mirror of
https://github.com/shadps4-emu/shadPS4.git
synced 2025-12-10 05:38:49 +00:00
video_core: Rework tile manager (#3374)
* video_core: Rework detiling * video_core: Support tiling and macrotile detiling * clang format * image_info: Cleanups * resource: Revert some changes * texture_cache: Fix small error * image_info: Set depth flag on depth promote * buffer_cache: Remove level check * tile_manager: Handle case of staging buffer causing flush * image_info: Add 2D thick array mode * image_info: Add slices to mip size * tile_manager: Set bank swizzle * buffer_cache: Support image copies from DmaData * vk_rasterizer: Accelerate trivial render target copies with compute Before tiling PR compute image copies were done with the following sequence vkCmdCopyImageToBuffer (in SynchronizeBufferFromImage) -> vkCmdDispatch (copy) -> vkCmdCopyBufferToImage (in RefreshImage) With the tiling PR it added extra tiling/detiling steps vkCmdCopyImageToBuffer -> vkCmdDispatch (tiling) -> vkCmdDispatch (copy) -> vkCmdDispatch (detiling) -> vkCmdCopyBufferToImage This is quite a bit of overhead for a simple image copy. This commit tries to detect trivial image copies i.e cs shaders that copy the full source image to all of the destination. So now all this sequence is just a vkCmdCopyImage. How much it triggers depends on the guest * texture_cache: Fix build * image: Copy all subresources with buffer too
This commit is contained in:
TheTurtle
GitHub
@@ -396,7 +396,9 @@ void BufferCache::CopyBuffer(VAddr dst, VAddr src, u32 num_bytes, bool dst_gds,
|
||||
// Avoid using ObtainBuffer here as that might give us the stream buffer.
|
||||
const BufferId buffer_id = FindBuffer(src, num_bytes);
|
||||
auto& buffer = slot_buffers[buffer_id];
|
||||
SynchronizeBuffer(buffer, src, num_bytes, false, false);
|
||||
if (SynchronizeBuffer(buffer, src, num_bytes, false, true)) {
|
||||
texture_cache.InvalidateMemoryFromGPU(dst, num_bytes);
|
||||
}
|
||||
return buffer;
|
||||
}();
|
||||
auto& dst_buffer = [&] -> const Buffer& {
|
||||
@@ -854,7 +856,7 @@ void BufferCache::ChangeRegister(BufferId buffer_id) {
|
||||
}
|
||||
}
|
||||
|
||||
void BufferCache::SynchronizeBuffer(Buffer& buffer, VAddr device_addr, u32 size, bool is_written,
|
||||
bool BufferCache::SynchronizeBuffer(Buffer& buffer, VAddr device_addr, u32 size, bool is_written,
|
||||
bool is_texel_buffer) {
|
||||
boost::container::small_vector<vk::BufferCopy, 4> copies;
|
||||
size_t total_size_bytes = 0;
|
||||
@@ -867,47 +869,47 @@ void BufferCache::SynchronizeBuffer(Buffer& buffer, VAddr device_addr, u32 size,
|
||||
total_size_bytes += range_size;
|
||||
},
|
||||
[&] { src_buffer = UploadCopies(buffer, copies, total_size_bytes); });
|
||||
SCOPE_EXIT {
|
||||
if (is_texel_buffer) {
|
||||
SynchronizeBufferFromImage(buffer, device_addr, size);
|
||||
}
|
||||
};
|
||||
if (!src_buffer) {
|
||||
return;
|
||||
|
||||
if (src_buffer) {
|
||||
scheduler.EndRendering();
|
||||
const auto cmdbuf = scheduler.CommandBuffer();
|
||||
const vk::BufferMemoryBarrier2 pre_barrier = {
|
||||
.srcStageMask = vk::PipelineStageFlagBits2::eAllCommands,
|
||||
.srcAccessMask = vk::AccessFlagBits2::eMemoryRead | vk::AccessFlagBits2::eMemoryWrite |
|
||||
vk::AccessFlagBits2::eTransferRead |
|
||||
vk::AccessFlagBits2::eTransferWrite,
|
||||
.dstStageMask = vk::PipelineStageFlagBits2::eTransfer,
|
||||
.dstAccessMask = vk::AccessFlagBits2::eTransferWrite,
|
||||
.buffer = buffer.Handle(),
|
||||
.offset = 0,
|
||||
.size = buffer.SizeBytes(),
|
||||
};
|
||||
const vk::BufferMemoryBarrier2 post_barrier = {
|
||||
.srcStageMask = vk::PipelineStageFlagBits2::eTransfer,
|
||||
.srcAccessMask = vk::AccessFlagBits2::eTransferWrite,
|
||||
.dstStageMask = vk::PipelineStageFlagBits2::eAllCommands,
|
||||
.dstAccessMask = vk::AccessFlagBits2::eMemoryRead | vk::AccessFlagBits2::eMemoryWrite,
|
||||
.buffer = buffer.Handle(),
|
||||
.offset = 0,
|
||||
.size = buffer.SizeBytes(),
|
||||
};
|
||||
cmdbuf.pipelineBarrier2(vk::DependencyInfo{
|
||||
.dependencyFlags = vk::DependencyFlagBits::eByRegion,
|
||||
.bufferMemoryBarrierCount = 1,
|
||||
.pBufferMemoryBarriers = &pre_barrier,
|
||||
});
|
||||
cmdbuf.copyBuffer(src_buffer, buffer.buffer, copies);
|
||||
cmdbuf.pipelineBarrier2(vk::DependencyInfo{
|
||||
.dependencyFlags = vk::DependencyFlagBits::eByRegion,
|
||||
.bufferMemoryBarrierCount = 1,
|
||||
.pBufferMemoryBarriers = &post_barrier,
|
||||
});
|
||||
TouchBuffer(buffer);
|
||||
}
|
||||
scheduler.EndRendering();
|
||||
const auto cmdbuf = scheduler.CommandBuffer();
|
||||
const vk::BufferMemoryBarrier2 pre_barrier = {
|
||||
.srcStageMask = vk::PipelineStageFlagBits2::eAllCommands,
|
||||
.srcAccessMask = vk::AccessFlagBits2::eMemoryRead | vk::AccessFlagBits2::eMemoryWrite |
|
||||
vk::AccessFlagBits2::eTransferRead | vk::AccessFlagBits2::eTransferWrite,
|
||||
.dstStageMask = vk::PipelineStageFlagBits2::eTransfer,
|
||||
.dstAccessMask = vk::AccessFlagBits2::eTransferWrite,
|
||||
.buffer = buffer.Handle(),
|
||||
.offset = 0,
|
||||
.size = buffer.SizeBytes(),
|
||||
};
|
||||
const vk::BufferMemoryBarrier2 post_barrier = {
|
||||
.srcStageMask = vk::PipelineStageFlagBits2::eTransfer,
|
||||
.srcAccessMask = vk::AccessFlagBits2::eTransferWrite,
|
||||
.dstStageMask = vk::PipelineStageFlagBits2::eAllCommands,
|
||||
.dstAccessMask = vk::AccessFlagBits2::eMemoryRead | vk::AccessFlagBits2::eMemoryWrite,
|
||||
.buffer = buffer.Handle(),
|
||||
.offset = 0,
|
||||
.size = buffer.SizeBytes(),
|
||||
};
|
||||
cmdbuf.pipelineBarrier2(vk::DependencyInfo{
|
||||
.dependencyFlags = vk::DependencyFlagBits::eByRegion,
|
||||
.bufferMemoryBarrierCount = 1,
|
||||
.pBufferMemoryBarriers = &pre_barrier,
|
||||
});
|
||||
cmdbuf.copyBuffer(src_buffer, buffer.buffer, copies);
|
||||
cmdbuf.pipelineBarrier2(vk::DependencyInfo{
|
||||
.dependencyFlags = vk::DependencyFlagBits::eByRegion,
|
||||
.bufferMemoryBarrierCount = 1,
|
||||
.pBufferMemoryBarriers = &post_barrier,
|
||||
});
|
||||
TouchBuffer(buffer);
|
||||
if (is_texel_buffer) {
|
||||
return SynchronizeBufferFromImage(buffer, device_addr, size);
|
||||
}
|
||||
return false;
|
||||
}
|
||||
|
||||
vk::Buffer BufferCache::UploadCopies(Buffer& buffer, std::span<vk::BufferCopy> copies,
|
||||
@@ -944,115 +946,81 @@ vk::Buffer BufferCache::UploadCopies(Buffer& buffer, std::span<vk::BufferCopy> c
|
||||
}
|
||||
|
||||
bool BufferCache::SynchronizeBufferFromImage(Buffer& buffer, VAddr device_addr, u32 size) {
|
||||
boost::container::small_vector<ImageId, 6> image_ids;
|
||||
texture_cache.ForEachImageInRegion(device_addr, size, [&](ImageId image_id, Image& image) {
|
||||
if (image.info.guest_address != device_addr) {
|
||||
return;
|
||||
}
|
||||
// Only perform sync if image is:
|
||||
// - GPU modified; otherwise there are no changes to synchronize.
|
||||
// - Not CPU dirty; otherwise we could overwrite CPU changes with stale GPU changes.
|
||||
// - Not GPU dirty; otherwise we could overwrite GPU changes with stale image data.
|
||||
if (False(image.flags & ImageFlagBits::GpuModified) ||
|
||||
True(image.flags & ImageFlagBits::Dirty)) {
|
||||
return;
|
||||
}
|
||||
image_ids.push_back(image_id);
|
||||
});
|
||||
if (image_ids.empty()) {
|
||||
const ImageId image_id = texture_cache.FindImageFromRange(device_addr, size);
|
||||
if (!image_id) {
|
||||
return false;
|
||||
}
|
||||
ImageId image_id{};
|
||||
if (image_ids.size() == 1) {
|
||||
// Sometimes image size might not exactly match with requested buffer size
|
||||
// If we only found 1 candidate image use it without too many questions.
|
||||
image_id = image_ids[0];
|
||||
} else {
|
||||
for (s32 i = 0; i < image_ids.size(); ++i) {
|
||||
Image& image = texture_cache.GetImage(image_ids[i]);
|
||||
if (image.info.guest_size == size) {
|
||||
image_id = image_ids[i];
|
||||
break;
|
||||
}
|
||||
}
|
||||
if (!image_id) {
|
||||
LOG_WARNING(Render_Vulkan,
|
||||
"Failed to find exact image match for copy addr={:#x}, size={:#x}",
|
||||
device_addr, size);
|
||||
return false;
|
||||
}
|
||||
}
|
||||
Image& image = texture_cache.GetImage(image_id);
|
||||
ASSERT_MSG(device_addr == image.info.guest_address,
|
||||
"Texel buffer aliases image subresources {:x} : {:x}", device_addr,
|
||||
image.info.guest_address);
|
||||
boost::container::small_vector<vk::BufferImageCopy, 8> copies;
|
||||
u32 offset = buffer.Offset(image.info.guest_address);
|
||||
const u32 num_layers = image.info.resources.layers;
|
||||
const u32 max_offset = offset + size;
|
||||
for (u32 m = 0; m < image.info.resources.levels; m++) {
|
||||
const u32 width = std::max(image.info.size.width >> m, 1u);
|
||||
const u32 height = std::max(image.info.size.height >> m, 1u);
|
||||
const u32 depth =
|
||||
image.info.props.is_volume ? std::max(image.info.size.depth >> m, 1u) : 1u;
|
||||
const auto [mip_size, mip_pitch, mip_height, mip_ofs] = image.info.mips_layout[m];
|
||||
offset += mip_ofs;
|
||||
if (offset + mip_size > max_offset) {
|
||||
const u32 buf_offset = buffer.Offset(image.info.guest_address);
|
||||
boost::container::small_vector<vk::BufferImageCopy, 8> buffer_copies;
|
||||
u32 copy_size = 0;
|
||||
for (u32 mip = 0; mip < image.info.resources.levels; mip++) {
|
||||
const auto& mip_info = image.info.mips_layout[mip];
|
||||
const u32 width = std::max(image.info.size.width >> mip, 1u);
|
||||
const u32 height = std::max(image.info.size.height >> mip, 1u);
|
||||
const u32 depth = std::max(image.info.size.depth >> mip, 1u);
|
||||
if (buf_offset + mip_info.offset + mip_info.size > buffer.SizeBytes()) {
|
||||
break;
|
||||
}
|
||||
copies.push_back({
|
||||
.bufferOffset = offset,
|
||||
.bufferRowLength = mip_pitch,
|
||||
.bufferImageHeight = mip_height,
|
||||
buffer_copies.push_back(vk::BufferImageCopy{
|
||||
.bufferOffset = mip_info.offset,
|
||||
.bufferRowLength = mip_info.pitch,
|
||||
.bufferImageHeight = mip_info.height,
|
||||
.imageSubresource{
|
||||
.aspectMask = image.aspect_mask & ~vk::ImageAspectFlagBits::eStencil,
|
||||
.mipLevel = m,
|
||||
.mipLevel = mip,
|
||||
.baseArrayLayer = 0,
|
||||
.layerCount = num_layers,
|
||||
.layerCount = image.info.resources.layers,
|
||||
},
|
||||
.imageOffset = {0, 0, 0},
|
||||
.imageExtent = {width, height, depth},
|
||||
});
|
||||
copy_size += mip_info.size;
|
||||
}
|
||||
if (!copies.empty()) {
|
||||
scheduler.EndRendering();
|
||||
const vk::BufferMemoryBarrier2 pre_barrier = {
|
||||
.srcStageMask = vk::PipelineStageFlagBits2::eAllCommands,
|
||||
.srcAccessMask = vk::AccessFlagBits2::eMemoryRead,
|
||||
.dstStageMask = vk::PipelineStageFlagBits2::eTransfer,
|
||||
.dstAccessMask = vk::AccessFlagBits2::eTransferWrite,
|
||||
.buffer = buffer.Handle(),
|
||||
.offset = max_offset - size,
|
||||
.size = size,
|
||||
};
|
||||
const vk::BufferMemoryBarrier2 post_barrier = {
|
||||
.srcStageMask = vk::PipelineStageFlagBits2::eTransfer,
|
||||
.srcAccessMask = vk::AccessFlagBits2::eTransferWrite,
|
||||
.dstStageMask = vk::PipelineStageFlagBits2::eAllCommands,
|
||||
.dstAccessMask = vk::AccessFlagBits2::eMemoryRead,
|
||||
.buffer = buffer.Handle(),
|
||||
.offset = max_offset - size,
|
||||
.size = size,
|
||||
};
|
||||
auto barriers = image.GetBarriers(vk::ImageLayout::eTransferSrcOptimal,
|
||||
vk::AccessFlagBits2::eTransferRead,
|
||||
vk::PipelineStageFlagBits2::eTransfer, {});
|
||||
const auto cmdbuf = scheduler.CommandBuffer();
|
||||
cmdbuf.pipelineBarrier2(vk::DependencyInfo{
|
||||
.dependencyFlags = vk::DependencyFlagBits::eByRegion,
|
||||
.bufferMemoryBarrierCount = 1,
|
||||
.pBufferMemoryBarriers = &pre_barrier,
|
||||
.imageMemoryBarrierCount = static_cast<u32>(barriers.size()),
|
||||
.pImageMemoryBarriers = barriers.data(),
|
||||
});
|
||||
cmdbuf.copyImageToBuffer(image.image, vk::ImageLayout::eTransferSrcOptimal, buffer.Handle(),
|
||||
copies);
|
||||
cmdbuf.pipelineBarrier2(vk::DependencyInfo{
|
||||
.dependencyFlags = vk::DependencyFlagBits::eByRegion,
|
||||
.bufferMemoryBarrierCount = 1,
|
||||
.pBufferMemoryBarriers = &post_barrier,
|
||||
});
|
||||
if (copy_size == 0) {
|
||||
return false;
|
||||
}
|
||||
scheduler.EndRendering();
|
||||
const vk::BufferMemoryBarrier2 pre_barrier = {
|
||||
.srcStageMask = vk::PipelineStageFlagBits2::eAllCommands,
|
||||
.srcAccessMask = vk::AccessFlagBits2::eMemoryRead,
|
||||
.dstStageMask = vk::PipelineStageFlagBits2::eTransfer,
|
||||
.dstAccessMask = vk::AccessFlagBits2::eTransferWrite,
|
||||
.buffer = buffer.Handle(),
|
||||
.offset = buf_offset,
|
||||
.size = copy_size,
|
||||
};
|
||||
const vk::BufferMemoryBarrier2 post_barrier = {
|
||||
.srcStageMask = vk::PipelineStageFlagBits2::eTransfer,
|
||||
.srcAccessMask = vk::AccessFlagBits2::eTransferWrite,
|
||||
.dstStageMask = vk::PipelineStageFlagBits2::eAllCommands,
|
||||
.dstAccessMask = vk::AccessFlagBits2::eMemoryRead,
|
||||
.buffer = buffer.Handle(),
|
||||
.offset = buf_offset,
|
||||
.size = copy_size,
|
||||
};
|
||||
auto barriers =
|
||||
image.GetBarriers(vk::ImageLayout::eTransferSrcOptimal, vk::AccessFlagBits2::eTransferRead,
|
||||
vk::PipelineStageFlagBits2::eTransfer, {});
|
||||
auto cmdbuf = scheduler.CommandBuffer();
|
||||
cmdbuf.pipelineBarrier2(vk::DependencyInfo{
|
||||
.dependencyFlags = vk::DependencyFlagBits::eByRegion,
|
||||
.bufferMemoryBarrierCount = 1,
|
||||
.pBufferMemoryBarriers = &pre_barrier,
|
||||
.imageMemoryBarrierCount = static_cast<u32>(barriers.size()),
|
||||
.pImageMemoryBarriers = barriers.data(),
|
||||
});
|
||||
auto& tile_manager = texture_cache.GetTileManager();
|
||||
tile_manager.TileImage(image.image, buffer_copies, buffer.Handle(), buf_offset, image.info);
|
||||
cmdbuf = scheduler.CommandBuffer();
|
||||
cmdbuf.pipelineBarrier2(vk::DependencyInfo{
|
||||
.dependencyFlags = vk::DependencyFlagBits::eByRegion,
|
||||
.bufferMemoryBarrierCount = 1,
|
||||
.pBufferMemoryBarriers = &post_barrier,
|
||||
});
|
||||
return true;
|
||||
}
|
||||
|
||||
|
||||
@@ -102,15 +102,14 @@ public:
|
||||
|
||||
/// Retrieves a utility buffer optimized for specified memory usage.
|
||||
StreamBuffer& GetUtilityBuffer(MemoryUsage usage) noexcept {
|
||||
switch (usage) {
|
||||
case MemoryUsage::Stream:
|
||||
if (usage == MemoryUsage::Stream) {
|
||||
return stream_buffer;
|
||||
case MemoryUsage::Download:
|
||||
} else if (usage == MemoryUsage::Download) {
|
||||
return download_buffer;
|
||||
case MemoryUsage::Upload:
|
||||
return staging_buffer;
|
||||
case MemoryUsage::DeviceLocal:
|
||||
} else if (usage == MemoryUsage::DeviceLocal) {
|
||||
return device_buffer;
|
||||
} else {
|
||||
return staging_buffer;
|
||||
}
|
||||
}
|
||||
|
||||
@@ -200,7 +199,7 @@ private:
|
||||
template <bool insert>
|
||||
void ChangeRegister(BufferId buffer_id);
|
||||
|
||||
void SynchronizeBuffer(Buffer& buffer, VAddr device_addr, u32 size, bool is_written,
|
||||
bool SynchronizeBuffer(Buffer& buffer, VAddr device_addr, u32 size, bool is_written,
|
||||
bool is_texel_buffer);
|
||||
|
||||
vk::Buffer UploadCopies(Buffer& buffer, std::span<vk::BufferCopy> copies,
|
||||
|
||||
Reference in New Issue
Block a user