[mlir][tosa] Replace StructAttrs with AttrDefs

[Mogball]

Summary

Depends on D127352

[rriddle]

mlir/include/mlir/Dialect/Tosa/IR/TosaOpBase.td

67

Please add a base class for Tosa attributes that passes in the dialect (and has a template param for the mnemonic). This section mentions this as well (+ gives an example): https://mlir.llvm.org/docs/AttributesAndTypes/#adding-a-new-attribute-or-type-definition

Same applies to some other commits in this chain as well.

88

Summary should be at the top of the def, same throughout.

[Mogball]

I fixed this in all the commits

[Mogball]

add dialect attr

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

 add_mlir_dialect(TosaOps tosa)
 add_mlir_doc(TosaOps TosaOps Dialects/ -gen-op-doc)
 add_mlir_interface(TosaInterfaces)
 
 set(LLVM_TARGET_DEFINITIONS TosaOps.td)
-mlir_tablegen(TosaStructs.h.inc -gen-struct-attr-decls)
-mlir_tablegen(TosaStructs.cpp.inc -gen-struct-attr-defs)
-add_public_tablegen_target(MLIRTosaStructsIncGen)
+mlir_tablegen(TosaAttributes.h.inc -gen-attrdef-decls)
+mlir_tablegen(TosaAttributes.cpp.inc -gen-attrdef-defs)
+add_public_tablegen_target(MLIRTosaAttributesIncGen)
 

mlir/include/mlir/Dialect/Tosa/IR/TosaOpBase.td

 //===-- TosaOpBase.td - TOSA dialect op builders -----------*- tablegen -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 // This file defines the common definitions for the TOSA dialect.
 //
 //===----------------------------------------------------------------------===//
 
-
 #ifndef TOSA_OP_BASE
 #define TOSA_OP_BASE
 
+include "mlir/IR/AttrTypeBase.td"
+include "mlir/IR/OpBase.td"
+
 //===----------------------------------------------------------------------===//
 // The TOSA Dialect.
 //===----------------------------------------------------------------------===//
+
 def Tosa_Dialect : Dialect {
   let name = "tosa";
 
   let description = [{
     The Tensor Operator Set Architecture (TOSA) dialect.
 
     This dialect implements the TOSA standard described at
     https://developer.mlplatform.org/w/tosa/ .
 
     Tensor Operator Set Architecture (TOSA) provides a set of whole-tensor
     operations commonly employed by Deep Neural Networks. The intent is to
     enable a variety of implementations running on a diverse range of
     processors, with the results at the TOSA level consistent across those
     implementations. Applications or frameworks which target TOSA can therefore
     be deployed on a wide range of different processors, such as CPUs or GPUs,
     with defined accuracy and compatibility constraints. Most operators from the
     common ML frameworks should be expressible in TOSA. It is expected that
     there will be tools to lower from the ML frameworks into TOSA.
   }];
 
   let dependentDialects = ["tensor::TensorDialect"];
 
   let cppNamespace = "mlir::tosa";
   let hasConstantMaterializer = 1;
+  let useDefaultAttributePrinterParser = 1;
 }
 
 //===----------------------------------------------------------------------===//
 // TOSA Operator Quantization Attributes.
 //===----------------------------------------------------------------------===//
 
 // Quantization attributes used across TOSA operators. Quantization attributes
 // feed numerical precision parameters to the functional implementation of TOSA
 // operators.
 // The functional behavior is defined in the TOSA specification maintained at
 // https://developer.mlplatform.org/w/tosa/ . TOSA leverages MLIR's built in
 // quantization support: https://mlir.llvm.org/docs/Quantization/, and supports
 // uniform quantization. Depending on datatype, asymmetric and symmetric
 // quantization are supported. The types themselves are described in
 // TosaTypesBase.td .
 
 // This quantization attribute expresses numerical behavior of operators where
 // the operator has a numerical relationship between a single input and output.
 // For example: tosa.negate.
-def Tosa_UnaryOpQuantizationAttr : StructAttr<"UnaryOpQuantizationAttr",
-  Tosa_Dialect, [
-    StructFieldAttr<"input_zp",         I32Attr>,
-    StructFieldAttr<"output_zp",        I32Attr>
-    ]> {
+def Tosa_UnaryOpQuantizationAttr : AttrDef<Tosa_Dialect, 
+                                           "UnaryOpQuantization"> {
+  let parameters = (ins "int64_t":$input_zp, "int64_t":$output_zp);
   let summary = "Attribute for UnaryOp quantization information.";
+  let mnemonic = "unary_quant";
+  let assemblyFormat = "`<` struct(params) `>`";
 }
 
 // There is no explicit BinaryOpQuantizationAttr for 2-input/1-output ops. In
 // this case, a tosa.rescale is used to express the inputs to the same scale.
 // TODO: Upload WIP legalization document describing this construction by
 // example.
 
 // This quantization attribute holds input and weight zero point. Both the
 // ConvOp and MatMulOp QuantizationAttrs follow a common design semantic where
 // their ownquantization attribute only expresses the numerical behavior at
 // the inputs.
 // The scaling of their accumulator output is done using an explicit
 // tosa.rescale operator that scales the accumulator result to output scale.
-def Tosa_ConvOpQuantizationAttr : StructAttr<"ConvOpQuantizationAttr",
-  Tosa_Dialect, [
-    StructFieldAttr<"input_zp",         I32Attr>,
-    StructFieldAttr<"weight_zp",        I32Attr>
-    ]> {
+def Tosa_ConvOpQuantizationAttr : AttrDef<Tosa_Dialect, "ConvOpQuantization"> { 
+  let parameters = (ins "int64_t":$input_zp, "int64_t":$weight_zp);
   let summary = "Attribute for Conv type op quantization information.";
+  let mnemonic = "conv_quant";
+  let assemblyFormat = "`<` struct(params) `>`";
 }
 
-def Tosa_MatMulOpQuantizationAttr : StructAttr<"MatMulOpQuantizationAttr",
-  Tosa_Dialect, [
-    StructFieldAttr<"a_zp",             I32Attr>,
-    StructFieldAttr<"b_zp",             I32Attr>
-    ]> {
+def Tosa_MatMulOpQuantizationAttr : AttrDef<Tosa_Dialect, 
+                                            "MatMulOpQuantization"> {
+  let parameters = (ins "int64_t":$a_zp, "int64_t":$b_zp);
   let summary = "Attribute for MatMulOp quantization information.";
+  let mnemonic = "matmul_quant";
+  let assemblyFormat = "`<` struct(params) `>`";
 }
 
 // This attribute holds input zero point correction applied to the padding
 // zeros to ensure numerical accuracy in the subsequent TOSA operations.
 // Its functional application is described in the tosa.pad() operator
 // description in the specification.
-def Tosa_PadOpQuantizationAttr : StructAttr<"PadOpQuantizationAttr",
-  Tosa_Dialect, [
-    StructFieldAttr<"input_zp",         I32Attr>
-    ]> {
+def Tosa_PadOpQuantizationAttr : AttrDef<Tosa_Dialect, "PadOpQuantization"> {
+  let parameters = (ins "int64_t":$input_zp);
   let summary = "Attribute for PadOp quantization information.";
+  let mnemonic = "pad_quant";
+  let assemblyFormat = "`<` struct(params) `>`";
 }
 
 //===----------------------------------------------------------------------===//
 // TOSA Operator Quantization Builders.
 //===----------------------------------------------------------------------===//
 
 // This builder is called on all convolution operators except for TransposeConv,
 // which has specialized output shape semantics. The builder also defines the

mlir/include/mlir/Dialect/Tosa/IR/TosaOps.h

 #include "mlir/Dialect/Traits.h"
 #include "mlir/Interfaces/InferTypeOpInterface.h"
 #include "mlir/Interfaces/LoopLikeInterface.h"
 #include "mlir/Interfaces/SideEffectInterfaces.h"
 
 //===----------------------------------------------------------------------===//
 // TOSA dialect and structs includes.
 //===----------------------------------------------------------------------===//
+
 #include "mlir/Dialect/Tosa/IR/TosaOpsDialect.h.inc"
 #include "mlir/Dialect/Tosa/IR/TosaStructs.h.inc"
 
 namespace mlir {
 class PatternRewriter;
 
 namespace tosa {
 
 #include "mlir/Dialect/Tosa/IR/TosaInterfaces.h.inc"
 
 } // namespace tosa
 } // namespace mlir
 
 //===----------------------------------------------------------------------===//
 // Utility Functions
 //===----------------------------------------------------------------------===//
 namespace mlir {
 namespace tosa {
 /// Appends the canonicalization patterns for all the TOSA ops to the `patterns`
 void populateTosaOpsCanonicalizationPatterns(MLIRContext *ctx,
                                              RewritePatternSet &patterns);
 } // namespace tosa
 } // namespace mlir
 
+#define GET_ATTRDEF_CLASSES
+#include "mlir/Dialect/Tosa/IR/TosaAttributes.h.inc"
+
 #define GET_OP_CLASSES
 #include "mlir/Dialect/Tosa/IR/TosaOps.h.inc"
 
 #endif // MLIR_DIALECT_TOSA_IR_TOSAOPS_H

mlir/lib/Conversion/TosaToLinalg/TosaToLinalg.cpp

auto constant =
         rewriter.create<arith::ConstantOp>(loc, IntegerAttr::get(elementTy, 0));
     return rewriter.create<arith::SubIOp>(loc, resultTypes, constant, args[0]);
   }
 
   if (isa<tosa::NegateOp>(op) && elementTy.isa<IntegerType>() &&
       cast<tosa::NegateOp>(op).quantization_info()) {
     auto quantizationInfo = cast<tosa::NegateOp>(op).quantization_info();
     int32_t inputBitWidth = elementTy.getIntOrFloatBitWidth();
-    int64_t inZp =
-        quantizationInfo.getValue().input_zp().getValue().getSExtValue();
-    int64_t outZp =
-        quantizationInfo.getValue().output_zp().getValue().getSExtValue();
+    int64_t inZp = quantizationInfo.getValue().getInput_zp();
+    int64_t outZp = quantizationInfo.getValue().getOutput_zp();
 
     // Compute the maximum value that can occur in the intermediate buffer.
     int64_t zpAdd = inZp + outZp;
     int64_t maxValue = APInt::getSignedMaxValue(inputBitWidth).getSExtValue() +
                        std::abs(zpAdd) + 1;
 
     // Convert that maximum value into the maximum bitwidth needed to represent
     // it. We assume 48-bit numbers may be supported further in the pipeline.
LogicalResult matchAndRewrite(tosa::PadOp padOp,
 
     Value padConstant;
 
     if (padOp.pad_const()) {
       padConstant = rewriter.createOrFold<tensor::ExtractOp>(
           loc, padOp.pad_const(), ValueRange({}));
     } else {
       Attribute constantAttr;
-      if (elementTy.isa<FloatType>())
+      if (elementTy.isa<FloatType>()) {
         constantAttr = rewriter.getFloatAttr(elementTy, 0.0);
-      else if (elementTy.isa<IntegerType>() && !padOp.quantization_info())
+      } else if (elementTy.isa<IntegerType>() && !padOp.quantization_info()) {
         constantAttr = rewriter.getIntegerAttr(elementTy, 0);
-      else if (elementTy.isa<IntegerType>() && padOp.quantization_info()) {
-        auto value = padOp.quantization_info().getValue().input_zp().getValue();
-        constantAttr = rewriter.getIntegerAttr(elementTy, value.getZExtValue());
+      } else if (elementTy.isa<IntegerType>() && padOp.quantization_info()) {
+        int64_t value = padOp.quantization_info().getValue().getInput_zp();
+        constantAttr = rewriter.getIntegerAttr(elementTy, value);
       }
       if (constantAttr)
         padConstant = rewriter.create<arith::ConstantOp>(loc, constantAttr);
     }
 
     if (!padConstant) {
       return rewriter.notifyMatchFailure(
           padOp, "tosa.pad was unable to determine the pad constant value.");

mlir/lib/Conversion/TosaToLinalg/TosaToLinalgNamed.cpp

matchAndRewrite(tosa::Conv2DOp op, OpAdaptor adaptor,
 
     auto weightShape = weightTy.getShape();
 
     // Apply padding as necessary.
     Attribute zeroAttr = rewriter.getZeroAttr(inputETy);
     if (isQuantized) {
       auto quantizationInfo =
           op->getAttr("quantization_info").cast<tosa::ConvOpQuantizationAttr>();
-      auto iZp = quantizationInfo.input_zp().getValue().getSExtValue();
+      int64_t iZp = quantizationInfo.getInput_zp();
 
       int64_t intMin =
           APInt::getSignedMinValue(inputETy.getIntOrFloatBitWidth())
               .getSExtValue();
       int64_t intMax =
           APInt::getSignedMaxValue(inputETy.getIntOrFloatBitWidth())
               .getSExtValue();
 
matchAndRewrite(tosa::Conv2DOp op, OpAdaptor adaptor,
     indexingMaps.push_back(rewriter.getMultiDimIdentityMap(resultTy.getRank()));
 
     Value biasInitTensor = rewriter.create<linalg::InitTensorOp>(
         loc, filteredDims, resultTy.getShape(), resultETy);
 
     if (isQuantized) {
       auto quantizationInfo =
           op->getAttr("quantization_info").cast<tosa::ConvOpQuantizationAttr>();
-      auto iZp = rewriter.getI32IntegerAttr(
-          quantizationInfo.input_zp().getValue().getSExtValue());
-      auto kZp = rewriter.getI32IntegerAttr(
-          quantizationInfo.weight_zp().getValue().getSExtValue());
+      auto iZp = rewriter.getI32IntegerAttr(quantizationInfo.getInput_zp());
+      auto kZp = rewriter.getI32IntegerAttr(quantizationInfo.getWeight_zp());
 
       auto iZpVal = rewriter.create<arith::ConstantOp>(loc, iZp);
       auto kZpVal = rewriter.create<arith::ConstantOp>(loc, kZp);
       Value conv =
           rewriter
               .create<linalg::Conv2DNhwcHwcfQOp>(
                   loc, resultTy, ValueRange{input, weight, iZpVal, kZpVal},
                   ValueRange{zeroTensor}, strideAttr, dilationAttr)
SmallVector<Value> filteredDims = inferDynamicDimsForConv(
         0, 1, rewriter);
 
     bool isQuantized = op->hasAttr("quantization_info");
     IntegerAttr iZp;
     IntegerAttr kZp;
     if (isQuantized) {
       auto quantizationInfo =
           op->getAttr("quantization_info").cast<tosa::ConvOpQuantizationAttr>();
-      iZp = rewriter.getI32IntegerAttr(
-          quantizationInfo.input_zp().getValue().getSExtValue());
-      kZp = rewriter.getI32IntegerAttr(
-          quantizationInfo.weight_zp().getValue().getSExtValue());
+      iZp = rewriter.getI32IntegerAttr(quantizationInfo.getInput_zp());
+      kZp = rewriter.getI32IntegerAttr(quantizationInfo.getWeight_zp());
     }
 
     auto weightShape = weightTy.getShape();
     auto resultShape = resultTy.getShape();
 
     // Apply padding as necessary.
     Attribute zeroAttr = rewriter.getZeroAttr(inputETy);
     if (isQuantized) {
       auto quantizationInfo =
           op->getAttr("quantization_info").cast<tosa::ConvOpQuantizationAttr>();
-      auto iZp = quantizationInfo.input_zp().getValue().getSExtValue();
+      int64_t iZp = quantizationInfo.getInput_zp();
 
       int64_t intMin =
           APInt::getSignedMinValue(inputETy.getIntOrFloatBitWidth())
               .getSExtValue();
       int64_t intMax =
           APInt::getSignedMaxValue(inputETy.getIntOrFloatBitWidth())
               .getSExtValue();
 
if (!op.quantization_info()) {
       rewriter.replaceOpWithNewOp<linalg::BatchMatmulOp>(
           op, TypeRange{op.getType()}, ValueRange{adaptor.a(), adaptor.b()},
           ValueRange{zeroTensor});
       return success();
     }
 
     auto quantizationInfo = op.quantization_info().getValue();
     auto aZp = rewriter.create<arith::ConstantOp>(
-        loc, rewriter.getI32IntegerAttr(
-                 quantizationInfo.a_zp().getValue().getSExtValue()));
+        loc, rewriter.getI32IntegerAttr(quantizationInfo.getA_zp()));
     auto bZp = rewriter.create<arith::ConstantOp>(
-        loc, rewriter.getI32IntegerAttr(
-                 quantizationInfo.b_zp().getValue().getSExtValue()));
+        loc, rewriter.getI32IntegerAttr(quantizationInfo.getB_zp()));
     rewriter.replaceOpWithNewOp<linalg::QuantizedBatchMatmulOp>(
         op, TypeRange{op.getType()},
         ValueRange{adaptor.a(), adaptor.b(), aZp, bZp}, zeroTensor);
 
     return success();
   }
 };
 
if (!op.quantization_info()) {
                   })
               .getResult(0);
       rewriter.replaceOp(op, result);
       return success();
     }
 
     auto quantizationInfo = op.quantization_info().getValue();
     auto inputZp = rewriter.create<arith::ConstantOp>(
-        loc, rewriter.getI32IntegerAttr(
-                 quantizationInfo.input_zp().getValue().getSExtValue()));
+        loc, rewriter.getI32IntegerAttr(quantizationInfo.getInput_zp()));
     auto outputZp = rewriter.create<arith::ConstantOp>(
-        loc, rewriter.getI32IntegerAttr(
-                 quantizationInfo.weight_zp().getValue().getSExtValue()));
+        loc, rewriter.getI32IntegerAttr(quantizationInfo.getWeight_zp()));
     Value matmul =
         rewriter
             .create<linalg::QuantizedMatmulOp>(
                 loc, TypeRange{op.getType()},
                 ValueRange{input, transposedWeight, inputZp, outputZp},
                 zeroTensor)
             ->getResult(0);
     Value result =
auto genericOp = rewriter.create<linalg::GenericOp>(
                           ->getResult(0);
           } else {
 
             // If we have quantization information we need to apply an offset
             // for the input zp value.
             if (op.quantization_info()) {
               auto quantizationInfo = op.quantization_info().getValue();
               auto inputZp = rewriter.create<arith::ConstantOp>(
-                  loc, quantizationInfo.input_zp());
+                  loc,
+                  b.getIntegerAttr(accETy, quantizationInfo.getInput_zp()));
               Value offset =
                   rewriter.create<arith::MulIOp>(loc, accETy, countI, inputZp);
               poolVal =
                   rewriter.create<arith::SubIOp>(loc, accETy, poolVal, offset);
             }
 
             // Compute the multiplier and shift values for the quantization
             // normalization. Preferably we would want to compute more bits
auto genericOp = rewriter.create<linalg::GenericOp>(
                         shiftVal, rewriter.getBoolAttr(false))
                     .getResult();
 
             // If we have quantization information we need to apply output
             // zeropoint.
             if (op.quantization_info()) {
               auto quantizationInfo = op.quantization_info().getValue();
               auto outputZp = rewriter.create<arith::ConstantOp>(
-                  loc, quantizationInfo.output_zp());
+                  loc, b.getIntegerAttr(scaled.getType(),
+                                        quantizationInfo.getOutput_zp()));
               scaled = rewriter.create<arith::AddIOp>(loc, scaled, outputZp)
                            .getResult();
             }
 
             // Apply Clip.
             int64_t outBitwidth = resultETy.getIntOrFloatBitWidth();
 
             auto min = rewriter.create<arith::ConstantIntOp>(

mlir/lib/Dialect/Tosa/CMakeLists.txt

 add_mlir_dialect_library(MLIRTosa
   Utils/ConversionUtils.cpp
   Utils/QuantUtils.cpp
   IR/TosaOps.cpp
 
   ADDITIONAL_HEADER_DIRS
   ${MLIR_MAIN_INCLUDE_DIR}/mlir/Dialect/Tosa
 
   DEPENDS
+  MLIRTosaAttributesIncGen
   MLIRTosaOpsIncGen
-  MLIRTosaStructsIncGen
   MLIRTosaInterfacesIncGen
 
   LINK_LIBS PUBLIC
   MLIRIR
   MLIRDialect
   MLIRCallInterfaces
   MLIRControlFlowInterfaces
   MLIRQuant
   MLIRQuantUtils
   MLIRSideEffectInterfaces
   MLIRTensor
   MLIRViewLikeInterface
   )
 
 add_subdirectory(Transforms)

mlir/lib/Dialect/Tosa/IR/TosaOps.cpp

 //===----------------------------------------------------------------------===//
 
 #include "mlir/Dialect/Tosa/IR/TosaOps.h"
 #include "mlir/Dialect/Quant/QuantOps.h"
 #include "mlir/Dialect/Tensor/IR/Tensor.h"
 #include "mlir/Dialect/Tosa/Utils/QuantUtils.h"
 #include "mlir/Dialect/Tosa/Utils/ShapeUtils.h"
 #include "mlir/IR/BuiltinTypes.h"
+#include "mlir/IR/DialectImplementation.h"
 #include "mlir/IR/Matchers.h"
 #include "mlir/IR/PatternMatch.h"
 #include "mlir/Transforms/FoldUtils.h"
 #include "mlir/Transforms/InliningUtils.h"
 #include "mlir/Transforms/RegionUtils.h"
 #include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/TypeSwitch.h"
 
 using namespace mlir;
 using namespace mlir::tosa;
 
 #include "mlir/Dialect/Tosa/IR/TosaOpsDialect.cpp.inc"
 
 //===----------------------------------------------------------------------===//
 // Tosa dialect structs and interface includes.

 // Tosa dialect initialization.
 //===----------------------------------------------------------------------===//
 
 void TosaDialect::initialize() {
   addOperations<
 #define GET_OP_LIST
 #include "mlir/Dialect/Tosa/IR/TosaOps.cpp.inc"
       >();
+  addAttributes<
+#define GET_ATTRDEF_LIST
+#include "mlir/Dialect/Tosa/IR/TosaAttributes.cpp.inc"
+      >();
   addInterfaces<TosaInlinerInterface>();
 }
 
 Operation *TosaDialect::materializeConstant(OpBuilder &builder, Attribute value,
                                             Type type, Location loc) {
   // Tosa dialect constants only support ElementsAttr unlike standard dialect
   // constant which supports all attributes.
   if (value.isa<ElementsAttr>())
LogicalResult matchAndRewrite(tosa::PadOp op,
 
     auto input = op.input1();
     auto padding = op.padding();
 
     ShapedType inputTy = input.getType().cast<ShapedType>();
     Type elementTy = inputTy.getElementType();
 
     Attribute constantAttr;
-    if (elementTy.isa<FloatType>())
+    if (elementTy.isa<FloatType>()) {
       constantAttr = rewriter.getFloatAttr(elementTy, 0.0);
-    else if (elementTy.isa<IntegerType>() && !op.quantization_info())
+    } else if (elementTy.isa<IntegerType>() && !op.quantization_info()) {
       constantAttr = rewriter.getIntegerAttr(elementTy, 0);
-    else if (elementTy.isa<IntegerType>() && op.quantization_info()) {
-      auto value = op.quantization_info().getValue().input_zp().getValue();
-      constantAttr = rewriter.getIntegerAttr(elementTy, value.getZExtValue());
+    } else if (elementTy.isa<IntegerType>() && op.quantization_info()) {
+      auto value = op.quantization_info().getValue().getInput_zp();
+      constantAttr = rewriter.getIntegerAttr(elementTy, value);
     }
 
     if (!constantAttr) {
       return rewriter.notifyMatchFailure(
           op,
           "tosa.pad to linalg lowering encountered an unknown element type");
     }
 
LogicalResult WhileOp::inferReturnTypeComponents(
   for (const ValueKnowledge &result : resultKnowledge) {
     inferredReturnShapes.push_back(result.getShapedTypeComponents());
   }
 
   return success();
 }
 
 //===----------------------------------------------------------------------===//
+// TOSA Attribute Definitions.
+//===----------------------------------------------------------------------===//
+
+#define GET_ATTRDEF_CLASSES
+#include "mlir/Dialect/Tosa/IR/TosaAttributes.cpp.inc"
+
+//===----------------------------------------------------------------------===//
 // TOSA Operator Definitions.
 //===----------------------------------------------------------------------===//
 
 #define GET_OP_CLASSES
 #include "mlir/Dialect/Tosa/IR/TosaOps.cpp.inc"

mlir/lib/Dialect/Tosa/Transforms/TosaDecomposeTransposeConv.cpp

LogicalResult matchAndRewrite(tosa::TransposeConv2DOp op,
     Value weightPaddingVal = createOpAndInfer<tosa::ConstOp>(
         rewriter, loc, weightPaddingAttr.getType(), weightPaddingAttr);
 
     if (op.quantization_info().hasValue()) {
       auto quantInfo = op.quantization_info().getValue();
       weight = createOpAndInfer<tosa::PadOp>(
           rewriter, loc, UnrankedTensorType::get(weightETy), weight,
           weightPaddingVal, nullptr,
-          PadOpQuantizationAttr::get(quantInfo.weight_zp(),
-                                     rewriter.getContext()));
+          rewriter.getAttr<PadOpQuantizationAttr>(quantInfo.getWeight_zp()));
 
     } else {
       weight = createOpAndInfer<tosa::PadOp>(rewriter, loc,
                                              UnrankedTensorType::get(weightETy),
                                              weight, weightPaddingVal);
     }
 
     weightTy = weight.getType().cast<ShapedType>();
LogicalResult matchAndRewrite(tosa::TransposeConv2DOp op,
     Value inputPaddingVal = createOpAndInfer<tosa::ConstOp>(
         rewriter, loc, inputPaddingAttr.getType(), inputPaddingAttr);
 
     if (op.quantization_info().hasValue()) {
       auto quantInfo = op.quantization_info().getValue();
       input = createOpAndInfer<tosa::PadOp>(
           rewriter, loc, UnrankedTensorType::get(inputETy), input,
           inputPaddingVal, nullptr,
-          PadOpQuantizationAttr::get(quantInfo.input_zp(),
-                                     rewriter.getContext()));
+          rewriter.getAttr<PadOpQuantizationAttr>(quantInfo.getInput_zp()));
     } else {
       input = createOpAndInfer<tosa::PadOp>(rewriter, loc,
                                             UnrankedTensorType::get(inputETy),
                                             input, inputPaddingVal);
     }
 
     // We use a zero bias as we need to broadcast the bias.
     auto zeroBias = rewriter.create<tosa::ConstOp>(

mlir/lib/Dialect/Tosa/Utils/QuantUtils.cpp

assert(!((bool)weightPerTensorQType && (bool)weightPerAxisQType) &&
          "Weights must be either per-tensor or per-axis quantized");
 
   // Either all quantized or all not quantized.
   assert(!((bool)inputQType ^
            ((bool)weightPerTensorQType || (bool)weightPerAxisQType)) &&
          "Inputs and weights must be all quantized or all not quantized");
 
   if (inputQType) {
-
     int64_t inputZp = inputQType.getZeroPoint();
     int64_t weightZp = 0;
 
     if (weightPerTensorQType) {
       weightZp = weightPerTensorQType.getZeroPoint();
     } else if (weightPerAxisQType) {
       weightZp = weightPerAxisQType.getZeroPoints().front();
     }
 
-    auto quantAttr = tosa::ConvOpQuantizationAttr::get(
-        builder.getI32IntegerAttr(inputZp), builder.getI32IntegerAttr(weightZp),
-        builder.getContext());
-
-    return quantAttr;
+    return builder.getAttr<tosa::ConvOpQuantizationAttr>(inputZp, weightZp);
   }
 
   return nullptr;
 }
 
 /// Builds MatMulOpQuantizationAttr, called from
 /// MatMulOpQuantInfoBuilder:
 /// aZp: input a zeropoint
mlir::tosa::buildMatMulOpQuantizationAttr(OpBuilder &builder, Value a,
   auto aQType = GET_UQTYPE(aType);
   auto bQType = GET_UQTYPE(bType);
 
   // A and B are either all quantized or all not quantized.
   assert(!((bool)aQType ^ (bool)bQType) &&
          "Matmul operands must be all quantized or all not quantized");
 
   if (aQType) {
-
-    int64_t aZp = aQType.getZeroPoint();
-    int64_t bZp = bQType.getZeroPoint();
-
-    auto quantAttr = tosa::MatMulOpQuantizationAttr::get(
-        builder.getI32IntegerAttr(aZp), builder.getI32IntegerAttr(bZp),
-        builder.getContext());
-
-    return quantAttr;
+    return builder.getAttr<tosa::MatMulOpQuantizationAttr>(
+        aQType.getZeroPoint(), bQType.getZeroPoint());
   }
 
   return nullptr;
 }
 
 /// Builds UnaryOpQuantizationAttr
 /// UnaryOpQuantInfoBuilder:
 /// inputZp: input zeropoint
mlir::tosa::buildUnaryOpQuantizationAttr(OpBuilder &builder, Value input,
   auto inputQType = GET_UQTYPE(inputType);
   auto outputQType = GET_UQTYPE(outputType);
 
   // Either all quantized or all not quantized.
   assert(!((bool)inputQType ^ (bool)outputQType) &&
          "Unary inputs/outputs must be all quantized or all not quantized");
 
   if (inputQType) {
-
-    int64_t inputZp = inputQType.getZeroPoint();
-    int64_t outputZp = outputQType.getZeroPoint();
-
-    auto quantAttr = tosa::UnaryOpQuantizationAttr::get(
-        builder.getI32IntegerAttr(inputZp), builder.getI32IntegerAttr(outputZp),
-        builder.getContext());
-
-    return quantAttr;
+    return builder.getAttr<UnaryOpQuantizationAttr>(inputQType.getZeroPoint(),
+                                                    outputQType.getZeroPoint());
   }
 
   return nullptr;
 }
 
 /// Builds PadOpQuantizationAttr, called from PadOpQuantInfoBuilder:
 /// inputZp: input zeropoint.
 PadOpQuantizationAttr mlir::tosa::buildPadOpQuantizationAttr(OpBuilder &builder,
                                                              Value input) {
 
   auto inputType = input.getType().dyn_cast<ShapedType>();
 
   if (!inputType)
     return nullptr;
 
   auto inputQType = GET_UQTYPE(inputType);
 
   if (inputQType) {
-
-    int64_t inputZp = inputQType.getZeroPoint();
-
-    auto quantAttr = tosa::PadOpQuantizationAttr::get(
-        builder.getI32IntegerAttr(inputZp), builder.getContext());
-
-    return quantAttr;
+    return builder.getAttr<tosa::PadOpQuantizationAttr>(
+        inputQType.getZeroPoint());
   }
 
   return nullptr;
 }
 
 /// Builds output type for a quantized ConvOp with the right bitwidth.
 /// This is called by the builder when dealing with quantized content.
 Type mlir::tosa::buildConvOpResultTypeInfo(OpBuilder &builder, Type outputType,

mlir/test/Conversion/TosaToLinalg/tosa-to-linalg-named.mlir

 // CHECK-LABEL: @matmul_quantized
 func.func @matmul_quantized(%arg0: tensor<1x5x3xi8>, %arg1: tensor<1x3x6xi8>) -> (tensor<1x5x6xi32>) {
   // CHECK: [[C0:%.+]] = arith.constant 0
   // CHECK: [[INIT:%.+]] = linalg.init_tensor [1, 5, 6]
   // CHECK: [[FILLED:%.+]] = linalg.fill ins([[C0]] : i32) outs([[INIT]] : tensor<1x5x6xi32>) -> tensor<1x5x6xi32>
   // CHECK: [[ONE:%.+]] = arith.constant 1
   // CHECK: [[TWO:%.+]] = arith.constant 2
   // CHECK: linalg.quantized_batch_matmul ins(%arg0, %arg1, [[ONE]], [[TWO]] : tensor<1x5x3xi8>, tensor<1x3x6xi8>, i32, i32) outs([[FILLED]] : tensor<1x5x6xi32>) -> tensor<1x5x6xi32>
-  %0 = "tosa.matmul"(%arg0, %arg1) {quantization_info = {a_zp = 1 : i32, b_zp = 2 : i32}} : (tensor<1x5x3xi8>, tensor<1x3x6xi8>) -> (tensor<1x5x6xi32>)
+  %0 = "tosa.matmul"(%arg0, %arg1) {quantization_info = #tosa.matmul_quant<a_zp = 1, b_zp = 2>} : (tensor<1x5x3xi8>, tensor<1x3x6xi8>) -> (tensor<1x5x6xi32>)
   return %0 : tensor<1x5x6xi32>
 }
 
 // -----
 
 // CHECK-LABEL: @matmul_dyn_batch
 func.func @matmul_dyn_batch(%arg0: tensor<?x5x3xf32>, %arg1: tensor<?x3x6xf32>) -> (tensor<?x5x6xf32>) {
   // CHECK: %[[C0:.+]] = arith.constant 0
func.func @quantized_fully_connected(%arg0: tensor<5x3xi8>, %arg1: tensor<6x3xi8>, %arg2: tensor<6xi32>) -> (tensor<5x6xi32>) {
   // CHECK: [[INITB:%.+]] = linalg.init_tensor [5, 6]
   // CHECK: [[ONE:%.+]] = arith.constant 1
   // CHECK: [[TWO:%.+]] = arith.constant 2
   // CHECK: [[MATMUL:%.+]] = linalg.quantized_matmul ins(%arg0, [[TRANSPOSE]], [[ONE]], [[TWO]] : tensor<5x3xi8>, tensor<3x6xi8>, i32, i32) outs([[FILL]] : tensor<5x6xi32>) -> tensor<5x6xi32>
   // CHECK: [[ADDED:%.+]] = linalg.generic {indexing_maps = [#[[$MAP1]], #[[$MAP2]], #[[$MAP2]]], iterator_types = ["parallel", "parallel"]} ins(%arg2, [[MATMUL]] : tensor<6xi32>, tensor<5x6xi32>) outs([[INITB]]
   // CHECK: ^bb0([[IN1:%.+]]: i32, [[IN2:%.+]]: i32, [[UNUSED:%.+]]: i32):
   // CHECK:   [[ADD:%.+]] = arith.addi
   // CHECK:   linalg.yield [[ADD]] : i32
-  %0 = "tosa.fully_connected"(%arg0, %arg1, %arg2) {quantization_info = {input_zp = 1:i32, weight_zp = 2:i32}} : (tensor<5x3xi8>, tensor<6x3xi8>, tensor<6xi32>)  -> (tensor<5x6xi32>)
+  %0 = "tosa.fully_connected"(%arg0, %arg1, %arg2) {quantization_info = #tosa.conv_quant<input_zp = 1, weight_zp = 2>} : (tensor<5x3xi8>, tensor<6x3xi8>, tensor<6xi32>)  -> (tensor<5x6xi32>)
   return %0 : tensor<5x6xi32>
 }
 
 // -----
 
 // CHECK: #[[$MAP1:.*]] = affine_map<(d0, d1) -> (d1)>
 // CHECK: #[[$MAP2:.*]] = affine_map<(d0, d1) -> (d0, d1)>
 
func.func @avg_pool_i8(%arg0 : tensor<1x128x128x2xi8>) -> () {
   // CHECK: %[[MIN:.+]] = arith.constant -128
   // CHECK: %[[MAX:.+]] = arith.constant 127
   // CHECK: %[[CMP_MIN:.+]] = arith.cmpi slt, %[[OUT]], %[[MIN]]
   // CHECK: %[[CLMP_MIN:.+]] = arith.select %[[CMP_MIN]], %[[MIN]], %[[OUT]]
   // CHECK: %[[CMP_MAX:.+]] = arith.cmpi slt, %[[MAX]], %[[OUT]]
   // CHECK: %[[CLMP_MAX:.+]] = arith.select %[[CMP_MAX]], %[[MAX]], %[[CLMP_MIN]]
   // CHECK: %[[TRUNC:.+]] = arith.trunci %[[CLMP_MAX]]
   // CHECK: linalg.yield %[[TRUNC]]
-  %0 = "tosa.avg_pool2d"(%arg0) {kernel = [4, 4], pad = [0, 0, 0, 0], quantization_info = {input_zp = -128 : i32, output_zp = -128 : i32}, stride = [4, 4]} : (tensor<1x128x128x2xi8>) -> tensor<1x32x32x2xi8>
+  %0 = "tosa.avg_pool2d"(%arg0) {kernel = [4, 4], pad = [0, 0, 0, 0], quantization_info = #tosa.unary_quant<input_zp = -128, output_zp = -128>, stride = [4, 4]} : (tensor<1x128x128x2xi8>) -> tensor<1x32x32x2xi8>
   return
 }
 
 // -----
 
 // CHECK-LABEL: @avg_pool_i16
 func.func @avg_pool_i16(%arg0 : tensor<1x128x128x2xi16>) -> () {
 
func.func @avg_pool_i16(%arg0 : tensor<1x128x128x2xi16>) -> () {
   // CHECK: %[[MIN:.+]] = arith.constant -32768
   // CHECK: %[[MAX:.+]] = arith.constant 32767
   // CHECK: %[[CMP_MIN:.+]] = arith.cmpi slt, %[[OUT]], %[[MIN]]
   // CHECK: %[[CLMP_MIN:.+]] = arith.select %[[CMP_MIN]], %[[MIN]], %[[OUT]]
   // CHECK: %[[CMP_MAX:.+]] = arith.cmpi slt, %[[MAX]], %[[OUT]]
   // CHECK: %[[CLMP_MAX:.+]] = arith.select %[[CMP_MAX]], %[[MAX]], %[[CLMP_MIN]]
   // CHECK: %[[TRUNC:.+]] = arith.trunci %[[CLMP_MAX]]
   // CHECK: linalg.yield %[[TRUNC]]
-  %0 = "tosa.avg_pool2d"(%arg0) {kernel = [4, 4], pad = [0, 0, 0, 0], quantization_info = {input_zp = -128 : i32, output_zp = -128 : i32}, stride = [4, 4]} : (tensor<1x128x128x2xi16>) -> tensor<1x32x32x2xi16>
+  %0 = "tosa.avg_pool2d"(%arg0) {kernel = [4, 4], pad = [0, 0, 0, 0], quantization_info = #tosa.unary_quant<input_zp = -128, output_zp = -128>, stride = [4, 4]} : (tensor<1x128x128x2xi16>) -> tensor<1x32x32x2xi16>
   return
 }
 
 // -----
 
 // CHECK: #[[$MAP1:.+]] = affine_map<(d0, d1, d2, d3) -> (d3)>
 // CHECK: #[[$MAP2:.+]] = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
 

 // -----
 
 // CHECK-LABEL: @conv2d_quant
 func.func @conv2d_quant(%arg0 : tensor<1x12x12x1xi8>, %arg1 : tensor<1024x3x3x1xi8>, %arg2 : tensor<1024xi32>) -> () {
   // CHECK:   %[[C22:.+]] = arith.constant -22
   // CHECK: tensor.pad %arg0 low[0, 1, 1, 0] high[0, 1, 1, 0]
   // CHECK:   tensor.yield %[[C22]]
   // CHECK: linalg.conv_2d_nhwc_hwcf_q
-  %0 = "tosa.conv2d"(%arg0, %arg1, %arg2) {dilation = [1, 1], pad = [1, 1, 1, 1], quantization_info = {input_zp = -22 : i32, weight_zp = 42 : i32}, stride = [1, 1]} : (tensor<1x12x12x1xi8>, tensor<1024x3x3x1xi8>, tensor<1024xi32>) -> tensor<1x12x12x1024xi32>
+  %0 = "tosa.conv2d"(%arg0, %arg1, %arg2) {dilation = [1, 1], pad = [1, 1, 1, 1], quantization_info = #tosa.conv_quant<input_zp = -22, weight_zp = 42>, stride = [1, 1]} : (tensor<1x12x12x1xi8>, tensor<1024x3x3x1xi8>, tensor<1024xi32>) -> tensor<1x12x12x1024xi32>
   return
 }
 
 // -----
 
 // CHECK: #[[$MAP0:.*]] = affine_map<(d0, d1, d2, d3) -> (d3)>
 // CHECK: #[[$MAP1:.*]] = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
 
func.func @depthwise_conv_quant(%arg0 : tensor<1x12x12x4xi8>, %arg1 : tensor<3x3x4x128xi8>, %arg2 : tensor<512xi32>) -> () {
   // CHECK: [[C42:%.+]] = arith.constant 42
   // CHECK: [[DEPTH:%.+]] = linalg.depthwise_conv_2d_nhwc_hwcm_q {dilations = dense<1> : tensor<2xi64>, strides = dense<1> : tensor<2xi64>} ins([[PAD]], %arg1, [[C128]], [[C42]] : tensor<1x14x14x4xi8>, tensor<3x3x4x128xi8>, i32, i32) outs([[FILL]] : tensor<1x12x12x4x128xi32>)
   // CHECK: [[COLLAPSED:%.+]] = tensor.collapse_shape [[DEPTH]] {{\[}}[0], [1], [2], [3, 4]]
   // CHECK: [[BIAS:%.+]] = linalg.generic {indexing_maps = [#[[$MAP0]], #[[$MAP1]], #[[$MAP1]]], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%arg2, [[COLLAPSED]] : tensor<512xi32>, tensor<1x12x12x512xi32>) outs([[OUT]] : tensor<1x12x12x512xi32>) {
   // CHECK: ^bb0(%arg3: i32, %arg4: i32, %arg5: i32):  
   // CHECK:   [[ADD:%.+]] = arith.addi %arg3, %arg4 : i32
   // CHECK:   linalg.yield [[ADD]] : i32
   // CHECK: } -> tensor<1x12x12x512xi32>
-  %0 = "tosa.depthwise_conv2d"(%arg0, %arg1, %arg2) {pad = [1, 1, 1, 1], quantization_info = {input_zp = -128 : i32, weight_zp = 42 : i32}, stride = [1, 1], dilation = [1, 1] } : (tensor<1x12x12x4xi8>, tensor<3x3x4x128xi8>, tensor<512xi32>)  -> tensor<1x12x12x512xi32>
+  %0 = "tosa.depthwise_conv2d"(%arg0, %arg1, %arg2) {pad = [1, 1, 1, 1], quantization_info = #tosa.conv_quant<input_zp = -128, weight_zp = 42>, stride = [1, 1], dilation = [1, 1] } : (tensor<1x12x12x4xi8>, tensor<3x3x4x128xi8>, tensor<512xi32>)  -> tensor<1x12x12x512xi32>
   return
 }
 
 // -----
 
 // CHECK: #[[$MAP0:.*]] = affine_map<(d0, d1, d2, d3) -> (d3)>
 // CHECK: #[[$MAP1:.*]] = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
 
 // CHECK-LABEL: @depthwise_conv_quant_dilations
 func.func @depthwise_conv_quant_dilations(%arg0 : tensor<1x14x14x4xi8>, %arg1 : tensor<3x3x4x128xi8>, %arg2 : tensor<512xi32>) -> () {
   // CHECK: [[INIT:%.+]] = linalg.init_tensor [1, 10, 10, 4, 128]
   // CHECK: [[CST0:%.+]] = arith.constant 0
   // CHECK: [[FILL:%.+]] = linalg.fill ins([[CST0]]{{.*}}outs([[INIT]]
   // CHECK: [[OUT:%.+]] = linalg.init_tensor [1, 10, 10, 512]
   // CHECK: [[C128:%.+]] = arith.constant -128
   // CHECK: [[C42:%.+]] = arith.constant 42
   // CHECK: [[DEPTH:%.+]] = linalg.depthwise_conv_2d_nhwc_hwcm_q {dilations = dense<2> : tensor<2xi64>, strides = dense<1> : tensor<2xi64>} ins(%arg0, %arg1, [[C128]], [[C42]] : tensor<1x14x14x4xi8>, tensor<3x3x4x128xi8>, i32, i32) outs([[FILL]] : tensor<1x10x10x4x128xi32>)
   // CHECK: [[COLLAPSED:%.+]] = tensor.collapse_shape [[DEPTH]] {{\[}}[0], [1], [2], [3, 4]]
   // CHECK: [[BIAS:%.+]] = linalg.generic {indexing_maps = [#[[$MAP0]], #[[$MAP1]], #[[$MAP1]]], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%arg2, [[COLLAPSED]] : tensor<512xi32>, tensor<1x10x10x512xi32>) outs([[OUT]] : tensor<1x10x10x512xi32>) {
   // CHECK: ^bb0(%arg3: i32, %arg4: i32, %arg5: i32):  
   // CHECK:   [[ADD:%.+]] = arith.addi %arg3, %arg4 : i32
   // CHECK:   linalg.yield [[ADD]] : i32
   // CHECK: } -> tensor<1x10x10x512xi32>
-  %0 = "tosa.depthwise_conv2d"(%arg0, %arg1, %arg2) {pad = [0, 0, 0, 0], quantization_info = {input_zp = -128 : i32, weight_zp = 42 : i32}, stride = [1, 1], dilation = [2, 2] } : (tensor<1x14x14x4xi8>, tensor<3x3x4x128xi8>, tensor<512xi32>)  -> tensor<1x10x10x512xi32>
+  %0 = "tosa.depthwise_conv2d"(%arg0, %arg1, %arg2) {pad = [0, 0, 0, 0], quantization_info = #tosa.conv_quant<input_zp = -128, weight_zp = 42>, stride = [1, 1], dilation = [2, 2] } : (tensor<1x14x14x4xi8>, tensor<3x3x4x128xi8>, tensor<512xi32>)  -> tensor<1x10x10x512xi32>
   return
 }
 
 // CHECK-LABEL: @depthwise_conv2d_dyn_w_h
 func.func @depthwise_conv2d_dyn_w_h(%arg0: tensor<2x?x?x3xf32>, %arg1: tensor<3x6x3x5xf32>, %arg2: tensor<15xf32>) {
   // CHECK: arith.addi
   // CHECK: arith.subi
   // CHECK: arith.muli

mlir/test/Conversion/TosaToLinalg/tosa-to-linalg.mlir

func.func @test_negate_quantized(%arg0: tensor<1xi8>) -> () {
   // CHECK: [[MIN:%.+]] = arith.constant -128
   // CHECK: [[MAX:%.+]] = arith.constant 127
   // CHECK: [[PRED1:%.+]] = arith.cmpi slt, [[SUB]], [[MIN]]
   // CHECK: [[LBOUND:%.+]] = arith.select [[PRED1]], [[MIN]], [[SUB]]
   // CHECK: [[PRED2:%.+]] = arith.cmpi slt, [[MAX]], [[SUB]]
   // CHECK: [[UBOUND:%.+]] = arith.select [[PRED2]], [[MAX]], [[LBOUND]]
   // CHECK: [[TRUNC:%.+]] = arith.trunci [[UBOUND]]
   // CHECK: linalg.yield [[TRUNC]]
-  %0 = "tosa.negate"(%arg0) {quantization_info = { input_zp = 0 : i32, output_zp = 0 : i32}} : (tensor<1xi8>) -> tensor<1xi8>
+  %0 = "tosa.negate"(%arg0) {quantization_info = #tosa.unary_quant<input_zp = 0, output_zp = 0>} : (tensor<1xi8>) -> tensor<1xi8>
 
   // CHECK: linalg.generic
   // CHECK: [[EXT:%.+]] = arith.extsi %arg1 : i8 to i16
-  %1 = "tosa.negate"(%arg0) {quantization_info = { input_zp = 32639 : i32, output_zp = 0 : i32}} : (tensor<1xi8>) -> tensor<1xi8>
+  %1 = "tosa.negate"(%arg0) {quantization_info = #tosa.unary_quant<input_zp = 32639, output_zp = 0>} : (tensor<1xi8>) -> tensor<1xi8>
 
   // CHECK: linalg.generic
   // CHECK: [[EXT:%.+]] = arith.extsi %arg1 : i8 to i32
-  %2 = "tosa.negate"(%arg0) {quantization_info = { input_zp = 32640 : i32, output_zp = 0 : i32}} : (tensor<1xi8>) -> tensor<1xi8>
+  %2 = "tosa.negate"(%arg0) {quantization_info = #tosa.unary_quant<input_zp = 32640, output_zp = 0>} : (tensor<1xi8>) -> tensor<1xi8>
 
   return
 }
 
 // -----
 
 // CHECK-LABEL: @test_reshape_downrank
 func.func @test_reshape_downrank(%arg0: tensor<2x3xf32>) -> tensor<6xf32> {
func.func @pad_int(%arg0 : tensor<1x2xi32>) -> (tensor<4x9xi32>) {
   return %1 : tensor<4x9xi32>
 }
 
 func.func @pad_quant(%arg0 : tensor<1x2xi32>) -> (tensor<4x9xi32>) {
   %0 = arith.constant dense<[[1, 2], [3, 4]]> : tensor<2x2xi32>
   // CHECK: [[CST:%.+]] = arith.constant 42 : i32
   // CHECK: tensor.pad
   // CHECK:   tensor.yield [[CST]]
-  %1 = "tosa.pad"(%arg0, %0) { quantization_info = { input_zp = 42 : i32}} : (tensor<1x2xi32>, tensor<2x2xi32>)  -> (tensor<4x9xi32>)
+  %1 = "tosa.pad"(%arg0, %0) {quantization_info = #tosa.pad_quant<input_zp = 42>} : (tensor<1x2xi32>, tensor<2x2xi32>)  -> (tensor<4x9xi32>)
   return %1 : tensor<4x9xi32>
 }
 
 // -----
 
 func.func @pad_float_explicit(%arg0 : tensor<1x2xf32>) -> (tensor<4x9xf32>) {
   %0 = arith.constant dense<[[1, 2], [3, 4]]> : tensor<2x2xi32>
   // TODO: Output contains multiple "arith.constant 1 : index".

mlir/test/Dialect/Tosa/canonicalize.mlir

func.func @pad_determine_val_f32(%arg0: tensor<?x?xf32>, %arg1 : tensor<2x2xi32>) -> tensor<?x?xf32> {
   return %1 : tensor<?x?xf32>
 }
 
 // CHECK-LABEL: @pad_determine_val_quant
 func.func @pad_determine_val_quant(%arg0: tensor<?x?xi32>, %arg1 : tensor<2x2xi32>) -> tensor<?x?xi32> {
   // CHECK: %[[ZERO:.+]] = "tosa.const"() {value = dense<42> : tensor<i32>}
   // CHECK: "tosa.pad"(%arg0, %arg1, %[[ZERO]])
   %0 = "tosa.const"() { value = dense<[[1, 0], [0, 1]]> : tensor<2x2xi32>} : () -> tensor<2x2xi32>
-  %1 = "tosa.pad"(%arg0, %arg1) { quantization_info = {input_zp = 42:i32} } : (tensor<?x?xi32>, tensor<2x2xi32>) -> tensor<?x?xi32>
+  %1 = "tosa.pad"(%arg0, %arg1) {quantization_info = #tosa.pad_quant<input_zp = 42>} : (tensor<?x?xi32>, tensor<2x2xi32>) -> tensor<?x?xi32>
   return %1 : tensor<?x?xi32>
 }
 
 // CHECK-LABEL: @mul_one_different_shape
 func.func @mul_one_different_shape(%arg0: tensor<2x3xf32>) -> tensor<4x2x3xf32> {
   // CHECK: tosa.mul
   %ones = "tosa.const"() {value = dense<1.0> : tensor<4x2x3xf32>} : () -> tensor<4x2x3xf32>
   %1 = "tosa.mul"(%arg0, %ones) {shift = 0 : i32} : (tensor<2x3xf32>, tensor<4x2x3xf32>) -> tensor<4x2x3xf32>

mlir/test/Dialect/Tosa/quant-test.mlir

 // RUN: mlir-opt --tosa-test-quant-utils %s | FileCheck %s
 
 // -----
 // CHECK-LABEL: test_build_qtype
 func.func @test_build_qtype(%arg0 : tensor<16x1x1x8x!quant.uniform<u8<1:255>:f32, 0.015680249780416489:128>>) -> tensor<16x1x1x8x!quant.uniform<u8<1:255>:f32, 0.015680249780416489:128>> {
   //  CHECK: tosa.negate
   %0 = "tosa.negate"(%arg0) : (tensor<16x1x1x8x!quant.uniform<u8<1:255>:f32, 0.015680249780416489:128>>) -> tensor<16x1x1x8x!quant.uniform<u8<1:255>:f32, 0.015680249780416489:128>>
   return %0 : tensor<16x1x1x8x!quant.uniform<u8<1:255>:f32, 0.015680249780416489:128>>
 }
 
 // -----
 // CHECK-LABEL: test_build_mult_and_shift
 func.func @test_build_mult_and_shift(%arg0: tensor<1x32x32x8x!quant.uniform<i8:f32, 0.015684768557548523>>, %arg1 : tensor<16x1x1x8x!quant.uniform<i8<-127:127>:f32, 0.015680249780416489>>, %arg2 : tensor<16xi32>) -> tensor<1x32x32x16x!quant.uniform<i8:f32, 0.078431375324726104>> {
   // CHECK: tosa.conv2d
-  %0 = "tosa.conv2d"(%arg0, %arg1, %arg2) {pad = [1, 1, 2, 2], dilation = [2, 1], stride = [1, 1], quantization_info = {input_zp = -1 : i32, weight_zp = 0 : i32}} : (tensor<1x32x32x8x!quant.uniform<i8:f32, 0.015684768557548523>>, tensor<16x1x1x8x!quant.uniform<i8<-127:127>:f32, 0.015680249780416489>>, tensor<16xi32>) -> tensor<1x32x32x16x!quant.uniform<i8:f32, 0.078431375324726104>>
+  %0 = "tosa.conv2d"(%arg0, %arg1, %arg2) {pad = [1, 1, 2, 2], dilation = [2, 1], stride = [1, 1], quantization_info = #tosa.conv_quant<input_zp = -1, weight_zp = 0>} : (tensor<1x32x32x8x!quant.uniform<i8:f32, 0.015684768557548523>>, tensor<16x1x1x8x!quant.uniform<i8<-127:127>:f32, 0.015680249780416489>>, tensor<16xi32>) -> tensor<1x32x32x16x!quant.uniform<i8:f32, 0.078431375324726104>>
   return %0 : tensor<1x32x32x16x!quant.uniform<i8:f32, 0.078431375324726104>>
 
 }

mlir/test/Dialect/Tosa/tosa-decompose-conv2d.mlir

 // CHECK-LABEL: @conv2d_as_fully_connected_quant
 func.func @conv2d_as_fully_connected_quant(%arg0: tensor<4x10x10x2xi8>, %arg1: tensor<3x1x1x2xi8>, %arg2: tensor<3xi32>) -> tensor<4x10x10x3xi32> {
   // CHECK-NOT: "tosa.conv2d"
   // CHECK: %[[VAR0:.*]] = "tosa.reshape"(%arg0) {new_shape = [400, 2]}
   // CHECK-SAME: -> tensor<400x2xi8>
   // CHECK: %[[VAR1:.*]] = "tosa.reshape"(%arg1) {new_shape = [3, 2]}
   // CHECK-SAME: -> tensor<3x2xi8>
   // CHECK: %[[VAR2:.*]] = "tosa.fully_connected"(%[[VAR0]], %[[VAR1]], %arg2)
-  // CHECK-SAME: quantization_info = {input_zp = 42 : i32, weight_zp = 24 : i32}
+  // CHECK-SAME: quantization_info = #tosa.conv_quant<input_zp = 42, weight_zp = 24>
   // CHECK-SAME: -> tensor<400x3xi32>
   // CHECK: %[[VAR3:.*]] = "tosa.reshape"(%[[VAR2]]) {new_shape = [4, 10, 10, 3]}
   // CHECK-SAME: -> tensor<4x10x10x3xi32>
   // CHECK: return %[[VAR3]]
-  %0 = "tosa.conv2d"(%arg0, %arg1, %arg2) {pad = [0, 0, 0, 0], stride = [1, 1], dilation = [1, 1], quantization_info = {input_zp = 42 : i32, weight_zp = 24 : i32}} : (tensor<4x10x10x2xi8>, tensor<3x1x1x2xi8>, tensor<3xi32>) -> tensor<4x10x10x3xi32>
+  %0 = "tosa.conv2d"(%arg0, %arg1, %arg2) {pad = [0, 0, 0, 0], stride = [1, 1], dilation = [1, 1], quantization_info = #tosa.conv_quant<input_zp = 42, weight_zp = 24>} : (tensor<4x10x10x2xi8>, tensor<3x1x1x2xi8>, tensor<3xi32>) -> tensor<4x10x10x3xi32>
   return %0 : tensor<4x10x10x3xi32>
 }
 
 // -----

mlir/test/Dialect/Tosa/tosa-decompose-depthwise.mlir

func.func @depthwise_conv2d_as_mul(%arg0: tensor<4x10x10x2xf32>, %arg1: tensor<1x1x2x3xf32>, %arg2: tensor<6xf32>) -> tensor<4x10x10x6xf32> {
   return %0 : tensor<4x10x10x6xf32>
 }
 
 // -----
 
 // CHECK-LABEL: @depthwise_conv2d_as_mul_q
 func.func @depthwise_conv2d_as_mul_q(%arg0: tensor<4x10x10x2xi8>, %arg1: tensor<1x1x2x3xi8>, %arg2: tensor<6xi32>) -> tensor<4x10x10x6xi32> {
   // CHECK: "tosa.depthwise_conv2d"
-  %0 = "tosa.depthwise_conv2d"(%arg0, %arg1, %arg2) {pad = [0, 0, 0, 0], stride = [1, 1], dilation = [1, 1], quantization_info = {input_zp = 0 : i32, weight_zp = 0 : i32}} : (tensor<4x10x10x2xi8>, tensor<1x1x2x3xi8>, tensor<6xi32>) -> tensor<4x10x10x6xi32>
+  %0 = "tosa.depthwise_conv2d"(%arg0, %arg1, %arg2) {pad = [0, 0, 0, 0], stride = [1, 1], dilation = [1, 1], quantization_info = #tosa.conv_quant<input_zp = 0, weight_zp = 0>} : (tensor<4x10x10x2xi8>, tensor<1x1x2x3xi8>, tensor<6xi32>) -> tensor<4x10x10x6xi32>
   return %0 : tensor<4x10x10x6xi32>
 }
 
 // -----

mlir/test/Dialect/Tosa/tosa-decompose-transpose-conv.mlir

 }
 
 // -----
 
 // CHECK-LABEL: @transpose_conv2d_quantized
 func.func @transpose_conv2d_quantized(%arg0: tensor<2x16x14x3xi8>, %arg1: tensor<5x3x6x3xi8>, %arg2: tensor<5xi32>) -> (tensor<2x18x19x5xi32>) {
   // CHECK: %[[REV1:.+]] = "tosa.reverse"(%arg1) {axis = 1 : i64}
   // CHECK: %[[REV2:.+]] = "tosa.reverse"(%[[REV1]]) {axis = 2 : i64}
-  // CHECK: "tosa.conv2d"(%arg0, %[[REV2]], %arg2) {dilation = [1, 1], pad = [2, 2, 5, 5], quantization_info = {input_zp = -22 : i32, weight_zp = 42 : i32}, stride = [1, 1]}
-  %0 = "tosa.transpose_conv2d"(%arg0, %arg1, %arg2) {dilation = [1, 1], out_pad = [0, 0], quantization_info = {input_zp = -22 : i32, weight_zp = 42 : i32}, out_shape = [-1, -1, -1, -1], stride = [1, 1]} : (tensor<2x16x14x3xi8>, tensor<5x3x6x3xi8>, tensor<5xi32>) -> tensor<2x18x19x5xi32>
+  // CHECK: "tosa.conv2d"(%arg0, %[[REV2]], %arg2) {dilation = [1, 1], pad = [2, 2, 5, 5], quantization_info = #tosa.conv_quant<input_zp = -22, weight_zp = 42>, stride = [1, 1]}
+  %0 = "tosa.transpose_conv2d"(%arg0, %arg1, %arg2) {dilation = [1, 1], out_pad = [0, 0], quantization_info = #tosa.conv_quant<input_zp = -22, weight_zp = 42>, out_shape = [-1, -1, -1, -1], stride = [1, 1]} : (tensor<2x16x14x3xi8>, tensor<5x3x6x3xi8>, tensor<5xi32>) -> tensor<2x18x19x5xi32>
   return %0 : tensor<2x18x19x5xi32>
 }
 
 // ----
 
 // CHECK-LABEL: @transpose_conv2d_dilated
 func.func @transpose_conv2d_dilated(%arg0: tensor<2x16x14x3xf32>, %arg1: tensor<5x3x6x3xf32>, %arg2: tensor<5xf32>) -> tensor<2x?x?x5xf32> {
   // CHECK: %[[REV1:.+]] = "tosa.reverse"(%arg1) {axis = 1 : i64}

 
 // ----
 
 // CHECK-LABEL: @transpose_conv2d_strided_quantized
 func.func @transpose_conv2d_strided_quantized(%arg0: tensor<2x17x15x3xi8>, %arg1: tensor<5x3x5x3xi8>, %arg2: tensor<5xi32>) -> (tensor<2x35x47x5xi32>) {
   // Manipulate the weight matrix to handle striding.
   // CHECK-DAG: %[[PADV:.+]]  = "tosa.const"() {value = dense<{{\[\[}}0, 0], [0, 1], [0, 1], [0, 0]]> : tensor<4x2xi32>}
   // CHECK-DAG: %[[TRANSV:.+]]  = "tosa.const"() {value = dense<[2, 4, 0, 1, 3, 5]> : tensor<6xi32>}
-  // CHECK-DAG: %[[PADW:.+]]  = "tosa.pad"(%arg1, %[[PADV]]) {quantization_info = {input_zp = 42 : i32}}
+  // CHECK-DAG: %[[PADW:.+]]  = "tosa.pad"(%arg1, %[[PADV]]) {quantization_info = #tosa.pad_quant<input_zp = 42>}
   // CHECK-DAG: %[[RESW1:.+]]  = "tosa.reshape"(%[[PADW]]) {new_shape = [5, 2, 2, 2, 3, 3]}
   // CHECK-DAG: %[[TRANS:.+]]  = "tosa.transpose"(%[[RESW1]], %[[TRANSV]])
   // CHECK-DAG: %[[RESW2:.+]]  = "tosa.reshape"(%[[TRANS]]) {new_shape = [30, 2, 2, 3]}
   // CHECK-DAG: %[[REV1:.+]]  = "tosa.reverse"(%[[RESW2]]) {axis = 1 : i64}
   // CHECK-DAG: %[[NEWWEIGHT:.+]] = "tosa.reverse"(%[[REV1]]) {axis = 2 : i64}
 
   // Pad out the input matrix to handle the transpose conv.
   // CHECK-DAG: %[[PAD:.+]]  = "tosa.const"() {value = dense<{{\[\[}}0, 0], [1, 1], [1, 1], [0, 0]]> : tensor<4x2xi32>}
   // CHECK-DAG: %[[TRANS2:.+]]  = "tosa.const"() {value = dense<[0, 1, 3, 2, 4, 5]> : tensor<6xi32>}
-  // CHECK-DAG: %[[NEWINPUT:.+]] = "tosa.pad"(%arg0, %[[PAD]]) {quantization_info = {input_zp = -22 : i32}}
+  // CHECK-DAG: %[[NEWINPUT:.+]] = "tosa.pad"(%arg0, %[[PAD]]) {quantization_info = #tosa.pad_quant<input_zp = -22>}
 
   // Manipulate the final shape.
   // CHECK-DAG: %[[BIAS:.+]]  = "tosa.const"() {value = dense<0> : tensor<30xi32>}
-  // CHECK-DAG: %[[CONV:.+]] = "tosa.conv2d"(%[[NEWINPUT]], %[[NEWWEIGHT]], %[[BIAS]]) {dilation = [1, 1], pad = [0, 0, 0, 0], quantization_info = {input_zp = -22 : i32, weight_zp = 42 : i32}, stride = [1, 1]}
+  // CHECK-DAG: %[[CONV:.+]] = "tosa.conv2d"(%[[NEWINPUT]], %[[NEWWEIGHT]], %[[BIAS]]) {dilation = [1, 1], pad = [0, 0, 0, 0], quantization_info = #tosa.conv_quant<input_zp = -22, weight_zp = 42>, stride = [1, 1]}
   // CHECK-DAG: %[[RESHAPE_OUT_1:.+]] = "tosa.reshape"(%[[CONV]]) {new_shape = [2, 18, 16, 2, 3, 5]}
   // CHECK-DAG: %[[TRANS_OUT:.+]] = "tosa.transpose"(%[[RESHAPE_OUT_1]], %[[TRANS2]])
   // CHECK-DAG: %[[RESHAPE_OUT_2:.+]] = "tosa.reshape"(%[[TRANS_OUT]]) {new_shape = [2, 36, 48, 5]}
   // CHECK-DAG: %[[SLICE:.+]] = "tosa.slice"(%[[RESHAPE_OUT_2]]) {size = [2, 35, 47, 5], start = [0, 0, 0, 0]}
   // CHECK: %[[ADD:.+]] = "tosa.add"(%[[SLICE]], %arg2)
-  %0 = "tosa.transpose_conv2d"(%arg0, %arg1, %arg2) {dilation = [1, 1], out_pad = [0, 0], quantization_info = {input_zp = -22 : i32, weight_zp = 42 : i32}, out_shape = [-1, -1, -1, -1], stride = [2, 3]} : (tensor<2x17x15x3xi8>, tensor<5x3x5x3xi8>, tensor<5xi32>) -> tensor<2x35x47x5xi32>
+  %0 = "tosa.transpose_conv2d"(%arg0, %arg1, %arg2) {dilation = [1, 1], out_pad = [0, 0], quantization_info = #tosa.conv_quant<input_zp = -22, weight_zp = 42>, out_shape = [-1, -1, -1, -1], stride = [2, 3]} : (tensor<2x17x15x3xi8>, tensor<5x3x5x3xi8>, tensor<5xi32>) -> tensor<2x35x47x5xi32>
   return %0 : tensor<2x35x47x5xi32>
 }

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

tbl_outs = [
             ["-gen-dialect-decls"],
             "include/mlir/Dialect/Tosa/IR/TosaOpsDialect.h.inc",
         ),
         (
             ["-gen-dialect-defs"],
             "include/mlir/Dialect/Tosa/IR/TosaOpsDialect.cpp.inc",
         ),
         (
+            ["-gen-attrdef-decls"],
+            "include/mlir/Dialect/Tosa/IR/TosaAttributes.h.inc",
+        ),
+        (
+            ["-gen-attrdef-defs"],
+            "include/mlir/Dialect/Tosa/IR/TosaAttributes.cpp.inc",
+        ),
+        (
             ["-gen-op-doc"],
             "g3doc/Dialects/Tosa/TosaOps.md",
         ),
     ],
     tblgen = ":mlir-tblgen",
     td_file = "include/mlir/Dialect/Tosa/IR/TosaOps.td",
     deps = [":TosaDialectTdFiles"],
 )