Change PadOp's padding to TOSA Shape

PadOp's padding input is now a tosa.shape type.
Changed serialization of PadOp to not store padding as attribute.
Changed deserialization of PadOp to not restore padding from attribute.

Signed-off-by: Tai Ly <tai.ly@arm.com>
Change-Id: I8a622978ea184b8d2779d311adba629c1a0d1fbd

2 files changed, 9 insertions, 26 deletions

diff --git a/src/TosaDeserialize.cpp b/src/TosaDeserialize.cpp
index 2421d79..5956bc8 100644
--- a/src/TosaDeserialize.cpp
+++ b/src/TosaDeserialize.cpp
@@ -1029,6 +1029,7 @@ template <>
 std::vector<mlir::Value>
 TosaMlirOperatorBuilder::build<Op_PAD>(TosaSerializationOperator *op) const {
   mlir::Value input_val = tensor_map->at(op->GetInputTensorNames()[0]);
+  mlir::Value padding_val = tensor_map->at(op->GetInputTensorNames()[1]);
   mlir::RankedTensorType input_type =
       tensor_type_map->at(op->GetInputTensorNames()[0]);
   mlir::RankedTensorType output_type =
@@ -1038,15 +1039,6 @@ TosaMlirOperatorBuilder::build<Op_PAD>(TosaSerializationOperator *op) const {
          Attribute_PadAttribute); // double check attribute type
   TosaPadAttribute *attr = static_cast<TosaPadAttribute *>(op->GetAttribute());
 
-  // padding has shape {rank(input_type), 2}
-  auto padding_type = mlir::RankedTensorType::get({input_type.getRank(), 2},
-                                                  op_builder->getI32Type());
-  auto padding_attr =
-      BuildDenseI32ElementsAttr(op_builder, padding_type, attr->padding());
-  mlir::Operation *mlir_const_op =
-      op_builder->create<mlir::tosa::ConstOp>(loc, padding_type, padding_attr);
-  block->push_back(mlir_const_op);
-  auto padding_value = mlir_const_op->getResult(0);
   auto pad_const_int = attr->pad_const_int();
   auto pad_const_fp = attr->pad_const_fp();
   // todo: int input_zp = attr->pad_input_zp();
@@ -1055,7 +1047,7 @@ TosaMlirOperatorBuilder::build<Op_PAD>(TosaSerializationOperator *op) const {
   if (pad_const_int == 0 && pad_const_fp == 0.0f) {
     // no pad_const input
     mlir_op = op_builder->create<mlir::tosa::PadOp>(loc, output_type, input_val,
-                                                    padding_value);
+                                                    padding_val);
   } else {
     // create a const value for pad_const input
     const auto input_element_type =
@@ -1083,7 +1075,7 @@ TosaMlirOperatorBuilder::build<Op_PAD>(TosaSerializationOperator *op) const {
       pad_const_value = pad_const_fp_op->getResult(0);
     }
     mlir_op = op_builder->create<mlir::tosa::PadOp>(
-        loc, output_type, input_val, padding_value, pad_const_value);
+        loc, output_type, input_val, padding_val, pad_const_value);
   }
   block->push_back(mlir_op);
   return std::vector<mlir::Value>({mlir_op->getResult(0)});
diff --git a/src/TosaSerialize.cpp b/src/TosaSerialize.cpp
index 4f2c358..de301fe 100644
--- a/src/TosaSerialize.cpp
+++ b/src/TosaSerialize.cpp
@@ -1205,19 +1205,10 @@ TosaSerializationOperator *
 TosaSerializationOperatorBuilder::build<mlir::tosa::PadOp>(
     mlir::Operation &op) const {
   std::string input_name = GetTensorName(op.getOperand(0));
+  std::string padding_name = GetTensorName(op.getOperand(1));
   std::string output_name = GetTensorName(op.getResult(0));
   auto pad_op = llvm::cast<mlir::tosa::PadOp>(op);
 
-  // Match padding tensor as compile-time constant attribute
-  mlir::ElementsAttr paddings_elems;
-  if (!matchPattern(pad_op.getPadding(), m_Constant(&paddings_elems)))
-    return nullptr;
-
-  std::vector<int> paddings;
-  for (int32_t val : paddings_elems.getValues<int32_t>()) {
-    paddings.push_back(val);
-  }
-
   auto quant_info = pad_op.getQuantizationInfoAttr();
   // pad_const includes the zero point if the tensor uses a zero point.
   int32_t pad_const_int = quant_info ? quant_info.getInputZp() : 0;
@@ -1242,12 +1233,12 @@ TosaSerializationOperatorBuilder::build<mlir::tosa::PadOp>(
     }
   }
 
-  TosaPadAttribute attribute(paddings, pad_const_int, pad_const_fp);
+  TosaPadAttribute attribute({}, pad_const_int, pad_const_fp);
 
-  TosaSerializationOperator *tyop =
-      new TosaSerializationOperator(Op_PAD, Attribute_PadAttribute, &attribute,
-                                    std::vector<std::string>{input_name},
-                                    std::vector<std::string>{output_name});
+  TosaSerializationOperator *tyop = new TosaSerializationOperator(
+      Op_PAD, Attribute_PadAttribute, &attribute,
+      std::vector<std::string>{input_name, padding_name},
+      std::vector<std::string>{output_name});
   return tyop;
 }