The-Homebrew-Cloud/skyline
2
0
Fork 0
mirror of https://github.com/skyline-emu/skyline.git synced 2024-06-26 18:26:03 +02:00
Code Issues Packages Projects Releases Wiki Activity
3e33d49faf
skyline/app/src/main/cpp/skyline/gpu/texture/texture.cpp

610 lines
28 KiB
C++
Raw Normal View History

Redesign Texture Class + Improve Presentation Engine This commit reworks the `Texture` class to include a Vulkan Image backing that can be optionally owning or non-owning and swapped in with consideration for Vulkan image layout, it also adds CPU-sided synchronization for the texture objects with FenceCycle. It also makes the appropriate changes to `PresentationEngine` and `GraphicBufferProducer` to work with the new `Texture` class while setting the groundwork for supporting swapchain recreation. It also fixes a log in `IpcResponse` and improves the display mode selection algorithm by further weighing refresh rate.
2021-05-22 19:46:28 +02:00
// SPDX-License-Identifier: MPL-2.0
// Copyright © 2020 Skyline Team and Contributors (https://github.com/skyline-emu/)
#include <gpu.h>
Use mirror mappings for Textures and Buffers This is a prerequisite to memory trapping as we need to write to the mirror to avoid a race condition with external threads writing to a texture/buffer while we do so ourselves for the sync on a read/write, it also avoids an additional `mprotect` to `-WX`/`RWX` on a read access. An additional advantage for textures especially is that we now support split-mapping textures due to laying them out in a contiguous mirror and they will not require costly algorithmic changes. Buffers should also benefit from not needing to iterate over every region when they are split into multiple mappings.
2022-03-06 15:57:13 +01:00
#include <kernel/memory.h>
Redesign Texture Class + Improve Presentation Engine This commit reworks the `Texture` class to include a Vulkan Image backing that can be optionally owning or non-owning and swapped in with consideration for Vulkan image layout, it also adds CPU-sided synchronization for the texture objects with FenceCycle. It also makes the appropriate changes to `PresentationEngine` and `GraphicBufferProducer` to work with the new `Texture` class while setting the groundwork for supporting swapchain recreation. It also fixes a log in `IpcResponse` and improves the display mode selection algorithm by further weighing refresh rate.
2021-05-22 19:46:28 +02:00
#include <common/trace.h>
#include <kernel/types/KProcess.h>
#include "texture.h"
Calculate Blocklinear Texture Aligned Size Correctly The size of blocklinear textures did not consider alignment to Block/ROB boundaries before, it is aligned to them now. Incorrect sizes led to textures not being aliased correctly due to different size calculations for GraphicBufferProducer surfaces and Maxwell3D color RTs.
2022-01-13 03:12:56 +01:00
#include "layout.h"
Redesign Texture Class + Improve Presentation Engine This commit reworks the `Texture` class to include a Vulkan Image backing that can be optionally owning or non-owning and swapped in with consideration for Vulkan image layout, it also adds CPU-sided synchronization for the texture objects with FenceCycle. It also makes the appropriate changes to `PresentationEngine` and `GraphicBufferProducer` to work with the new `Texture` class while setting the groundwork for supporting swapchain recreation. It also fixes a log in `IpcResponse` and improves the display mode selection algorithm by further weighing refresh rate.
2021-05-22 19:46:28 +02:00
namespace skyline::gpu {
Calculate Blocklinear Texture Aligned Size Correctly The size of blocklinear textures did not consider alignment to Block/ROB boundaries before, it is aligned to them now. Incorrect sizes led to textures not being aliased correctly due to different size calculations for GraphicBufferProducer surfaces and Maxwell3D color RTs.
2022-01-13 03:12:56 +01:00
u32 GuestTexture::GetLayerSize() {
switch (tileConfig.mode) {
case texture::TileMode::Linear:
return layerStride = static_cast<u32>(format->GetSize(dimensions));
case texture::TileMode::Pitch:
return layerStride = dimensions.height * tileConfig.pitch;
case texture::TileMode::Block:
return layerStride = static_cast<u32>(texture::GetBlockLinearLayerSize(*this));
}
}
Order `TextureView` Definition Correctly The definition of the `TextureView` class was spread across `texture.cpp` and has now been moved to the top of the file above the other half of the definition.
2021-12-08 09:50:39 +01:00
TextureView::TextureView(std::shared_ptr<Texture> texture, vk::ImageViewType type, vk::ImageSubresourceRange range, texture::Format format, vk::ComponentMapping mapping) : texture(std::move(texture)), type(type), format(format), mapping(mapping), range(range) {}
vk::ImageView TextureView::GetView() {
if (view)
return **view;
auto viewType{[&]() {
switch (texture->dimensions.GetType()) {
case vk::ImageType::e1D:
return range.layerCount > 1 ? vk::ImageViewType::e1DArray : vk::ImageViewType::e1D;
case vk::ImageType::e2D:
return range.layerCount > 1 ? vk::ImageViewType::e2DArray : vk::ImageViewType::e2D;
case vk::ImageType::e3D:
return vk::ImageViewType::e3D;
}
}()};
vk::ImageViewCreateInfo createInfo{
.image = texture->GetBacking(),
.viewType = viewType,
.format = format ? *format : *texture->format,
.components = mapping,
.subresourceRange = range,
};
return *view.emplace(texture->gpu.vkDevice, createInfo);
}
Rework `TextureView` to be disconnected from `Texture` We want `TextureView`(s) to be disconnected from the backing on the host and instead represent a specific texture on the guest with a backing that can change depending on mapping of new textures which'd invalidate the backing but should now be automatically repointed to an appropriate new backing. This approach also requires locking of the backing to function as it is mutable till it has been locked or the backing has an attached `FenceCycle` that hasn't been signaled which will be added for `CommandExecutor` in a subsequent commit.
2021-12-06 22:13:14 +01:00
void TextureView::lock() {
Cleanup `BufferView`/`TextureView` Locking Code Renames the variable to be neater and less confusing alongside adding comments for `try_lock()` to make the goal of the function more apparent.
2021-12-07 22:12:20 +01:00
auto backing{std::atomic_load(&texture)};
Rework `TextureView` to be disconnected from `Texture` We want `TextureView`(s) to be disconnected from the backing on the host and instead represent a specific texture on the guest with a backing that can change depending on mapping of new textures which'd invalidate the backing but should now be automatically repointed to an appropriate new backing. This approach also requires locking of the backing to function as it is mutable till it has been locked or the backing has an attached `FenceCycle` that hasn't been signaled which will be added for `CommandExecutor` in a subsequent commit.
2021-12-06 22:13:14 +01:00
while (true) {
Cleanup `BufferView`/`TextureView` Locking Code Renames the variable to be neater and less confusing alongside adding comments for `try_lock()` to make the goal of the function more apparent.
2021-12-07 22:12:20 +01:00
backing->lock();
Rework `TextureView` to be disconnected from `Texture` We want `TextureView`(s) to be disconnected from the backing on the host and instead represent a specific texture on the guest with a backing that can change depending on mapping of new textures which'd invalidate the backing but should now be automatically repointed to an appropriate new backing. This approach also requires locking of the backing to function as it is mutable till it has been locked or the backing has an attached `FenceCycle` that hasn't been signaled which will be added for `CommandExecutor` in a subsequent commit.
2021-12-06 22:13:14 +01:00
Cleanup `BufferView`/`TextureView` Locking Code Renames the variable to be neater and less confusing alongside adding comments for `try_lock()` to make the goal of the function more apparent.
2021-12-07 22:12:20 +01:00
auto latestBacking{std::atomic_load(&texture)};
if (backing == latestBacking)
Rework `TextureView` to be disconnected from `Texture` We want `TextureView`(s) to be disconnected from the backing on the host and instead represent a specific texture on the guest with a backing that can change depending on mapping of new textures which'd invalidate the backing but should now be automatically repointed to an appropriate new backing. This approach also requires locking of the backing to function as it is mutable till it has been locked or the backing has an attached `FenceCycle` that hasn't been signaled which will be added for `CommandExecutor` in a subsequent commit.
2021-12-06 22:13:14 +01:00
return;
Cleanup `BufferView`/`TextureView` Locking Code Renames the variable to be neater and less confusing alongside adding comments for `try_lock()` to make the goal of the function more apparent.
2021-12-07 22:12:20 +01:00
backing->unlock();
backing = latestBacking;
Rework `TextureView` to be disconnected from `Texture` We want `TextureView`(s) to be disconnected from the backing on the host and instead represent a specific texture on the guest with a backing that can change depending on mapping of new textures which'd invalidate the backing but should now be automatically repointed to an appropriate new backing. This approach also requires locking of the backing to function as it is mutable till it has been locked or the backing has an attached `FenceCycle` that hasn't been signaled which will be added for `CommandExecutor` in a subsequent commit.
2021-12-06 22:13:14 +01:00
}
}
void TextureView::unlock() {
texture->unlock();
}
bool TextureView::try_lock() {
Cleanup `BufferView`/`TextureView` Locking Code Renames the variable to be neater and less confusing alongside adding comments for `try_lock()` to make the goal of the function more apparent.
2021-12-07 22:12:20 +01:00
auto backing{std::atomic_load(&texture)};
Rework `TextureView` to be disconnected from `Texture` We want `TextureView`(s) to be disconnected from the backing on the host and instead represent a specific texture on the guest with a backing that can change depending on mapping of new textures which'd invalidate the backing but should now be automatically repointed to an appropriate new backing. This approach also requires locking of the backing to function as it is mutable till it has been locked or the backing has an attached `FenceCycle` that hasn't been signaled which will be added for `CommandExecutor` in a subsequent commit.
2021-12-06 22:13:14 +01:00
while (true) {
Cleanup `BufferView`/`TextureView` Locking Code Renames the variable to be neater and less confusing alongside adding comments for `try_lock()` to make the goal of the function more apparent.
2021-12-07 22:12:20 +01:00
bool success{backing->try_lock()};
Rework `TextureView` to be disconnected from `Texture` We want `TextureView`(s) to be disconnected from the backing on the host and instead represent a specific texture on the guest with a backing that can change depending on mapping of new textures which'd invalidate the backing but should now be automatically repointed to an appropriate new backing. This approach also requires locking of the backing to function as it is mutable till it has been locked or the backing has an attached `FenceCycle` that hasn't been signaled which will be added for `CommandExecutor` in a subsequent commit.
2021-12-06 22:13:14 +01:00
Cleanup `BufferView`/`TextureView` Locking Code Renames the variable to be neater and less confusing alongside adding comments for `try_lock()` to make the goal of the function more apparent.
2021-12-07 22:12:20 +01:00
auto latestBacking{std::atomic_load(&texture)};
if (backing == latestBacking)
// We want to ensure that the try_lock() was on the latest backing and not on an outdated one
Rework `TextureView` to be disconnected from `Texture` We want `TextureView`(s) to be disconnected from the backing on the host and instead represent a specific texture on the guest with a backing that can change depending on mapping of new textures which'd invalidate the backing but should now be automatically repointed to an appropriate new backing. This approach also requires locking of the backing to function as it is mutable till it has been locked or the backing has an attached `FenceCycle` that hasn't been signaled which will be added for `CommandExecutor` in a subsequent commit.
2021-12-06 22:13:14 +01:00
return success;
if (success)
Cleanup `BufferView`/`TextureView` Locking Code Renames the variable to be neater and less confusing alongside adding comments for `try_lock()` to make the goal of the function more apparent.
2021-12-07 22:12:20 +01:00
// We only unlock() if the try_lock() was successful and we acquired the mutex
backing->unlock();
backing = latestBacking;
Rework `TextureView` to be disconnected from `Texture` We want `TextureView`(s) to be disconnected from the backing on the host and instead represent a specific texture on the guest with a backing that can change depending on mapping of new textures which'd invalidate the backing but should now be automatically repointed to an appropriate new backing. This approach also requires locking of the backing to function as it is mutable till it has been locked or the backing has an attached `FenceCycle` that hasn't been signaled which will be added for `CommandExecutor` in a subsequent commit.
2021-12-06 22:13:14 +01:00
}
}
Use mirror mappings for Textures and Buffers This is a prerequisite to memory trapping as we need to write to the mirror to avoid a race condition with external threads writing to a texture/buffer while we do so ourselves for the sync on a read/write, it also avoids an additional `mprotect` to `-WX`/`RWX` on a read access. An additional advantage for textures especially is that we now support split-mapping textures due to laying them out in a contiguous mirror and they will not require costly algorithmic changes. Buffers should also benefit from not needing to iterate over every region when they are split into multiple mappings.
2022-03-06 15:57:13 +01:00
void Texture::SetupGuestMappings() {
auto &mappings{guest->mappings};
if (mappings.size() == 1) {
auto mapping{mappings.front()};
u8 *alignedData{util::AlignDown(mapping.data(), PAGE_SIZE)};
size_t alignedSize{static_cast<size_t>(util::AlignUp(mapping.data() + mapping.size(), PAGE_SIZE) - alignedData)};
alignedMirror = gpu.state.process->memory.CreateMirror(alignedData, alignedSize);
mirror = alignedMirror.subspan(static_cast<size_t>(mapping.data() - alignedData), mapping.size());
} else {
std::vector<span<u8>> alignedMappings;
const auto &frontMapping{mappings.front()};
u8 *alignedData{util::AlignDown(frontMapping.data(), PAGE_SIZE)};
alignedMappings.emplace_back(alignedData, (frontMapping.data() + frontMapping.size()) - alignedData);
size_t totalSize{frontMapping.size()};
for (auto it{std::next(mappings.begin())}; it != std::prev(mappings.end()); ++it) {
auto mappingSize{it->size()};
alignedMappings.emplace_back(it->data(), mappingSize);
totalSize += mappingSize;
}
const auto &backMapping{mappings.back()};
totalSize += backMapping.size();
alignedMappings.emplace_back(backMapping.data(), util::AlignUp(backMapping.size(), PAGE_SIZE));
alignedMirror = gpu.state.process->memory.CreateMirrors(alignedMappings);
mirror = alignedMirror.subspan(static_cast<size_t>(frontMapping.data() - alignedData), totalSize);
}
}
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
std::shared_ptr<memory::StagingBuffer> Texture::SynchronizeHostImpl(const std::shared_ptr<FenceCycle> &pCycle) {
if (!guest)
throw exception("Synchronization of host textures requires a valid guest texture to synchronize from");
else if (guest->dimensions != dimensions)
throw exception("Guest and host dimensions being different is not supported currently");
Use mirror mappings for Textures and Buffers This is a prerequisite to memory trapping as we need to write to the mirror to avoid a race condition with external threads writing to a texture/buffer while we do so ourselves for the sync on a read/write, it also avoids an additional `mprotect` to `-WX`/`RWX` on a read access. An additional advantage for textures especially is that we now support split-mapping textures due to laying them out in a contiguous mirror and they will not require costly algorithmic changes. Buffers should also benefit from not needing to iterate over every region when they are split into multiple mappings.
2022-03-06 15:57:13 +01:00
auto pointer{mirror.data()};
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
auto size{format->GetSize(dimensions)};
WaitOnBacking();
u8 *bufferData;
auto stagingBuffer{[&]() -> std::shared_ptr<memory::StagingBuffer> {
if (tiling == vk::ImageTiling::eOptimal || !std::holds_alternative<memory::Image>(backing)) {
// We need a staging buffer for all optimal copies (since we aren't aware of the host optimal layout) and linear textures which we cannot map on the CPU since we do not have access to their backing VkDeviceMemory
auto stagingBuffer{gpu.memory.AllocateStagingBuffer(size)};
bufferData = stagingBuffer->data();
return stagingBuffer;
} else if (tiling == vk::ImageTiling::eLinear) {
// We can optimize linear texture sync on a UMA by mapping the texture onto the CPU and copying directly into it rather than a staging buffer
bufferData = std::get<memory::Image>(backing).data();
if (cycle.lock() != pCycle)
WaitOnFence();
return nullptr;
} else {
throw exception("Guest -> Host synchronization of images tiled as '{}' isn't implemented", vk::to_string(tiling));
}
}()};
if (guest->tileConfig.mode == texture::TileMode::Block)
Calculate Blocklinear Texture Aligned Size Correctly The size of blocklinear textures did not consider alignment to Block/ROB boundaries before, it is aligned to them now. Incorrect sizes led to textures not being aliased correctly due to different size calculations for GraphicBufferProducer surfaces and Maxwell3D color RTs.
2022-01-13 03:12:56 +01:00
texture::CopyBlockLinearToLinear(*guest, pointer, bufferData);
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
else if (guest->tileConfig.mode == texture::TileMode::Pitch)
Calculate Blocklinear Texture Aligned Size Correctly The size of blocklinear textures did not consider alignment to Block/ROB boundaries before, it is aligned to them now. Incorrect sizes led to textures not being aliased correctly due to different size calculations for GraphicBufferProducer surfaces and Maxwell3D color RTs.
2022-01-13 03:12:56 +01:00
texture::CopyPitchLinearToLinear(*guest, pointer, bufferData);
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
else if (guest->tileConfig.mode == texture::TileMode::Linear)
std::memcpy(bufferData, pointer, size);
if (stagingBuffer && cycle.lock() != pCycle)
WaitOnFence();
return stagingBuffer;
}
void Texture::CopyFromStagingBuffer(const vk::raii::CommandBuffer &commandBuffer, const std::shared_ptr<memory::StagingBuffer> &stagingBuffer) {
auto image{GetBacking()};
Improve Vulkan `Texture` Synchronization The Vulkan Pipeline Barriers were unoptimal and incorrect to some degree prior as we purely synchronized images and not staging buffers. This has now been fixed and improved in general with more relevant synchronization.
2021-11-09 16:38:03 +01:00
if (layout == vk::ImageLayout::eUndefined)
Fix Texture Synchronization Bug Fixes texture corruption due to incorrect synchronization, the barrier would not enforce waiting till the texture was entirely rendered causing an incomplete texture to be downloaded which lead to rendering bugs for certain GPUs including ARM's Mali GPUs.
2022-01-09 23:46:56 +01:00
commandBuffer.pipelineBarrier(vk::PipelineStageFlagBits::eHost, vk::PipelineStageFlagBits::eTransfer, {}, {}, {}, vk::ImageMemoryBarrier{
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
.image = image,
Improve Vulkan `Texture` Synchronization The Vulkan Pipeline Barriers were unoptimal and incorrect to some degree prior as we purely synchronized images and not staging buffers. This has now been fixed and improved in general with more relevant synchronization.
2021-11-09 16:38:03 +01:00
.srcAccessMask = vk::AccessFlagBits::eMemoryRead,
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
.dstAccessMask = vk::AccessFlagBits::eTransferWrite,
Improve Vulkan `Texture` Synchronization The Vulkan Pipeline Barriers were unoptimal and incorrect to some degree prior as we purely synchronized images and not staging buffers. This has now been fixed and improved in general with more relevant synchronization.
2021-11-09 16:38:03 +01:00
.oldLayout = std::exchange(layout, vk::ImageLayout::eGeneral),
.newLayout = vk::ImageLayout::eGeneral,
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.subresourceRange = {
.aspectMask = format->vkAspect,
.levelCount = mipLevels,
.layerCount = layerCount,
},
});
Add Depth/Stencil Format Support to `Texture` Sets `VkImageUsageFlags` correctly rather than hardcoding it for color attachments and adds multiple `VkBufferImageCopy` to `VkCmdCopyBufferToImage` for Color/Depth/Stencil aspects of an image.
2021-12-07 21:22:22 +01:00
boost::container::static_vector<const vk::BufferImageCopy, 3> bufferImageCopies;
auto pushBufferImageCopyWithAspect{[&](vk::ImageAspectFlagBits aspect) {
bufferImageCopies.emplace_back(
vk::BufferImageCopy{
.imageExtent = dimensions,
.imageSubresource = {
.aspectMask = aspect,
.layerCount = layerCount,
},
});
}};
if (format->vkAspect & vk::ImageAspectFlagBits::eColor)
pushBufferImageCopyWithAspect(vk::ImageAspectFlagBits::eColor);
if (format->vkAspect & vk::ImageAspectFlagBits::eDepth)
pushBufferImageCopyWithAspect(vk::ImageAspectFlagBits::eDepth);
if (format->vkAspect & vk::ImageAspectFlagBits::eStencil)
pushBufferImageCopyWithAspect(vk::ImageAspectFlagBits::eStencil);
commandBuffer.copyBufferToImage(stagingBuffer->vkBuffer, image, layout, vk::ArrayProxy(static_cast<u32>(bufferImageCopies.size()), bufferImageCopies.data()));
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
}
void Texture::CopyIntoStagingBuffer(const vk::raii::CommandBuffer &commandBuffer, const std::shared_ptr<memory::StagingBuffer> &stagingBuffer) {
auto image{GetBacking()};
Fix Texture Synchronization Bug Fixes texture corruption due to incorrect synchronization, the barrier would not enforce waiting till the texture was entirely rendered causing an incomplete texture to be downloaded which lead to rendering bugs for certain GPUs including ARM's Mali GPUs.
2022-01-09 23:46:56 +01:00
commandBuffer.pipelineBarrier(vk::PipelineStageFlagBits::eBottomOfPipe, vk::PipelineStageFlagBits::eTransfer, {}, {}, {}, vk::ImageMemoryBarrier{
Improve Vulkan `Texture` Synchronization The Vulkan Pipeline Barriers were unoptimal and incorrect to some degree prior as we purely synchronized images and not staging buffers. This has now been fixed and improved in general with more relevant synchronization.
2021-11-09 16:38:03 +01:00
.image = image,
.srcAccessMask = vk::AccessFlagBits::eMemoryWrite,
.dstAccessMask = vk::AccessFlagBits::eTransferRead,
.oldLayout = layout,
.newLayout = layout,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.subresourceRange = {
.aspectMask = format->vkAspect,
.levelCount = mipLevels,
.layerCount = layerCount,
},
});
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
Support Multi-Aspect Copy in `Texture::CopyIntoStagingBuffer` Only copying a single aspect was supported by `CopyIntoStagingBuffer` earlier due to not supplying a `VkBufferImageCopy` for each aspect separately, this has now been done with Color/Depth/Stencil aspects having their own `VkBufferImageCopy` for the `VkCmdCopyImageToBuffer` command.
2021-12-08 09:58:45 +01:00
boost::container::static_vector<const vk::BufferImageCopy, 3> bufferImageCopies;
auto pushBufferImageCopyWithAspect{[&](vk::ImageAspectFlagBits aspect) {
bufferImageCopies.emplace_back(
vk::BufferImageCopy{
.imageExtent = dimensions,
.imageSubresource = {
.aspectMask = aspect,
.layerCount = layerCount,
},
});
}};
if (format->vkAspect & vk::ImageAspectFlagBits::eColor)
pushBufferImageCopyWithAspect(vk::ImageAspectFlagBits::eColor);
if (format->vkAspect & vk::ImageAspectFlagBits::eDepth)
pushBufferImageCopyWithAspect(vk::ImageAspectFlagBits::eDepth);
if (format->vkAspect & vk::ImageAspectFlagBits::eStencil)
pushBufferImageCopyWithAspect(vk::ImageAspectFlagBits::eStencil);
commandBuffer.copyImageToBuffer(image, layout, stagingBuffer->vkBuffer, vk::ArrayProxy(static_cast<u32>(bufferImageCopies.size()), bufferImageCopies.data()));
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
Improve Vulkan `Texture` Synchronization The Vulkan Pipeline Barriers were unoptimal and incorrect to some degree prior as we purely synchronized images and not staging buffers. This has now been fixed and improved in general with more relevant synchronization.
2021-11-09 16:38:03 +01:00
commandBuffer.pipelineBarrier(vk::PipelineStageFlagBits::eTransfer, vk::PipelineStageFlagBits::eHost, {}, {}, vk::BufferMemoryBarrier{
.srcAccessMask = vk::AccessFlagBits::eTransferWrite,
.dstAccessMask = vk::AccessFlagBits::eHostRead,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.buffer = stagingBuffer->vkBuffer,
.offset = 0,
.size = stagingBuffer->size(),
}, {});
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
}
void Texture::CopyToGuest(u8 *hostBuffer) {
Use mirror mappings for Textures and Buffers This is a prerequisite to memory trapping as we need to write to the mirror to avoid a race condition with external threads writing to a texture/buffer while we do so ourselves for the sync on a read/write, it also avoids an additional `mprotect` to `-WX`/`RWX` on a read access. An additional advantage for textures especially is that we now support split-mapping textures due to laying them out in a contiguous mirror and they will not require costly algorithmic changes. Buffers should also benefit from not needing to iterate over every region when they are split into multiple mappings.
2022-03-06 15:57:13 +01:00
auto guestOutput{mirror.data()};
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
if (guest->tileConfig.mode == texture::TileMode::Block)
Calculate Blocklinear Texture Aligned Size Correctly The size of blocklinear textures did not consider alignment to Block/ROB boundaries before, it is aligned to them now. Incorrect sizes led to textures not being aliased correctly due to different size calculations for GraphicBufferProducer surfaces and Maxwell3D color RTs.
2022-01-13 03:12:56 +01:00
texture::CopyLinearToBlockLinear(*guest, hostBuffer, guestOutput);
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
else if (guest->tileConfig.mode == texture::TileMode::Pitch)
Calculate Blocklinear Texture Aligned Size Correctly The size of blocklinear textures did not consider alignment to Block/ROB boundaries before, it is aligned to them now. Incorrect sizes led to textures not being aliased correctly due to different size calculations for GraphicBufferProducer surfaces and Maxwell3D color RTs.
2022-01-13 03:12:56 +01:00
texture::CopyLinearToPitchLinear(*guest, hostBuffer, guestOutput);
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
else if (guest->tileConfig.mode == texture::TileMode::Linear)
std::memcpy(hostBuffer, guestOutput, format->GetSize(dimensions));
}
Texture::TextureBufferCopy::TextureBufferCopy(std::shared_ptr<Texture> texture, std::shared_ptr<memory::StagingBuffer> stagingBuffer) : texture(std::move(texture)), stagingBuffer(std::move(stagingBuffer)) {}
Texture::TextureBufferCopy::~TextureBufferCopy() {
Rework `TextureView` to be disconnected from `Texture` We want `TextureView`(s) to be disconnected from the backing on the host and instead represent a specific texture on the guest with a backing that can change depending on mapping of new textures which'd invalidate the backing but should now be automatically repointed to an appropriate new backing. This approach also requires locking of the backing to function as it is mutable till it has been locked or the backing has an attached `FenceCycle` that hasn't been signaled which will be added for `CommandExecutor` in a subsequent commit.
2021-12-06 22:13:14 +01:00
TRACE_EVENT("gpu", "Texture::TextureBufferCopy");
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
texture->CopyToGuest(stagingBuffer ? stagingBuffer->data() : std::get<memory::Image>(texture->backing).data());
}
Validation Layer Filter + Fix `Texture`, GPU & `PresentationEngine` bugs This commit implements a filter by type for any validation layer output, this allows filtering out any logs which may be unnecessary and additionally triggering a breakpoint as required. An issue concerning the `NDEBUG` flag never being set was fixed, it's now supplied as a release compiler flag. The issue can manifest itself by always relying on a validation layer even though it shouldn't on release, this is why the validation layer was mistakenly disabled entirely previously by using `#ifndef` rather than `#ifdef`. An issue with the initial layout of a texture being supplied as neither `VK_IMAGE_LAYOUT_UNDEFINED` or `VK_IMAGE_LAYOUT_PREINITIALIZED` was fixed, these cases are now handled by transitioning to those layouts after creating the image rather than supplying it within `initialLayout`. Another issue was fixed regarding not maintaining a transformation after a surface has been destroyed and recreated existed and manifested itself when the user would go out of the app and come back in, they would see the surface having an identity transformation rather than the desired one.
2021-09-14 17:43:16 +02:00
Texture::Texture(GPU &gpu, BackingType &&backing, GuestTexture guest, texture::Dimensions dimensions, texture::Format format, vk::ImageLayout layout, vk::ImageTiling tiling, u32 mipLevels, u32 layerCount, vk::SampleCountFlagBits sampleCount)
: gpu(gpu),
backing(std::move(backing)),
layout(layout),
guest(std::move(guest)),
dimensions(dimensions),
format(format),
tiling(tiling),
mipLevels(mipLevels),
layerCount(layerCount),
sampleCount(sampleCount) {
Use mirror mappings for Textures and Buffers This is a prerequisite to memory trapping as we need to write to the mirror to avoid a race condition with external threads writing to a texture/buffer while we do so ourselves for the sync on a read/write, it also avoids an additional `mprotect` to `-WX`/`RWX` on a read access. An additional advantage for textures especially is that we now support split-mapping textures due to laying them out in a contiguous mirror and they will not require costly algorithmic changes. Buffers should also benefit from not needing to iterate over every region when they are split into multiple mappings.
2022-03-06 15:57:13 +01:00
SetupGuestMappings();
Initial Texture Manager Implementation + Maxwell3D Render Target Implement the groundwork for the texture manager to be able to report basic overlaps and be extended to support more in the future. The Maxwell3D registers `RenderTargetControl`, `RenderTarget` and a stub for `ClearBuffers` were implemented. A lot of changes were also made to `GuestTexture`/`Texture` for supporting mipmapping and multiple array layers alongside significant architectural changes to `GuestTexture` effectively disconnecting it from `Texture` with it no longer being a parent rather an object that can be used to create a `Texture` object. Note: Support for fragmented CPU mappings hasn't been added for texture synchronization yet
2021-08-20 22:17:13 +02:00
if (GetBacking())
SynchronizeHost();
Redesign Texture Class + Improve Presentation Engine This commit reworks the `Texture` class to include a Vulkan Image backing that can be optionally owning or non-owning and swapped in with consideration for Vulkan image layout, it also adds CPU-sided synchronization for the texture objects with FenceCycle. It also makes the appropriate changes to `PresentationEngine` and `GraphicBufferProducer` to work with the new `Texture` class while setting the groundwork for supporting swapchain recreation. It also fixes a log in `IpcResponse` and improves the display mode selection algorithm by further weighing refresh rate.
2021-05-22 19:46:28 +02:00
}
Validation Layer Filter + Fix `Texture`, GPU & `PresentationEngine` bugs This commit implements a filter by type for any validation layer output, this allows filtering out any logs which may be unnecessary and additionally triggering a breakpoint as required. An issue concerning the `NDEBUG` flag never being set was fixed, it's now supplied as a release compiler flag. The issue can manifest itself by always relying on a validation layer even though it shouldn't on release, this is why the validation layer was mistakenly disabled entirely previously by using `#ifndef` rather than `#ifdef`. An issue with the initial layout of a texture being supplied as neither `VK_IMAGE_LAYOUT_UNDEFINED` or `VK_IMAGE_LAYOUT_PREINITIALIZED` was fixed, these cases are now handled by transitioning to those layouts after creating the image rather than supplying it within `initialLayout`. Another issue was fixed regarding not maintaining a transformation after a surface has been destroyed and recreated existed and manifested itself when the user would go out of the app and come back in, they would see the surface having an identity transformation rather than the desired one.
2021-09-14 17:43:16 +02:00
Texture::Texture(GPU &gpu, BackingType &&backing, texture::Dimensions dimensions, texture::Format format, vk::ImageLayout layout, vk::ImageTiling tiling, u32 mipLevels, u32 layerCount, vk::SampleCountFlagBits sampleCount)
: gpu(gpu),
backing(std::move(backing)),
dimensions(dimensions),
format(format),
layout(layout),
tiling(tiling),
mipLevels(mipLevels),
layerCount(layerCount),
sampleCount(sampleCount) {}
Initial Texture Manager Implementation + Maxwell3D Render Target Implement the groundwork for the texture manager to be able to report basic overlaps and be extended to support more in the future. The Maxwell3D registers `RenderTargetControl`, `RenderTarget` and a stub for `ClearBuffers` were implemented. A lot of changes were also made to `GuestTexture`/`Texture` for supporting mipmapping and multiple array layers alongside significant architectural changes to `GuestTexture` effectively disconnecting it from `Texture` with it no longer being a parent rather an object that can be used to create a `Texture` object. Note: Support for fragmented CPU mappings hasn't been added for texture synchronization yet
2021-08-20 22:17:13 +02:00
Texture::Texture(GPU &pGpu, GuestTexture pGuest)
: gpu(pGpu),
guest(std::move(pGuest)),
dimensions(guest->dimensions),
format(guest->format),
Validation Layer Filter + Fix `Texture`, GPU & `PresentationEngine` bugs This commit implements a filter by type for any validation layer output, this allows filtering out any logs which may be unnecessary and additionally triggering a breakpoint as required. An issue concerning the `NDEBUG` flag never being set was fixed, it's now supplied as a release compiler flag. The issue can manifest itself by always relying on a validation layer even though it shouldn't on release, this is why the validation layer was mistakenly disabled entirely previously by using `#ifndef` rather than `#ifdef`. An issue with the initial layout of a texture being supplied as neither `VK_IMAGE_LAYOUT_UNDEFINED` or `VK_IMAGE_LAYOUT_PREINITIALIZED` was fixed, these cases are now handled by transitioning to those layouts after creating the image rather than supplying it within `initialLayout`. Another issue was fixed regarding not maintaining a transformation after a surface has been destroyed and recreated existed and manifested itself when the user would go out of the app and come back in, they would see the surface having an identity transformation rather than the desired one.
2021-09-14 17:43:16 +02:00
layout(vk::ImageLayout::eUndefined),
Force Textures to be Optimal on Host GPU We don't respect the host subresource layout in synchronizing linear textures from the guest to the host when mapped to memory directly, this leads to texture corruption and while the real fix would involve respecting the host subresource layout, this has been deferred for later as real world performance advantages/disadvantages associated with this change can be observed more carefully to determine if it's worth it.
2022-01-11 21:28:46 +01:00
tiling(vk::ImageTiling::eOptimal), // Force Optimal due to not adhering to host subresource layout during Linear synchronization
Initial Texture Manager Implementation + Maxwell3D Render Target Implement the groundwork for the texture manager to be able to report basic overlaps and be extended to support more in the future. The Maxwell3D registers `RenderTargetControl`, `RenderTarget` and a stub for `ClearBuffers` were implemented. A lot of changes were also made to `GuestTexture`/`Texture` for supporting mipmapping and multiple array layers alongside significant architectural changes to `GuestTexture` effectively disconnecting it from `Texture` with it no longer being a parent rather an object that can be used to create a `Texture` object. Note: Support for fragmented CPU mappings hasn't been added for texture synchronization yet
2021-08-20 22:17:13 +02:00
mipLevels(1),
layerCount(guest->layerCount),
sampleCount(vk::SampleCountFlagBits::e1) {
Implement Maxwell3D Sampled Textures The descriptor sets should now contain a combined image and sampler handle for any sampled textures in the guest shader from the supplied offset into the texture constant buffer. Note: Games tend to rely on inline constant buffer updates for writing the texture constant buffer and due to it not being implemented, the value will be read as 0 which is incorrect.
2022-01-05 20:29:54 +01:00
vk::ImageUsageFlags usage{vk::ImageUsageFlagBits::eTransferSrc | vk::ImageUsageFlagBits::eTransferDst | vk::ImageUsageFlagBits::eSampled};
Don't set COLOR_ATTACHMENT_BIT for compressed formats. The better solution would be to only set this for formats that support it on original HW but this will get rid of the validation errors for now.
2022-02-28 22:44:26 +01:00
if ((format->vkAspect & vk::ImageAspectFlagBits::eColor) && !format->IsCompressed())
Add Depth/Stencil Format Support to `Texture` Sets `VkImageUsageFlags` correctly rather than hardcoding it for color attachments and adds multiple `VkBufferImageCopy` to `VkCmdCopyBufferToImage` for Color/Depth/Stencil aspects of an image.
2021-12-07 21:22:22 +01:00
usage |= vk::ImageUsageFlagBits::eColorAttachment;
if (format->vkAspect & (vk::ImageAspectFlagBits::eDepth | vk::ImageAspectFlagBits::eStencil))
usage |= vk::ImageUsageFlagBits::eDepthStencilAttachment;
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
vk::ImageCreateInfo imageCreateInfo{
Initial Texture Manager Implementation + Maxwell3D Render Target Implement the groundwork for the texture manager to be able to report basic overlaps and be extended to support more in the future. The Maxwell3D registers `RenderTargetControl`, `RenderTarget` and a stub for `ClearBuffers` were implemented. A lot of changes were also made to `GuestTexture`/`Texture` for supporting mipmapping and multiple array layers alongside significant architectural changes to `GuestTexture` effectively disconnecting it from `Texture` with it no longer being a parent rather an object that can be used to create a `Texture` object. Note: Support for fragmented CPU mappings hasn't been added for texture synchronization yet
2021-08-20 22:17:13 +02:00
.imageType = guest->dimensions.GetType(),
.format = *guest->format,
.extent = guest->dimensions,
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
.mipLevels = 1,
Initial Texture Manager Implementation + Maxwell3D Render Target Implement the groundwork for the texture manager to be able to report basic overlaps and be extended to support more in the future. The Maxwell3D registers `RenderTargetControl`, `RenderTarget` and a stub for `ClearBuffers` were implemented. A lot of changes were also made to `GuestTexture`/`Texture` for supporting mipmapping and multiple array layers alongside significant architectural changes to `GuestTexture` effectively disconnecting it from `Texture` with it no longer being a parent rather an object that can be used to create a `Texture` object. Note: Support for fragmented CPU mappings hasn't been added for texture synchronization yet
2021-08-20 22:17:13 +02:00
.arrayLayers = guest->layerCount,
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
.samples = vk::SampleCountFlagBits::e1,
.tiling = tiling,
Add Depth/Stencil Format Support to `Texture` Sets `VkImageUsageFlags` correctly rather than hardcoding it for color attachments and adds multiple `VkBufferImageCopy` to `VkCmdCopyBufferToImage` for Color/Depth/Stencil aspects of an image.
2021-12-07 21:22:22 +01:00
.usage = usage,
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
.sharingMode = vk::SharingMode::eExclusive,
.queueFamilyIndexCount = 1,
Initial Texture Manager Implementation + Maxwell3D Render Target Implement the groundwork for the texture manager to be able to report basic overlaps and be extended to support more in the future. The Maxwell3D registers `RenderTargetControl`, `RenderTarget` and a stub for `ClearBuffers` were implemented. A lot of changes were also made to `GuestTexture`/`Texture` for supporting mipmapping and multiple array layers alongside significant architectural changes to `GuestTexture` effectively disconnecting it from `Texture` with it no longer being a parent rather an object that can be used to create a `Texture` object. Note: Support for fragmented CPU mappings hasn't been added for texture synchronization yet
2021-08-20 22:17:13 +02:00
.pQueueFamilyIndices = &gpu.vkQueueFamilyIndex,
.initialLayout = layout,
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
};
Initial Texture Manager Implementation + Maxwell3D Render Target Implement the groundwork for the texture manager to be able to report basic overlaps and be extended to support more in the future. The Maxwell3D registers `RenderTargetControl`, `RenderTarget` and a stub for `ClearBuffers` were implemented. A lot of changes were also made to `GuestTexture`/`Texture` for supporting mipmapping and multiple array layers alongside significant architectural changes to `GuestTexture` effectively disconnecting it from `Texture` with it no longer being a parent rather an object that can be used to create a `Texture` object. Note: Support for fragmented CPU mappings hasn't been added for texture synchronization yet
2021-08-20 22:17:13 +02:00
backing = tiling != vk::ImageTiling::eLinear ? gpu.memory.AllocateImage(imageCreateInfo) : gpu.memory.AllocateMappedImage(imageCreateInfo);
Validation Layer Filter + Fix `Texture`, GPU & `PresentationEngine` bugs This commit implements a filter by type for any validation layer output, this allows filtering out any logs which may be unnecessary and additionally triggering a breakpoint as required. An issue concerning the `NDEBUG` flag never being set was fixed, it's now supplied as a release compiler flag. The issue can manifest itself by always relying on a validation layer even though it shouldn't on release, this is why the validation layer was mistakenly disabled entirely previously by using `#ifndef` rather than `#ifdef`. An issue with the initial layout of a texture being supplied as neither `VK_IMAGE_LAYOUT_UNDEFINED` or `VK_IMAGE_LAYOUT_PREINITIALIZED` was fixed, these cases are now handled by transitioning to those layouts after creating the image rather than supplying it within `initialLayout`. Another issue was fixed regarding not maintaining a transformation after a surface has been destroyed and recreated existed and manifested itself when the user would go out of the app and come back in, they would see the surface having an identity transformation rather than the desired one.
2021-09-14 17:43:16 +02:00
TransitionLayout(vk::ImageLayout::eGeneral);
Use mirror mappings for Textures and Buffers This is a prerequisite to memory trapping as we need to write to the mirror to avoid a race condition with external threads writing to a texture/buffer while we do so ourselves for the sync on a read/write, it also avoids an additional `mprotect` to `-WX`/`RWX` on a read access. An additional advantage for textures especially is that we now support split-mapping textures due to laying them out in a contiguous mirror and they will not require costly algorithmic changes. Buffers should also benefit from not needing to iterate over every region when they are split into multiple mappings.
2022-03-06 15:57:13 +01:00
SetupGuestMappings();
Redesign Texture Class + Improve Presentation Engine This commit reworks the `Texture` class to include a Vulkan Image backing that can be optionally owning or non-owning and swapped in with consideration for Vulkan image layout, it also adds CPU-sided synchronization for the texture objects with FenceCycle. It also makes the appropriate changes to `PresentationEngine` and `GraphicBufferProducer` to work with the new `Texture` class while setting the groundwork for supporting swapchain recreation. It also fixes a log in `IpcResponse` and improves the display mode selection algorithm by further weighing refresh rate.
2021-05-22 19:46:28 +02:00
}
Validation Layer Filter + Fix `Texture`, GPU & `PresentationEngine` bugs This commit implements a filter by type for any validation layer output, this allows filtering out any logs which may be unnecessary and additionally triggering a breakpoint as required. An issue concerning the `NDEBUG` flag never being set was fixed, it's now supplied as a release compiler flag. The issue can manifest itself by always relying on a validation layer even though it shouldn't on release, this is why the validation layer was mistakenly disabled entirely previously by using `#ifndef` rather than `#ifdef`. An issue with the initial layout of a texture being supplied as neither `VK_IMAGE_LAYOUT_UNDEFINED` or `VK_IMAGE_LAYOUT_PREINITIALIZED` was fixed, these cases are now handled by transitioning to those layouts after creating the image rather than supplying it within `initialLayout`. Another issue was fixed regarding not maintaining a transformation after a surface has been destroyed and recreated existed and manifested itself when the user would go out of the app and come back in, they would see the surface having an identity transformation rather than the desired one.
2021-09-14 17:43:16 +02:00
Texture::Texture(GPU &gpu, texture::Dimensions dimensions, texture::Format format, vk::ImageLayout initialLayout, vk::ImageUsageFlags usage, vk::ImageTiling tiling, u32 mipLevels, u32 layerCount, vk::SampleCountFlagBits sampleCount)
: gpu(gpu),
dimensions(dimensions),
format(format),
layout(initialLayout == vk::ImageLayout::ePreinitialized ? vk::ImageLayout::ePreinitialized : vk::ImageLayout::eUndefined),
Force Textures to be Optimal on Host GPU We don't respect the host subresource layout in synchronizing linear textures from the guest to the host when mapped to memory directly, this leads to texture corruption and while the real fix would involve respecting the host subresource layout, this has been deferred for later as real world performance advantages/disadvantages associated with this change can be observed more carefully to determine if it's worth it.
2022-01-11 21:28:46 +01:00
tiling(vk::ImageTiling::eOptimal), // Same as above
Validation Layer Filter + Fix `Texture`, GPU & `PresentationEngine` bugs This commit implements a filter by type for any validation layer output, this allows filtering out any logs which may be unnecessary and additionally triggering a breakpoint as required. An issue concerning the `NDEBUG` flag never being set was fixed, it's now supplied as a release compiler flag. The issue can manifest itself by always relying on a validation layer even though it shouldn't on release, this is why the validation layer was mistakenly disabled entirely previously by using `#ifndef` rather than `#ifdef`. An issue with the initial layout of a texture being supplied as neither `VK_IMAGE_LAYOUT_UNDEFINED` or `VK_IMAGE_LAYOUT_PREINITIALIZED` was fixed, these cases are now handled by transitioning to those layouts after creating the image rather than supplying it within `initialLayout`. Another issue was fixed regarding not maintaining a transformation after a surface has been destroyed and recreated existed and manifested itself when the user would go out of the app and come back in, they would see the surface having an identity transformation rather than the desired one.
2021-09-14 17:43:16 +02:00
mipLevels(mipLevels),
layerCount(layerCount),
sampleCount(sampleCount) {
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
vk::ImageCreateInfo imageCreateInfo{
.imageType = dimensions.GetType(),
Initial Texture Manager Implementation + Maxwell3D Render Target Implement the groundwork for the texture manager to be able to report basic overlaps and be extended to support more in the future. The Maxwell3D registers `RenderTargetControl`, `RenderTarget` and a stub for `ClearBuffers` were implemented. A lot of changes were also made to `GuestTexture`/`Texture` for supporting mipmapping and multiple array layers alongside significant architectural changes to `GuestTexture` effectively disconnecting it from `Texture` with it no longer being a parent rather an object that can be used to create a `Texture` object. Note: Support for fragmented CPU mappings hasn't been added for texture synchronization yet
2021-08-20 22:17:13 +02:00
.format = *format,
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
.extent = dimensions,
Initial Texture Manager Implementation + Maxwell3D Render Target Implement the groundwork for the texture manager to be able to report basic overlaps and be extended to support more in the future. The Maxwell3D registers `RenderTargetControl`, `RenderTarget` and a stub for `ClearBuffers` were implemented. A lot of changes were also made to `GuestTexture`/`Texture` for supporting mipmapping and multiple array layers alongside significant architectural changes to `GuestTexture` effectively disconnecting it from `Texture` with it no longer being a parent rather an object that can be used to create a `Texture` object. Note: Support for fragmented CPU mappings hasn't been added for texture synchronization yet
2021-08-20 22:17:13 +02:00
.mipLevels = mipLevels,
.arrayLayers = layerCount,
.samples = sampleCount,
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
.tiling = tiling,
.usage = usage | vk::ImageUsageFlagBits::eTransferSrc | vk::ImageUsageFlagBits::eTransferDst,
.sharingMode = vk::SharingMode::eExclusive,
.queueFamilyIndexCount = 1,
.pQueueFamilyIndices = &gpu.vkQueueFamilyIndex,
Validation Layer Filter + Fix `Texture`, GPU & `PresentationEngine` bugs This commit implements a filter by type for any validation layer output, this allows filtering out any logs which may be unnecessary and additionally triggering a breakpoint as required. An issue concerning the `NDEBUG` flag never being set was fixed, it's now supplied as a release compiler flag. The issue can manifest itself by always relying on a validation layer even though it shouldn't on release, this is why the validation layer was mistakenly disabled entirely previously by using `#ifndef` rather than `#ifdef`. An issue with the initial layout of a texture being supplied as neither `VK_IMAGE_LAYOUT_UNDEFINED` or `VK_IMAGE_LAYOUT_PREINITIALIZED` was fixed, these cases are now handled by transitioning to those layouts after creating the image rather than supplying it within `initialLayout`. Another issue was fixed regarding not maintaining a transformation after a surface has been destroyed and recreated existed and manifested itself when the user would go out of the app and come back in, they would see the surface having an identity transformation rather than the desired one.
2021-09-14 17:43:16 +02:00
.initialLayout = layout,
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
};
backing = tiling != vk::ImageTiling::eLinear ? gpu.memory.AllocateImage(imageCreateInfo) : gpu.memory.AllocateMappedImage(imageCreateInfo);
Validation Layer Filter + Fix `Texture`, GPU & `PresentationEngine` bugs This commit implements a filter by type for any validation layer output, this allows filtering out any logs which may be unnecessary and additionally triggering a breakpoint as required. An issue concerning the `NDEBUG` flag never being set was fixed, it's now supplied as a release compiler flag. The issue can manifest itself by always relying on a validation layer even though it shouldn't on release, this is why the validation layer was mistakenly disabled entirely previously by using `#ifndef` rather than `#ifdef`. An issue with the initial layout of a texture being supplied as neither `VK_IMAGE_LAYOUT_UNDEFINED` or `VK_IMAGE_LAYOUT_PREINITIALIZED` was fixed, these cases are now handled by transitioning to those layouts after creating the image rather than supplying it within `initialLayout`. Another issue was fixed regarding not maintaining a transformation after a surface has been destroyed and recreated existed and manifested itself when the user would go out of the app and come back in, they would see the surface having an identity transformation rather than the desired one.
2021-09-14 17:43:16 +02:00
if (initialLayout != layout)
TransitionLayout(initialLayout);
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
}
Use mirror mappings for Textures and Buffers This is a prerequisite to memory trapping as we need to write to the mirror to avoid a race condition with external threads writing to a texture/buffer while we do so ourselves for the sync on a read/write, it also avoids an additional `mprotect` to `-WX`/`RWX` on a read access. An additional advantage for textures especially is that we now support split-mapping textures due to laying them out in a contiguous mirror and they will not require costly algorithmic changes. Buffers should also benefit from not needing to iterate over every region when they are split into multiple mappings.
2022-03-06 15:57:13 +01:00
Texture::~Texture() {
std::lock_guard lock(*this);
SynchronizeGuest(true);
if (alignedMirror.valid())
munmap(alignedMirror.data(), alignedMirror.size());
}
Redesign Texture Class + Improve Presentation Engine This commit reworks the `Texture` class to include a Vulkan Image backing that can be optionally owning or non-owning and swapped in with consideration for Vulkan image layout, it also adds CPU-sided synchronization for the texture objects with FenceCycle. It also makes the appropriate changes to `PresentationEngine` and `GraphicBufferProducer` to work with the new `Texture` class while setting the groundwork for supporting swapchain recreation. It also fixes a log in `IpcResponse` and improves the display mode selection algorithm by further weighing refresh rate.
2021-05-22 19:46:28 +02:00
bool Texture::WaitOnBacking() {
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
TRACE_EVENT("gpu", "Texture::WaitOnBacking");
Redesign Texture Class + Improve Presentation Engine This commit reworks the `Texture` class to include a Vulkan Image backing that can be optionally owning or non-owning and swapped in with consideration for Vulkan image layout, it also adds CPU-sided synchronization for the texture objects with FenceCycle. It also makes the appropriate changes to `PresentationEngine` and `GraphicBufferProducer` to work with the new `Texture` class while setting the groundwork for supporting swapchain recreation. It also fixes a log in `IpcResponse` and improves the display mode selection algorithm by further weighing refresh rate.
2021-05-22 19:46:28 +02:00
if (GetBacking()) [[likely]] {
return false;
} else {
std::unique_lock lock(mutex, std::adopt_lock);
backingCondition.wait(lock, [&]() -> bool { return GetBacking(); });
lock.release();
return true;
}
}
void Texture::WaitOnFence() {
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
TRACE_EVENT("gpu", "Texture::WaitOnFence");
Subpass Support + More RT Formats + Fix `FenceCycle` Cyclic Dependencies Support for subpasses was added by reworking attachment reuse code to account for preserved attachments and subpass dependencies. A lot of RT formats were also added to allow SMO to boot up entirely, it should be noted that it doesn't render anything. `FenceCycle` had a cyclic dependency which broke clean exit, we now utilize `std::weak_ptr<FenceCycle>` inside the `Texture` object. A minor fix for broken stack traces was also made caused by supplying a `nullptr` C-string to libfmt when a symbol was unresolved which caused an `abort` due to invocation of `strlen` with it.
2021-09-22 05:30:11 +02:00
auto lCycle{cycle.lock()};
if (lCycle) {
lCycle->Wait();
Redesign Texture Class + Improve Presentation Engine This commit reworks the `Texture` class to include a Vulkan Image backing that can be optionally owning or non-owning and swapped in with consideration for Vulkan image layout, it also adds CPU-sided synchronization for the texture objects with FenceCycle. It also makes the appropriate changes to `PresentationEngine` and `GraphicBufferProducer` to work with the new `Texture` class while setting the groundwork for supporting swapchain recreation. It also fixes a log in `IpcResponse` and improves the display mode selection algorithm by further weighing refresh rate.
2021-05-22 19:46:28 +02:00
cycle.reset();
}
}
void Texture::SwapBacking(BackingType &&pBacking, vk::ImageLayout pLayout) {
WaitOnFence();
backing = std::move(pBacking);
layout = pLayout;
if (GetBacking())
backingCondition.notify_all();
}
void Texture::TransitionLayout(vk::ImageLayout pLayout) {
WaitOnBacking();
WaitOnFence();
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
TRACE_EVENT("gpu", "Texture::TransitionLayout");
Improve Vulkan `Texture` Synchronization The Vulkan Pipeline Barriers were unoptimal and incorrect to some degree prior as we purely synchronized images and not staging buffers. This has now been fixed and improved in general with more relevant synchronization.
2021-11-09 16:38:03 +01:00
if (layout != pLayout)
Redesign Texture Class + Improve Presentation Engine This commit reworks the `Texture` class to include a Vulkan Image backing that can be optionally owning or non-owning and swapped in with consideration for Vulkan image layout, it also adds CPU-sided synchronization for the texture objects with FenceCycle. It also makes the appropriate changes to `PresentationEngine` and `GraphicBufferProducer` to work with the new `Texture` class while setting the groundwork for supporting swapchain recreation. It also fixes a log in `IpcResponse` and improves the display mode selection algorithm by further weighing refresh rate.
2021-05-22 19:46:28 +02:00
cycle = gpu.scheduler.Submit([&](vk::raii::CommandBuffer &commandBuffer) {
Improve Vulkan `Texture` Synchronization The Vulkan Pipeline Barriers were unoptimal and incorrect to some degree prior as we purely synchronized images and not staging buffers. This has now been fixed and improved in general with more relevant synchronization.
2021-11-09 16:38:03 +01:00
commandBuffer.pipelineBarrier(vk::PipelineStageFlagBits::eTopOfPipe, vk::PipelineStageFlagBits::eBottomOfPipe, {}, {}, {}, vk::ImageMemoryBarrier{
Redesign Texture Class + Improve Presentation Engine This commit reworks the `Texture` class to include a Vulkan Image backing that can be optionally owning or non-owning and swapped in with consideration for Vulkan image layout, it also adds CPU-sided synchronization for the texture objects with FenceCycle. It also makes the appropriate changes to `PresentationEngine` and `GraphicBufferProducer` to work with the new `Texture` class while setting the groundwork for supporting swapchain recreation. It also fixes a log in `IpcResponse` and improves the display mode selection algorithm by further weighing refresh rate.
2021-05-22 19:46:28 +02:00
.image = GetBacking(),
Improve Vulkan `Texture` Synchronization The Vulkan Pipeline Barriers were unoptimal and incorrect to some degree prior as we purely synchronized images and not staging buffers. This has now been fixed and improved in general with more relevant synchronization.
2021-11-09 16:38:03 +01:00
.srcAccessMask = vk::AccessFlagBits::eNoneKHR,
.dstAccessMask = vk::AccessFlagBits::eNoneKHR,
.oldLayout = std::exchange(layout, pLayout),
Redesign Texture Class + Improve Presentation Engine This commit reworks the `Texture` class to include a Vulkan Image backing that can be optionally owning or non-owning and swapped in with consideration for Vulkan image layout, it also adds CPU-sided synchronization for the texture objects with FenceCycle. It also makes the appropriate changes to `PresentationEngine` and `GraphicBufferProducer` to work with the new `Texture` class while setting the groundwork for supporting swapchain recreation. It also fixes a log in `IpcResponse` and improves the display mode selection algorithm by further weighing refresh rate.
2021-05-22 19:46:28 +02:00
.newLayout = pLayout,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.subresourceRange = {
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
.aspectMask = format->vkAspect,
.levelCount = mipLevels,
.layerCount = layerCount,
Redesign Texture Class + Improve Presentation Engine This commit reworks the `Texture` class to include a Vulkan Image backing that can be optionally owning or non-owning and swapped in with consideration for Vulkan image layout, it also adds CPU-sided synchronization for the texture objects with FenceCycle. It also makes the appropriate changes to `PresentationEngine` and `GraphicBufferProducer` to work with the new `Texture` class while setting the groundwork for supporting swapchain recreation. It also fixes a log in `IpcResponse` and improves the display mode selection algorithm by further weighing refresh rate.
2021-05-22 19:46:28 +02:00
},
});
});
}
void Texture::SynchronizeHost() {
TRACE_EVENT("gpu", "Texture::SynchronizeHost");
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
auto stagingBuffer{SynchronizeHostImpl(nullptr)};
if (stagingBuffer) {
auto lCycle{gpu.scheduler.Submit([&](vk::raii::CommandBuffer &commandBuffer) {
CopyFromStagingBuffer(commandBuffer, stagingBuffer);
})};
lCycle->AttachObjects(stagingBuffer, shared_from_this());
cycle = lCycle;
}
}
Redesign Texture Class + Improve Presentation Engine This commit reworks the `Texture` class to include a Vulkan Image backing that can be optionally owning or non-owning and swapped in with consideration for Vulkan image layout, it also adds CPU-sided synchronization for the texture objects with FenceCycle. It also makes the appropriate changes to `PresentationEngine` and `GraphicBufferProducer` to work with the new `Texture` class while setting the groundwork for supporting swapchain recreation. It also fixes a log in `IpcResponse` and improves the display mode selection algorithm by further weighing refresh rate.
2021-05-22 19:46:28 +02:00
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
void Texture::SynchronizeHostWithBuffer(const vk::raii::CommandBuffer &commandBuffer, const std::shared_ptr<FenceCycle> &pCycle) {
TRACE_EVENT("gpu", "Texture::SynchronizeHostWithBuffer");
Redesign Texture Class + Improve Presentation Engine This commit reworks the `Texture` class to include a Vulkan Image backing that can be optionally owning or non-owning and swapped in with consideration for Vulkan image layout, it also adds CPU-sided synchronization for the texture objects with FenceCycle. It also makes the appropriate changes to `PresentationEngine` and `GraphicBufferProducer` to work with the new `Texture` class while setting the groundwork for supporting swapchain recreation. It also fixes a log in `IpcResponse` and improves the display mode selection algorithm by further weighing refresh rate.
2021-05-22 19:46:28 +02:00
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
auto stagingBuffer{SynchronizeHostImpl(pCycle)};
if (stagingBuffer) {
CopyFromStagingBuffer(commandBuffer, stagingBuffer);
pCycle->AttachObjects(stagingBuffer, shared_from_this());
cycle = pCycle;
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
}
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
}
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
void Texture::SynchronizeGuest() {
if (!guest)
throw exception("Synchronization of guest textures requires a valid guest texture to synchronize to");
else if (layout == vk::ImageLayout::eUndefined)
return; // If the state of the host texture is undefined then so can the guest
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
TRACE_EVENT("gpu", "Texture::SynchronizeGuest");
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
Improve Vulkan `Texture` Synchronization The Vulkan Pipeline Barriers were unoptimal and incorrect to some degree prior as we purely synchronized images and not staging buffers. This has now been fixed and improved in general with more relevant synchronization.
2021-11-09 16:38:03 +01:00
if (layout == vk::ImageLayout::eUndefined)
return; // We don't need to synchronize the image if it is in an undefined state on the host
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
WaitOnBacking();
WaitOnFence();
if (tiling == vk::ImageTiling::eOptimal || !std::holds_alternative<memory::Image>(backing)) {
auto size{format->GetSize(dimensions)};
auto stagingBuffer{gpu.memory.AllocateStagingBuffer(size)};
auto lCycle{gpu.scheduler.Submit([&](vk::raii::CommandBuffer &commandBuffer) {
CopyIntoStagingBuffer(commandBuffer, stagingBuffer);
Subpass Support + More RT Formats + Fix `FenceCycle` Cyclic Dependencies Support for subpasses was added by reworking attachment reuse code to account for preserved attachments and subpass dependencies. A lot of RT formats were also added to allow SMO to boot up entirely, it should be noted that it doesn't render anything. `FenceCycle` had a cyclic dependency which broke clean exit, we now utilize `std::weak_ptr<FenceCycle>` inside the `Texture` object. A minor fix for broken stack traces was also made caused by supplying a `nullptr` C-string to libfmt when a symbol was unresolved which caused an `abort` due to invocation of `strlen` with it.
2021-09-22 05:30:11 +02:00
})};
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
lCycle->AttachObject(std::make_shared<TextureBufferCopy>(shared_from_this(), stagingBuffer));
Subpass Support + More RT Formats + Fix `FenceCycle` Cyclic Dependencies Support for subpasses was added by reworking attachment reuse code to account for preserved attachments and subpass dependencies. A lot of RT formats were also added to allow SMO to boot up entirely, it should be noted that it doesn't render anything. `FenceCycle` had a cyclic dependency which broke clean exit, we now utilize `std::weak_ptr<FenceCycle>` inside the `Texture` object. A minor fix for broken stack traces was also made caused by supplying a `nullptr` C-string to libfmt when a symbol was unresolved which caused an `abort` due to invocation of `strlen` with it.
2021-09-22 05:30:11 +02:00
cycle = lCycle;
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
} else if (tiling == vk::ImageTiling::eLinear) {
// We can optimize linear texture sync on a UMA by mapping the texture onto the CPU and copying directly from it rather than using a staging buffer
CopyToGuest(std::get<memory::Image>(backing).data());
} else {
throw exception("Host -> Guest synchronization of images tiled as '{}' isn't implemented", vk::to_string(tiling));
Redesign Texture Class + Improve Presentation Engine This commit reworks the `Texture` class to include a Vulkan Image backing that can be optionally owning or non-owning and swapped in with consideration for Vulkan image layout, it also adds CPU-sided synchronization for the texture objects with FenceCycle. It also makes the appropriate changes to `PresentationEngine` and `GraphicBufferProducer` to work with the new `Texture` class while setting the groundwork for supporting swapchain recreation. It also fixes a log in `IpcResponse` and improves the display mode selection algorithm by further weighing refresh rate.
2021-05-22 19:46:28 +02:00
}
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
}
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
void Texture::SynchronizeGuestWithBuffer(const vk::raii::CommandBuffer &commandBuffer, const std::shared_ptr<FenceCycle> &pCycle) {
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
if (!guest)
throw exception("Synchronization of guest textures requires a valid guest texture to synchronize to");
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
else if (layout == vk::ImageLayout::eUndefined)
return; // If the state of the host texture is undefined then so can the guest
TRACE_EVENT("gpu", "Texture::SynchronizeGuestWithBuffer");
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
Improve Vulkan `Texture` Synchronization The Vulkan Pipeline Barriers were unoptimal and incorrect to some degree prior as we purely synchronized images and not staging buffers. This has now been fixed and improved in general with more relevant synchronization.
2021-11-09 16:38:03 +01:00
if (layout == vk::ImageLayout::eUndefined)
return;
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
WaitOnBacking();
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
if (cycle.lock() != pCycle)
WaitOnFence();
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
if (tiling == vk::ImageTiling::eOptimal || !std::holds_alternative<memory::Image>(backing)) {
auto size{format->GetSize(dimensions)};
auto stagingBuffer{gpu.memory.AllocateStagingBuffer(size)};
CopyIntoStagingBuffer(commandBuffer, stagingBuffer);
pCycle->AttachObject(std::make_shared<TextureBufferCopy>(shared_from_this(), stagingBuffer));
cycle = pCycle;
} else if (tiling == vk::ImageTiling::eLinear) {
CopyToGuest(std::get<memory::Image>(backing).data());
pCycle->AttachObject(std::make_shared<TextureBufferCopy>(shared_from_this()));
cycle = pCycle;
} else {
throw exception("Host -> Guest synchronization of images tiled as '{}' isn't implemented", vk::to_string(tiling));
}
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
}
Redesign Texture Class + Improve Presentation Engine This commit reworks the `Texture` class to include a Vulkan Image backing that can be optionally owning or non-owning and swapped in with consideration for Vulkan image layout, it also adds CPU-sided synchronization for the texture objects with FenceCycle. It also makes the appropriate changes to `PresentationEngine` and `GraphicBufferProducer` to work with the new `Texture` class while setting the groundwork for supporting swapchain recreation. It also fixes a log in `IpcResponse` and improves the display mode selection algorithm by further weighing refresh rate.
2021-05-22 19:46:28 +02:00
Rework `TextureView` to be disconnected from `Texture` We want `TextureView`(s) to be disconnected from the backing on the host and instead represent a specific texture on the guest with a backing that can change depending on mapping of new textures which'd invalidate the backing but should now be automatically repointed to an appropriate new backing. This approach also requires locking of the backing to function as it is mutable till it has been locked or the backing has an attached `FenceCycle` that hasn't been signaled which will be added for `CommandExecutor` in a subsequent commit.
2021-12-06 22:13:14 +01:00
std::shared_ptr<TextureView> Texture::GetView(vk::ImageViewType type, vk::ImageSubresourceRange range, texture::Format pFormat, vk::ComponentMapping mapping) {
for (const auto &viewWeak : views) {
auto view{viewWeak.lock()};
if (view && type == view->type && pFormat == view->format && range == view->range && mapping == view->mapping)
return view;
}
auto view{std::make_shared<TextureView>(shared_from_this(), type, range, pFormat, mapping)};
views.push_back(view);
return view;
}
Initial Texture Manager Implementation + Maxwell3D Render Target Implement the groundwork for the texture manager to be able to report basic overlaps and be extended to support more in the future. The Maxwell3D registers `RenderTargetControl`, `RenderTarget` and a stub for `ClearBuffers` were implemented. A lot of changes were also made to `GuestTexture`/`Texture` for supporting mipmapping and multiple array layers alongside significant architectural changes to `GuestTexture` effectively disconnecting it from `Texture` with it no longer being a parent rather an object that can be used to create a `Texture` object. Note: Support for fragmented CPU mappings hasn't been added for texture synchronization yet
2021-08-20 22:17:13 +02:00
void Texture::CopyFrom(std::shared_ptr<Texture> source, const vk::ImageSubresourceRange &subresource) {
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
WaitOnBacking();
Redesign Texture Class + Improve Presentation Engine This commit reworks the `Texture` class to include a Vulkan Image backing that can be optionally owning or non-owning and swapped in with consideration for Vulkan image layout, it also adds CPU-sided synchronization for the texture objects with FenceCycle. It also makes the appropriate changes to `PresentationEngine` and `GraphicBufferProducer` to work with the new `Texture` class while setting the groundwork for supporting swapchain recreation. It also fixes a log in `IpcResponse` and improves the display mode selection algorithm by further weighing refresh rate.
2021-05-22 19:46:28 +02:00
WaitOnFence();
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
source->WaitOnBacking();
source->WaitOnFence();
if (source->layout == vk::ImageLayout::eUndefined)
throw exception("Cannot copy from image with undefined layout");
else if (source->dimensions != dimensions)
throw exception("Cannot copy from image with different dimensions");
else if (source->format != format)
throw exception("Cannot copy from image with different format");
Introduce Texture Always Sync + Wait on GPU Execution + More RT Formats Infrastructure for always syncing textures has been introduced now, they will be synced prior to and after every execution. This does considerably reduce the performance alongside waiting on GPU execution to finish but it will be partially recouped once conditional syncing is performed.
2021-10-06 10:50:54 +02:00
TRACE_EVENT("gpu", "Texture::CopyFrom");
Subpass Support + More RT Formats + Fix `FenceCycle` Cyclic Dependencies Support for subpasses was added by reworking attachment reuse code to account for preserved attachments and subpass dependencies. A lot of RT formats were also added to allow SMO to boot up entirely, it should be noted that it doesn't render anything. `FenceCycle` had a cyclic dependency which broke clean exit, we now utilize `std::weak_ptr<FenceCycle>` inside the `Texture` object. A minor fix for broken stack traces was also made caused by supplying a `nullptr` C-string to libfmt when a symbol was unresolved which caused an `abort` due to invocation of `strlen` with it.
2021-09-22 05:30:11 +02:00
auto lCycle{gpu.scheduler.Submit([&](vk::raii::CommandBuffer &commandBuffer) {
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
auto sourceBacking{source->GetBacking()};
if (source->layout != vk::ImageLayout::eTransferSrcOptimal) {
commandBuffer.pipelineBarrier(vk::PipelineStageFlagBits::eTopOfPipe, vk::PipelineStageFlagBits::eTransfer, {}, {}, {}, vk::ImageMemoryBarrier{
.image = sourceBacking,
.srcAccessMask = vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite,
.dstAccessMask = vk::AccessFlagBits::eTransferRead,
.oldLayout = source->layout,
.newLayout = vk::ImageLayout::eTransferSrcOptimal,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
Initial Texture Manager Implementation + Maxwell3D Render Target Implement the groundwork for the texture manager to be able to report basic overlaps and be extended to support more in the future. The Maxwell3D registers `RenderTargetControl`, `RenderTarget` and a stub for `ClearBuffers` were implemented. A lot of changes were also made to `GuestTexture`/`Texture` for supporting mipmapping and multiple array layers alongside significant architectural changes to `GuestTexture` effectively disconnecting it from `Texture` with it no longer being a parent rather an object that can be used to create a `Texture` object. Note: Support for fragmented CPU mappings hasn't been added for texture synchronization yet
2021-08-20 22:17:13 +02:00
.subresourceRange = subresource,
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
});
}
auto destinationBacking{GetBacking()};
Redesign Texture Class + Improve Presentation Engine This commit reworks the `Texture` class to include a Vulkan Image backing that can be optionally owning or non-owning and swapped in with consideration for Vulkan image layout, it also adds CPU-sided synchronization for the texture objects with FenceCycle. It also makes the appropriate changes to `PresentationEngine` and `GraphicBufferProducer` to work with the new `Texture` class while setting the groundwork for supporting swapchain recreation. It also fixes a log in `IpcResponse` and improves the display mode selection algorithm by further weighing refresh rate.
2021-05-22 19:46:28 +02:00
if (layout != vk::ImageLayout::eTransferDstOptimal) {
commandBuffer.pipelineBarrier(layout != vk::ImageLayout::eUndefined ? vk::PipelineStageFlagBits::eTopOfPipe : vk::PipelineStageFlagBits::eBottomOfPipe, vk::PipelineStageFlagBits::eTransfer, {}, {}, {}, vk::ImageMemoryBarrier{
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
.image = destinationBacking,
Redesign Texture Class + Improve Presentation Engine This commit reworks the `Texture` class to include a Vulkan Image backing that can be optionally owning or non-owning and swapped in with consideration for Vulkan image layout, it also adds CPU-sided synchronization for the texture objects with FenceCycle. It also makes the appropriate changes to `PresentationEngine` and `GraphicBufferProducer` to work with the new `Texture` class while setting the groundwork for supporting swapchain recreation. It also fixes a log in `IpcResponse` and improves the display mode selection algorithm by further weighing refresh rate.
2021-05-22 19:46:28 +02:00
.srcAccessMask = vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite,
.dstAccessMask = vk::AccessFlagBits::eTransferWrite,
.oldLayout = layout,
.newLayout = vk::ImageLayout::eTransferDstOptimal,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
Initial Texture Manager Implementation + Maxwell3D Render Target Implement the groundwork for the texture manager to be able to report basic overlaps and be extended to support more in the future. The Maxwell3D registers `RenderTargetControl`, `RenderTarget` and a stub for `ClearBuffers` were implemented. A lot of changes were also made to `GuestTexture`/`Texture` for supporting mipmapping and multiple array layers alongside significant architectural changes to `GuestTexture` effectively disconnecting it from `Texture` with it no longer being a parent rather an object that can be used to create a `Texture` object. Note: Support for fragmented CPU mappings hasn't been added for texture synchronization yet
2021-08-20 22:17:13 +02:00
.subresourceRange = subresource,
Redesign Texture Class + Improve Presentation Engine This commit reworks the `Texture` class to include a Vulkan Image backing that can be optionally owning or non-owning and swapped in with consideration for Vulkan image layout, it also adds CPU-sided synchronization for the texture objects with FenceCycle. It also makes the appropriate changes to `PresentationEngine` and `GraphicBufferProducer` to work with the new `Texture` class while setting the groundwork for supporting swapchain recreation. It also fixes a log in `IpcResponse` and improves the display mode selection algorithm by further weighing refresh rate.
2021-05-22 19:46:28 +02:00
});
if (layout == vk::ImageLayout::eUndefined)
layout = vk::ImageLayout::eTransferDstOptimal;
}
Initial Texture Manager Implementation + Maxwell3D Render Target Implement the groundwork for the texture manager to be able to report basic overlaps and be extended to support more in the future. The Maxwell3D registers `RenderTargetControl`, `RenderTarget` and a stub for `ClearBuffers` were implemented. A lot of changes were also made to `GuestTexture`/`Texture` for supporting mipmapping and multiple array layers alongside significant architectural changes to `GuestTexture` effectively disconnecting it from `Texture` with it no longer being a parent rather an object that can be used to create a `Texture` object. Note: Support for fragmented CPU mappings hasn't been added for texture synchronization yet
2021-08-20 22:17:13 +02:00
vk::ImageSubresourceLayers subresourceLayers{
.aspectMask = subresource.aspectMask,
.mipLevel = subresource.baseMipLevel,
.baseArrayLayer = subresource.baseArrayLayer,
.layerCount = subresource.layerCount == VK_REMAINING_ARRAY_LAYERS ? layerCount - subresource.baseArrayLayer : subresource.layerCount,
};
for (; subresourceLayers.mipLevel < (subresource.levelCount == VK_REMAINING_MIP_LEVELS ? mipLevels - subresource.baseMipLevel : subresource.levelCount); subresourceLayers.mipLevel++)
commandBuffer.copyImage(sourceBacking, vk::ImageLayout::eTransferSrcOptimal, destinationBacking, vk::ImageLayout::eTransferDstOptimal, vk::ImageCopy{
.srcSubresource = subresourceLayers,
.dstSubresource = subresourceLayers,
.extent = dimensions,
});
Redesign Texture Class + Improve Presentation Engine This commit reworks the `Texture` class to include a Vulkan Image backing that can be optionally owning or non-owning and swapped in with consideration for Vulkan image layout, it also adds CPU-sided synchronization for the texture objects with FenceCycle. It also makes the appropriate changes to `PresentationEngine` and `GraphicBufferProducer` to work with the new `Texture` class while setting the groundwork for supporting swapchain recreation. It also fixes a log in `IpcResponse` and improves the display mode selection algorithm by further weighing refresh rate.
2021-05-22 19:46:28 +02:00
if (layout != vk::ImageLayout::eTransferDstOptimal)
commandBuffer.pipelineBarrier(vk::PipelineStageFlagBits::eTransfer, vk::PipelineStageFlagBits::eTransfer, {}, {}, {}, vk::ImageMemoryBarrier{
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
.image = destinationBacking,
Redesign Texture Class + Improve Presentation Engine This commit reworks the `Texture` class to include a Vulkan Image backing that can be optionally owning or non-owning and swapped in with consideration for Vulkan image layout, it also adds CPU-sided synchronization for the texture objects with FenceCycle. It also makes the appropriate changes to `PresentationEngine` and `GraphicBufferProducer` to work with the new `Texture` class while setting the groundwork for supporting swapchain recreation. It also fixes a log in `IpcResponse` and improves the display mode selection algorithm by further weighing refresh rate.
2021-05-22 19:46:28 +02:00
.srcAccessMask = vk::AccessFlagBits::eTransferWrite,
.dstAccessMask = vk::AccessFlagBits::eMemoryRead,
.oldLayout = vk::ImageLayout::eTransferDstOptimal,
.newLayout = layout,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
Initial Texture Manager Implementation + Maxwell3D Render Target Implement the groundwork for the texture manager to be able to report basic overlaps and be extended to support more in the future. The Maxwell3D registers `RenderTargetControl`, `RenderTarget` and a stub for `ClearBuffers` were implemented. A lot of changes were also made to `GuestTexture`/`Texture` for supporting mipmapping and multiple array layers alongside significant architectural changes to `GuestTexture` effectively disconnecting it from `Texture` with it no longer being a parent rather an object that can be used to create a `Texture` object. Note: Support for fragmented CPU mappings hasn't been added for texture synchronization yet
2021-08-20 22:17:13 +02:00
.subresourceRange = subresource,
Redesign Texture Class + Improve Presentation Engine This commit reworks the `Texture` class to include a Vulkan Image backing that can be optionally owning or non-owning and swapped in with consideration for Vulkan image layout, it also adds CPU-sided synchronization for the texture objects with FenceCycle. It also makes the appropriate changes to `PresentationEngine` and `GraphicBufferProducer` to work with the new `Texture` class while setting the groundwork for supporting swapchain recreation. It also fixes a log in `IpcResponse` and improves the display mode selection algorithm by further weighing refresh rate.
2021-05-22 19:46:28 +02:00
});
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
if (layout != vk::ImageLayout::eTransferSrcOptimal)
commandBuffer.pipelineBarrier(vk::PipelineStageFlagBits::eTransfer, vk::PipelineStageFlagBits::eTransfer, {}, {}, {}, vk::ImageMemoryBarrier{
.image = sourceBacking,
.srcAccessMask = vk::AccessFlagBits::eTransferRead,
.dstAccessMask = vk::AccessFlagBits::eMemoryWrite,
.oldLayout = vk::ImageLayout::eTransferSrcOptimal,
.newLayout = source->layout,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
Initial Texture Manager Implementation + Maxwell3D Render Target Implement the groundwork for the texture manager to be able to report basic overlaps and be extended to support more in the future. The Maxwell3D registers `RenderTargetControl`, `RenderTarget` and a stub for `ClearBuffers` were implemented. A lot of changes were also made to `GuestTexture`/`Texture` for supporting mipmapping and multiple array layers alongside significant architectural changes to `GuestTexture` effectively disconnecting it from `Texture` with it no longer being a parent rather an object that can be used to create a `Texture` object. Note: Support for fragmented CPU mappings hasn't been added for texture synchronization yet
2021-08-20 22:17:13 +02:00
.subresourceRange = subresource,
Separate Guest and Host Presentation + AChoreographer V-Sync Event We had issues when combining host and guest presentation since certain configurations in guest presentation such as double buffering were very unoptimal for the host and would significantly affect the FPS. As a result of this, we've now made host presentation have its own presentation textures which are copied into from the guest at presentation time, allowing us to change parameters of the host presentation independently of the guest. We've implemented the infrastructure for this which includes being able to create images from host GPU memory using VMA, an optimized linear texture sync and a method to do on-GPU texture-to-texture copies. We've also moved to driving the V-Sync event using AChoreographer on its on thread in this PR, which more accurately encapsulates HOS behavior and allows games such as ARMS to boot as they depend on the V-Sync event being signalled even when the game isn't presenting.
2021-06-18 12:55:19 +02:00
});
Subpass Support + More RT Formats + Fix `FenceCycle` Cyclic Dependencies Support for subpasses was added by reworking attachment reuse code to account for preserved attachments and subpass dependencies. A lot of RT formats were also added to allow SMO to boot up entirely, it should be noted that it doesn't render anything. `FenceCycle` had a cyclic dependency which broke clean exit, we now utilize `std::weak_ptr<FenceCycle>` inside the `Texture` object. A minor fix for broken stack traces was also made caused by supplying a `nullptr` C-string to libfmt when a symbol was unresolved which caused an `abort` due to invocation of `strlen` with it.
2021-09-22 05:30:11 +02:00
})};
lCycle->AttachObjects(std::move(source), shared_from_this());
cycle = lCycle;
Initial Texture Manager Implementation + Maxwell3D Render Target Implement the groundwork for the texture manager to be able to report basic overlaps and be extended to support more in the future. The Maxwell3D registers `RenderTargetControl`, `RenderTarget` and a stub for `ClearBuffers` were implemented. A lot of changes were also made to `GuestTexture`/`Texture` for supporting mipmapping and multiple array layers alongside significant architectural changes to `GuestTexture` effectively disconnecting it from `Texture` with it no longer being a parent rather an object that can be used to create a `Texture` object. Note: Support for fragmented CPU mappings hasn't been added for texture synchronization yet
2021-08-20 22:17:13 +02:00
}
Redesign Texture Class + Improve Presentation Engine This commit reworks the `Texture` class to include a Vulkan Image backing that can be optionally owning or non-owning and swapped in with consideration for Vulkan image layout, it also adds CPU-sided synchronization for the texture objects with FenceCycle. It also makes the appropriate changes to `PresentationEngine` and `GraphicBufferProducer` to work with the new `Texture` class while setting the groundwork for supporting swapchain recreation. It also fixes a log in `IpcResponse` and improves the display mode selection algorithm by further weighing refresh rate.
2021-05-22 19:46:28 +02:00
}
Reference in New Issue Copy Permalink