void Collect( VkCommandBuffer cmdbuf )
{
ZoneScopedC( Color::Red4 );
if( m_tail == m_head ) return;
#ifdef TRACY_ON_DEMAND
if( !s_profiler.IsConnected() )
{
vkCmdResetQueryPool( cmdbuf, m_query, 0, QueryCount );
m_head = m_tail = 0;
return;
}
#endif
unsigned int cnt;
if( m_oldCnt != 0 )
{
cnt = m_oldCnt;
m_oldCnt = 0;
}
else
{
cnt = m_head < m_tail ? QueryCount - m_tail : m_head - m_tail;
}
int64_t res[QueryCount];
if( vkGetQueryPoolResults( m_device, m_query, m_tail, cnt, sizeof( res ), res, sizeof( *res ), VK_QUERY_RESULT_64_BIT ) == VK_NOT_READY )
{
m_oldCnt = cnt;
return;
}
Magic magic;
auto& token = s_token.ptr;
auto& tail = token->get_tail_index();
for( unsigned int idx=0; idx<cnt; idx++ )
{
auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
MemWrite( &item->hdr.type, QueueType::GpuTime );
MemWrite( &item->gpuTime.gpuTime, res[idx] );
MemWrite( &item->gpuTime.queryId, uint16_t( m_tail + idx ) );
MemWrite( &item->gpuTime.context, m_context );
tail.store( magic + 1, std::memory_order_release );
}
vkCmdResetQueryPool( cmdbuf, m_query, m_tail, cnt );
m_tail += cnt;
if( m_tail == QueryCount ) m_tail = 0;
}
void PerfQuery::QueueCopyQueryResults(VkCommandBuffer command_buffer, VkFence fence,
u32 start_index, u32 query_count)
{
DEBUG_LOG(VIDEO, "queue copy of queries %u-%u", start_index, start_index + query_count - 1);
// Transition buffer for GPU write
// TODO: Is this needed?
m_readback_buffer->PrepareForGPUWrite(command_buffer, VK_ACCESS_TRANSFER_WRITE_BIT,
VK_PIPELINE_STAGE_TRANSFER_BIT);
// Copy from queries -> buffer
vkCmdCopyQueryPoolResults(command_buffer, m_query_pool, start_index, query_count,
m_readback_buffer->GetBuffer(), start_index * sizeof(PerfQueryDataType),
sizeof(PerfQueryDataType), VK_QUERY_RESULT_WAIT_BIT);
// Prepare for host readback
m_readback_buffer->FlushGPUCache(command_buffer, VK_ACCESS_TRANSFER_WRITE_BIT,
VK_PIPELINE_STAGE_TRANSFER_BIT);
// Reset queries so they're ready to use again
vkCmdResetQueryPool(command_buffer, m_query_pool, start_index, query_count);
// Flag all queries as available, but with a fence that has to be completed first
for (u32 i = 0; i < query_count; i++)
{
u32 index = start_index + i;
ActiveQuery& entry = m_query_buffer[index];
entry.pending_fence = fence;
entry.available = true;
entry.active = false;
}
}
nvh::Profiler::SectionID ProfilerVK::beginSection(const char* name, VkCommandBuffer cmd)
{
nvh::Profiler::gpuTimeProvider_fn fnProvider = [&](SectionID i, uint32_t queryFrame, double& gpuTime) {
uint32_t idxBegin = getTimerIdx(i, queryFrame, true);
uint32_t idxEnd = getTimerIdx(i, queryFrame, false);
uint64_t times[2];
VkResult result = vkGetQueryPoolResults(m_device, m_queryPool, idxBegin, 2, sizeof(uint64_t) * 2, times, 0, VK_QUERY_RESULT_64_BIT);
if(result == VK_SUCCESS)
{
gpuTime = (double(times[1] - times[0]) * double(m_frequency)) / double(1000);
return true;
}
else
{
return false;
}
};
SectionID slot = Profiler::beginSection(name, "VK ", fnProvider);
if (getRequiredTimers() > m_queryPoolSize) {
resize();
}
uint32_t idx = getTimerIdx(slot, getSubFrame(), true);
// clear begin and end
vkCmdResetQueryPool(cmd, m_queryPool, idx, 2); // not ideal to do this per query
vkCmdWriteTimestamp(cmd, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, m_queryPool, idx);
return slot;
}
void PerfQuery::ResetQuery()
{
m_query_count = 0;
m_query_readback_pos = 0;
m_query_next_pos = 0;
std::fill_n(m_results, ArraySize(m_results), 0);
// Reset entire query pool, ensuring all queries are ready to write to.
StateTracker::GetInstance()->EndRenderPass();
vkCmdResetQueryPool(g_command_buffer_mgr->GetCurrentCommandBuffer(), m_query_pool, 0,
PERF_QUERY_BUFFER_SIZE);
std::memset(m_query_buffer.data(), 0, sizeof(ActiveQuery) * m_query_buffer.size());
}
//------------------------------------------------------------------------------
// method for GPUInterface
//------------------------------------------------------------------------------
void RendererVk::TimerSetup(nv_helpers::Profiler::TimerIdx idx)
{
VkResult result = VK_ERROR_INITIALIZATION_FAILED;
if(m_bValid == false) return;
::VkCommandBuffer timerCmd;
timerCmd = nvk.vkAllocateCommandBuffer(m_cmdPool, true);
nvk.vkBeginCommandBuffer(timerCmd, true);
vkCmdResetQueryPool(timerCmd, m_timePool, idx, 1); // not ideal to do this per query
vkCmdWriteTimestamp(timerCmd, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, m_timePool, idx);
nvk.vkEndCommandBuffer(timerCmd);
nvk.vkQueueSubmit(NVK::VkSubmitInfo(0, NULL, NULL, 1, &timerCmd, 0, NULL), NULL);
}
void PerfQuery::ResetQuery()
{
m_query_count = 0;
m_query_read_pos = 0;
std::fill_n(m_results, ArraySize(m_results), 0);
// Reset entire query pool, ensuring all queries are ready to write to.
StateTracker::GetInstance()->EndRenderPass();
vkCmdResetQueryPool(g_command_buffer_mgr->GetCurrentCommandBuffer(), m_query_pool, 0,
PERF_QUERY_BUFFER_SIZE);
for (auto& entry : m_query_buffer)
{
entry.pending_fence = VK_NULL_HANDLE;
entry.available = false;
entry.active = false;
}
}
VkCtx( VkPhysicalDevice physdev, VkDevice device, VkQueue queue, VkCommandBuffer cmdbuf )
: m_device( device )
, m_queue( queue )
, m_context( s_gpuCtxCounter.fetch_add( 1, std::memory_order_relaxed ) )
, m_head( 0 )
, m_tail( 0 )
, m_oldCnt( 0 )
{
assert( m_context != 255 );
VkPhysicalDeviceProperties prop;
vkGetPhysicalDeviceProperties( physdev, &prop );
const float period = prop.limits.timestampPeriod;
VkQueryPoolCreateInfo poolInfo = {};
poolInfo.sType = VK_STRUCTURE_TYPE_QUERY_POOL_CREATE_INFO;
poolInfo.queryCount = QueryCount;
poolInfo.queryType = VK_QUERY_TYPE_TIMESTAMP;
vkCreateQueryPool( device, &poolInfo, nullptr, &m_query );
VkCommandBufferBeginInfo beginInfo = {};
beginInfo.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO;
beginInfo.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT;
VkSubmitInfo submitInfo = {};
submitInfo.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
submitInfo.commandBufferCount = 1;
submitInfo.pCommandBuffers = &cmdbuf;
vkBeginCommandBuffer( cmdbuf, &beginInfo );
vkCmdResetQueryPool( cmdbuf, m_query, 0, QueryCount );
vkEndCommandBuffer( cmdbuf );
vkQueueSubmit( queue, 1, &submitInfo, VK_NULL_HANDLE );
vkQueueWaitIdle( queue );
vkBeginCommandBuffer( cmdbuf, &beginInfo );
vkCmdWriteTimestamp( cmdbuf, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, m_query, 0 );
vkEndCommandBuffer( cmdbuf );
vkQueueSubmit( queue, 1, &submitInfo, VK_NULL_HANDLE );
vkQueueWaitIdle( queue );
int64_t tcpu = Profiler::GetTime();
int64_t tgpu;
vkGetQueryPoolResults( device, m_query, 0, 1, sizeof( tgpu ), &tgpu, sizeof( tgpu ), VK_QUERY_RESULT_64_BIT | VK_QUERY_RESULT_WAIT_BIT );
vkBeginCommandBuffer( cmdbuf, &beginInfo );
vkCmdResetQueryPool( cmdbuf, m_query, 0, 1 );
vkEndCommandBuffer( cmdbuf );
vkQueueSubmit( queue, 1, &submitInfo, VK_NULL_HANDLE );
vkQueueWaitIdle( queue );
Magic magic;
auto& token = s_token.ptr;
auto& tail = token->get_tail_index();
auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
MemWrite( &item->hdr.type, QueueType::GpuNewContext );
MemWrite( &item->gpuNewContext.cpuTime, tcpu );
MemWrite( &item->gpuNewContext.gpuTime, tgpu );
memset( &item->gpuNewContext.thread, 0, sizeof( item->gpuNewContext.thread ) );
MemWrite( &item->gpuNewContext.period, period );
MemWrite( &item->gpuNewContext.context, m_context );
MemWrite( &item->gpuNewContext.accuracyBits, uint8_t( 0 ) );
#ifdef TRACY_ON_DEMAND
s_profiler.DeferItem( *item );
#endif
tail.store( magic + 1, std::memory_order_release );
}
int main(int argc, char *argv[]) {
VkResult U_ASSERT_ONLY res;
bool U_ASSERT_ONLY pass;
struct sample_info info = {};
char sample_title[] = "Draw Cube";
process_command_line_args(info, argc, argv);
init_global_layer_properties(info);
info.instance_extension_names.push_back(VK_KHR_SURFACE_EXTENSION_NAME);
#ifdef _WIN32
info.instance_extension_names.push_back(
VK_KHR_WIN32_SURFACE_EXTENSION_NAME);
#else
info.instance_extension_names.push_back(VK_KHR_XCB_SURFACE_EXTENSION_NAME);
#endif
info.device_extension_names.push_back(VK_KHR_SWAPCHAIN_EXTENSION_NAME);
init_instance(info, sample_title);
init_enumerate_device(info);
init_window_size(info, 500, 500);
init_connection(info);
init_window(info);
init_swapchain_extension(info);
init_device(info);
init_command_pool(info);
init_command_buffer(info);
execute_begin_command_buffer(info);
init_device_queue(info);
init_swap_chain(info);
init_depth_buffer(info);
init_uniform_buffer(info);
init_descriptor_and_pipeline_layouts(info, false);
init_renderpass(info, DEPTH_PRESENT);
init_shaders(info, vertShaderText, fragShaderText);
init_framebuffers(info, DEPTH_PRESENT);
init_vertex_buffer(info, g_vb_solid_face_colors_Data,
sizeof(g_vb_solid_face_colors_Data),
sizeof(g_vb_solid_face_colors_Data[0]), false);
init_descriptor_pool(info, false);
init_descriptor_set(info, false);
init_pipeline_cache(info);
init_pipeline(info, DEPTH_PRESENT);
/* VULKAN_KEY_START */
VkClearValue clear_values[2];
clear_values[0].color.float32[0] = 0.2f;
clear_values[0].color.float32[1] = 0.2f;
clear_values[0].color.float32[2] = 0.2f;
clear_values[0].color.float32[3] = 0.2f;
clear_values[1].depthStencil.depth = 1.0f;
clear_values[1].depthStencil.stencil = 0;
VkSemaphore presentCompleteSemaphore;
VkSemaphoreCreateInfo presentCompleteSemaphoreCreateInfo;
presentCompleteSemaphoreCreateInfo.sType =
VK_STRUCTURE_TYPE_SEMAPHORE_CREATE_INFO;
presentCompleteSemaphoreCreateInfo.pNext = NULL;
presentCompleteSemaphoreCreateInfo.flags = VK_FENCE_CREATE_SIGNALED_BIT;
res = vkCreateSemaphore(info.device, &presentCompleteSemaphoreCreateInfo,
NULL, &presentCompleteSemaphore);
assert(res == VK_SUCCESS);
// Get the index of the next available swapchain image:
res = vkAcquireNextImageKHR(info.device, info.swap_chain, UINT64_MAX,
presentCompleteSemaphore, NULL,
&info.current_buffer);
// TODO: Deal with the VK_SUBOPTIMAL_KHR and VK_ERROR_OUT_OF_DATE_KHR
// return codes
assert(res == VK_SUCCESS);
/* Allocate a uniform buffer that will take query results. */
VkBuffer query_result_buf;
VkDeviceMemory query_result_mem;
VkBufferCreateInfo buf_info = {};
buf_info.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
buf_info.pNext = NULL;
buf_info.usage =
VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT;
buf_info.size = 4 * sizeof(uint64_t);
buf_info.queueFamilyIndexCount = 0;
buf_info.pQueueFamilyIndices = NULL;
buf_info.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
buf_info.flags = 0;
res = vkCreateBuffer(info.device, &buf_info, NULL, &query_result_buf);
assert(res == VK_SUCCESS);
VkMemoryRequirements mem_reqs;
vkGetBufferMemoryRequirements(info.device, query_result_buf, &mem_reqs);
VkMemoryAllocateInfo alloc_info = {};
alloc_info.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO;
alloc_info.pNext = NULL;
alloc_info.memoryTypeIndex = 0;
alloc_info.allocationSize = mem_reqs.size;
pass = memory_type_from_properties(info, mem_reqs.memoryTypeBits,
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT,
&alloc_info.memoryTypeIndex);
assert(pass);
res = vkAllocateMemory(info.device, &alloc_info, NULL, &query_result_mem);
assert(res == VK_SUCCESS);
res =
vkBindBufferMemory(info.device, query_result_buf, query_result_mem, 0);
assert(res == VK_SUCCESS);
VkQueryPool query_pool;
VkQueryPoolCreateInfo query_pool_info;
query_pool_info.sType = VK_STRUCTURE_TYPE_QUERY_POOL_CREATE_INFO;
query_pool_info.pNext = NULL;
query_pool_info.queryType = VK_QUERY_TYPE_OCCLUSION;
query_pool_info.flags = 0;
query_pool_info.queryCount = 2;
query_pool_info.pipelineStatistics = 0;
res = vkCreateQueryPool(info.device, &query_pool_info, NULL, &query_pool);
assert(res == VK_SUCCESS);
vkCmdResetQueryPool(info.cmd, query_pool, 0 /*startQuery*/,
2 /*queryCount*/);
VkRenderPassBeginInfo rp_begin;
rp_begin.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO;
rp_begin.pNext = NULL;
rp_begin.renderPass = info.render_pass;
rp_begin.framebuffer = info.framebuffers[info.current_buffer];
rp_begin.renderArea.offset.x = 0;
rp_begin.renderArea.offset.y = 0;
rp_begin.renderArea.extent.width = info.width;
rp_begin.renderArea.extent.height = info.height;
rp_begin.clearValueCount = 2;
rp_begin.pClearValues = clear_values;
vkCmdBeginRenderPass(info.cmd, &rp_begin, VK_SUBPASS_CONTENTS_INLINE);
vkCmdBindPipeline(info.cmd, VK_PIPELINE_BIND_POINT_GRAPHICS, info.pipeline);
vkCmdBindDescriptorSets(info.cmd, VK_PIPELINE_BIND_POINT_GRAPHICS,
info.pipeline_layout, 0, NUM_DESCRIPTOR_SETS,
info.desc_set.data(), 0, NULL);
const VkDeviceSize offsets[1] = {0};
vkCmdBindVertexBuffers(info.cmd, 0, 1, &info.vertex_buffer.buf, offsets);
VkViewport viewport;
viewport.height = (float)info.height;
viewport.width = (float)info.width;
viewport.minDepth = (float)0.0f;
viewport.maxDepth = (float)1.0f;
viewport.x = 0;
viewport.y = 0;
vkCmdSetViewport(info.cmd, 0, NUM_VIEWPORTS, &viewport);
VkRect2D scissor;
scissor.extent.width = info.width;
scissor.extent.height = info.height;
scissor.offset.x = 0;
scissor.offset.y = 0;
vkCmdSetScissor(info.cmd, 0, NUM_SCISSORS, &scissor);
vkCmdBeginQuery(info.cmd, query_pool, 0 /*slot*/, 0 /*flags*/);
vkCmdEndQuery(info.cmd, query_pool, 0 /*slot*/);
vkCmdBeginQuery(info.cmd, query_pool, 1 /*slot*/, 0 /*flags*/);
vkCmdDraw(info.cmd, 12 * 3, 1, 0, 0);
vkCmdEndRenderPass(info.cmd);
vkCmdEndQuery(info.cmd, query_pool, 1 /*slot*/);
vkCmdCopyQueryPoolResults(
info.cmd, query_pool, 0 /*firstQuery*/, 2 /*queryCount*/,
query_result_buf, 0 /*dstOffset*/, sizeof(uint64_t) /*stride*/,
VK_QUERY_RESULT_64_BIT | VK_QUERY_RESULT_WAIT_BIT);
VkImageMemoryBarrier prePresentBarrier = {};
prePresentBarrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER;
prePresentBarrier.pNext = NULL;
prePresentBarrier.srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
prePresentBarrier.dstAccessMask = 0;
prePresentBarrier.oldLayout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
prePresentBarrier.newLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR;
prePresentBarrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
prePresentBarrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
prePresentBarrier.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
prePresentBarrier.subresourceRange.baseMipLevel = 0;
prePresentBarrier.subresourceRange.levelCount = 1;
prePresentBarrier.subresourceRange.baseArrayLayer = 0;
prePresentBarrier.subresourceRange.layerCount = 1;
prePresentBarrier.image = info.buffers[info.current_buffer].image;
vkCmdPipelineBarrier(info.cmd, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, 0, 0, NULL, 0, NULL,
1, &prePresentBarrier);
res = vkEndCommandBuffer(info.cmd);
const VkCommandBuffer cmd_bufs[] = {info.cmd};
VkFenceCreateInfo fenceInfo;
VkFence drawFence;
fenceInfo.sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO;
fenceInfo.pNext = NULL;
fenceInfo.flags = 0;
vkCreateFence(info.device, &fenceInfo, NULL, &drawFence);
VkPipelineStageFlags pipe_stage_flags =
VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT;
VkSubmitInfo submit_info[1] = {};
submit_info[0].pNext = NULL;
submit_info[0].sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
submit_info[0].waitSemaphoreCount = 1;
submit_info[0].pWaitSemaphores = &presentCompleteSemaphore;
submit_info[0].pWaitDstStageMask = &pipe_stage_flags;
submit_info[0].commandBufferCount = 1;
submit_info[0].pCommandBuffers = cmd_bufs;
submit_info[0].signalSemaphoreCount = 0;
submit_info[0].pSignalSemaphores = NULL;
/* Queue the command buffer for execution */
res = vkQueueSubmit(info.queue, 1, submit_info, drawFence);
assert(res == VK_SUCCESS);
res = vkQueueWaitIdle(info.queue);
assert(res == VK_SUCCESS);
uint64_t samples_passed[4];
samples_passed[0] = 0;
samples_passed[1] = 0;
res = vkGetQueryPoolResults(
info.device, query_pool, 0 /*firstQuery*/, 2 /*queryCount*/,
sizeof(samples_passed) /*dataSize*/, samples_passed,
sizeof(uint64_t) /*stride*/,
VK_QUERY_RESULT_64_BIT | VK_QUERY_RESULT_WAIT_BIT);
assert(res == VK_SUCCESS);
std::cout << "vkGetQueryPoolResults data"
<< "\n";
std::cout << "samples_passed[0] = " << samples_passed[0] << "\n";
std::cout << "samples_passed[1] = " << samples_passed[1] << "\n";
/* Read back query result from buffer */
uint64_t *samples_passed_ptr;
res = vkMapMemory(info.device, query_result_mem, 0, mem_reqs.size, 0,
(void **)&samples_passed_ptr);
assert(res == VK_SUCCESS);
std::cout << "vkCmdCopyQueryPoolResults data"
<< "\n";
std::cout << "samples_passed[0] = " << samples_passed_ptr[0] << "\n";
std::cout << "samples_passed[1] = " << samples_passed_ptr[1] << "\n";
vkUnmapMemory(info.device, query_result_mem);
/* Now present the image in the window */
VkPresentInfoKHR present;
present.sType = VK_STRUCTURE_TYPE_PRESENT_INFO_KHR;
present.pNext = NULL;
present.swapchainCount = 1;
present.pSwapchains = &info.swap_chain;
present.pImageIndices = &info.current_buffer;
present.pWaitSemaphores = NULL;
present.waitSemaphoreCount = 0;
present.pResults = NULL;
/* Make sure command buffer is finished before presenting */
do {
res =
vkWaitForFences(info.device, 1, &drawFence, VK_TRUE, FENCE_TIMEOUT);
} while (res == VK_TIMEOUT);
assert(res == VK_SUCCESS);
res = vkQueuePresentKHR(info.queue, &present);
assert(res == VK_SUCCESS);
wait_seconds(1);
/* VULKAN_KEY_END */
if (info.save_images)
write_ppm(info, "occlusion_query");
vkDestroyBuffer(info.device, query_result_buf, NULL);
vkFreeMemory(info.device, query_result_mem, NULL);
vkDestroySemaphore(info.device, presentCompleteSemaphore, NULL);
vkDestroyQueryPool(info.device, query_pool, NULL);
vkDestroyFence(info.device, drawFence, NULL);
destroy_pipeline(info);
destroy_pipeline_cache(info);
destroy_descriptor_pool(info);
destroy_vertex_buffer(info);
destroy_framebuffers(info);
destroy_shaders(info);
destroy_renderpass(info);
destroy_descriptor_and_pipeline_layouts(info);
destroy_uniform_buffer(info);
destroy_depth_buffer(info);
destroy_swap_chain(info);
destroy_command_buffer(info);
destroy_command_pool(info);
destroy_device(info);
destroy_window(info);
destroy_instance(info);
return 0;
}
void buildCommandBuffers()
{
VkCommandBufferBeginInfo cmdBufInfo = vks::initializers::commandBufferBeginInfo();
VkClearValue clearValues[2];
clearValues[0].color = defaultClearColor;
clearValues[1].depthStencil = { 1.0f, 0 };
VkRenderPassBeginInfo renderPassBeginInfo = vks::initializers::renderPassBeginInfo();
renderPassBeginInfo.renderPass = renderPass;
renderPassBeginInfo.renderArea.offset.x = 0;
renderPassBeginInfo.renderArea.offset.y = 0;
renderPassBeginInfo.renderArea.extent.width = width;
renderPassBeginInfo.renderArea.extent.height = height;
renderPassBeginInfo.clearValueCount = 2;
renderPassBeginInfo.pClearValues = clearValues;
for (int32_t i = 0; i < drawCmdBuffers.size(); ++i) {
renderPassBeginInfo.framebuffer = frameBuffers[i];
VK_CHECK_RESULT(vkBeginCommandBuffer(drawCmdBuffers[i], &cmdBufInfo));
// Reset timestamp query pool
vkCmdResetQueryPool(drawCmdBuffers[i], queryPool, 0, static_cast<uint32_t>(pipelineStats.size()));
vkCmdBeginRenderPass(drawCmdBuffers[i], &renderPassBeginInfo, VK_SUBPASS_CONTENTS_INLINE);
VkViewport viewport = vks::initializers::viewport((float)width, (float)height, 0.0f, 1.0f);
vkCmdSetViewport(drawCmdBuffers[i], 0, 1, &viewport);
VkRect2D scissor = vks::initializers::rect2D(width, height, 0, 0);
vkCmdSetScissor(drawCmdBuffers[i], 0, 1, &scissor);
VkDeviceSize offsets[1] = { 0 };
// Start capture of pipeline statistics
vkCmdBeginQuery(drawCmdBuffers[i], queryPool, 0, VK_QUERY_CONTROL_PRECISE_BIT);
vkCmdBindPipeline(drawCmdBuffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
vkCmdBindDescriptorSets(drawCmdBuffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, pipelineLayout, 0, 1, &descriptorSet, 0, NULL);
vkCmdBindVertexBuffers(drawCmdBuffers[i], 0, 1, &models.objects[models.objectIndex].vertices.buffer, offsets);
vkCmdBindIndexBuffer(drawCmdBuffers[i], models.objects[models.objectIndex].indices.buffer, 0, VK_INDEX_TYPE_UINT32);
for (int32_t y = 0; y < gridSize; y++) {
for (int32_t x = 0; x < gridSize; x++) {
glm::vec3 pos = glm::vec3(float(x - (gridSize / 2.0f)) * 2.5f, 0.0f, float(y - (gridSize / 2.0f)) * 2.5f);
vkCmdPushConstants(drawCmdBuffers[i], pipelineLayout, VK_SHADER_STAGE_VERTEX_BIT, 0, sizeof(glm::vec3), &pos);
vkCmdDrawIndexed(drawCmdBuffers[i], models.objects[models.objectIndex].indexCount, 1, 0, 0, 0);
}
}
// End capture of pipeline statistics
vkCmdEndQuery(drawCmdBuffers[i], queryPool, 0);
drawUI(drawCmdBuffers[i]);
vkCmdEndRenderPass(drawCmdBuffers[i]);
VK_CHECK_RESULT(vkEndCommandBuffer(drawCmdBuffers[i]));
}
}
void buildCommandBuffers()
{
VkCommandBufferBeginInfo cmdBufInfo = vks::initializers::commandBufferBeginInfo();
VkClearValue clearValues[2];
clearValues[0].color = defaultClearColor;
clearValues[1].depthStencil = { 1.0f, 0 };
VkRenderPassBeginInfo renderPassBeginInfo = vks::initializers::renderPassBeginInfo();
renderPassBeginInfo.renderPass = renderPass;
renderPassBeginInfo.renderArea.offset.x = 0;
renderPassBeginInfo.renderArea.offset.y = 0;
renderPassBeginInfo.renderArea.extent.width = width;
renderPassBeginInfo.renderArea.extent.height = height;
renderPassBeginInfo.clearValueCount = 2;
renderPassBeginInfo.pClearValues = clearValues;
for (int32_t i = 0; i < drawCmdBuffers.size(); ++i)
{
renderPassBeginInfo.framebuffer = frameBuffers[i];
VK_CHECK_RESULT(vkBeginCommandBuffer(drawCmdBuffers[i], &cmdBufInfo));
if (deviceFeatures.pipelineStatisticsQuery) {
vkCmdResetQueryPool(drawCmdBuffers[i], queryPool, 0, 2);
}
vkCmdBeginRenderPass(drawCmdBuffers[i], &renderPassBeginInfo, VK_SUBPASS_CONTENTS_INLINE);
VkViewport viewport = vks::initializers::viewport((float)width, (float)height, 0.0f, 1.0f);
vkCmdSetViewport(drawCmdBuffers[i], 0, 1, &viewport);
VkRect2D scissor = vks::initializers::rect2D(width, height, 0, 0);
vkCmdSetScissor(drawCmdBuffers[i], 0, 1, &scissor);
vkCmdSetLineWidth(drawCmdBuffers[i], 1.0f);
VkDeviceSize offsets[1] = { 0 };
// Skysphere
vkCmdBindPipeline(drawCmdBuffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, pipelines.skysphere);
vkCmdBindDescriptorSets(drawCmdBuffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, pipelineLayouts.skysphere, 0, 1, &descriptorSets.skysphere, 0, NULL);
vkCmdBindVertexBuffers(drawCmdBuffers[i], VERTEX_BUFFER_BIND_ID, 1, &models.skysphere.vertices.buffer, offsets);
vkCmdBindIndexBuffer(drawCmdBuffers[i], models.skysphere.indices.buffer, 0, VK_INDEX_TYPE_UINT32);
vkCmdDrawIndexed(drawCmdBuffers[i], models.skysphere.indexCount, 1, 0, 0, 0);
// Terrain
if (deviceFeatures.pipelineStatisticsQuery) {
// Begin pipeline statistics query
vkCmdBeginQuery(drawCmdBuffers[i], queryPool, 0, VK_QUERY_CONTROL_PRECISE_BIT);
}
// Render
vkCmdBindPipeline(drawCmdBuffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, wireframe ? pipelines.wireframe : pipelines.terrain);
vkCmdBindDescriptorSets(drawCmdBuffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, pipelineLayouts.terrain, 0, 1, &descriptorSets.terrain, 0, NULL);
vkCmdBindVertexBuffers(drawCmdBuffers[i], VERTEX_BUFFER_BIND_ID, 1, &models.terrain.vertices.buffer, offsets);
vkCmdBindIndexBuffer(drawCmdBuffers[i], models.terrain.indices.buffer, 0, VK_INDEX_TYPE_UINT32);
vkCmdDrawIndexed(drawCmdBuffers[i], models.terrain.indexCount, 1, 0, 0, 0);
if (deviceFeatures.pipelineStatisticsQuery) {
// End pipeline statistics query
vkCmdEndQuery(drawCmdBuffers[i], queryPool, 0);
}
drawUI(drawCmdBuffers[i]);
vkCmdEndRenderPass(drawCmdBuffers[i]);
VK_CHECK_RESULT(vkEndCommandBuffer(drawCmdBuffers[i]));
}
}
void VulkanQuery::reset(VkCommandBuffer cmdBuf)
{
vkCmdResetQueryPool(cmdBuf, mPool, mQueryIdx, 1);
}
void buildCommandBuffers()
{
VkCommandBufferBeginInfo cmdBufInfo = vkTools::initializers::commandBufferBeginInfo();
VkClearValue clearValues[2];
clearValues[0].color = defaultClearColor;
clearValues[1].depthStencil = { 1.0f, 0 };
VkRenderPassBeginInfo renderPassBeginInfo = vkTools::initializers::renderPassBeginInfo();
renderPassBeginInfo.renderPass = renderPass;
renderPassBeginInfo.renderArea.offset.x = 0;
renderPassBeginInfo.renderArea.offset.y = 0;
renderPassBeginInfo.renderArea.extent.width = width;
renderPassBeginInfo.renderArea.extent.height = height;
renderPassBeginInfo.clearValueCount = 2;
renderPassBeginInfo.pClearValues = clearValues;
VkResult err;
for (int32_t i = 0; i < drawCmdBuffers.size(); ++i)
{
// Set target frame buffer
renderPassBeginInfo.framebuffer = frameBuffers[i];
err = vkBeginCommandBuffer(drawCmdBuffers[i], &cmdBufInfo);
assert(!err);
// Reset query pool
// Must be done outside of render pass
vkCmdResetQueryPool(drawCmdBuffers[i], queryPool, 0, 2);
vkCmdBeginRenderPass(drawCmdBuffers[i], &renderPassBeginInfo, VK_SUBPASS_CONTENTS_INLINE);
VkViewport viewport = vkTools::initializers::viewport(
(float)width,
(float)height,
0.0f,
1.0f);
vkCmdSetViewport(drawCmdBuffers[i], 0, 1, &viewport);
VkRect2D scissor = vkTools::initializers::rect2D(
width,
height,
0,
0);
vkCmdSetScissor(drawCmdBuffers[i], 0, 1, &scissor);
VkDeviceSize offsets[1] = { 0 };
glm::mat4 modelMatrix = glm::mat4();
// Occlusion pass
vkCmdBindPipeline(drawCmdBuffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, pipelines.simple);
// Occluder first
vkCmdBindDescriptorSets(drawCmdBuffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, pipelineLayout, 0, 1, &descriptorSet, 0, NULL);
vkCmdBindVertexBuffers(drawCmdBuffers[i], VERTEX_BUFFER_BIND_ID, 1, &meshes.plane.vertices.buf, offsets);
vkCmdBindIndexBuffer(drawCmdBuffers[i], meshes.plane.indices.buf, 0, VK_INDEX_TYPE_UINT32);
vkCmdDrawIndexed(drawCmdBuffers[i], meshes.plane.indexCount, 1, 0, 0, 0);
// Teapot
vkCmdBeginQuery(drawCmdBuffers[i], queryPool, 0, VK_FLAGS_NONE);
vkCmdBindDescriptorSets(drawCmdBuffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, pipelineLayout, 0, 1, &descriptorSets.teapot, 0, NULL);
vkCmdBindVertexBuffers(drawCmdBuffers[i], VERTEX_BUFFER_BIND_ID, 1, &meshes.teapot.vertices.buf, offsets);
vkCmdBindIndexBuffer(drawCmdBuffers[i], meshes.teapot.indices.buf, 0, VK_INDEX_TYPE_UINT32);
vkCmdDrawIndexed(drawCmdBuffers[i], meshes.teapot.indexCount, 1, 0, 0, 0);
vkCmdEndQuery(drawCmdBuffers[i], queryPool, 0);
// Sphere
vkCmdBeginQuery(drawCmdBuffers[i], queryPool, 1, VK_FLAGS_NONE);
vkCmdBindDescriptorSets(drawCmdBuffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, pipelineLayout, 0, 1, &descriptorSets.sphere, 0, NULL);
vkCmdBindVertexBuffers(drawCmdBuffers[i], VERTEX_BUFFER_BIND_ID, 1, &meshes.sphere.vertices.buf, offsets);
vkCmdBindIndexBuffer(drawCmdBuffers[i], meshes.sphere.indices.buf, 0, VK_INDEX_TYPE_UINT32);
vkCmdDrawIndexed(drawCmdBuffers[i], meshes.sphere.indexCount, 1, 0, 0, 0);
vkCmdEndQuery(drawCmdBuffers[i], queryPool, 1);
// Query results
vkCmdCopyQueryPoolResults(
drawCmdBuffers[i],
queryPool,
0,
2,
queryResult.buffer,
0,
sizeof(uint64_t),
VK_QUERY_RESULT_64_BIT | VK_QUERY_RESULT_WAIT_BIT);
// Visible pass
// Clear color and depth attachments
VkClearAttachment clearAttachments[2] = {};
clearAttachments[0].aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
clearAttachments[0].clearValue.color = defaultClearColor;
clearAttachments[0].colorAttachment = 0;
clearAttachments[1].aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT;
clearAttachments[1].clearValue.depthStencil = { 1.0f, 0 };
VkClearRect clearRect = {};
clearRect.layerCount = 1;
clearRect.rect.offset = { 0, 0 };
clearRect.rect.extent = { width, height };
vkCmdClearAttachments(
drawCmdBuffers[i],
2,
clearAttachments,
1,
&clearRect);
vkCmdBindPipeline(drawCmdBuffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, pipelines.solid);
// Teapot
vkCmdBindDescriptorSets(drawCmdBuffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, pipelineLayout, 0, 1, &descriptorSets.teapot, 0, NULL);
vkCmdBindVertexBuffers(drawCmdBuffers[i], VERTEX_BUFFER_BIND_ID, 1, &meshes.teapot.vertices.buf, offsets);
vkCmdBindIndexBuffer(drawCmdBuffers[i], meshes.teapot.indices.buf, 0, VK_INDEX_TYPE_UINT32);
vkCmdDrawIndexed(drawCmdBuffers[i], meshes.teapot.indexCount, 1, 0, 0, 0);
// Sphere
vkCmdBindDescriptorSets(drawCmdBuffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, pipelineLayout, 0, 1, &descriptorSets.sphere, 0, NULL);
vkCmdBindVertexBuffers(drawCmdBuffers[i], VERTEX_BUFFER_BIND_ID, 1, &meshes.sphere.vertices.buf, offsets);
vkCmdBindIndexBuffer(drawCmdBuffers[i], meshes.sphere.indices.buf, 0, VK_INDEX_TYPE_UINT32);
vkCmdDrawIndexed(drawCmdBuffers[i], meshes.sphere.indexCount, 1, 0, 0, 0);
// Occluder
vkCmdBindPipeline(drawCmdBuffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, pipelines.occluder);
vkCmdBindDescriptorSets(drawCmdBuffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, pipelineLayout, 0, 1, &descriptorSet, 0, NULL);
vkCmdBindVertexBuffers(drawCmdBuffers[i], VERTEX_BUFFER_BIND_ID, 1, &meshes.plane.vertices.buf, offsets);
vkCmdBindIndexBuffer(drawCmdBuffers[i], meshes.plane.indices.buf, 0, VK_INDEX_TYPE_UINT32);
vkCmdDrawIndexed(drawCmdBuffers[i], meshes.plane.indexCount, 1, 0, 0, 0);
vkCmdEndRenderPass(drawCmdBuffers[i]);
VkImageMemoryBarrier prePresentBarrier = vkTools::prePresentBarrier(swapChain.buffers[i].image);
vkCmdPipelineBarrier(
drawCmdBuffers[i],
VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT,
VK_FLAGS_NONE,
0, nullptr,
0, nullptr,
1, &prePresentBarrier);
err = vkEndCommandBuffer(drawCmdBuffers[i]);
assert(!err);
}
}
