[TOSA] Change PadOp padding to tosa.shape (#123133)

This patch changes PadOp's padding input to type !tosa.shape<2 * rank>,
(where rank is the rank of the PadOp's input), instead of a <rank x 2>
tensor.

This patch is also a part of TOSA v1.0 effort:
https://discourse.llvm.org/t/rfc-tosa-dialect-increment-to-v1-0/83708

This patch updates the PadOp to match all against the TOSA v1.0 form. 

Original Authors include: 
@Tai78641 
@wonjeon

Co-authored-by: Tai Ly <[email protected]>

Author: Jerry-Ge

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

@@ -1557,21 +1557,21 @@ def Tosa_PadOp : Tosa_InferShapedTypeOp<"pad"> {
     Example:
 
     ```mlir
-    %0 = arith.constant dense<[1, 2, 3, 4]> : tensor<4xi32>
-    tosa.pad %arg0, %0 : (tensor<1x2xf32>, tensor<4xi32>)  -> (tensor<4x9xf32>)
+    %0 = tosa.const_shape { value = dense<[1, 2, 3, 4]> : tensor<4xindex> } : () -> !tosa.shape<4>
+    tosa.pad %arg0, %0 : (tensor<1x2xf32>, !tosa.shape<4>)  -> (tensor<4x9xf32>)
     ```
 
     Example 2:
 
     ```mlir
-    %0 = arith.constant dense<[-1, 2, 3, 4]> : tensor<4xi32>
-    tosa.pad %arg0, %0 : (tensor<1x2xf32>, tensor<4xi32>)  -> (tensor<?x9xf32>)
+    %0 = tosa.const_shape { value = dense<[-1, 2, 3, 4]> : tensor<4xindex> } : () -> !tosa.shape<4>
+    tosa.pad %arg0, %0 : (tensor<1x2xf32>, !tosa.shape<4>)  -> (tensor<?x9xf32>)
     ```
   }];
 
   let arguments = (ins
     Tosa_RankedTensor:$input1,
-    TosaTensorRankOf<[Tosa_Int32Or64], [1]>:$padding,
+    Tosa_Shape:$padding,
     Optional<Tosa_ScalarTensor>:$pad_const,
     OptionalAttr<Tosa_PadOpQuantizationAttr>:$quantization_info
   );

mlir/include/mlir/Dialect/Tosa/Utils/ConversionUtils.h

@@ -229,6 +229,14 @@ SmallVector<T> applyTOSAPermutation(ArrayRef<T> input,
   return permuted;
 }
 
+// Computes shape value using tosa const_shape op.
+Value getTosaConstShape(PatternRewriter &rewriter, Location loc,
+                        llvm::ArrayRef<int64_t> shape);
+SmallVector<int64_t> convertFromMlirShape(ArrayRef<int64_t> shape);
+
+bool getConstShapeValue(Operation *op,
+                        llvm::SmallVector<int64_t> &result_shape);
+
 } // namespace tosa
 } // namespace mlir
 

mlir/lib/Conversion/TosaToTensor/TosaToTensor.cpp

@@ -306,7 +306,16 @@ class PadConverter : public OpConversionPattern<tosa::PadOp> {
                   ConversionPatternRewriter &rewriter) const final {
     auto loc = padOp.getLoc();
     auto input = padOp.getInput1();
-    auto padding = padOp.getPadding();
+
+    ElementsAttr paddingElems;
+    if (!matchPattern(padOp.getPadding(), m_Constant(&paddingElems))) {
+      return rewriter.notifyMatchFailure(
+          padOp, "padding must be a static shape value");
+    }
+    llvm::SmallVector<int64_t> paddingVals;
+    for (auto idx : paddingElems.getValues<IntegerAttr>()) {
+      paddingVals.push_back(static_cast<int64_t>(idx.getInt()));
+    }
 
     ShapedType inputTy = cast<ShapedType>(input.getType());
     Type elementTy = inputTy.getElementType();
@@ -345,18 +354,10 @@ class PadConverter : public OpConversionPattern<tosa::PadOp> {
     highValues.reserve(rank);
 
     for (int i = 0; i < rank; i++) {
-      Value lowIndex = rewriter.create<arith::ConstantIndexOp>(loc, 2 * i);
-      Value highIndex = rewriter.create<arith::ConstantIndexOp>(loc, 2 * i + 1);
-      Value lowVal = rewriter.createOrFold<tensor::ExtractOp>(
-          loc, padding, ValueRange({lowIndex}));
-      Value highVal = rewriter.createOrFold<tensor::ExtractOp>(
-          loc, padding, ValueRange({highIndex}));
-
-      lowVal = rewriter.createOrFold<arith::IndexCastOp>(
-          loc, rewriter.getIndexType(), lowVal);
-      highVal = rewriter.createOrFold<arith::IndexCastOp>(
-          loc, rewriter.getIndexType(), highVal);
-
+      Value lowVal = rewriter.create<arith::ConstantOp>(
+          loc, rewriter.getIndexAttr(paddingVals[2 * i]));
+      Value highVal = rewriter.create<arith::ConstantOp>(
+          loc, rewriter.getIndexAttr(paddingVals[2 * i + 1]));
       lowValues.push_back(lowVal);
       highValues.push_back(highVal);
     }

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

@@ -36,6 +36,7 @@ using namespace mlir;
 using namespace mlir::tosa;
 
 #include "mlir/Dialect/Tosa/IR/TosaOpsDialect.cpp.inc"
+#include "mlir/Dialect/Tosa/Utils/ConversionUtils.h"
 
 //===----------------------------------------------------------------------===//
 // Tosa dialect interface includes.
@@ -822,51 +823,42 @@ LogicalResult tosa::PadOp::inferReturnTypeComponents(
     PadOp::Adaptor adaptor,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   ShapeAdaptor inputShape(adaptor.getInput1().getType());
-  ShapeAdaptor paddingShape(adaptor.getPadding().getType());
+  auto paddingRank =
+      cast<tosa::shapeType>(adaptor.getPadding().getType()).getRank();
   SmallVector<int64_t> outputShape;
 
-  // If both inputs have unknown shape, we cannot determine the shape of the
-  // output.
-  if (!inputShape.hasRank() && !paddingShape.hasRank()) {
-    inferredReturnShapes.push_back(ShapedTypeComponents());
-    return success();
-  }
-
-  // If the input rank is unknown we can info the output rank using the
-  // padding shape's first dim.
+  // If the input rank is unknown, we can infer the output rank using the
+  // padding shape's rank divided by 2.
   if (!inputShape.hasRank()) {
-    if (paddingShape.isDynamicDim(0)) {
-      inferredReturnShapes.push_back(ShapedTypeComponents());
-      return success();
-    }
-
-    outputShape.resize(paddingShape.getDimSize(0) / 2, ShapedType::kDynamic);
+    outputShape.resize(paddingRank / 2, ShapedType::kDynamic);
     inferredReturnShapes.push_back(ShapedTypeComponents(outputShape));
     return success();
   }
 
-  DenseIntElementsAttr paddings;
+  SmallVector<int64_t> paddingValues;
   // If the paddings value is not a constant, all dimensions must be dynamic.
-  if (!matchPattern(adaptor.getPadding(), m_Constant(&paddings))) {
+  if (!tosa::getConstShapeValue(adaptor.getPadding().getDefiningOp(),
+                                paddingValues)) {
     outputShape.resize(inputShape.getRank(), ShapedType::kDynamic);
     inferredReturnShapes.push_back(ShapedTypeComponents(outputShape));
     return success();
   }
 
-  SmallVector<int64_t> paddingValues;
-  for (auto val : paddings) {
-    paddingValues.push_back(val.getSExtValue());
-  }
-
   outputShape.reserve(inputShape.getRank());
   for (int i = 0, s = inputShape.getRank(); i < s; i++) {
     if (inputShape.isDynamicDim(i)) {
       outputShape.push_back(ShapedType::kDynamic);
       continue;
     }
+    auto padFront = paddingValues[i * 2];
+    auto padBack = paddingValues[i * 2 + 1];
+    if (padFront < 0 || padBack < 0) {
+      // if either padding for dim i is -1, output dim is unknown
+      outputShape.push_back(ShapedType::kDynamic);
+      continue;
+    }
 
-    outputShape.push_back(inputShape.getDimSize(i) + paddingValues[i * 2] +
-                          paddingValues[i * 2 + 1]);
+    outputShape.push_back(inputShape.getDimSize(i) + padFront + padBack);
   }
 
   inferredReturnShapes.push_back(ShapedTypeComponents(outputShape));
@@ -876,17 +868,15 @@ LogicalResult tosa::PadOp::inferReturnTypeComponents(
 LogicalResult tosa::PadOp::verify() {
   RankedTensorType inputType = getInput1().getType();
   RankedTensorType outputType = getOutput().getType();
-  RankedTensorType paddingType = getPadding().getType();
+  auto paddingRank = cast<tosa::shapeType>(getPadding().getType()).getRank();
 
   if (inputType.getRank() != outputType.getRank())
     return emitOpError() << "expect same input and output tensor rank.";
 
-  if (!paddingType.isDynamicDim(0) &&
-      paddingType.getDimSize(0) != inputType.getRank() * 2)
+  if (paddingRank != inputType.getRank() * 2)
     return emitOpError() << "expected padding tensor dim 0 to have size "
                          << inputType.getRank() * 2
-                         << " (2*rank(shape1)) but got size "
-                         << paddingType.getDimSize(0);
+                         << " (2*rank(shape1)) but got size " << paddingRank;
 
   return success();
 }

mlir/lib/Dialect/Tosa/Transforms/TosaDecomposeConv2D.cpp

@@ -81,11 +81,7 @@ struct Conv2DIsFullyConnected : public OpRewritePattern<tosa::Conv2DOp> {
         }
       }
 
-      auto padSizeTy = RankedTensorType::get({8}, rewriter.getI64Type());
-      auto padSize =
-          DenseIntElementsAttr::get(padSizeTy, ArrayRef<int64_t>(pad));
-      Value padSizeVal =
-          rewriter.create<tosa::ConstOp>(op->getLoc(), padSizeTy, padSize);
+      Value padSizeVal = getTosaConstShape(rewriter, op->getLoc(), pad);
 
       auto padTy = RankedTensorType::get({}, inputETy);
       auto padAttr = DenseElementsAttr::get(padTy, zeroAttr);

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

@@ -108,11 +108,7 @@ struct DepthwiseConv2DIsMul : public OpRewritePattern<tosa::DepthwiseConv2DOp> {
         }
       }
 
-      auto padSizeTy = RankedTensorType::get({10}, rewriter.getI64Type());
-      auto padSize =
-          DenseIntElementsAttr::get(padSizeTy, ArrayRef<int64_t>(pad));
-      Value padSizeVal =
-          rewriter.create<tosa::ConstOp>(op->getLoc(), padSizeTy, padSize);
+      Value padSizeVal = getTosaConstShape(rewriter, op->getLoc(), pad);
 
       auto padTy = RankedTensorType::get({}, inputETy);
       auto padAttr = DenseElementsAttr::get(padTy, zeroAttr);

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

@@ -135,15 +135,14 @@ class TransposeConvStridedConverter
     int64_t inputChannels = weightTy.getDimSize(3);
 
     // Pad the weight so that it is modulo of the striding.
-    llvm::SmallVector<int32_t, 8> weightPadding = {0, 0, 0, 0, 0, 0, 0, 0};
+    llvm::SmallVector<int64_t, 8> weightPadding = {0, 0, 0, 0, 0, 0, 0, 0};
     weightPadding[3] =
         (weightHeight % stride[0]) ? (stride[0] - weightHeight % stride[0]) : 0;
     weightPadding[5] =
-        (weightWidth % stride[1]) ? (stride[1] - weightWidth % stride[1]) : 0;
-    DenseElementsAttr weightPaddingAttr = DenseIntElementsAttr::get(
-        RankedTensorType::get({8}, rewriter.getI32Type()), weightPadding);
-    Value weightPaddingVal = CreateOpAndInferShape<tosa::ConstOp>(
-        rewriter, loc, weightPaddingAttr.getType(), weightPaddingAttr);
+        weightWidth % stride[1] ? stride[1] - weightWidth % stride[1] : 0;
+
+    Value weightPaddingVal =
+        getTosaConstShape(rewriter, op->getLoc(), weightPadding);
 
     if (op.getQuantizationInfo().has_value()) {
       auto quantInfo = op.getQuantizationInfo().value();
@@ -197,17 +196,14 @@ class TransposeConvStridedConverter
         /* axis = */ rewriter.getI32IntegerAttr(2));
 
     // We need to pad the input far enough that we can pull all values.
-    llvm::SmallVector<int32_t, 8> inputPadding = {0, 0, 0, 0, 0, 0, 0, 0};
+    llvm::SmallVector<int64_t, 8> inputPadding = {0, 0, 0, 0, 0, 0, 0, 0};
     inputPadding[2] += restridedWeightTy.getDimSize(1) - 1;
     inputPadding[3] += restridedWeightTy.getDimSize(1) - 1;
     inputPadding[4] += restridedWeightTy.getDimSize(2) - 1;
     inputPadding[5] += restridedWeightTy.getDimSize(2) - 1;
 
-    DenseElementsAttr inputPaddingAttr = DenseIntElementsAttr::get(
-        RankedTensorType::get({8}, rewriter.getI32Type()), inputPadding);
-
-    Value inputPaddingVal = CreateOpAndInferShape<tosa::ConstOp>(
-        rewriter, loc, inputPaddingAttr.getType(), inputPaddingAttr);
+    Value inputPaddingVal =
+        getTosaConstShape(rewriter, op->getLoc(), inputPadding);
 
     if (op.getQuantizationInfo().has_value()) {
       auto quantInfo = op.getQuantizationInfo().value();
@@ -310,17 +306,14 @@ class TransposeConvStridedConverter
                      rewriter.getDenseI64ArrayAttr(sliceSize))
                      .getResult();
 
-    llvm::SmallVector<int32_t, 8> resultPadding = {0, 0, 0, 0, 0, 0, 0, 0};
+    llvm::SmallVector<int64_t, 8> resultPadding = {0, 0, 0, 0, 0, 0, 0, 0};
     resultPadding[2] = resultPadTop;
     resultPadding[3] = resultTy.getDimSize(1) - resultPadTop - sliceSize[1];
     resultPadding[4] = resultPadLeft;
     resultPadding[5] = resultTy.getDimSize(2) - resultPadLeft - sliceSize[2];
 
-    DenseElementsAttr resultPaddingAttr = DenseIntElementsAttr::get(
-        RankedTensorType::get({8}, rewriter.getI32Type()), resultPadding);
-
-    Value resultPaddingVal = CreateOpAndInferShape<tosa::ConstOp>(
-        rewriter, loc, resultPaddingAttr.getType(), resultPaddingAttr);
+    Value resultPaddingVal =
+        getTosaConstShape(rewriter, op->getLoc(), resultPadding);
 
     Value resultPad = CreateOpAndInferShape<tosa::PadOp>(
         rewriter, loc, UnrankedTensorType::get(resultETy), slice,

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

@@ -160,3 +160,36 @@ LogicalResult mlir::tosa::EqualizeRanks(ImplicitLocOpBuilder &builder,
 
   return success();
 }
+
+Value mlir::tosa::getTosaConstShape(PatternRewriter &rewriter, Location loc,
+                                    llvm::ArrayRef<int64_t> shape) {
+  auto attr = rewriter.getIndexTensorAttr(shape);
+  auto type = mlir::tosa::shapeType::get(rewriter.getContext(), shape.size());
+  mlir::Operation *mlir_op =
+      rewriter.create<tosa::ConstShapeOp>(loc, type, attr);
+  return mlir_op->getResult(0);
+}
+
+SmallVector<int64_t> mlir::tosa::convertFromMlirShape(ArrayRef<int64_t> shape) {
+  return to_vector(llvm::map_range(shape, [](int64_t dim) {
+    return ShapedType::isDynamic(dim) ? -1 : dim;
+  }));
+}
+
+bool mlir::tosa::getConstShapeValue(Operation *op,
+                                    llvm::SmallVector<int64_t> &result_shape) {
+  if (!op) {
+    return false;
+  }
+  if (auto constOp = mlir::dyn_cast<tosa::ConstShapeOp>(op)) {
+    Attribute constOpAttr = constOp->getAttr("value");
+    DenseElementsAttr elementsAttr = cast<DenseElementsAttr>(constOpAttr);
+    for (int i = 0; i < elementsAttr.size(); i++) {
+      int64_t val = elementsAttr.getValues<int64_t>()[i];
+      result_shape.push_back(val);
+    }
+    return true;
+  }
+  // for undefined op, return false.
+  return false;
+}