Refactored compute ray tracing example, use dedicated graphics queue if available

This commit is contained in:
saschawillems 2016-08-20 12:02:02 +02:00
parent 1c9492acba
commit a104110b30
5 changed files with 163 additions and 254 deletions

View file

@ -3,9 +3,6 @@
#extension GL_ARB_separate_shader_objects : enable
#extension GL_ARB_shading_language_420pack : enable
layout (location = 0) in vec3 inPos;
layout (location = 1) in vec2 inUV;
layout (location = 0) out vec2 outUV;
out gl_PerVertex
@ -15,6 +12,6 @@ out gl_PerVertex
void main()
{
outUV = inUV;
gl_Position = vec4(inPos.xyz, 1.0);
outUV = vec2((gl_VertexIndex << 1) & 2, gl_VertexIndex & 2);
gl_Position = vec4(outUV * 2.0f + -1.0f, 0.0f, 1.0f);
}

View file

@ -23,35 +23,43 @@
#define VERTEX_BUFFER_BIND_ID 0
#define ENABLE_VALIDATION false
#if defined(__ANDROID__)
#define TEX_DIM 1024
#else
#define TEX_DIM 2048
// Vertex layout for this example
struct Vertex {
float pos[3];
float uv[2];
};
#endif
class VulkanExample : public VulkanExampleBase
{
private:
vkTools::VulkanTexture textureComputeTarget;
public:
vkTools::VulkanTexture textureComputeTarget;
// Resources for the graphics part of the example
struct {
VkPipelineVertexInputStateCreateInfo inputState;
std::vector<VkVertexInputBindingDescription> bindingDescriptions;
std::vector<VkVertexInputAttributeDescription> attributeDescriptions;
} vertices;
VkDescriptorSetLayout descriptorSetLayout; // Raytraced image display shader binding layout
VkDescriptorSet descriptorSetPreCompute; // Raytraced image display shader bindings before compute shader image manipulation
VkDescriptorSet descriptorSet; // Raytraced image display shader bindings after compute shader image manipulation
VkPipeline pipeline; // Raytraced image display pipeline
VkPipelineLayout pipelineLayout; // Layout of the graphics pipeline
} graphics;
// Resources for the compute part of the example
struct {
VkQueue queue; // Separate queue for compute commands (queue family may differ from the one used for graphics)
VkCommandPool commandPool; // Use a separate command pool (queue family may differ from the one used for graphics)
VkCommandBuffer commandBuffer; // Command buffer storing the dispatch commands and barriers
VkFence fence; // Synchronization fence to avoid rewriting compute CB if still in use
VkDescriptorSetLayout descriptorSetLayout; // Compute shader binding layout
VkDescriptorSet descriptorSet; // Compute shader bindings
VkPipelineLayout pipelineLayout; // Layout of the compute pipeline
VkPipeline pipeline; // Compute raytracing pipeline
} compute;
vk::Buffer uniformDataCompute;
struct {
vkMeshLoader::MeshBuffer quad;
} meshes;
vkTools::UniformData uniformDataCompute;
struct {
glm::vec3 lightPos;
// Aspect ratio of the viewport
float aspectRatio;
glm::vec3 lightPos;
float aspectRatio; // Aspect ratio of the viewport
glm::vec4 fogColor = glm::vec4(0.0f);
struct {
glm::vec3 pos = glm::vec3(0.0f, 1.5f, 4.0f);
@ -60,28 +68,10 @@ public:
} camera;
} uboCompute;
struct {
VkPipeline display;
VkPipeline compute;
} pipelines;
int vertexBufferSize;
VkQueue computeQueue;
VkCommandBuffer computeCmdBuffer;
VkPipelineLayout computePipelineLayout;
VkDescriptorSet computeDescriptorSet;
VkDescriptorSetLayout computeDescriptorSetLayout;
VkDescriptorPool computeDescriptorPool;
VkPipelineLayout pipelineLayout;
VkDescriptorSet descriptorSetPostCompute;
VkDescriptorSetLayout descriptorSetLayout;
VulkanExample() : VulkanExampleBase(ENABLE_VALIDATION)
{
zoom = -2.0f;
title = "Vulkan Example - Compute shader ray tracing";
enableTextOverlay = true;
uboCompute.aspectRatio = (float)width / (float)height;
paused = true;
timerSpeed *= 0.5f;
@ -89,26 +79,20 @@ public:
~VulkanExample()
{
// Clean up used Vulkan resources
// Note : Inherited destructor cleans up resources stored in base class
vkDestroyPipeline(device, pipelines.display, nullptr);
vkDestroyPipeline(device, pipelines.compute, nullptr);
vkDestroyPipelineLayout(device, pipelineLayout, nullptr);
vkDestroyDescriptorSetLayout(device, descriptorSetLayout, nullptr);
vkMeshLoader::freeMeshBufferResources(device, &meshes.quad);
vkTools::destroyUniformData(device, &uniformDataCompute);
vkFreeCommandBuffers(device, cmdPool, 1, &computeCmdBuffer);
textureLoader->destroyTexture(textureComputeTarget);
// Graphics
vkDestroyPipeline(device, graphics.pipeline, nullptr);
vkDestroyPipelineLayout(device, graphics.pipelineLayout, nullptr);
vkDestroyDescriptorSetLayout(device, graphics.descriptorSetLayout, nullptr);
// Compute
vkDestroyPipelineLayout(device, computePipelineLayout, nullptr);
vkDestroyDescriptorSetLayout(device, computeDescriptorSetLayout, nullptr);
vkDestroyPipeline(device, compute.pipeline, nullptr);
vkDestroyPipelineLayout(device, compute.pipelineLayout, nullptr);
vkDestroyDescriptorSetLayout(device, compute.descriptorSetLayout, nullptr);
vkDestroyFence(device, compute.fence, nullptr);
vkDestroyCommandPool(device, compute.commandPool, nullptr);
uniformDataCompute.destroy();
textureLoader->destroyTexture(textureComputeTarget);
}
// Prepare a texture target that is used to store compute shader calculations
@ -134,9 +118,7 @@ public:
imageCreateInfo.tiling = VK_IMAGE_TILING_OPTIMAL;
imageCreateInfo.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
// Image will be sampled in the fragment shader and used as storage target in the compute shader
imageCreateInfo.usage =
VK_IMAGE_USAGE_SAMPLED_BIT |
VK_IMAGE_USAGE_STORAGE_BIT;
imageCreateInfo.usage = VK_IMAGE_USAGE_SAMPLED_BIT | VK_IMAGE_USAGE_STORAGE_BIT;
imageCreateInfo.flags = 0;
VkMemoryAllocateInfo memAllocInfo = vkTools::initializers::memoryAllocateInfo();
@ -166,7 +148,7 @@ public:
sampler.magFilter = VK_FILTER_LINEAR;
sampler.minFilter = VK_FILTER_LINEAR;
sampler.mipmapMode = VK_SAMPLER_MIPMAP_MODE_LINEAR;
sampler.addressModeU = VK_SAMPLER_ADDRESS_MODE_REPEAT;
sampler.addressModeU = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER;
sampler.addressModeV = sampler.addressModeU;
sampler.addressModeW = sampler.addressModeU;
sampler.mipLodBias = 0.0f;
@ -185,6 +167,11 @@ public:
view.subresourceRange = { VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1 };
view.image = tex->image;
VK_CHECK_RESULT(vkCreateImageView(device, &view, nullptr, &tex->view));
// Initialize a descriptor for later use
tex->descriptor.imageLayout = tex->imageLayout;
tex->descriptor.imageView = tex->view;
tex->descriptor.sampler = tex->sampler;
}
void buildCommandBuffers()
@ -219,22 +206,19 @@ public:
VK_CHECK_RESULT(vkBeginCommandBuffer(drawCmdBuffers[i], &cmdBufInfo));
// Image memory barrier to make sure that compute
// shader writes are finished before sampling
// from the texture
// Image memory barrier to make sure that compute shader writes are finished before sampling from the texture
VkImageMemoryBarrier imageMemoryBarrier = {};
imageMemoryBarrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER;
imageMemoryBarrier.pNext = NULL;
imageMemoryBarrier.oldLayout = VK_IMAGE_LAYOUT_GENERAL;
imageMemoryBarrier.newLayout = VK_IMAGE_LAYOUT_GENERAL;
imageMemoryBarrier.image = textureComputeTarget.image;
imageMemoryBarrier.subresourceRange = { VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1 };
imageMemoryBarrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT;
imageMemoryBarrier.dstAccessMask = VK_ACCESS_INPUT_ATTACHMENT_READ_BIT;
imageMemoryBarrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT;
vkCmdPipelineBarrier(
drawCmdBuffers[i],
VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT,
VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT,
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT,
VK_FLAGS_NONE,
0, nullptr,
0, nullptr,
@ -248,16 +232,11 @@ public:
VkRect2D scissor = vkTools::initializers::rect2D(width, height, 0, 0);
vkCmdSetScissor(drawCmdBuffers[i], 0, 1, &scissor);
VkDeviceSize offsets[1] = { 0 };
vkCmdBindVertexBuffers(drawCmdBuffers[i], VERTEX_BUFFER_BIND_ID, 1, &meshes.quad.vertices.buf, offsets);
vkCmdBindIndexBuffer(drawCmdBuffers[i], meshes.quad.indices.buf, 0, VK_INDEX_TYPE_UINT32);
// Display ray traced image generated by compute shader as a full screen quad
vkCmdBindDescriptorSets(drawCmdBuffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, pipelineLayout, 0, 1, &descriptorSetPostCompute, 0, NULL);
vkCmdBindPipeline(drawCmdBuffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, pipelines.display);
vkCmdDrawIndexed(drawCmdBuffers[i], meshes.quad.indexCount, 1, 0, 0, 0);
// Quad vertices are generated in the vertex shader
vkCmdBindDescriptorSets(drawCmdBuffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, graphics.pipelineLayout, 0, 1, &graphics.descriptorSet, 0, NULL);
vkCmdBindPipeline(drawCmdBuffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, graphics.pipeline);
vkCmdDraw(drawCmdBuffers[i], 3, 1, 0, 0);
vkCmdEndRenderPass(drawCmdBuffers[i]);
@ -270,82 +249,14 @@ public:
{
VkCommandBufferBeginInfo cmdBufInfo = vkTools::initializers::commandBufferBeginInfo();
VK_CHECK_RESULT(vkBeginCommandBuffer(computeCmdBuffer, &cmdBufInfo));
VK_CHECK_RESULT(vkBeginCommandBuffer(compute.commandBuffer, &cmdBufInfo));
vkCmdBindPipeline(computeCmdBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, pipelines.compute);
vkCmdBindDescriptorSets(computeCmdBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, computePipelineLayout, 0, 1, &computeDescriptorSet, 0, 0);
vkCmdBindPipeline(compute.commandBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, compute.pipeline);
vkCmdBindDescriptorSets(compute.commandBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, compute.pipelineLayout, 0, 1, &compute.descriptorSet, 0, 0);
vkCmdDispatch(computeCmdBuffer, textureComputeTarget.width / 16, textureComputeTarget.height / 16, 1);
vkCmdDispatch(compute.commandBuffer, textureComputeTarget.width / 16, textureComputeTarget.height / 16, 1);
vkEndCommandBuffer(computeCmdBuffer);
}
// Setup vertices for a single uv-mapped quad
void generateQuad()
{
#define dim 1.0f
std::vector<Vertex> vertexBuffer =
{
{ { dim, dim, 0.0f }, { 1.0f, 1.0f } },
{ { -dim, dim, 0.0f }, { 0.0f, 1.0f } },
{ { -dim, -dim, 0.0f }, { 0.0f, 0.0f } },
{ { dim, -dim, 0.0f }, { 1.0f, 0.0f } }
};
#undef dim
createBuffer(
VK_BUFFER_USAGE_VERTEX_BUFFER_BIT,
vertexBuffer.size() * sizeof(Vertex),
vertexBuffer.data(),
&meshes.quad.vertices.buf,
&meshes.quad.vertices.mem);
// Setup indices
std::vector<uint32_t> indexBuffer = { 0,1,2, 2,3,0 };
meshes.quad.indexCount = indexBuffer.size();
createBuffer(
VK_BUFFER_USAGE_INDEX_BUFFER_BIT,
indexBuffer.size() * sizeof(uint32_t),
indexBuffer.data(),
&meshes.quad.indices.buf,
&meshes.quad.indices.mem);
}
void setupVertexDescriptions()
{
// Binding description
vertices.bindingDescriptions.resize(1);
vertices.bindingDescriptions[0] =
vkTools::initializers::vertexInputBindingDescription(
VERTEX_BUFFER_BIND_ID,
sizeof(Vertex),
VK_VERTEX_INPUT_RATE_VERTEX);
// Attribute descriptions
// Describes memory layout and shader positions
vertices.attributeDescriptions.resize(2);
// Location 0 : Position
vertices.attributeDescriptions[0] =
vkTools::initializers::vertexInputAttributeDescription(
VERTEX_BUFFER_BIND_ID,
0,
VK_FORMAT_R32G32B32_SFLOAT,
0);
// Location 1 : Texture coordinates
vertices.attributeDescriptions[1] =
vkTools::initializers::vertexInputAttributeDescription(
VERTEX_BUFFER_BIND_ID,
1,
VK_FORMAT_R32G32_SFLOAT,
sizeof(float) * 3);
// Assign to vertex buffer
vertices.inputState = vkTools::initializers::pipelineVertexInputStateCreateInfo();
vertices.inputState.vertexBindingDescriptionCount = vertices.bindingDescriptions.size();
vertices.inputState.pVertexBindingDescriptions = vertices.bindingDescriptions.data();
vertices.inputState.vertexAttributeDescriptionCount = vertices.attributeDescriptions.size();
vertices.inputState.pVertexAttributeDescriptions = vertices.attributeDescriptions.data();
vkEndCommandBuffer(compute.commandBuffer);
}
void setupDescriptorPool()
@ -384,14 +295,14 @@ public:
setLayoutBindings.data(),
setLayoutBindings.size());
VK_CHECK_RESULT(vkCreateDescriptorSetLayout(device, &descriptorLayout, nullptr, &descriptorSetLayout));
VK_CHECK_RESULT(vkCreateDescriptorSetLayout(device, &descriptorLayout, nullptr, &graphics.descriptorSetLayout));
VkPipelineLayoutCreateInfo pPipelineLayoutCreateInfo =
vkTools::initializers::pipelineLayoutCreateInfo(
&descriptorSetLayout,
&graphics.descriptorSetLayout,
1);
VK_CHECK_RESULT(vkCreatePipelineLayout(device, &pPipelineLayoutCreateInfo, nullptr, &pipelineLayout));
VK_CHECK_RESULT(vkCreatePipelineLayout(device, &pPipelineLayoutCreateInfo, nullptr, &graphics.pipelineLayout));
}
void setupDescriptorSet()
@ -399,43 +310,24 @@ public:
VkDescriptorSetAllocateInfo allocInfo =
vkTools::initializers::descriptorSetAllocateInfo(
descriptorPool,
&descriptorSetLayout,
&graphics.descriptorSetLayout,
1);
VK_CHECK_RESULT(vkAllocateDescriptorSets(device, &allocInfo, &descriptorSetPostCompute));
// Image descriptor for the color map texture
VkDescriptorImageInfo texDescriptor =
vkTools::initializers::descriptorImageInfo(
textureComputeTarget.sampler,
textureComputeTarget.view,
VK_IMAGE_LAYOUT_GENERAL);
VK_CHECK_RESULT(vkAllocateDescriptorSets(device, &allocInfo, &graphics.descriptorSet));
std::vector<VkWriteDescriptorSet> writeDescriptorSets =
{
// Binding 0 : Fragment shader texture sampler
vkTools::initializers::writeDescriptorSet(
descriptorSetPostCompute,
graphics.descriptorSet,
VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
0,
&texDescriptor)
&textureComputeTarget.descriptor)
};
vkUpdateDescriptorSets(device, writeDescriptorSets.size(), writeDescriptorSets.data(), 0, NULL);
}
// Create a separate command buffer for compute commands
void createComputeCommandBuffer()
{
VkCommandBufferAllocateInfo cmdBufAllocateInfo =
vkTools::initializers::commandBufferAllocateInfo(
cmdPool,
VK_COMMAND_BUFFER_LEVEL_PRIMARY,
1);
VK_CHECK_RESULT(vkAllocateCommandBuffers(device, &cmdBufAllocateInfo, &computeCmdBuffer));
}
void preparePipelines()
{
VkPipelineInputAssemblyStateCreateInfo inputAssemblyState =
@ -447,7 +339,7 @@ public:
VkPipelineRasterizationStateCreateInfo rasterizationState =
vkTools::initializers::pipelineRasterizationStateCreateInfo(
VK_POLYGON_MODE_FILL,
VK_CULL_MODE_NONE,
VK_CULL_MODE_FRONT_BIT,
VK_FRONT_FACE_COUNTER_CLOCKWISE,
0);
@ -463,8 +355,8 @@ public:
VkPipelineDepthStencilStateCreateInfo depthStencilState =
vkTools::initializers::pipelineDepthStencilStateCreateInfo(
VK_TRUE,
VK_TRUE,
VK_FALSE,
VK_FALSE,
VK_COMPARE_OP_LESS_OR_EQUAL);
VkPipelineViewportStateCreateInfo viewportState =
@ -493,11 +385,18 @@ public:
VkGraphicsPipelineCreateInfo pipelineCreateInfo =
vkTools::initializers::pipelineCreateInfo(
pipelineLayout,
graphics.pipelineLayout,
renderPass,
0);
pipelineCreateInfo.pVertexInputState = &vertices.inputState;
VkPipelineVertexInputStateCreateInfo emptyInputState{};
emptyInputState.sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO;
emptyInputState.vertexAttributeDescriptionCount = 0;
emptyInputState.pVertexAttributeDescriptions = nullptr;
emptyInputState.vertexBindingDescriptionCount = 0;
emptyInputState.pVertexBindingDescriptions = nullptr;
pipelineCreateInfo.pVertexInputState = &emptyInputState;
pipelineCreateInfo.pInputAssemblyState = &inputAssemblyState;
pipelineCreateInfo.pRasterizationState = &rasterizationState;
pipelineCreateInfo.pColorBlendState = &colorBlendState;
@ -509,12 +408,23 @@ public:
pipelineCreateInfo.pStages = shaderStages.data();
pipelineCreateInfo.renderPass = renderPass;
VK_CHECK_RESULT(vkCreateGraphicsPipelines(device, pipelineCache, 1, &pipelineCreateInfo, nullptr, &pipelines.display));
VK_CHECK_RESULT(vkCreateGraphicsPipelines(device, pipelineCache, 1, &pipelineCreateInfo, nullptr, &graphics.pipeline));
}
// Prepare the compute pipeline that generates the ray traced image
void prepareCompute()
{
// Create a compute capable device queue
// The VulkanDevice::createLogicalDevice functions finds a compute capable queue and prefers queue families that only support compute
// Depending on the implementation this may result in different queue family indices for graphics and computes,
// requiring proper synchronization (see the memory barriers in buildComputeCommandBuffer)
VkDeviceQueueCreateInfo queueCreateInfo = {};
queueCreateInfo.sType = VK_STRUCTURE_TYPE_DEVICE_QUEUE_CREATE_INFO;
queueCreateInfo.pNext = NULL;
queueCreateInfo.queueFamilyIndex = vulkanDevice->queueFamilyIndices.compute;
queueCreateInfo.queueCount = 1;
vkGetDeviceQueue(device, vulkanDevice->queueFamilyIndices.compute, 0, &compute.queue);
std::vector<VkDescriptorSetLayoutBinding> setLayoutBindings = {
// Binding 0 : Sampled image (write)
vkTools::initializers::descriptorSetLayoutBinding(
@ -533,42 +443,34 @@ public:
setLayoutBindings.data(),
setLayoutBindings.size());
VK_CHECK_RESULT(vkCreateDescriptorSetLayout(device, &descriptorLayout, nullptr, &computeDescriptorSetLayout));
VK_CHECK_RESULT(vkCreateDescriptorSetLayout(device, &descriptorLayout, nullptr, &compute.descriptorSetLayout));
VkPipelineLayoutCreateInfo pPipelineLayoutCreateInfo =
vkTools::initializers::pipelineLayoutCreateInfo(
&computeDescriptorSetLayout,
&compute.descriptorSetLayout,
1);
VK_CHECK_RESULT(vkCreatePipelineLayout(device, &pPipelineLayoutCreateInfo, nullptr, &computePipelineLayout));
VK_CHECK_RESULT(vkCreatePipelineLayout(device, &pPipelineLayoutCreateInfo, nullptr, &compute.pipelineLayout));
VkDescriptorSetAllocateInfo allocInfo =
vkTools::initializers::descriptorSetAllocateInfo(
descriptorPool,
&computeDescriptorSetLayout,
&compute.descriptorSetLayout,
1);
VK_CHECK_RESULT(vkAllocateDescriptorSets(device, &allocInfo, &computeDescriptorSet));
std::vector<VkDescriptorImageInfo> computeTexDescriptors =
{
vkTools::initializers::descriptorImageInfo(
VK_NULL_HANDLE,
textureComputeTarget.view,
VK_IMAGE_LAYOUT_GENERAL)
};
VK_CHECK_RESULT(vkAllocateDescriptorSets(device, &allocInfo, &compute.descriptorSet));
std::vector<VkWriteDescriptorSet> computeWriteDescriptorSets =
{
// Binding 0 : Output storage image
vkTools::initializers::writeDescriptorSet(
computeDescriptorSet,
compute.descriptorSet,
VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
0,
&computeTexDescriptors[0]),
&textureComputeTarget.descriptor),
// Binding 1 : Uniform buffer block
vkTools::initializers::writeDescriptorSet(
computeDescriptorSet,
compute.descriptorSet,
VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER,
1,
&uniformDataCompute.descriptor)
@ -576,28 +478,48 @@ public:
vkUpdateDescriptorSets(device, computeWriteDescriptorSets.size(), computeWriteDescriptorSets.data(), 0, NULL);
// Create compute shader pipelines
VkComputePipelineCreateInfo computePipelineCreateInfo =
vkTools::initializers::computePipelineCreateInfo(
computePipelineLayout,
compute.pipelineLayout,
0);
computePipelineCreateInfo.stage = loadShader(getAssetPath() + "shaders/raytracing/raytracing.comp.spv", VK_SHADER_STAGE_COMPUTE_BIT);
VK_CHECK_RESULT(vkCreateComputePipelines(device, pipelineCache, 1, &computePipelineCreateInfo, nullptr, &pipelines.compute));
VK_CHECK_RESULT(vkCreateComputePipelines(device, pipelineCache, 1, &computePipelineCreateInfo, nullptr, &compute.pipeline));
// Separate command pool as queue family for compute may be different than graphics
VkCommandPoolCreateInfo cmdPoolInfo = {};
cmdPoolInfo.sType = VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO;
cmdPoolInfo.queueFamilyIndex = vulkanDevice->queueFamilyIndices.compute;
cmdPoolInfo.flags = VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT;
VK_CHECK_RESULT(vkCreateCommandPool(device, &cmdPoolInfo, nullptr, &compute.commandPool));
// Create a command buffer for compute operations
VkCommandBufferAllocateInfo cmdBufAllocateInfo =
vkTools::initializers::commandBufferAllocateInfo(
compute.commandPool,
VK_COMMAND_BUFFER_LEVEL_PRIMARY,
1);
VK_CHECK_RESULT(vkAllocateCommandBuffers(device, &cmdBufAllocateInfo, &compute.commandBuffer));
// Fence for compute CB sync
VkFenceCreateInfo fenceCreateInfo = vkTools::initializers::fenceCreateInfo(VK_FENCE_CREATE_SIGNALED_BIT);
VK_CHECK_RESULT(vkCreateFence(device, &fenceCreateInfo, nullptr, &compute.fence));
// Build a single command buffer containing the compute dispatch commands
buildComputeCommandBuffer();
}
// Prepare and initialize uniform buffer containing shader uniforms
void prepareUniformBuffers()
{
// Vertex shader uniform buffer block
createBuffer(
// Compute shader parameter uniform buffer block
vulkanDevice->createBuffer(
VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT,
sizeof(uboCompute),
&uboCompute,
&uniformDataCompute.buffer,
&uniformDataCompute.memory,
&uniformDataCompute.descriptor);
VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
&uniformDataCompute,
sizeof(uboCompute));
updateUniformBuffers();
}
@ -608,35 +530,9 @@ public:
uboCompute.lightPos.y = 5.0f;
uboCompute.lightPos.z = 1.0f;
uboCompute.lightPos.z = 0.0f + cos(glm::radians(timer * 360.0f)) * 2.0f;
uint8_t *pData;
VK_CHECK_RESULT(vkMapMemory(device, uniformDataCompute.memory, 0, sizeof(uboCompute), 0, (void **)&pData));
memcpy(pData, &uboCompute, sizeof(uboCompute));
vkUnmapMemory(device, uniformDataCompute.memory);
}
// Find and create a compute capable device queue
void getComputeQueue()
{
uint32_t queueIndex = 0;
uint32_t queueCount;
vkGetPhysicalDeviceQueueFamilyProperties(physicalDevice, &queueCount, NULL);
assert(queueCount >= 1);
std::vector<VkQueueFamilyProperties> queueProps;
queueProps.resize(queueCount);
vkGetPhysicalDeviceQueueFamilyProperties(physicalDevice, &queueCount, queueProps.data());
for (queueIndex = 0; queueIndex < queueCount; queueIndex++)
{
if (queueProps[queueIndex].queueFlags & VK_QUEUE_COMPUTE_BIT)
break;
}
assert(queueIndex < queueCount);
VkDeviceQueueCreateInfo queueCreateInfo = {};
queueCreateInfo.queueFamilyIndex = queueIndex;
queueCreateInfo.queueCount = 1;
vkGetDeviceQueue(device, queueIndex, 0, &computeQueue);
VK_CHECK_RESULT(uniformDataCompute.map());
memcpy(uniformDataCompute.mapped, &uboCompute, sizeof(uboCompute));
uniformDataCompute.unmap();
}
void draw()
@ -646,28 +542,25 @@ public:
// Command buffer to be sumitted to the queue
submitInfo.commandBufferCount = 1;
submitInfo.pCommandBuffers = &drawCmdBuffers[currentBuffer];
// Submit to queue
VK_CHECK_RESULT(vkQueueSubmit(queue, 1, &submitInfo, VK_NULL_HANDLE));
VulkanExampleBase::submitFrame();
// Compute
// Submit compute commands
// Use a fence to ensure that compute command buffer has finished executin before using it again
vkWaitForFences(device, 1, &compute.fence, VK_TRUE, UINT64_MAX);
vkResetFences(device, 1, &compute.fence);
VkSubmitInfo computeSubmitInfo = vkTools::initializers::submitInfo();
computeSubmitInfo.commandBufferCount = 1;
computeSubmitInfo.pCommandBuffers = &computeCmdBuffer;
computeSubmitInfo.pCommandBuffers = &compute.commandBuffer;
VK_CHECK_RESULT(vkQueueSubmit(computeQueue, 1, &computeSubmitInfo, VK_NULL_HANDLE));
VK_CHECK_RESULT(vkQueueWaitIdle(computeQueue));
VK_CHECK_RESULT(vkQueueSubmit(compute.queue, 1, &computeSubmitInfo, compute.fence));
}
void prepare()
{
VulkanExampleBase::prepare();
generateQuad();
getComputeQueue();
createComputeCommandBuffer();
setupVertexDescriptions();
prepareUniformBuffers();
prepareTextureTarget(&textureComputeTarget, TEX_DIM, TEX_DIM, VK_FORMAT_R8G8B8A8_UNORM);
setupDescriptorSetLayout();
@ -676,7 +569,6 @@ public:
setupDescriptorSet();
prepareCompute();
buildCommandBuffers();
buildComputeCommandBuffer();
prepared = true;
}
@ -693,6 +585,7 @@ public:
virtual void viewChanged()
{
uboCompute.aspectRatio = (float)width / (float)height;
updateUniformBuffers();
}
};

View file

@ -90,6 +90,11 @@
<ClInclude Include="..\base\vulkanexamplebase.h" />
<ClInclude Include="..\base\vulkantools.h" />
</ItemGroup>
<ItemGroup>
<None Include="..\data\shaders\raytracing\raytracing.comp" />
<None Include="..\data\shaders\raytracing\texture.frag" />
<None Include="..\data\shaders\raytracing\texture.vert" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
</ImportGroup>

View file

@ -13,6 +13,9 @@
<UniqueIdentifier>{67DA6AB6-F800-4c08-8B7A-83BB121AAD01}</UniqueIdentifier>
<Extensions>rc;ico;cur;bmp;dlg;rc2;rct;bin;rgs;gif;jpg;jpeg;jpe;resx;tiff;tif;png;wav</Extensions>
</Filter>
<Filter Include="Shaders">
<UniqueIdentifier>{a50f0cd4-8bf4-41c9-adaa-ac4dc27aaea3}</UniqueIdentifier>
</Filter>
</ItemGroup>
<ItemGroup>
<ClCompile Include="..\base\vulkandebug.cpp">
@ -39,4 +42,15 @@
<Filter>Header Files</Filter>
</ClInclude>
</ItemGroup>
<ItemGroup>
<None Include="..\data\shaders\raytracing\raytracing.comp">
<Filter>Shaders</Filter>
</None>
<None Include="..\data\shaders\raytracing\texture.frag">
<Filter>Shaders</Filter>
</None>
<None Include="..\data\shaders\raytracing\texture.vert">
<Filter>Shaders</Filter>
</None>
</ItemGroup>
</Project>