update vulkan samples with SPIR-V shaders

Samples/simpleVulkan/Build_instructions.txt

@@ -19,8 +19,17 @@ For Linux:
 -- Install "libxcb1-dev" and "xorg-dev" as GLFW3 is depended on it
 -- Add Vulkan and GLFW3 libraries directories to LD_LIBRARY_PATH
 
+
 For Linux aarch64(L4T):
 -- Install GLFW3 library using "sudo apt-get install libglfw3-dev" this will provide glfw3 
 -- install above will also provide libvulkan-dev as dependencies
 -- Add Vulkan and GLFW3 libraries directories to LD_LIBRARY_PATH
 -- Pass path to vulkan sdk while building 'make VULKAN_SDK_PATH=<PATH_TO_VULKAN_SDK>', VULKAN_SDK_PATH in this scenario is typically "/usr"
+
+
+For Shader changes:
+-- Update the sinewave.vert and/or sinewave.frag shader source file as required
+-- Use the glslc shader compiler from the installed Vulkan SDK's bin directory to compile shaders as:
+    glslc sinewave.vert -o vert.spv
+    glslc sinewave.frag -o frag.spv
+** Make sure to add glslc's path in your PATH environment variable **

Samples/simpleVulkan/main.cpp

@@ -92,9 +92,9 @@ class VulkanCudaSineWave : public VulkanBaseApp {
     }
     // Add our compiled vulkan shader files
     char *vertex_shader_path =
-        sdkFindFilePath("sinewave.vert", execution_path.c_str());
+        sdkFindFilePath("vert.spv", execution_path.c_str());
     char *fragment_shader_path =
-        sdkFindFilePath("sinewave.frag", execution_path.c_str());
+        sdkFindFilePath("frag.spv", execution_path.c_str());
     m_shaderFiles.push_back(
         std::make_pair(VK_SHADER_STAGE_VERTEX_BIT, vertex_shader_path));
     m_shaderFiles.push_back(

Samples/simpleVulkanMMAP/Build_instructions.txt

@@ -0,0 +1,35 @@
+For Windows:
+Follow these steps once you have installed Vulkan SDK for Windows from https://www.lunarg.com/vulkan-sdk/
+-- Install GLFW3 library at suitable location
+-- Open the simpleVulkan VS project file.
+To add the GLFW3 library path
+-- Right click on Project name "simpleVulkan" click on "Properties"
+-- In Property pages window go to Linker -> General. Here in "Additional Libraries Directories" edit and add path to glfw3dll.lib
+To add the GLFW3 headers path
+-- Right click on Project name "simpleVulkan" click on "Properties"
+-- In Property pages window go to "VC++ Directories" section. Here in "Include Directories" edit and add path to GLFW3 headers include directory location.
+** Make sure to add path to glfw3.dll in your PATH environment variable**
+
+
+For Linux:
+-- Install the Vulkan SDK from https://www.lunarg.com/vulkan-sdk/  and follow environment setup instructions.
+-- Install GLFW3 library through your OS package repository. For example: apt-get for Ubuntu and dnf for RHEL/CentOS. Below is for Ubuntu:
+    sudo apt-get install libglfw3
+    sudo apt-get install libglfw3-dev
+-- Install "libxcb1-dev" and "xorg-dev" as GLFW3 is depended on it
+-- Add Vulkan and GLFW3 libraries directories to LD_LIBRARY_PATH
+
+
+For Linux aarch64(L4T):
+-- Install GLFW3 library using "sudo apt-get install libglfw3-dev" this will provide glfw3 
+-- install above will also provide libvulkan-dev as dependencies
+-- Add Vulkan and GLFW3 libraries directories to LD_LIBRARY_PATH
+-- Pass path to vulkan sdk while building 'make VULKAN_SDK_PATH=<PATH_TO_VULKAN_SDK>', VULKAN_SDK_PATH in this scenario is typically "/usr"
+
+
+For Shader changes:
+-- Update the montecarlo.vert and/or montecarlo.frag shader source file as required
+-- Use the glslc shader compiler from the installed Vulkan SDK's bin directory to compile shaders as:
+    glslc montecarlo.vert -o vert.spv
+    glslc montecarlo.frag -o frag.spv
+** Make sure to add glslc's path in your PATH environment variable **

Samples/simpleVulkanMMAP/MonteCarloPi.cu

@@ -45,7 +45,8 @@
 // ipcHandleTypeFlag variable is a convenience variable and is passed by value
 // to individual requests.
 #if defined(__linux__)
-CUmemAllocationHandleType ipcHandleTypeFlag = CU_MEM_HANDLE_TYPE_POSIX_FILE_DESCRIPTOR;
+CUmemAllocationHandleType ipcHandleTypeFlag =
+    CU_MEM_HANDLE_TYPE_POSIX_FILE_DESCRIPTOR;
 #else
 CUmemAllocationHandleType ipcHandleTypeFlag = CU_MEM_HANDLE_TYPE_WIN32;
 #endif
@@ -78,8 +79,9 @@ void getDefaultSecurityDescriptor(CUmemAllocationProp *prop) {
 #endif
 }
 
-__global__ void monte_carlo_kernel(vec2 *xyVector, float *pointsInsideCircle, float *numPointsInCircle, unsigned int numPoints, float time)
+__global__ void monte_carlo_kernel(vec2 *xyVector, float *pointsInsideCircle,
-{
+                                   float *numPointsInCircle,
+                                   unsigned int numPoints, float time) {
   const size_t stride = gridDim.x * blockDim.x;
   size_t tid = blockIdx.x * blockDim.x + threadIdx.x;
   float count = 0.0f;
@@ -98,24 +100,22 @@ __global__ void monte_carlo_kernel(vec2 *xyVector, float *pointsInsideCircle, fl
     // Compute the distance of this point form the center(0, 0)
     float dist = sqrtf((x * x) + (y * y));
 
-        // If distance is less than the radius of the unit circle, the point lies in the circle.
+    // If distance is less than the radius of the unit circle, the point lies in
+    // the circle.
     pointsInsideCircle[tid] = (dist <= 1.0f);
     count += (dist <= 1.0f);
   }
   atomicAdd(numPointsInCircle, count);
 }
 
-MonteCarloPiSimulation::MonteCarloPiSimulation(size_t num_points) :
+MonteCarloPiSimulation::MonteCarloPiSimulation(size_t num_points)
-    m_xyVector(nullptr),
+    : m_xyVector(nullptr),
       m_pointsInsideCircle(nullptr),
       m_totalPointsInsideCircle(0),
       m_totalPointsSimulated(0),
-    m_numPoints(num_points)
+      m_numPoints(num_points) {}
-{
-}
 
-MonteCarloPiSimulation::~MonteCarloPiSimulation()
+MonteCarloPiSimulation::~MonteCarloPiSimulation() {
-{
   if (m_numPointsInCircle) {
     checkCudaErrors(cudaFree(m_numPointsInCircle));
     m_numPointsInCircle = nullptr;
@@ -128,65 +128,77 @@ MonteCarloPiSimulation::~MonteCarloPiSimulation()
   cleanupSimulationAllocations();
 }
 
-void MonteCarloPiSimulation::initSimulation(int cudaDevice, cudaStream_t stream)
+void MonteCarloPiSimulation::initSimulation(int cudaDevice,
-{
+                                            cudaStream_t stream) {
   m_cudaDevice = cudaDevice;
   getIdealExecutionConfiguration();
 
-    // Allocate a position buffer that contains random location of the points in XY cartesian plane.
+  // Allocate a position buffer that contains random location of the points in
-    // Allocate a bitmap buffer which holds information of whether a point in the position buffer is inside the unit circle or not.
+  // XY cartesian plane.
+  // Allocate a bitmap buffer which holds information of whether a point in the
+  // position buffer is inside the unit circle or not.
   setupSimulationAllocations();
 
-    checkCudaErrors(cudaMalloc((float **)&m_numPointsInCircle, sizeof(*m_numPointsInCircle)));
+  checkCudaErrors(
-    checkCudaErrors(cudaMallocHost((float **)&m_hostNumPointsInCircle, sizeof(*m_hostNumPointsInCircle)));
+      cudaMalloc((float **)&m_numPointsInCircle, sizeof(*m_numPointsInCircle)));
+  checkCudaErrors(cudaMallocHost((float **)&m_hostNumPointsInCircle,
+                                 sizeof(*m_hostNumPointsInCircle)));
 }
 
-void MonteCarloPiSimulation::stepSimulation(float time, cudaStream_t stream)
+void MonteCarloPiSimulation::stepSimulation(float time, cudaStream_t stream) {
-{
+  checkCudaErrors(cudaMemsetAsync(m_numPointsInCircle, 0,
+                                  sizeof(*m_numPointsInCircle), stream));
 
-    checkCudaErrors(cudaMemsetAsync(m_numPointsInCircle, 0, sizeof(*m_numPointsInCircle), stream));
+  monte_carlo_kernel<<<m_blocks, m_threads, 0, stream>>>(
-
+      m_xyVector, m_pointsInsideCircle, m_numPointsInCircle, m_numPoints, time);
-    monte_carlo_kernel << < m_blocks, m_threads, 0, stream >> > (m_xyVector, m_pointsInsideCircle, m_numPointsInCircle, m_numPoints, time);
   getLastCudaError("Failed to launch CUDA simulation");
 
-    checkCudaErrors(cudaMemcpyAsync(m_hostNumPointsInCircle, m_numPointsInCircle, sizeof(*m_numPointsInCircle), cudaMemcpyDeviceToHost, stream));
+  checkCudaErrors(cudaMemcpyAsync(m_hostNumPointsInCircle, m_numPointsInCircle,
+                                  sizeof(*m_numPointsInCircle),
+                                  cudaMemcpyDeviceToHost, stream));
 
   // Queue up a stream callback to compute and print the PI value.
-    checkCudaErrors(cudaLaunchHostFunc(stream, this->computePiCallback, (void *)this));
+  checkCudaErrors(
+      cudaLaunchHostFunc(stream, this->computePiCallback, (void *)this));
 }
 
-void MonteCarloPiSimulation::computePiCallback(void *args)
+void MonteCarloPiSimulation::computePiCallback(void *args) {
-{
   MonteCarloPiSimulation *cbData = (MonteCarloPiSimulation *)args;
   cbData->m_totalPointsInsideCircle += *(cbData->m_hostNumPointsInCircle);
   cbData->m_totalPointsSimulated += cbData->m_numPoints;
-    double piValue = 4.0 * ((double)cbData->m_totalPointsInsideCircle / (double)cbData->m_totalPointsSimulated);
+  double piValue = 4.0 * ((double)cbData->m_totalPointsInsideCircle /
-    printf("Approximate Pi value for %zd data points: %lf \n", cbData->m_totalPointsSimulated, piValue);
+                          (double)cbData->m_totalPointsSimulated);
+  printf("Approximate Pi value for %zd data points: %lf \n",
+         cbData->m_totalPointsSimulated, piValue);
 }
 
-void MonteCarloPiSimulation::getIdealExecutionConfiguration()
+void MonteCarloPiSimulation::getIdealExecutionConfiguration() {
-{
   int warpSize = 0;
   int multiProcessorCount = 0;
 
   checkCudaErrors(cudaSetDevice(m_cudaDevice));
-    checkCudaErrors(cudaDeviceGetAttribute(&warpSize, cudaDevAttrWarpSize, m_cudaDevice));
+  checkCudaErrors(
+      cudaDeviceGetAttribute(&warpSize, cudaDevAttrWarpSize, m_cudaDevice));
 
-    // We don't need large block sizes, since there's not much inter-thread communication
+  // We don't need large block sizes, since there's not much inter-thread
+  // communication
   m_threads = warpSize;
 
   // Use the occupancy calculator and fill the gpu as best as we can
-    checkCudaErrors(cudaOccupancyMaxActiveBlocksPerMultiprocessor(&m_blocks, monte_carlo_kernel, warpSize, 0));
+  checkCudaErrors(cudaOccupancyMaxActiveBlocksPerMultiprocessor(
+      &m_blocks, monte_carlo_kernel, warpSize, 0));
 
-    checkCudaErrors(cudaDeviceGetAttribute(&multiProcessorCount, cudaDevAttrMultiProcessorCount, m_cudaDevice));
+  checkCudaErrors(cudaDeviceGetAttribute(
+      &multiProcessorCount, cudaDevAttrMultiProcessorCount, m_cudaDevice));
   m_blocks *= multiProcessorCount;
 
-    // Go ahead and the clamp the blocks to the minimum needed for this height/width
+  // Go ahead and the clamp the blocks to the minimum needed for this
-    m_blocks = std::min(m_blocks, (int)((m_numPoints + m_threads - 1) / m_threads));
+  // height/width
+  m_blocks =
+      std::min(m_blocks, (int)((m_numPoints + m_threads - 1) / m_threads));
 }
 
-void MonteCarloPiSimulation::setupSimulationAllocations()
+void MonteCarloPiSimulation::setupSimulationAllocations() {
-{
   CUdeviceptr d_ptr = 0U;
   size_t granularity = 0;
   CUmemGenericAllocationHandle cudaPositionHandle, cudaInCircleHandle;
@@ -205,30 +217,39 @@ void MonteCarloPiSimulation::setupSimulationAllocations()
   getDefaultSecurityDescriptor(&allocProp);
 
   // Get the recommended granularity for m_cudaDevice.
-    checkCudaErrors(cuMemGetAllocationGranularity(&granularity, &allocProp, CU_MEM_ALLOC_GRANULARITY_RECOMMENDED));
+  checkCudaErrors(cuMemGetAllocationGranularity(
+      &granularity, &allocProp, CU_MEM_ALLOC_GRANULARITY_RECOMMENDED));
 
   size_t xyPositionVecSize = m_numPoints * sizeof(*m_xyVector);
   size_t inCircleVecSize = m_numPoints * sizeof(*m_pointsInsideCircle);
 
-    size_t xyPositionSize = ROUND_UP_TO_GRANULARITY(xyPositionVecSize, granularity);
+  size_t xyPositionSize =
+      ROUND_UP_TO_GRANULARITY(xyPositionVecSize, granularity);
   size_t inCircleSize = ROUND_UP_TO_GRANULARITY(inCircleVecSize, granularity);
   m_totalAllocationSize = (xyPositionSize + inCircleSize);
 
   // Reserve the required contiguous VA space for the allocations
-    checkCudaErrors(cuMemAddressReserve(&d_ptr, m_totalAllocationSize, granularity, 0U, 0));
+  checkCudaErrors(
+      cuMemAddressReserve(&d_ptr, m_totalAllocationSize, granularity, 0U, 0));
 
   // Create the allocations as a pinned allocation on this device.
-    // Create an allocation to store all the positions of points on the xy plane and a second
+  // Create an allocation to store all the positions of points on the xy plane
-    // allocation which stores information if the corresponding position is inside the unit circle or not.
+  // and a second allocation which stores information if the corresponding
-    checkCudaErrors(cuMemCreate(&cudaPositionHandle, xyPositionSize, &allocProp, 0));
+  // position is inside the unit circle or not.
-    checkCudaErrors(cuMemCreate(&cudaInCircleHandle, inCircleSize, &allocProp, 0));
+  checkCudaErrors(
+      cuMemCreate(&cudaPositionHandle, xyPositionSize, &allocProp, 0));
+  checkCudaErrors(
+      cuMemCreate(&cudaInCircleHandle, inCircleSize, &allocProp, 0));
 
   // Export the allocation to a platform-specific handle. The type of handle
   // requested here must match the requestedHandleTypes field in the prop
-    // structure passed to cuMemCreate. The handle obtained here will be passed to vulkan
+  // structure passed to cuMemCreate. The handle obtained here will be passed to
-    // to import the allocation.
+  // vulkan to import the allocation.
-    checkCudaErrors(cuMemExportToShareableHandle((void *)&m_posShareableHandle, cudaPositionHandle, ipcHandleTypeFlag, 0));
+  checkCudaErrors(cuMemExportToShareableHandle(
-    checkCudaErrors(cuMemExportToShareableHandle((void *)&m_inCircleShareableHandle, cudaInCircleHandle, ipcHandleTypeFlag, 0));
+      (void *)&m_posShareableHandle, cudaPositionHandle, ipcHandleTypeFlag, 0));
+  checkCudaErrors(
+      cuMemExportToShareableHandle((void *)&m_inCircleShareableHandle,
+                                   cudaInCircleHandle, ipcHandleTypeFlag, 0));
 
   CUdeviceptr va_position = d_ptr;
   CUdeviceptr va_InCircle = va_position + xyPositionSize;
@@ -236,12 +257,15 @@ void MonteCarloPiSimulation::setupSimulationAllocations()
   m_xyVector = (vec2 *)va_position;
 
   // Assign the chunk to the appropriate VA range
-    checkCudaErrors(cuMemMap(va_position, xyPositionSize, 0, cudaPositionHandle, 0));
+  checkCudaErrors(
-    checkCudaErrors(cuMemMap(va_InCircle, inCircleSize, 0, cudaInCircleHandle, 0));
+      cuMemMap(va_position, xyPositionSize, 0, cudaPositionHandle, 0));
+  checkCudaErrors(
+      cuMemMap(va_InCircle, inCircleSize, 0, cudaInCircleHandle, 0));
 
-    // Release the handles for the allocation. Since the allocation is currently mapped to a VA range
+  // Release the handles for the allocation. Since the allocation is currently
-    // with a previous call to cuMemMap the actual freeing of memory allocation will happen on an eventual call to
+  // mapped to a VA range with a previous call to cuMemMap the actual freeing of
-    // cuMemUnmap. Thus the allocation will be kept live until it is unmapped.
+  // memory allocation will happen on an eventual call to cuMemUnmap. Thus the
+  // allocation will be kept live until it is unmapped.
   checkCudaErrors(cuMemRelease(cudaPositionHandle));
   checkCudaErrors(cuMemRelease(cudaInCircleHandle));
 
@@ -250,12 +274,13 @@ void MonteCarloPiSimulation::setupSimulationAllocations()
   accessDescriptor.location.type = CU_MEM_LOCATION_TYPE_DEVICE;
   accessDescriptor.flags = CU_MEM_ACCESS_FLAGS_PROT_READWRITE;
 
-    // Apply the access descriptor to the whole VA range. Essentially enables Read-Write access to the range.
+  // Apply the access descriptor to the whole VA range. Essentially enables
-    checkCudaErrors(cuMemSetAccess(d_ptr, m_totalAllocationSize, &accessDescriptor, 1));
+  // Read-Write access to the range.
+  checkCudaErrors(
+      cuMemSetAccess(d_ptr, m_totalAllocationSize, &accessDescriptor, 1));
 }
 
-void MonteCarloPiSimulation::cleanupSimulationAllocations()
+void MonteCarloPiSimulation::cleanupSimulationAllocations() {
-{
   if (m_xyVector && m_pointsInsideCircle) {
     // Unmap the mapped virtual memory region
     // Since the handles to the mapped backing stores have already been released
@@ -267,7 +292,8 @@ void MonteCarloPiSimulation::cleanupSimulationAllocations()
     checkIpcErrors(ipcCloseShareableHandle(m_inCircleShareableHandle));
 
     // Free the virtual address region.
-        checkCudaErrors(cuMemAddressFree((CUdeviceptr)m_xyVector, m_totalAllocationSize));
+    checkCudaErrors(
+        cuMemAddressFree((CUdeviceptr)m_xyVector, m_totalAllocationSize));
 
     m_xyVector = nullptr;
     m_pointsInsideCircle = nullptr;

Samples/simpleVulkanMMAP/MonteCarloPi.h

@@ -39,33 +39,35 @@
 
 typedef float vec2[2];
 
-class MonteCarloPiSimulation
+class MonteCarloPiSimulation {
-{
   size_t m_numPoints;
 
-    // Pointers to Cuda allocated buffers which are imported and used by vulkan as vertex buffer
+  // Pointers to Cuda allocated buffers which are imported and used by vulkan as
+  // vertex buffer
   vec2 *m_xyVector;
   float *m_pointsInsideCircle;
 
-    // Pointers to device and host allocated memories storing number of points that are inside the unit circle
+  // Pointers to device and host allocated memories storing number of points
+  // that are inside the unit circle
   float *m_numPointsInCircle;
   float *m_hostNumPointsInCircle;
 
   int m_blocks, m_threads;
 
-    // Total size of allocations created by cuMemMap Apis. This size is the sum of sizes of
+  // Total size of allocations created by cuMemMap Apis. This size is the sum of
-    // m_xyVector and m_pointsInsideCircle buffers.
+  // sizes of m_xyVector and m_pointsInsideCircle buffers.
   size_t m_totalAllocationSize;
 
-    // Shareable Handles(a file descriptor on Linux and NT Handle on Windows), used for sharing cuda
+  // Shareable Handles(a file descriptor on Linux and NT Handle on Windows),
+  // used for sharing cuda
   // allocated memory with Vulkan
   ShareableHandle m_posShareableHandle, m_inCircleShareableHandle;
 
   // Cuda Device corresponding to the Vulkan Physical device
   int m_cudaDevice;
 
-    // Track and accumulate total points that have been simulated since start of the sample.
+  // Track and accumulate total points that have been simulated since start of
-    // The idea is to get a closer approximation to PI with time.
+  // the sample. The idea is to get a closer approximation to PI with time.
   size_t m_totalPointsInsideCircle;
   size_t m_totalPointsSimulated;
 
@@ -80,21 +82,14 @@ public:
   void stepSimulation(float time, cudaStream_t stream = 0);
   static void computePiCallback(void *args);
 
-    size_t getNumPoints() const {
+  size_t getNumPoints() const { return m_numPoints; }
-        return m_numPoints;
-    }
 
-    float getNumPointsInCircle() const {
+  float getNumPointsInCircle() const { return *m_hostNumPointsInCircle; }
-        return *m_hostNumPointsInCircle;
-    }
 
-    ShareableHandle &getPositionShareableHandle() {
+  ShareableHandle &getPositionShareableHandle() { return m_posShareableHandle; }
-        return m_posShareableHandle;
-    }
   ShareableHandle &getInCircleShareableHandle() {
     return m_inCircleShareableHandle;
   }
-
 };
 
 #endif  // __PISIM_H__

Samples/simpleVulkanMMAP/VulkanBaseApp.h

@@ -40,25 +40,37 @@
 
 struct GLFWwindow;
 
-class VulkanBaseApp
+class VulkanBaseApp {
-{
  public:
   VulkanBaseApp(const std::string& appName, bool enableValidation = false);
   static VkExternalSemaphoreHandleTypeFlagBits getDefaultSemaphoreHandleType();
   static VkExternalMemoryHandleTypeFlagBits getDefaultMemHandleType();
   virtual ~VulkanBaseApp();
   void init();
-    void *getMemHandle(VkDeviceMemory memory, VkExternalMemoryHandleTypeFlagBits handleType);
+  void* getMemHandle(VkDeviceMemory memory,
-    void *getSemaphoreHandle(VkSemaphore semaphore, VkExternalSemaphoreHandleTypeFlagBits handleType);
+                     VkExternalMemoryHandleTypeFlagBits handleType);
+  void* getSemaphoreHandle(VkSemaphore semaphore,
+                           VkExternalSemaphoreHandleTypeFlagBits handleType);
   bool isVkPhysicalDeviceUuid(void* Uuid);
-    void createExternalSemaphore(VkSemaphore& semaphore, VkExternalSemaphoreHandleTypeFlagBits handleType);
+  void createExternalSemaphore(
-    void createBuffer(VkDeviceSize size, VkBufferUsageFlags usage, VkMemoryPropertyFlags properties, VkBuffer& buffer, VkDeviceMemory& bufferMemory);
+      VkSemaphore& semaphore, VkExternalSemaphoreHandleTypeFlagBits handleType);
-    void createExternalBuffer(VkDeviceSize size, VkBufferUsageFlags usage, VkMemoryPropertyFlags properties, VkExternalMemoryHandleTypeFlagsKHR extMemHandleType, VkBuffer& buffer, VkDeviceMemory& bufferMemory);
+  void createBuffer(VkDeviceSize size, VkBufferUsageFlags usage,
-    void importExternalBuffer(void *handle, VkExternalMemoryHandleTypeFlagBits handleType, size_t size, VkBufferUsageFlags usage, VkMemoryPropertyFlags properties, VkBuffer& buffer, VkDeviceMemory& memory);
+                    VkMemoryPropertyFlags properties, VkBuffer& buffer,
+                    VkDeviceMemory& bufferMemory);
+  void createExternalBuffer(VkDeviceSize size, VkBufferUsageFlags usage,
+                            VkMemoryPropertyFlags properties,
+                            VkExternalMemoryHandleTypeFlagsKHR extMemHandleType,
+                            VkBuffer& buffer, VkDeviceMemory& bufferMemory);
+  void importExternalBuffer(void* handle,
+                            VkExternalMemoryHandleTypeFlagBits handleType,
+                            size_t size, VkBufferUsageFlags usage,
+                            VkMemoryPropertyFlags properties, VkBuffer& buffer,
+                            VkDeviceMemory& memory);
   void copyBuffer(VkBuffer dst, VkBuffer src, VkDeviceSize size);
   VkCommandBuffer beginSingleTimeCommands();
   void endSingleTimeCommands(VkCommandBuffer commandBuffer);
   void mainLoop();
+
  protected:
   const std::string m_appName;
   const bool m_enableValidation;
@@ -101,13 +113,19 @@ protected:
   virtual void fillRenderingCommandBuffer(VkCommandBuffer& buffer) {}
   virtual std::vector<const char*> getRequiredExtensions() const;
   virtual std::vector<const char*> getRequiredDeviceExtensions() const;
-    virtual void getVertexDescriptions(std::vector<VkVertexInputBindingDescription>& bindingDesc, std::vector<VkVertexInputAttributeDescription>& attribDesc);
+  virtual void getVertexDescriptions(
-    virtual void getAssemblyStateInfo(VkPipelineInputAssemblyStateCreateInfo& info);
+      std::vector<VkVertexInputBindingDescription>& bindingDesc,
-    virtual void getWaitFrameSemaphores(std::vector<VkSemaphore>& wait, std::vector< VkPipelineStageFlags>& waitStages) const;
+      std::vector<VkVertexInputAttributeDescription>& attribDesc);
+  virtual void getAssemblyStateInfo(
+      VkPipelineInputAssemblyStateCreateInfo& info);
+  virtual void getWaitFrameSemaphores(
+      std::vector<VkSemaphore>& wait,
+      std::vector<VkPipelineStageFlags>& waitStages) const;
   virtual void getSignalFrameSemaphores(std::vector<VkSemaphore>& signal) const;
   virtual VkDeviceSize getUniformSize() const;
   virtual void updateUniformBuffer(uint32_t imageIndex, size_t globalFrame);
   virtual void drawFrame();
+
  private:
   GLFWwindow* m_window;
 

Samples/simpleVulkanMMAP/VulkanCudaInterop.h

@@ -35,7 +35,6 @@
 #include <helper_cuda.h>
 
 bool isDeviceCompatible(void *Uuid, size_t size) {
-
   int cudaDevice = cudaInvalidDeviceId;
   int deviceCount;
   checkCudaErrors(cudaGetDeviceCount(&deviceCount));
@@ -56,20 +55,28 @@ bool isDeviceCompatible(void *Uuid, size_t size) {
   int attributeVal = 0;
   int deviceComputeMode = 0;
 
-    checkCudaErrors(cuDeviceGetAttribute(&deviceComputeMode, CU_DEVICE_ATTRIBUTE_COMPUTE_MODE, cudaDevice));
+  checkCudaErrors(cuDeviceGetAttribute(
-    checkCudaErrors(cuDeviceGetAttribute(&attributeVal, CU_DEVICE_ATTRIBUTE_VIRTUAL_ADDRESS_MANAGEMENT_SUPPORTED, cudaDevice));
+      &deviceComputeMode, CU_DEVICE_ATTRIBUTE_COMPUTE_MODE, cudaDevice));
+  checkCudaErrors(cuDeviceGetAttribute(
+      &attributeVal, CU_DEVICE_ATTRIBUTE_VIRTUAL_ADDRESS_MANAGEMENT_SUPPORTED,
+      cudaDevice));
 
 #if defined(__linux__)
-    checkCudaErrors(cuDeviceGetAttribute(&deviceSupportsHandle, CU_DEVICE_ATTRIBUTE_HANDLE_TYPE_POSIX_FILE_DESCRIPTOR_SUPPORTED, cudaDevice));
+  checkCudaErrors(cuDeviceGetAttribute(
+      &deviceSupportsHandle,
+      CU_DEVICE_ATTRIBUTE_HANDLE_TYPE_POSIX_FILE_DESCRIPTOR_SUPPORTED,
+      cudaDevice));
 #else
-    checkCudaErrors(cuDeviceGetAttribute(&deviceSupportsHandle, CU_DEVICE_ATTRIBUTE_HANDLE_TYPE_WIN32_HANDLE_SUPPORTED, cudaDevice));
+  checkCudaErrors(cuDeviceGetAttribute(
+      &deviceSupportsHandle,
+      CU_DEVICE_ATTRIBUTE_HANDLE_TYPE_WIN32_HANDLE_SUPPORTED, cudaDevice));
 #endif
 
-    if ((deviceComputeMode != CU_COMPUTEMODE_DEFAULT) || !attributeVal || !deviceSupportsHandle) {
+  if ((deviceComputeMode != CU_COMPUTEMODE_DEFAULT) || !attributeVal ||
+      !deviceSupportsHandle) {
     return false;
   }
   return true;
 }
 
 #endif  // __VKCUDA_H__
-

Samples/simpleVulkanMMAP/main.cpp

@@ -27,8 +27,9 @@
 
 /*
  * This sample demonstrates CUDA Interop with Vulkan using cuMemMap APIs.
-  * Allocating device memory and updating values in those allocations are performed by CUDA
+ * Allocating device memory and updating values in those allocations are
-  * and the contents of the allocation are visualized by Vulkan.
+ * performed by CUDA and the contents of the allocation are visualized by
+ * Vulkan.
  */
 
 #include "VulkanBaseApp.h"
@@ -55,11 +56,8 @@
 
 std::string execution_path;
 
-class VulkanCudaPi : public VulkanBaseApp
+class VulkanCudaPi : public VulkanBaseApp {
-{
+  typedef struct UniformBufferObject_st { float frame; } UniformBufferObject;
-    typedef struct UniformBufferObject_st {
-        float frame;
-    } UniformBufferObject;
 
   VkBuffer m_inCircleBuffer, m_xyPositionBuffer;
   VkDeviceMemory m_inCircleMemory, m_xyPositionMemory;
@@ -71,9 +69,10 @@ class VulkanCudaPi : public VulkanBaseApp
   using chrono_tp = std::chrono::time_point<std::chrono::high_resolution_clock>;
   chrono_tp m_lastTime;
   size_t m_lastFrame;
+
  public:
-    VulkanCudaPi(size_t num_points) :
+  VulkanCudaPi(size_t num_points)
-        VulkanBaseApp("simpleVulkanMMAP", ENABLE_VALIDATION),
+      : VulkanBaseApp("simpleVulkanMMAP", ENABLE_VALIDATION),
         m_inCircleBuffer(VK_NULL_HANDLE),
         m_xyPositionBuffer(VK_NULL_HANDLE),
         m_inCircleMemory(VK_NULL_HANDLE),
@@ -86,12 +85,15 @@ public:
         m_cudaWaitSemaphore(),
         m_cudaSignalSemaphore(),
         m_lastFrame(0) {
-
     // Add our compiled vulkan shader files
-        char* vertex_shader_path = sdkFindFilePath("montecarlo.vert", execution_path.c_str());
+    char* vertex_shader_path =
-        char* fragment_shader_path = sdkFindFilePath("montecarlo.frag", execution_path.c_str());
+        sdkFindFilePath("vert.spv", execution_path.c_str());
-        m_shaderFiles.push_back(std::make_pair(VK_SHADER_STAGE_VERTEX_BIT, vertex_shader_path));
+    char* fragment_shader_path =
-        m_shaderFiles.push_back(std::make_pair(VK_SHADER_STAGE_FRAGMENT_BIT, fragment_shader_path));
+        sdkFindFilePath("frag.spv", execution_path.c_str());
+    m_shaderFiles.push_back(
+        std::make_pair(VK_SHADER_STAGE_VERTEX_BIT, vertex_shader_path));
+    m_shaderFiles.push_back(
+        std::make_pair(VK_SHADER_STAGE_FRAGMENT_BIT, fragment_shader_path));
   }
 
   ~VulkanCudaPi() {
@@ -126,11 +128,15 @@ public:
   void fillRenderingCommandBuffer(VkCommandBuffer& commandBuffer) {
     VkBuffer vertexBuffers[] = {m_inCircleBuffer, m_xyPositionBuffer};
     VkDeviceSize offsets[] = {0, 0};
-        vkCmdBindVertexBuffers(commandBuffer, 0, sizeof(vertexBuffers) / sizeof(vertexBuffers[0]), vertexBuffers, offsets);
+    vkCmdBindVertexBuffers(commandBuffer, 0,
+                           sizeof(vertexBuffers) / sizeof(vertexBuffers[0]),
+                           vertexBuffers, offsets);
     vkCmdDraw(commandBuffer, (uint32_t)(m_sim.getNumPoints()), 1, 0, 0);
   }
 
-    void getVertexDescriptions(std::vector<VkVertexInputBindingDescription>& bindingDesc, std::vector<VkVertexInputAttributeDescription>& attribDesc) {
+  void getVertexDescriptions(
+      std::vector<VkVertexInputBindingDescription>& bindingDesc,
+      std::vector<VkVertexInputAttributeDescription>& attribDesc) {
     bindingDesc.resize(2);
     attribDesc.resize(2);
 
@@ -159,25 +165,32 @@ public:
     info.primitiveRestartEnable = VK_FALSE;
   }
 
-    void getWaitFrameSemaphores(std::vector<VkSemaphore>& wait, std::vector< VkPipelineStageFlags>& waitStages) const {
+  void getWaitFrameSemaphores(
+      std::vector<VkSemaphore>& wait,
+      std::vector<VkPipelineStageFlags>& waitStages) const {
     if (m_currentFrame != 0) {
-            // Have vulkan wait until cuda is done with the vertex buffer before rendering
+      // Have vulkan wait until cuda is done with the vertex buffer before
-            // We don't do this on the first frame, as the wait semaphore hasn't been initialized yet
+      // rendering
+      // We don't do this on the first frame, as the wait semaphore hasn't been
+      // initialized yet
       wait.push_back(m_vkWaitSemaphore);
-            // We want to wait until all the pipeline commands are complete before letting cuda work
+      // We want to wait until all the pipeline commands are complete before
+      // letting cuda work
       waitStages.push_back(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT);
     }
   }
 
   void getSignalFrameSemaphores(std::vector<VkSemaphore>& signal) const {
-        // Add this semaphore for vulkan to signal once the vertex buffer is ready for cuda to modify
+    // Add this semaphore for vulkan to signal once the vertex buffer is ready
+    // for cuda to modify
     signal.push_back(m_vkSignalSemaphore);
   }
 
   void initVulkanApp() {
     const size_t nVerts = m_sim.getNumPoints();
 
-        // Obtain cuda device id for the device corresponding to the Vulkan physical device
+    // Obtain cuda device id for the device corresponding to the Vulkan physical
+    // device
     int deviceCount;
     int cudaDevice = cudaInvalidDeviceId;
     checkCudaErrors(cudaGetDeviceCount(&deviceCount));
@@ -195,62 +208,78 @@ public:
 
     // On the corresponding cuda device, create the cuda stream we'll using
     checkCudaErrors(cudaSetDevice(cudaDevice));
-        checkCudaErrors(cudaStreamCreateWithFlags(&m_stream, cudaStreamNonBlocking));
+    checkCudaErrors(
+        cudaStreamCreateWithFlags(&m_stream, cudaStreamNonBlocking));
     m_sim.initSimulation(cudaDevice, m_stream);
 
-        importExternalBuffer((void *)(uintptr_t)m_sim.getPositionShareableHandle(), getDefaultMemHandleType(), nVerts * sizeof(vec2),
+    importExternalBuffer(
+        (void*)(uintptr_t)m_sim.getPositionShareableHandle(),
+        getDefaultMemHandleType(), nVerts * sizeof(vec2),
         VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_VERTEX_BUFFER_BIT,
-            VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT, m_xyPositionBuffer, m_xyPositionMemory);
+        VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT, m_xyPositionBuffer,
+        m_xyPositionMemory);
 
-        importExternalBuffer((void *)(uintptr_t)m_sim.getInCircleShareableHandle(), getDefaultMemHandleType(), nVerts * sizeof(float),
+    importExternalBuffer(
+        (void*)(uintptr_t)m_sim.getInCircleShareableHandle(),
+        getDefaultMemHandleType(), nVerts * sizeof(float),
         VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_VERTEX_BUFFER_BIT,
-            VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT, m_inCircleBuffer, m_inCircleMemory);
+        VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT, m_inCircleBuffer,
+        m_inCircleMemory);
 
-        // Create the semaphore vulkan will signal when it's done with the vertex buffer
+    // Create the semaphore vulkan will signal when it's done with the vertex
-        createExternalSemaphore(m_vkSignalSemaphore, getDefaultSemaphoreHandleType());
+    // buffer
+    createExternalSemaphore(m_vkSignalSemaphore,
+                            getDefaultSemaphoreHandleType());
     // Create the semaphore vulkan will wait for before using the vertex buffer
     createExternalSemaphore(m_vkWaitSemaphore, getDefaultSemaphoreHandleType());
     // Import the semaphore cuda will use -- vulkan's signal will be cuda's wait
-        importCudaExternalSemaphore(m_cudaWaitSemaphore, m_vkSignalSemaphore, getDefaultSemaphoreHandleType());
+    importCudaExternalSemaphore(m_cudaWaitSemaphore, m_vkSignalSemaphore,
+                                getDefaultSemaphoreHandleType());
     // Import the semaphore cuda will use -- cuda's signal will be vulkan's wait
-        importCudaExternalSemaphore(m_cudaSignalSemaphore, m_vkWaitSemaphore, getDefaultSemaphoreHandleType());
+    importCudaExternalSemaphore(m_cudaSignalSemaphore, m_vkWaitSemaphore,
+                                getDefaultSemaphoreHandleType());
   }
 
-    void importCudaExternalSemaphore(cudaExternalSemaphore_t& cudaSem, VkSemaphore& vkSem, VkExternalSemaphoreHandleTypeFlagBits handleType) {
+  void importCudaExternalSemaphore(
+      cudaExternalSemaphore_t& cudaSem, VkSemaphore& vkSem,
+      VkExternalSemaphoreHandleTypeFlagBits handleType) {
     cudaExternalSemaphoreHandleDesc externalSemaphoreHandleDesc = {};
 
     if (handleType & VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32_BIT) {
-            externalSemaphoreHandleDesc.type = cudaExternalSemaphoreHandleTypeOpaqueWin32;
+      externalSemaphoreHandleDesc.type =
-        }
+          cudaExternalSemaphoreHandleTypeOpaqueWin32;
-        else if (handleType & VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32_KMT_BIT) {
+    } else if (handleType &
-            externalSemaphoreHandleDesc.type = cudaExternalSemaphoreHandleTypeOpaqueWin32Kmt;
+               VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32_KMT_BIT) {
-        }
+      externalSemaphoreHandleDesc.type =
-        else if (handleType & VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT) {
+          cudaExternalSemaphoreHandleTypeOpaqueWin32Kmt;
-            externalSemaphoreHandleDesc.type = cudaExternalSemaphoreHandleTypeOpaqueFd;
+    } else if (handleType & VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT) {
-        }
+      externalSemaphoreHandleDesc.type =
-        else {
+          cudaExternalSemaphoreHandleTypeOpaqueFd;
+    } else {
       throw std::runtime_error("Unknown handle type requested!");
     }
 
 #ifdef _WIN64
-        externalSemaphoreHandleDesc.handle.win32.handle = (HANDLE)getSemaphoreHandle(vkSem, handleType);
+    externalSemaphoreHandleDesc.handle.win32.handle =
+        (HANDLE)getSemaphoreHandle(vkSem, handleType);
 #else
-        externalSemaphoreHandleDesc.handle.fd = (int)(uintptr_t)getSemaphoreHandle(vkSem, handleType);
+    externalSemaphoreHandleDesc.handle.fd =
+        (int)(uintptr_t)getSemaphoreHandle(vkSem, handleType);
 #endif
 
     externalSemaphoreHandleDesc.flags = 0;
 
-        checkCudaErrors(cudaImportExternalSemaphore(&cudaSem, &externalSemaphoreHandleDesc));
+    checkCudaErrors(
+        cudaImportExternalSemaphore(&cudaSem, &externalSemaphoreHandleDesc));
   }
 
-    VkDeviceSize getUniformSize() const {
+  VkDeviceSize getUniformSize() const { return sizeof(UniformBufferObject); }
-        return sizeof(UniformBufferObject);
-    }
 
   void updateUniformBuffer(uint32_t imageIndex, size_t globalFrame) {
     m_ubo.frame = (float)globalFrame;
     void* data;
-        vkMapMemory(m_device, m_uniformMemory[imageIndex], 0, getUniformSize(), 0, &data);
+    vkMapMemory(m_device, m_uniformMemory[imageIndex], 0, getUniformSize(), 0,
+                &data);
     memcpy(data, &m_ubo, sizeof(m_ubo));
     vkUnmapMemory(m_device, m_uniformMemory[imageIndex]);
   }
@@ -259,6 +288,9 @@ public:
     std::vector<const char*> extensions;
     extensions.push_back(VK_KHR_EXTERNAL_MEMORY_CAPABILITIES_EXTENSION_NAME);
     extensions.push_back(VK_KHR_EXTERNAL_SEMAPHORE_CAPABILITIES_EXTENSION_NAME);
+    extensions.push_back(VK_KHR_EXTERNAL_FENCE_CAPABILITIES_EXTENSION_NAME);
+    extensions.push_back(
+        VK_KHR_GET_PHYSICAL_DEVICE_PROPERTIES_2_EXTENSION_NAME);
     return extensions;
   }
 
@@ -281,7 +313,9 @@ public:
     static chrono_tp startTime = std::chrono::high_resolution_clock::now();
 
     chrono_tp currentTime = std::chrono::high_resolution_clock::now();
-        float time = std::chrono::duration<float, std::chrono::seconds::period>(currentTime - startTime).count();
+    float time = std::chrono::duration<float, std::chrono::seconds::period>(
+                     currentTime - startTime)
+                     .count();
 
     if (m_currentFrame == 0) {
       m_lastTime = startTime;
@@ -298,17 +332,18 @@ public:
     // Have vulkan draw the current frame...
     VulkanBaseApp::drawFrame();
     // Wait for vulkan to complete it's work
-        checkCudaErrors(cudaWaitExternalSemaphoresAsync(&m_cudaWaitSemaphore, &waitParams, 1, m_stream));
+    checkCudaErrors(cudaWaitExternalSemaphoresAsync(&m_cudaWaitSemaphore,
+                                                    &waitParams, 1, m_stream));
     // Now step the simulation
     m_sim.stepSimulation(time, m_stream);
 
     // Signal vulkan to continue with the updated buffers
-        checkCudaErrors(cudaSignalExternalSemaphoresAsync(&m_cudaSignalSemaphore, &signalParams, 1, m_stream));
+    checkCudaErrors(cudaSignalExternalSemaphoresAsync(
+        &m_cudaSignalSemaphore, &signalParams, 1, m_stream));
   }
 };
 
-int main(int argc, char **argv)
+int main(int argc, char** argv) {
-{
   execution_path = argv[0];
   VulkanCudaPi app(NUM_SIMULATION_POINTS);
   app.init();

Samples/vulkanImageCUDA/Build_instructions.txt

@@ -19,8 +19,17 @@ For Linux:
 -- Install "libxcb1-dev" and "xorg-dev" as GLFW3 is depended on it
 -- Add Vulkan and GLFW3 libraries directories to LD_LIBRARY_PATH
 
+
 For Linux aarch64(L4T):
 -- Install GLFW3 library using "sudo apt-get install libglfw3-dev" this will provide glfw3 
 -- install above will also provide libvulkan-dev as dependencies
 -- Add Vulkan and GLFW3 libraries directories to LD_LIBRARY_PATH
 -- Pass path to vulkan sdk while building 'make VULKAN_SDK_PATH=<PATH_TO_VULKAN_SDK>', VULKAN_SDK_PATH in this scenario is typically "/usr"
+
+
+For Shader changes:
+-- Update the shader.vert and/or shader.frag shader source file as required
+-- Use the glslc shader compiler from the installed Vulkan SDK's bin directory to compile shaders as:
+    glslc shader.vert -o vert.spv
+    glslc shader.frag -o frag.spv
+** Make sure to add glslc's path in your PATH environment variable **

Samples/vulkanImageCUDA/vulkanImageCUDA.cu

@@ -69,7 +69,7 @@ const std::vector<const char*> validationLayers = {
     "VK_LAYER_KHRONOS_validation"};
 
 #ifdef NDEBUG
-const bool enableValidationLayers = false;
+const bool enableValidationLayers = true;
 #else
 const bool enableValidationLayers = false;
 #endif
@@ -494,7 +494,7 @@ class vulkanImageCUDA {
 
   unsigned int* image_data = NULL;
   unsigned int imageWidth, imageHeight;
-  unsigned int mipLevels;
+  unsigned int mipLevels = 1;
   size_t totalImageMemSize;
 
   // CUDA objects
@@ -630,6 +630,9 @@ class vulkanImageCUDA {
     vkDestroyBuffer(device, vertexBuffer, nullptr);
     vkFreeMemory(device, vertexBufferMemory, nullptr);
 
+    vkDestroySemaphore(device, cudaUpdateVkSemaphore, nullptr);
+    vkDestroySemaphore(device, vkUpdateCudaSemaphore, nullptr);
+
     for (size_t i = 0; i < MAX_FRAMES; i++) {
       vkDestroySemaphore(device, renderFinishedSemaphores[i], nullptr);
       vkDestroySemaphore(device, imageAvailableSemaphores[i], nullptr);
@@ -686,7 +689,7 @@ class vulkanImageCUDA {
     appInfo.applicationVersion = VK_MAKE_VERSION(1, 0, 0);
     appInfo.pEngineName = "No Engine";
     appInfo.engineVersion = VK_MAKE_VERSION(1, 0, 0);
-    appInfo.apiVersion = VK_API_VERSION_1_0;
+    appInfo.apiVersion = VK_API_VERSION_1_1;
 
     VkInstanceCreateInfo createInfo = {};
     createInfo.sType = VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO;
@@ -905,6 +908,7 @@ class vulkanImageCUDA {
     }
 
     VkPhysicalDeviceFeatures deviceFeatures = {};
+    deviceFeatures.samplerAnisotropy = VK_TRUE;
 
     VkDeviceCreateInfo createInfo = {};
     createInfo.sType = VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO;
@@ -1078,8 +1082,8 @@ class vulkanImageCUDA {
   }
 
   void createGraphicsPipeline() {
-    auto vertShaderCode = readFile("shader.vert");
+    auto vertShaderCode = readFile("vert.spv");
-    auto fragShaderCode = readFile("shader.frag");
+    auto fragShaderCode = readFile("frag.spv");
 
     VkShaderModule vertShaderModule = createShaderModule(vertShaderCode);
     VkShaderModule fragShaderModule = createShaderModule(fragShaderCode);
@@ -1268,7 +1272,7 @@ class vulkanImageCUDA {
 
     // VK_FORMAT_R8G8B8A8_UNORM changed to VK_FORMAT_R8G8B8A8_UINT
     createImage(
-        imageWidth, imageHeight, VK_FORMAT_R8G8B8A8_UINT,
+        imageWidth, imageHeight, VK_FORMAT_R8G8B8A8_UNORM,
         VK_IMAGE_TILING_OPTIMAL,
         VK_IMAGE_USAGE_STORAGE_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT |
             VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_SAMPLED_BIT,
@@ -1280,9 +1284,6 @@ class vulkanImageCUDA {
     copyBufferToImage(stagingBuffer, textureImage,
                       static_cast<uint32_t>(imageWidth),
                       static_cast<uint32_t>(imageHeight));
-    transitionImageLayout(textureImage, VK_FORMAT_R8G8B8A8_UINT,
-                          VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
-                          VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL);
 
     vkDestroyBuffer(device, stagingBuffer, nullptr);
     vkFreeMemory(device, stagingBufferMemory, nullptr);
@@ -1523,8 +1524,13 @@ class vulkanImageCUDA {
     vkExternalMemImageCreateInfo.sType =
         VK_STRUCTURE_TYPE_EXTERNAL_MEMORY_IMAGE_CREATE_INFO;
     vkExternalMemImageCreateInfo.pNext = NULL;
+#ifdef _WIN64
+    vkExternalMemImageCreateInfo.handleTypes =
+        VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_BIT;
+#else
     vkExternalMemImageCreateInfo.handleTypes =
         VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT_KHR;
+#endif
 
     imageInfo.pNext = &vkExternalMemImageCreateInfo;
 
@@ -2201,7 +2207,6 @@ class vulkanImageCUDA {
       throw std::runtime_error(
           "failed to create synchronization objects for a CUDA-Vulkan!");
     }
-
   }
 
   void updateUniformBuffer() {
@@ -2333,8 +2338,8 @@ class vulkanImageCUDA {
     submitInfo.signalSemaphoreCount = 2;
     submitInfo.pSignalSemaphores = signalSemaphores;
 
-    if (vkQueueSubmit(graphicsQueue, 1, &submitInfo, inFlightFences[currentFrame]) !=
+    if (vkQueueSubmit(graphicsQueue, 1, &submitInfo,
-        VK_SUCCESS) {
+                      inFlightFences[currentFrame]) != VK_SUCCESS) {
       throw std::runtime_error("failed to submit draw command buffer!");
     }
   }
@@ -2360,8 +2365,8 @@ class vulkanImageCUDA {
     submitInfo.signalSemaphoreCount = 2;
     submitInfo.pSignalSemaphores = signalSemaphores;
 
-    if (vkQueueSubmit(graphicsQueue, 1, &submitInfo, inFlightFences[currentFrame]) !=
+    if (vkQueueSubmit(graphicsQueue, 1, &submitInfo,
-        VK_SUCCESS) {
+                      inFlightFences[currentFrame]) != VK_SUCCESS) {
       throw std::runtime_error("failed to submit draw command buffer!");
     }
   }