Files
WiVRn/client/scenes/stream_defoveator.cpp

465 lines
15 KiB
C++

/*
* WiVRn VR streaming
* Copyright (C) 2022 Guillaume Meunier <[email protected]>
* Copyright (C) 2022 Patrick Nicolas <[email protected]>
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program. If not, see <https://www.gnu.org/licenses/>.
*/
#include "stream_defoveator.h"
#include "application.h"
#include "utils/ranges.h"
#include "vk/allocation.h"
#include "vk/pipeline.h"
#include "vk/shader.h"
#include "vk/specialization_constants.h"
#include <array>
#include <glm/glm.hpp>
#include <glm/gtx/quaternion.hpp>
#include <vk_mem_alloc.h>
#include <vulkan/vulkan_core.h>
#include <vulkan/vulkan_raii.hpp>
struct stream_defoveator::vertex
{
// output image position
alignas(8) glm::vec2 position;
// input texture coordinates
alignas(8) glm::uvec2 uv;
};
struct vert_pc
{
glm::ivec4 rgb_rect;
glm::ivec4 a_rect;
std::array<float, 4> scale;
std::array<float, 4> bias;
};
void stream_defoveator::ensure_vertices(size_t num_vertices)
{
vk::BufferCreateInfo create_info{
.size = num_vertices * sizeof(vertex) * view_count,
.usage = vk::BufferUsageFlagBits::eVertexBuffer,
.sharingMode = vk::SharingMode::eExclusive,
};
if (create_info.size <= buffer.info().size)
return;
VmaAllocationCreateInfo alloc_info{
.requiredFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT | VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
};
buffer = buffer_allocation(device, create_info, alloc_info);
vertices_size = num_vertices * sizeof(vertex);
}
stream_defoveator::vertex * stream_defoveator::get_vertices(size_t view)
{
assert(buffer);
return reinterpret_cast<vertex *>(reinterpret_cast<uintptr_t>(buffer.map()) + view * vertices_size);
}
stream_defoveator::pipeline_t & stream_defoveator::ensure_pipeline(size_t view, vk::Sampler rgb, vk::Sampler a)
{
auto & target = a ? pipeline_a[view] : pipeline_rgb[view];
if (*target.pipeline)
return target;
std::array samplers{rgb, a};
int32_t alpha = a ? 1 : 0;
// Create VkDescriptorSetLayout
std::array layout_binding{
vk::DescriptorSetLayoutBinding{
.binding = 0,
.descriptorType = vk::DescriptorType::eCombinedImageSampler,
.descriptorCount = uint32_t(alpha + 1),
.stageFlags = vk::ShaderStageFlagBits::eFragment,
.pImmutableSamplers = samplers.data(),
},
};
target.descriptor_set_layout = device.createDescriptorSetLayout(vk::DescriptorSetLayoutCreateInfo{
.bindingCount = layout_binding.size(),
.pBindings = layout_binding.data(),
});
target.ds = device.allocateDescriptorSets(vk::DescriptorSetAllocateInfo{
.descriptorPool = *ds_pool,
.descriptorSetCount = 1,
.pSetLayouts = &*target.descriptor_set_layout,
})[0]
.release();
// pipeline layout
vk::PushConstantRange pc_range{
.stageFlags = vk::ShaderStageFlagBits::eVertex | vk::ShaderStageFlagBits::eFragment,
.size = sizeof(vert_pc),
};
vk::PipelineLayoutCreateInfo pipeline_layout_info{
.setLayoutCount = 1,
.pSetLayouts = &*target.descriptor_set_layout,
.pushConstantRangeCount = 1,
.pPushConstantRanges = &pc_range,
};
target.layout = vk::raii::PipelineLayout(device, pipeline_layout_info);
// Vertex shader
auto vertex_shader = load_shader(device, "reprojection.vert");
// Fragment shader
auto specialization = make_specialization_constants(
int32_t(alpha),
VkBool32(application::get_hmd_traits().needs_srgb_conversion));
auto fragment_shader = load_shader(device, "reprojection.frag");
vk::pipeline_builder pipeline_info{
.flags = {},
.Stages = {
{
.stage = vk::ShaderStageFlagBits::eVertex,
.module = *vertex_shader,
.pName = "main",
},
{
.stage = vk::ShaderStageFlagBits::eFragment,
.module = *fragment_shader,
.pName = "main",
.pSpecializationInfo = specialization,
},
},
.VertexBindingDescriptions = {
{
.binding = 0,
.stride = sizeof(vertex),
.inputRate = vk::VertexInputRate::eVertex,
},
},
.VertexAttributeDescriptions = {
{
.location = 0,
.binding = 0,
.format = vk::Format::eR32G32Sfloat,
.offset = offsetof(vertex, position),
},
{
.location = 1,
.binding = 0,
.format = vk::Format::eR32G32Uint,
.offset = offsetof(vertex, uv),
},
},
.InputAssemblyState = {{
.topology = vk::PrimitiveTopology::eTriangleStrip,
}},
.Viewports = {{}},
.Scissors = {{}},
.RasterizationState = {{
.polygonMode = vk::PolygonMode::eFill,
.lineWidth = 1,
}},
.MultisampleState = {{
.rasterizationSamples = vk::SampleCountFlagBits::e1,
}},
.ColorBlendState = {.flags = {}},
.ColorBlendAttachments = {{
.colorWriteMask = vk::ColorComponentFlagBits::eR | vk::ColorComponentFlagBits::eG | vk::ColorComponentFlagBits::eB | vk::ColorComponentFlagBits::eA,
}},
.DynamicStates = {vk::DynamicState::eViewport, vk::DynamicState::eScissor},
.layout = *target.layout,
.renderPass = *renderpass,
.subpass = 0,
};
target.pipeline = device.createGraphicsPipeline(application::get_pipeline_cache(), pipeline_info);
return target;
}
stream_defoveator::stream_defoveator(
vk::raii::Device & device,
vk::raii::PhysicalDevice & physical_device,
std::vector<vk::Image> output_images_,
vk::Extent2D output_extent,
vk::Format format) :
device(device),
physical_device(physical_device),
output_images(std::move(output_images_)),
output_extent(output_extent)
{
// Create renderpass
vk::AttachmentDescription attachment{
.format = format,
.samples = vk::SampleCountFlagBits::e1,
.loadOp = vk::AttachmentLoadOp::eDontCare,
.storeOp = vk::AttachmentStoreOp::eStore,
.finalLayout = vk::ImageLayout::eColorAttachmentOptimal,
};
vk::AttachmentReference color_ref{
.attachment = 0,
.layout = vk::ImageLayout::eColorAttachmentOptimal,
};
vk::SubpassDescription subpass{
.pipelineBindPoint = vk::PipelineBindPoint::eGraphics,
.colorAttachmentCount = 1,
.pColorAttachments = &color_ref,
};
vk::StructureChain renderpass_info{
vk::RenderPassCreateInfo{
.attachmentCount = 1,
.pAttachments = &attachment,
.subpassCount = 1,
.pSubpasses = &subpass,
},
};
renderpass = vk::raii::RenderPass(device, renderpass_info.get());
vk::DescriptorPoolSize pool_size{
.type = vk::DescriptorType::eCombinedImageSampler,
.descriptorCount = view_count * 4,
};
ds_pool = device.createDescriptorPool(vk::DescriptorPoolCreateInfo{
.maxSets = view_count * 2,
.poolSizeCount = 1,
.pPoolSizes = &pool_size,
});
// Create image views and framebuffers
output_image_views.reserve(output_images.size() * view_count);
framebuffers.reserve(output_images.size() * view_count);
for (vk::Image image: output_images)
{
for (uint32_t view = 0; view < view_count; ++view)
{
vk::ImageViewCreateInfo iv_info{
.image = image,
.viewType = vk::ImageViewType::e2DArray,
.format = format,
.components = {
.r = vk::ComponentSwizzle::eIdentity,
.g = vk::ComponentSwizzle::eIdentity,
.b = vk::ComponentSwizzle::eIdentity,
.a = vk::ComponentSwizzle::eIdentity,
},
.subresourceRange = {
.aspectMask = vk::ImageAspectFlagBits::eColor,
.baseMipLevel = 0,
.levelCount = 1,
.baseArrayLayer = view,
.layerCount = 1,
},
};
output_image_views.emplace_back(device, iv_info);
vk::FramebufferCreateInfo fb_create_info{
.renderPass = *renderpass,
.width = output_extent.width,
.height = output_extent.height,
.layers = 1,
};
fb_create_info.setAttachments(*output_image_views.back());
framebuffers.emplace_back(device, fb_create_info);
}
}
}
void stream_defoveator::reset_pipelines()
{
for (auto & p: pipeline_rgb)
p = {};
for (auto & p: pipeline_a)
p = {};
}
static size_t required_vertices(const wivrn::to_headset::foveation_parameter & p)
{
// strips are constructed like this:
// 0 2 4
// 1 3 5 5*
// there is one such line per value in y
// the last element is repeated to break the line
return (2 * (p.x.size() + 1) + 1) * p.y.size();
}
void stream_defoveator::defoveate(vk::raii::CommandBuffer & command_buffer,
const std::array<wivrn::to_headset::foveation_parameter, 2> & foveation,
const std::array<input, 2> & inputs,
std::array<float, 4> scale,
std::array<float, 4> bias,
int destination)
{
if (destination < 0 || destination >= (int)output_images.size())
throw std::runtime_error("Invalid destination image index");
ensure_vertices(std::max(required_vertices(foveation[0]), required_vertices(foveation[1])));
for (size_t view = 0; view < view_count; ++view)
{
const auto out_size = defoveated_size(foveation[view]);
auto vertices = get_vertices(view);
const auto & [px, py] = foveation[view];
assert(px.size() % 2 == 1);
assert(py.size() % 2 == 1);
const int n_ratio_y = (py.size() - 1) / 2;
const int n_ratio_x = (px.size() - 1) / 2;
command_buffer.setScissor(
0,
vk::Rect2D{
.extent = {.width = uint32_t(out_size.width), .height = uint32_t(out_size.height)},
});
command_buffer.setViewport(
0,
vk::Viewport{
.x = 0,
.y = 0,
.width = float(out_size.width),
.height = float(out_size.height),
.minDepth = 0,
.maxDepth = 1,
});
glm::uvec2 in(0);
glm::vec2 out(-0.5 * out_size.width, -0.5 * out_size.height); // pixel coordinates
glm::vec2 out_pixel_size(2. / out_size.width,
2. / out_size.height);
for (auto [iy, n_out_y]: utils::enumerate_range(py))
{
// number of output pixels per source pixels
const int ratio_y = std::abs(n_ratio_y - int(iy)) + 1;
in.x = 0;
out.x = -0.5 * out_size.width;
for (auto [ix, n_out_x]: utils::enumerate_range(px))
{
const int ratio_x = std::abs(n_ratio_x - int(ix)) + 1;
*vertices++ = {
.position = out * out_pixel_size,
.uv = in,
};
*vertices++ = {
.position = (out + glm::vec2(0, n_out_y * ratio_y)) * out_pixel_size,
.uv = (in + glm::uvec2(0, n_out_y)),
};
in.x += n_out_x;
out.x += n_out_x * ratio_x;
}
*vertices++ = {
.position = out * out_pixel_size,
.uv = in,
};
in.y += n_out_y;
out.y += n_out_y * ratio_y;
*vertices++ = {
.position = out * out_pixel_size,
.uv = in,
};
*vertices++ = {
.position = out * out_pixel_size,
.uv = in,
};
}
}
for (size_t view = 0; view < view_count; ++view)
{
vk::RenderPassBeginInfo begin_info{
.renderPass = *renderpass,
.framebuffer = *framebuffers[destination * view_count + view],
.renderArea = {
.offset = {0, 0},
.extent = output_extent,
},
};
const auto & input = inputs[view];
auto & pipeline = ensure_pipeline(view, input.sampler_rgb, input.sampler_a);
std::array image_info{
vk::DescriptorImageInfo{
.sampler = input.sampler_rgb,
.imageView = input.rgb,
.imageLayout = input.layout_rgb,
},
vk::DescriptorImageInfo{
.sampler = input.sampler_a,
.imageView = input.a,
.imageLayout = input.layout_a,
},
};
std::array descriptor_writes{
vk::WriteDescriptorSet{
.dstSet = pipeline.ds,
.dstBinding = 0,
.descriptorCount = input.sampler_a ? 2u : 1u,
.descriptorType = vk::DescriptorType::eCombinedImageSampler,
.pImageInfo = image_info.data(),
},
};
vert_pc pc{
.rgb_rect = glm::ivec4(input.rect_rgb.offset.x,
input.rect_rgb.offset.y,
input.rect_rgb.extent.width,
input.rect_rgb.extent.height),
.a_rect = glm::ivec4(input.rect_a.offset.x,
input.rect_a.offset.y,
input.rect_a.extent.width,
input.rect_a.extent.height),
.scale = scale,
.bias = bias,
};
device.updateDescriptorSets(descriptor_writes, {});
command_buffer.beginRenderPass(begin_info, vk::SubpassContents::eInline);
command_buffer.bindPipeline(vk::PipelineBindPoint::eGraphics, *pipeline.pipeline);
command_buffer.bindDescriptorSets(vk::PipelineBindPoint::eGraphics, *pipeline.layout, 0, pipeline.ds, {});
command_buffer.pushConstants<vert_pc>(*pipeline.layout, vk::ShaderStageFlagBits::eVertex | vk::ShaderStageFlagBits::eFragment, 0, pc);
command_buffer.bindVertexBuffers(0, vk::Buffer(buffer), vertices_size * view);
command_buffer.draw(required_vertices(foveation[view]), 1, 0, 0);
command_buffer.endRenderPass();
}
}
static uint16_t count_pixels(const std::vector<uint16_t> & param)
{
uint16_t res = 0;
const int n_ratio = (param.size() - 1) / 2;
for (auto [i, n_out]: utils::enumerate_range(param))
{
// number of output pixels per source pixels
const int ratio = std::abs(n_ratio - int(i)) + 1;
res += ratio * n_out;
}
return res;
}
XrExtent2Di stream_defoveator::defoveated_size(const wivrn::to_headset::foveation_parameter & view)
{
return {
count_pixels(view.x),
count_pixels(view.y),
};
}