-bufferize -split-input-file -verify-diagnostics
+// RUN: mlir-opt %s -allow-unregistered-dialect -one-shot-bufferize="bufferize-function-boundaries=1" -split-input-file -verify-diagnostics
 
 func private @foo() -> tensor<?xf32>
 
 func @bar() -> tensor<?xf32> {
   %foo = constant @foo : () -> (tensor<?xf32>)
 // expected-error @+1 {{expected a CallOp}}
   %res = call_indirect %foo() : () -> (tensor<?xf32>)
   return %res : tensor<?xf32>
^bb1:
     return %T#0, %T#1 : tensor<f32>, tensor<f32>
   ^bb2:
     return %t2, %t1 : tensor<f32>, tensor<f32>
 }
 
 // -----
 
 func @scf_if_not_equivalent(
-    %cond: i1, %t1: tensor<?xf32> {linalg.inplaceable = true},
+    %cond: i1, %t1: tensor<?xf32> {bufferization.writable = true},
     %idx: index) -> tensor<?xf32> {
   %r = scf.if %cond -> (tensor<?xf32>) {
     scf.yield %t1 : tensor<?xf32>
   } else {
     // This buffer aliases, but it is not equivalent.
     %t2 = tensor.extract_slice %t1 [%idx] [%idx] [1] : tensor<?xf32> to tensor<?xf32>
     // expected-error @+1 {{operand #0 of ReturnLike op does not satisfy destination passing style}}
     scf.yield %t2 : tensor<?xf32>
   }
   // expected-error @+1 {{operand #0 of ReturnLike op does not satisfy destination passing style}}
   return %r : tensor<?xf32>
 }
 
 // -----
 
 func @scf_if_not_aliasing(
-    %cond: i1, %t1: tensor<?xf32> {linalg.inplaceable = true},
+    %cond: i1, %t1: tensor<?xf32> {bufferization.writable = true},
     %idx: index) -> f32 {
   %r = scf.if %cond -> (tensor<?xf32>) {
     scf.yield %t1 : tensor<?xf32>
   } else {
     // This buffer aliases.
     %t2 = linalg.init_tensor [%idx] : tensor<?xf32>
     // expected-error @+1 {{operand #0 of ReturnLike op does not satisfy destination passing style}}
     scf.yield %t2 : tensor<?xf32>

 func @bar() {
   call @foo() : () -> ()
   return
 }
 
 // -----
 
 func @scf_for(%A : tensor<?xf32>,
-              %B : tensor<?xf32> {linalg.inplaceable = true},
+              %B : tensor<?xf32> {bufferization.writable = true},
               %C : tensor<4xf32>,
               %lb : index, %ub : index, %step : index)
   -> (f32, f32)
 {
   %r0:2 = scf.for %i = %lb to %ub step %step iter_args(%tA = %A, %tB = %B)
       -> (tensor<?xf32>, tensor<?xf32>)
   {
     %ttA = tensor.insert_slice %C into %tA[0][4][1] : tensor<4xf32> into tensor<?xf32>
     %ttB = tensor.insert_slice %C into %tB[0][4][1] : tensor<4xf32> into tensor<?xf32>
 
     // Throw a wrench in the system by swapping yielded values: this result in a
     // ping-pong of values at each iteration on which we currently want to fail.
 
     // expected-error @+1 {{Yield operand #0 does not bufferize to a buffer that is aliasing}}
     scf.yield %ttB, %ttA : tensor<?xf32>, tensor<?xf32>
   }
 
   %f0 = tensor.extract %r0#0[%step] : tensor<?xf32>
   %f1 = tensor.extract %r0#1[%step] : tensor<?xf32>
   return %f0, %f1: f32, f32
 }
 
 // -----
 
-func private @fun_with_side_effects(%A: tensor<?xf32> {linalg.inplaceable = true})
+func private @fun_with_side_effects(%A: tensor<?xf32> {bufferization.writable = true})
 
-func @foo(%A: tensor<?xf32> {linalg.inplaceable = true}) -> (tensor<?xf32>) {
+func @foo(%A: tensor<?xf32> {bufferization.writable = true}) -> (tensor<?xf32>) {
   call @fun_with_side_effects(%A) : (tensor<?xf32>) -> ()
   return %A: tensor<?xf32>
 }
 
-func @scf_yield_needs_copy(%A : tensor<?xf32> {linalg.inplaceable = true}, %iters : index) {
+func @scf_yield_needs_copy(%A : tensor<?xf32> {bufferization.writable = true}, %iters : index) {
   %c0 = arith.constant 0 : index
   %c1 = arith.constant 1 : index
   %res = scf.for %arg0 = %c0 to %iters step %c1 iter_args(%bbarg = %A) -> (tensor<?xf32>) {
     %r = call @foo(%A) : (tensor<?xf32>) -> (tensor<?xf32>)
     // expected-error @+1 {{Yield operand #0 does not bufferize to a buffer that is aliasing}}
     scf.yield %r : tensor<?xf32>
   }
   call @fun_with_side_effects(%res) : (tensor<?xf32>) -> ()
   return
 }
 
 // -----
 
-func @extract_slice_fun(%A : tensor<?xf32> {linalg.inplaceable = true})
+func @extract_slice_fun(%A : tensor<?xf32> {bufferization.writable = true})
   ->  tensor<4xf32>
 {
   // This bufferizes to a pattern that the cross-function boundary pass needs to
   // convert into a new memref argument at all call site; this may be either:
   //   - an externally created aliasing subview (if we want to allow aliasing
   //     function arguments).
   //   - a new alloc + copy (more expensive but does not create new function
   //     argument aliasing).
func @mini_test_case1() -> tensor<10x20xf32> {
   return %r : tensor<10x20xf32>
 }
 
 // -----
 
 func @main() -> tensor<4xi32> {
   %r = scf.execute_region -> tensor<4xi32> {
     %A = arith.constant dense<[1, 2, 3, 4]> : tensor<4xi32>
+    // expected-error @+1 {{operand #0 of ReturnLike op does not satisfy destination passing style}}
     scf.yield %A: tensor<4xi32>
   }
 
   // expected-error @+1 {{operand #0 of ReturnLike op does not satisfy destination passing style}}
   return %r: tensor<4xi32>
 }
 
 // -----
 
 func @to_memref_op_is_writing(
-    %t1: tensor<?xf32> {linalg.inplaceable = true}, %idx1: index,
+    %t1: tensor<?xf32> {bufferization.writable = true}, %idx1: index,
     %idx2: index, %idx3: index, %v1: vector<5xf32>) -> (vector<5xf32>, vector<5xf32>) {
   // This is a RaW conflict because to_memref is an inplace write and %t1 is
   // read further down. This will likely have to change with partial
   // bufferization.
 
   // expected-error @+1 {{input IR has RaW conflict}}
   %0 = bufferization.to_memref %t1 : memref<?xf32>
 

mlir/test/Dialect/Bufferization/Transforms/one-shot-module-bufferize.mlir

-// RUN: mlir-opt %s -linalg-comprehensive-module-bufferize -split-input-file | FileCheck %s
+// RUN: mlir-opt %s -one-shot-bufferize="bufferize-function-boundaries=1" -split-input-file | FileCheck %s
 
 // Run fuzzer with different seeds.
-// RUN: mlir-opt %s -linalg-comprehensive-module-bufferize="allow-return-allocs test-analysis-only analysis-fuzzer-seed=23" -split-input-file -o /dev/null
-// RUN: mlir-opt %s -linalg-comprehensive-module-bufferize="allow-return-allocs test-analysis-only analysis-fuzzer-seed=59" -split-input-file -o /dev/null
-// RUN: mlir-opt %s -linalg-comprehensive-module-bufferize="allow-return-allocs test-analysis-only analysis-fuzzer-seed=91" -split-input-file -o /dev/null
+// RUN: mlir-opt %s -one-shot-bufferize="bufferize-function-boundaries=1 allow-return-allocs test-analysis-only analysis-fuzzer-seed=23" -split-input-file -o /dev/null
+// RUN: mlir-opt %s -one-shot-bufferize="bufferize-function-boundaries=1 allow-return-allocs test-analysis-only analysis-fuzzer-seed=59" -split-input-file -o /dev/null
+// RUN: mlir-opt %s -one-shot-bufferize="bufferize-function-boundaries=1 allow-return-allocs test-analysis-only analysis-fuzzer-seed=91" -split-input-file -o /dev/null
 
 // Test bufferization using memref types that have no layout map.
-// RUN: mlir-opt %s -linalg-comprehensive-module-bufferize="allow-return-allocs fully-dynamic-layout-maps=0" -split-input-file -o /dev/null
+// RUN: mlir-opt %s -one-shot-bufferize="bufferize-function-boundaries=1 allow-return-allocs fully-dynamic-layout-maps=0" -split-input-file | FileCheck %s --check-prefix=CHECK-NO-LAYOUT-MAP-LABEL
 
 // Bufferization of bodiless function with no tensor return value.
 
 // CHECK-LABEL: func private @private_func
 func private @private_func(tensor<?xf32>) -> ()
 
 // CHECK-LABEL: func @empty_func()
 func @empty_func() -> () {
   return
 }
 
 // -----
 
 // CHECK-LABEL: func private @private_func
 func private @private_func(tensor<?xf32>) -> (f32)
 
 // private_func may modify the buffer arg, but that's OK because %t is writable.
 // No alloc/copy should be inserted.
 
 // CHECK-LABEL: func @main(
 //  CHECK-SAME:     %[[t:.*]]: memref<?xf32
 //   CHECK-NOT: alloc
 //   CHECK-NOT: copy
 //       CHECK: call @private_func(%[[t]])
-func @main(%t: tensor<?xf32> {linalg.inplaceable = true}) -> (f32) {
+func @main(%t: tensor<?xf32> {bufferization.writable = true}) -> (f32) {
   %0 = call @private_func(%t) : (tensor<?xf32>) -> (f32)
   return %0 : f32
 }
 
 // -----
 
 // CHECK-LABEL: func private @private_func
 func private @private_func(tensor<?xf32>) -> (f32)
 
 // private_func may modify the buffer arg, %t is not writable. A copy is needed.
 
 // CHECK-LABEL: func @main(
 //  CHECK-SAME:     %[[t:.*]]: memref<?xf32
 //       CHECK: %[[alloc:.*]] = memref.alloc
 //   CHECK-DAG: memref.copy %[[t]], %[[alloc]]
 //   CHECK-DAG: %[[casted:.*]] = memref.cast %[[alloc]]
 //       CHECK: call @private_func(%[[casted]])
 //       CHECK: memref.dealloc %[[alloc]]
-func @main(%t: tensor<?xf32> {linalg.inplaceable = false}) -> (f32) {
+func @main(%t: tensor<?xf32> {bufferization.writable = false}) -> (f32) {
   %0 = call @private_func(%t) : (tensor<?xf32>) -> (f32)
   return %0 : f32
 }
 
 // -----
 
 // Test bufferization of a function without tensor args.
 
func @inner_func(%t: tensor<?xf32>) -> (tensor<?xf32>, f32) {
   %1 = tensor.extract %0[%c1] : tensor<?xf32>
   // CHECK: return %[[load]] : f32
   return %0, %1 : tensor<?xf32>, f32
 }
 
 // CHECK-LABEL: func @call_func_with_non_tensor_return(
 //  CHECK-SAME:     %[[arg0:.*]]: memref<?xf32
 func @call_func_with_non_tensor_return(
-    %t0: tensor<?xf32> {linalg.inplaceable = true}) -> (f32, tensor<?xf32>) {
+    %t0: tensor<?xf32> {bufferization.writable = true}) -> (f32, tensor<?xf32>) {
   // CHECK-NOT: alloc
   // CHECK-NOT: copy
   // CHECK: %[[call:.*]] = call @inner_func(%[[arg0]])
   %0, %1 = call @inner_func(%t0) : (tensor<?xf32>) -> (tensor<?xf32>, f32)
   // CHECK: return %[[call]] : f32
   return %1, %0 : f32, tensor<?xf32>
 }
 
func @inner_func(%t: tensor<?xf32>) -> (tensor<?xf32>, f32) {
   %1 = tensor.extract %0[%c1] : tensor<?xf32>
   // CHECK: return %[[load]] : f32
   return %0, %1 : tensor<?xf32>, f32
 }
 
 // CHECK-LABEL: func @call_func_with_non_tensor_return(
 //  CHECK-SAME:     %[[arg0:.*]]: memref<?xf32
 func @call_func_with_non_tensor_return(
-    %t0: tensor<?xf32> {linalg.inplaceable = false}) -> (f32, tensor<?xf32>) {
+    %t0: tensor<?xf32> {bufferization.writable = false}) -> (f32, tensor<?xf32>) {
   // CHECK: %[[alloc:.*]] = memref.alloc
   // CHECK-DAG: memref.copy %[[arg0]], %[[alloc]]
   // CHECK-DAG: %[[casted:.*]] = memref.cast %[[alloc]]
   // CHECK: %[[call:.*]] = call @inner_func(%[[casted]])
   %0, %1 = call @inner_func(%t0) : (tensor<?xf32>) -> (tensor<?xf32>, f32)
 
   // Note: The tensor return value has folded away.
   // CHECK: return %[[call]] : f32

 
 // CHECK-LABEL: func @main(
 //  CHECK-SAME:     %[[t3:.*]]: memref<?xf32
 //       CHECK: %[[alloc:.*]] = memref.alloc
 //   CHECK-DAG: memref.copy %[[t3]], %[[alloc]]
 //   CHECK-DAG: %[[casted:.*]] = memref.cast %[[alloc]]
 //       CHECK: call @f2(%[[casted]])
 //       CHECK: memref.dealloc %[[alloc]]
-func @main(%t: tensor<?xf32> {linalg.inplaceable = false}) -> (f32) {
+func @main(%t: tensor<?xf32> {bufferization.writable = false}) -> (f32) {
   %0 = call @f2(%t) : (tensor<?xf32>) -> (f32)
   return %0 : f32
 }
 
 // -----
 
 // This function does not read, just write. We need an alloc, but no copy.
 

 //  CHECK-SAME:     %[[t:.*]]: memref<?xf32
 //       CHECK:   %[[alloc:.*]] = memref.alloc
 //   CHECK-NOT:   copy
 //       CHECK: %[[casted:.*]] = memref.cast %[[alloc]]
 //   CHECK-NOT:   copy
 //       CHECK:   call @does_not_read(%[[casted]])
 //       CHECK:   %[[r:.*]] = memref.load %[[alloc]]
 //       CHECK:   memref.dealloc %[[alloc]]
-func @main(%t: tensor<?xf32> {linalg.inplaceable = false}) -> f32 {
+func @main(%t: tensor<?xf32> {bufferization.writable = false}) -> f32 {
   %0 = call @does_not_read(%t) : (tensor<?xf32>) -> (tensor<?xf32>)
   %idx = arith.constant 4 : index
   %r = tensor.extract %0[%idx] : tensor<?xf32>
   return %r : f32
 }
 
 // -----
 
func @scf_for_with_tensor_insert_slice(
   return %r0#1, %r0#0: tensor<?xf32>, tensor<?xf32>
 }
 
 //      CHECK:  func @bar(
 // CHECK-SAME:    %[[A:[a-zA-Z0-9]*]]: memref<?xf32, #[[$DYN_1D_MAP]]>
 // CHECK-SAME:    %[[B:[a-zA-Z0-9]*]]: memref<?xf32, #[[$DYN_1D_MAP]]>
 // CHECK-SAME:    %[[C:[a-zA-Z0-9]*]]: memref<4xf32, #[[$DYN_1D_MAP]]>
 func @bar(
-    %A : tensor<?xf32> {linalg.inplaceable = true},
-    %B : tensor<?xf32> {linalg.inplaceable = true},
-    %C : tensor<4xf32> {linalg.inplaceable = true},
+    %A : tensor<?xf32> {bufferization.writable = true},
+    %B : tensor<?xf32> {bufferization.writable = true},
+    %C : tensor<4xf32> {bufferization.writable = true},
     %lb : index, %ub : index, %step : index)
   -> (tensor<?xf32>, tensor<?xf32>)
 {
 //  CHECK-DAG:   call @scf_for_with_tensor_insert_slice(%[[A]], %[[B]], %[[C]]
   %r0:2 = call @scf_for_with_tensor_insert_slice(%A, %B, %C, %lb, %ub, %step) :
       (tensor<?xf32>, tensor<?xf32>, tensor<4xf32>, index, index, index)
         -> (tensor<?xf32>, tensor<?xf32>)
 

 
 // CHECK: func private @external_func(memref<?xf32, #[[$DYNAMIC]]>)
 func private @external_func(tensor<?xf32>)
 
 //      CHECK: func @callee(
 // CHECK-SAME:   %[[A:[0-9a-zA-Z]*]]: memref<?xf32>
 // CHECK-SAME:   %[[B:[0-9a-zA-Z]*]]: memref<?xf32, #[[$DYNAMIC]]>
 // CHECK-SAME:   %[[C:[0-9a-zA-Z]*]]: memref<?xf32, #[[$DYNAMIC]]>
-func @callee(%A : tensor<?xf32> {linalg.buffer_layout = affine_map<(i)[s0, s1] -> (i)>},
+func @callee(%A : tensor<?xf32> {bufferization.buffer_layout = affine_map<(i)[s0, s1] -> (i)>},
              %B : tensor<?xf32>,
              %C : tensor<?xf32>) {
 // CHECK-NEXT: %[[CASTED:.*]] = memref.cast %[[A]] : memref<?xf32> to memref<?xf32, #[[$DYNAMIC]]>
 // CHECK-NEXT: call @external_func(%[[CASTED]]) : (memref<?xf32, #[[$DYNAMIC]]>) -> ()
   call @external_func(%A) : (tensor<?xf32>) -> ()
 
 // CHECK-NEXT: call @external_func(%[[B]]) : (memref<?xf32, #[[$DYNAMIC]]>) -> ()
   call @external_func(%B) : (tensor<?xf32>) -> ()
 
 // CHECK-NEXT: call @external_func(%[[C]]) : (memref<?xf32, #[[$DYNAMIC]]>) -> ()
   call @external_func(%C) : (tensor<?xf32>) -> ()
 
   return
 }
 
 //      CHECK: func @entry(
 // CHECK-SAME:   %[[A:[0-9a-zA-Z]*]]: memref<?xf32>
 // CHECK-SAME:   %[[B:[0-9a-zA-Z]*]]: memref<?xf32>
 // CHECK-SAME:   %[[C:[0-9a-zA-Z]*]]: memref<?xf32, #[[$DYNAMIC]]>
-func @entry(%A : tensor<?xf32> {linalg.buffer_layout = affine_map<(i)[s0, s1] -> (i)>, linalg.inplaceable = false},
-            %B : tensor<?xf32> {linalg.buffer_layout = affine_map<(i)[s0, s1] -> (i)>, linalg.inplaceable = false},
-            %C : tensor<?xf32> {linalg.inplaceable = false}) {
+func @entry(%A : tensor<?xf32> {bufferization.buffer_layout = affine_map<(i)[s0, s1] -> (i)>, bufferization.writable = false},
+            %B : tensor<?xf32> {bufferization.buffer_layout = affine_map<(i)[s0, s1] -> (i)>, bufferization.writable = false},
+            %C : tensor<?xf32> {bufferization.writable = false}) {
 // Note: `callee` does not write to its bbArg directly, but `external_func`
 // does. Inside `callee`, the writes via `external_func` do not cause a
 // conflict. However, inside `entry`, the writes do cause a conflict because
 // %A, %B and %C are not inplaceable. This test case shows that this kind of
 // conflict detection has a "transitive" nature.
 //  CHECK-DAG: %[[ALLOC_C:.*]] = memref.alloc
 //  CHECK-DAG: %[[CASTED_C:.*]] = memref.cast %[[ALLOC_C]]
 //  CHECK-DAG: %[[ALLOC_B:.*]] = memref.alloc
func @inner_func(%t: tensor<?xf32>) -> tensor<?xf32> {
   %c0 = arith.constant 0 : index
   // CHECK: memref.store %{{.*}}, %[[arg0]]
   %0 = tensor.insert %f into %t[%c0] : tensor<?xf32>
   return %0 : tensor<?xf32>
 }
 
 // CHECK-LABEL: func @equivalent_func_arg(
 //  CHECK-SAME:     %[[arg0:.*]]: memref<?xf32
-func @equivalent_func_arg(%t0: tensor<?xf32> {linalg.inplaceable = true},
+func @equivalent_func_arg(%t0: tensor<?xf32> {bufferization.writable = true},
                           %c0: index, %c10: index, %c1: index) -> tensor<?xf32> {
   // CHECK-NOT: alloc
   // CHECK-NOT: copy
   %1 = scf.for %iv = %c0 to %c10 step %c1 iter_args(%t1 = %t0) -> (tensor<?xf32>) {
     // CHECK: call @inner_func(%[[arg0]])
     %3 = call @inner_func(%t1) : (tensor<?xf32>) -> tensor<?xf32>
     scf.yield %3 : tensor<?xf32>
   }
func @inner_func_2(%t: tensor<?xf32>) -> tensor<?xf32> {
   %c0 = arith.constant 0 : index
   // CHECK: memref.store %{{.*}}, %[[arg0]]
   %0 = tensor.insert %f into %t[%c0] : tensor<?xf32>
   return %0 : tensor<?xf32>
 }
 
 // CHECK-LABEL: func @equivalent_func_arg_2(
 //  CHECK-SAME:     %[[arg0:.*]]: memref<?xf32
-func @equivalent_func_arg_2(%t0: tensor<?xf32> {linalg.inplaceable = true},
+func @equivalent_func_arg_2(%t0: tensor<?xf32> {bufferization.writable = true},
                             %c0: index, %c10: index, %c1: index) -> tensor<?xf32> {
   // CHECK: scf.for {{.*}} {
   %1 = scf.for %iv = %c0 to %c10 step %c1 iter_args(%t1 = %t0) -> (tensor<?xf32>) {
     // CHECK: %[[alloc:.*]] = memref.alloc
     // CHECK: %[[casted:.*]] = memref.cast %[[alloc]]
     // CHECK: memref.copy %[[arg0]], %[[alloc]]
     // CHECK: call @inner_func_2(%[[casted]])
     // CHECK: memref.dealloc %[[alloc]]
     // CHECK-NOT: scf.yield
     %3 = call @inner_func_2(%t1) : (tensor<?xf32>) -> tensor<?xf32>
     scf.yield %t1 : tensor<?xf32>
   }
   return %1: tensor<?xf32>
 }
+
+// -----
+
+// Bufferize without fully dynamic layout maps.
+
+// CHECK-LABEL: func @transfer_read(%{{.*}}: memref<?xf32, #map>) -> vector<4xf32> {
+// CHECK-NO-LAYOUT-MAP-LABEL: func @transfer_read(%{{.*}}: memref<?xf32>) -> vector<4xf32>
+func @transfer_read(
+    %A : tensor<?xf32> {bufferization.writable = false})
+  -> (vector<4xf32>)
+{
+  %c0 = arith.constant 0 : index
+  %f0 = arith.constant 0.0 : f32
+
+//       CHECK: %[[RES:.*]] = vector.transfer_read {{.*}} : memref<?xf32, #{{.*}}>, vector<4xf32>
+  %0 = vector.transfer_read %A[%c0], %f0 : tensor<?xf32>, vector<4xf32>
+
+//       CHECK: return %[[RES]] : vector<4xf32>
+  return %0 : vector<4xf32>
+}

mlir/test/Dialect/Linalg/comprehensive-bufferize-analysis-2fill-extract-matmul-all-perms.mlir

 // RUN: mlir-opt %s -linalg-comprehensive-module-bufferize=test-analysis-only -split-input-file | FileCheck %s
 
 /// All combinations of matmul(fill(extract(init_tensor)), fill(extract(%init_tensor)), %arg2)
 /// These should all be inplaceable except the first op.
 
 // -----
 
 // CHECK-LABEL: func @fill_extract_matmul_
 func @fill_extract_matmul_1234(
-    %arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+    %arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
   -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
   %cst_0 = arith.constant 1.000000e+00 : f32
   %0 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["none", "false"]}
   %1 = linalg.fill ins(%cst : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %2 = linalg.fill ins(%cst_0 : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %3 = tensor.extract_slice %1[0, 0] [256, 16] [1, 1] : tensor<256x256xf32> to tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %4 = tensor.extract_slice %2[0, 0] [16, 256] [1, 1] : tensor<256x256xf32> to tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true", "true", "true"]}
   %5 = linalg.matmul ins(%3, %4 : tensor<256x16xf32>, tensor<16x256xf32>) outs(%arg2 : tensor<256x256xf32>) -> tensor<256x256xf32>
   return %5 : tensor<256x256xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @fill_extract_matmul_
 func @fill_extract_matmul_1243(
-    %arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+    %arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
   -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
   %cst_0 = arith.constant 1.000000e+00 : f32
   %0 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["none", "false"]}
   %1 = linalg.fill ins(%cst : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %2 = linalg.fill ins(%cst_0 : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %4 = tensor.extract_slice %2[0, 0] [16, 256] [1, 1] : tensor<256x256xf32> to tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %3 = tensor.extract_slice %1[0, 0] [256, 16] [1, 1] : tensor<256x256xf32> to tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["true", "true", "true"]}
   %5 = linalg.matmul ins(%3, %4 : tensor<256x16xf32>, tensor<16x256xf32>) outs(%arg2 : tensor<256x256xf32>) -> tensor<256x256xf32>
   return %5 : tensor<256x256xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @fill_extract_matmul_
-func @fill_extract_matmul_1324(%arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-                        %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-                        %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+func @fill_extract_matmul_1324(%arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+                        %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+                        %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
   -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
   %cst_0 = arith.constant 1.000000e+00 : f32
   %0 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["none", "false"]}
   %1 = linalg.fill ins(%cst : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %3 = tensor.extract_slice %1[0, 0] [256, 16] [1, 1] : tensor<256x256xf32> to tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %2 = linalg.fill ins(%cst_0 : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %4 = tensor.extract_slice %2[0, 0] [16, 256] [1, 1] : tensor<256x256xf32> to tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true", "true", "true"]}
   %5 = linalg.matmul ins(%3, %4 : tensor<256x16xf32>, tensor<16x256xf32>) outs(%arg2 : tensor<256x256xf32>) -> tensor<256x256xf32>
   return %5 : tensor<256x256xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @fill_extract_matmul_
-func @fill_extract_matmul_1342(%arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-                        %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-                        %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+func @fill_extract_matmul_1342(%arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+                        %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+                        %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
   -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
   %cst_0 = arith.constant 1.000000e+00 : f32
   %0 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["none", "false"]}
   %1 = linalg.fill ins(%cst : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %3 = tensor.extract_slice %1[0, 0] [256, 16] [1, 1] : tensor<256x256xf32> to tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %4 = tensor.extract_slice %0[0, 0] [16, 256] [1, 1] : tensor<256x256xf32> to tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %2 = linalg.fill ins(%cst_0 : f32) outs(%4 : tensor<16x256xf32>) -> tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true", "true", "true"]}
   %5 = linalg.matmul ins(%3, %2 : tensor<256x16xf32>, tensor<16x256xf32>) outs(%arg2 : tensor<256x256xf32>) -> tensor<256x256xf32>
   return %5 : tensor<256x256xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @fill_extract_matmul_
-func @fill_extract_matmul_1423(%arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-                        %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-                        %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+func @fill_extract_matmul_1423(%arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+                        %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+                        %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
   -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
   %cst_0 = arith.constant 1.000000e+00 : f32
   %0 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["none", "false"]}
   %1 = linalg.fill ins(%cst : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %4 = tensor.extract_slice %0[0, 0] [16, 256] [1, 1] : tensor<256x256xf32> to tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %2 = linalg.fill ins(%cst_0 : f32) outs(%4 : tensor<16x256xf32>) -> tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %3 = tensor.extract_slice %1[0, 0] [256, 16] [1, 1] : tensor<256x256xf32> to tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["true", "true", "true"]}
   %5 = linalg.matmul ins(%3, %2 : tensor<256x16xf32>, tensor<16x256xf32>) outs(%arg2 : tensor<256x256xf32>) -> tensor<256x256xf32>
   return %5 : tensor<256x256xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @fill_extract_matmul_
-func @fill_extract_matmul_1432(%arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-                        %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-                        %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+func @fill_extract_matmul_1432(%arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+                        %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+                        %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
   -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
   %cst_0 = arith.constant 1.000000e+00 : f32
   %0 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["none", "false"]}
   %1 = linalg.fill ins(%cst : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %4 = tensor.extract_slice %0[0, 0] [16, 256] [1, 1] : tensor<256x256xf32> to tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %3 = tensor.extract_slice %1[0, 0] [256, 16] [1, 1] : tensor<256x256xf32> to tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %2 = linalg.fill ins(%cst_0 : f32) outs(%4 : tensor<16x256xf32>) -> tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true", "true", "true"]}
   %5 = linalg.matmul ins(%3, %2 : tensor<256x16xf32>, tensor<16x256xf32>) outs(%arg2 : tensor<256x256xf32>) -> tensor<256x256xf32>
   return %5 : tensor<256x256xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @fill_extract_matmul_
 func @fill_extract_matmul_2134(
-    %arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+    %arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
   -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
   %cst_0 = arith.constant 1.000000e+00 : f32
   %0 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["none", "false"]}
   %2 = linalg.fill ins(%cst_0 : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %1 = linalg.fill ins(%cst : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %3 = tensor.extract_slice %1[0, 0] [256, 16] [1, 1] : tensor<256x256xf32> to tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %4 = tensor.extract_slice %2[0, 0] [16, 256] [1, 1] : tensor<256x256xf32> to tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true", "true", "true"]}
   %5 = linalg.matmul ins(%3, %4 : tensor<256x16xf32>, tensor<16x256xf32>) outs(%arg2 : tensor<256x256xf32>) -> tensor<256x256xf32>
   return %5 : tensor<256x256xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @fill_extract_matmul_
 func @fill_extract_matmul_2143(
-    %arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+    %arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
   -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
   %cst_0 = arith.constant 1.000000e+00 : f32
   %0 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["none", "false"]}
   %2 = linalg.fill ins(%cst_0 : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %1 = linalg.fill ins(%cst : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %4 = tensor.extract_slice %2[0, 0] [16, 256] [1, 1] : tensor<256x256xf32> to tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %3 = tensor.extract_slice %1[0, 0] [256, 16] [1, 1] : tensor<256x256xf32> to tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["true", "true", "true"]}
   %5 = linalg.matmul ins(%3, %4 : tensor<256x16xf32>, tensor<16x256xf32>) outs(%arg2 : tensor<256x256xf32>) -> tensor<256x256xf32>
   return %5 : tensor<256x256xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @fill_extract_matmul_
 func @fill_extract_matmul_2314(
-    %arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+    %arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
   -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
   %cst_0 = arith.constant 1.000000e+00 : f32
   %0 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["none", "false"]}
   %2 = linalg.fill ins(%cst_0 : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %3 = tensor.extract_slice %0[0, 0] [256, 16] [1, 1] : tensor<256x256xf32> to tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %1 = linalg.fill ins(%cst : f32) outs(%3 : tensor<256x16xf32>) -> tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %4 = tensor.extract_slice %2[0, 0] [16, 256] [1, 1] : tensor<256x256xf32> to tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true", "true", "true"]}
   %5 = linalg.matmul ins(%1, %4 : tensor<256x16xf32>, tensor<16x256xf32>) outs(%arg2 : tensor<256x256xf32>) -> tensor<256x256xf32>
   return %5 : tensor<256x256xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @fill_extract_matmul_
 func @fill_extract_matmul_2341(
-    %arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+    %arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
   -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
   %cst_0 = arith.constant 1.000000e+00 : f32
   %0 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["none", "false"]}
   %2 = linalg.fill ins(%cst_0 : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %3 = tensor.extract_slice %0[0, 0] [256, 16] [1, 1] : tensor<256x256xf32> to tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %4 = tensor.extract_slice %2[0, 0] [16, 256] [1, 1] : tensor<256x256xf32> to tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %1 = linalg.fill ins(%cst : f32) outs(%3 : tensor<256x16xf32>) -> tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["true", "true", "true"]}
   %5 = linalg.matmul ins(%1, %4 : tensor<256x16xf32>, tensor<16x256xf32>) outs(%arg2 : tensor<256x256xf32>) -> tensor<256x256xf32>
   return %5 : tensor<256x256xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @fill_extract_matmul_
 func @fill_extract_matmul_2413(
-    %arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+    %arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
   -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
   %cst_0 = arith.constant 1.000000e+00 : f32
   %0 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["none", "false"]}
   %2 = linalg.fill ins(%cst_0 : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %4 = tensor.extract_slice %2[0, 0] [16, 256] [1, 1] : tensor<256x256xf32> to tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %1 = linalg.fill ins(%cst : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %3 = tensor.extract_slice %1[0, 0] [256, 16] [1, 1] : tensor<256x256xf32> to tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["true", "true", "true"]}
   %5 = linalg.matmul ins(%3, %4 : tensor<256x16xf32>, tensor<16x256xf32>) outs(%arg2 : tensor<256x256xf32>) -> tensor<256x256xf32>
   return %5 : tensor<256x256xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @fill_extract_matmul_
 func @fill_extract_matmul_2431(
-    %arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+    %arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
   -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
   %cst_0 = arith.constant 1.000000e+00 : f32
   %0 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["none", "false"]}
   %2 = linalg.fill ins(%cst_0 : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %4 = tensor.extract_slice %2[0, 0] [16, 256] [1, 1] : tensor<256x256xf32> to tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %3 = tensor.extract_slice %0[0, 0] [256, 16] [1, 1] : tensor<256x256xf32> to tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %1 = linalg.fill ins(%cst : f32) outs(%3 : tensor<256x16xf32>) -> tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["true", "true", "true"]}
   %5 = linalg.matmul ins(%1, %4 : tensor<256x16xf32>, tensor<16x256xf32>) outs(%arg2 : tensor<256x256xf32>) -> tensor<256x256xf32>
   return %5 : tensor<256x256xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @fill_extract_matmul_
 func @fill_extract_matmul_3124(
-    %arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+    %arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
   -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
   %cst_0 = arith.constant 1.000000e+00 : f32
   %0 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["false"]}
   %3 = tensor.extract_slice %0[0, 0] [256, 16] [1, 1] : tensor<256x256xf32> to tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %1 = linalg.fill ins(%cst : f32) outs(%3 : tensor<256x16xf32>) -> tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %2 = linalg.fill ins(%cst_0 : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %4 = tensor.extract_slice %2[0, 0] [16, 256] [1, 1] : tensor<256x256xf32> to tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true", "true", "true"]}
   %5 = linalg.matmul ins(%1, %4 : tensor<256x16xf32>, tensor<16x256xf32>) outs(%arg2 : tensor<256x256xf32>) -> tensor<256x256xf32>
   return %5 : tensor<256x256xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @fill_extract_matmul_
 func @fill_extract_matmul_3142(
-    %arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+    %arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
   -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
   %cst_0 = arith.constant 1.000000e+00 : f32
   %0 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["false"]}
   %3 = tensor.extract_slice %0[0, 0] [256, 16] [1, 1] : tensor<256x256xf32> to tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %1 = linalg.fill ins(%cst : f32) outs(%3 : tensor<256x16xf32>) -> tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %4 = tensor.extract_slice %0[0, 0] [16, 256] [1, 1] : tensor<256x256xf32> to tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %2 = linalg.fill ins(%cst_0 : f32) outs(%4 : tensor<16x256xf32>) -> tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true", "true", "true"]}
   %5 = linalg.matmul ins(%1, %2 : tensor<256x16xf32>, tensor<16x256xf32>) outs(%arg2 : tensor<256x256xf32>) -> tensor<256x256xf32>
   return %5 : tensor<256x256xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @fill_extract_matmul_
 func @fill_extract_matmul_3214(
-    %arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+    %arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
   -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
   %cst_0 = arith.constant 1.000000e+00 : f32
   %0 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["false"]}
   %3 = tensor.extract_slice %0[0, 0] [256, 16] [1, 1] : tensor<256x256xf32> to tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %2 = linalg.fill ins(%cst_0 : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %1 = linalg.fill ins(%cst : f32) outs(%3 : tensor<256x16xf32>) -> tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %4 = tensor.extract_slice %2[0, 0] [16, 256] [1, 1] : tensor<256x256xf32> to tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true", "true", "true"]}
   %5 = linalg.matmul ins(%1, %4 : tensor<256x16xf32>, tensor<16x256xf32>) outs(%arg2 : tensor<256x256xf32>) -> tensor<256x256xf32>
   return %5 : tensor<256x256xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @fill_extract_matmul_
 func @fill_extract_matmul_3241(
-    %arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+    %arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
   -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
   %cst_0 = arith.constant 1.000000e+00 : f32
   %0 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["false"]}
   %3 = tensor.extract_slice %0[0, 0] [256, 16] [1, 1] : tensor<256x256xf32> to tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %2 = linalg.fill ins(%cst_0 : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %4 = tensor.extract_slice %2[0, 0] [16, 256] [1, 1] : tensor<256x256xf32> to tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %1 = linalg.fill ins(%cst : f32) outs(%3 : tensor<256x16xf32>) -> tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["true", "true", "true"]}
   %5 = linalg.matmul ins(%1, %4 : tensor<256x16xf32>, tensor<16x256xf32>) outs(%arg2 : tensor<256x256xf32>) -> tensor<256x256xf32>
   return %5 : tensor<256x256xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @fill_extract_matmul_
 func @fill_extract_matmul_3412(
-    %arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+    %arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
   -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
   %cst_0 = arith.constant 1.000000e+00 : f32
   %0 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["false"]}
   %3 = tensor.extract_slice %0[0, 0] [256, 16] [1, 1] : tensor<256x256xf32> to tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %4 = tensor.extract_slice %0[0, 0] [16, 256] [1, 1] : tensor<256x256xf32> to tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %1 = linalg.fill ins(%cst : f32) outs(%3 : tensor<256x16xf32>) -> tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %2 = linalg.fill ins(%cst_0 : f32) outs(%4 : tensor<16x256xf32>) -> tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true", "true", "true"]}
   %5 = linalg.matmul ins(%1, %2 : tensor<256x16xf32>, tensor<16x256xf32>) outs(%arg2 : tensor<256x256xf32>) -> tensor<256x256xf32>
   return %5 : tensor<256x256xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @fill_extract_matmul_
 func @fill_extract_matmul_3421(
-    %arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+    %arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
   -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
   %cst_0 = arith.constant 1.000000e+00 : f32
   %0 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["false"]}
   %3 = tensor.extract_slice %0[0, 0] [256, 16] [1, 1] : tensor<256x256xf32> to tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %4 = tensor.extract_slice %0[0, 0] [16, 256] [1, 1] : tensor<256x256xf32> to tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %2 = linalg.fill ins(%cst_0 : f32) outs(%4 : tensor<16x256xf32>) -> tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %1 = linalg.fill ins(%cst : f32) outs(%3 : tensor<256x16xf32>) -> tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["true", "true", "true"]}
   %5 = linalg.matmul ins(%1, %2 : tensor<256x16xf32>, tensor<16x256xf32>) outs(%arg2 : tensor<256x256xf32>) -> tensor<256x256xf32>
   return %5 : tensor<256x256xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @fill_extract_matmul_
 func @fill_extract_matmul_4123(
-    %arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+    %arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
   -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
   %cst_0 = arith.constant 1.000000e+00 : f32
   %0 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["false"]}
   %4 = tensor.extract_slice %0[0, 0] [16, 256] [1, 1] : tensor<256x256xf32> to tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %1 = linalg.fill ins(%cst : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %2 = linalg.fill ins(%cst_0 : f32) outs(%4 : tensor<16x256xf32>) -> tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %3 = tensor.extract_slice %1[0, 0] [256, 16] [1, 1] : tensor<256x256xf32> to tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["true", "true", "true"]}
   %5 = linalg.matmul ins(%3, %2 : tensor<256x16xf32>, tensor<16x256xf32>) outs(%arg2 : tensor<256x256xf32>) -> tensor<256x256xf32>
   return %5 : tensor<256x256xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @fill_extract_matmul_
 func @fill_extract_matmul_4132(
-    %arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+    %arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
   -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
   %cst_0 = arith.constant 1.000000e+00 : f32
   %0 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["false"]}
   %4 = tensor.extract_slice %0[0, 0] [16, 256] [1, 1] : tensor<256x256xf32> to tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %1 = linalg.fill ins(%cst : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %3 = tensor.extract_slice %1[0, 0] [256, 16] [1, 1] : tensor<256x256xf32> to tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %2 = linalg.fill ins(%cst_0 : f32) outs(%4 : tensor<16x256xf32>) -> tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true", "true", "true"]}
   %5 = linalg.matmul ins(%3, %2 : tensor<256x16xf32>, tensor<16x256xf32>) outs(%arg2 : tensor<256x256xf32>) -> tensor<256x256xf32>
   return %5 : tensor<256x256xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @fill_extract_matmul_
 func @fill_extract_matmul_4213(
-    %arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+    %arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
   -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
   %cst_0 = arith.constant 1.000000e+00 : f32
   %0 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["false"]}
   %4 = tensor.extract_slice %0[0, 0] [16, 256] [1, 1] : tensor<256x256xf32> to tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %2 = linalg.fill ins(%cst_0 : f32) outs(%4 : tensor<16x256xf32>) -> tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %1 = linalg.fill ins(%cst : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %3 = tensor.extract_slice %1[0, 0] [256, 16] [1, 1] : tensor<256x256xf32> to tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["true", "true", "true"]}
   %5 = linalg.matmul ins(%3, %2 : tensor<256x16xf32>, tensor<16x256xf32>) outs(%arg2 : tensor<256x256xf32>) -> tensor<256x256xf32>
   return %5 : tensor<256x256xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @fill_extract_matmul_
 func @fill_extract_matmul_4231(
-    %arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+    %arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
   -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
   %cst_0 = arith.constant 1.000000e+00 : f32
   %0 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["false"]}
   %4 = tensor.extract_slice %0[0, 0] [16, 256] [1, 1] : tensor<256x256xf32> to tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %2 = linalg.fill ins(%cst_0 : f32) outs(%4 : tensor<16x256xf32>) -> tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %3 = tensor.extract_slice %0[0, 0] [256, 16] [1, 1] : tensor<256x256xf32> to tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %1 = linalg.fill ins(%cst : f32) outs(%3 : tensor<256x16xf32>) -> tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["true", "true", "true"]}
   %5 = linalg.matmul ins(%1, %2 : tensor<256x16xf32>, tensor<16x256xf32>) outs(%arg2 : tensor<256x256xf32>) -> tensor<256x256xf32>
   return %5 : tensor<256x256xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @fill_extract_matmul_
 func @fill_extract_matmul_4312(
-    %arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+    %arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
   -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
   %cst_0 = arith.constant 1.000000e+00 : f32
   %0 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["false"]}
   %4 = tensor.extract_slice %0[0, 0] [16, 256] [1, 1] : tensor<256x256xf32> to tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true"]}
   %3 = tensor.extract_slice %0[0, 0] [256, 16] [1, 1] : tensor<256x256xf32> to tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %1 = linalg.fill ins(%cst : f32) outs(%3 : tensor<256x16xf32>) -> tensor<256x16xf32>
   // CHECK: {__inplace_operands_attr__ = ["none", "true"]}
   %2 = linalg.fill ins(%cst_0 : f32) outs(%4 : tensor<16x256xf32>) -> tensor<16x256xf32>
   // CHECK: {__inplace_operands_attr__ = ["true", "true", "true"]}
   %5 = linalg.matmul ins(%1, %2 : tensor<256x16xf32>, tensor<16x256xf32>) outs(%arg2 : tensor<256x256xf32>) -> tensor<256x256xf32>
   return %5 : tensor<256x256xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @fill_extract_matmul_
 func @fill_extract_matmul_4321(
-    %arg0: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg1: tensor<518x518xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %arg2: tensor<256x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+    %arg0: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg1: tensor<518x518xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
+    %arg2: tensor<256x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
   -> tensor<256x256xf32>
 {
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f32
   %cst_0 = arith.constant 1.000000e+00 : f32
   %0 = linalg.init_tensor [256, 256] : tensor<256x256xf32>
 
   // CHECK: {__inplace_operands_attr__ = ["false"]}

mlir/test/Dialect/Linalg/comprehensive-module-bufferize-aliasing-in.mlir

 // RUN: mlir-opt %s -linalg-comprehensive-module-bufferize="allow-return-allocs always-aliasing-with-dest=0" -split-input-file | FileCheck %s
 
 // CHECK-LABEL: func @linalg_op_bufferizes_inplace_with_input
 //  CHECK-SAME:     %[[t1:.*]]: memref<?x?xf32, #{{.*}}>, %[[t2:.*]]: memref<?xf32, #{{.*}}>, %[[t3:.*]]: memref<?x?xf32, #{{.*}}>
 func @linalg_op_bufferizes_inplace_with_input(
-    %t1: tensor<?x?xf32> {linalg.inplaceable = true},
-    %t2: tensor<?xf32> {linalg.inplaceable = false},
-    %t3: tensor<?x?xf32> {linalg.inplaceable = false},
+    %t1: tensor<?x?xf32> {bufferization.writable = true},
+    %t2: tensor<?xf32> {bufferization.writable = false},
+    %t3: tensor<?x?xf32> {bufferization.writable = false},
     %s1: index, %s2: index, %cst: f32) -> tensor<?x?xf32> {
   // CHECK: linalg.generic {{.*}} ins(%[[t1]], %[[t2]] : {{.*}}) outs(%[[t1]] : {{.*}})
   %r = linalg.generic {
     indexing_maps = [affine_map<(d0, d1) -> (d0, d1)>,
                      affine_map<(d0, d1) -> (d1)>,
                      affine_map<(d0, d1)-> (d0, d1)>],
     iterator_types = ["parallel", "parallel"]}
     ins(%t1, %t2 : tensor<?x?xf32>, tensor<?xf32>)
     outs(%t3 : tensor<?x?xf32>) {
       ^bb0(%arg0 : f32, %arg1 : f32, %arg2 : f32) :
         %add = arith.addf %arg0, %arg1 : f32
         linalg.yield %add : f32
     } -> tensor<?x?xf32>
   return %r : tensor<?x?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @linalg_op_bufferizes_out_of_place_with_input
 //  CHECK-SAME:     %[[t1:.*]]: memref<?x?xf32, #{{.*}}>, %[[t2:.*]]: memref<?xf32, #{{.*}}>, %[[t3:.*]]: memref<?x?xf32, #{{.*}}>
 func @linalg_op_bufferizes_out_of_place_with_input(
-    %t1: tensor<?x?xf32> {linalg.inplaceable = false},
-    %t2: tensor<?xf32> {linalg.inplaceable = false},
-    %t3: tensor<?x?xf32> {linalg.inplaceable = false},
+    %t1: tensor<?x?xf32> {bufferization.writable = false},
+    %t2: tensor<?xf32> {bufferization.writable = false},
+    %t3: tensor<?x?xf32> {bufferization.writable = false},
     %s1: index, %s2: index, %cst: f32) -> tensor<?x?xf32> {
   // CHECK: %[[alloc:.*]] = memref.alloc
   // CHECK: memref.copy %[[t1]], %[[alloc]]
   // CHECK: linalg.generic {{.*}} ins(%[[t1]], %[[t2]] : {{.*}}) outs(%[[alloc]] : {{.*}})
   %r = linalg.generic {
     indexing_maps = [affine_map<(d0, d1) -> (d0, d1)>,
                      affine_map<(d0, d1) -> (d1)>,
                      affine_map<(d0, d1)-> (d0, d1)>],
     iterator_types = ["parallel", "parallel"]}
     ins(%t1, %t2 : tensor<?x?xf32>, tensor<?xf32>)
     outs(%t3 : tensor<?x?xf32>) {
       ^bb0(%arg0 : f32, %arg1 : f32, %arg2 : f32) :
         %add = arith.addf %arg0, %arg1 : f32
         linalg.yield %add : f32
     } -> tensor<?x?xf32>
   // CHECK: return %[[alloc]]
   return %r : tensor<?x?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @linalg_op_output_cannot_alias_with_input
 //  CHECK-SAME:     %[[t1:.*]]: memref<?x?xf32, #{{.*}}>, %[[t2:.*]]: memref<?xf32, #{{.*}}>, %[[t3:.*]]: memref<?x?xf32, #{{.*}}>
 func @linalg_op_output_cannot_alias_with_input(
-    %t1: tensor<?x?xf32> {linalg.inplaceable = true},
-    %t2: tensor<?xf32> {linalg.inplaceable = false},
-    %t3: tensor<?x?xf32> {linalg.inplaceable = true},
+    %t1: tensor<?x?xf32> {bufferization.writable = true},
+    %t2: tensor<?xf32> {bufferization.writable = false},
+    %t3: tensor<?x?xf32> {bufferization.writable = true},
     %s1: index, %s2: index, %cst: f32) -> tensor<?x?xf32> {
   // CHECK: linalg.generic {{.*}} ins(%[[t1]], %[[t2]] : {{.*}}) outs(%[[t3]] : {{.*}})
   %r = linalg.generic {
     indexing_maps = [affine_map<(d0, d1) -> (d1, d0)>,
                      affine_map<(d0, d1) -> (d1)>,
                      affine_map<(d0, d1)-> (d0, d1)>],
     iterator_types = ["parallel", "parallel"]}
     ins(%t1, %t2 : tensor<?x?xf32>, tensor<?xf32>)
     outs(%t3 : tensor<?x?xf32>) {
       ^bb0(%arg0 : f32, %arg1 : f32, %arg2 : f32) :
         %add = arith.addf %arg0, %arg1 : f32
         linalg.yield %add : f32
     } -> tensor<?x?xf32>
   return %r : tensor<?x?xf32>
 }
 

mlir/test/Dialect/Linalg/comprehensive-module-bufferize-analysis-aliasing-in.mlir

 ]
 #trait = {
   indexing_maps = #accesses,
   iterator_types = ["parallel"]
 }
 
 // CHECK-LABEL: func @linalg_op_same_out_tensors(
 func @linalg_op_same_out_tensors(
-    %t1: tensor<?xf32> {linalg.inplaceable = true},
+    %t1: tensor<?xf32> {bufferization.writable = true},
 // CHECK-SAME:          bufferization.access = "read-write"
-    %t2: tensor<?xf32> {linalg.inplaceable = true})
+    %t2: tensor<?xf32> {bufferization.writable = true})
 // CHECK-SAME:          bufferization.access = "write"
   -> (tensor<?xf32>, tensor<?xf32>){
 
   // %1 and %2 are not used in the computation, so the two OpResults do not
   // necessarily have to bufferize in-place with the two "out" OpOperands. They
   // bufferize in-place with the first and second OpOperand (one of which is an
   // "in" OpOperand).
   //      CHECK: linalg.generic

 ]
 #trait = {
   indexing_maps = #accesses,
   iterator_types = ["parallel"]
 }
 
 // CHECK-LABEL: func @linalg_op_same_out_tensors_2(
 func @linalg_op_same_out_tensors_2(
-    %t1: tensor<?xf32> {linalg.inplaceable = true},
+    %t1: tensor<?xf32> {bufferization.writable = true},
 // CHECK-SAME:          bufferization.access = "read-write"
-    %t2: tensor<?xf32> {linalg.inplaceable = true})
+    %t2: tensor<?xf32> {bufferization.writable = true})
 // CHECK-SAME:          bufferization.access = "write"
         -> (tensor<?xf32>, tensor<?xf32>, tensor<?xf32>){
 
   // %1, %2 and %3 are not used in the computation, so the three OpResults do
   // not necessarily have to bufferize in-place with the three "out" OpOperands.
   // They bufferize in-place with the first, second and third OpOperand (one of
   // which is an "in" OpOperand).
   // In contrast to the previous test case, two of the chosen OpOperands are the

mlir/test/Dialect/Linalg/comprehensive-module-bufferize-analysis-init-tensor-elimination.mlir

 // RUN: mlir-opt %s -linalg-comprehensive-module-bufferize="test-analysis-only allow-return-allocs init-tensor-elimination" -split-input-file | FileCheck %s
 
 // -----
 
 //===----------------------------------------------------------------------===//
 // InitTensorOp elimination
 //===----------------------------------------------------------------------===//
 
 // CHECK-LABEL: func @buffer_forwarding_conflict
-func @buffer_forwarding_conflict(%arg0: tensor<?xf32> {linalg.inplaceable = true}, %arg1: index) -> (tensor<?xf32>, tensor<?xf32>) {
+func @buffer_forwarding_conflict(%arg0: tensor<?xf32> {bufferization.writable = true}, %arg1: index) -> (tensor<?xf32>, tensor<?xf32>) {
   %cst = arith.constant 0.000000e+00 : f32
   //      CHECK: tensor.extract_slice
   // CHECK-SAME: {__inplace_operands_attr__ = ["false", "none"]
   // Instead of allocating, share buffer with some inplace bufferization?
   %0 = linalg.init_tensor [%arg1] : tensor<?xf32>
 
   //      CHECK: linalg.fill
   // CHECK-SAME: {__inplace_operands_attr__ = ["none", "true"]
func @buffer_forwarding_conflict(%arg0: tensor<?xf32> {bufferization.writable = true}, %arg1: index) -> (tensor<?xf32>, tensor<?xf32>) {
   //      CHECK: return
   // CHECK-SAME: __equivalent_func_args__ = [-1, 0]
   return %2, %3 : tensor<?xf32>, tensor<?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @buffer_forwarding_no_conflict
-func @buffer_forwarding_no_conflict(%arg0: tensor<?xf32> {linalg.inplaceable = true}, %arg1: index) -> (tensor<?xf32>, tensor<?xf32>) {
+func @buffer_forwarding_no_conflict(%arg0: tensor<?xf32> {bufferization.writable = true}, %arg1: index) -> (tensor<?xf32>, tensor<?xf32>) {
   %cst = arith.constant 0.000000e+00 : f32
   //      CHECK: tensor.extract_slice
   // CHECK-SAME: {__inplace_operands_attr__ = ["true", "none"]
   // Instead of allocating, share buffer with some inplace bufferization?
   %0 = linalg.init_tensor [%arg1] : tensor<?xf32>
 
   //      CHECK: linalg.fill
   // CHECK-SAME: {__inplace_operands_attr__ = ["none", "true"]

mlir/test/Dialect/Linalg/comprehensive-module-bufferize-init-tensor-elimination.mlir

 // RUN: mlir-opt %s -linalg-comprehensive-module-bufferize="allow-return-allocs init-tensor-elimination" -canonicalize -split-input-file | FileCheck %s
 
 // -----
 
 //      CHECK: func @buffer_forwarding_conflict(
 // CHECK-SAME:   %[[FUNC_ARG:[0-9a-zA-Z]*]]: memref<?xf32>
 // CHECK-SAME:   %[[sz:[0-9a-zA-Z]*]]: index
 func @buffer_forwarding_conflict(
-  %t: tensor<?xf32> {linalg.buffer_layout = affine_map<(d0) -> (d0)>, linalg.inplaceable = true},
+  %t: tensor<?xf32> {bufferization.buffer_layout = affine_map<(d0) -> (d0)>, bufferization.writable = true},
   %sz: index)
     -> (tensor<?xf32>, tensor<?xf32>)
 {
   %f0 = arith.constant 0.0: f32
   // Alloc is needed for the **first** insert_slice (due to backward traversal during analysis).
   //     CHECK: %[[DIM:.*]] = memref.dim %[[FUNC_ARG]]
   // This allocs the whole dim to allow for a full clone of t.
   //     CHECK: %[[ALLOC:.*]] = memref.alloc(%[[DIM]])

 }
 
 // -----
 
 //      CHECK: func @buffer_forwarding_no_conflict(
 // CHECK-SAME:   %[[FUNC_ARG:[0-9a-zA-Z]*]]: memref<?xf32>
 // CHECK-SAME:   %[[sz:[0-9a-zA-Z]*]]: index
 func @buffer_forwarding_no_conflict(
-  %t: tensor<?xf32> {linalg.buffer_layout = affine_map<(d0) -> (d0)>, linalg.inplaceable = true},
+  %t: tensor<?xf32> {bufferization.buffer_layout = affine_map<(d0) -> (d0)>, bufferization.writable = true},
   %sz: index)
     -> (tensor<?xf32>)
 {
   %f0 = arith.constant 0.0: f32
 
   // init_tensor itself does not alloc but forwards to the insert_slice.
   // InitTensorOp replaces the init_tensor with an inplace extract_slice.
   // CHECK: %[[T_SUBVIEW:.*]] =  memref.subview %[[FUNC_ARG]][42] [%[[sz]]] [1]

mlir/test/Dialect/Linalg/comprehensive-module-bufferize.mlir

 // RUN: mlir-opt %s -linalg-comprehensive-module-bufferize=allow-return-allocs -split-input-file | FileCheck %s
 
 // Run fuzzer with different seeds.
 // RUN: mlir-opt %s -linalg-comprehensive-module-bufferize="allow-return-allocs test-analysis-only analysis-fuzzer-seed=23" -split-input-file -o /dev/null
 // RUN: mlir-opt %s -linalg-comprehensive-module-bufferize="allow-return-allocs test-analysis-only analysis-fuzzer-seed=59" -split-input-file -o /dev/null
 // RUN: mlir-opt %s -linalg-comprehensive-module-bufferize="allow-return-allocs test-analysis-only analysis-fuzzer-seed=91" -split-input-file -o /dev/null
 
 // Test bufferization using memref types that have no layout map.
 // RUN: mlir-opt %s -linalg-comprehensive-module-bufferize="allow-return-allocs fully-dynamic-layout-maps=0" -split-input-file | FileCheck %s --check-prefix=CHECK-NO-LAYOUT-MAP
 
-// CHECK-LABEL: func @transfer_read(%{{.*}}: memref<?xf32, #map>) -> vector<4xf32> {
-// CHECK-NO-LAYOUT-MAP-LABEL: func @transfer_read(%{{.*}}: memref<?xf32>) -> vector<4xf32>
-func @transfer_read(
-    %A : tensor<?xf32> {linalg.inplaceable = false})
-  -> (vector<4xf32>)
-{
-  %c0 = arith.constant 0 : index
-  %f0 = arith.constant 0.0 : f32
-
-//       CHECK: %[[RES:.*]] = vector.transfer_read {{.*}} : memref<?xf32, #{{.*}}>, vector<4xf32>
-  %0 = vector.transfer_read %A[%c0], %f0 : tensor<?xf32>, vector<4xf32>
-
-//       CHECK: return %[[RES]] : vector<4xf32>
-  return %0 : vector<4xf32>
-}
-
-// -----
-
 // CHECK-DAG: #[[$map_1d_dyn:.*]] = affine_map<(d0)[s0, s1] -> (d0 * s1 + s0)>
 
 // CHECK-LABEL: func @fill_inplace(
 //  CHECK-SAME:   %[[A:[a-zA-Z0-9]*]]: memref<?xf32, #[[$map_1d_dyn]]>
 // CHECK-NO-LAYOUT-MAP-LABEL: func @fill_inplace(%{{.*}}: memref<?xf32>) {
 func @fill_inplace(
-    %A : tensor<?xf32> {linalg.inplaceable = true})
+    %A : tensor<?xf32> {bufferization.writable = true})
   -> tensor<?xf32>
 {
   //     CHECK: %[[F0:.*]] = arith.constant 0.000000e+00 : f32
   %f0 = arith.constant 0.0 : f32
 
   /// Inplaceable, no alloc
   // CHECK-NOT: alloc
   //     CHECK: linalg.fill ins(%[[F0]] : f32) outs(%[[A]] : memref<?xf32, #[[$map_1d_dyn]]>)
   %r = linalg.fill ins(%f0 : f32) outs(%A : tensor<?xf32>) -> tensor<?xf32>
 
   //     CHECK: return
   // CHECK-NOT: tensor
   return %r: tensor<?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @tensor_extract(%{{.*}}: memref<?xf32, #{{.*}}>) -> f32 {
-func @tensor_extract(%A : tensor<?xf32> {linalg.inplaceable = false}) -> (f32) {
+func @tensor_extract(%A : tensor<?xf32> {bufferization.writable = false}) -> (f32) {
   %c0 = arith.constant 0 : index
 
 //       CHECK: %[[RES:.*]] = memref.load {{.*}} : memref<?xf32, #{{.*}}>
   %0 = tensor.extract %A[%c0] : tensor<?xf32>
 
 //       CHECK: return %[[RES]] : f32
   return %0 : f32
 }
 
 // -----
 
 // CHECK-DAG: #[[$map_1d_dyn:.*]] = affine_map<(d0)[s0, s1] -> (d0 * s1 + s0)>
 
-/// No linalg.inplaceable flag, must allocate.
+/// No bufferization.writable flag, must allocate.
 // CHECK-LABEL: func @not_inplace(
 //  CHECK-SAME:   %[[A:[a-zA-Z0-9]*]]: memref<?xf32, #[[$map_1d_dyn]]>) -> memref<?xf32> {
 // CHECK-NO-LAYOUT-MAP-LABEL: func @not_inplace(%{{.*}}: memref<?xf32>) -> memref<?xf32>
 func @not_inplace(
-    %A : tensor<?xf32> {linalg.inplaceable = false})
+    %A : tensor<?xf32> {bufferization.writable = false})
   -> tensor<?xf32>
 {
   //     CHECK: %[[F0:.*]] = arith.constant 0.000000e+00 : f32
   %f0 = arith.constant 0.0 : f32
 
   //     CHECK: %[[D0:.*]] = memref.dim %[[A]], {{.*}} : memref<?xf32, #[[$map_1d_dyn]]>
   //     CHECK: %[[ALLOC:.*]] = memref.alloc(%[[D0]]) {alignment = 128 : i64} : memref<?xf32>
   //     CHECK: linalg.fill ins(%[[F0]] : f32) outs(%[[ALLOC]] : memref<?xf32>)
   %r = linalg.fill ins(%f0 : f32) outs(%A : tensor<?xf32>) -> tensor<?xf32>
 
   // CHECK-NOT: dealloc
   //     CHECK: return %[[ALLOC]] : memref<?xf32>
   return %r: tensor<?xf32>
 }
 
 // -----
 
 // CHECK-DAG: #[[$map_2d_dyn:.*]] = affine_map<(d0, d1)[s0, s1, s2] -> (d0 * s1 + s0 + d1 * s2)>
 
 // CHECK-LABEL: func @not_inplace
 //  CHECK-SAME:   %[[A:[a-zA-Z0-9]*]]: memref<?x?xf32, #[[$map_2d_dyn]]>) {
 // CHECK-NO-LAYOUT-MAP-LABEL: func @not_inplace(%{{.*}}: memref<?x?xf32>) {
 func @not_inplace(
-    %A : tensor<?x?xf32> {linalg.inplaceable = true})
+    %A : tensor<?x?xf32> {bufferization.writable = true})
   -> tensor<?x?xf32>
 {
   %f0 = arith.constant 0.0 : f32
 
   /// Cross-op multiple uses of %A, the first op which has interfering reads must alloc.
   //       CHECK: %[[ALLOC:.*]] = memref.alloc
   //       CHECK: linalg.fill ins({{.*}}{{.*}}outs(%[[ALLOC]]
   %f = linalg.fill ins(%f0 : f32) outs(%A : tensor<?x?xf32>) -> tensor<?x?xf32>
func @not_inplace(
   //     CHECK: return
   // CHECK-NOT: tensor
   return %r: tensor<?x?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @not_inplace
-func @not_inplace(%A : tensor<?x?xf32> {linalg.inplaceable = true}) -> tensor<?x?xf32> {
+func @not_inplace(%A : tensor<?x?xf32> {bufferization.writable = true}) -> tensor<?x?xf32> {
   /// Within op multiple uses of %A, must alloc.
   // CHECK: alloc
   %r = linalg.matmul  ins(%A, %A: tensor<?x?xf32>, tensor<?x?xf32>)
                      outs(%A: tensor<?x?xf32>)
     -> tensor<?x?xf32>
   // CHECK-NOT: dealloc
   return %r: tensor<?x?xf32>
 }
 // -----
 
 // CHECK-LABEL: func @vec_inplace
-func @vec_inplace(%A : tensor<?xf32> {linalg.inplaceable = true}, %vec : vector<4xf32>)
+func @vec_inplace(%A : tensor<?xf32> {bufferization.writable = true}, %vec : vector<4xf32>)
     -> tensor<?xf32>
 {
   %c0 = arith.constant 0 : index
 
   // CHECK-NOT: alloc
   %r = vector.transfer_write %vec, %A[%c0] : vector<4xf32>, tensor<?xf32>
 
   //     CHECK: return
   // CHECK-NOT: tensor
   return %r: tensor<?xf32>
 }
 
 // -----
 
 // CHECK-DAG: #[[$map_1d_dyn:.*]] = affine_map<(d0)[s0, s1] -> (d0 * s1 + s0)>
 
 // CHECK-LABEL: func @vec_not_inplace
 //  CHECK-SAME:   %[[A:[a-zA-Z0-9]*]]: memref<?xf32, #[[$map_1d_dyn]]>
-func @vec_not_inplace(%A : tensor<?xf32> {linalg.inplaceable = true}, %vec : vector<4xf32>)
+func @vec_not_inplace(%A : tensor<?xf32> {bufferization.writable = true}, %vec : vector<4xf32>)
     -> (tensor<?xf32>, tensor<?xf32>)
 {
   %c0 = arith.constant 0 : index
   %c1 = arith.constant 1 : index
 
   /// Cross-op multiple uses of %A, the first vector.transfer which has interfering reads must alloc.
   //      CHECK: %[[ALLOC:.*]] = memref.alloc
   //      CHECK: memref.copy {{.*}}, %[[ALLOC]]

 
 // CHECK-DAG: #[[$map_1d_dyn:.*]] = affine_map<(d0)[s0, s1] -> (d0 * s1 + s0)>
 
 // CHECK-LABEL: func @insert_slice_fun
 //  CHECK-SAME:   %[[A0:[a-zA-Z0-9]*]]: memref<?xf32, #[[$map_1d_dyn]]>,
 //  CHECK-SAME:   %[[A1:[a-zA-Z0-9]*]]: memref<?xf32, #[[$map_1d_dyn]]>,
 //  CHECK-SAME:   %[[t0:[a-zA-Z0-9]*]]: memref<4xf32, #[[$map_1d_dyn]]>,
 //  CHECK-SAME:   %[[t1:[a-zA-Z0-9]*]]: memref<4xf32, #[[$map_1d_dyn]]>
-func @insert_slice_fun(%A0 : tensor<?xf32> {linalg.inplaceable = false},
-                       %A1 : tensor<?xf32> {linalg.inplaceable = true},
-                       %t0 : tensor<4xf32> {linalg.inplaceable = false},
-                       %t1 : tensor<4xf32> {linalg.inplaceable = true})
+func @insert_slice_fun(%A0 : tensor<?xf32> {bufferization.writable = false},
+                       %A1 : tensor<?xf32> {bufferization.writable = true},
+                       %t0 : tensor<4xf32> {bufferization.writable = false},
+                       %t1 : tensor<4xf32> {bufferization.writable = true})
   ->  (tensor<?xf32>, tensor<?xf32>, tensor<?xf32>, tensor<?xf32>)
 {
   // Hoisted allocs.
   //      CHECK: %[[REALLOC1:.*]] = memref.alloc
   //      CHECK: %[[REALLOC2:.*]] = memref.alloc
   //      CHECK: %[[REALLOC3:.*]] = memref.alloc
 
   // Alloc and copy the whole result tensor. Copy the tensor.extract_slice.

 // -----
 
 // CHECK-DAG: #[[$map_1d_dyn:.*]] = affine_map<(d0)[s0, s1] -> (d0 * s1 + s0)>
 
 // CHECK-LABEL: func @insert_slice_fun
 //  CHECK-SAME:   %[[A:[a-zA-Z0-9]*]]: memref<?xf32, #[[$map_1d_dyn]]>
 //  CHECK-SAME:   %[[t:[a-zA-Z0-9]*]]: memref<4xf32, #[[$map_1d_dyn]]>
 func @insert_slice_fun(
-    %A : tensor<?xf32> {linalg.inplaceable = true},
-    %t : tensor<4xf32> {linalg.inplaceable = false})
+    %A : tensor<?xf32> {bufferization.writable = true},
+    %t : tensor<4xf32> {bufferization.writable = false})
   -> tensor<?xf32>
 {
   %f0 = arith.constant 0.0 : f32
 
   //  CHECK-NOT: alloc
   //      CHECK: %[[SV_A:.*]] = memref.subview %[[A]]
   //      CHECK: memref.copy %[[t]], %[[SV_A]]
   %r0 = tensor.insert_slice %t into %A[0][4][1] : tensor<4xf32> into tensor<?xf32>

 // -----
 
 // CHECK-DAG: #[[$map_1d_dyn:.*]] = affine_map<(d0)[s0, s1] -> (d0 * s1 + s0)>
 
 // CHECK-LABEL: func @insert_slice_fun
 //  CHECK-SAME:   %[[A:[a-zA-Z0-9]*]]: memref<?xf32, #[[$map_1d_dyn]]>
 //  CHECK-SAME:   %[[t:[a-zA-Z0-9]*]]: memref<4xf32, #[[$map_1d_dyn]]>
 func @insert_slice_fun(
-    %A : tensor<?xf32> {linalg.inplaceable = true},
-    %t : tensor<4xf32> {linalg.inplaceable = false})
+    %A : tensor<?xf32> {bufferization.writable = true},
+    %t : tensor<4xf32> {bufferization.writable = false})
   -> tensor<?xf32>
 {
   %f0 = arith.constant 0.0 : f32
 
   //      CHECK: linalg.fill ins({{.*}}{{.*}}outs(%[[A]]
   %r0 = linalg.fill ins(%f0 : f32) outs(%A : tensor<?xf32>) -> tensor<?xf32>
 
   //  CHECK-NOT: alloc

 // -----
 
 // CHECK-DAG: #[[$map_1d_dyn:.*]] = affine_map<(d0)[s0, s1] -> (d0 * s1 + s0)>
 
 // CHECK-LABEL: func @insert_slice_fun_not_inplace
 //  CHECK-SAME:   %[[A:[a-zA-Z0-9]*]]: memref<?xf32, #[[$map_1d_dyn]]>
 //  CHECK-SAME:   %[[t:[a-zA-Z0-9]*]]: memref<4xf32, #[[$map_1d_dyn]]>
 func @insert_slice_fun_not_inplace(
-    %A : tensor<?xf32> {linalg.inplaceable = false},
-    %t : tensor<4xf32> {linalg.inplaceable = false})
+    %A : tensor<?xf32> {bufferization.writable = false},
+    %t : tensor<4xf32> {bufferization.writable = false})
   -> tensor<?xf32>
 {
   //      CHECK: %[[ALLOC:.*]] = memref.alloc(%{{.*}}) {alignment = 128 : i64} : memref<?xf32>
   //      CHECK: memref.copy %[[A]], %[[ALLOC]] : memref<?xf32{{.*}} to memref<?xf32>
   //      CHECK: %[[SV:.*]] = memref.subview %[[ALLOC]][0] [4] [1] : memref<?xf32> to memref<4xf32>
   //      CHECK: memref.copy %[[t]], %[[SV]] : memref<4xf32, #map> to memref<4xf32>
   %r0 = tensor.insert_slice %t into %A[0][4][1] : tensor<4xf32> into tensor<?xf32>
 
   //     CHECK: return %{{.*}} : memref<?xf32>
   return %r0: tensor<?xf32>
 }
 
 //===----------------------------------------------------------------------===//
 // Simple loop cases
 //===----------------------------------------------------------------------===//
 
 // -----
 
 // CHECK-DAG: #[[$map_1d_dyn:.*]] = affine_map<(d0)[s0, s1] -> (d0 * s1 + s0)>
 
 // CHECK-LABEL: func @scf_for_yield_only
 //  CHECK-SAME:   %[[A:[a-zA-Z0-9]*]]: memref<?xf32, #[[$map_1d_dyn]]>
 //  CHECK-SAME:   %[[t:[a-zA-Z0-9]*]]: memref<?xf32, #[[$map_1d_dyn]]>
-func @scf_for_yield_only(%A : tensor<?xf32> {linalg.inplaceable = false},
-                         %B : tensor<?xf32> {linalg.inplaceable = true},
+func @scf_for_yield_only(%A : tensor<?xf32> {bufferization.writable = false},
+                         %B : tensor<?xf32> {bufferization.writable = true},
                          %lb : index, %ub : index, %step : index)
   -> (tensor<?xf32>, tensor<?xf32>)
 {
   //     CHECK:   %[[ALLOC_FOR_A:.*]] = memref.alloc
   //     CHECK:   %[[CASTED:.*]] = memref.cast %[[ALLOC_FOR_A]]
   //     CHECK:   memref.copy %[[A]], %[[ALLOC_FOR_A]]
 
   // The first scf.for remains but just turns into dead code.

 
 // -----
 
 // Ensure that the function bufferizes without error. This tests pre-order
 // traversal of scf.for loops during bufferization. No need to check the IR,
 // just want to make sure that it does not crash.
 
 // CHECK-LABEL: func @nested_scf_for
-func @nested_scf_for(%A : tensor<?xf32> {linalg.inplaceable = true},
+func @nested_scf_for(%A : tensor<?xf32> {bufferization.writable = true},
                      %v : vector<5xf32>) -> tensor<?xf32> {
   %c0 = arith.constant 0 : index
   %c1 = arith.constant 1 : index
   %c10 = arith.constant 10 : index
   %r1 = scf.for %i = %c0 to %c10 step %c1 iter_args(%B = %A) -> tensor<?xf32> {
     %r2 = scf.for %j = %c0 to %c10 step %c1 iter_args(%C = %B) -> tensor<?xf32> {
       %w = vector.transfer_write %v, %C[%c0] : vector<5xf32>, tensor<?xf32>
       scf.yield %w : tensor<?xf32>
     }
     scf.yield %r2 : tensor<?xf32>
   }
   return %r1 : tensor<?xf32>
 }
 
 // -----
 
 // CHECK-DAG: #[[$map_1d_dyn:.*]] = affine_map<(d0)[s0, s1] -> (d0 * s1 + s0)>
 
 // CHECK-LABEL: func @scf_for_with_tensor.insert_slice
 //  CHECK-SAME:   %[[A:[a-zA-Z0-9]*]]: memref<?xf32, #[[$map_1d_dyn]]>
 //  CHECK-SAME:   %[[B:[a-zA-Z0-9]*]]: memref<?xf32, #[[$map_1d_dyn]]>
 //  CHECK-SAME:   %[[C:[a-zA-Z0-9]*]]: memref<4xf32, #[[$map_1d_dyn]]>
 func @scf_for_with_tensor.insert_slice(
-   %A : tensor<?xf32> {linalg.inplaceable = false},
-   %B : tensor<?xf32> {linalg.inplaceable = true},
-   %C : tensor<4xf32> {linalg.inplaceable = false},
+   %A : tensor<?xf32> {bufferization.writable = false},
+   %B : tensor<?xf32> {bufferization.writable = true},
+   %C : tensor<4xf32> {bufferization.writable = false},
    %lb : index, %ub : index, %step : index)
   -> (tensor<?xf32>, tensor<?xf32>)
 {
   //     CHECK:   %[[ALLOC_FOR_A:.*]] = memref.alloc
   //     CHECK:   %[[CASTED:.*]] = memref.cast %[[ALLOC_FOR_A]]
   //     CHECK:   memref.copy %[[A]], %[[ALLOC_FOR_A]]
 
   //     CHECK: %[[svA:.*]] = memref.subview %[[ALLOC_FOR_A]][0] [4] [1]

 
 //===----------------------------------------------------------------------===//
 // Cross function boundary cases.
 //===----------------------------------------------------------------------===//
 
 
 // CHECK-LABEL: func @execute_region_with_conflict(
 //  CHECK-SAME:     %[[m1:.*]]: memref<?xf32
-func @execute_region_with_conflict(%t1 : tensor<?xf32> {linalg.inplaceable = "true"})
+func @execute_region_with_conflict(%t1 : tensor<?xf32> {bufferization.writable = "true"})
     -> (f32, tensor<?xf32>, f32)
 {
   %f1 = arith.constant 0.0 : f32
   %idx = arith.constant 7 : index
 
   // scf.execute_region is canonicalized away after bufferization. So just the
   // memref.store is left over.
 

 
 // -----
 
 //      CHECK: func @matmul(
 // CHECK-SAME:   %[[A:[0-9a-zA-Z]*]]: memref<128x256xf32>
 // CHECK-SAME:   %[[B:[0-9a-zA-Z]*]]: memref<256x192xf32>
 // CHECK-SAME:   %[[C:[0-9a-zA-Z]*]]: memref<128x192xf32>
 func @matmul(
-    %A: tensor<128x256xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %B: tensor<256x192xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = false},
-    %C: tensor<128x192xf32> {linalg.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, linalg.inplaceable = true})
+    %A: tensor<128x256xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, bufferization.writable = false},
+    %B: tensor<256x192xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, bufferization.writable = false},
+    %C: tensor<128x192xf32> {bufferization.buffer_layout = affine_map<(d0, d1) -> (d0, d1)>, bufferization.writable = true})
       -> tensor<128x192xf32> {
   %c0 = arith.constant 0 : index
   %c256 = arith.constant 256 : index
   %c32 = arith.constant 32 : index
   %cst = arith.constant 0.000000e+00 : f32
   %c128 = arith.constant 128 : index
   %c192 = arith.constant 192 : index
   %c8 = arith.constant 8 : index

 
 // CHECK-LABEL: func @tensor_cast_not_in_place(
 //  CHECK-SAME:     %[[A:.*]]: memref<?xf32{{.*}}>, %[[B:.*]]: memref<?xf32{{.*}}>
 //       CHECK:   %[[alloc:.*]] = memref.alloc
 //       CHECK:   memref.copy %[[A]], %[[alloc]]
 //       CHECK:   %[[subview:.*]] = memref.subview %[[A]][{{.*}}] [4] [1] : {{.*}} to memref<4xf32
 //       CHECK:   memref.copy %[[alloc]], %[[subview]]
 func @tensor_cast_not_in_place(
-    %A : tensor<?xf32> {linalg.inplaceable = true},
-    %B : tensor<?xf32> {linalg.inplaceable = false}, %idx: index)
+    %A : tensor<?xf32> {bufferization.writable = true},
+    %B : tensor<?xf32> {bufferization.writable = false}, %idx: index)
   -> (tensor<?xf32>)
 {
   %r0 = tensor.cast %A : tensor<?xf32> to tensor<4xf32>
   %r1 = tensor.insert_slice %r0 into %A[%idx][4][1] : tensor<4xf32> into tensor<?xf32>
   return %r1 : tensor<?xf32>
 }
 
 // -----
 
 //===----------------------------------------------------------------------===//
 // Insertion point cases.
 //===----------------------------------------------------------------------===//
 
 /// These tests just check the produced IR is valid and does not have dominance
 /// errors in the def-use chains.
 
 // CHECK-LABEL: func @dominance_violation_bug_1
 func @dominance_violation_bug_1(
-    %A : tensor<?x?xf32> {linalg.inplaceable = false},
+    %A : tensor<?x?xf32> {bufferization.writable = false},
     %idx : index)
   -> tensor<?x?xf32>
 {
   %f0 = arith.constant 0.0 : f32
 
   %sA = tensor.extract_slice %A[0, 0][%idx, %idx][1, 1] : tensor<?x?xf32> to tensor<?x?xf32>
   %ssA = tensor.extract_slice %sA[0, 0][4, 4][1, 1] : tensor<?x?xf32> to tensor<4x4xf32>
   %FA = linalg.fill ins(%f0 : f32) outs(%ssA : tensor<4x4xf32>) -> tensor<4x4xf32>
   %rsA = tensor.insert_slice %FA into %sA[0, 0][4, 4][1, 1] : tensor<4x4xf32> into tensor<?x?xf32>
   %rA = tensor.insert_slice %rsA into %A[0, 0][%idx, %idx][1, 1] : tensor<?x?xf32> into tensor<?x?xf32>
 
   return %rA : tensor<?x?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @scf_if_inplace(
 //  CHECK-SAME:     %[[cond:.*]]: i1, %[[t1:.*]]: memref<?xf32{{.*}}>, %[[v:.*]]: vector
 func @scf_if_inplace(%cond: i1,
-                     %t1: tensor<?xf32> {linalg.inplaceable = true},
+                     %t1: tensor<?xf32> {bufferization.writable = true},
                      %v: vector<5xf32>, %idx: index) -> tensor<?xf32> {
 
   //      CHECK: scf.if %[[cond]] {
   // CHECK-NEXT: } else {
   // CHECK-NEXT:   vector.transfer_write %[[v]], %[[t1]]
   // CHECK-NEXT: }
   // CHECK-NEXT: return
   %r = scf.if %cond -> (tensor<?xf32>) {

 //   CHECK-DAG:   %[[c1:.*]] = arith.constant 1 : index
 //   CHECK-DAG:   %[[c10:.*]] = arith.constant 10 : index
 //       CHECK:   scf.for %{{.*}} = %[[c0]] to %[[c10]] step %[[c1]] {
 //       CHECK:     scf.if %{{.*}} {
 //       CHECK:     } else {
 //       CHECK:       vector.transfer_write
 //       CHECK:     }
 //       CHECK:   }
-func @scf_if_inside_scf_for(%t1: tensor<?xf32> {linalg.inplaceable = true},
+func @scf_if_inside_scf_for(%t1: tensor<?xf32> {bufferization.writable = true},
                       %v: vector<5xf32>, %idx: index,
                       %cond: i1) -> tensor<?xf32> {
   %c0 = arith.constant 0 : index
   %c1 = arith.constant 1 : index
   %c10 = arith.constant 10 : index
   %r = scf.for %iv = %c0 to %c10 step %c1 iter_args(%bb = %t1) -> (tensor<?xf32>) {
     %r2 = scf.if %cond -> (tensor<?xf32>) {
       scf.yield %bb : tensor<?xf32>
     } else {
       %t2 = vector.transfer_write %v, %bb[%idx] : vector<5xf32>, tensor<?xf32>
       scf.yield %t2 : tensor<?xf32>
     }
     scf.yield %r2 : tensor<?xf32>
   }
   return %r : tensor<?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @scf_if_non_equiv_yields(
 //  CHECK-SAME:     %[[cond:.*]]: i1, %[[A:.*]]: memref<{{.*}}>, %[[B:.*]]: memref<{{.*}}>) -> memref<{{.*}}>
 func @scf_if_non_equiv_yields(
     %b : i1,
-    %A : tensor<4xf32> {linalg.inplaceable = false},
-    %B : tensor<4xf32> {linalg.inplaceable = false})
+    %A : tensor<4xf32> {bufferization.writable = false},
+    %B : tensor<4xf32> {bufferization.writable = false})
   -> tensor<4xf32>
 {
   // CHECK: %[[r:.*]] = arith.select %[[cond]], %[[A]], %[[B]]
   %r = scf.if %b -> (tensor<4xf32>) {
     scf.yield %A : tensor<4xf32>
   } else {
     scf.yield %B : tensor<4xf32>
   }
   // CHECK: return %[[r]]
   return %r: tensor<4xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @insert_op
 //  CHECK-SAME:     %[[t1:.*]]: memref<?xf32, {{.*}}>, %[[s:.*]]: f32, %[[i:.*]]: index
-func @insert_op(%t1 : tensor<?xf32> {linalg.inplaceable = true},
+func @insert_op(%t1 : tensor<?xf32> {bufferization.writable = true},
                 %s : f32, %i : index) -> tensor<?xf32> {
   // CHECK: memref.store %[[s]], %[[t1]][%[[i]]]
   %0 = tensor.insert %s into %t1[%i] : tensor<?xf32>
   // CHECK: return
   return %0 : tensor<?xf32>
 }
 
 // -----
 
 func @gather_like(
-    %arg0 : tensor<?x?xf32> {linalg.inplaceable = false},
-    %arg1 : tensor<?xi32> {linalg.inplaceable = false},
-    %arg2 : tensor<?x?xf32> {linalg.inplaceable = true}) -> tensor<?x?xf32> {
+    %arg0 : tensor<?x?xf32> {bufferization.writable = false},
+    %arg1 : tensor<?xi32> {bufferization.writable = false},
+    %arg2 : tensor<?x?xf32> {bufferization.writable = true}) -> tensor<?x?xf32> {
   %0 = linalg.generic {
       indexing_maps = [affine_map<(d0, d1) -> (d0)>,
                        affine_map<(d0, d1) -> (d0, d1)>],
       iterator_types = ["parallel", "parallel"]}
       ins(%arg1 : tensor<?xi32>) outs(%arg2 : tensor<?x?xf32>) {
       ^bb0(%arg3: i32, %arg4 : f32):
         %iv1 = linalg.index 1 : index
         %1 = arith.index_cast %arg3: i32 to index

 //       CHECK:     %[[YIELD:.+]] = memref.load %[[ARG0]]
 //       CHECK:     linalg.yield %[[YIELD]]
 
 // -----
 
 // CHECK-LABEL: func @linalg_op_bufferizes_inplace_with_input
 //  CHECK-SAME:     %[[t1:.*]]: memref<?x?xf32, #{{.*}}>, %[[t2:.*]]: memref<?xf32, #{{.*}}>, %[[t3:.*]]: memref<?x?xf32, #{{.*}}>
 func @linalg_op_bufferizes_inplace_with_input(
-    %t1: tensor<?x?xf32> {linalg.inplaceable = true},
-    %t2: tensor<?xf32> {linalg.inplaceable = true},
-    %t3: tensor<?x?xf32> {linalg.inplaceable = true},
+    %t1: tensor<?x?xf32> {bufferization.writable = true},
+    %t2: tensor<?xf32> {bufferization.writable = true},
+    %t3: tensor<?x?xf32> {bufferization.writable = true},
     %s1: index, %s2: index, %cst: f32) -> tensor<?x?xf32> {
   // CHECK: linalg.generic {{.*}} ins(%[[t1]], %[[t2]] : {{.*}}) outs(%[[t3]] : {{.*}})
   %r = linalg.generic {
     indexing_maps = [affine_map<(d0, d1) -> (d0, d1)>,
                      affine_map<(d0, d1) -> (d1)>,
                      affine_map<(d0, d1)-> (d0, d1)>],
     iterator_types = ["parallel", "parallel"]}
     ins(%t1, %t2 : tensor<?x?xf32>, tensor<?xf32>)

 #trait = {
   indexing_maps = #accesses,
   iterator_types = ["parallel"]
 }
 
 // CHECK-LABEL: func @op_is_reading_but_following_ops_are_not
 //  CHECK-SAME:     %[[t0:.*]]: memref<?xf32
 func @op_is_reading_but_following_ops_are_not(
-    %t0 : tensor<?xf32> {linalg.inplaceable = false},
+    %t0 : tensor<?xf32> {bufferization.writable = false},
     %cst : f32)
   -> tensor<?xf32>
 {
   // Make sure that a copy is inserted here.
   // CHECK: %[[ALLOC:.*]] = memref.alloc
   // CHECK: memref.copy %[[t0]], %[[ALLOC]]
   // CHECK: linalg.generic {{.*}} outs(%[[ALLOC]] : memref
   %r0 =linalg.generic #trait outs (%t0 : tensor<?xf32>) {
func @depthwise_conv_1d_nwc_wc(%arg0: index, %arg1: index, %arg2: tensor<8x18x32xf32>)
   return %3 : tensor<?x1x6x8xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @write_to_select_op_source
 //  CHECK-SAME:     %[[t1:.*]]: memref<?xf32, #{{.*}}>, %[[t2:.*]]: memref<?xf32, #{{.*}}>
 func @write_to_select_op_source(
-    %t1 : tensor<?xf32> {linalg.inplaceable = true},
-    %t2 : tensor<?xf32> {linalg.inplaceable = true},
+    %t1 : tensor<?xf32> {bufferization.writable = true},
+    %t2 : tensor<?xf32> {bufferization.writable = true},
     %c : i1)
   -> (tensor<?xf32>, tensor<?xf32>)
 {
   %cst = arith.constant 0.0 : f32
   %idx = arith.constant 0 : index
   // CHECK: %[[alloc:.*]] = memref.alloc
   // CHECK: memref.copy %[[t1]], %[[alloc]]
   // CHECK: memref.store %{{.*}}, %[[alloc]]
   %w = tensor.insert %cst into %t1[%idx] : tensor<?xf32>
   // CHECK: %[[select:.*]] = arith.select %{{.*}}, %[[t1]], %[[t2]]
   %s = arith.select %c, %t1, %t2 : tensor<?xf32>
   // CHECK: return %[[select]], %[[alloc]]
   return %s, %w : tensor<?xf32>, tensor<?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @write_after_select_read_one
 //  CHECK-SAME:     %[[t1:.*]]: memref<?xf32, #{{.*}}>, %[[t2:.*]]: memref<?xf32, #{{.*}}>
 func @write_after_select_read_one(
-    %t1 : tensor<?xf32> {linalg.inplaceable = true},
-    %t2 : tensor<?xf32> {linalg.inplaceable = true},
+    %t1 : tensor<?xf32> {bufferization.writable = true},
+    %t2 : tensor<?xf32> {bufferization.writable = true},
     %c : i1)
   -> (f32, tensor<?xf32>)
 {
   %cst = arith.constant 0.0 : f32
   %idx = arith.constant 0 : index
 
   // CHECK: %[[alloc:.*]] = memref.alloc
   // CHECK: %[[casted:.*]] = memref.cast %[[alloc]]

 
 // TODO: The scf.yield could bufferize to 1 alloc and 2 copies (instead of
 // 2 allocs and 2 copies).
 
 // CHECK-LABEL: func @scf_for_swapping_yields(
 //  CHECK-SAME:     %[[A:.*]]: memref<?xf32, #{{.*}}>, %[[B:.*]]: memref<?xf32, #{{.*}}>
 
 func @scf_for_swapping_yields(
-    %A : tensor<?xf32>, %B : tensor<?xf32> {linalg.inplaceable = true},
+    %A : tensor<?xf32>, %B : tensor<?xf32> {bufferization.writable = true},
     %C : tensor<4xf32>, %lb : index, %ub : index, %step : index)
   -> (f32, f32)
 {
 //   CHECK-DAG:   %[[clone1:.*]] = bufferization.clone %[[A]]
 //   CHECK-DAG:   %[[clone2:.*]] = bufferization.clone %[[B]]
 //       CHECK:   %[[for:.*]]:2 = scf.for {{.*}} iter_args(%[[iter1:.*]] = %[[clone1]], %[[iter2:.*]] = %[[clone2]])
   %r0:2 = scf.for %i = %lb to %ub step %step iter_args(%tA = %A, %tB = %B)
       -> (tensor<?xf32>, tensor<?xf32>)

utils/bazel/llvm-project-overlay/mlir/BUILD.bazel

deps = [
         ":FuncDialect",
         ":FuncTransforms",
         ":IR",
         ":LinalgOps",
         ":LinalgPassIncGen",
         ":LinalgStructuredOpsIncGen",
         ":MathDialect",
         ":MemRefDialect",
-        ":ModuleBufferization",
         ":Pass",
         ":SCFDialect",
         ":SCFTransforms",
         ":SCFUtils",
         ":SparseTensor",
         ":Support",
         ":TensorDialect",
         ":TensorTilingInterfaceImpl",
         ":TensorTransforms",
         ":TensorUtils",
         ":TransformUtils",
         ":Transforms",
         ":VectorOps",
         ":VectorToSCF",
         ":VectorTransforms",
         ":VectorUtils",
         ":X86VectorTransforms",
         "//llvm:Support",
     ],
 )
 
 cc_library(
-    name = "ModuleBufferization",
-    srcs = [
-        "lib/Dialect/Linalg/ComprehensiveBufferize/ModuleBufferization.cpp",
-    ],
-    hdrs = [
-        "include/mlir/Dialect/Linalg/ComprehensiveBufferize/ModuleBufferization.h",
-    ],
-    includes = ["include"],
-    deps = [
-        ":BufferizationDialect",
-        ":BufferizationTransforms",
-        ":FuncDialect",
-        ":IR",
-        ":MemRefDialect",
-        "//llvm:Support",
-    ],
-)
-
-cc_library(
     name = "TilingInterface",
     srcs = ["lib/Interfaces/TilingInterface.cpp"],
     hdrs = ["include/mlir/Interfaces/TilingInterface.h"],
     includes = ["include"],
     deps = [
         ":IR",
         ":Support",
         ":TilingInterfaceIncGen",