fixed issue with cudaFree

Author: alphaville

include/tensor.cuh

@@ -177,6 +177,8 @@ public:
      * @return Total allocated bytes
      */
     size_t totalAllocatedBytes() const { return bytesAllocated; }
+
+    void incrementAllocatedBytes(size_t s) { bytesAllocated += s; }
 };
 
 
@@ -216,10 +218,12 @@ private:
         if (m_doDestroyData) {
             if (m_d_data) gpuErrChk(cudaFree(m_d_data));
             m_d_data = nullptr;
+            Session::getInstance().incrementAllocatedBytes(-m_numRows * m_numCols * m_numMats * sizeof(T));
         }
         if (m_doDestroyPtrMatrices) {
             if (m_d_ptrMatrices) gpuErrChk(cudaFree(m_d_ptrMatrices));
             m_d_ptrMatrices = nullptr;
+            Session::getInstance().incrementAllocatedBytes(-m_numMats * sizeof(T *));
         }
     }
 

main.cu

@@ -4,20 +4,26 @@
 #include <string>
 #include <vector>
 
-
-int main() {
+void xyz() {
     /* Write to binary file */
     auto r = DTensor<double>::createRandomTensor(3, 6, 4, -1, 1);
     auto r2 = DTensor<double>::createRandomTensor(300, 600, 4, -1, 1);
     std::string fName = "tensor.bt"; // binary tensor file extension: .bt
-    r.saveToFile(fName);
 
     /* Parse binary file */
     auto recov = DTensor<double>::parseFromFile(fName);
     auto err = r - recov;
     std::cout << "max error : " << err.maxAbs() << std::endl;
     std::cout << "Memory: " << std::setprecision(3)
             << (float) Session::getInstance().totalAllocatedBytes() / 1e6
-    << " MB" << std::endl;
+            << " MB" << std::endl;
+}
+
+
+int main() {
+    xyz();
+    std::cout << "Memory (outside): " << std::setprecision(3)
+            << (float) Session::getInstance().totalAllocatedBytes() / 1e6
+            << " MB" << std::endl;
     return 0;
 }

test/testTensor.cu

@@ -950,16 +950,16 @@ TEST_F(TensorTest, tensorTranspose) {
 }
 
 /* ---------------------------------------
- * Tensor: transpose
+ * Tensor: total bytes allocated
  * --------------------------------------- */
 TEMPLATE_WITH_TYPE_T
 void tensorBytesAllocated() {
-    size_t previouslyAllocatedBytes = Session::getInstance().totalAllocatedBytes();
+    const size_t previouslyAllocatedBytes = Session::getInstance().totalAllocatedBytes();
     size_t m = 10, n = 10, k = 20;
     DTensor<T> A(m, n, k);
-    size_t allocatedBytes = Session::getInstance().totalAllocatedBytes();
-    size_t currentAllocatedBytes = m * n * k * sizeof(T) + k * sizeof(T *);
-    size_t expectedBytes = currentAllocatedBytes + previouslyAllocatedBytes;
+    const size_t allocatedBytes = Session::getInstance().totalAllocatedBytes();
+    const size_t currentAllocatedBytes = m * n * k * sizeof(T) + k * sizeof(T *);
+    const size_t expectedBytes = currentAllocatedBytes + previouslyAllocatedBytes;
     EXPECT_EQ(expectedBytes, allocatedBytes);
 }
 
@@ -969,6 +969,30 @@ TEST_F(TensorTest, tensorBytesAllocated) {
     tensorBytesAllocated<int>();
 }
 
+
+/* ---------------------------------------
+ * Tensor: total bytes allocated
+ * --------------------------------------- */
+TEMPLATE_WITH_TYPE_T
+void tensorBytesAllocatedDeallocated() {
+    const size_t previouslyAllocatedBytes = Session::getInstance().totalAllocatedBytes();
+    size_t m = 10, n = 10, k = 20;
+    auto *A = new DTensor<T>(m, n, k); // new allocation (increments session)
+    const size_t allocatedBytes = Session::getInstance().totalAllocatedBytes();
+    const size_t expectedBytes = previouslyAllocatedBytes + m * n * k * sizeof(T) + k * sizeof(T *);
+    EXPECT_EQ(expectedBytes, allocatedBytes);
+    delete A;
+    EXPECT_EQ(previouslyAllocatedBytes, Session::getInstance().totalAllocatedBytes());
+}
+
+TEST_F(TensorTest, tensorBytesAllocatedDeallocated) {
+    tensorBytesAllocatedDeallocated<float>();
+    tensorBytesAllocatedDeallocated<double>();
+    tensorBytesAllocatedDeallocated<int>();
+}
+
+
+
 /* ================================================================================================
  *  LEAST SQUARES TESTS
  * ================================================================================================ */