VkResult VKAPI_CALL vkAllocateCommandBuffers(
VkDevice device,
const VkCommandBufferAllocateInfo* pAllocateInfo,
VkCommandBuffer* pCommandBuffers)
{
dispatch_key key = get_dispatch_key(device);
layer_data *my_data = get_my_data_ptr(key, layer_data_map);
VkLayerDispatchTable *pTable = my_data->device_dispatch_table;
VkResult result;
startReadObject(my_data, device);
startWriteObject(my_data, pAllocateInfo->commandPool);
result = pTable->AllocateCommandBuffers(device, pAllocateInfo, pCommandBuffers);
finishReadObject(my_data, device);
finishWriteObject(my_data, pAllocateInfo->commandPool);
// Record mapping from command buffer to command pool
if (VK_SUCCESS == result) {
for (int index=0;index<pAllocateInfo->commandBufferCount;index++) {
loader_platform_thread_lock_mutex(&threadingLock);
command_pool_map[pCommandBuffers[index]] = pAllocateInfo->commandPool;
loader_platform_thread_unlock_mutex(&threadingLock);
}
}
return result;
}
/* Various dispatchable objects will use the same underlying dispatch table if
* they
* are created from that "parent" object. Thus use pointer to dispatch table
* as the key to these table maps.
* Instance -> PhysicalDevice
* Device -> CommandBuffer or Queue
* If use the object themselves as key to map then implies Create entrypoints
* have to be intercepted
* and a new key inserted into map */
VkLayerDebugMarkerDispatchTable *initDebugMarkerTable(VkDevice device) {
VkLayerDebugMarkerDispatchTable *pDebugMarkerTable;
assert(device);
VkLayerDispatchTable *pDisp = *(VkLayerDispatchTable **)device;
std::unordered_map<void *,
VkLayerDebugMarkerDispatchTable *>::const_iterator it =
tableDebugMarkerMap.find((void *)pDisp);
if (it == tableDebugMarkerMap.end()) {
pDebugMarkerTable = new VkLayerDebugMarkerDispatchTable;
tableDebugMarkerMap[(void *)pDisp] = pDebugMarkerTable;
} else {
return it->second;
}
pDebugMarkerTable->CmdDbgMarkerBegin =
(PFN_vkCmdDbgMarkerBegin)pDisp->GetDeviceProcAddr(
device, "vkCmdDbgMarkerBegin");
pDebugMarkerTable->CmdDbgMarkerEnd =
(PFN_vkCmdDbgMarkerEnd)pDisp->GetDeviceProcAddr(device,
"vkCmdDbgMarkerEnd");
pDebugMarkerTable->DbgSetObjectTag =
(PFN_vkDbgSetObjectTag)pDisp->GetDeviceProcAddr(device,
"vkDbgSetObjectTag");
pDebugMarkerTable->DbgSetObjectName =
(PFN_vkDbgSetObjectName)pDisp->GetDeviceProcAddr(device,
"vkDbgSetObjectName");
return pDebugMarkerTable;
}
VK_LAYER_EXPORT VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
vkGetDeviceProcAddr(VkDevice dev, const char *funcName) {
#define ADD_HOOK(fn) \
if (!strncmp(#fn, funcName, sizeof(#fn))) \
return (PFN_vkVoidFunction)fn
ADD_HOOK(vkGetDeviceProcAddr);
ADD_HOOK(vkDestroyDevice);
ADD_HOOK(vkCreateSwapchainKHR);
ADD_HOOK(vkGetSwapchainImagesKHR);
ADD_HOOK(vkQueuePresentKHR);
ADD_HOOK(vkDestroySwapchainKHR);
ADD_HOOK(vkQueueSubmit);
#undef ADD_HOOK
if (dev == NULL)
return NULL;
layer_data *dev_data;
dev_data = get_my_data_ptr(get_dispatch_key(dev), layer_data_map);
VkLayerDispatchTable *pTable = dev_data->device_dispatch_table;
if (pTable->GetDeviceProcAddr == NULL)
return NULL;
return pTable->GetDeviceProcAddr(dev, funcName);
}
void VKAPI_CALL vkFreeCommandBuffers(
VkDevice device,
VkCommandPool commandPool,
uint32_t commandBufferCount,
const VkCommandBuffer* pCommandBuffers)
{
dispatch_key key = get_dispatch_key(device);
layer_data *my_data = get_my_data_ptr(key, layer_data_map);
VkLayerDispatchTable *pTable = my_data->device_dispatch_table;
const bool lockCommandPool = false; // pool is already directly locked
startReadObject(my_data, device);
startWriteObject(my_data, commandPool);
for (int index=0;index<commandBufferCount;index++) {
startWriteObject(my_data, pCommandBuffers[index], lockCommandPool);
}
pTable->FreeCommandBuffers(device,commandPool,commandBufferCount,pCommandBuffers);
finishReadObject(my_data, device);
finishWriteObject(my_data, commandPool);
for (int index=0;index<commandBufferCount;index++) {
finishWriteObject(my_data, pCommandBuffers[index], lockCommandPool);
loader_platform_thread_lock_mutex(&threadingLock);
command_pool_map.erase(pCommandBuffers[index]);
loader_platform_thread_unlock_mutex(&threadingLock);
}
}
VK_LAYER_EXPORT VKAPI_ATTR VkResult VKAPI_CALL vkQueueSubmit(VkQueue queue, uint32_t submitCount, const VkSubmitInfo *pSubmits,
VkFence fence) {
layer_data *my_data = GetLayerDataPtr(get_dispatch_key(queue), layer_data_map);
VkLayerDispatchTable *pTable = my_data->device_dispatch_table;
my_data->cmdBuffersThisFrame += submitCount; // XXX WRONG
return pTable->QueueSubmit(queue, submitCount, pSubmits, fence);
}
VK_LAYER_EXPORT VKAPI_ATTR void VKAPI_CALL vkDestroyDevice(VkDevice device, const VkAllocationCallbacks *pAllocator) {
dispatch_key key = get_dispatch_key(device);
layer_data *my_data = GetLayerDataPtr(key, layer_data_map);
my_data->Cleanup();
VkLayerDispatchTable *pTable = my_data->device_dispatch_table;
pTable->DeviceWaitIdle(device);
pTable->DestroyDevice(device, pAllocator);
delete pTable;
layer_data_map.erase(key);
}
void WsiImageData::Cleanup(VkDevice dev) {
layer_data *my_data =
get_my_data_ptr(get_dispatch_key(dev), layer_data_map);
VkLayerDispatchTable *pTable = my_data->device_dispatch_table;
// XXX: needs device data
// pTable->FreeCommandBuffers(dev, cmd, nullptr);
pTable->DestroyFramebuffer(dev, framebuffer, nullptr);
pTable->DestroyImageView(dev, view, nullptr);
pTable->DestroyBuffer(dev, vertexBuffer, nullptr);
pTable->FreeMemory(dev, vertexBufferMemory, nullptr);
}
void SwapChainData::Cleanup(VkDevice dev) {
layer_data *my_data = GetLayerDataPtr(get_dispatch_key(dev), layer_data_map);
VkLayerDispatchTable *pTable = my_data->device_dispatch_table;
for (uint32_t i = 0; i < presentableImages.size(); i++) {
presentableImages[i]->Cleanup(dev);
delete presentableImages[i];
}
presentableImages.clear();
pTable->DestroyPipeline(dev, pipeline, nullptr);
pTable->DestroyRenderPass(dev, render_pass, nullptr);
}
VKAPI_ATTR void VKAPI_CALL
DestroyDevice(VkDevice device, const VkAllocationCallbacks *pAllocator) {
DeviceMapStruct *devMap = get_dev_info(device);
assert(devMap);
VkLayerDispatchTable *pDisp = devMap->device_dispatch_table;
pDisp->DestroyDevice(device, pAllocator);
loader_platform_thread_lock_mutex(&globalLock);
delete pDisp;
delete devMap;
deviceMap.erase(device);
loader_platform_thread_unlock_mutex(&globalLock);
}
VKAPI_ATTR VkResult VKAPI_CALL
GetSwapchainImagesKHR(VkDevice device, VkSwapchainKHR swapchain,
uint32_t *pCount, VkImage *pSwapchainImages) {
DeviceMapStruct *devMap = get_dev_info(device);
assert(devMap);
VkLayerDispatchTable *pDisp = devMap->device_dispatch_table;
VkResult result = pDisp->GetSwapchainImagesKHR(device, swapchain, pCount,
pSwapchainImages);
// Save the swapchain images in a map if we are taking screenshots
loader_platform_thread_lock_mutex(&globalLock);
if (screenshotEnvQueried && screenshotFrames.empty()) {
// No screenshots in the list to take
loader_platform_thread_unlock_mutex(&globalLock);
return result;
}
if (result == VK_SUCCESS && pSwapchainImages && !swapchainMap.empty() &&
swapchainMap.find(swapchain) != swapchainMap.end()) {
unsigned i;
for (i = 0; i < *pCount; i++) {
// Create a mapping for an image to a device, image extent, and
// format
if (imageMap[pSwapchainImages[i]] == NULL) {
ImageMapStruct *imageMapElem = new ImageMapStruct;
imageMap[pSwapchainImages[i]] = imageMapElem;
}
imageMap[pSwapchainImages[i]]->device =
swapchainMap[swapchain]->device;
imageMap[pSwapchainImages[i]]->imageExtent =
swapchainMap[swapchain]->imageExtent;
imageMap[pSwapchainImages[i]]->format =
swapchainMap[swapchain]->format;
}
// Add list of images to swapchain to image map
SwapchainMapStruct *swapchainMapElem = swapchainMap[swapchain];
if (i >= 1 && swapchainMapElem) {
VkImage *imageList = new VkImage[i];
swapchainMapElem->imageList = imageList;
for (unsigned j = 0; j < i; j++) {
swapchainMapElem->imageList[j] = pSwapchainImages[j];
}
}
}
loader_platform_thread_unlock_mutex(&globalLock);
return result;
}
VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL GetDeviceProcAddr(VkDevice device, const char *funcName) {
PFN_vkVoidFunction addr;
layer_data *dev_data;
assert(device);
addr = layer_intercept_proc(funcName);
if (addr)
return addr;
dev_data = get_my_data_ptr(get_dispatch_key(device), layer_data_map);
VkLayerDispatchTable *pTable = dev_data->device_dispatch_table;
if (pTable->GetDeviceProcAddr == NULL)
return NULL;
return pTable->GetDeviceProcAddr(device, funcName);
}
VK_LAYER_EXPORT PFN_vkVoidFunction VKAPI_CALL vkGetDeviceProcAddr(VkDevice device, const char *funcName) {
PFN_vkVoidFunction addr;
layer_data *dev_data;
if (device == VK_NULL_HANDLE) {
return NULL;
}
addr = layer_intercept_proc(funcName);
if (addr)
return addr;
dev_data = get_my_data_ptr(get_dispatch_key(device), layer_data_map);
VkLayerDispatchTable *pTable = dev_data->device_dispatch_table;
if (pTable->GetDeviceProcAddr == NULL)
return NULL;
return pTable->GetDeviceProcAddr(device, funcName);
}
VKAPI_ATTR VkResult VKAPI_CALL CreateSwapchainKHR(
VkDevice device, const VkSwapchainCreateInfoKHR *pCreateInfo,
const VkAllocationCallbacks *pAllocator, VkSwapchainKHR *pSwapchain) {
DeviceMapStruct *devMap = get_dev_info(device);
assert(devMap);
VkLayerDispatchTable *pDisp = devMap->device_dispatch_table;
// This layer does an image copy later on, and the copy command expects the
// transfer src bit to be on.
VkSwapchainCreateInfoKHR myCreateInfo = *pCreateInfo;
myCreateInfo.imageUsage |= VK_IMAGE_USAGE_TRANSFER_SRC_BIT;
VkResult result = pDisp->CreateSwapchainKHR(device, &myCreateInfo,
pAllocator, pSwapchain);
// Save the swapchain in a map of we are taking screenshots.
loader_platform_thread_lock_mutex(&globalLock);
if (screenshotEnvQueried && screenshotFrames.empty()) {
// No screenshots in the list to take
loader_platform_thread_unlock_mutex(&globalLock);
return result;
}
if (result == VK_SUCCESS) {
// Create a mapping for a swapchain to a device, image extent, and
// format
SwapchainMapStruct *swapchainMapElem = new SwapchainMapStruct;
swapchainMapElem->device = device;
swapchainMapElem->imageExtent = pCreateInfo->imageExtent;
swapchainMapElem->format = pCreateInfo->imageFormat;
swapchainMap.insert(make_pair(*pSwapchain, swapchainMapElem));
// Create a mapping for the swapchain object into the dispatch table
// TODO is this needed? screenshot_device_table_map.emplace((void
// *)pSwapchain, pTable);
}
loader_platform_thread_unlock_mutex(&globalLock);
return result;
}
VKAPI_ATTR VkResult VKAPI_CALL CreateCommandPool(
VkDevice device, const VkCommandPoolCreateInfo *pCreateInfo,
const VkAllocationCallbacks *pAllocator, VkCommandPool *pCommandPool) {
DeviceMapStruct *devMap = get_dev_info(device);
assert(devMap);
VkLayerDispatchTable *pDisp = devMap->device_dispatch_table;
VkResult result =
pDisp->CreateCommandPool(device, pCreateInfo, pAllocator, pCommandPool);
// Save the command pool on a map if we are taking screenshots.
loader_platform_thread_lock_mutex(&globalLock);
if (screenshotEnvQueried && screenshotFrames.empty()) {
// No screenshots in the list to take
loader_platform_thread_unlock_mutex(&globalLock);
return result;
}
// Create a mapping from a device to a commandPool
devMap->commandPool = *pCommandPool;
loader_platform_thread_unlock_mutex(&globalLock);
return result;
}
VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
GetDeviceProcAddr(VkDevice dev, const char *funcName) {
PFN_vkVoidFunction proc = intercept_core_device_command(funcName);
if (proc)
return proc;
if (dev == NULL) {
return NULL;
}
proc = intercept_khr_swapchain_command(funcName, dev);
if (proc)
return proc;
DeviceMapStruct *devMap = get_dev_info(dev);
assert(devMap);
VkLayerDispatchTable *pDisp = devMap->device_dispatch_table;
if (pDisp->GetDeviceProcAddr == NULL)
return NULL;
return pDisp->GetDeviceProcAddr(dev, funcName);
}
VK_LAYER_EXPORT VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL vkGetDeviceProcAddr(VkDevice dev, const char *funcName) {
if (!strcmp(funcName, "vkGetDeviceProcAddr"))
return (PFN_vkVoidFunction)vkGetDeviceProcAddr;
if (!strcmp(funcName, "vkDestroyDevice"))
return (PFN_vkVoidFunction)vkDestroyDevice;
if (!strcmp(funcName, "vkGetDeviceQueue"))
return (PFN_vkVoidFunction)vkGetDeviceQueue;
if (!strcmp(funcName, "vkCreateRenderPass"))
return (PFN_vkVoidFunction)vkCreateRenderPass;
if (!strcmp(funcName, "vkCreateCommandPool"))
return (PFN_vkVoidFunction)vkCreateCommandPool;
if (!strcmp(funcName, "vkDestroyCommandPool"))
return (PFN_vkVoidFunction)vkDestroyCommandPool;
if (!strcmp(funcName, "vkResetCommandPool"))
return (PFN_vkVoidFunction)vkResetCommandPool;
if (!strcmp(funcName, "vkAllocateCommandBuffers"))
return (PFN_vkVoidFunction)vkAllocateCommandBuffers;
if (!strcmp(funcName, "vkFreeCommandBuffers"))
return (PFN_vkVoidFunction)vkFreeCommandBuffers;
if (!strcmp(funcName, "vkBeginCommandBuffer"))
return (PFN_vkVoidFunction)vkBeginCommandBuffer;
if (!strcmp(funcName, "vkCmdUpdateBuffer"))
return (PFN_vkVoidFunction)vkCmdUpdateBuffer;
if (!strcmp(funcName, "vkUpdateDescriptorSets"))
return (PFN_vkVoidFunction)vkUpdateDescriptorSets;
if (!strcmp(funcName, "vkCmdFillBuffer"))
return (PFN_vkVoidFunction)vkCmdFillBuffer;
if (dev == NULL)
return NULL;
layer_data *my_data = get_my_data_ptr(get_dispatch_key(dev), layer_data_map);
VkLayerDispatchTable *pTable = my_data->device_dispatch_table;
{
if (pTable->GetDeviceProcAddr == NULL)
return NULL;
return pTable->GetDeviceProcAddr(dev, funcName);
}
}
VKAPI_ATTR VkResult VKAPI_CALL
AllocateCommandBuffers(VkDevice device, const VkCommandBufferAllocateInfo *pAllocateInfo, VkCommandBuffer *pCommandBuffers) {
dispatch_key key = get_dispatch_key(device);
layer_data *my_data = get_my_data_ptr(key, layer_data_map);
VkLayerDispatchTable *pTable = my_data->device_dispatch_table;
VkResult result;
startReadObject(my_data, device);
startWriteObject(my_data, pAllocateInfo->commandPool);
result = pTable->AllocateCommandBuffers(device, pAllocateInfo, pCommandBuffers);
finishReadObject(my_data, device);
finishWriteObject(my_data, pAllocateInfo->commandPool);
// Record mapping from command buffer to command pool
if (VK_SUCCESS == result) {
for (uint32_t index = 0; index < pAllocateInfo->commandBufferCount; index++) {
std::lock_guard<std::mutex> lock(global_lock);
command_pool_map[pCommandBuffers[index]] = pAllocateInfo->commandPool;
}
}
return result;
}
VKAPI_ATTR void VKAPI_CALL GetDeviceQueue(VkDevice device,
uint32_t queueNodeIndex,
uint32_t queueIndex,
VkQueue *pQueue) {
DeviceMapStruct *devMap = get_dev_info(device);
assert(devMap);
VkLayerDispatchTable *pDisp = devMap->device_dispatch_table;
pDisp->GetDeviceQueue(device, queueNodeIndex, queueIndex, pQueue);
// Save the device queue in a map if we are taking screenshots.
loader_platform_thread_lock_mutex(&globalLock);
if (screenshotEnvQueried && screenshotFrames.empty()) {
// No screenshots in the list to take
loader_platform_thread_unlock_mutex(&globalLock);
return;
}
VkDevice que = static_cast<VkDevice>(static_cast<void *>(*pQueue));
deviceMap.emplace(que, devMap);
// Create a mapping from a device to a queue
devMap->queue = *pQueue;
loader_platform_thread_unlock_mutex(&globalLock);
}
VK_LAYER_EXPORT VKAPI_ATTR VkResult VKAPI_CALL vkGetSwapchainImagesKHR(VkDevice device, VkSwapchainKHR swapChain, uint32_t *pCount,
VkImage *pImages) {
layer_data *my_data = GetLayerDataPtr(get_dispatch_key(device), layer_data_map);
VkLayerDispatchTable *pTable = my_data->device_dispatch_table;
VkResult result = my_data->pfnGetSwapchainImagesKHR(device, swapChain, pCount, pImages);
VkResult U_ASSERT_ONLY err;
/* GetSwapChainImagesWSI may be called without an images buffer, in which
* case it
* just returns the count to the caller. We're only interested in acting on
* the
* /actual/ fetch of the images.
*/
if (pImages) {
auto data = (*my_data->swapChains)[swapChain];
for (uint32_t i = 0; i < *pCount; i++) {
/* Create attachment view for each */
VkImageViewCreateInfo ivci;
ivci.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO;
ivci.viewType = VK_IMAGE_VIEW_TYPE_2D;
ivci.pNext = nullptr;
ivci.format = data->format;
ivci.components.r = VK_COMPONENT_SWIZZLE_R;
ivci.components.g = VK_COMPONENT_SWIZZLE_G;
ivci.components.b = VK_COMPONENT_SWIZZLE_B;
ivci.components.a = VK_COMPONENT_SWIZZLE_A;
ivci.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
ivci.subresourceRange.baseMipLevel = 0;
ivci.subresourceRange.levelCount = 1;
ivci.subresourceRange.baseArrayLayer = 0;
ivci.subresourceRange.layerCount = 1;
ivci.image = pImages[i];
ivci.flags = 0;
VkImageView v;
pTable->CreateImageView(device, &ivci, nullptr, &v);
/* Create framebuffer for each */
VkFramebufferCreateInfo fci;
fci.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO;
fci.pNext = nullptr;
fci.flags = 0;
fci.renderPass = data->render_pass;
fci.attachmentCount = 1;
fci.pAttachments = &v;
fci.width = data->width;
fci.height = data->height;
fci.layers = 1;
VkFramebuffer fb;
pTable->CreateFramebuffer(device, &fci, nullptr, &fb);
/* Create command buffer for each */
VkCommandBufferAllocateInfo cbai;
cbai.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO;
cbai.pNext = nullptr;
cbai.commandPool = my_data->pool;
cbai.level = VK_COMMAND_BUFFER_LEVEL_PRIMARY;
cbai.commandBufferCount = 1;
VkCommandBuffer cmd;
pTable->AllocateCommandBuffers(device, &cbai, &cmd);
/* We have just created a dispatchable object, but the dispatch
* table has not been placed in the object yet.
* When a "normal" application creates a command buffer,
* the dispatch table is installed by the top-level binding
* (trampoline.c).
* But here, we have to do it ourselves. */
if (!my_data->pfn_dev_init) {
*((const void **)cmd) = *(void **)device;
} else {
err = my_data->pfn_dev_init(device, (void *)cmd);
assert(!err);
}
/* Create vertex buffer */
VkBufferCreateInfo bci;
memset(&bci, 0, sizeof(bci));
bci.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
bci.usage = VK_BUFFER_USAGE_VERTEX_BUFFER_BIT;
bci.size = sizeof(vertex) * MAX_TEXT_VERTICES;
VkBuffer buf;
err = pTable->CreateBuffer(device, &bci, nullptr, &buf);
assert(!err);
VkMemoryRequirements mem_reqs;
pTable->GetBufferMemoryRequirements(device, buf, &mem_reqs);
assert(!err);
VkMemoryAllocateInfo mem_alloc;
memset(&mem_alloc, 0, sizeof(mem_alloc));
mem_alloc.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO;
mem_alloc.allocationSize = mem_reqs.size;
mem_alloc.memoryTypeIndex = choose_memory_type(
my_data->gpu, mem_reqs.memoryTypeBits, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT);
VkDeviceMemory mem;
err = pTable->AllocateMemory(device, &mem_alloc, nullptr, &mem);
assert(!err);
err = pTable->BindBufferMemory(device, buf, mem, 0);
assert(!err);
auto imageData = new WsiImageData;
imageData->image = pImages[i];
imageData->view = v;
imageData->framebuffer = fb;
imageData->cmd = cmd;
imageData->vertexBuffer = buf;
imageData->vertexBufferMemory = mem;
imageData->numVertices = 0;
imageData->vertexBufferSize = mem_alloc.allocationSize;
data->presentableImages.push_back(imageData);
}
}
return result;
}
VkResult WrappedVulkan::vkQueuePresentKHR(VkQueue queue, const VkPresentInfoKHR *pPresentInfo)
{
if(m_State == WRITING_IDLE)
{
RenderDoc::Inst().Tick();
GetResourceManager()->FlushPendingDirty();
}
m_FrameCounter++; // first present becomes frame #1, this function is at the end of the frame
if(pPresentInfo->swapchainCount > 1 && (m_FrameCounter % 100) == 0)
{
RDCWARN("Presenting multiple swapchains at once - only first will be processed");
}
vector<VkSwapchainKHR> unwrappedSwaps;
vector<VkSemaphore> unwrappedSems;
VkPresentInfoKHR unwrappedInfo = *pPresentInfo;
for(uint32_t i = 0; i < unwrappedInfo.swapchainCount; i++)
unwrappedSwaps.push_back(Unwrap(unwrappedInfo.pSwapchains[i]));
for(uint32_t i = 0; i < unwrappedInfo.waitSemaphoreCount; i++)
unwrappedSems.push_back(Unwrap(unwrappedInfo.pWaitSemaphores[i]));
unwrappedInfo.pSwapchains = unwrappedInfo.swapchainCount ? &unwrappedSwaps[0] : NULL;
unwrappedInfo.pWaitSemaphores = unwrappedInfo.waitSemaphoreCount ? &unwrappedSems[0] : NULL;
// Don't support any extensions for present info
RDCASSERT(pPresentInfo->pNext == NULL);
VkResourceRecord *swaprecord = GetRecord(pPresentInfo->pSwapchains[0]);
RDCASSERT(swaprecord->swapInfo);
SwapchainInfo &swapInfo = *swaprecord->swapInfo;
bool activeWindow = RenderDoc::Inst().IsActiveWindow(LayerDisp(m_Instance), swapInfo.wndHandle);
// need to record which image was last flipped so we can get the correct backbuffer
// for a thumbnail in EndFrameCapture
swapInfo.lastPresent = pPresentInfo->pImageIndices[0];
m_LastSwap = swaprecord->GetResourceID();
if(m_State == WRITING_IDLE)
{
uint32_t overlay = RenderDoc::Inst().GetOverlayBits();
if(overlay & eRENDERDOC_Overlay_Enabled)
{
VkRenderPass rp = swapInfo.rp;
VkImage im = swapInfo.images[pPresentInfo->pImageIndices[0]].im;
VkFramebuffer fb = swapInfo.images[pPresentInfo->pImageIndices[0]].fb;
VkLayerDispatchTable *vt = ObjDisp(GetDev());
TextPrintState textstate = {
GetNextCmd(), rp, fb, swapInfo.extent.width, swapInfo.extent.height, swapInfo.format};
VkCommandBufferBeginInfo beginInfo = {VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO, NULL,
VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT};
VkResult vkr = vt->BeginCommandBuffer(Unwrap(textstate.cmd), &beginInfo);
RDCASSERTEQUAL(vkr, VK_SUCCESS);
VkImageMemoryBarrier bbBarrier = {VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
NULL,
0,
0,
VK_IMAGE_LAYOUT_PRESENT_SRC_KHR,
VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL,
VK_QUEUE_FAMILY_IGNORED,
VK_QUEUE_FAMILY_IGNORED,
Unwrap(im),
{VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1}};
bbBarrier.srcAccessMask = VK_ACCESS_ALL_READ_BITS;
bbBarrier.dstAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
DoPipelineBarrier(textstate.cmd, 1, &bbBarrier);
GetDebugManager()->BeginText(textstate);
int flags = activeWindow ? RenderDoc::eOverlay_ActiveWindow : 0;
string overlayText = RenderDoc::Inst().GetOverlayText(RDC_Vulkan, m_FrameCounter, flags);
if(!overlayText.empty())
GetDebugManager()->RenderText(textstate, 0.0f, 0.0f, overlayText.c_str());
GetDebugManager()->EndText(textstate);
std::swap(bbBarrier.oldLayout, bbBarrier.newLayout);
bbBarrier.srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
bbBarrier.dstAccessMask = VK_ACCESS_ALL_READ_BITS;
DoPipelineBarrier(textstate.cmd, 1, &bbBarrier);
ObjDisp(textstate.cmd)->EndCommandBuffer(Unwrap(textstate.cmd));
SubmitCmds();
FlushQ();
}
}
VkResult vkr = ObjDisp(queue)->QueuePresentKHR(Unwrap(queue), &unwrappedInfo);
if(!activeWindow)
return vkr;
RenderDoc::Inst().SetCurrentDriver(RDC_Vulkan);
// kill any current capture that isn't application defined
if(m_State == WRITING_CAPFRAME && !m_AppControlledCapture)
RenderDoc::Inst().EndFrameCapture(LayerDisp(m_Instance), swapInfo.wndHandle);
if(RenderDoc::Inst().ShouldTriggerCapture(m_FrameCounter) && m_State == WRITING_IDLE)
{
RenderDoc::Inst().StartFrameCapture(LayerDisp(m_Instance), swapInfo.wndHandle);
m_AppControlledCapture = false;
}
return vkr;
}
// Save an image to a PPM image file.
//
// This function issues commands to copy/convert the swapchain image
// from whatever compatible format the swapchain image uses
// to a single format (VK_FORMAT_R8G8B8A8_UNORM) so that the converted
// result can be easily written to a PPM file.
//
// Error handling: If there is a problem, this function should silently
// fail without affecting the Present operation going on in the caller.
// The numerous debug asserts are to catch programming errors and are not
// expected to assert. Recovery and clean up are implemented for image memory
// allocation failures.
// (TODO) It would be nice to pass any failure info to DebugReport or something.
static void writePPM(const char *filename, VkImage image1) {
VkResult err;
bool pass;
// Bail immediately if we can't find the image.
if (imageMap.empty() || imageMap.find(image1) == imageMap.end())
return;
// Collect object info from maps. This info is generally recorded
// by the other functions hooked in this layer.
VkDevice device = imageMap[image1]->device;
VkPhysicalDevice physicalDevice = deviceMap[device]->physicalDevice;
VkInstance instance = physDeviceMap[physicalDevice]->instance;
VkQueue queue = deviceMap[device]->queue;
DeviceMapStruct *devMap = get_dev_info(device);
if (NULL == devMap) {
assert(0);
return;
}
VkLayerDispatchTable *pTableDevice = devMap->device_dispatch_table;
VkLayerDispatchTable *pTableQueue =
get_dev_info(static_cast<VkDevice>(static_cast<void *>(queue)))
->device_dispatch_table;
VkLayerInstanceDispatchTable *pInstanceTable;
pInstanceTable = instance_dispatch_table(instance);
// Gather incoming image info and check image format for compatibility with
// the target format.
// This function supports both 24-bit and 32-bit swapchain images.
VkFormat const target32bitFormat = VK_FORMAT_R8G8B8A8_UNORM;
VkFormat const target24bitFormat = VK_FORMAT_R8G8B8_UNORM;
uint32_t const width = imageMap[image1]->imageExtent.width;
uint32_t const height = imageMap[image1]->imageExtent.height;
VkFormat const format = imageMap[image1]->format;
uint32_t const numChannels = vk_format_get_channel_count(format);
if ((vk_format_get_compatibility_class(target24bitFormat) !=
vk_format_get_compatibility_class(format)) &&
(vk_format_get_compatibility_class(target32bitFormat) !=
vk_format_get_compatibility_class(format))) {
assert(0);
return;
}
if ((3 != numChannels) && (4 != numChannels)) {
assert(0);
return;
}
// General Approach
//
// The idea here is to copy/convert the swapchain image into another image
// that can be mapped and read by the CPU to produce a PPM file.
// The image must be untiled and converted to a specific format for easy
// parsing. The memory for the final image must be host-visible.
// Note that in Vulkan, a BLIT operation must be used to perform a format
// conversion.
//
// Devices vary in their ability to blit to/from linear and optimal tiling.
// So we must query the device properties to get this information.
//
// If the device cannot BLIT to a LINEAR image, then the operation must be
// done in two steps:
// 1) BLIT the swapchain image (image1) to a temp image (image2) that is
// created with TILING_OPTIMAL.
// 2) COPY image2 to another temp image (image3) that is created with
// TILING_LINEAR.
// 3) Map image 3 and write the PPM file.
//
// If the device can BLIT to a LINEAR image, then:
// 1) BLIT the swapchain image (image1) to a temp image (image2) that is
// created with TILING_LINEAR.
// 2) Map image 2 and write the PPM file.
//
// There seems to be no way to tell if the swapchain image (image1) is tiled
// or not. We therefore assume that the BLIT operation can always read from
// both linear and optimal tiled (swapchain) images.
// There is therefore no point in looking at the BLIT_SRC properties.
//
// There is also the optimization where the incoming and target formats are
// the same. In this case, just do a COPY.
VkFormatProperties targetFormatProps;
pInstanceTable->GetPhysicalDeviceFormatProperties(
physicalDevice,
(3 == numChannels) ? target24bitFormat : target32bitFormat,
&targetFormatProps);
bool need2steps = false;
bool copyOnly = false;
if ((target24bitFormat == format) || (target32bitFormat == format)) {
copyOnly = true;
} else {
bool const bltLinear = targetFormatProps.linearTilingFeatures &
VK_FORMAT_FEATURE_BLIT_DST_BIT
? true
: false;
bool const bltOptimal = targetFormatProps.optimalTilingFeatures &
VK_FORMAT_FEATURE_BLIT_DST_BIT
? true
: false;
if (!bltLinear && !bltOptimal) {
// Cannot blit to either target tiling type. It should be pretty
// unlikely to have a device that cannot blit to either type.
// But punt by just doing a copy and possibly have the wrong
// colors. This should be quite rare.
copyOnly = true;
} else if (!bltLinear && bltOptimal) {
// Cannot blit to a linear target but can blt to optimal, so copy
// after blit is needed.
need2steps = true;
}
// Else bltLinear is available and only 1 step is needed.
}
// Put resources that need to be cleaned up in a struct with a destructor
// so that things get cleaned up when this function is exited.
WritePPMCleanupData data = {};
data.device = device;
data.pTableDevice = pTableDevice;
// Set up the image creation info for both the blit and copy images, in case
// both are needed.
VkImageCreateInfo imgCreateInfo2 = {
VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
NULL,
0,
VK_IMAGE_TYPE_2D,
VK_FORMAT_R8G8B8A8_UNORM,
{width, height, 1},
1,
1,
VK_SAMPLE_COUNT_1_BIT,
VK_IMAGE_TILING_LINEAR,
VK_IMAGE_USAGE_TRANSFER_DST_BIT,
VK_SHARING_MODE_EXCLUSIVE,
0,
NULL,
VK_IMAGE_LAYOUT_UNDEFINED,
};
VkImageCreateInfo imgCreateInfo3 = imgCreateInfo2;
// If we need both images, set up image2 to be read/write and tiled.
if (need2steps) {
imgCreateInfo2.tiling = VK_IMAGE_TILING_OPTIMAL;
imgCreateInfo2.usage =
VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT;
}
VkMemoryAllocateInfo memAllocInfo = {
VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO, NULL,
0, // allocationSize, queried later
0 // memoryTypeIndex, queried later
};
VkMemoryRequirements memRequirements;
VkPhysicalDeviceMemoryProperties memoryProperties;
// Create image2 and allocate its memory. It could be the intermediate or
// final image.
err =
pTableDevice->CreateImage(device, &imgCreateInfo2, NULL, &data.image2);
assert(!err);
if (VK_SUCCESS != err)
return;
pTableDevice->GetImageMemoryRequirements(device, data.image2,
&memRequirements);
memAllocInfo.allocationSize = memRequirements.size;
pInstanceTable->GetPhysicalDeviceMemoryProperties(physicalDevice,
&memoryProperties);
pass = memory_type_from_properties(
&memoryProperties, memRequirements.memoryTypeBits,
need2steps ? VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
: VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT,
&memAllocInfo.memoryTypeIndex);
assert(pass);
err = pTableDevice->AllocateMemory(device, &memAllocInfo, NULL, &data.mem2);
assert(!err);
if (VK_SUCCESS != err)
return;
err = pTableQueue->BindImageMemory(device, data.image2, data.mem2, 0);
assert(!err);
if (VK_SUCCESS != err)
return;
// Create image3 and allocate its memory, if needed.
if (need2steps) {
err = pTableDevice->CreateImage(device, &imgCreateInfo3, NULL,
&data.image3);
assert(!err);
if (VK_SUCCESS != err)
return;
pTableDevice->GetImageMemoryRequirements(device, data.image3,
&memRequirements);
memAllocInfo.allocationSize = memRequirements.size;
pInstanceTable->GetPhysicalDeviceMemoryProperties(physicalDevice,
&memoryProperties);
pass = memory_type_from_properties(
&memoryProperties, memRequirements.memoryTypeBits,
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT, &memAllocInfo.memoryTypeIndex);
assert(pass);
err = pTableDevice->AllocateMemory(device, &memAllocInfo, NULL,
&data.mem3);
assert(!err);
if (VK_SUCCESS != err)
return;
err = pTableQueue->BindImageMemory(device, data.image3, data.mem3, 0);
assert(!err);
if (VK_SUCCESS != err)
return;
}
// Set up the command buffer. We get a command buffer from a pool we saved
// in a hooked function, which would be the application's pool.
const VkCommandBufferAllocateInfo allocCommandBufferInfo = {
VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO, NULL,
deviceMap[device]->commandPool, VK_COMMAND_BUFFER_LEVEL_PRIMARY, 1};
data.commandPool = deviceMap[device]->commandPool;
err = pTableDevice->AllocateCommandBuffers(device, &allocCommandBufferInfo,
&data.commandBuffer);
assert(!err);
if (VK_SUCCESS != err)
return;
VkDevice cmdBuf =
static_cast<VkDevice>(static_cast<void *>(data.commandBuffer));
deviceMap.emplace(cmdBuf, devMap);
VkLayerDispatchTable *pTableCommandBuffer;
pTableCommandBuffer = get_dev_info(cmdBuf)->device_dispatch_table;
// We have just created a dispatchable object, but the dispatch table has
// not been placed in the object yet. When a "normal" application creates
// a command buffer, the dispatch table is installed by the top-level api
// binding (trampoline.c). But here, we have to do it ourselves.
if (!devMap->pfn_dev_init) {
*((const void **)data.commandBuffer) = *(void **)device;
} else {
err = devMap->pfn_dev_init(device, (void *)data.commandBuffer);
assert(!err);
}
const VkCommandBufferBeginInfo commandBufferBeginInfo = {
VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO, NULL,
VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT,
};
err = pTableCommandBuffer->BeginCommandBuffer(data.commandBuffer,
&commandBufferBeginInfo);
assert(!err);
// This barrier is used to transition from/to present Layout
VkImageMemoryBarrier presentMemoryBarrier = {
VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
NULL,
VK_ACCESS_TRANSFER_WRITE_BIT,
VK_ACCESS_TRANSFER_READ_BIT,
VK_IMAGE_LAYOUT_PRESENT_SRC_KHR,
VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
VK_QUEUE_FAMILY_IGNORED,
VK_QUEUE_FAMILY_IGNORED,
image1,
{VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1}};
// This barrier is used to transition from a newly-created layout to a blt
// or copy destination layout.
VkImageMemoryBarrier destMemoryBarrier = {
VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
NULL,
0,
VK_ACCESS_TRANSFER_WRITE_BIT,
VK_IMAGE_LAYOUT_UNDEFINED,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
VK_QUEUE_FAMILY_IGNORED,
VK_QUEUE_FAMILY_IGNORED,
data.image2,
{VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1}};
// This barrier is used to transition a dest layout to general layout.
VkImageMemoryBarrier generalMemoryBarrier = {
VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
NULL,
VK_ACCESS_TRANSFER_WRITE_BIT,
VK_ACCESS_TRANSFER_READ_BIT,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
VK_IMAGE_LAYOUT_GENERAL,
VK_QUEUE_FAMILY_IGNORED,
VK_QUEUE_FAMILY_IGNORED,
data.image2,
{VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1}};
VkPipelineStageFlags srcStages = VK_PIPELINE_STAGE_TRANSFER_BIT;
VkPipelineStageFlags dstStages = VK_PIPELINE_STAGE_TRANSFER_BIT;
// The source image needs to be transitioned from present to transfer
// source.
pTableCommandBuffer->CmdPipelineBarrier(data.commandBuffer, srcStages,
dstStages, 0, 0, NULL, 0, NULL, 1,
&presentMemoryBarrier);
// image2 needs to be transitioned from its undefined state to transfer
// destination.
pTableCommandBuffer->CmdPipelineBarrier(data.commandBuffer, srcStages,
dstStages, 0, 0, NULL, 0, NULL, 1,
&destMemoryBarrier);
const VkImageCopy imageCopyRegion = {{VK_IMAGE_ASPECT_COLOR_BIT, 0, 0, 1},
{0, 0, 0},
{VK_IMAGE_ASPECT_COLOR_BIT, 0, 0, 1},
{0, 0, 0},
{width, height, 1}};
if (copyOnly) {
pTableCommandBuffer->CmdCopyImage(
data.commandBuffer, image1, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
data.image2, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1,
&imageCopyRegion);
} else {
VkImageBlit imageBlitRegion = {};
imageBlitRegion.srcSubresource.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
imageBlitRegion.srcSubresource.baseArrayLayer = 0;
imageBlitRegion.srcSubresource.layerCount = 1;
imageBlitRegion.srcSubresource.mipLevel = 0;
imageBlitRegion.srcOffsets[1].x = width;
imageBlitRegion.srcOffsets[1].y = height;
imageBlitRegion.srcOffsets[1].z = 1;
imageBlitRegion.dstSubresource.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
imageBlitRegion.dstSubresource.baseArrayLayer = 0;
imageBlitRegion.dstSubresource.layerCount = 1;
imageBlitRegion.dstSubresource.mipLevel = 0;
imageBlitRegion.dstOffsets[1].x = width;
imageBlitRegion.dstOffsets[1].y = height;
imageBlitRegion.dstOffsets[1].z = 1;
pTableCommandBuffer->CmdBlitImage(
data.commandBuffer, image1, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
data.image2, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1,
&imageBlitRegion, VK_FILTER_NEAREST);
if (need2steps) {
// image 3 needs to be transitioned from its undefined state to a
// transfer destination.
destMemoryBarrier.image = data.image3;
pTableCommandBuffer->CmdPipelineBarrier(
data.commandBuffer, srcStages, dstStages, 0, 0, NULL, 0, NULL,
1, &destMemoryBarrier);
// Transition image2 so that it can be read for the upcoming copy to
// image 3.
destMemoryBarrier.oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL;
destMemoryBarrier.newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL;
destMemoryBarrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
destMemoryBarrier.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT;
destMemoryBarrier.image = data.image2;
pTableCommandBuffer->CmdPipelineBarrier(
data.commandBuffer, srcStages, dstStages, 0, 0, NULL, 0, NULL,
1, &destMemoryBarrier);
// This step essentially untiles the image.
pTableCommandBuffer->CmdCopyImage(
data.commandBuffer, data.image2,
VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, data.image3,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &imageCopyRegion);
generalMemoryBarrier.image = data.image3;
}
}
// The destination needs to be transitioned from the optimal copy format to
// the format we can read with the CPU.
pTableCommandBuffer->CmdPipelineBarrier(data.commandBuffer, srcStages,
dstStages, 0, 0, NULL, 0, NULL, 1,
&generalMemoryBarrier);
// Restore the swap chain image layout to what it was before.
// This may not be strictly needed, but it is generally good to restore
// things to original state.
presentMemoryBarrier.oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL;
presentMemoryBarrier.newLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR;
presentMemoryBarrier.srcAccessMask = VK_ACCESS_TRANSFER_READ_BIT;
presentMemoryBarrier.dstAccessMask = 0;
pTableCommandBuffer->CmdPipelineBarrier(data.commandBuffer, srcStages,
dstStages, 0, 0, NULL, 0, NULL, 1,
&presentMemoryBarrier);
err = pTableCommandBuffer->EndCommandBuffer(data.commandBuffer);
assert(!err);
VkFence nullFence = {VK_NULL_HANDLE};
VkSubmitInfo submitInfo;
submitInfo.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
submitInfo.pNext = NULL;
submitInfo.waitSemaphoreCount = 0;
submitInfo.pWaitSemaphores = NULL;
submitInfo.pWaitDstStageMask = NULL;
submitInfo.commandBufferCount = 1;
submitInfo.pCommandBuffers = &data.commandBuffer;
submitInfo.signalSemaphoreCount = 0;
submitInfo.pSignalSemaphores = NULL;
err = pTableQueue->QueueSubmit(queue, 1, &submitInfo, nullFence);
assert(!err);
err = pTableQueue->QueueWaitIdle(queue);
assert(!err);
err = pTableDevice->DeviceWaitIdle(device);
assert(!err);
// Map the final image so that the CPU can read it.
const VkImageSubresource sr = {VK_IMAGE_ASPECT_COLOR_BIT, 0, 0};
VkSubresourceLayout srLayout;
const char *ptr;
if (!need2steps) {
pTableDevice->GetImageSubresourceLayout(device, data.image2, &sr,
&srLayout);
err = pTableDevice->MapMemory(device, data.mem2, 0, VK_WHOLE_SIZE, 0,
(void **)&ptr);
assert(!err);
if (VK_SUCCESS != err)
return;
data.mem2mapped = true;
} else {
pTableDevice->GetImageSubresourceLayout(device, data.image3, &sr,
&srLayout);
err = pTableDevice->MapMemory(device, data.mem3, 0, VK_WHOLE_SIZE, 0,
(void **)&ptr);
assert(!err);
if (VK_SUCCESS != err)
return;
data.mem3mapped = true;
}
// Write the data to a PPM file.
ofstream file(filename, ios::binary);
file << "P6\n";
file << width << "\n";
file << height << "\n";
file << 255 << "\n";
ptr += srLayout.offset;
if (3 == numChannels) {
for (uint32_t y = 0; y < height; y++) {
file.write(ptr, 3 * width);
ptr += srLayout.rowPitch;
}
} else if (4 == numChannels) {
for (uint32_t y = 0; y < height; y++) {
const unsigned int *row = (const unsigned int *)ptr;
for (uint32_t x = 0; x < width; x++) {
file.write((char *)row, 3);
row++;
}
ptr += srLayout.rowPitch;
}
}
file.close();
// Clean up handled by ~WritePPMCleanupData()
}
static void before_present(VkQueue queue, layer_data *my_data, SwapChainData *swapChain, unsigned imageIndex) {
VkLayerDispatchTable *pTable = my_data->device_dispatch_table;
if (!my_data->fontUploadComplete) {
VkSubmitInfo si = {};
si.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
si.pNext = nullptr;
si.waitSemaphoreCount = 0;
si.commandBufferCount = 1;
si.signalSemaphoreCount = 0;
si.pCommandBuffers = &my_data->fontUploadCmdBuffer;
pTable->QueueSubmit(queue, 1, &si, VK_NULL_HANDLE);
my_data->fontUploadComplete = true;
#ifdef OVERLAY_DEBUG
printf("Font image layout transition queued\n");
#endif
}
WsiImageData *id = swapChain->presentableImages[imageIndex];
/* update the overlay content */
vertex *vertices = nullptr;
/* guaranteed not in flight due to WSI surface being available */
VkResult U_ASSERT_ONLY err =
pTable->MapMemory(my_data->dev, id->vertexBufferMemory, 0, id->vertexBufferSize, 0, (void **)&vertices);
assert(!err);
/* write vertices for string in here */
id->numVertices = fill_vertex_buffer(my_data, vertices, imageIndex);
pTable->UnmapMemory(my_data->dev, id->vertexBufferMemory);
/* JIT record a command buffer to draw the overlay */
VkCommandBufferBeginInfo cbbi;
cbbi.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO;
cbbi.pNext = nullptr;
cbbi.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT;
cbbi.pInheritanceInfo = nullptr;
VkImageMemoryBarrier imb;
imb.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER;
imb.pNext = nullptr;
imb.dstAccessMask = VK_ACCESS_MEMORY_READ_BIT;
imb.srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
imb.oldLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR;
imb.newLayout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
imb.image = id->image;
imb.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
imb.subresourceRange.baseMipLevel = 0;
imb.subresourceRange.levelCount = 1;
imb.subresourceRange.baseArrayLayer = 0;
imb.subresourceRange.layerCount = 1;
imb.srcQueueFamilyIndex = my_data->graphicsQueueFamilyIndex;
imb.dstQueueFamilyIndex = my_data->graphicsQueueFamilyIndex;
VkRenderPassBeginInfo rpbi;
rpbi.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO;
rpbi.pNext = nullptr;
rpbi.renderPass = swapChain->render_pass;
rpbi.framebuffer = id->framebuffer;
rpbi.renderArea.offset.x = 0;
rpbi.renderArea.offset.y = 0;
rpbi.renderArea.extent.width = swapChain->width;
rpbi.renderArea.extent.height = swapChain->height;
rpbi.clearValueCount = 0;
rpbi.pClearValues = nullptr;
pTable->BeginCommandBuffer(id->cmd, &cbbi);
pTable->CmdPipelineBarrier(id->cmd, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT,
0 /* dependency flags */, 0, nullptr, /* memory barriers */
0, nullptr, /* buffer memory barriers */
1, &imb); /* image memory barriers */
pTable->CmdBeginRenderPass(id->cmd, &rpbi, VK_SUBPASS_CONTENTS_INLINE);
pTable->CmdBindPipeline(id->cmd, VK_PIPELINE_BIND_POINT_GRAPHICS, swapChain->pipeline);
pTable->CmdBindDescriptorSets(id->cmd, VK_PIPELINE_BIND_POINT_GRAPHICS, my_data->pl, 0, 1, &my_data->desc_set, 0, nullptr);
VkDeviceSize offsets[] = {0};
VkBuffer buffers[] = {id->vertexBuffer};
pTable->CmdBindVertexBuffers(id->cmd, 0, 1, buffers, offsets);
pTable->CmdDraw(id->cmd, id->numVertices, 1, 0, 0);
pTable->CmdEndRenderPass(id->cmd);
pTable->EndCommandBuffer(id->cmd);
/* Schedule this command buffer for execution. TODO: Do we need to protect
* ourselves from an app that didn't wait for the presentation image to
* be idle before mangling it? If the app is well-behaved, our command
* buffer is guaranteed to have been retired before the app tries to
* present it again.
*/
VkSubmitInfo si = {};
si.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
si.pNext = nullptr;
si.waitSemaphoreCount = 0;
si.commandBufferCount = 1;
si.signalSemaphoreCount = 0;
si.pCommandBuffers = &id->cmd;
pTable->WaitForFences(my_data->dev, 1, &my_data->fence, VK_TRUE, UINT64_MAX);
pTable->ResetFences(my_data->dev, 1, &my_data->fence);
pTable->QueueSubmit(queue, 1, &si, my_data->fence);
/* Reset per-frame stats */
my_data->cmdBuffersThisFrame = 0;
}
VkResult WrappedVulkan::vkQueuePresentKHR(
VkQueue queue,
const VkPresentInfoKHR* pPresentInfo)
{
if(m_State == WRITING_IDLE)
{
RenderDoc::Inst().Tick();
GetResourceManager()->FlushPendingDirty();
}
m_FrameCounter++; // first present becomes frame #1, this function is at the end of the frame
if(pPresentInfo->swapchainCount > 1 && (m_FrameCounter % 100) == 0)
{
RDCWARN("Presenting multiple swapchains at once - only first will be processed");
}
vector<VkSwapchainKHR> unwrappedSwaps;
vector<VkSemaphore> unwrappedSems;
VkPresentInfoKHR unwrappedInfo = *pPresentInfo;
for(uint32_t i=0; i < unwrappedInfo.swapchainCount; i++)
unwrappedSwaps.push_back(Unwrap(unwrappedInfo.pSwapchains[i]));
for(uint32_t i=0; i < unwrappedInfo.waitSemaphoreCount; i++)
unwrappedSems.push_back(Unwrap(unwrappedInfo.pWaitSemaphores[i]));
unwrappedInfo.pSwapchains = unwrappedInfo.swapchainCount ? &unwrappedSwaps[0] : NULL;
unwrappedInfo.pWaitSemaphores = unwrappedInfo.waitSemaphoreCount ? &unwrappedSems[0] : NULL;
// Don't support any extensions for present info
RDCASSERT(pPresentInfo->pNext == NULL);
VkResourceRecord *swaprecord = GetRecord(pPresentInfo->pSwapchains[0]);
RDCASSERT(swaprecord->swapInfo);
SwapchainInfo &swapInfo = *swaprecord->swapInfo;
bool activeWindow = RenderDoc::Inst().IsActiveWindow(LayerDisp(m_Instance), swapInfo.wndHandle);
// need to record which image was last flipped so we can get the correct backbuffer
// for a thumbnail in EndFrameCapture
swapInfo.lastPresent = pPresentInfo->pImageIndices[0];
m_LastSwap = swaprecord->GetResourceID();
VkImage backbuffer = swapInfo.images[pPresentInfo->pImageIndices[0]].im;
if(m_State == WRITING_IDLE)
{
m_FrameTimes.push_back(m_FrameTimer.GetMilliseconds());
m_TotalTime += m_FrameTimes.back();
m_FrameTimer.Restart();
// update every second
if(m_TotalTime > 1000.0)
{
m_MinFrametime = 10000.0;
m_MaxFrametime = 0.0;
m_AvgFrametime = 0.0;
m_TotalTime = 0.0;
for(size_t i=0; i < m_FrameTimes.size(); i++)
{
m_AvgFrametime += m_FrameTimes[i];
if(m_FrameTimes[i] < m_MinFrametime)
m_MinFrametime = m_FrameTimes[i];
if(m_FrameTimes[i] > m_MaxFrametime)
m_MaxFrametime = m_FrameTimes[i];
}
m_AvgFrametime /= double(m_FrameTimes.size());
m_FrameTimes.clear();
}
uint32_t overlay = RenderDoc::Inst().GetOverlayBits();
if(overlay & eRENDERDOC_Overlay_Enabled)
{
VkRenderPass rp = swapInfo.rp;
VkImage im = swapInfo.images[pPresentInfo->pImageIndices[0]].im;
VkFramebuffer fb = swapInfo.images[pPresentInfo->pImageIndices[0]].fb;
VkLayerDispatchTable *vt = ObjDisp(GetDev());
TextPrintState textstate = { GetNextCmd(), rp, fb, swapInfo.extent.width, swapInfo.extent.height };
VkCommandBufferBeginInfo beginInfo = { VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO, NULL, VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT };
VkResult vkr = vt->BeginCommandBuffer(Unwrap(textstate.cmd), &beginInfo);
RDCASSERTEQUAL(vkr, VK_SUCCESS);
VkImageMemoryBarrier bbBarrier = {
VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, NULL,
0, 0, VK_IMAGE_LAYOUT_PRESENT_SRC_KHR, VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL,
VK_QUEUE_FAMILY_IGNORED, VK_QUEUE_FAMILY_IGNORED,
Unwrap(im),
{ VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1 }
};
bbBarrier.srcAccessMask = VK_ACCESS_ALL_READ_BITS;
bbBarrier.dstAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
DoPipelineBarrier(textstate.cmd, 1, &bbBarrier);
GetDebugManager()->BeginText(textstate);
if(activeWindow)
{
vector<RENDERDOC_InputButton> keys = RenderDoc::Inst().GetCaptureKeys();
string overlayText = "Vulkan. ";
for(size_t i=0; i < keys.size(); i++)
{
if(i > 0)
overlayText += ", ";
overlayText += ToStr::Get(keys[i]);
}
if(!keys.empty())
overlayText += " to capture.";
if(overlay & eRENDERDOC_Overlay_FrameNumber)
{
overlayText += StringFormat::Fmt(" Frame: %d.", m_FrameCounter);
}
if(overlay & eRENDERDOC_Overlay_FrameRate)
{
overlayText += StringFormat::Fmt(" %.2lf ms (%.2lf .. %.2lf) (%.0lf FPS)",
m_AvgFrametime, m_MinFrametime, m_MaxFrametime, 1000.0f/m_AvgFrametime);
}
float y=0.0f;
if(!overlayText.empty())
{
GetDebugManager()->RenderText(textstate, 0.0f, y, overlayText.c_str());
y += 1.0f;
}
if(overlay & eRENDERDOC_Overlay_CaptureList)
{
GetDebugManager()->RenderText(textstate, 0.0f, y, "%d Captures saved.\n", (uint32_t)m_FrameRecord.size());
y += 1.0f;
uint64_t now = Timing::GetUnixTimestamp();
for(size_t i=0; i < m_FrameRecord.size(); i++)
{
if(now - m_FrameRecord[i].frameInfo.captureTime < 20)
{
GetDebugManager()->RenderText(textstate, 0.0f, y, "Captured frame %d.\n", m_FrameRecord[i].frameInfo.frameNumber);
y += 1.0f;
}
}
}
#if !defined(RELEASE)
GetDebugManager()->RenderText(textstate, 0.0f, y, "%llu chunks - %.2f MB", Chunk::NumLiveChunks(), float(Chunk::TotalMem())/1024.0f/1024.0f);
y += 1.0f;
#endif
}
else
{
vector<RENDERDOC_InputButton> keys = RenderDoc::Inst().GetFocusKeys();
string str = "Vulkan. Inactive swapchain.";
for(size_t i=0; i < keys.size(); i++)
{
if(i == 0)
str += " ";
else
str += ", ";
str += ToStr::Get(keys[i]);
}
if(!keys.empty())
str += " to cycle between swapchains";
GetDebugManager()->RenderText(textstate, 0.0f, 0.0f, str.c_str());
}
GetDebugManager()->EndText(textstate);
std::swap(bbBarrier.oldLayout, bbBarrier.newLayout);
bbBarrier.srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
bbBarrier.dstAccessMask = VK_ACCESS_ALL_READ_BITS;
DoPipelineBarrier(textstate.cmd, 1, &bbBarrier);
ObjDisp(textstate.cmd)->EndCommandBuffer(Unwrap(textstate.cmd));
SubmitCmds();
FlushQ();
}
}
VkResult vkr = ObjDisp(queue)->QueuePresentKHR(Unwrap(queue), &unwrappedInfo);
if(!activeWindow)
return vkr;
RenderDoc::Inst().SetCurrentDriver(RDC_Vulkan);
// kill any current capture that isn't application defined
if(m_State == WRITING_CAPFRAME && !m_AppControlledCapture)
RenderDoc::Inst().EndFrameCapture(LayerDisp(m_Instance), swapInfo.wndHandle);
if(RenderDoc::Inst().ShouldTriggerCapture(m_FrameCounter) && m_State == WRITING_IDLE)
{
RenderDoc::Inst().StartFrameCapture(LayerDisp(m_Instance), swapInfo.wndHandle);
m_AppControlledCapture = false;
}
return vkr;
}
VK_LAYER_EXPORT VKAPI_ATTR VkResult VKAPI_CALL vkCreateSwapchainKHR(VkDevice device, const VkSwapchainCreateInfoKHR *pCreateInfo,
const VkAllocationCallbacks *pAllocator,
VkSwapchainKHR *pSwapChain) {
layer_data *my_data = GetLayerDataPtr(get_dispatch_key(device), layer_data_map);
VkLayerDispatchTable *pTable = my_data->device_dispatch_table;
VkResult result = my_data->pfnCreateSwapchainKHR(device, pCreateInfo, pAllocator, pSwapChain);
if (result == VK_SUCCESS) {
auto &data = (*my_data->swapChains)[*pSwapChain];
data = new SwapChainData;
data->width = pCreateInfo->imageExtent.width;
data->height = pCreateInfo->imageExtent.height;
data->format = pCreateInfo->imageFormat;
#ifdef OVERLAY_DEBUG
printf(
"Creating resources for scribbling on swapchain format %u width "
"%u height %u\n",
data->format, data->width, data->height);
#endif
/* Create a renderpass for drawing into this swapchain */
VkAttachmentDescription ad;
ad.flags = 0;
ad.format = data->format;
ad.samples = VK_SAMPLE_COUNT_1_BIT;
ad.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
ad.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
ad.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
ad.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
/* TODO: deal with the image possibly being in a different
* layout - we need to care about general,
* VK_IMAGE_LAYOUT_PRESENT_SOURCE_WSI,
* etc etc
*/
ad.initialLayout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
ad.finalLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR;
VkAttachmentReference ar;
ar.attachment = 0;
/* TODO: see previous layout comment */
ar.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
VkAttachmentReference dr;
dr.attachment = VK_ATTACHMENT_UNUSED;
dr.layout = VK_IMAGE_LAYOUT_UNDEFINED;
VkSubpassDescription sd;
sd.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS;
sd.flags = 0;
sd.inputAttachmentCount = 0;
sd.pInputAttachments = nullptr;
sd.colorAttachmentCount = 1;
sd.pColorAttachments = &ar;
sd.pResolveAttachments = nullptr;
sd.pDepthStencilAttachment = &dr;
/* TODO: do we need to mark the color attachment here? */
sd.preserveAttachmentCount = 0;
sd.pPreserveAttachments = nullptr;
VkRenderPassCreateInfo rpci;
rpci.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO;
rpci.pNext = nullptr;
rpci.flags = 0;
rpci.attachmentCount = 1;
rpci.pAttachments = &ad;
rpci.subpassCount = 1;
rpci.pSubpasses = &sd;
rpci.dependencyCount = 0;
rpci.pDependencies = nullptr;
pTable->CreateRenderPass(device, &rpci, nullptr, &data->render_pass);
/* Create the pipeline to use in this renderpass */
VkPipelineShaderStageCreateInfo stages[2];
stages[0].sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO;
stages[0].pNext = nullptr;
stages[0].stage = VK_SHADER_STAGE_VERTEX_BIT;
stages[0].module = my_data->vsShaderModule;
stages[0].pName = "main";
stages[0].flags = 0;
stages[0].pSpecializationInfo = nullptr;
stages[1].sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO;
stages[1].pNext = nullptr;
stages[1].stage = VK_SHADER_STAGE_FRAGMENT_BIT;
stages[1].module = my_data->fsShaderModule;
stages[1].pName = "main";
stages[1].flags = 0;
stages[1].pSpecializationInfo = nullptr;
VkPipelineInputAssemblyStateCreateInfo piasci;
memset(&piasci, 0, sizeof(piasci));
piasci.sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO;
piasci.topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST;
VkViewport viewport;
memset(&viewport, 0, sizeof(viewport));
viewport.width = (float)data->width;
viewport.height = (float)data->height;
viewport.minDepth = 0.0f;
viewport.maxDepth = 1.0f;
VkRect2D scissor;
memset(&scissor, 0, sizeof(scissor));
scissor.extent.width = data->width;
scissor.extent.height = data->height;
VkPipelineViewportStateCreateInfo pvsci;
memset(&pvsci, 0, sizeof(pvsci));
pvsci.sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO;
pvsci.viewportCount = 1;
pvsci.pViewports = &viewport;
pvsci.scissorCount = 1;
pvsci.pScissors = &scissor;
VkPipelineRasterizationStateCreateInfo prsci;
memset(&prsci, 0, sizeof(prsci));
prsci.sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO;
prsci.polygonMode = VK_POLYGON_MODE_FILL;
prsci.cullMode = VK_CULL_MODE_NONE;
prsci.lineWidth = 1.0f;
VkPipelineMultisampleStateCreateInfo pmsci;
memset(&pmsci, 0, sizeof(pmsci));
pmsci.sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO;
pmsci.rasterizationSamples = VK_SAMPLE_COUNT_1_BIT;
VkPipelineDepthStencilStateCreateInfo pdssci;
memset(&pdssci, 0, sizeof(pdssci));
pdssci.sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO;
pdssci.minDepthBounds = 0.0f;
pdssci.maxDepthBounds = 1.0f;
VkPipelineColorBlendAttachmentState pcbas;
memset(&pcbas, 0, sizeof(pcbas));
pcbas.colorWriteMask =
VK_COLOR_COMPONENT_R_BIT | VK_COLOR_COMPONENT_G_BIT | VK_COLOR_COMPONENT_B_BIT | VK_COLOR_COMPONENT_A_BIT;
VkPipelineColorBlendStateCreateInfo pcbsci;
memset(&pcbsci, 0, sizeof(pcbsci));
pcbsci.sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO;
pcbsci.attachmentCount = 1;
pcbsci.pAttachments = &pcbas;
pcbsci.blendConstants[0] = 1.0f;
pcbsci.blendConstants[1] = 1.0f;
pcbsci.blendConstants[2] = 1.0f;
pcbsci.blendConstants[3] = 1.0f;
VkVertexInputBindingDescription bindings[] = {
{0, sizeof(vertex), VK_VERTEX_INPUT_RATE_VERTEX},
};
VkVertexInputAttributeDescription attribs[] = {
{0, 0, VK_FORMAT_R32G32_SFLOAT, offsetof(vertex, x)}, {1, 0, VK_FORMAT_R32G32_SFLOAT, offsetof(vertex, u)},
};
VkPipelineVertexInputStateCreateInfo pvisci;
memset(&pvisci, 0, sizeof(pvisci));
pvisci.sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO;
pvisci.vertexBindingDescriptionCount = sizeof(bindings) / sizeof(*bindings);
pvisci.pVertexBindingDescriptions = &bindings[0];
pvisci.vertexAttributeDescriptionCount = sizeof(attribs) / sizeof(*attribs);
pvisci.pVertexAttributeDescriptions = &attribs[0];
VkGraphicsPipelineCreateInfo gpci;
gpci.sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO;
gpci.pNext = nullptr;
gpci.stageCount = sizeof(stages) / sizeof(*stages);
gpci.pStages = &stages[0];
gpci.pVertexInputState = &pvisci;
gpci.pInputAssemblyState = &piasci;
gpci.pTessellationState = nullptr;
gpci.pViewportState = &pvsci;
gpci.pRasterizationState = &prsci;
gpci.pMultisampleState = &pmsci;
gpci.pDepthStencilState = &pdssci;
gpci.pColorBlendState = &pcbsci;
gpci.pDynamicState = nullptr;
gpci.flags = 0;
gpci.layout = my_data->pl;
gpci.renderPass = data->render_pass;
gpci.subpass = 0;
gpci.basePipelineHandle = VK_NULL_HANDLE;
gpci.basePipelineIndex = 0;
pTable->CreateGraphicsPipelines(device, VK_NULL_HANDLE, 1, &gpci, nullptr, &data->pipeline);
}
return result;
}
void layer_data::Cleanup() {
VkLayerDispatchTable *pTable = this->device_dispatch_table;
pTable->DestroySampler(dev, sampler, nullptr);
pTable->DestroyDescriptorPool(dev, desc_pool, nullptr);
pTable->DestroyPipelineLayout(dev, pl, nullptr);
pTable->DestroyDescriptorSetLayout(dev, dsl, nullptr);
pTable->DestroyImageView(dev, fontGlyphsImageView, nullptr);
pTable->DestroyImage(dev, fontGlyphsImage, nullptr);
pTable->FreeMemory(dev, fontGlyphsMemory, nullptr);
pTable->FreeCommandBuffers(dev, pool, 1, &fontUploadCmdBuffer);
pTable->DestroyCommandPool(dev, pool, nullptr);
pTable->DestroyShaderModule(dev, vsShaderModule, nullptr);
pTable->DestroyShaderModule(dev, fsShaderModule, nullptr);
pTable->DestroyFence(dev, fence, nullptr);
}
static void after_device_create(VkPhysicalDevice gpu, VkDevice device, layer_data *data) {
VkResult U_ASSERT_ONLY err;
data->gpu = gpu;
data->dev = device;
data->frame = 0;
data->cmdBuffersThisFrame = 0;
VkLayerDispatchTable *pTable = data->device_dispatch_table;
/* Get our WSI hooks in. */
data->pfnCreateSwapchainKHR = (PFN_vkCreateSwapchainKHR)pTable->GetDeviceProcAddr(device, "vkCreateSwapchainKHR");
data->pfnGetSwapchainImagesKHR = (PFN_vkGetSwapchainImagesKHR)pTable->GetDeviceProcAddr(device, "vkGetSwapchainImagesKHR");
data->pfnQueuePresentKHR = (PFN_vkQueuePresentKHR)pTable->GetDeviceProcAddr(device, "vkQueuePresentKHR");
data->pfnDestroySwapchainKHR = (PFN_vkDestroySwapchainKHR)pTable->GetDeviceProcAddr(device, "vkDestroySwapchainKHR");
data->swapChains = new std::unordered_map<VkSwapchainKHR, SwapChainData *>;
/* Command pool */
VkCommandPoolCreateInfo cpci;
cpci.sType = VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO;
cpci.pNext = nullptr;
cpci.queueFamilyIndex = data->graphicsQueueFamilyIndex;
cpci.flags = VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT;
err = pTable->CreateCommandPool(device, &cpci, nullptr, &data->pool);
assert(!err);
/* Create the objects we need */
/* Compile the shaders */
compile_shader(device, VULKAN_SAMPLES_BASE_DIR "/Layer-Samples/data/overlay-vert.spv", &data->vsShaderModule);
compile_shader(device, VULKAN_SAMPLES_BASE_DIR "/Layer-Samples/data/overlay-frag.spv", &data->fsShaderModule);
/* Upload the font bitmap */
VkImageCreateInfo ici;
memset(&ici, 0, sizeof(ici));
ici.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO;
ici.imageType = VK_IMAGE_TYPE_2D;
ici.format = VK_FORMAT_R8_UNORM;
ici.extent.width = FONT_ATLAS_SIZE;
ici.extent.height = FONT_ATLAS_SIZE;
ici.extent.depth = 1;
ici.mipLevels = 1;
ici.arrayLayers = 1;
ici.samples = VK_SAMPLE_COUNT_1_BIT;
ici.tiling = VK_IMAGE_TILING_LINEAR;
ici.usage = VK_IMAGE_USAGE_SAMPLED_BIT;
ici.initialLayout = VK_IMAGE_LAYOUT_PREINITIALIZED;
err = pTable->CreateImage(device, &ici, nullptr, &data->fontGlyphsImage);
assert(!err);
VkMemoryRequirements mem_reqs;
pTable->GetImageMemoryRequirements(device, data->fontGlyphsImage, &mem_reqs);
VkMemoryAllocateInfo mem_alloc;
memset(&mem_alloc, 0, sizeof(mem_alloc));
mem_alloc.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO;
mem_alloc.allocationSize = mem_reqs.size;
mem_alloc.memoryTypeIndex = choose_memory_type(gpu, mem_reqs.memoryTypeBits,
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT);
err = pTable->AllocateMemory(device, &mem_alloc, nullptr, &data->fontGlyphsMemory);
assert(!err);
err = pTable->BindImageMemory(device, data->fontGlyphsImage, data->fontGlyphsMemory, 0);
assert(!err);
VkImageSubresource subres;
subres.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
subres.mipLevel = 0;
subres.arrayLayer = 0;
VkSubresourceLayout layout;
void *bits;
pTable->GetImageSubresourceLayout(device, data->fontGlyphsImage, &subres, &layout);
/* ensure we can directly upload into this layout */
assert(!layout.offset);
assert(layout.size >= FONT_ATLAS_SIZE * FONT_ATLAS_SIZE);
assert(layout.rowPitch == FONT_ATLAS_SIZE);
err = pTable->MapMemory(device, data->fontGlyphsMemory, 0, VK_WHOLE_SIZE, 0, &bits);
assert(!err);
/* Load the font glyphs directly into the mapped buffer */
std::vector<unsigned char> fontData;
get_file_contents(VULKAN_SAMPLES_BASE_DIR "/Layer-Samples/data/FreeSans.ttf", fontData);
stbtt_BakeFontBitmap(&fontData[0], 0, FONT_SIZE_PIXELS, (unsigned char *)bits, FONT_ATLAS_SIZE, FONT_ATLAS_SIZE, 32, 96,
data->glyphs);
pTable->UnmapMemory(device, data->fontGlyphsMemory);
VkImageViewCreateInfo ivci;
ivci.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO;
ivci.viewType = VK_IMAGE_VIEW_TYPE_2D;
ivci.pNext = nullptr;
ivci.format = ici.format;
ivci.components.r = VK_COMPONENT_SWIZZLE_R;
ivci.components.g = VK_COMPONENT_SWIZZLE_G;
ivci.components.b = VK_COMPONENT_SWIZZLE_B;
ivci.components.a = VK_COMPONENT_SWIZZLE_A;
ivci.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
ivci.subresourceRange.baseMipLevel = 0;
ivci.subresourceRange.levelCount = 1;
ivci.subresourceRange.baseArrayLayer = 0;
ivci.subresourceRange.layerCount = 1;
ivci.image = data->fontGlyphsImage;
ivci.flags = 0;
err = pTable->CreateImageView(device, &ivci, nullptr, &data->fontGlyphsImageView);
assert(!err);
/* transition from undefined layout to shader readonly so we can use it.
* requires a command buffer. */
VkCommandBufferAllocateInfo cbai;
cbai.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO;
cbai.pNext = nullptr;
cbai.commandPool = data->pool;
cbai.level = VK_COMMAND_BUFFER_LEVEL_PRIMARY;
cbai.commandBufferCount = 1;
VkCommandBuffer cmd;
err = pTable->AllocateCommandBuffers(device, &cbai, &cmd);
assert(!err);
/* We have just created a dispatchable object, but the dispatch table has
* not been placed in the object yet.
* When a "normal" application creates a command buffer,
* the dispatch table is installed by the top-level binding (trampoline.c).
* But here, we have to do it ourselves. */
if (!data->pfn_dev_init) {
*((const void **)cmd) = *(void **)device;
} else {
err = data->pfn_dev_init(device, (void *)cmd);
assert(!err);
}
VkCommandBufferBeginInfo cbbi;
cbbi.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO;
cbbi.pNext = nullptr;
cbbi.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT;
cbbi.pInheritanceInfo = nullptr;
err = pTable->BeginCommandBuffer(cmd, &cbbi);
assert(!err);
VkImageMemoryBarrier imb;
imb.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER;
imb.pNext = nullptr;
imb.dstAccessMask = VK_ACCESS_HOST_WRITE_BIT | VK_ACCESS_TRANSFER_WRITE_BIT;
imb.srcAccessMask = 0;
imb.oldLayout = VK_IMAGE_LAYOUT_UNDEFINED;
imb.newLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
imb.image = data->fontGlyphsImage;
imb.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
imb.subresourceRange.baseMipLevel = 0;
imb.subresourceRange.levelCount = 1;
imb.subresourceRange.baseArrayLayer = 0;
imb.subresourceRange.layerCount = 1;
imb.srcQueueFamilyIndex = data->graphicsQueueFamilyIndex;
imb.dstQueueFamilyIndex = data->graphicsQueueFamilyIndex;
pTable->CmdPipelineBarrier(cmd, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, 0 /* dependency flags */,
0, nullptr, /* memory barriers */
0, nullptr, /* buffer memory barriers */
1, &imb); /* image memory barriers */
pTable->EndCommandBuffer(cmd);
data->fontUploadCmdBuffer = cmd;
data->fontUploadComplete = false; /* we will schedule this at first present on this device */
#ifdef OVERLAY_DEBUG
printf("Font upload done.\n");
#endif
/* create a sampler to use with the texture */
VkSamplerCreateInfo sci;
sci.sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO;
sci.pNext = nullptr;
sci.flags = 0;
sci.magFilter = VK_FILTER_NEAREST;
sci.minFilter = VK_FILTER_NEAREST;
sci.mipmapMode = VK_SAMPLER_MIPMAP_MODE_NEAREST;
sci.addressModeU = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE;
sci.addressModeV = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE;
sci.addressModeW = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE;
sci.mipLodBias = 0.0f;
sci.anisotropyEnable = false;
sci.maxAnisotropy = 1;
sci.compareEnable = false;
sci.compareOp = VK_COMPARE_OP_NEVER;
sci.minLod = 0.0f;
sci.maxLod = 0.0f;
sci.borderColor = VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE;
sci.unnormalizedCoordinates = VK_FALSE;
err = pTable->CreateSampler(device, &sci, nullptr, &data->sampler);
assert(!err);
/* descriptor set stuff so we can use the texture from a shader. */
VkDescriptorSetLayoutBinding dslb[1];
dslb[0].binding = 0;
dslb[0].descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
dslb[0].descriptorCount = 1;
dslb[0].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT;
dslb[0].pImmutableSamplers = nullptr;
VkDescriptorSetLayoutCreateInfo dslci;
memset(&dslci, 0, sizeof(dslci));
dslci.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO;
dslci.pNext = nullptr;
dslci.bindingCount = 1;
dslci.pBindings = dslb;
err = pTable->CreateDescriptorSetLayout(device, &dslci, nullptr, &data->dsl);
assert(!err);
VkPipelineLayoutCreateInfo plci;
memset(&plci, 0, sizeof(plci));
plci.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO;
plci.setLayoutCount = 1;
plci.pSetLayouts = &data->dsl;
err = pTable->CreatePipelineLayout(device, &plci, nullptr, &data->pl);
assert(!err);
VkDescriptorPoolSize dtc[1];
dtc[0].type = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
dtc[0].descriptorCount = 1;
VkDescriptorPoolCreateInfo dpci;
dpci.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO;
dpci.pNext = nullptr;
dpci.flags = 0;
dpci.maxSets = 1;
dpci.poolSizeCount = 1;
dpci.pPoolSizes = dtc;
err = pTable->CreateDescriptorPool(device, &dpci, nullptr, &data->desc_pool);
assert(!err);
VkDescriptorSetAllocateInfo dsai;
dsai.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO;
dsai.pNext = nullptr;
dsai.descriptorPool = data->desc_pool;
dsai.descriptorSetCount = 1;
dsai.pSetLayouts = &data->dsl;
err = pTable->AllocateDescriptorSets(device, &dsai, &data->desc_set);
assert(!err);
VkDescriptorImageInfo descs[1];
descs[0].sampler = data->sampler;
descs[0].imageView = data->fontGlyphsImageView;
descs[0].imageLayout = VK_IMAGE_LAYOUT_GENERAL; // TODO: cube does this, is it correct?
VkWriteDescriptorSet writes[1];
memset(&writes, 0, sizeof(writes));
writes[0].sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
writes[0].dstSet = data->desc_set;
writes[0].dstBinding = 0;
writes[0].descriptorCount = 1;
writes[0].descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
writes[0].pImageInfo = descs;
pTable->UpdateDescriptorSets(device, 1, writes, 0, nullptr);
VkFenceCreateInfo fci;
fci.sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO;
fci.pNext = NULL;
fci.flags = VK_FENCE_CREATE_SIGNALED_BIT;
pTable->CreateFence(device, &fci, NULL, &data->fence);
}
VKAPI_ATTR VkResult VKAPI_CALL
QueuePresentKHR(VkQueue queue, const VkPresentInfoKHR *pPresentInfo) {
static int frameNumber = 0;
if (frameNumber == 10) {
fflush(stdout); /* *((int*)0)=0; */
}
DeviceMapStruct *devMap = get_dev_info((VkDevice)queue);
assert(devMap);
VkLayerDispatchTable *pDisp = devMap->device_dispatch_table;
VkResult result = pDisp->QueuePresentKHR(queue, pPresentInfo);
loader_platform_thread_lock_mutex(&globalLock);
if (!screenshotEnvQueried) {
const char *_vk_screenshot = local_getenv("_VK_SCREENSHOT");
if (_vk_screenshot && *_vk_screenshot) {
string spec(_vk_screenshot), word;
size_t start = 0, comma = 0;
while (start < spec.size()) {
int frameToAdd;
comma = spec.find(',', start);
if (comma == string::npos)
word = string(spec, start);
else
word = string(spec, start, comma - start);
frameToAdd = atoi(word.c_str());
// Add the frame number to set, but only do it if the word
// started with a digit and if
// it's not already in the list
if (*(word.c_str()) >= '0' && *(word.c_str()) <= '9') {
screenshotFrames.insert(frameToAdd);
}
if (comma == string::npos)
break;
start = comma + 1;
}
}
local_free_getenv(_vk_screenshot);
screenshotEnvQueried = true;
}
if (result == VK_SUCCESS && !screenshotFrames.empty()) {
set<int>::iterator it;
it = screenshotFrames.find(frameNumber);
if (it != screenshotFrames.end()) {
string fileName;
fileName = to_string(frameNumber) + ".ppm";
VkImage image;
VkSwapchainKHR swapchain;
// We'll dump only one image: the first
swapchain = pPresentInfo->pSwapchains[0];
image = swapchainMap[swapchain]
->imageList[pPresentInfo->pImageIndices[0]];
writePPM(fileName.c_str(), image);
screenshotFrames.erase(it);
if (screenshotFrames.empty()) {
// Free all our maps since we are done with them.
for (auto it = swapchainMap.begin(); it != swapchainMap.end();
it++) {
SwapchainMapStruct *swapchainMapElem = it->second;
delete swapchainMapElem;
}
for (auto it = imageMap.begin(); it != imageMap.end(); it++) {
ImageMapStruct *imageMapElem = it->second;
delete imageMapElem;
}
for (auto it = physDeviceMap.begin(); it != physDeviceMap.end();
it++) {
PhysDeviceMapStruct *physDeviceMapElem = it->second;
delete physDeviceMapElem;
}
swapchainMap.clear();
imageMap.clear();
physDeviceMap.clear();
}
}
}
frameNumber++;
loader_platform_thread_unlock_mutex(&globalLock);
return result;
}
static void before_present(VkQueue queue, layer_data *my_data,
SwapChainData *swapChain, unsigned imageIndex) {
VkLayerDispatchTable *pTable = my_data->device_dispatch_table;
if (!my_data->fontUploadComplete) {
VkSubmitInfo si;
si.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
si.pNext = nullptr;
si.waitSemaphoreCount = 0;
si.commandBufferCount = 1;
si.signalSemaphoreCount = 0;
si.pCommandBuffers = &my_data->fontUploadCmdBuffer;
pTable->QueueSubmit(queue, 1, &si, VK_NULL_HANDLE);
my_data->fontUploadComplete = true;
#ifdef OVERLAY_DEBUG
printf("Font image layout transition queued\n");
#endif
}
WsiImageData *id = swapChain->presentableImages[imageIndex];
/* update the overlay content */
vertex *vertices = nullptr;
/* guaranteed not in flight due to WSI surface being available */
VkResult U_ASSERT_ONLY err =
pTable->MapMemory(my_data->dev, id->vertexBufferMemory, 0,
id->vertexBufferSize, 0, (void **)&vertices);
assert(!err);
/* write vertices for string in here */
id->numVertices = fill_vertex_buffer(my_data, vertices, imageIndex);
pTable->UnmapMemory(my_data->dev, id->vertexBufferMemory);
/* JIT record a command buffer to draw the overlay */
VkCommandBufferBeginInfo cbbi;
cbbi.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO;
cbbi.pNext = nullptr;
cbbi.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT;
cbbi.pInheritanceInfo = nullptr;
VkRenderPassBeginInfo rpbi;
rpbi.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO;
rpbi.pNext = nullptr;
rpbi.renderPass = swapChain->render_pass;
rpbi.framebuffer = id->framebuffer;
rpbi.renderArea.offset.x = 0;
rpbi.renderArea.offset.y = 0;
rpbi.renderArea.extent.width = swapChain->width;
rpbi.renderArea.extent.height = swapChain->height;
rpbi.clearValueCount = 0;
rpbi.pClearValues = nullptr;
pTable->BeginCommandBuffer(id->cmd, &cbbi);
pTable->CmdBeginRenderPass(id->cmd, &rpbi, VK_SUBPASS_CONTENTS_INLINE);
pTable->CmdBindPipeline(id->cmd, VK_PIPELINE_BIND_POINT_GRAPHICS,
swapChain->pipeline);
pTable->CmdBindDescriptorSets(id->cmd, VK_PIPELINE_BIND_POINT_GRAPHICS,
my_data->pl, 0, 1, &my_data->desc_set, 0,
nullptr);
VkDeviceSize offsets[] = {0};
VkBuffer buffers[] = {id->vertexBuffer};
pTable->CmdBindVertexBuffers(id->cmd, 0, 1, buffers, offsets);
pTable->CmdDraw(id->cmd, id->numVertices, 1, 0, 0);
pTable->CmdEndRenderPass(id->cmd);
pTable->EndCommandBuffer(id->cmd);
/* Schedule this command buffer for execution. TODO: Do we need to protect
* ourselves from an app that didn't wait for the presentation image to
* be idle before mangling it? If the app is well-behaved, our command
* buffer is guaranteed to have been retired before the app tries to
* present it again.
*/
VkFence null_fence = VK_NULL_HANDLE;
VkSubmitInfo si;
si.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
si.pNext = nullptr;
si.waitSemaphoreCount = 0;
si.commandBufferCount = 1;
si.signalSemaphoreCount = 0;
si.pCommandBuffers = &id->cmd;
pTable->QueueSubmit(queue, 1, &si, null_fence);
/* Reset per-frame stats */
my_data->cmdBuffersThisFrame = 0;
}
