Import third_party/hexagon (#1028)

Importing third_party/hexagon from codelinario at commit:
5bee22654219b89efdd35e447cc89d50369ad168

This is generated by running ci/import_third_party_hexagon.sh

BUG=https://b/227665919

third_party/hexagon/fully_connected.cc

@@ -72,13 +72,17 @@ TfLiteStatus EvalFloat(TfLiteContext* context, TfLiteNode* node,
   tflite::FullyConnectedParams op_params;
   op_params.float_activation_min = output_activation_min;
   op_params.float_activation_max = output_activation_max;
+
+  const float* bias_data =
+      nullptr != bias ? tflite::micro::GetTensorData<float>(bias) : nullptr;
+
   tflite::reference_ops::FullyConnected(
       op_params, tflite::micro::GetTensorShape(input),
       tflite::micro::GetTensorData<float>(input),
       tflite::micro::GetTensorShape(filter),
       tflite::micro::GetTensorData<float>(filter),
       tflite::micro::GetTensorShape(bias),
-      tflite::micro::GetTensorData<float>(bias),
+      bias_data,
       tflite::micro::GetTensorShape(output),
       tflite::micro::GetTensorData<float>(output));
   return kTfLiteOk;

third_party/hexagon/fully_connected_int8.cc

@@ -53,7 +53,9 @@ ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "tensorflow/lite/kernels/kernel_util.h"
 #include "tensorflow/lite/micro/kernels/fully_connected.h"
 #include "tensorflow/lite/micro/kernels/kernel_util.h"
+
 #include "third_party/hexagon/hexagon_fully_connected.h"
+#include "third_party/hexagon/hexagon_tflm_translation_fully_connected.h"
 
 namespace tflite {
 namespace {
@@ -70,19 +72,22 @@ TfLiteStatus EvalQuantizedInt8(TfLiteContext* context, TfLiteNode* node,
   op_params.output_offset = data.reference_op_data.output_zero_point;
   op_params.output_multiplier = data.reference_op_data.output_multiplier;
   // TODO(b/138810107): Figure out whether output shift should be inverted
-  op_params.output_shift = -data.reference_op_data.output_shift;
+  op_params.output_shift = data.reference_op_data.output_shift;
   op_params.quantized_activation_min =
       data.reference_op_data.output_activation_min;
   op_params.quantized_activation_max =
       data.reference_op_data.output_activation_max;
 
+  const int32_t* bias_data =
+      nullptr != bias ? tflite::micro::GetTensorData<int32_t>(bias) : nullptr;
+
   reference_integer_ops::FullyConnected(
       op_params, tflite::micro::GetTensorShape(input),
       tflite::micro::GetTensorData<int8_t>(input),
       tflite::micro::GetTensorShape(filter),
       tflite::micro::GetTensorData<int8_t>(filter),
       tflite::micro::GetTensorShape(bias),
-      tflite::micro::GetTensorData<int32_t>(bias),
+      bias_data,
       tflite::micro::GetTensorShape(output),
       tflite::micro::GetTensorData<int8_t>(output));
 
@@ -132,9 +137,19 @@ TfLiteStatus HexagonFullyConnectedPrepare(TfLiteContext* context, TfLiteNode* no
       AllocateTempOutputTensor(node, kFullyConnectedOutputTensor);
   TF_LITE_ENSURE(context, output != nullptr);
 
+  TF_LITE_ENSURE_OK(context, CalculateOpDataFullyConnected(
+                                 context, params->activation, input->type,
+                                 input, filter, bias, output, &data->reference_op_data));
+
+  TF_LITE_ENSURE_TYPES_EQ(context, input->type, output->type);
+  TF_LITE_ENSURE_MSG(context, input->type == filter->type,
+                     "Hybrid models are not supported on TFLite Micro.");
+
   micro_context->DeallocateTempTfLiteTensor(input);
   micro_context->DeallocateTempTfLiteTensor(filter);
-  micro_context->DeallocateTempTfLiteTensor(bias);
+  if (bias != nullptr) {
+    micro_context->DeallocateTempTfLiteTensor(bias);
+  }
   micro_context->DeallocateTempTfLiteTensor(output);
 
   TF_LITE_ENSURE_TYPES_EQ(context, input->type, output->type);
@@ -145,10 +160,8 @@ TfLiteStatus HexagonFullyConnectedPrepare(TfLiteContext* context, TfLiteNode* no
 
   if (tflite::hexagon_fully_connected::HexagonOptimizable(context, node)) {
     return tflite::hexagon_fully_connected::HexagonPrepare(context, node);
-  } else {
-    return CalculateOpDataFullyConnected(context, params->activation, input->type, input,
-                           filter, bias, output, &data->reference_op_data);
   }
+  return kTfLiteOk;
 }
 
 TfLiteStatus HexagonFullyConnectedEvalInt8(TfLiteContext* context, TfLiteNode* node) {
@@ -170,7 +183,9 @@ TfLiteStatus HexagonFullyConnectedEvalInt8(TfLiteContext* context, TfLiteNode* n
   // This kernel only implements the int8 version of the fully_connected kernel.
   TFLITE_DCHECK(input->type == kTfLiteInt8);
   TFLITE_DCHECK(filter->type == kTfLiteInt8);
-  TFLITE_DCHECK(bias->type == kTfLiteInt32);
+  if (bias != nullptr) {
+    TFLITE_DCHECK(bias->type == kTfLiteInt32);
+  }
   TFLITE_DCHECK(output->type == kTfLiteInt8);
 
   if (tflite::hexagon_fully_connected::HexagonOptimizable(context, node)) {