dolphin/Source/Core/VideoBackends/Vulkan/Renderer.cpp

1526 lines
56 KiB
C++
Raw Normal View History

2016-08-13 12:57:50 +00:00
// Copyright 2016 Dolphin Emulator Project
// Licensed under GPLv2+
// Refer to the license.txt file included.
#include <cstddef>
2016-08-13 12:57:50 +00:00
#include <cstdio>
#include <limits>
#include <string>
#include <tuple>
#include "Common/Assert.h"
#include "Common/CommonTypes.h"
#include "Common/Logging/Log.h"
#include "Common/MsgHandler.h"
2016-08-13 12:57:50 +00:00
#include "Core/Core.h"
2016-08-13 12:57:50 +00:00
#include "VideoBackends/Vulkan/BoundingBox.h"
#include "VideoBackends/Vulkan/CommandBufferManager.h"
#include "VideoBackends/Vulkan/FramebufferManager.h"
#include "VideoBackends/Vulkan/ObjectCache.h"
#include "VideoBackends/Vulkan/PostProcessing.h"
2016-08-13 12:57:50 +00:00
#include "VideoBackends/Vulkan/RasterFont.h"
#include "VideoBackends/Vulkan/Renderer.h"
#include "VideoBackends/Vulkan/StagingTexture2D.h"
2016-08-13 12:57:50 +00:00
#include "VideoBackends/Vulkan/StateTracker.h"
#include "VideoBackends/Vulkan/SwapChain.h"
#include "VideoBackends/Vulkan/TextureCache.h"
2016-08-13 12:57:50 +00:00
#include "VideoBackends/Vulkan/Util.h"
#include "VideoBackends/Vulkan/VKTexture.h"
2016-08-13 12:57:50 +00:00
#include "VideoBackends/Vulkan/VulkanContext.h"
#include "VideoCommon/AVIDump.h"
2016-08-13 12:57:50 +00:00
#include "VideoCommon/BPFunctions.h"
#include "VideoCommon/BPMemory.h"
#include "VideoCommon/DriverDetails.h"
2016-08-13 12:57:50 +00:00
#include "VideoCommon/OnScreenDisplay.h"
#include "VideoCommon/PixelEngine.h"
#include "VideoCommon/PixelShaderManager.h"
#include "VideoCommon/RenderState.h"
2016-08-13 12:57:50 +00:00
#include "VideoCommon/SamplerCommon.h"
#include "VideoCommon/TextureCacheBase.h"
#include "VideoCommon/VideoBackendBase.h"
2016-08-13 12:57:50 +00:00
#include "VideoCommon/VideoConfig.h"
#include "VideoCommon/XFMemory.h"
2016-08-13 12:57:50 +00:00
namespace Vulkan
{
Renderer::Renderer(std::unique_ptr<SwapChain> swap_chain)
: ::Renderer(swap_chain ? static_cast<int>(swap_chain->GetWidth()) : 1,
swap_chain ? static_cast<int>(swap_chain->GetHeight()) : 0),
m_swap_chain(std::move(swap_chain))
2016-08-13 12:57:50 +00:00
{
UpdateActiveConfig();
2016-08-13 12:57:50 +00:00
// Set to something invalid, forcing all states to be re-initialized.
for (size_t i = 0; i < m_sampler_states.size(); i++)
m_sampler_states[i].bits = std::numeric_limits<decltype(m_sampler_states[i].bits)>::max();
}
Renderer::~Renderer()
{
UpdateActiveConfig();
// Ensure all frames are written to frame dump at shutdown.
if (m_frame_dumping_active)
EndFrameDumping();
DestroyFrameDumpResources();
DestroyShaders();
2016-08-13 12:57:50 +00:00
DestroySemaphores();
}
Renderer* Renderer::GetInstance()
{
return static_cast<Renderer*>(g_renderer.get());
}
bool Renderer::Initialize()
2016-08-13 12:57:50 +00:00
{
BindEFBToStateTracker();
if (!CreateSemaphores())
{
PanicAlert("Failed to create semaphores.");
return false;
}
if (!CompileShaders())
{
PanicAlert("Failed to compile shaders.");
return false;
}
m_raster_font = std::make_unique<RasterFont>();
if (!m_raster_font->Initialize())
{
PanicAlert("Failed to initialize raster font.");
return false;
}
m_bounding_box = std::make_unique<BoundingBox>();
if (!m_bounding_box->Initialize())
{
PanicAlert("Failed to initialize bounding box.");
return false;
}
if (g_vulkan_context->SupportsBoundingBox())
{
// Bind bounding box to state tracker
StateTracker::GetInstance()->SetBBoxBuffer(m_bounding_box->GetGPUBuffer(),
m_bounding_box->GetGPUBufferOffset(),
m_bounding_box->GetGPUBufferSize());
2016-08-13 12:57:50 +00:00
}
// Initialize post processing.
m_post_processor = std::make_unique<VulkanPostProcessing>();
if (!static_cast<VulkanPostProcessing*>(m_post_processor.get())
->Initialize(m_raster_font->GetTexture()))
{
PanicAlert("failed to initialize post processor.");
return false;
}
2016-08-13 12:57:50 +00:00
// Various initialization routines will have executed commands on the command buffer.
// Execute what we have done before beginning the first frame.
g_command_buffer_mgr->PrepareToSubmitCommandBuffer();
g_command_buffer_mgr->SubmitCommandBuffer(false);
BeginFrame();
return true;
}
bool Renderer::CreateSemaphores()
{
// Create two semaphores, one that is triggered when the swapchain buffer is ready, another after
// submit and before present
VkSemaphoreCreateInfo semaphore_info = {
VK_STRUCTURE_TYPE_SEMAPHORE_CREATE_INFO, // VkStructureType sType
nullptr, // const void* pNext
0 // VkSemaphoreCreateFlags flags
};
VkResult res;
if ((res = vkCreateSemaphore(g_vulkan_context->GetDevice(), &semaphore_info, nullptr,
&m_image_available_semaphore)) != VK_SUCCESS ||
(res = vkCreateSemaphore(g_vulkan_context->GetDevice(), &semaphore_info, nullptr,
&m_rendering_finished_semaphore)) != VK_SUCCESS)
{
LOG_VULKAN_ERROR(res, "vkCreateSemaphore failed: ");
return false;
}
return true;
}
void Renderer::DestroySemaphores()
{
if (m_image_available_semaphore)
{
vkDestroySemaphore(g_vulkan_context->GetDevice(), m_image_available_semaphore, nullptr);
m_image_available_semaphore = VK_NULL_HANDLE;
2016-08-13 12:57:50 +00:00
}
if (m_rendering_finished_semaphore)
{
vkDestroySemaphore(g_vulkan_context->GetDevice(), m_rendering_finished_semaphore, nullptr);
m_rendering_finished_semaphore = VK_NULL_HANDLE;
2016-08-13 12:57:50 +00:00
}
}
void Renderer::RenderText(const std::string& text, int left, int top, u32 color)
{
u32 backbuffer_width = m_swap_chain->GetWidth();
u32 backbuffer_height = m_swap_chain->GetHeight();
m_raster_font->PrintMultiLineText(m_swap_chain->GetRenderPass(), text,
left * 2.0f / static_cast<float>(backbuffer_width) - 1,
1 - top * 2.0f / static_cast<float>(backbuffer_height),
backbuffer_width, backbuffer_height, color);
}
u32 Renderer::AccessEFB(EFBAccessType type, u32 x, u32 y, u32 poke_data)
{
if (type == EFBAccessType::PeekColor)
2016-08-13 12:57:50 +00:00
{
u32 color = FramebufferManager::GetInstance()->PeekEFBColor(x, y);
2016-08-13 12:57:50 +00:00
// a little-endian value is expected to be returned
color = ((color & 0xFF00FF00) | ((color >> 16) & 0xFF) | ((color << 16) & 0xFF0000));
// check what to do with the alpha channel (GX_PokeAlphaRead)
PixelEngine::UPEAlphaReadReg alpha_read_mode = PixelEngine::GetAlphaReadMode();
if (bpmem.zcontrol.pixel_format == PEControl::RGBA6_Z24)
{
color = RGBA8ToRGBA6ToRGBA8(color);
}
else if (bpmem.zcontrol.pixel_format == PEControl::RGB565_Z16)
{
color = RGBA8ToRGB565ToRGBA8(color);
}
if (bpmem.zcontrol.pixel_format != PEControl::RGBA6_Z24)
{
color |= 0xFF000000;
}
if (alpha_read_mode.ReadMode == 2)
{
return color; // GX_READ_NONE
}
else if (alpha_read_mode.ReadMode == 1)
{
return color | 0xFF000000; // GX_READ_FF
}
else /*if(alpha_read_mode.ReadMode == 0)*/
{
return color & 0x00FFFFFF; // GX_READ_00
}
}
else // if (type == EFBAccessType::PeekZ)
2016-08-13 12:57:50 +00:00
{
// Depth buffer is inverted for improved precision near far plane
float depth = 1.0f - FramebufferManager::GetInstance()->PeekEFBDepth(x, y);
2016-08-13 12:57:50 +00:00
u32 ret = 0;
if (bpmem.zcontrol.pixel_format == PEControl::RGB565_Z16)
{
// if Z is in 16 bit format you must return a 16 bit integer
ret = MathUtil::Clamp<u32>(static_cast<u32>(depth * 65536.0f), 0, 0xFFFF);
}
else
{
ret = MathUtil::Clamp<u32>(static_cast<u32>(depth * 16777216.0f), 0, 0xFFFFFF);
}
return ret;
}
}
void Renderer::PokeEFB(EFBAccessType type, const EfbPokeData* points, size_t num_points)
{
if (type == EFBAccessType::PokeColor)
2016-08-13 12:57:50 +00:00
{
for (size_t i = 0; i < num_points; i++)
{
// Convert to expected format (BGRA->RGBA)
// TODO: Check alpha, depending on mode?
const EfbPokeData& point = points[i];
u32 color = ((point.data & 0xFF00FF00) | ((point.data >> 16) & 0xFF) |
((point.data << 16) & 0xFF0000));
FramebufferManager::GetInstance()->PokeEFBColor(point.x, point.y, color);
2016-08-13 12:57:50 +00:00
}
}
else // if (type == EFBAccessType::PokeZ)
2016-08-13 12:57:50 +00:00
{
for (size_t i = 0; i < num_points; i++)
{
// Convert to floating-point depth.
const EfbPokeData& point = points[i];
float depth = (1.0f - float(point.data & 0xFFFFFF) / 16777216.0f);
FramebufferManager::GetInstance()->PokeEFBDepth(point.x, point.y, depth);
2016-08-13 12:57:50 +00:00
}
}
}
u16 Renderer::BBoxRead(int index)
{
s32 value = m_bounding_box->Get(static_cast<size_t>(index));
2016-08-13 12:57:50 +00:00
// Here we get the min/max value of the truncated position of the upscaled framebuffer.
// So we have to correct them to the unscaled EFB sizes.
if (index < 2)
{
// left/right
value = value * EFB_WIDTH / m_target_width;
2016-08-13 12:57:50 +00:00
}
else
{
// up/down
value = value * EFB_HEIGHT / m_target_height;
2016-08-13 12:57:50 +00:00
}
// fix max values to describe the outer border
if (index & 1)
value++;
return static_cast<u16>(value);
}
void Renderer::BBoxWrite(int index, u16 value)
{
s32 scaled_value = static_cast<s32>(value);
// fix max values to describe the outer border
if (index & 1)
scaled_value--;
// scale to internal resolution
if (index < 2)
{
// left/right
scaled_value = scaled_value * m_target_width / EFB_WIDTH;
2016-08-13 12:57:50 +00:00
}
else
{
// up/down
scaled_value = scaled_value * m_target_height / EFB_HEIGHT;
2016-08-13 12:57:50 +00:00
}
m_bounding_box->Set(static_cast<size_t>(index), scaled_value);
2016-08-13 12:57:50 +00:00
}
TargetRectangle Renderer::ConvertEFBRectangle(const EFBRectangle& rc)
{
TargetRectangle result;
result.left = EFBToScaledX(rc.left);
result.top = EFBToScaledY(rc.top);
result.right = EFBToScaledX(rc.right);
result.bottom = EFBToScaledY(rc.bottom);
return result;
}
void Renderer::BeginFrame()
{
// Activate a new command list, and restore state ready for the next draw
g_command_buffer_mgr->ActivateCommandBuffer();
// Restore the EFB color texture to color attachment ready for rendering the next frame.
FramebufferManager::GetInstance()->GetEFBColorTexture()->TransitionToLayout(
g_command_buffer_mgr->GetCurrentCommandBuffer(), VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL);
2016-08-13 12:57:50 +00:00
// Ensure that the state tracker rebinds everything, and allocates a new set
// of descriptors out of the next pool.
StateTracker::GetInstance()->InvalidateDescriptorSets();
StateTracker::GetInstance()->InvalidateConstants();
StateTracker::GetInstance()->SetPendingRebind();
2016-08-13 12:57:50 +00:00
}
void Renderer::ClearScreen(const EFBRectangle& rc, bool color_enable, bool alpha_enable,
bool z_enable, u32 color, u32 z)
{
// Native -> EFB coordinates
TargetRectangle target_rc = Renderer::ConvertEFBRectangle(rc);
// Size we pass this size to vkBeginRenderPass, it has to be clamped to the framebuffer
// dimensions. The other backends just silently ignore this case.
target_rc.ClampUL(0, 0, m_target_width, m_target_height);
VkRect2D target_vk_rc = {
{target_rc.left, target_rc.top},
{static_cast<uint32_t>(target_rc.GetWidth()), static_cast<uint32_t>(target_rc.GetHeight())}};
// Determine whether the EFB has an alpha channel. If it doesn't, we can clear the alpha
// channel to 0xFF. This hopefully allows us to use the fast path in most cases.
if (bpmem.zcontrol.pixel_format == PEControl::RGB565_Z16 ||
bpmem.zcontrol.pixel_format == PEControl::RGB8_Z24 ||
bpmem.zcontrol.pixel_format == PEControl::Z24)
{
// Force alpha writes, and clear the alpha channel. This is different to the other backends,
// where the existing values of the alpha channel are preserved.
alpha_enable = true;
color &= 0x00FFFFFF;
}
// Convert RGBA8 -> floating-point values.
VkClearValue clear_color_value = {};
VkClearValue clear_depth_value = {};
clear_color_value.color.float32[0] = static_cast<float>((color >> 16) & 0xFF) / 255.0f;
clear_color_value.color.float32[1] = static_cast<float>((color >> 8) & 0xFF) / 255.0f;
clear_color_value.color.float32[2] = static_cast<float>((color >> 0) & 0xFF) / 255.0f;
clear_color_value.color.float32[3] = static_cast<float>((color >> 24) & 0xFF) / 255.0f;
clear_depth_value.depthStencil.depth = (1.0f - (static_cast<float>(z & 0xFFFFFF) / 16777216.0f));
// If we're not in a render pass (start of the frame), we can use a clear render pass
// to discard the data, rather than loading and then clearing.
bool use_clear_render_pass = (color_enable && alpha_enable && z_enable);
if (StateTracker::GetInstance()->InRenderPass())
2016-08-13 12:57:50 +00:00
{
// Prefer not to end a render pass just to do a clear.
use_clear_render_pass = false;
2016-08-13 12:57:50 +00:00
}
// Fastest path: Use a render pass to clear the buffers.
if (use_clear_render_pass)
2016-08-13 12:57:50 +00:00
{
VkClearValue clear_values[2] = {clear_color_value, clear_depth_value};
StateTracker::GetInstance()->BeginClearRenderPass(target_vk_rc, clear_values);
return;
2016-08-13 12:57:50 +00:00
}
// Fast path: Use vkCmdClearAttachments to clear the buffers within a render path
// We can't use this when preserving alpha but clearing color.
if (g_ActiveConfig.iMultisamples == 1 ||
!DriverDetails::HasBug(DriverDetails::BUG_BROKEN_MSAA_VKCMDCLEARATTACHMENTS))
{
VkClearAttachment clear_attachments[2];
uint32_t num_clear_attachments = 0;
if (color_enable && alpha_enable)
{
clear_attachments[num_clear_attachments].aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
clear_attachments[num_clear_attachments].colorAttachment = 0;
clear_attachments[num_clear_attachments].clearValue = clear_color_value;
num_clear_attachments++;
color_enable = false;
alpha_enable = false;
}
if (z_enable)
{
clear_attachments[num_clear_attachments].aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT;
clear_attachments[num_clear_attachments].colorAttachment = 0;
clear_attachments[num_clear_attachments].clearValue = clear_depth_value;
num_clear_attachments++;
z_enable = false;
}
if (num_clear_attachments > 0)
{
VkClearRect vk_rect = {target_vk_rc, 0, FramebufferManager::GetInstance()->GetEFBLayers()};
if (!StateTracker::GetInstance()->IsWithinRenderArea(
target_vk_rc.offset.x, target_vk_rc.offset.y, target_vk_rc.extent.width,
target_vk_rc.extent.height))
{
StateTracker::GetInstance()->EndClearRenderPass();
}
StateTracker::GetInstance()->BeginRenderPass();
2016-08-13 12:57:50 +00:00
vkCmdClearAttachments(g_command_buffer_mgr->GetCurrentCommandBuffer(), num_clear_attachments,
clear_attachments, 1, &vk_rect);
}
2016-08-13 12:57:50 +00:00
}
// Anything left over for the slow path?
2016-08-13 12:57:50 +00:00
if (!color_enable && !alpha_enable && !z_enable)
return;
// Clearing must occur within a render pass.
if (!StateTracker::GetInstance()->IsWithinRenderArea(target_vk_rc.offset.x, target_vk_rc.offset.y,
target_vk_rc.extent.width,
target_vk_rc.extent.height))
{
StateTracker::GetInstance()->EndClearRenderPass();
}
StateTracker::GetInstance()->BeginRenderPass();
StateTracker::GetInstance()->SetPendingRebind();
2016-08-13 12:57:50 +00:00
// Mask away the appropriate colors and use a shader
BlendingState blend_state = Util::GetNoBlendingBlendState();
blend_state.colorupdate = color_enable;
blend_state.alphaupdate = alpha_enable;
2016-08-13 12:57:50 +00:00
DepthStencilState depth_state = Util::GetNoDepthTestingDepthStencilState();
depth_state.test_enable = z_enable ? VK_TRUE : VK_FALSE;
depth_state.write_enable = z_enable ? VK_TRUE : VK_FALSE;
depth_state.compare_op = VK_COMPARE_OP_ALWAYS;
RasterizationState rs_state = Util::GetNoCullRasterizationState();
rs_state.per_sample_shading = g_ActiveConfig.bSSAA ? VK_TRUE : VK_FALSE;
rs_state.samples = FramebufferManager::GetInstance()->GetEFBSamples();
2016-08-13 12:57:50 +00:00
// No need to start a new render pass, but we do need to restore viewport state
UtilityShaderDraw draw(g_command_buffer_mgr->GetCurrentCommandBuffer(),
g_object_cache->GetPipelineLayout(PIPELINE_LAYOUT_STANDARD),
FramebufferManager::GetInstance()->GetEFBLoadRenderPass(),
g_shader_cache->GetPassthroughVertexShader(),
g_shader_cache->GetPassthroughGeometryShader(), m_clear_fragment_shader);
2016-08-13 12:57:50 +00:00
draw.SetRasterizationState(rs_state);
draw.SetDepthStencilState(depth_state);
draw.SetBlendState(blend_state);
draw.DrawColoredQuad(target_rc.left, target_rc.top, target_rc.GetWidth(), target_rc.GetHeight(),
clear_color_value.color.float32[0], clear_color_value.color.float32[1],
clear_color_value.color.float32[2], clear_color_value.color.float32[3],
clear_depth_value.depthStencil.depth);
2016-08-13 12:57:50 +00:00
}
void Renderer::ReinterpretPixelData(unsigned int convtype)
{
StateTracker::GetInstance()->EndRenderPass();
StateTracker::GetInstance()->SetPendingRebind();
FramebufferManager::GetInstance()->ReinterpretPixelData(convtype);
2016-08-13 12:57:50 +00:00
// EFB framebuffer has now changed, so update accordingly.
BindEFBToStateTracker();
}
void Renderer::SwapImpl(u32 xfb_addr, u32 fb_width, u32 fb_stride, u32 fb_height,
2016-10-08 10:56:28 +00:00
const EFBRectangle& rc, u64 ticks, float gamma)
2016-08-13 12:57:50 +00:00
{
// Pending/batched EFB pokes should be included in the final image.
FramebufferManager::GetInstance()->FlushEFBPokes();
2016-08-13 12:57:50 +00:00
// Check that we actually have an image to render in XFB-on modes.
if ((!m_xfb_written && !g_ActiveConfig.RealXFBEnabled()) || !fb_width || !fb_height)
{
Core::Callback_VideoCopiedToXFB(false);
return;
}
u32 xfb_count = 0;
const XFBSourceBase* const* xfb_sources =
FramebufferManager::GetXFBSource(xfb_addr, fb_stride, fb_height, &xfb_count);
if (g_ActiveConfig.VirtualXFBEnabled() && (!xfb_sources || xfb_count == 0))
{
Core::Callback_VideoCopiedToXFB(false);
return;
}
2016-08-13 12:57:50 +00:00
// End the current render pass.
StateTracker::GetInstance()->EndRenderPass();
StateTracker::GetInstance()->OnEndFrame();
2016-08-13 12:57:50 +00:00
// There are a few variables which can alter the final window draw rectangle, and some of them
// are determined by guest state. Currently, the only way to catch these is to update every frame.
UpdateDrawRectangle();
// Scale the source rectangle to the internal resolution when XFB is disabled.
TargetRectangle scaled_efb_rect = Renderer::ConvertEFBRectangle(rc);
// If MSAA is enabled, and we're not using XFB, we need to resolve the EFB framebuffer before
// rendering the final image to the screen, or dumping the frame. This is because we can't resolve
// an image within a render pass, which will have already started by the time it is used.
TransitionBuffersForSwap(scaled_efb_rect, xfb_sources, xfb_count);
// Render the frame dump image if enabled.
if (IsFrameDumping())
{
// If we haven't dumped a single frame yet, set up frame dumping.
if (!m_frame_dumping_active)
StartFrameDumping();
DrawFrameDump(scaled_efb_rect, xfb_addr, xfb_sources, xfb_count, fb_width, fb_stride, fb_height,
ticks);
}
else
{
// If frame dumping was previously enabled, flush all frames and remove the fence callback.
if (m_frame_dumping_active)
EndFrameDumping();
}
2016-08-13 12:57:50 +00:00
// Ensure the worker thread is not still submitting a previous command buffer.
// In other words, the last frame has been submitted (otherwise the next call would
// be a race, as the image may not have been consumed yet).
g_command_buffer_mgr->PrepareToSubmitCommandBuffer();
// Draw to the screen if we have a swap chain.
if (m_swap_chain)
{
DrawScreen(scaled_efb_rect, xfb_addr, xfb_sources, xfb_count, fb_width, fb_stride, fb_height);
2016-08-13 12:57:50 +00:00
// Submit the current command buffer, signaling rendering finished semaphore when it's done
// Because this final command buffer is rendering to the swap chain, we need to wait for
// the available semaphore to be signaled before executing the buffer. This final submission
// can happen off-thread in the background while we're preparing the next frame.
g_command_buffer_mgr->SubmitCommandBuffer(
true, m_image_available_semaphore, m_rendering_finished_semaphore,
2016-08-13 12:57:50 +00:00
m_swap_chain->GetSwapChain(), m_swap_chain->GetCurrentImageIndex());
}
else
{
// No swap chain, just execute command buffer.
g_command_buffer_mgr->SubmitCommandBuffer(true);
2016-08-13 12:57:50 +00:00
}
// NOTE: It is important that no rendering calls are made to the EFB between submitting the
// (now-previous) frame and after the below config checks are completed. If the target size
// changes, as the resize methods to not defer the destruction of the framebuffer, the current
// command buffer will contain references to a now non-existent framebuffer.
2016-08-13 12:57:50 +00:00
// Prep for the next frame (get command buffer ready) before doing anything else.
BeginFrame();
// Determine what (if anything) has changed in the config.
CheckForConfigChanges();
2016-08-13 12:57:50 +00:00
// Handle host window resizes.
CheckForSurfaceChange();
2016-08-13 12:57:50 +00:00
// Handle output size changes from the guest.
// There is a downside to doing this here is that if the game changes its XFB source area,
// the changes will be delayed by one frame. For the moment it has to be done here because
// this can cause a target size change, which would result in a black frame if done earlier.
2016-08-13 12:57:50 +00:00
CheckForTargetResize(fb_width, fb_stride, fb_height);
// Update the window size based on the frame that was just rendered.
// Due to depending on guest state, we need to call this every frame.
SetWindowSize(static_cast<int>(fb_stride), static_cast<int>(fb_height));
// Clean up stale textures.
TextureCache::GetInstance()->Cleanup(frameCount);
2017-07-20 05:25:35 +00:00
// Pull in now-ready async shaders.
g_shader_cache->RetrieveAsyncShaders();
2016-08-13 12:57:50 +00:00
}
void Renderer::TransitionBuffersForSwap(const TargetRectangle& scaled_rect,
const XFBSourceBase* const* xfb_sources, u32 xfb_count)
{
VkCommandBuffer command_buffer = g_command_buffer_mgr->GetCurrentCommandBuffer();
if (!g_ActiveConfig.bUseXFB)
{
// Drawing EFB direct.
if (g_ActiveConfig.iMultisamples > 1)
{
// While the source rect can be out-of-range when drawing, the resolve rectangle must be
// within the bounds of the texture.
VkRect2D region = {
{scaled_rect.left, scaled_rect.top},
{static_cast<u32>(scaled_rect.GetWidth()), static_cast<u32>(scaled_rect.GetHeight())}};
region = Util::ClampRect2D(region, FramebufferManager::GetInstance()->GetEFBWidth(),
FramebufferManager::GetInstance()->GetEFBHeight());
Vulkan::Texture2D* rtex = FramebufferManager::GetInstance()->ResolveEFBColorTexture(region);
rtex->TransitionToLayout(command_buffer, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL);
}
else
{
FramebufferManager::GetInstance()->GetEFBColorTexture()->TransitionToLayout(
command_buffer, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL);
}
return;
}
// Drawing XFB sources, so transition all of them.
// Don't need the EFB, so leave it as-is.
for (u32 i = 0; i < xfb_count; i++)
{
const XFBSource* xfb_source = static_cast<const XFBSource*>(xfb_sources[i]);
xfb_source->GetTexture()->GetRawTexIdentifier()->TransitionToLayout(
command_buffer, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL);
}
}
void Renderer::DrawFrame(VkRenderPass render_pass, const TargetRectangle& target_rect,
const TargetRectangle& scaled_efb_rect, u32 xfb_addr,
const XFBSourceBase* const* xfb_sources, u32 xfb_count, u32 fb_width,
u32 fb_stride, u32 fb_height)
{
if (!g_ActiveConfig.bUseXFB)
DrawEFB(render_pass, target_rect, scaled_efb_rect);
else if (!g_ActiveConfig.bUseRealXFB)
DrawVirtualXFB(render_pass, target_rect, xfb_addr, xfb_sources, xfb_count, fb_width, fb_stride,
fb_height);
else
DrawRealXFB(render_pass, target_rect, xfb_sources, xfb_count, fb_width, fb_stride, fb_height);
}
void Renderer::DrawEFB(VkRenderPass render_pass, const TargetRectangle& target_rect,
const TargetRectangle& scaled_efb_rect)
{
// Transition the EFB render target to a shader resource.
Texture2D* efb_color_texture =
g_ActiveConfig.iMultisamples > 1 ?
FramebufferManager::GetInstance()->GetResolvedEFBColorTexture() :
FramebufferManager::GetInstance()->GetEFBColorTexture();
// Copy EFB -> backbuffer
BlitScreen(render_pass, target_rect, scaled_efb_rect, efb_color_texture);
}
void Renderer::DrawVirtualXFB(VkRenderPass render_pass, const TargetRectangle& target_rect,
u32 xfb_addr, const XFBSourceBase* const* xfb_sources, u32 xfb_count,
u32 fb_width, u32 fb_stride, u32 fb_height)
{
for (u32 i = 0; i < xfb_count; ++i)
{
const XFBSource* xfb_source = static_cast<const XFBSource*>(xfb_sources[i]);
TargetRectangle source_rect = xfb_source->sourceRc;
TargetRectangle draw_rect;
int xfb_width = static_cast<int>(xfb_source->srcWidth);
int xfb_height = static_cast<int>(xfb_source->srcHeight);
int h_offset = (static_cast<s32>(xfb_source->srcAddr) - static_cast<s32>(xfb_addr)) /
(static_cast<s32>(fb_stride) * 2);
draw_rect.top =
target_rect.top + h_offset * target_rect.GetHeight() / static_cast<s32>(fb_height);
draw_rect.bottom =
target_rect.top +
(h_offset + xfb_height) * target_rect.GetHeight() / static_cast<s32>(fb_height);
draw_rect.left = target_rect.left +
(target_rect.GetWidth() -
xfb_width * target_rect.GetWidth() / static_cast<s32>(fb_stride)) /
2;
draw_rect.right = target_rect.left +
(target_rect.GetWidth() +
xfb_width * target_rect.GetWidth() / static_cast<s32>(fb_stride)) /
2;
source_rect.right -= Renderer::EFBToScaledX(fb_stride - fb_width);
BlitScreen(render_pass, draw_rect, source_rect,
xfb_source->GetTexture()->GetRawTexIdentifier());
}
}
void Renderer::DrawRealXFB(VkRenderPass render_pass, const TargetRectangle& target_rect,
const XFBSourceBase* const* xfb_sources, u32 xfb_count, u32 fb_width,
u32 fb_stride, u32 fb_height)
{
for (u32 i = 0; i < xfb_count; ++i)
{
const XFBSource* xfb_source = static_cast<const XFBSource*>(xfb_sources[i]);
TargetRectangle source_rect = xfb_source->sourceRc;
TargetRectangle draw_rect = target_rect;
source_rect.right -= fb_stride - fb_width;
BlitScreen(render_pass, draw_rect, source_rect,
xfb_source->GetTexture()->GetRawTexIdentifier());
}
}
void Renderer::DrawScreen(const TargetRectangle& scaled_efb_rect, u32 xfb_addr,
const XFBSourceBase* const* xfb_sources, u32 xfb_count, u32 fb_width,
u32 fb_stride, u32 fb_height)
2016-08-13 12:57:50 +00:00
{
// Grab the next image from the swap chain in preparation for drawing the window.
VkResult res = m_swap_chain->AcquireNextImage(m_image_available_semaphore);
if (res == VK_SUBOPTIMAL_KHR || res == VK_ERROR_OUT_OF_DATE_KHR)
{
// There's an issue here. We can't resize the swap chain while the GPU is still busy with it,
// but calling WaitForGPUIdle would create a deadlock as PrepareToSubmitCommandBuffer has been
// called by SwapImpl. WaitForGPUIdle waits on the semaphore, which PrepareToSubmitCommandBuffer
// has already done, so it blocks indefinitely. To work around this, we submit the current
// command buffer, resize the swap chain (which calls WaitForGPUIdle), and then finally call
// PrepareToSubmitCommandBuffer to return to the state that the caller expects.
g_command_buffer_mgr->SubmitCommandBuffer(false);
2016-08-13 12:57:50 +00:00
ResizeSwapChain();
g_command_buffer_mgr->PrepareToSubmitCommandBuffer();
2016-08-13 12:57:50 +00:00
res = m_swap_chain->AcquireNextImage(m_image_available_semaphore);
}
if (res != VK_SUCCESS)
PanicAlert("Failed to grab image from swap chain");
// Transition from undefined (or present src, but it can be substituted) to
// color attachment ready for writing. These transitions must occur outside
// a render pass, unless the render pass declares a self-dependency.
Texture2D* backbuffer = m_swap_chain->GetCurrentTexture();
backbuffer->OverrideImageLayout(VK_IMAGE_LAYOUT_UNDEFINED);
backbuffer->TransitionToLayout(g_command_buffer_mgr->GetCurrentCommandBuffer(),
VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL);
// Begin render pass for rendering to the swap chain.
2016-08-13 12:57:50 +00:00
VkClearValue clear_value = {{{0.0f, 0.0f, 0.0f, 1.0f}}};
VkRenderPassBeginInfo info = {VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
nullptr,
m_swap_chain->GetRenderPass(),
m_swap_chain->GetCurrentFramebuffer(),
{{0, 0}, {backbuffer->GetWidth(), backbuffer->GetHeight()}},
1,
&clear_value};
vkCmdBeginRenderPass(g_command_buffer_mgr->GetCurrentCommandBuffer(), &info,
VK_SUBPASS_CONTENTS_INLINE);
2016-08-13 12:57:50 +00:00
// Draw guest buffers (EFB or XFB)
DrawFrame(m_swap_chain->GetRenderPass(), GetTargetRectangle(), scaled_efb_rect, xfb_addr,
xfb_sources, xfb_count, fb_width, fb_stride, fb_height);
2016-08-13 12:57:50 +00:00
// Draw OSD
2016-08-13 12:57:50 +00:00
Util::SetViewportAndScissor(g_command_buffer_mgr->GetCurrentCommandBuffer(), 0, 0,
backbuffer->GetWidth(), backbuffer->GetHeight());
DrawDebugText();
OSD::DoCallbacks(OSD::CallbackType::OnFrame);
OSD::DrawMessages();
// End drawing to backbuffer
vkCmdEndRenderPass(g_command_buffer_mgr->GetCurrentCommandBuffer());
2016-08-13 12:57:50 +00:00
// Transition the backbuffer to PRESENT_SRC to ensure all commands drawing
// to it have finished before present.
backbuffer->TransitionToLayout(g_command_buffer_mgr->GetCurrentCommandBuffer(),
VK_IMAGE_LAYOUT_PRESENT_SRC_KHR);
}
bool Renderer::DrawFrameDump(const TargetRectangle& scaled_efb_rect, u32 xfb_addr,
const XFBSourceBase* const* xfb_sources, u32 xfb_count, u32 fb_width,
u32 fb_stride, u32 fb_height, u64 ticks)
{
TargetRectangle target_rect = CalculateFrameDumpDrawRectangle();
u32 width = std::max(1u, static_cast<u32>(target_rect.GetWidth()));
u32 height = std::max(1u, static_cast<u32>(target_rect.GetHeight()));
if (!ResizeFrameDumpBuffer(width, height))
return false;
// If there was a previous frame dumped, we'll still be in TRANSFER_SRC layout.
m_frame_dump_render_texture->TransitionToLayout(g_command_buffer_mgr->GetCurrentCommandBuffer(),
VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL);
VkClearValue clear_value = {{{0.0f, 0.0f, 0.0f, 1.0f}}};
VkClearRect clear_rect = {{{0, 0}, {width, height}}, 0, 1};
VkClearAttachment clear_attachment = {VK_IMAGE_ASPECT_COLOR_BIT, 0, clear_value};
VkRenderPassBeginInfo info = {
VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
nullptr,
FramebufferManager::GetInstance()->GetColorCopyForReadbackRenderPass(),
m_frame_dump_framebuffer,
{{0, 0}, {width, height}},
1,
&clear_value};
vkCmdBeginRenderPass(g_command_buffer_mgr->GetCurrentCommandBuffer(), &info,
VK_SUBPASS_CONTENTS_INLINE);
vkCmdClearAttachments(g_command_buffer_mgr->GetCurrentCommandBuffer(), 1, &clear_attachment, 1,
&clear_rect);
DrawFrame(FramebufferManager::GetInstance()->GetColorCopyForReadbackRenderPass(), target_rect,
scaled_efb_rect, xfb_addr, xfb_sources, xfb_count, fb_width, fb_stride, fb_height);
vkCmdEndRenderPass(g_command_buffer_mgr->GetCurrentCommandBuffer());
// Prepare the readback texture for copying.
StagingTexture2D* readback_texture = PrepareFrameDumpImage(width, height, ticks);
if (!readback_texture)
return false;
// Queue a copy to the current frame dump buffer. It will be written to the frame dump later.
m_frame_dump_render_texture->TransitionToLayout(g_command_buffer_mgr->GetCurrentCommandBuffer(),
VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL);
readback_texture->CopyFromImage(g_command_buffer_mgr->GetCurrentCommandBuffer(),
m_frame_dump_render_texture->GetImage(),
VK_IMAGE_ASPECT_COLOR_BIT, 0, 0, width, height, 0, 0);
return true;
}
void Renderer::StartFrameDumping()
{
_assert_(!m_frame_dumping_active);
// Register fence callback so that we know when frames are ready to be written to the dump.
// This is done by clearing the fence pointer, so WriteFrameDumpFrame doesn't have to wait.
auto queued_callback = [](VkCommandBuffer, VkFence) {};
auto signaled_callback = std::bind(&Renderer::OnFrameDumpImageReady, this, std::placeholders::_1);
// We use the array pointer as a key here, that way if Renderer needed fence callbacks in
// the future it could be used without conflicting.
// We're not interested in when fences are submitted, so the first callback is a no-op.
g_command_buffer_mgr->AddFencePointCallback(
m_frame_dump_images.data(), std::move(queued_callback), std::move(signaled_callback));
m_frame_dumping_active = true;
}
void Renderer::EndFrameDumping()
{
_assert_(m_frame_dumping_active);
// Write any pending frames to the frame dump.
FlushFrameDump();
// Remove the fence callback that we registered earlier, one less function that needs to be
// called when preparing a command buffer.
g_command_buffer_mgr->RemoveFencePointCallback(m_frame_dump_images.data());
m_frame_dumping_active = false;
}
void Renderer::OnFrameDumpImageReady(VkFence fence)
{
for (FrameDumpImage& frame : m_frame_dump_images)
{
// fence being a null handle means that we don't have to wait to re-use this image.
if (frame.fence == fence)
frame.fence = VK_NULL_HANDLE;
}
}
void Renderer::WriteFrameDumpImage(size_t index)
{
FrameDumpImage& frame = m_frame_dump_images[index];
_assert_(frame.pending);
// Check fence has been signaled.
// The callback here should set fence to null.
if (frame.fence != VK_NULL_HANDLE)
{
g_command_buffer_mgr->WaitForFence(frame.fence);
_assert_(frame.fence == VK_NULL_HANDLE);
}
// Copy the now-populated image data to the output file.
DumpFrameData(reinterpret_cast<const u8*>(frame.readback_texture->GetMapPointer()),
static_cast<int>(frame.readback_texture->GetWidth()),
static_cast<int>(frame.readback_texture->GetHeight()),
static_cast<int>(frame.readback_texture->GetRowStride()), frame.dump_state);
frame.pending = false;
}
StagingTexture2D* Renderer::PrepareFrameDumpImage(u32 width, u32 height, u64 ticks)
{
// Ensure the last frame that was sent to the frame dump has completed encoding before we send
// the next image to it.
FinishFrameData();
// If the last image hasn't been written to the frame dump yet, write it now.
// This is necessary so that the worker thread is no more than one frame behind, and the pointer
// (which is actually the buffer) is safe for us to re-use next time.
if (m_frame_dump_images[m_current_frame_dump_image].pending)
WriteFrameDumpImage(m_current_frame_dump_image);
// Move to the next image buffer
m_current_frame_dump_image = (m_current_frame_dump_image + 1) % FRAME_DUMP_BUFFERED_FRAMES;
FrameDumpImage& image = m_frame_dump_images[m_current_frame_dump_image];
// Ensure the dimensions of the readback texture are sufficient.
if (!image.readback_texture || width != image.readback_texture->GetWidth() ||
height != image.readback_texture->GetHeight())
{
// Allocate a new readback texture.
// The reset() call is here so that the memory is released before allocating the new texture.
image.readback_texture.reset();
image.readback_texture = StagingTexture2D::Create(STAGING_BUFFER_TYPE_READBACK, width, height,
EFB_COLOR_TEXTURE_FORMAT);
if (!image.readback_texture || !image.readback_texture->Map())
{
// Not actually fatal, just means we can't dump this frame.
PanicAlert("Failed to allocate frame dump readback texture.");
image.readback_texture.reset();
return nullptr;
}
}
// The copy happens immediately after this function returns, so flag this frame as pending.
image.fence = g_command_buffer_mgr->GetCurrentCommandBufferFence();
image.dump_state = AVIDump::FetchState(ticks);
image.pending = true;
return image.readback_texture.get();
}
void Renderer::FlushFrameDump()
{
// We must write frames in order, so this is why we use a counter rather than a range.
for (size_t i = 0; i < FRAME_DUMP_BUFFERED_FRAMES; i++)
{
if (m_frame_dump_images[m_current_frame_dump_image].pending)
WriteFrameDumpImage(m_current_frame_dump_image);
m_current_frame_dump_image = (m_current_frame_dump_image + 1) % FRAME_DUMP_BUFFERED_FRAMES;
}
// Since everything has been written now, may as well start at index zero.
// count-1 here because the index is incremented before usage.
m_current_frame_dump_image = FRAME_DUMP_BUFFERED_FRAMES - 1;
}
2016-08-13 12:57:50 +00:00
void Renderer::BlitScreen(VkRenderPass render_pass, const TargetRectangle& dst_rect,
const TargetRectangle& src_rect, const Texture2D* src_tex)
2016-08-13 12:57:50 +00:00
{
VulkanPostProcessing* post_processor = static_cast<VulkanPostProcessing*>(m_post_processor.get());
2016-08-13 12:57:50 +00:00
if (g_ActiveConfig.iStereoMode == STEREO_SBS || g_ActiveConfig.iStereoMode == STEREO_TAB)
{
TargetRectangle left_rect;
TargetRectangle right_rect;
std::tie(left_rect, right_rect) = ConvertStereoRectangle(dst_rect);
2016-08-13 12:57:50 +00:00
post_processor->BlitFromTexture(left_rect, src_rect, src_tex, 0, render_pass);
post_processor->BlitFromTexture(right_rect, src_rect, src_tex, 1, render_pass);
2016-08-13 12:57:50 +00:00
}
else if (g_ActiveConfig.iStereoMode == STEREO_QUADBUFFER)
{
post_processor->BlitFromTexture(dst_rect, src_rect, src_tex, -1, render_pass);
}
2016-08-13 12:57:50 +00:00
else
{
post_processor->BlitFromTexture(dst_rect, src_rect, src_tex, 0, render_pass);
2016-08-13 12:57:50 +00:00
}
}
bool Renderer::ResizeFrameDumpBuffer(u32 new_width, u32 new_height)
{
if (m_frame_dump_render_texture && m_frame_dump_render_texture->GetWidth() == new_width &&
m_frame_dump_render_texture->GetHeight() == new_height)
{
return true;
}
// Ensure all previous frames have been dumped, since we are destroying a framebuffer
// that may still be in use.
FlushFrameDump();
if (m_frame_dump_framebuffer != VK_NULL_HANDLE)
{
vkDestroyFramebuffer(g_vulkan_context->GetDevice(), m_frame_dump_framebuffer, nullptr);
m_frame_dump_framebuffer = VK_NULL_HANDLE;
}
m_frame_dump_render_texture =
Texture2D::Create(new_width, new_height, 1, 1, EFB_COLOR_TEXTURE_FORMAT,
VK_SAMPLE_COUNT_1_BIT, VK_IMAGE_VIEW_TYPE_2D, VK_IMAGE_TILING_OPTIMAL,
VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT);
if (!m_frame_dump_render_texture)
{
WARN_LOG(VIDEO, "Failed to resize frame dump render texture");
m_frame_dump_render_texture.reset();
return false;
}
VkImageView attachment = m_frame_dump_render_texture->GetView();
VkFramebufferCreateInfo info = {};
info.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO;
info.renderPass = FramebufferManager::GetInstance()->GetColorCopyForReadbackRenderPass();
info.attachmentCount = 1;
info.pAttachments = &attachment;
info.width = new_width;
info.height = new_height;
info.layers = 1;
VkResult res =
vkCreateFramebuffer(g_vulkan_context->GetDevice(), &info, nullptr, &m_frame_dump_framebuffer);
if (res != VK_SUCCESS)
{
WARN_LOG(VIDEO, "Failed to create frame dump framebuffer");
m_frame_dump_render_texture.reset();
return false;
}
// Render pass expects texture is in transfer src to start with.
m_frame_dump_render_texture->TransitionToLayout(g_command_buffer_mgr->GetCurrentCommandBuffer(),
VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL);
return true;
}
void Renderer::DestroyFrameDumpResources()
{
if (m_frame_dump_framebuffer != VK_NULL_HANDLE)
{
vkDestroyFramebuffer(g_vulkan_context->GetDevice(), m_frame_dump_framebuffer, nullptr);
m_frame_dump_framebuffer = VK_NULL_HANDLE;
}
m_frame_dump_render_texture.reset();
for (FrameDumpImage& image : m_frame_dump_images)
{
image.readback_texture.reset();
image.fence = VK_NULL_HANDLE;
image.dump_state = {};
image.pending = false;
}
m_current_frame_dump_image = FRAME_DUMP_BUFFERED_FRAMES - 1;
}
2016-08-13 12:57:50 +00:00
void Renderer::CheckForTargetResize(u32 fb_width, u32 fb_stride, u32 fb_height)
{
if (FramebufferManagerBase::LastXfbWidth() == fb_stride &&
FramebufferManagerBase::LastXfbHeight() == fb_height)
{
return;
}
u32 new_width = (fb_stride < 1 || fb_stride > MAX_XFB_WIDTH) ? MAX_XFB_WIDTH : fb_stride;
u32 new_height = (fb_height < 1 || fb_height > MAX_XFB_HEIGHT) ? MAX_XFB_HEIGHT : fb_height;
FramebufferManagerBase::SetLastXfbWidth(new_width);
FramebufferManagerBase::SetLastXfbHeight(new_height);
// Changing the XFB source area may alter the target size.
if (CalculateTargetSize())
ResizeEFBTextures();
2016-08-13 12:57:50 +00:00
}
void Renderer::CheckForSurfaceChange()
{
if (!m_surface_needs_change.IsSet())
2016-08-13 12:57:50 +00:00
return;
u32 old_width = m_swap_chain ? m_swap_chain->GetWidth() : 0;
u32 old_height = m_swap_chain ? m_swap_chain->GetHeight() : 0;
// Fast path, if the surface handle is the same, the window has just been resized.
if (m_swap_chain && m_new_surface_handle == m_swap_chain->GetNativeHandle())
2016-08-13 12:57:50 +00:00
{
INFO_LOG(VIDEO, "Detected window resize.");
ResizeSwapChain();
// Notify the main thread we are done.
m_surface_needs_change.Clear();
m_new_surface_handle = nullptr;
m_surface_changed.Set();
2016-08-13 12:57:50 +00:00
}
else
{
// Wait for the GPU to catch up since we're going to destroy the swap chain.
g_command_buffer_mgr->WaitForGPUIdle();
2016-08-13 12:57:50 +00:00
// Did we previously have a swap chain?
if (m_swap_chain)
{
if (!m_new_surface_handle)
2016-08-13 12:57:50 +00:00
{
// If there is no surface now, destroy the swap chain.
m_swap_chain.reset();
}
else
{
// Recreate the surface. If this fails we're in trouble.
if (!m_swap_chain->RecreateSurface(m_new_surface_handle))
2016-08-13 12:57:50 +00:00
PanicAlert("Failed to recreate Vulkan surface. Cannot continue.");
}
}
else
{
// Previously had no swap chain. So create one.
VkSurfaceKHR surface = SwapChain::CreateVulkanSurface(g_vulkan_context->GetVulkanInstance(),
m_new_surface_handle);
2016-08-13 12:57:50 +00:00
if (surface != VK_NULL_HANDLE)
{
m_swap_chain = SwapChain::Create(m_new_surface_handle, surface, g_ActiveConfig.IsVSync());
2016-08-13 12:57:50 +00:00
if (!m_swap_chain)
PanicAlert("Failed to create swap chain.");
}
else
{
PanicAlert("Failed to create surface.");
}
}
// Notify calling thread.
m_surface_needs_change.Clear();
m_new_surface_handle = nullptr;
m_surface_changed.Set();
2016-08-13 12:57:50 +00:00
}
if (m_swap_chain)
{
// Handle case where the dimensions are now different
if (old_width != m_swap_chain->GetWidth() || old_height != m_swap_chain->GetHeight())
OnSwapChainResized();
}
}
void Renderer::CheckForConfigChanges()
{
// Save the video config so we can compare against to determine which settings have changed.
int old_anisotropy = g_ActiveConfig.iMaxAnisotropy;
int old_aspect_ratio = g_ActiveConfig.iAspectRatio;
int old_efb_scale = g_ActiveConfig.iEFBScale;
bool old_force_filtering = g_ActiveConfig.bForceFiltering;
bool old_use_xfb = g_ActiveConfig.bUseXFB;
bool old_use_realxfb = g_ActiveConfig.bUseRealXFB;
2016-08-13 12:57:50 +00:00
// Copy g_Config to g_ActiveConfig.
// NOTE: This can potentially race with the UI thread, however if it does, the changes will be
// delayed until the next time CheckForConfigChanges is called.
2016-08-13 12:57:50 +00:00
UpdateActiveConfig();
// Determine which (if any) settings have changed.
bool anisotropy_changed = old_anisotropy != g_ActiveConfig.iMaxAnisotropy;
bool force_texture_filtering_changed = old_force_filtering != g_ActiveConfig.bForceFiltering;
bool efb_scale_changed = old_efb_scale != g_ActiveConfig.iEFBScale;
bool aspect_changed = old_aspect_ratio != g_ActiveConfig.iAspectRatio;
bool use_xfb_changed = old_use_xfb != g_ActiveConfig.bUseXFB;
bool use_realxfb_changed = old_use_realxfb != g_ActiveConfig.bUseRealXFB;
// Update texture cache settings with any changed options.
TextureCache::GetInstance()->OnConfigChanged(g_ActiveConfig);
// Handle settings that can cause the target rectangle to change.
if (efb_scale_changed || aspect_changed || use_xfb_changed || use_realxfb_changed)
2016-08-13 12:57:50 +00:00
{
if (CalculateTargetSize())
2016-08-13 12:57:50 +00:00
ResizeEFBTextures();
}
// MSAA samples changed, we need to recreate the EFB render pass.
// If the stereoscopy mode changed, we need to recreate the buffers as well.
// SSAA changed on/off, we have to recompile shaders.
// Changing stereoscopy from off<->on also requires shaders to be recompiled.
if (CheckForHostConfigChanges())
2016-08-13 12:57:50 +00:00
{
g_command_buffer_mgr->WaitForGPUIdle();
FramebufferManager::GetInstance()->RecreateRenderPass();
FramebufferManager::GetInstance()->ResizeEFBTextures();
2016-08-13 12:57:50 +00:00
BindEFBToStateTracker();
RecompileShaders();
FramebufferManager::GetInstance()->RecompileShaders();
g_shader_cache->ReloadShaderAndPipelineCaches();
g_shader_cache->RecompileSharedShaders();
StateTracker::GetInstance()->InvalidateShaderPointers();
StateTracker::GetInstance()->ReloadPipelineUIDCache();
2016-08-13 12:57:50 +00:00
}
// For vsync, we need to change the present mode, which means recreating the swap chain.
if (m_swap_chain && g_ActiveConfig.IsVSync() != m_swap_chain->IsVSyncEnabled())
{
g_command_buffer_mgr->WaitForGPUIdle();
m_swap_chain->SetVSync(g_ActiveConfig.IsVSync());
}
2016-08-13 12:57:50 +00:00
// For quad-buffered stereo we need to change the layer count, so recreate the swap chain.
if (m_swap_chain &&
(g_ActiveConfig.iStereoMode == STEREO_QUADBUFFER) != m_swap_chain->IsStereoEnabled())
ResizeSwapChain();
2016-08-13 12:57:50 +00:00
// Wipe sampler cache if force texture filtering or anisotropy changes.
if (anisotropy_changed || force_texture_filtering_changed)
ResetSamplerStates();
// Check for a changed post-processing shader and recompile if needed.
static_cast<VulkanPostProcessing*>(m_post_processor.get())->UpdateConfig();
2016-08-13 12:57:50 +00:00
}
void Renderer::OnSwapChainResized()
{
m_backbuffer_width = m_swap_chain->GetWidth();
m_backbuffer_height = m_swap_chain->GetHeight();
UpdateDrawRectangle();
if (CalculateTargetSize())
2016-08-13 12:57:50 +00:00
ResizeEFBTextures();
}
void Renderer::BindEFBToStateTracker()
{
// Update framebuffer in state tracker
VkRect2D framebuffer_size = {{0, 0},
{FramebufferManager::GetInstance()->GetEFBWidth(),
FramebufferManager::GetInstance()->GetEFBHeight()}};
StateTracker::GetInstance()->SetRenderPass(
FramebufferManager::GetInstance()->GetEFBLoadRenderPass(),
FramebufferManager::GetInstance()->GetEFBClearRenderPass());
StateTracker::GetInstance()->SetFramebuffer(
FramebufferManager::GetInstance()->GetEFBFramebuffer(), framebuffer_size);
2016-08-13 12:57:50 +00:00
// Update rasterization state with MSAA info
RasterizationState rs_state = {};
rs_state.bits = StateTracker::GetInstance()->GetRasterizationState().bits;
rs_state.samples = FramebufferManager::GetInstance()->GetEFBSamples();
2016-08-13 12:57:50 +00:00
rs_state.per_sample_shading = g_ActiveConfig.bSSAA ? VK_TRUE : VK_FALSE;
StateTracker::GetInstance()->SetRasterizationState(rs_state);
2016-08-13 12:57:50 +00:00
}
void Renderer::ResizeEFBTextures()
{
// Ensure the GPU is finished with the current EFB textures.
g_command_buffer_mgr->WaitForGPUIdle();
FramebufferManager::GetInstance()->ResizeEFBTextures();
2016-08-13 12:57:50 +00:00
BindEFBToStateTracker();
// Viewport and scissor rect have to be reset since they will be scaled differently.
SetViewport();
BPFunctions::SetScissor();
}
void Renderer::ResizeSwapChain()
{
// The worker thread may still be submitting a present on this swap chain.
g_command_buffer_mgr->WaitForGPUIdle();
// It's now safe to resize the swap chain.
if (!m_swap_chain->ResizeSwapChain())
PanicAlert("Failed to resize swap chain.");
OnSwapChainResized();
}
void Renderer::ApplyState()
2016-08-13 12:57:50 +00:00
{
}
void Renderer::ResetAPIState()
{
// End the EFB render pass if active
StateTracker::GetInstance()->EndRenderPass();
2016-08-13 12:57:50 +00:00
}
void Renderer::RestoreAPIState()
{
// Instruct the state tracker to re-bind everything before the next draw
StateTracker::GetInstance()->SetPendingRebind();
2016-08-13 12:57:50 +00:00
}
void Renderer::SetGenerationMode()
{
RasterizationState new_rs_state = {};
new_rs_state.bits = StateTracker::GetInstance()->GetRasterizationState().bits;
2016-08-13 12:57:50 +00:00
switch (bpmem.genMode.cullmode)
{
case GenMode::CULL_NONE:
new_rs_state.cull_mode = VK_CULL_MODE_NONE;
break;
case GenMode::CULL_BACK:
new_rs_state.cull_mode = VK_CULL_MODE_BACK_BIT;
break;
case GenMode::CULL_FRONT:
new_rs_state.cull_mode = VK_CULL_MODE_FRONT_BIT;
break;
case GenMode::CULL_ALL:
new_rs_state.cull_mode = VK_CULL_MODE_FRONT_AND_BACK;
break;
default:
new_rs_state.cull_mode = VK_CULL_MODE_NONE;
break;
}
StateTracker::GetInstance()->SetRasterizationState(new_rs_state);
2016-08-13 12:57:50 +00:00
}
void Renderer::SetDepthMode()
{
DepthStencilState new_ds_state = {};
new_ds_state.test_enable = bpmem.zmode.testenable ? VK_TRUE : VK_FALSE;
new_ds_state.write_enable = bpmem.zmode.updateenable ? VK_TRUE : VK_FALSE;
// Inverted depth, hence these are swapped
switch (bpmem.zmode.func)
{
case ZMode::NEVER:
new_ds_state.compare_op = VK_COMPARE_OP_NEVER;
break;
case ZMode::LESS:
new_ds_state.compare_op = VK_COMPARE_OP_GREATER;
break;
case ZMode::EQUAL:
new_ds_state.compare_op = VK_COMPARE_OP_EQUAL;
break;
case ZMode::LEQUAL:
new_ds_state.compare_op = VK_COMPARE_OP_GREATER_OR_EQUAL;
break;
case ZMode::GREATER:
new_ds_state.compare_op = VK_COMPARE_OP_LESS;
break;
case ZMode::NEQUAL:
new_ds_state.compare_op = VK_COMPARE_OP_NOT_EQUAL;
break;
case ZMode::GEQUAL:
new_ds_state.compare_op = VK_COMPARE_OP_LESS_OR_EQUAL;
break;
case ZMode::ALWAYS:
new_ds_state.compare_op = VK_COMPARE_OP_ALWAYS;
break;
default:
new_ds_state.compare_op = VK_COMPARE_OP_ALWAYS;
break;
}
StateTracker::GetInstance()->SetDepthStencilState(new_ds_state);
2016-08-13 12:57:50 +00:00
}
void Renderer::SetBlendMode(bool force_update)
{
BlendingState state;
state.Generate(bpmem);
2016-08-13 12:57:50 +00:00
StateTracker::GetInstance()->SetBlendState(state);
2016-08-13 12:57:50 +00:00
}
void Renderer::SetSamplerState(int stage, int texindex, bool custom_tex)
{
const FourTexUnits& tex = bpmem.tex[texindex];
const TexMode0& tm0 = tex.texMode0[stage];
const TexMode1& tm1 = tex.texMode1[stage];
SamplerState new_state = {};
if (g_ActiveConfig.bForceFiltering)
{
new_state.min_filter = VK_FILTER_LINEAR;
new_state.mag_filter = VK_FILTER_LINEAR;
new_state.mipmap_mode = SamplerCommon::AreBpTexMode0MipmapsEnabled(tm0) ?
VK_SAMPLER_MIPMAP_MODE_LINEAR :
VK_SAMPLER_MIPMAP_MODE_NEAREST;
}
else
{
// Constants for these?
new_state.min_filter = (tm0.min_filter & 4) != 0 ? VK_FILTER_LINEAR : VK_FILTER_NEAREST;
new_state.mipmap_mode = SamplerCommon::AreBpTexMode0MipmapsEnabled(tm0) ?
VK_SAMPLER_MIPMAP_MODE_LINEAR :
VK_SAMPLER_MIPMAP_MODE_NEAREST;
new_state.mag_filter = tm0.mag_filter != 0 ? VK_FILTER_LINEAR : VK_FILTER_NEAREST;
}
// If mipmaps are disabled, clamp min/max lod
new_state.max_lod = SamplerCommon::AreBpTexMode0MipmapsEnabled(tm0) ? tm1.max_lod : 0;
new_state.min_lod = std::min(new_state.max_lod.Value(), tm1.min_lod);
new_state.lod_bias = SamplerCommon::AreBpTexMode0MipmapsEnabled(tm0) ? tm0.lod_bias : 0;
2016-08-13 12:57:50 +00:00
// Custom textures may have a greater number of mips
if (custom_tex)
new_state.max_lod = 255;
// Address modes
static const VkSamplerAddressMode address_modes[] = {
VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE, VK_SAMPLER_ADDRESS_MODE_REPEAT,
VK_SAMPLER_ADDRESS_MODE_MIRRORED_REPEAT, VK_SAMPLER_ADDRESS_MODE_REPEAT};
new_state.wrap_u = address_modes[tm0.wrap_s];
new_state.wrap_v = address_modes[tm0.wrap_t];
// Only use anisotropic filtering for textures that would be linearly filtered.
new_state.enable_anisotropic_filtering = SamplerCommon::IsBpTexMode0PointFiltering(tm0) ? 0 : 1;
2016-08-13 12:57:50 +00:00
// Skip lookup if the state hasn't changed.
size_t bind_index = (texindex * 4) + stage;
if (m_sampler_states[bind_index].bits == new_state.bits)
return;
// Look up new state and replace in state tracker.
VkSampler sampler = g_object_cache->GetSampler(new_state);
if (sampler == VK_NULL_HANDLE)
{
ERROR_LOG(VIDEO, "Failed to create sampler");
sampler = g_object_cache->GetPointSampler();
}
StateTracker::GetInstance()->SetSampler(bind_index, sampler);
2016-08-13 12:57:50 +00:00
m_sampler_states[bind_index].bits = new_state.bits;
}
void Renderer::ResetSamplerStates()
{
// Ensure none of the sampler objects are in use.
// This assumes that none of the samplers are in use on the command list currently being recorded.
2016-08-13 12:57:50 +00:00
g_command_buffer_mgr->WaitForGPUIdle();
// Invalidate all sampler states, next draw will re-initialize them.
for (size_t i = 0; i < m_sampler_states.size(); i++)
{
m_sampler_states[i].bits = std::numeric_limits<decltype(m_sampler_states[i].bits)>::max();
StateTracker::GetInstance()->SetSampler(i, g_object_cache->GetPointSampler());
2016-08-13 12:57:50 +00:00
}
// Invalidate all sampler objects (some will be unused now).
g_object_cache->ClearSamplerCache();
}
void Renderer::SetInterlacingMode()
{
}
void Renderer::SetScissorRect(const EFBRectangle& rc)
{
TargetRectangle target_rc = ConvertEFBRectangle(rc);
VkRect2D scissor = {
{target_rc.left, target_rc.top},
{static_cast<uint32_t>(target_rc.GetWidth()), static_cast<uint32_t>(target_rc.GetHeight())}};
StateTracker::GetInstance()->SetScissor(scissor);
2016-08-13 12:57:50 +00:00
}
void Renderer::SetViewport()
{
int scissor_x_offset = bpmem.scissorOffset.x * 2;
int scissor_y_offset = bpmem.scissorOffset.y * 2;
float x = Renderer::EFBToScaledXf(xfmem.viewport.xOrig - xfmem.viewport.wd - scissor_x_offset);
float y = Renderer::EFBToScaledYf(xfmem.viewport.yOrig + xfmem.viewport.ht - scissor_y_offset);
float width = Renderer::EFBToScaledXf(2.0f * xfmem.viewport.wd);
float height = Renderer::EFBToScaledYf(-2.0f * xfmem.viewport.ht);
float min_depth = (xfmem.viewport.farZ - xfmem.viewport.zRange) / 16777216.0f;
float max_depth = xfmem.viewport.farZ / 16777216.0f;
2016-08-13 12:57:50 +00:00
if (width < 0.0f)
{
x += width;
width = -width;
}
if (height < 0.0f)
{
y += height;
height = -height;
}
// If an oversized or inverted depth range is used, we need to calculate the depth range in the
// vertex shader.
// TODO: Inverted depth ranges are bugged in all drivers, which should be added to DriverDetails.
if (UseVertexDepthRange())
{
// We need to ensure depth values are clamped the maximum value supported by the console GPU.
min_depth = 0.0f;
max_depth = GX_MAX_DEPTH;
}
2016-08-13 12:57:50 +00:00
// We use an inverted depth range here to apply the Reverse Z trick.
// This trick makes sure we match the precision provided by the 1:0
// clipping depth range on the hardware.
VkViewport viewport = {x, y, width, height, 1.0f - max_depth, 1.0f - min_depth};
StateTracker::GetInstance()->SetViewport(viewport);
2016-08-13 12:57:50 +00:00
}
void Renderer::ChangeSurface(void* new_surface_handle)
{
// Called by the main thread when the window is resized.
m_new_surface_handle = new_surface_handle;
m_surface_needs_change.Set();
m_surface_changed.Set();
2016-08-13 12:57:50 +00:00
}
void Renderer::RecompileShaders()
{
DestroyShaders();
if (!CompileShaders())
PanicAlert("Failed to recompile shaders.");
}
bool Renderer::CompileShaders()
{
static const char CLEAR_FRAGMENT_SHADER_SOURCE[] = R"(
layout(location = 0) in float3 uv0;
layout(location = 1) in float4 col0;
layout(location = 0) out float4 ocol0;
void main()
{
ocol0 = col0;
}
)";
std::string source = g_shader_cache->GetUtilityShaderHeader() + CLEAR_FRAGMENT_SHADER_SOURCE;
2016-08-13 12:57:50 +00:00
m_clear_fragment_shader = Util::CompileAndCreateFragmentShader(source);
return m_clear_fragment_shader != VK_NULL_HANDLE;
2016-08-13 12:57:50 +00:00
}
void Renderer::DestroyShaders()
{
auto DestroyShader = [this](VkShaderModule& shader) {
if (shader != VK_NULL_HANDLE)
{
vkDestroyShaderModule(g_vulkan_context->GetDevice(), shader, nullptr);
shader = VK_NULL_HANDLE;
}
};
DestroyShader(m_clear_fragment_shader);
}
} // namespace Vulkan