[NNOP unittest] Use TensorWrapper for FC (#651)

Use TensorWrapper for FC unit tests instead of directly access primitives.

Signed-off-by: Junghyun Kim <jh0822.kim@samsung.com>

diff --git a/src/kernel/acl/src/cl/FullyConnected.test.cpp b/src/kernel/acl/src/cl/FullyConnected.test.cpp
index b3c2f92..fc9678c 100644 (file)
--- a/src/kernel/acl/src/cl/FullyConnected.test.cpp
+++ b/src/kernel/acl/src/cl/FullyConnected.test.cpp
@@ -9,73 +9,77 @@
 using namespace nnfw::kernel::acl;
 
 TEST(KernelACL_TC, fcFloat32_1) {
-  const android::nn::Shape inputShape = {OperandType::FLOAT32, {1, 1, 1, 100}, 1.0, 0};
-  float inputData[100];
 
-  const android::nn::Shape weightsShape = {OperandType::FLOAT32, {1, 1, 100, 1}, 1.0, 0};
-  float weightsData[100];
-
-  const android::nn::Shape biasShape = {OperandType::FLOAT32, {1, 1, 1, 1}, 1.0, 0};
-  float biasData[1];
-
-  const android::nn::Shape outputShape = { OperandType::FLOAT32, {1,1,1,1}, 1.0, 0 };
-  float outputData[1];
+  util::TensorWrapper input({1,1,1,100});
+  util::TensorWrapper weights({1,1,100,1});
+  util::TensorWrapper bias({1,1,1,1});
+  util::TensorWrapper output({1,1,1,1});
 
   int32_t activation = static_cast<int32_t>(FusedActivationFunc::RELU);
 
-  util::initData(inputData, sizeof(inputData) / sizeof(inputData[0]), 1.0);
-  util::initData(weightsData, sizeof(weightsData) / sizeof(weightsData[0]), 1.0);
-  util::initData(biasData, sizeof(biasData) / sizeof(biasData[0]), 0.0);
-  util::initData(outputData, sizeof(outputData) / sizeof(outputData[0]), 0.0);
-
-  bool bret = fullyConnectedFloat32(inputData, inputShape,
-      weightsData, weightsShape,
-      biasData, biasShape,
+  input.initValue([](uint32_t n, uint32_t c, uint32_t h, uint32_t w) {
+    return 1.f;
+  });
+  weights.initValue([](uint32_t n, uint32_t c, uint32_t h, uint32_t w) {
+    return 1.f;
+  });
+  bias.initValue([](uint32_t n, uint32_t c, uint32_t h, uint32_t w) {
+    return 0.f;
+  });
+  output.initValue([](uint32_t n, uint32_t c, uint32_t h, uint32_t w) {
+    return 0.f;
+  });
+
+  bool bret = fullyConnectedFloat32(input.ptr<float>(), input.shape(),
+      weights.ptr<float>(), weights.shape(),
+      bias.ptr<float>(), bias.shape(),
       activation,
-      outputData, outputShape);
+      output.ptr<float>(), output.shape());
 
   EXPECT_EQ(bret, true);
 
-  float expectData[] = {
-    100.0f
-  };
-  bret = util::compareData(outputData, expectData, outputShape);
-  EXPECT_EQ(bret, true);
+  util::TensorWrapper expected({1,1,1,1});
+  expected.initValue([](uint32_t n, uint32_t c, uint32_t h, uint32_t w) {
+    return 100.f;
+  });
 
+  EXPECT_EQ(output, expected);
 }
 
 TEST(KernelACL_TC, fcFloat32_relu) {
-  const android::nn::Shape inputShape = {OperandType::FLOAT32, {1, 1, 1, 100}, 1.0, 0};
-  float inputData[100];
-
-  const android::nn::Shape weightsShape = {OperandType::FLOAT32, {1, 1, 100, 1}, 1.0, 0};
-  float weightsData[100];
 
-  const android::nn::Shape biasShape = {OperandType::FLOAT32, {1, 1, 1, 1}, 1.0, 0};
-  float biasData[1];
+  util::TensorWrapper input({1,1,1,100});
+  util::TensorWrapper weights({1,1,100,1});
+  util::TensorWrapper bias({1,1,1,1});
+  util::TensorWrapper output({1,1,1,1});
 
-  const android::nn::Shape outputShape = { OperandType::FLOAT32, {1,1,1,1}, 1.0, 0 };
-  float outputData[1];
-
-  int32_t activation = ANEURALNETWORKS_FUSED_RELU;
-
-  util::initData(inputData, sizeof(inputData) / sizeof(inputData[0]), 1.0);
-  util::initData(weightsData, sizeof(weightsData) / sizeof(weightsData[0]), -1.0);
-  util::initData(biasData, sizeof(biasData) / sizeof(biasData[0]), 0.0);
-  util::initData(outputData, sizeof(outputData) / sizeof(outputData[0]), 0.0);
+  int32_t activation = static_cast<int32_t>(FusedActivationFunc::RELU);
 
-  bool bret = fullyConnectedFloat32(inputData, inputShape,
-      weightsData, weightsShape,
-      biasData, biasShape,
+  input.initValue([](uint32_t n, uint32_t c, uint32_t h, uint32_t w) {
+    return 1.f;
+  });
+  weights.initValue([](uint32_t n, uint32_t c, uint32_t h, uint32_t w) {
+    return -1.f;
+  });
+  bias.initValue([](uint32_t n, uint32_t c, uint32_t h, uint32_t w) {
+    return 0.f;
+  });
+  output.initValue([](uint32_t n, uint32_t c, uint32_t h, uint32_t w) {
+    return 0.f;
+  });
+
+  bool bret = fullyConnectedFloat32(input.ptr<float>(), input.shape(),
+      weights.ptr<float>(), weights.shape(),
+      bias.ptr<float>(), bias.shape(),
       activation,
-      outputData, outputShape);
+      output.ptr<float>(), output.shape());
 
   EXPECT_EQ(bret, true);
 
-  float expectData[] = {
-    0.0f
-  };
-  bret = util::compareData(outputData, expectData, outputShape);
-  EXPECT_EQ(bret, true);
+  util::TensorWrapper expected({1,1,1,1});
+  expected.initValue([](uint32_t n, uint32_t c, uint32_t h, uint32_t w) {
+    return 0.f;
+  });
 
+  EXPECT_EQ(output, expected);
 }