Take better care of heap allocation.

Author: larsgeb

01-MetalAdder/MetalAdder.cpp

@@ -29,6 +29,7 @@ MetalAdder::MetalAdder(MTL::Device *device)
 
     auto str = NS::String::string("add_arrays", NS::ASCIIStringEncoding);
     MTL::Function *addFunction = defaultLibrary->newFunction(str);
+    defaultLibrary->release();
 
     if (addFunction == nullptr)
     {
@@ -38,6 +39,7 @@ MetalAdder::MetalAdder(MTL::Device *device)
 
     // Create a compute pipeline state object.
     _mAddFunctionPSO = _mDevice->newComputePipelineState(addFunction, &error);
+    addFunction->release();
 
     if (_mAddFunctionPSO == nullptr)
     {
@@ -55,16 +57,16 @@ MetalAdder::MetalAdder(MTL::Device *device)
         return;
     }
 
+    // Allocate three buffers to hold our initial data and the result.
+    _mBufferA = _mDevice->newBuffer(bufferSize, MTL::ResourceStorageModeShared);
+    _mBufferB = _mDevice->newBuffer(bufferSize, MTL::ResourceStorageModeShared);
+    _mBufferResult = _mDevice->newBuffer(bufferSize, MTL::ResourceStorageModeShared);
+
     prepareData();
 }
 
 void MetalAdder::prepareData()
 {
-    // Allocate three buffers to hold our initial data and the result.
-
-    _mBufferA = _mDevice->newBuffer(bufferSize, MTL::ResourceStorageModeShared);
-    _mBufferB = _mDevice->newBuffer(bufferSize, MTL::ResourceStorageModeShared);
-    _mBufferResult = _mDevice->newBuffer(bufferSize, MTL::ResourceStorageModeShared);
 
     generateRandomFloatData(_mBufferA);
     generateRandomFloatData(_mBufferB);
@@ -143,3 +145,13 @@ void MetalAdder::verifyResults()
         }
     }
 }
+
+MetalAdder::~MetalAdder()
+{
+    _mBufferA->release();
+    _mBufferB->release();
+    _mBufferResult->release();
+
+    _mAddFunctionPSO->release();
+    _mCommandQueue->release();
+}

01-MetalAdder/MetalAdder.hpp

@@ -34,6 +34,8 @@ class MetalAdder
     MTL::Buffer *_mBufferResult;
 
     MetalAdder(MTL::Device *device);
+    ~MetalAdder();
+
     void prepareData();
     void sendComputeCommand();
     void verifyResults();

01-MetalAdder/main.cpp

@@ -53,14 +53,15 @@ int main(int argc, char *argv[])
               << std::endl;
 
     // Verify serial code --------------------------------------------------------------
-    // Let's randomize the data again, making sure that the result buffer starts out
-    // incorrect
-    adder->prepareData();
     // Get buffers pointers for CPU code. Using MTL::ResourceStorageModeShared should
     // make them accessible to both GPU and CPU, perfect!
     auto array_a = ((float *)adder->_mBufferA->contents());
     auto array_b = ((float *)adder->_mBufferB->contents());
     auto array_c = ((float *)adder->_mBufferResult->contents());
+
+    // Let's randomize the data again, making sure that the result buffer starts out
+    // incorrect
+    adder->prepareData();
     add_array_serial(array_a, array_b, array_c, arrayLength);
     adder->verifyResults();
 
@@ -85,9 +86,6 @@ int main(int argc, char *argv[])
         // Verify OpenMP code --------------------------------------------------------------
         omp_set_num_threads(threads);
         adder->prepareData();
-        array_a = ((float *)adder->_mBufferA->contents());
-        array_b = ((float *)adder->_mBufferB->contents());
-        array_c = ((float *)adder->_mBufferResult->contents());
         add_array_openmp(array_a, array_b, array_c, arrayLength);
         adder->verifyResults();
 
@@ -105,6 +103,10 @@ int main(int argc, char *argv[])
         std::cout << array_mean << unit_name << " \t +/- " << array_std << unit_name << std::endl
                   << std::endl;
     }
+
+    delete[] durations;
+    delete adder;
+    device->release();
 }
 
 template <class T>