ConcreteFunction fix and performance improvements (#364)

Signed-off-by: Ryan Nett <[email protected]>

Co-authored-by: Ryan Nett <[email protected]>

Author: karllessard

tensorflow-core/tensorflow-core-api/src/main/java/org/tensorflow/ConcreteFunction.java

@@ -19,10 +19,10 @@
 import static org.tensorflow.internal.c_api.global.tensorflow.TF_GraphToFunction;
 
 import java.util.ArrayList;
-import java.util.Arrays;
 import java.util.Collection;
 import java.util.Collections;
 import java.util.HashSet;
+import java.util.Iterator;
 import java.util.LinkedHashMap;
 import java.util.List;
 import java.util.Map;
@@ -66,7 +66,7 @@
  * Map<String, Tensor> outputTensorMap = myFunction.call(inputTensorMap);
  * }</pre>
  */
-public class ConcreteFunction implements AutoCloseable, TensorFunction {
+public final class ConcreteFunction implements AutoCloseable, TensorFunction {
 
   /**
    * Creates a function by building a new graph.
@@ -220,11 +220,11 @@ public String toString() {
   public Map<String, Operand<?>> call(Scope scope, Map<String, Operand<?>> arguments) {
     List<Operand<?>> inputList = new ArrayList<>(signature.inputNames().size());
 
-    for (String inputName : signature().inputNames()) {
+    for (String inputName : signature.inputNames()) {
       if (!arguments.containsKey(inputName)) {
         throw new IllegalArgumentException(
             "Function "
-                + signature().methodName()
+                + signature.methodName()
                 + " has parameter \""
                 + inputName
                 + "\", but no argument was passed for it.");
@@ -241,30 +241,30 @@ public Map<String, Operand<?>> call(Scope scope, Map<String, Operand<?>> argumen
     }
 
     List<Output<?>> outputList =
-        PartitionedCall.create(
-                scope,
-                inputList,
-                Arrays.stream(outputDtypes)
-                    .map(x -> TensorTypeRegistry.find(x).type())
-                    .collect(Collectors.toList()),
-                this)
-            .output();
-
-    Map<String, Operand<?>> namedOutputs = new LinkedHashMap<>(signature().outputNames().size());
-
-    List<String> outputNames = new ArrayList<>(signature().outputNames());
-    for (int i = 0; i < outputNames.size(); i++) {
-      String outputName = outputNames.get(i);
-
-      if (i > outputList.size()) {
-        throw new IllegalStateException(
-            "Somehow, not all required outputs were returned from the function");
-      }
+        PartitionedCall.create(scope, inputList, outputTypes, this).output();
 
+    if (signature.outputNames().size() == 0) {
+      return Collections.emptyMap();
+    }
+    if (signature.outputNames().size() == 1) {
+      return Collections.singletonMap(signature.outputNames().iterator().next(), outputList.get(0));
+    }
+    if (outputList.size() < signature.outputNames().size()) {
+      throw new IllegalStateException(
+          "Somehow, not all required outputs were returned from the function"
+              + "(expected: "
+              + signature.outputNames().size()
+              + ", returned: "
+              + outputList.size()
+              + ")");
+    }
+    Map<String, Operand<?>> namedOutputs = new LinkedHashMap<>(signature.outputNames().size());
+    Iterator<String> outputNames = signature.outputNames().iterator();
+    for (int i = 0; outputNames.hasNext(); i++) {
+      String outputName = outputNames.next();
       Operand<?> output = outputList.get(i);
       namedOutputs.put(outputName, output);
     }
-
     return Collections.unmodifiableMap(namedOutputs);
   }
 
@@ -291,10 +291,7 @@ public Operand<?> call(Scope scope, Operand<?> argument) {
     }
     String outputName = signatureDef.getOutputsMap().keySet().iterator().next();
 
-    Map<String, Operand<?>> inputMap = new LinkedHashMap<>();
-    inputMap.put(inputName, argument);
-
-    return call(scope, inputMap).get(outputName);
+    return call(scope, Collections.singletonMap(inputName, argument)).get(outputName);
   }
 
   @Override
@@ -395,8 +392,7 @@ static ConcreteFunction fromNativeHandle(
   private final NativeFunction nativeFunction;
   private final PointerScope scope;
   private final Set<TF_Function> dependencies;
-  private final DataType[] inputDtypes;
-  private final DataType[] outputDtypes;
+  private final List<Class<? extends TType>> outputTypes;
 
   /** All native functions should have deallocators registered */
   private ConcreteFunction(
@@ -405,7 +401,7 @@ private ConcreteFunction(
     this.nativeFunction = nativeFunction;
     this.dependencies = Collections.unmodifiableSet(dependencies);
 
-    if (this.signature.getInputs().size()
+    if (signature.getInputs().size()
         != nativeFunction.getFunctionDef().getSignature().getInputArgCount()) {
       throw new IllegalArgumentException(
           "Signature must have the same number of inputs as the native function.  Expected "
@@ -414,7 +410,7 @@ private ConcreteFunction(
               + this.signature.getInputs().size());
     }
 
-    if (this.signature.getOutputs().size()
+    if (signature.getOutputs().size()
         != nativeFunction.getFunctionDef().getSignature().getOutputArgCount()) {
       throw new IllegalArgumentException(
           "New signature must have the same number of outputs as the native function.  Expected "
@@ -423,10 +419,8 @@ private ConcreteFunction(
               + this.signature.getOutputs().size());
     }
 
-    inputDtypes =
-        this.signature.getInputs().values().stream().map(x -> x.dataType).toArray(DataType[]::new);
-
-    List<DataType> inputs = Arrays.asList(inputDtypes);
+    List<DataType> inputs =
+        signature.getInputs().values().stream().map(x -> x.dataType).collect(Collectors.toList());
     List<DataType> nativeInputs =
         nativeFunction.getFunctionDef().getSignature().getInputArgList().stream()
             .map(ArgDef::getType)
@@ -440,10 +434,8 @@ private ConcreteFunction(
               + inputs);
     }
 
-    outputDtypes =
-        signature().getOutputs().values().stream().map(x -> x.dataType).toArray(DataType[]::new);
-
-    List<DataType> outputs = Arrays.asList(outputDtypes);
+    List<DataType> outputs =
+        signature.getOutputs().values().stream().map(x -> x.dataType).collect(Collectors.toList());
     List<DataType> nativeOutputs =
         nativeFunction.getFunctionDef().getSignature().getOutputArgList().stream()
             .map(ArgDef::getType)
@@ -457,6 +449,9 @@ private ConcreteFunction(
               + outputs);
     }
 
+    outputTypes =
+        outputs.stream().map(x -> TensorTypeRegistry.find(x).type()).collect(Collectors.toList());
+
     try (PointerScope scope = new PointerScope()) {
       this.scope = scope;
       scope.extend();
@@ -469,6 +464,8 @@ private ConcreteFunction(
    * FIXME: This causes native errors when I use it (Linux GPU, 6.1 CC), but I'm leaving it because
    * how to enable XLA JIT is extremely non-obvious.
    *
+   * <p>See https://github.com/tensorflow/java/issues/347
+   *
    * <p>Causes {@code OP_REQUIRES failed at xla_ops.cc:363 : Not found: could not find registered
    * platform with id: 0x7f75af03e6e8} (it's a warning, but the resulting TF_Status fails).
    */

tensorflow-core/tensorflow-core-api/src/test/java/org/tensorflow/ConcreteFunctionTest.java

@@ -1,24 +1,26 @@
 /* Copyright 2020-2021 The TensorFlow Authors. All Rights Reserved.
 
- Licensed under the Apache License, Version 2.0 (the "License");
- you may not use this file except in compliance with the License.
- You may obtain a copy of the License at
-
-     http://www.apache.org/licenses/LICENSE-2.0
-
- Unless required by applicable law or agreed to in writing, software
- distributed under the License is distributed on an "AS IS" BASIS,
- WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- See the License for the specific language governing permissions and
- limitations under the License.
- =======================================================================
- */
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+=======================================================================
+*/
 package org.tensorflow;
 
 import static org.junit.jupiter.api.Assertions.assertEquals;
 import static org.junit.jupiter.api.Assertions.assertNotNull;
 
 import java.util.Arrays;
+import java.util.HashMap;
+import java.util.Map;
 import org.junit.jupiter.api.Test;
 import org.tensorflow.op.Ops;
 import org.tensorflow.op.core.Init;
@@ -27,6 +29,7 @@
 import org.tensorflow.op.math.Sub;
 import org.tensorflow.proto.framework.DataType;
 import org.tensorflow.types.TFloat32;
+import org.tensorflow.types.TInt32;
 
 public class ConcreteFunctionTest {
 
@@ -144,6 +147,28 @@ public void testNestedFunctionGraph() {
     }
   }
 
+  @Test
+  public void testFunctionWithTwoOutputs() {
+    ConcreteFunction cf =
+        ConcreteFunction.create(
+            tf -> {
+              Placeholder<TInt32> x = tf.placeholder(TInt32.class);
+              Operand<TInt32> dblX = tf.math.add(x, x);
+              Operand<TInt32> tripX = tf.math.add(x, dblX);
+              return Signature.builder()
+                  .input("x", x)
+                  .output("dbl", dblX)
+                  .output("trpl", tripX)
+                  .build();
+            });
+
+    Map<String, Tensor> inputs = new HashMap<>();
+    inputs.put("x", TInt32.scalarOf(2));
+    Map<String, Tensor> outputs = cf.call(inputs);
+    assertEquals(4, ((TInt32) outputs.get("dbl")).getInt());
+    assertEquals(6, ((TInt32) outputs.get("trpl")).getInt());
+  }
+
   private static Signature square(Ops tf) {
     Placeholder<TFloat32> input = tf.placeholder(TFloat32.class);
     Operand<TFloat32> output = tf.math.square(input);