/*
    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"

void RenderPasses::Reset() {
	for (auto& renderPass : renderPasses)
		renderPass.reset();
}

vk::UniqueRenderPass RenderPasses::MakeRenderPass(bool initial, bool last, bool loadClear)
{
	vk::AttachmentDescription attach0 = GetAttachment0Description(initial, last, loadClear);
    std::array<vk::AttachmentDescription, 4> attachmentDescriptions = {
    		// Swap chain image
    		attach0,
			// OP+PT color attachment
			vk::AttachmentDescription(vk::AttachmentDescriptionFlags(), vk::Format::eR8G8B8A8Unorm, vk::SampleCountFlagBits::e1,
					loadClear && initial ? vk::AttachmentLoadOp::eClear : vk::AttachmentLoadOp::eLoad,
					last ? vk::AttachmentStoreOp::eDontCare : vk::AttachmentStoreOp::eStore,
					vk::AttachmentLoadOp::eDontCare, vk::AttachmentStoreOp::eDontCare,
					loadClear && initial ? vk::ImageLayout::eUndefined : attach0.finalLayout, attach0.finalLayout), // initial layout is eUndefined for rtt, so use final layout instead
			// OP+PT depth attachment
			vk::AttachmentDescription(vk::AttachmentDescriptionFlags(), GetContext()->GetDepthFormat(), vk::SampleCountFlagBits::e1,
					initial ? vk::AttachmentLoadOp::eClear : vk::AttachmentLoadOp::eLoad,
					last ? vk::AttachmentStoreOp::eDontCare : vk::AttachmentStoreOp::eStore,
					vk::AttachmentLoadOp::eClear, vk::AttachmentStoreOp::eDontCare,
					initial ? vk::ImageLayout::eUndefined : vk::ImageLayout::eDepthStencilAttachmentOptimal, vk::ImageLayout::eDepthStencilAttachmentOptimal),
			// OP+PT depth attachment for subpass 1
			vk::AttachmentDescription(vk::AttachmentDescriptionFlags(), GetContext()->GetDepthFormat(), vk::SampleCountFlagBits::e1,
					initial ? vk::AttachmentLoadOp::eClear : vk::AttachmentLoadOp::eLoad,
					last ? vk::AttachmentStoreOp::eDontCare : vk::AttachmentStoreOp::eStore,
					vk::AttachmentLoadOp::eClear, vk::AttachmentStoreOp::eDontCare,
					initial ? vk::ImageLayout::eUndefined : vk::ImageLayout::eDepthStencilAttachmentOptimal, vk::ImageLayout::eDepthStencilAttachmentOptimal),
    };
    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, vk::ImageLayout::eDepthStencilReadOnlyOptimal);
    vk::AttachmentReference depthReference2(3, vk::ImageLayout::eDepthStencilAttachmentOptimal);
    vk::AttachmentReference colorInput(1, vk::ImageLayout::eShaderReadOnlyOptimal);

    std::array<vk::SubpassDescription, 3> subpasses = {
    	// Depth and modvol pass	FIXME subpass 0 shouldn't reference the color attachment
    	vk::SubpassDescription(vk::SubpassDescriptionFlags(), vk::PipelineBindPoint::eGraphics,
    			nullptr,
				colorReference,
				nullptr,
				&depthReference),
    	// Color pass
    	vk::SubpassDescription(vk::SubpassDescriptionFlags(), vk::PipelineBindPoint::eGraphics,
    			depthReadOnlyRef,
				colorReference,
				nullptr,
				&depthReference2),
    	// Final pass
    	vk::SubpassDescription(vk::SubpassDescriptionFlags(), vk::PipelineBindPoint::eGraphics,
    			colorInput,
				swapChainReference,
				nullptr,
				&depthReference),	// depth-only Tr pass when continuation
    };

    std::vector<vk::SubpassDependency> dependencies = GetSubpassDependencies();
    dependencies.emplace_back(vk::SubpassExternal, 1, vk::PipelineStageFlagBits::eFragmentShader, vk::PipelineStageFlagBits::eColorAttachmentOutput,
    		vk::AccessFlagBits::eInputAttachmentRead, vk::AccessFlagBits::eColorAttachmentWrite, vk::DependencyFlagBits::eByRegion);
    // The OP+PT color and depth/stencil attachments and the OIT storage buffers are shared by consecutive
    // render passes and by all the frames in flight. Previous reads (input attachments) and writes (color, depth, a-buffers)
    // must be done before these attachments are cleared or loaded and the a-buffers are written again.
    // Attachments 1 and 2 are first used in subpass 0 and depth attachment 3 in subpass 1.
    constexpr vk::PipelineStageFlags oitStages = vk::PipelineStageFlagBits::eEarlyFragmentTests | vk::PipelineStageFlagBits::eLateFragmentTests
    		| vk::PipelineStageFlagBits::eFragmentShader | vk::PipelineStageFlagBits::eColorAttachmentOutput;
    for (u32 subpass : { 0u, 1u })
    	dependencies.emplace_back(vk::SubpassExternal, 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);
    // Attachment 0 is first used in subpass 2. With multiple render passes, it was the OP+PT color attachment
    // of the previous render pass (written as color and read as input attachment).
    dependencies.emplace_back(vk::SubpassExternal, 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);
    // Chains the external dependency above with the OP+PT color attachment transition to 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 OP+PT color 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(1, 1, vk::PipelineStageFlagBits::eFragmentShader, vk::PipelineStageFlagBits::eFragmentShader,
    		vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite,
			vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite,
			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);
	if (GetContext()->GetVendorID() == VulkanContext::VENDOR_ARM) {
		// Avoid glitches in upper left corner with Mali GPUs.
		// Using eTopOfPipe as destination since eFragmentShader/eInputAttachmentRead|eShaderRead fails to fix the problem, which is odd.
		dependencies.emplace_back(2, vk::SubpassExternal, vk::PipelineStageFlagBits::eColorAttachmentOutput, vk::PipelineStageFlagBits::eTopOfPipe,
				vk::AccessFlagBits::eColorAttachmentWrite, vk::AccessFlagBits::eNone);
	}

    return GetContext()->GetDevice().createRenderPassUnique(vk::RenderPassCreateInfo(vk::RenderPassCreateFlags(),
    		attachmentDescriptions,
    		subpasses,
			dependencies));
}

vk::AttachmentDescription RenderPasses::GetAttachment0Description(bool initial, bool last, bool loadClear) const
{
	return vk::AttachmentDescription(vk::AttachmentDescriptionFlags(), vk::Format::eR8G8B8A8Unorm, vk::SampleCountFlagBits::e1,
			initial && loadClear ? vk::AttachmentLoadOp::eClear : vk::AttachmentLoadOp::eLoad, vk::AttachmentStoreOp::eStore,
			vk::AttachmentLoadOp::eDontCare, vk::AttachmentStoreOp::eDontCare,
			config::EmulateFramebuffer && initial ? vk::ImageLayout::eTransferSrcOptimal : vk::ImageLayout::eShaderReadOnlyOptimal,
			config::EmulateFramebuffer && last ? vk::ImageLayout::eTransferSrcOptimal : vk::ImageLayout::eShaderReadOnlyOptimal);
}

std::vector<vk::SubpassDependency> RenderPasses::GetSubpassDependencies() const
{
	if (config::EmulateFramebuffer)
		return { { 2, vk::SubpassExternal,
				vk::PipelineStageFlagBits::eColorAttachmentOutput, vk::PipelineStageFlagBits::eTransfer | vk::PipelineStageFlagBits::eHost,
				vk::AccessFlagBits::eColorAttachmentWrite, vk::AccessFlagBits::eTransferRead | vk::AccessFlagBits::eHostRead, vk::DependencyFlagBits::eByRegion } };
	else
		return { { 2, vk::SubpassExternal,
				vk::PipelineStageFlagBits::eColorAttachmentOutput, vk::PipelineStageFlagBits::eFragmentShader,
				vk::AccessFlagBits::eColorAttachmentWrite, vk::AccessFlagBits::eShaderRead, vk::DependencyFlagBits::eByRegion } };
}

vk::AttachmentDescription RttRenderPasses::GetAttachment0Description(bool initial, bool last, bool loadClear) const
{
	return vk::AttachmentDescription(vk::AttachmentDescriptionFlags(), vk::Format::eR8G8B8A8Unorm, vk::SampleCountFlagBits::e1,
			vk::AttachmentLoadOp::eClear, vk::AttachmentStoreOp::eStore,
			vk::AttachmentLoadOp::eDontCare, vk::AttachmentStoreOp::eDontCare,
			vk::ImageLayout::eUndefined,
			config::RenderToTextureBuffer && last ? vk::ImageLayout::eTransferSrcOptimal : vk::ImageLayout::eShaderReadOnlyOptimal);
}

std::vector<vk::SubpassDependency> RttRenderPasses::GetSubpassDependencies() const
{
	if (config::RenderToTextureBuffer)
		return { { 2, vk::SubpassExternal,
				vk::PipelineStageFlagBits::eColorAttachmentOutput, vk::PipelineStageFlagBits::eTransfer | vk::PipelineStageFlagBits::eHost,
				vk::AccessFlagBits::eColorAttachmentWrite, vk::AccessFlagBits::eTransferRead | vk::AccessFlagBits::eHostRead } };
	else
		return { { 2, vk::SubpassExternal, vk::PipelineStageFlagBits::eColorAttachmentOutput, vk::PipelineStageFlagBits::eFragmentShader,
				vk::AccessFlagBits::eColorAttachmentWrite, vk::AccessFlagBits::eShaderRead } };
}