partial work of resnet

.travis.yml

@@ -15,6 +15,7 @@ cache:
 
 before_script:
   - scripts/download_squeezenet.sh
+  - scripts/download_resnet.sh
 
 script:
   - sbt test

scripts/download_squeezenet.sh

@@ -10,4 +10,3 @@ then
 	tar xzf $HOME/tmp/squeezenet.tar.gz -C $HOME/onnx_models
 	rm -f $HOME/tmp/squeezenet.tar.gz
 fi
-

src/test/scala/lantern/test_onnx.scala

@@ -23,12 +23,24 @@ import java.nio.FloatBuffer;
 
 class ONNXTest extends FunSuite {
 
-  val model_file = s"""${sys.env("HOME")}/onnx_models/squeezenet/model.onnx"""
-  val model_dir = s"""${sys.env("HOME")}/onnx_models/squeezenet/"""
+  val model_file_all = (name: String) => s"""${sys.env("HOME")}/onnx_models/$name/model.onnx"""
+  val model_dir_all = (name: String) => s"""${sys.env("HOME")}/onnx_models/$name/"""
   val gene_dir = "/tmp/"
+  def runTest(snippet: DslDriverC[String, Unit]) = {
+    val test = new PrintWriter(new File("/tmp/snippet.cpp"))
+    test.println(snippet.code)
+    test.flush()
+    new java.io.File("/tmp/snippet").delete
+    import scala.sys.process._
+    System.out.println("Compile C++ code")
+    (s"g++ -std=c++11 -O1 /tmp/snippet.cpp -o /tmp/snippet": ProcessBuilder).lines.foreach(System.out.println)
+    System.out.println("Run C++ code")
+    (s"/tmp/snippet a": ProcessBuilder).lines.foreach(System.out.println)
+  }
 
   test("onnx_reading_basic") {
 
+    val model_file = model_file_all("squeezenet")
     System.out.println(s"testing reading onnx models from $model_file")
     val model = onnx_ml.ModelProto.parseFrom(new FileInputStream(model_file))
 
@@ -215,6 +227,7 @@ class ONNXTest extends FunSuite {
       @virtualize
       def snippet(a: Rep[String]): Rep[Unit] = {
 
+        val model_file = model_file_all("squeezenet")
         val model = onnx_ml.ModelProto.parseFrom(new FileInputStream(model_file))
         val graph = model.getGraph
 
@@ -484,16 +497,16 @@ class ONNXTest extends FunSuite {
       }
     }
 
-    println(s"testing reading ONNX models from $model_file")
-
     val squeezenet_file = new PrintWriter(new File(gene_dir + "squeezenet.cpp"))
     squeezenet_file.println(squeezenet.code)
     squeezenet_file.flush()
 
   }
 
-  test("inference") {
+  test("squeezenet_inference") {
 
+    val model_file = model_file_all("squeezenet")
+    val model_dir = model_dir_all("squeezenet")
     System.out.println(s"testing reading ONNX model using library from $model_file")
 
     val inference_func = new DslDriverC[String, Unit] with ONNXLib {
@@ -510,17 +523,20 @@ class ONNXTest extends FunSuite {
         val input = readTensor(input_file).tensor
         val output = readTensor(output_file).tensor
         val output1 = func(input)
+        Tensor.assertEqual(output, output1.resize(1, 1000, 1, 1))
       }
     }
 
     val squeezenet_file = new PrintWriter(new File(gene_dir + "squeezenet.cpp"))
     squeezenet_file.println(inference_func.code)
     squeezenet_file.flush()
-
+    runTest(inference_func)
   }
 
-  test("training") {
+  test("squeezenet_training") {
 
+    val model_file = model_file_all("squeezenet")
+    val model_dir = model_dir_all("squeezenet")
     System.out.println(s"testing reading ONNX model using library from $model_file for training")
 
     val training_func = new DslDriverC[String, Unit] with ONNXLib {
@@ -551,4 +567,32 @@ class ONNXTest extends FunSuite {
     squeezenet_file.println(training_func.code)
     squeezenet_file.flush()
   }
+
+  test("resnet_inference") {
+
+    val model_file = model_file_all("resnet")
+    val model_dir = model_dir_all("resnet")
+    System.out.println(s"testing reading ONNX model using library from $model_file")
+
+    val inference_func = new DslDriverC[String, Unit] with ONNXLib {
+
+      @virtualize
+      def snippet(a: Rep[String]): Rep[Unit] = {
+        val model = readONNX(model_file)
+        val (func, x_dims) = (model.inference_func, model.x_dims)
+
+        // get test data as TensorProto
+        val input_file = model_dir + "test_data_set_0/input_0.pb"
+        val output_file = model_dir + "test_data_set_0/output_0.pb"
+        val input = readTensor(input_file).tensor
+        val output = readTensor(output_file).tensor
+        val output1 = func(input)
+        Tensor.assertEqual(output, output1)
+      }
+    }
+    val resnet_file = new PrintWriter(new File(gene_dir + "resnet.cpp"))
+    resnet_file.println(inference_func.code)
+    resnet_file.flush()
+    runTest(inference_func)
+  }
 }