GitHub/skyline
2
0
Fork 0
mirror of https://github.com/skyline-emu/skyline.git synced 2024-12-29 13:35:30 +03:00
Code Issues Packages Projects Releases Wiki Activity

Drop exclusiveSubpass in favour of only ending RPs when attachments change

Browse Source 
Significantly helps Mali performance by avoiding creating an excessive number of RPs.
This commit is contained in:
Billy Laws 2022-09-29 20:43:24 +01:00
parent a3f38c0cf7
commit 7133c5d6b3
2 changed files with 13 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
app/src/main/cpp/skyline/gpu/interconnect/command_executor.cpp
View File

@ -152,7 +152,7 @@ namespace skyline::gpu::interconnect {
return gpu.megaBufferAllocator;
}
bool CommandExecutor::CreateRenderPassWithSubpass(vk::Rect2D renderArea, span<TextureView *> inputAttachments, span<TextureView *> colorAttachments, TextureView *depthStencilAttachment) {
bool CommandExecutor::CreateRenderPassWithSubpass(vk::Rect2D renderArea, span<TextureView *> inputAttachments, span<TextureView *> colorAttachments, TextureView *depthStencilAttachment, bool noSubpassCreation) {
auto addSubpass{[&] {
renderPass->AddSubpass(inputAttachments, colorAttachments, depthStencilAttachment, gpu);
@ -175,7 +175,12 @@ namespace skyline::gpu::interconnect {
lastSubpassDepthStencilAttachment = depthStencilAttachment;
}};
if (renderPass == nullptr || renderPass->renderArea != renderArea || subpassCount >= gpu.traits.quirks.maxSubpassCount) {
bool attachmentsMatch{ranges::equal(lastSubpassInputAttachments, inputAttachments) &&
ranges::equal(lastSubpassColorAttachments, colorAttachments) &&
lastSubpassDepthStencilAttachment == depthStencilAttachment};
if (renderPass == nullptr || renderPass->renderArea != renderArea ||
((noSubpassCreation || subpassCount >= gpu.traits.quirks.maxSubpassCount) && !attachmentsMatch)) {
// We need to create a render pass if one doesn't already exist or the current one isn't compatible
if (renderPass != nullptr)
slot->nodes.emplace_back(std::in_place_type_t<node::RenderPassEndNode>());
@ -184,9 +189,7 @@ namespace skyline::gpu::interconnect {
subpassCount = 1;
return false;
} else {
if (ranges::equal(lastSubpassInputAttachments, inputAttachments) &&
ranges::equal(lastSubpassColorAttachments, colorAttachments) &&
lastSubpassDepthStencilAttachment == depthStencilAttachment) {
if (attachmentsMatch) {
// The last subpass had the same attachments, so we can reuse them
return false;
} else {
@ -285,18 +288,12 @@ namespace skyline::gpu::interconnect {
cycle->AttachObject(dependency);
}
void CommandExecutor::AddSubpass(std::function<void(vk::raii::CommandBuffer &, const std::shared_ptr<FenceCycle> &, GPU &, vk::RenderPass, u32)> &&function, vk::Rect2D renderArea, span<TextureView *> inputAttachments, span<TextureView *> colorAttachments, TextureView *depthStencilAttachment, bool exclusiveSubpass) {
if (exclusiveSubpass)
FinishRenderPass();
bool gotoNext{CreateRenderPassWithSubpass(renderArea, inputAttachments, colorAttachments, depthStencilAttachment ? &*depthStencilAttachment : nullptr)};
void CommandExecutor::AddSubpass(std::function<void(vk::raii::CommandBuffer &, const std::shared_ptr<FenceCycle> &, GPU &, vk::RenderPass, u32)> &&function, vk::Rect2D renderArea, span<TextureView *> inputAttachments, span<TextureView *> colorAttachments, TextureView *depthStencilAttachment, bool noSubpassCreation) {
bool gotoNext{CreateRenderPassWithSubpass(renderArea, inputAttachments, colorAttachments, depthStencilAttachment ? &*depthStencilAttachment : nullptr, noSubpassCreation)};
if (gotoNext)
slot->nodes.emplace_back(std::in_place_type_t<node::NextSubpassFunctionNode>(), std::forward<decltype(function)>(function));
else
slot->nodes.emplace_back(std::in_place_type_t<node::SubpassFunctionNode>(), std::forward<decltype(function)>(function));
if (exclusiveSubpass)
FinishRenderPass();
}
void CommandExecutor::AddOutsideRpCommand(std::function<void(vk::raii::CommandBuffer &, const std::shared_ptr<FenceCycle> &, GPU &)> &&function) {

5
app/src/main/cpp/skyline/gpu/interconnect/command_executor.h
View File

@ -127,10 +127,11 @@ namespace skyline::gpu::interconnect {
/**
* @brief Create a new render pass and subpass with the specified attachments, if one doesn't already exist or the current one isn't compatible
* @param noSubpassCreation Forces creation of a renderpass when a new subpass would otherwise be created
* @note This also checks for subpass coalescing and will merge the new subpass with the previous one when possible
* @return If the next subpass must be started prior to issuing any commands
*/
bool CreateRenderPassWithSubpass(vk::Rect2D renderArea, span<TextureView *> inputAttachments, span<TextureView *> colorAttachments, TextureView *depthStencilAttachment);
bool CreateRenderPassWithSubpass(vk::Rect2D renderArea, span<TextureView *> inputAttachments, span<TextureView *> colorAttachments, TextureView *depthStencilAttachment, bool noSubpassCreation = false);
/**
* @brief Ends a render pass if one is currently active and resets all corresponding state
@ -218,7 +219,7 @@ namespace skyline::gpu::interconnect {
* @param exclusiveSubpass If this subpass should be the only subpass in a render pass
* @note Any supplied texture should be attached prior and not undergo any persistent layout transitions till execution
*/
void AddSubpass(std::function<void(vk::raii::CommandBuffer &, const std::shared_ptr<FenceCycle> &, GPU &, vk::RenderPass, u32)> &&function, vk::Rect2D renderArea, span<TextureView *> inputAttachments = {}, span<TextureView *> colorAttachments = {}, TextureView *depthStencilAttachment = {}, bool exclusiveSubpass = false);
void AddSubpass(std::function<void(vk::raii::CommandBuffer &, const std::shared_ptr<FenceCycle> &, GPU &, vk::RenderPass, u32)> &&function, vk::Rect2D renderArea, span<TextureView *> inputAttachments = {}, span<TextureView *> colorAttachments = {}, TextureView *depthStencilAttachment = {}, bool noSubpassCreation = false);
/**
* @brief Adds a subpass that clears the entirety of the specified attachment with a color value, it may utilize VK_ATTACHMENT_LOAD_OP_CLEAR for a more efficient clear when possible