mirror of
https://github.com/libretro/parallel-dc.git
synced 2026-11-10 17:26:21 -10:00
libretroadmin
fab5ee678c
The per-pixel (OIT) renderer renders at the size the frontend's temporal upscaler asks for, under its jitter, keeps each frame's depth and draws its motion vectors over it, as the per-triangle renderer does. Its depth is one image a frame slot while the upscaler has the frames, stored by the last render pass instead of dropped, with the render passes' framebuffers made for it; SSAA is off then, as the upscaler is the anti-aliasing and reads a single-sample depth. The motion vector pass is one piece of code for both renderers (rend/vulkan/motion_pass.cpp): the pass, its pipelines by cull mode, the vectors' images and the strips' previous positions. While the upscaler is active the Internal Resolution and SSAA core options have no effect, and are hidden. tools/vulkan/vk_hdr.sh and vk_sanitize.sh run the per-pixel renderer with the upscaler, the inputs read back and checked under the validation layer and the sanitizers.
231 lines
15 KiB
C++
231 lines
15 KiB
C++
#define PVR_REGS_FOR_RENDERER // see hw/pvr/pvr_regs.h
|
|
/*
|
|
Created on: Nov 10, 2019
|
|
|
|
Copyright 2019 flyinghead
|
|
|
|
This file is part of Flycast.
|
|
|
|
Flycast 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 2 of the License, or
|
|
(at your option) any later version.
|
|
|
|
Flycast 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 Flycast. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
#include "oit_renderpass.h"
|
|
|
|
vk::UniqueRenderPass RenderPasses::MakeRenderPass(bool initial, bool last, bool depthWrites, bool resumed)
|
|
{
|
|
if (samples > 1)
|
|
return MakeSsaaRenderPass(initial, last, depthWrites, resumed);
|
|
vk::AttachmentDescription attachmentDescriptions[] = {
|
|
// Swap chain image
|
|
GetAttachment0Description(initial, last),
|
|
// OP+PT color attachment
|
|
vk::AttachmentDescription(vk::AttachmentDescriptionFlags(), GetColorFormat(), vk::SampleCountFlagBits::e1,
|
|
initial ? vk::AttachmentLoadOp::eClear : vk::AttachmentLoadOp::eLoad,
|
|
last ? vk::AttachmentStoreOp::eDontCare : vk::AttachmentStoreOp::eStore,
|
|
vk::AttachmentLoadOp::eDontCare, vk::AttachmentStoreOp::eDontCare,
|
|
initial ? vk::ImageLayout::eUndefined : vk::ImageLayout::eShaderReadOnlyOptimal, vk::ImageLayout::eShaderReadOnlyOptimal),
|
|
// OP+PT depth attachment (kept after the last render pass for the upscaler, when there is one)
|
|
vk::AttachmentDescription(vk::AttachmentDescriptionFlags(), GetContext()->GetDepthFormat(), vk::SampleCountFlagBits::e1,
|
|
initial ? vk::AttachmentLoadOp::eClear : vk::AttachmentLoadOp::eLoad,
|
|
last && !keepDepth ? vk::AttachmentStoreOp::eDontCare : vk::AttachmentStoreOp::eStore,
|
|
vk::AttachmentLoadOp::eClear, vk::AttachmentStoreOp::eDontCare,
|
|
initial ? vk::ImageLayout::eUndefined : vk::ImageLayout::eDepthStencilReadOnlyOptimal, vk::ImageLayout::eDepthStencilReadOnlyOptimal),
|
|
};
|
|
if (resumed)
|
|
{
|
|
attachmentDescriptions[0].loadOp = vk::AttachmentLoadOp::eLoad;
|
|
attachmentDescriptions[0].initialLayout = attachmentDescriptions[0].finalLayout;
|
|
}
|
|
vk::AttachmentReference swapChainReference(0, vk::ImageLayout::eColorAttachmentOptimal);
|
|
vk::AttachmentReference colorReference(1, vk::ImageLayout::eColorAttachmentOptimal);
|
|
vk::AttachmentReference depthReference(2, vk::ImageLayout::eDepthStencilAttachmentOptimal);
|
|
|
|
vk::AttachmentReference depthReadOnlyRef(2, DepthInputLayout(depthWrites));
|
|
vk::AttachmentReference colorInput(1, vk::ImageLayout::eShaderReadOnlyOptimal);
|
|
|
|
vk::SubpassDescription subpasses[] = {
|
|
// Depth and modvol pass FIXME subpass 0 shouldn't reference the color attachment
|
|
vk::SubpassDescription(vk::SubpassDescriptionFlags(), vk::PipelineBindPoint::eGraphics,
|
|
0, nullptr,
|
|
1, &colorReference,
|
|
nullptr,
|
|
&depthReference),
|
|
// Color pass
|
|
vk::SubpassDescription(vk::SubpassDescriptionFlags(), vk::PipelineBindPoint::eGraphics,
|
|
1, &depthReadOnlyRef,
|
|
1, &colorReference,
|
|
nullptr,
|
|
&depthReadOnlyRef),
|
|
// Final pass
|
|
vk::SubpassDescription(vk::SubpassDescriptionFlags(), vk::PipelineBindPoint::eGraphics,
|
|
1, &colorInput,
|
|
1, &swapChainReference,
|
|
nullptr,
|
|
&depthReference), // depth-only Tr pass when continuation
|
|
};
|
|
|
|
std::vector<vk::SubpassDependency> dependencies = GetSubpassDependencies();
|
|
dependencies.emplace_back(VK_SUBPASS_EXTERNAL, 0, vk::PipelineStageFlagBits::eFragmentShader,
|
|
vk::PipelineStageFlagBits::eEarlyFragmentTests | vk::PipelineStageFlagBits::eLateFragmentTests,
|
|
vk::AccessFlagBits::eInputAttachmentRead | vk::AccessFlagBits::eShaderRead,
|
|
vk::AccessFlagBits::eDepthStencilAttachmentRead | vk::AccessFlagBits::eDepthStencilAttachmentWrite,
|
|
vk::DependencyFlagBits::eByRegion);
|
|
dependencies.emplace_back(VK_SUBPASS_EXTERNAL, 1, vk::PipelineStageFlagBits::eFragmentShader, vk::PipelineStageFlagBits::eColorAttachmentOutput,
|
|
vk::AccessFlagBits::eInputAttachmentRead, vk::AccessFlagBits::eColorAttachmentWrite, vk::DependencyFlagBits::eByRegion);
|
|
/* The colour and depth attachments the polygons are drawn into, and the
|
|
* per-pixel lists, are the same ones for every render pass and every
|
|
* frame: the one before may still be reading or writing them when this
|
|
* one clears or loads them and writes the lists again. Nothing said it
|
|
* had to be done first. Most drivers work that out for themselves; the
|
|
* ones that do not show one frame's leavings in the next. (From
|
|
* upstream, which names MoltenVK on macOS 15.)
|
|
* These two attachments are first used in subpasses 0 and 1. */
|
|
const vk::PipelineStageFlags oitStages = vk::PipelineStageFlagBits::eEarlyFragmentTests | vk::PipelineStageFlagBits::eLateFragmentTests
|
|
| vk::PipelineStageFlagBits::eFragmentShader | vk::PipelineStageFlagBits::eColorAttachmentOutput;
|
|
for (u32 subpass = 0; subpass < 2; subpass++)
|
|
dependencies.emplace_back(VK_SUBPASS_EXTERNAL, subpass, oitStages, oitStages,
|
|
vk::AccessFlagBits::eColorAttachmentWrite | vk::AccessFlagBits::eDepthStencilAttachmentWrite | vk::AccessFlagBits::eShaderWrite,
|
|
vk::AccessFlagBits::eColorAttachmentRead | vk::AccessFlagBits::eColorAttachmentWrite
|
|
| vk::AccessFlagBits::eDepthStencilAttachmentRead | vk::AccessFlagBits::eDepthStencilAttachmentWrite
|
|
| vk::AccessFlagBits::eInputAttachmentRead | vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite);
|
|
/* The picture's attachment is first used in subpass 2. With several
|
|
* render passes it was the polygons' colour attachment of the one
|
|
* before, written and then read as an input. */
|
|
dependencies.emplace_back(VK_SUBPASS_EXTERNAL, 2,
|
|
vk::PipelineStageFlagBits::eFragmentShader | vk::PipelineStageFlagBits::eColorAttachmentOutput,
|
|
vk::PipelineStageFlagBits::eFragmentShader | vk::PipelineStageFlagBits::eColorAttachmentOutput,
|
|
vk::AccessFlagBits::eColorAttachmentWrite,
|
|
vk::AccessFlagBits::eColorAttachmentRead | vk::AccessFlagBits::eColorAttachmentWrite);
|
|
dependencies.emplace_back(0, 1, vk::PipelineStageFlagBits::eLateFragmentTests, vk::PipelineStageFlagBits::eFragmentShader,
|
|
vk::AccessFlagBits::eDepthStencilAttachmentRead | vk::AccessFlagBits::eDepthStencilAttachmentWrite,
|
|
vk::AccessFlagBits::eInputAttachmentRead | vk::AccessFlagBits::eShaderRead, vk::DependencyFlagBits::eByRegion);
|
|
// carries the dependency above on to where the colour attachment changes hands for subpass 2
|
|
dependencies.emplace_back(0, 1, vk::PipelineStageFlagBits::eColorAttachmentOutput, vk::PipelineStageFlagBits::eColorAttachmentOutput,
|
|
vk::AccessFlagBits::eColorAttachmentWrite, vk::AccessFlagBits::eColorAttachmentRead | vk::AccessFlagBits::eColorAttachmentWrite,
|
|
vk::DependencyFlagBits::eByRegion);
|
|
// the depth attachment is stored at the end of subpass 2, and changes layout after that - not before
|
|
dependencies.emplace_back(2, VK_SUBPASS_EXTERNAL,
|
|
vk::PipelineStageFlagBits::eEarlyFragmentTests | vk::PipelineStageFlagBits::eLateFragmentTests, oitStages,
|
|
vk::AccessFlagBits::eDepthStencilAttachmentWrite,
|
|
vk::AccessFlagBits::eDepthStencilAttachmentRead | vk::AccessFlagBits::eDepthStencilAttachmentWrite
|
|
| vk::AccessFlagBits::eInputAttachmentRead | vk::AccessFlagBits::eShaderRead);
|
|
// (the colour attachment is stored at the end of subpass 2)
|
|
dependencies.emplace_back(1, 2, vk::PipelineStageFlagBits::eColorAttachmentOutput,
|
|
vk::PipelineStageFlagBits::eFragmentShader | vk::PipelineStageFlagBits::eColorAttachmentOutput,
|
|
vk::AccessFlagBits::eColorAttachmentWrite, vk::AccessFlagBits::eInputAttachmentRead | vk::AccessFlagBits::eColorAttachmentWrite,
|
|
vk::DependencyFlagBits::eByRegion);
|
|
// This dependency is only needed if the render pass isn't the last: it's needed for the depth-only Tr pass
|
|
// Unfortunately we want all render passes to be compatible, and that means all attachments must be identical
|
|
dependencies.emplace_back(1, 2, vk::PipelineStageFlagBits::eFragmentShader,
|
|
vk::PipelineStageFlagBits::eEarlyFragmentTests | vk::PipelineStageFlagBits::eLateFragmentTests,
|
|
vk::AccessFlagBits::eInputAttachmentRead | vk::AccessFlagBits::eShaderRead,
|
|
vk::AccessFlagBits::eDepthStencilAttachmentRead | vk::AccessFlagBits::eDepthStencilAttachmentWrite,
|
|
vk::DependencyFlagBits::eByRegion);
|
|
dependencies.emplace_back(2, 2, vk::PipelineStageFlagBits::eFragmentShader, vk::PipelineStageFlagBits::eFragmentShader,
|
|
vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite,
|
|
vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite,
|
|
vk::DependencyFlagBits::eByRegion);
|
|
|
|
return GetContext()->GetDevice().createRenderPassUnique(vk::RenderPassCreateInfo(vk::RenderPassCreateFlags(),
|
|
ARRAY_SIZE(attachmentDescriptions), attachmentDescriptions,
|
|
ARRAY_SIZE(subpasses), subpasses,
|
|
dependencies.size(), dependencies.data()));
|
|
}
|
|
|
|
|
|
/* SSAA: attachment 0 is the picture, of one sample; 1 and 2 the colour
|
|
* and depth of the samples, one image each for all the render passes of a
|
|
* frame. The first render pass clears them; one that is not the last
|
|
* stores them for the next, the last leaves them: with one render pass
|
|
* they are never stored, and a tiler never backs them. The picture is all
|
|
* written by the last subpass of the last render pass, so it is never
|
|
* loaded, and stored by the last only. The colour samples are read and
|
|
* written in subpass 2 (each put together where it is) and read in 3:
|
|
* they are in the general layout there. */
|
|
vk::UniqueRenderPass RenderPasses::MakeSsaaRenderPass(bool initial, bool last, bool depthWrites, bool resumed)
|
|
{
|
|
const vk::SampleCountFlagBits n = (vk::SampleCountFlagBits)samples;
|
|
const vk::ImageLayout depthLayout = vk::ImageLayout::eDepthStencilAttachmentOptimal;
|
|
vk::AttachmentDescription attachmentDescriptions[] = {
|
|
vk::AttachmentDescription(vk::AttachmentDescriptionFlags(), GetColorFormat(), vk::SampleCountFlagBits::e1,
|
|
vk::AttachmentLoadOp::eDontCare, last ? vk::AttachmentStoreOp::eStore : vk::AttachmentStoreOp::eDontCare,
|
|
vk::AttachmentLoadOp::eDontCare, vk::AttachmentStoreOp::eDontCare,
|
|
vk::ImageLayout::eUndefined, vk::ImageLayout::eShaderReadOnlyOptimal),
|
|
vk::AttachmentDescription(vk::AttachmentDescriptionFlags(), GetColorFormat(), n,
|
|
initial ? vk::AttachmentLoadOp::eClear : vk::AttachmentLoadOp::eLoad,
|
|
last ? vk::AttachmentStoreOp::eDontCare : vk::AttachmentStoreOp::eStore,
|
|
vk::AttachmentLoadOp::eDontCare, vk::AttachmentStoreOp::eDontCare,
|
|
initial ? vk::ImageLayout::eUndefined : vk::ImageLayout::eGeneral, vk::ImageLayout::eGeneral),
|
|
vk::AttachmentDescription(vk::AttachmentDescriptionFlags(), GetContext()->GetDepthFormat(), n,
|
|
initial ? vk::AttachmentLoadOp::eClear : vk::AttachmentLoadOp::eLoad,
|
|
last ? vk::AttachmentStoreOp::eDontCare : vk::AttachmentStoreOp::eStore,
|
|
vk::AttachmentLoadOp::eClear, vk::AttachmentStoreOp::eDontCare,
|
|
initial ? vk::ImageLayout::eUndefined : depthLayout, depthLayout),
|
|
};
|
|
if (resumed)
|
|
{
|
|
attachmentDescriptions[0].loadOp = vk::AttachmentLoadOp::eLoad;
|
|
attachmentDescriptions[0].initialLayout = attachmentDescriptions[0].finalLayout;
|
|
}
|
|
vk::AttachmentReference pictureReference(0, vk::ImageLayout::eColorAttachmentOptimal);
|
|
vk::AttachmentReference colorReference(1, vk::ImageLayout::eColorAttachmentOptimal);
|
|
vk::AttachmentReference depthReference(2, depthLayout);
|
|
vk::AttachmentReference depthReadOnlyRef(2, DepthInputLayout(depthWrites));
|
|
vk::AttachmentReference colorGeneral(1, vk::ImageLayout::eGeneral);
|
|
|
|
vk::SubpassDescription subpasses[] = {
|
|
// depth and stencil, modifier volumes
|
|
vk::SubpassDescription(vk::SubpassDescriptionFlags(), vk::PipelineBindPoint::eGraphics,
|
|
0, nullptr, 1, &colorReference, nullptr, &depthReference),
|
|
// colour, and the translucent lists
|
|
vk::SubpassDescription(vk::SubpassDescriptionFlags(), vk::PipelineBindPoint::eGraphics,
|
|
1, &depthReadOnlyRef, 1, &colorReference, nullptr, &depthReadOnlyRef),
|
|
// translucent modifier volumes; each sample put together and the translucent depth, before another render pass
|
|
vk::SubpassDescription(vk::SubpassDescriptionFlags(), vk::PipelineBindPoint::eGraphics,
|
|
1, &colorGeneral, 1, &colorGeneral, nullptr, &depthReference),
|
|
// the picture put together and resolved
|
|
vk::SubpassDescription(vk::SubpassDescriptionFlags(), vk::PipelineBindPoint::eGraphics,
|
|
1, &colorGeneral, 1, &pictureReference, nullptr, nullptr),
|
|
};
|
|
|
|
const vk::PipelineStageFlags fbStages = vk::PipelineStageFlagBits::eEarlyFragmentTests | vk::PipelineStageFlagBits::eLateFragmentTests
|
|
| vk::PipelineStageFlagBits::eFragmentShader | vk::PipelineStageFlagBits::eColorAttachmentOutput;
|
|
const vk::AccessFlags writes = vk::AccessFlagBits::eColorAttachmentWrite | vk::AccessFlagBits::eDepthStencilAttachmentWrite
|
|
| vk::AccessFlagBits::eShaderWrite;
|
|
const vk::AccessFlags all = writes | vk::AccessFlagBits::eColorAttachmentRead | vk::AccessFlagBits::eDepthStencilAttachmentRead
|
|
| vk::AccessFlagBits::eInputAttachmentRead | vk::AccessFlagBits::eShaderRead;
|
|
std::vector<vk::SubpassDependency> dependencies;
|
|
/* The samples, the lists and the picture's image are the same ones for
|
|
* every render pass and every frame: what came before is done with
|
|
* them before any subpass uses them */
|
|
for (u32 subpass = 0; subpass < 4; subpass++)
|
|
dependencies.emplace_back(VK_SUBPASS_EXTERNAL, subpass, fbStages, fbStages, writes | vk::AccessFlagBits::eShaderRead, all);
|
|
// each subpass reads and writes what the one before wrote, where it is
|
|
for (u32 subpass = 0; subpass < 3; subpass++)
|
|
dependencies.emplace_back(subpass, subpass + 1, fbStages, fbStages, writes, all, vk::DependencyFlagBits::eByRegion);
|
|
// (the barriers between the passes over the lists in subpasses 2 and 3)
|
|
for (u32 subpass = 2; subpass < 4; subpass++)
|
|
dependencies.emplace_back(subpass, subpass, vk::PipelineStageFlagBits::eFragmentShader, vk::PipelineStageFlagBits::eFragmentShader,
|
|
vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite, vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite,
|
|
vk::DependencyFlagBits::eByRegion);
|
|
// what the render pass leaves - the picture, to be shown or sampled; the samples, for the next render pass
|
|
for (u32 subpass = 2; subpass < 4; subpass++)
|
|
dependencies.emplace_back(subpass, VK_SUBPASS_EXTERNAL, fbStages, fbStages | vk::PipelineStageFlagBits::eTransfer,
|
|
writes, all | vk::AccessFlagBits::eTransferRead);
|
|
|
|
return GetContext()->GetDevice().createRenderPassUnique(vk::RenderPassCreateInfo(vk::RenderPassCreateFlags(),
|
|
ARRAY_SIZE(attachmentDescriptions), attachmentDescriptions,
|
|
ARRAY_SIZE(subpasses), subpasses,
|
|
dependencies.size(), dependencies.data()));
|
|
}
|