From 67e9fc539dd014745f8e2559b967489b8479a8f8 Mon Sep 17 00:00:00 2001
From: Luke Hutton <luke.hutton@arm.com>
Date: Thu, 12 Jan 2023 11:30:33 +0000
Subject: Support translation of FFT2d

Signed-off-by: Luke Hutton <luke.hutton@arm.com>
Change-Id: I4aae94438380d394b9c13015aa69ac52f9b73f74
---
 include/operator.def    |  1 +
 src/TosaDeserialize.cpp | 36 +++++++++++++++++++++++++++++-------
 src/TosaSerialize.cpp   | 22 ++++++++++++++++++++++
 3 files changed, 52 insertions(+), 7 deletions(-)

diff --git a/include/operator.def b/include/operator.def
index 51eaf82..3c766c4 100644
--- a/include/operator.def
+++ b/include/operator.def
@@ -27,6 +27,7 @@ DEF_OPERATOR(AvgPool2d)
 DEF_OPERATOR(Conv2D)
 DEF_OPERATOR(Conv3D)
 DEF_OPERATOR(DepthwiseConv2D)
+DEF_OPERATOR(FFT2d)
 DEF_OPERATOR(FullyConnected)
 DEF_OPERATOR(MatMul)
 DEF_OPERATOR(MaxPool2d)
diff --git a/src/TosaDeserialize.cpp b/src/TosaDeserialize.cpp
index f6f78fc..d30a6f2 100644
--- a/src/TosaDeserialize.cpp
+++ b/src/TosaDeserialize.cpp
@@ -1057,6 +1057,28 @@ TosaMlirOperatorBuilder::build<Op_RFFT2D>(TosaSerializationOperator *op) const {
       {mlir_op->getResult(0), mlir_op->getResult(1)});
 }
 
+template <>
+std::vector<mlir::Value>
+TosaMlirOperatorBuilder::build<Op_FFT2D>(TosaSerializationOperator *op) const {
+  mlir::Value input0_val = tensor_map->at(op->GetInputTensorNames()[0]);
+  mlir::Value input1_val = tensor_map->at(op->GetInputTensorNames()[1]);
+  mlir::RankedTensorType output0_type =
+      tensor_type_map->at(op->GetOutputTensorNames()[0]);
+  mlir::RankedTensorType output1_type =
+      tensor_type_map->at(op->GetOutputTensorNames()[1]);
+
+  assert(op->GetAttributeType() == Attribute_FFTAttribute);
+  TosaFFTAttribute *attr =
+    static_cast<TosaFFTAttribute *>(op->GetAttribute());
+  auto inverse = op_builder->getBoolAttr(attr->inverse());
+
+  mlir::Operation *mlir_op = op_builder->create<mlir::tosa::FFT2dOp>(
+      loc, output0_type, output1_type, input0_val, input1_val, inverse);
+  block->push_back(mlir_op);
+  return std::vector<mlir::Value>(
+      {mlir_op->getResult(0), mlir_op->getResult(1)});
+}
+
 class TosaMlirRegionBuilder {
 public:
   TosaMlirRegionBuilder(TosaSerializationRegion* _ser_region,
@@ -1125,7 +1147,7 @@ mlir::LogicalResult TosaMlirBlockBuilder::BuildAllOpsInBlock(
     tensor_type_map[ts_name] = type;
     tensor_built[ts_name] = false;
   }
-  
+
   for (auto op : ser_block->GetOperators()) {
     operator_built[op] = false;
     for (auto ts_name : op->GetInputTensorNames()) {
@@ -1200,7 +1222,7 @@ mlir::LogicalResult TosaMlirBlockBuilder::BuildAllOpsInBlock(
       llvm::errs() << "ERROR: unsupported opcode=" << EnumNamesOp()[op->GetOp()] << "\n";
       return mlir::failure();
     }
-    
+
     // Sanity check if number of built mlir::Value is expected
     if (op->GetOutputTensorNames().size() != output_values.size()) {
       llvm::errs() << "ERROR: number of built mlir::Value is not matching number of operator output tensor\n";
@@ -1262,7 +1284,7 @@ mlir::LogicalResult TosaMlirRegionBuilder::BuildAllBlocksInRegion(
 mlir::LogicalResult buildTosaMlir(mlir::func::FuncOp& func,
                                   mlir::MLIRContext& context,
                                   tosa::TosaSerializationHandler& tsh) {
-                                    
+
   mlir::Region* main_region = func.getCallableRegion();
   if (!main_region) {
     llvm::errs() << "Invalid MLIR: doesn't have valid \"main\" region\n";
@@ -1274,7 +1296,7 @@ mlir::LogicalResult buildTosaMlir(mlir::func::FuncOp& func,
                     "must contain exactly one region\n";
     return mlir::failure();
   }
-  
+
   TosaSerializationRegion* ser_main_region = tsh.GetRegions().front();
 
   auto loc = func.getLoc();
@@ -1328,7 +1350,7 @@ public:
 
     auto function = getOperation();
     auto& context = getContext();
-    
+
     if (buildTosaMlir(function, context, tsh).failed()) {
       llvm::errs() << "Failed to deserialize flatbuffer " << tosa_deserialize_filename << "\n";
       return signalPassFailure();
@@ -1341,7 +1363,7 @@ class TosaDeserializeJSON
 public:
   void runOnOperation() final {
     TosaSerializationHandler tsh;
-    
+
     // must load tosa schema before loading json file
     if (loadTosaSchema(tsh).failed()) {
       return signalPassFailure();
@@ -1354,7 +1376,7 @@ public:
 
     auto function = getOperation();
     auto& context = getContext();
-    
+
     if (buildTosaMlir(function, context, tsh).failed()) {
       llvm::errs() << "Failed to deserialize flatbuffer " << tosa_deserialize_filename << "\n";
       return signalPassFailure();
diff --git a/src/TosaSerialize.cpp b/src/TosaSerialize.cpp
index e69fcba..8a95b68 100644
--- a/src/TosaSerialize.cpp
+++ b/src/TosaSerialize.cpp
@@ -1367,6 +1367,28 @@ TosaSerializationOperatorBuilder::build<mlir::tosa::RFFT2dOp>(
   return tyop;
 }
 
+template<>
+TosaSerializationOperator *
+TosaSerializationOperatorBuilder::build<mlir::tosa::FFT2dOp>(
+    mlir::Operation &op) const {
+
+  bool inverse = op.getAttr("inverse").dyn_cast<mlir::BoolAttr>().getValue();
+
+  std::string input_real_name = GetTensorName(op.getOperand(0));
+  std::string input_imag_name = GetTensorName(op.getOperand(1));
+  std::string output_real_name = GetTensorName(op.getResult(0));
+  std::string output_imag_name = GetTensorName(op.getResult(1));
+
+  TosaFFTAttribute attribute(inverse);
+
+  TosaSerializationOperator *tyop = new TosaSerializationOperator(
+    Op_FFT2D, Attribute_FFTAttribute, &attribute,
+    std::vector<std::string>{input_real_name, input_imag_name},
+    std::vector<std::string>{output_real_name, output_imag_name});
+
+  return tyop;
+}
+
 /* End translating TOSA operator */
 mlir::LogicalResult TosaSerializationRegionBuilder::BuildAllBlocksInRegion(std::vector<mlir::Value>& return_values) {
   std::string region_name = ser_region->GetName();
-- 
cgit v1.2.1