diff options
author | tcmal <me@aria.rip> | 2024-08-25 17:44:22 +0100 |
---|---|---|
committer | tcmal <me@aria.rip> | 2024-08-25 17:44:22 +0100 |
commit | c48b54f3fb7bbe9046915eb99eca02fa84dc55c9 (patch) | |
tree | 752831451d2bd3a658485df724a01ae39e80fae3 /stockton-render/src/draw/texture/load.rs | |
parent | b437109ebf4da243fd643f0a31546d0d0155b0a4 (diff) |
feat(render): multithreaded texture loading
also a bunch of supporting changes
Diffstat (limited to 'stockton-render/src/draw/texture/load.rs')
-rw-r--r-- | stockton-render/src/draw/texture/load.rs | 352 |
1 files changed, 352 insertions, 0 deletions
diff --git a/stockton-render/src/draw/texture/load.rs b/stockton-render/src/draw/texture/load.rs new file mode 100644 index 0000000..011ae29 --- /dev/null +++ b/stockton-render/src/draw/texture/load.rs @@ -0,0 +1,352 @@ +use super::{ + block::LoadedImage, block::TexturesBlock, loader::TextureLoader, repo::BLOCK_SIZE, + resolver::TextureResolver, staging_buffer::StagingBuffer, LoadableImage, PIXEL_SIZE, +}; +use crate::types::*; +use stockton_levels::prelude::*; + +use arrayvec::ArrayVec; +use hal::{ + command::{BufferImageCopy, CommandBufferFlags}, + format::{Aspects, Format, Swizzle}, + image::{ + Extent, Filter, Layout, Offset, SamplerDesc, SubresourceLayers, SubresourceRange, + Usage as ImgUsage, ViewKind, WrapMode, + }, + memory::{Barrier, Dependencies}, + prelude::*, + pso::{Descriptor, DescriptorSetWrite, PipelineStage, ShaderStageFlags}, + queue::Submission, +}; +use rendy_descriptor::{DescriptorRanges, DescriptorSetLayoutBinding, DescriptorType}; +use rendy_memory::{Allocator, Block}; +use std::mem::ManuallyDrop; + +pub struct QueuedLoad<B: Block<back::Backend>> { + pub fence: Fence, + pub buf: CommandBuffer, + pub block: TexturesBlock<B>, + pub staging_bufs: ArrayVec<[StagingBuffer; BLOCK_SIZE]>, +} + +impl<B: Block<back::Backend>> QueuedLoad<B> { + pub fn dissolve( + self, + ) -> ( + (Fence, CommandBuffer), + ArrayVec<[StagingBuffer; BLOCK_SIZE]>, + TexturesBlock<B>, + ) { + ((self.fence, self.buf), self.staging_bufs, self.block) + } +} + +impl<'a, T: HasTextures, R: TextureResolver<I>, I: LoadableImage> TextureLoader<'a, T, R, I> { + const FORMAT: Format = Format::Rgba8Srgb; + const RESOURCES: SubresourceRange = SubresourceRange { + aspects: Aspects::COLOR, + levels: 0..1, + layers: 0..1, + }; + const LAYERS: SubresourceLayers = SubresourceLayers { + aspects: Aspects::COLOR, + level: 0, + layers: 0..1, + }; + + pub(crate) unsafe fn attempt_queue_load( + &mut self, + block_ref: usize, + ) -> Option<QueuedLoad<DynamicBlock>> { + // Get assets to use + let (fence, mut buf) = self.buffers.pop_front()?; + + // Create descriptor set + let descriptor_set = self.create_descriptor_set(); + + // Get a command buffer + buf.begin_primary(CommandBufferFlags::ONE_TIME_SUBMIT); + + let mut copy_cmds: ArrayVec<[_; BLOCK_SIZE]> = ArrayVec::new(); + let mut imgs: ArrayVec<[_; BLOCK_SIZE]> = ArrayVec::new(); + let mut staging_bufs: ArrayVec<[_; BLOCK_SIZE]> = ArrayVec::new(); + + // For each texture in block + for tex_idx in (block_ref * BLOCK_SIZE)..(block_ref + 1) * BLOCK_SIZE { + // Get texture and Resolve image + let tex = self.textures.get_texture(tex_idx as u32); + if tex.is_none() { + break; // Past the end + // TODO: We should actually write blank descriptors + } + let tex = tex.unwrap(); + + let img_data = self.resolver.resolve(tex).expect("Invalid texture"); + + // Calculate buffer size + let (row_size, total_size) = + tex_size_info(&img_data, self.optimal_buffer_copy_pitch_alignment); + + // Create staging buffer + let mut staging_buffer = StagingBuffer::new( + &mut self.device, + &mut self.staging_allocator, + total_size as u64, + self.staging_memory_type, + ) + .expect("Couldn't create staging buffer"); + + // Write to staging buffer + let mapped_memory = staging_buffer + .map_memory(&mut self.device) + .expect("Error mapping staged memory"); + + img_data.copy_into(mapped_memory, row_size); + + staging_buffer.unmap_memory(&mut self.device); + + // Create image + let (img_mem, img) = create_image_view( + &mut self.device, + &mut *self.tex_allocator, + Self::FORMAT, + ImgUsage::SAMPLED, + &img_data, + ) + .unwrap(); + + // Create image view + let img_view = self + .device + .create_image_view( + &img, + ViewKind::D2, + Self::FORMAT, + Swizzle::NO, + Self::RESOURCES, + ) + .expect("Error creating image view"); + + // Queue copy from buffer to image + copy_cmds.push(BufferImageCopy { + buffer_offset: 0, + buffer_width: (row_size / super::PIXEL_SIZE) as u32, + buffer_height: img_data.height(), + image_layers: Self::LAYERS, + image_offset: Offset { x: 0, y: 0, z: 0 }, + image_extent: Extent { + width: img_data.width(), + height: img_data.height(), + depth: 1, + }, + }); + + // Create sampler + let sampler = self + .device + .create_sampler(&SamplerDesc::new(Filter::Nearest, WrapMode::Tile)) + .expect("Failed to create sampler"); + + // Write to descriptor set + { + self.device.write_descriptor_sets(vec![ + DescriptorSetWrite { + set: descriptor_set.raw(), + binding: 0, + array_offset: tex_idx % BLOCK_SIZE, + descriptors: Some(Descriptor::Image( + &img_view, + Layout::ShaderReadOnlyOptimal, + )), + }, + DescriptorSetWrite { + set: descriptor_set.raw(), + binding: 1, + array_offset: tex_idx % BLOCK_SIZE, + descriptors: Some(Descriptor::Sampler(&sampler)), + }, + ]); + } + + imgs.push(LoadedImage { + mem: ManuallyDrop::new(img_mem), + img: ManuallyDrop::new(img), + img_view: ManuallyDrop::new(img_view), + sampler: ManuallyDrop::new(sampler), + row_size, + height: img_data.height(), + width: img_data.width(), + }); + + staging_bufs.push(staging_buffer); + } + + // Add start pipeline barriers + for li in imgs.iter() { + use hal::image::Access; + + buf.pipeline_barrier( + PipelineStage::TOP_OF_PIPE..PipelineStage::TRANSFER, + Dependencies::empty(), + &[Barrier::Image { + states: (Access::empty(), Layout::Undefined) + ..(Access::TRANSFER_WRITE, Layout::TransferDstOptimal), + target: &*li.img, + families: None, + range: SubresourceRange { + aspects: Aspects::COLOR, + levels: 0..1, + layers: 0..1, + }, + }], + ); + } + + // Record copy commands + for (li, sb) in imgs.iter().zip(staging_bufs.iter()) { + buf.copy_buffer_to_image( + &*sb.buf, + &*li.img, + Layout::TransferDstOptimal, + &[BufferImageCopy { + buffer_offset: 0, + buffer_width: (li.row_size / super::PIXEL_SIZE) as u32, + buffer_height: li.height, + image_layers: SubresourceLayers { + aspects: Aspects::COLOR, + level: 0, + layers: 0..1, + }, + image_offset: Offset { x: 0, y: 0, z: 0 }, + image_extent: gfx_hal::image::Extent { + width: li.width, + height: li.height, + depth: 1, + }, + }], + ); + } + for li in imgs.iter() { + use hal::image::Access; + + buf.pipeline_barrier( + PipelineStage::TOP_OF_PIPE..PipelineStage::TRANSFER, + Dependencies::empty(), + &[Barrier::Image { + states: (Access::TRANSFER_WRITE, Layout::TransferDstOptimal) + ..(Access::SHADER_READ, Layout::ShaderReadOnlyOptimal), + target: &*li.img, + families: None, + range: Self::RESOURCES, + }], + ); + } + + buf.finish(); + + // Submit command buffer + self.gpu.queue_groups[self.cmd_queue_idx].queues[0].submit::<_, _, Semaphore, _, _>( + Submission { + command_buffers: &[&buf], + signal_semaphores: std::iter::empty(), + wait_semaphores: std::iter::empty(), + }, + Some(&fence), + ); + + Some(QueuedLoad { + staging_bufs, + fence, + buf, + block: TexturesBlock { + id: block_ref, + imgs, + descriptor_set: ManuallyDrop::new(descriptor_set), + }, + }) + } + + pub(crate) unsafe fn create_descriptor_set(&mut self) -> RDescriptorSet { + let mut v: ArrayVec<[RDescriptorSet; 1]> = ArrayVec::new(); + self.descriptor_allocator + .allocate( + self.device, + &*self.ds_layout, + DescriptorRanges::from_bindings(&[ + DescriptorSetLayoutBinding { + binding: 0, + ty: DescriptorType::SampledImage, + count: BLOCK_SIZE, + stage_flags: ShaderStageFlags::FRAGMENT, + immutable_samplers: false, + }, + DescriptorSetLayoutBinding { + binding: 1, + ty: DescriptorType::Sampler, + count: BLOCK_SIZE, + stage_flags: ShaderStageFlags::FRAGMENT, + immutable_samplers: false, + }, + ]), + 1, + &mut v, + ) + .unwrap(); + + v.pop().unwrap() + } +} + +pub fn tex_size_info<T: LoadableImage>(img: &T, obcpa: hal::buffer::Offset) -> (usize, usize) { + let initial_row_size = PIXEL_SIZE * img.width() as usize; + let row_alignment_mask = obcpa as u32 - 1; + + let row_size = ((initial_row_size as u32 + row_alignment_mask) & !row_alignment_mask) as usize; + let total_size = (row_size * (img.height() as usize)) as u64; + debug_assert!(row_size as usize >= initial_row_size); + + (row_size, total_size as usize) +} + +fn create_image_view<T, I>( + device: &mut Device, + allocator: &mut T, + format: Format, + usage: ImgUsage, + img: &I, +) -> Result<(T::Block, Image), &'static str> +where + T: Allocator<back::Backend>, + I: LoadableImage, +{ + // Make the image + let mut image_ref = unsafe { + use hal::image::{Kind, Tiling, ViewCapabilities}; + + device.create_image( + Kind::D2(img.width(), img.height(), 1, 1), + 1, + format, + Tiling::Optimal, + usage, + ViewCapabilities::empty(), + ) + } + .map_err(|_| "Couldn't create image")?; + + // Allocate memory + let (block, _) = unsafe { + let requirements = device.get_image_requirements(&image_ref); + + allocator.alloc(device, requirements.size, requirements.alignment) + } + .map_err(|_| "Out of memory")?; + + unsafe { + device + .bind_image_memory(&block.memory(), block.range().start, &mut image_ref) + .map_err(|_| "Couldn't bind memory to image")?; + } + + Ok((block, image_ref)) +} |