alex/librashader
1
0
Fork 0
Code Issues Pull requests Packages Projects Releases Wiki Activity

rt(d3d12): pull out descriptor heap implementation into its own crate

Browse source
This commit is contained in:
chyyran 2024-08-28 01:30:02 -04:00 committed by Ronny Chan
parent e934f175ef
commit ae76bf9cc1
13 changed files with 121 additions and 540 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

13
Cargo.lock generated
View file

@ -853,6 +853,18 @@ dependencies = [
"winapi",
]
[[package]]
name = "d3d12-descriptor-heap"
version = "0.1.2"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "825c5ff0f3af86b775d1e842a11fb42ee5eac1acec44ff46572a756219be3ff5"
dependencies = [
"array-init",
"bitvec",
"thiserror",
"windows 0.58.0",
]
[[package]]
name = "data-encoding"
version = "2.6.0"
@ -1774,6 +1786,7 @@ dependencies = [
"array-init",
"bitvec",
"bytemuck",
"d3d12-descriptor-heap",
"gfx-maths",
"gpu-allocator 0.27.0",
"librashader-cache",

1
librashader-runtime-d3d12/Cargo.toml
View file

@ -31,6 +31,7 @@ mach-siegbert-vogt-dxcsa = "0.1.3"
rayon = "1.6.1"
gpu-allocator = { version = "0.27.0", features = ["d3d12"], default-features = false}
parking_lot = "0.12.3"
d3d12-descriptor-heap = "0.1.0"
[target.'cfg(windows)'.dependencies.windows]
workspace = true

304
librashader-runtime-d3d12/src/descriptor_heap.rs
View file

@ -1,26 +1,11 @@
use crate::error;
use bitvec::bitvec;
use bitvec::boxed::BitBox;
use bitvec::order::Lsb0;
use std::marker::PhantomData;
use std::ops::Deref;
use std::sync::atomic::{AtomicUsize, Ordering};
use std::sync::Arc;
use crate::error::FilterChainError;
use windows::Win32::Graphics::Direct3D12::{
ID3D12DescriptorHeap, ID3D12Device, D3D12_CPU_DESCRIPTOR_HANDLE, D3D12_DESCRIPTOR_HEAP_DESC,
D3D12_DESCRIPTOR_HEAP_FLAG_NONE, D3D12_DESCRIPTOR_HEAP_FLAG_SHADER_VISIBLE,
D3D12_DESCRIPTOR_HEAP_TYPE, D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV,
D3D12_DESCRIPTOR_HEAP_DESC, D3D12_DESCRIPTOR_HEAP_FLAG_NONE,
D3D12_DESCRIPTOR_HEAP_FLAG_SHADER_VISIBLE, D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV,
D3D12_DESCRIPTOR_HEAP_TYPE_RTV, D3D12_DESCRIPTOR_HEAP_TYPE_SAMPLER,
D3D12_GPU_DESCRIPTOR_HANDLE,
};
pub trait D3D12HeapType {
fn get_desc(size: usize) -> D3D12_DESCRIPTOR_HEAP_DESC;
}
use d3d12_descriptor_heap::{D3D12DescriptorHeapType, D3D12ShaderVisibleDescriptorHeapType};
pub trait D3D12ShaderVisibleHeapType: D3D12HeapType {}
#[derive(Clone)]
pub struct SamplerPaletteHeap;
@ -36,9 +21,9 @@ pub struct ResourceWorkHeap;
#[derive(Clone)]
pub struct SamplerWorkHeap;
impl D3D12HeapType for SamplerPaletteHeap {
impl D3D12DescriptorHeapType for SamplerPaletteHeap {
// sampler palettes just get set directly
fn get_desc(size: usize) -> D3D12_DESCRIPTOR_HEAP_DESC {
fn create_desc(size: usize) -> D3D12_DESCRIPTOR_HEAP_DESC {
D3D12_DESCRIPTOR_HEAP_DESC {
Type: D3D12_DESCRIPTOR_HEAP_TYPE_SAMPLER,
NumDescriptors: size as u32,
@ -48,9 +33,9 @@ impl D3D12HeapType for SamplerPaletteHeap {
}
}
impl D3D12HeapType for CpuStagingHeap {
impl D3D12DescriptorHeapType for CpuStagingHeap {
// Lut texture heaps are CPU only and get bound to the descriptor heap of the shader.
fn get_desc(size: usize) -> D3D12_DESCRIPTOR_HEAP_DESC {
fn create_desc(size: usize) -> D3D12_DESCRIPTOR_HEAP_DESC {
D3D12_DESCRIPTOR_HEAP_DESC {
Type: D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV,
NumDescriptors: size as u32,
@ -60,9 +45,9 @@ impl D3D12HeapType for CpuStagingHeap {
}
}
impl D3D12HeapType for RenderTargetHeap {
impl D3D12DescriptorHeapType for RenderTargetHeap {
// Lut texture heaps are CPU only and get bound to the descriptor heap of the shader.
fn get_desc(size: usize) -> D3D12_DESCRIPTOR_HEAP_DESC {
fn create_desc(size: usize) -> D3D12_DESCRIPTOR_HEAP_DESC {
D3D12_DESCRIPTOR_HEAP_DESC {
Type: D3D12_DESCRIPTOR_HEAP_TYPE_RTV,
NumDescriptors: size as u32,
@ -72,10 +57,10 @@ impl D3D12HeapType for RenderTargetHeap {
}
}
impl D3D12ShaderVisibleHeapType for ResourceWorkHeap {}
impl D3D12HeapType for ResourceWorkHeap {
unsafe impl D3D12ShaderVisibleDescriptorHeapType for ResourceWorkHeap {}
impl D3D12DescriptorHeapType for ResourceWorkHeap {
// Lut texture heaps are CPU only and get bound to the descriptor heap of the shader.
fn get_desc(size: usize) -> D3D12_DESCRIPTOR_HEAP_DESC {
fn create_desc(size: usize) -> D3D12_DESCRIPTOR_HEAP_DESC {
D3D12_DESCRIPTOR_HEAP_DESC {
Type: D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV,
NumDescriptors: size as u32,
@ -85,10 +70,10 @@ impl D3D12HeapType for ResourceWorkHeap {
}
}
impl D3D12ShaderVisibleHeapType for SamplerWorkHeap {}
impl D3D12HeapType for SamplerWorkHeap {
unsafe impl D3D12ShaderVisibleDescriptorHeapType for SamplerWorkHeap {}
impl D3D12DescriptorHeapType for SamplerWorkHeap {
// Lut texture heaps are CPU only and get bound to the descriptor heap of the shader.
fn get_desc(size: usize) -> D3D12_DESCRIPTOR_HEAP_DESC {
fn create_desc(size: usize) -> D3D12_DESCRIPTOR_HEAP_DESC {
D3D12_DESCRIPTOR_HEAP_DESC {
Type: D3D12_DESCRIPTOR_HEAP_TYPE_SAMPLER,
NumDescriptors: size as u32,
@ -97,262 +82,3 @@ impl D3D12HeapType for SamplerWorkHeap {
}
}
}
pub type D3D12DescriptorHeapSlot<T> = Arc<D3D12DescriptorHeapSlotInner<T>>;
pub struct D3D12DescriptorHeapSlotInner<T> {
cpu_handle: D3D12_CPU_DESCRIPTOR_HANDLE,
gpu_handle: Option<D3D12_GPU_DESCRIPTOR_HANDLE>,
heap: Arc<D3D12DescriptorHeapInner>,
slot: usize,
_pd: PhantomData<T>,
}
impl<T> D3D12DescriptorHeapSlotInner<T> {
/// Get the index of the resource within the heap.
pub fn index(&self) -> usize {
self.slot
}
/// unsafe because type must match
pub unsafe fn copy_descriptor(&self, source: D3D12_CPU_DESCRIPTOR_HANDLE) {
unsafe {
let heap = self.heap.deref();
heap.device
.CopyDescriptorsSimple(1, self.cpu_handle, source, heap.ty)
}
}
}
impl<T> AsRef<D3D12_CPU_DESCRIPTOR_HANDLE> for D3D12DescriptorHeapSlotInner<T> {
fn as_ref(&self) -> &D3D12_CPU_DESCRIPTOR_HANDLE {
&self.cpu_handle
}
}
impl<T: D3D12ShaderVisibleHeapType> AsRef<D3D12_GPU_DESCRIPTOR_HANDLE>
for D3D12DescriptorHeapSlotInner<T>
{
fn as_ref(&self) -> &D3D12_GPU_DESCRIPTOR_HANDLE {
// SAFETY: D3D12ShaderVisibleHeapType must have a GPU handle.
self.gpu_handle.as_ref().unwrap()
}
}
impl<T: D3D12ShaderVisibleHeapType> From<&D3D12DescriptorHeap<T>> for ID3D12DescriptorHeap {
fn from(value: &D3D12DescriptorHeap<T>) -> Self {
value.0.heap.clone()
}
}
#[derive(Debug)]
struct D3D12DescriptorHeapInner {
device: ID3D12Device,
heap: ID3D12DescriptorHeap,
ty: D3D12_DESCRIPTOR_HEAP_TYPE,
cpu_start: D3D12_CPU_DESCRIPTOR_HANDLE,
gpu_start: Option<D3D12_GPU_DESCRIPTOR_HANDLE>,
handle_size: usize,
start: AtomicUsize,
num_descriptors: usize,
map: BitBox<AtomicUsize>,
}
pub struct D3D12DescriptorHeap<T>(Arc<D3D12DescriptorHeapInner>, PhantomData<T>);
impl<T: D3D12HeapType> D3D12DescriptorHeap<T> {
pub fn new(device: &ID3D12Device, size: usize) -> error::Result<D3D12DescriptorHeap<T>> {
let desc = T::get_desc(size);
unsafe { D3D12DescriptorHeap::new_with_desc(device, desc) }
}
}
impl<T> D3D12DescriptorHeap<T> {
/// Gets a cloned handle to the inner heap
pub fn handle(&self) -> ID3D12DescriptorHeap {
let inner = &self.0;
inner.heap.clone()
}
pub unsafe fn new_with_desc(
device: &ID3D12Device,
desc: D3D12_DESCRIPTOR_HEAP_DESC,
) -> error::Result<D3D12DescriptorHeap<T>> {
unsafe {
let heap: ID3D12DescriptorHeap = device.CreateDescriptorHeap(&desc)?;
let cpu_start = heap.GetCPUDescriptorHandleForHeapStart();
let gpu_start = if (desc.Flags & D3D12_DESCRIPTOR_HEAP_FLAG_SHADER_VISIBLE).0 != 0 {
Some(heap.GetGPUDescriptorHandleForHeapStart())
} else {
None
};
Ok(D3D12DescriptorHeap(
Arc::new(D3D12DescriptorHeapInner {
device: device.clone(),
heap,
ty: desc.Type,
cpu_start,
gpu_start,
handle_size: device.GetDescriptorHandleIncrementSize(desc.Type) as usize,
start: AtomicUsize::new(0),
num_descriptors: desc.NumDescriptors as usize,
map: bitvec![AtomicUsize, Lsb0; 0; desc.NumDescriptors as usize]
.into_boxed_bitslice(),
}),
PhantomData::default(),
))
}
}
/// suballocates this heap into equally sized chunks.
/// if there aren't enough descriptors, throws an error.
///
/// it is UB (programmer error) to call this if the descriptor heap already has
/// descriptors allocated for it.
///
/// size must also divide equally into the size of the heap.
pub unsafe fn suballocate(
self,
size: usize,
reserved: usize,
) -> (
Vec<D3D12DescriptorHeap<T>>,
Option<D3D12DescriptorHeap<T>>,
ID3D12DescriptorHeap,
) {
// has to be called right after creation.
assert_eq!(
Arc::strong_count(&self.0),
1,
"D3D12DescriptorHeap::suballocate can only be callled immediately after creation."
);
let Ok(inner) = Arc::try_unwrap(self.0) else {
panic!("[d3d12] undefined behaviour to suballocate a descriptor heap with live descriptors.")
};
let num_descriptors = inner.num_descriptors - reserved;
// number of suballocated heaps
let num_heaps = num_descriptors / size;
let remainder = num_descriptors % size;
assert_eq!(
remainder, 0,
"D3D12DescriptorHeap::suballocate \
must be called with a size that equally divides the number of descriptors"
);
let mut heaps = Vec::new();
let mut start = 0;
let root_cpu_ptr = inner.cpu_start.ptr;
let root_gpu_ptr = inner.gpu_start.map(|p| p.ptr);
for _ in 0..num_heaps {
let new_cpu_start = root_cpu_ptr + (start * inner.handle_size);
let new_gpu_start = root_gpu_ptr.map(|r| D3D12_GPU_DESCRIPTOR_HANDLE {
ptr: r + (start as u64 * inner.handle_size as u64),
});
heaps.push(D3D12DescriptorHeapInner {
device: inner.device.clone(),
heap: inner.heap.clone(),
ty: inner.ty,
cpu_start: D3D12_CPU_DESCRIPTOR_HANDLE { ptr: new_cpu_start },
gpu_start: new_gpu_start,
handle_size: inner.handle_size,
start: AtomicUsize::new(0),
num_descriptors: size,
map: bitvec![AtomicUsize, Lsb0; 0; size].into_boxed_bitslice(),
});
start += size;
}
let mut reserved_heap = None;
if reserved != 0 {
assert_eq!(
reserved,
inner.num_descriptors - start,
"The input heap could not fit the number of requested reserved descriptors."
);
let new_cpu_start = root_cpu_ptr + (start * inner.handle_size);
let new_gpu_start = root_gpu_ptr.map(|r| D3D12_GPU_DESCRIPTOR_HANDLE {
ptr: r + (start as u64 * inner.handle_size as u64),
});
reserved_heap = Some(D3D12DescriptorHeapInner {
device: inner.device.clone(),
heap: inner.heap.clone(),
ty: inner.ty,
cpu_start: D3D12_CPU_DESCRIPTOR_HANDLE { ptr: new_cpu_start },
gpu_start: new_gpu_start,
handle_size: inner.handle_size,
start: AtomicUsize::new(0),
num_descriptors: reserved,
map: bitvec![AtomicUsize, Lsb0; 0; reserved].into_boxed_bitslice(),
});
}
(
heaps
.into_iter()
.map(|inner| D3D12DescriptorHeap(Arc::new(inner), PhantomData::default()))
.collect(),
reserved_heap.map(|inner| D3D12DescriptorHeap(Arc::new(inner), PhantomData::default())),
inner.heap,
)
}
pub fn alloc_slot(&mut self) -> error::Result<D3D12DescriptorHeapSlot<T>> {
let mut handle = D3D12_CPU_DESCRIPTOR_HANDLE { ptr: 0 };
let inner = &self.0;
let start = inner.start.load(Ordering::Acquire);
for i in start..inner.num_descriptors {
if !inner.map[i] {
inner.map.set_aliased(i, true);
handle.ptr = inner.cpu_start.ptr + (i * inner.handle_size);
inner.start.store(i + 1, Ordering::Release);
let gpu_handle = inner
.gpu_start
.map(|gpu_start| D3D12_GPU_DESCRIPTOR_HANDLE {
ptr: (handle.ptr as u64 - inner.cpu_start.ptr as u64) + gpu_start.ptr,
});
return Ok(Arc::new(D3D12DescriptorHeapSlotInner {
cpu_handle: handle,
slot: i,
heap: Arc::clone(&self.0),
gpu_handle,
_pd: Default::default(),
}));
}
}
Err(FilterChainError::DescriptorHeapOverflow(
inner.num_descriptors,
))
}
pub fn alloc_range<const NUM_DESC: usize>(
&mut self,
) -> error::Result<[D3D12DescriptorHeapSlot<T>; NUM_DESC]> {
let dest = array_init::try_array_init(|_| self.alloc_slot())?;
Ok(dest)
}
}
impl<T> Drop for D3D12DescriptorHeapSlotInner<T> {
fn drop(&mut self) {
let inner = &self.heap;
inner.map.set_aliased(self.slot, false);
// inner.start > self.slot => inner.start = self.slot
inner.start.fetch_min(self.slot, Ordering::AcqRel);
}
}

6
librashader-runtime-d3d12/src/error.rs
View file

@ -1,5 +1,7 @@
//! Direct3D 12 shader runtime errors.
//!
use d3d12_descriptor_heap::D3D12DescriptorHeapError;
use thiserror::Error;
/// Cumulative error type for Direct3D12 filter chains.
@ -19,8 +21,8 @@ pub enum FilterChainError {
ShaderReflectError(#[from] ShaderReflectError),
#[error("lut loading error")]
LutLoadError(#[from] ImageError),
#[error("heap overflow")]
DescriptorHeapOverflow(usize),
#[error("heap error")]
HeapError(#[from] D3D12DescriptorHeapError),
#[error("allocation error")]
AllocationError(#[from] gpu_allocator::AllocationError),
}

73
librashader-runtime-d3d12/src/filter_chain.rs
View file

@ -1,8 +1,5 @@
use crate::buffer::{D3D12Buffer, RawD3D12Buffer};
use crate::descriptor_heap::{
CpuStagingHeap, D3D12DescriptorHeap, D3D12DescriptorHeapSlot, RenderTargetHeap,
ResourceWorkHeap,
};
use crate::descriptor_heap::{CpuStagingHeap, RenderTargetHeap, ResourceWorkHeap};
use crate::draw_quad::DrawQuad;
use crate::error::FilterChainError;
use crate::filter_pass::FilterPass;
@ -14,6 +11,9 @@ use crate::options::{FilterChainOptionsD3D12, FrameOptionsD3D12};
use crate::samplers::SamplerSet;
use crate::texture::{D3D12InputImage, D3D12OutputView, InputTexture, OutputDescriptor};
use crate::{error, util};
use d3d12_descriptor_heap::{
D3D12DescriptorHeap, D3D12DescriptorHeapSlot, D3D12PartitionableHeap, D3D12PartitionedHeap,
};
use gpu_allocator::d3d12::{Allocator, AllocatorCreateDesc, ID3D12DeviceVersion};
use librashader_common::map::FastHashMap;
use librashader_common::{ImageFormat, Size, Viewport};
@ -308,18 +308,22 @@ impl FilterChainD3D12 {
})?));
let draw_quad = DrawQuad::new(&allocator)?;
let mut staging_heap = D3D12DescriptorHeap::new(
device,
(MAX_BINDINGS_COUNT as usize) * shader_count
+ MIPMAP_RESERVED_WORKHEAP_DESCRIPTORS
+ lut_count,
)?;
let rtv_heap = D3D12DescriptorHeap::new(
device,
(MAX_BINDINGS_COUNT as usize) * shader_count
+ MIPMAP_RESERVED_WORKHEAP_DESCRIPTORS
+ lut_count,
)?;
let mut staging_heap = unsafe {
D3D12DescriptorHeap::new(
device,
(MAX_BINDINGS_COUNT as usize) * shader_count
+ MIPMAP_RESERVED_WORKHEAP_DESCRIPTORS
+ lut_count,
)
}?;
let rtv_heap = unsafe {
D3D12DescriptorHeap::new(
device,
(MAX_BINDINGS_COUNT as usize) * shader_count
+ MIPMAP_RESERVED_WORKHEAP_DESCRIPTORS
+ lut_count,
)
}?;
let root_signature = D3D12RootSignature::new(device)?;
@ -473,22 +477,31 @@ impl FilterChainD3D12 {
D3D12DescriptorHeap<ResourceWorkHeap>,
)> {
let shader_count = passes.len();
let work_heap = D3D12DescriptorHeap::<ResourceWorkHeap>::new(
device,
(MAX_BINDINGS_COUNT as usize) * shader_count + MIPMAP_RESERVED_WORKHEAP_DESCRIPTORS,
)?;
let (work_heaps, mipmap_heap, texture_heap_handle) = unsafe {
work_heap.suballocate(
let D3D12PartitionedHeap {
partitioned: work_heaps,
reserved: mipmap_heap,
handle: texture_heap_handle,
} = unsafe {
let work_heap = D3D12PartitionableHeap::<ResourceWorkHeap>::new(
device,
(MAX_BINDINGS_COUNT as usize) * shader_count + MIPMAP_RESERVED_WORKHEAP_DESCRIPTORS,
)?;
work_heap.into_partitioned(
MAX_BINDINGS_COUNT as usize,
MIPMAP_RESERVED_WORKHEAP_DESCRIPTORS,
)
)?
};
let sampler_work_heap =
D3D12DescriptorHeap::new(device, (MAX_BINDINGS_COUNT as usize) * shader_count)?;
let (sampler_work_heaps, _, sampler_heap_handle) =
unsafe { sampler_work_heap.suballocate(MAX_BINDINGS_COUNT as usize, 0) };
let D3D12PartitionedHeap {
partitioned: sampler_work_heaps,
reserved: _,
handle: sampler_heap_handle,
} = unsafe {
let sampler_heap =
D3D12PartitionableHeap::new(device, (MAX_BINDINGS_COUNT as usize) * shader_count)?;
sampler_heap.into_partitioned(MAX_BINDINGS_COUNT as usize, 0)?
};
let filters: Vec<error::Result<_>> = passes
.into_par_iter()
@ -578,8 +591,8 @@ impl FilterChainD3D12 {
let uniform_bindings =
reflection.meta.create_binding_map(|param| param.offset());
let texture_heap = texture_heap.alloc_range()?;
let sampler_heap = sampler_heap.alloc_range()?;
let texture_heap = texture_heap.allocate_descriptor_range()?;
let sampler_heap = sampler_heap.allocate_descriptor_range()?;
Ok(FilterPass {
reflection,

16
librashader-runtime-d3d12/src/filter_pass.rs
View file

@ -1,11 +1,12 @@
use crate::buffer::RawD3D12Buffer;
use crate::descriptor_heap::{D3D12DescriptorHeapSlot, ResourceWorkHeap, SamplerWorkHeap};
use crate::descriptor_heap::{ResourceWorkHeap, SamplerWorkHeap};
use crate::error;
use crate::filter_chain::FilterCommon;
use crate::graphics_pipeline::D3D12GraphicsPipeline;
use crate::options::FrameOptionsD3D12;
use crate::samplers::SamplerSet;
use crate::texture::{D3D12OutputView, InputTexture};
use d3d12_descriptor_heap::D3D12DescriptorHeapSlot;
use librashader_common::map::FastHashMap;
use librashader_common::{ImageFormat, Size, Viewport};
use librashader_preprocess::ShaderSource;
@ -17,7 +18,6 @@ use librashader_runtime::filter_pass::FilterPassMeta;
use librashader_runtime::quad::QuadType;
use librashader_runtime::render_target::RenderTarget;
use librashader_runtime::uniforms::{NoUniformBinder, UniformStorage};
use std::ops::Deref;
use windows::core::Interface;
use windows::Win32::Foundation::RECT;
use windows::Win32::Graphics::Direct3D12::{
@ -66,12 +66,8 @@ impl BindSemantics<NoUniformBinder, Option<()>, RawD3D12Buffer, RawD3D12Buffer>
unsafe {
texture_binding[binding.binding as usize].copy_descriptor(*texture.descriptor.as_ref());
sampler_binding[binding.binding as usize].copy_descriptor(
*samplers
.get(texture.wrap_mode, texture.filter)
.deref()
.as_ref(),
)
sampler_binding[binding.binding as usize]
.copy_descriptor(*samplers.get(texture.wrap_mode, texture.filter).as_ref())
}
}
}
@ -177,8 +173,8 @@ impl FilterPass {
}
unsafe {
cmd.SetGraphicsRootDescriptorTable(0, *self.texture_heap[0].deref().as_ref());
cmd.SetGraphicsRootDescriptorTable(1, *self.sampler_heap[0].deref().as_ref());
cmd.SetGraphicsRootDescriptorTable(0, *self.texture_heap[0].as_ref());
cmd.SetGraphicsRootDescriptorTable(1, *self.sampler_heap[0].as_ref());
}
// todo: check for non-renderpass.

12
librashader-runtime-d3d12/src/framebuffer.rs
View file

@ -1,9 +1,10 @@
use crate::descriptor_heap::{CpuStagingHeap, D3D12DescriptorHeap, RenderTargetHeap};
use crate::descriptor_heap::{CpuStagingHeap, RenderTargetHeap};
use crate::error::FilterChainError;
use crate::filter_chain::FrameResiduals;
use crate::texture::{D3D12OutputView, InputTexture};
use crate::util::d3d12_get_closest_format;
use crate::{error, util};
use d3d12_descriptor_heap::D3D12DescriptorHeap;
use gpu_allocator::d3d12::{
Allocator, Resource, ResourceCategory, ResourceCreateDesc, ResourceStateOrBarrierLayout,
ResourceType,
@ -14,7 +15,6 @@ use librashader_presets::Scale2D;
use librashader_runtime::scaling::{MipmapSize, ScaleFramebuffer, ViewportSize};
use parking_lot::Mutex;
use std::mem::ManuallyDrop;
use std::ops::Deref;
use std::sync::Arc;
use windows::Win32::Graphics::Direct3D12::{
ID3D12Device, ID3D12GraphicsCommandList, D3D12_BOX, D3D12_DEFAULT_SHADER_4_COMPONENT_MAPPING,
@ -275,7 +275,7 @@ impl OwnedImage {
filter: FilterMode,
wrap_mode: WrapMode,
) -> error::Result<InputTexture> {
let descriptor = heap.alloc_slot()?;
let descriptor = heap.allocate_descriptor()?;
unsafe {
let srv_desc = D3D12_SHADER_RESOURCE_VIEW_DESC {
@ -293,7 +293,7 @@ impl OwnedImage {
self.device.CreateShaderResourceView(
self.handle.resource(),
Some(&srv_desc),
*descriptor.deref().as_ref(),
*descriptor.as_ref(),
);
}
@ -311,7 +311,7 @@ impl OwnedImage {
&self,
heap: &mut D3D12DescriptorHeap<RenderTargetHeap>,
) -> error::Result<D3D12OutputView> {
let descriptor = heap.alloc_slot()?;
let descriptor = heap.allocate_descriptor()?;
unsafe {
let rtv_desc = D3D12_RENDER_TARGET_VIEW_DESC {
@ -328,7 +328,7 @@ impl OwnedImage {
self.device.CreateRenderTargetView(
self.handle.resource(),
Some(&rtv_desc),
*descriptor.deref().as_ref(),
*descriptor.as_ref(),
);
}

8
librashader-runtime-d3d12/src/luts.rs
View file

@ -1,9 +1,10 @@
use crate::descriptor_heap::{CpuStagingHeap, D3D12DescriptorHeap};
use crate::descriptor_heap::CpuStagingHeap;
use crate::error;
use crate::filter_chain::FrameResiduals;
use crate::mipmap::MipmapGenContext;
use crate::texture::InputTexture;
use crate::util::{d3d12_get_closest_format, d3d12_resource_transition, d3d12_update_subresources};
use d3d12_descriptor_heap::D3D12DescriptorHeap;
use gpu_allocator::d3d12::{
Allocator, Resource, ResourceCategory, ResourceCreateDesc, ResourceStateOrBarrierLayout,
ResourceType,
@ -14,7 +15,6 @@ use librashader_runtime::image::Image;
use librashader_runtime::scaling::MipmapSize;
use parking_lot::Mutex;
use std::mem::ManuallyDrop;
use std::ops::Deref;
use std::sync::Arc;
use windows::Win32::Graphics::Direct3D12::{
ID3D12Device, ID3D12GraphicsCommandList, D3D12_DEFAULT_SHADER_4_COMPONENT_MAPPING,
@ -80,7 +80,7 @@ impl LutTexture {
}
desc.Format = d3d12_get_closest_format(device, format_support);
let descriptor = heap.alloc_slot()?;
let descriptor = heap.allocate_descriptor()?;
// create handles on GPU
let resource = allocator.lock().create_resource(&ResourceCreateDesc {
@ -112,7 +112,7 @@ impl LutTexture {
device.CreateShaderResourceView(
resource.resource(),
Some(&srv_desc),
*descriptor.deref().as_ref(),
*descriptor.as_ref(),
);
}

16
librashader-runtime-d3d12/src/mipmap.rs
View file

@ -1,11 +1,11 @@
use crate::descriptor_heap::{D3D12DescriptorHeap, D3D12DescriptorHeapSlot, ResourceWorkHeap};
use crate::descriptor_heap::ResourceWorkHeap;
use crate::util::dxc_validate_shader;
use crate::{error, util};
use bytemuck::{Pod, Zeroable};
use d3d12_descriptor_heap::{D3D12DescriptorHeap, D3D12DescriptorHeapSlot};
use librashader_common::Size;
use librashader_runtime::scaling::MipmapSize;
use std::mem::ManuallyDrop;
use std::ops::Deref;
use windows::Win32::Graphics::Direct3D::Dxc::{
CLSID_DxcLibrary, CLSID_DxcValidator, DxcCreateInstance,
};
@ -191,7 +191,7 @@ impl D3D12MipmapGen {
Vec<D3D12_RESOURCE_BARRIER>,
)> {
// create views for mipmap generation
let srv = work_heap.alloc_slot()?;
let srv = work_heap.allocate_descriptor()?;
unsafe {
let srv_desc = D3D12_SHADER_RESOURCE_VIEW_DESC {
Format: format,
@ -206,14 +206,14 @@ impl D3D12MipmapGen {
};
self.device
.CreateShaderResourceView(resource, Some(&srv_desc), *srv.deref().as_ref());
.CreateShaderResourceView(resource, Some(&srv_desc), *srv.as_ref());
}
let mut heap_slots = Vec::with_capacity(miplevels as usize);
heap_slots.push(srv);
for i in 1..miplevels {
let descriptor = work_heap.alloc_slot()?;
let descriptor = work_heap.allocate_descriptor()?;
let desc = D3D12_UNORDERED_ACCESS_VIEW_DESC {
Format: format,
ViewDimension: D3D12_UAV_DIMENSION_TEXTURE2D,
@ -230,14 +230,14 @@ impl D3D12MipmapGen {
resource,
None,
Some(&desc),
*descriptor.deref().as_ref(),
*descriptor.as_ref(),
);
}
heap_slots.push(descriptor);
}
unsafe {
cmd.SetComputeRootDescriptorTable(0, *heap_slots[0].deref().as_ref());
cmd.SetComputeRootDescriptorTable(0, *heap_slots[0].as_ref());
}
let mut residual_barriers = Vec::new();
@ -269,7 +269,7 @@ impl D3D12MipmapGen {
cmd.ResourceBarrier(&barriers);
residual_barriers.extend(barriers);
cmd.SetComputeRootDescriptorTable(1, *heap_slots[i as usize].deref().as_ref());
cmd.SetComputeRootDescriptorTable(1, *heap_slots[i as usize].as_ref());
cmd.SetComputeRoot32BitConstants(
2,
(std::mem::size_of::<MipConstants>() / std::mem::size_of::<u32>()) as u32,

10
librashader-runtime-d3d12/src/samplers.rs
View file

@ -1,8 +1,8 @@
use crate::descriptor_heap::{D3D12DescriptorHeap, D3D12DescriptorHeapSlot, SamplerPaletteHeap};
use crate::descriptor_heap::SamplerPaletteHeap;
use crate::error;
use d3d12_descriptor_heap::{D3D12DescriptorHeap, D3D12DescriptorHeapSlot};
use librashader_common::map::FastHashMap;
use librashader_common::{FilterMode, WrapMode};
use std::ops::Deref;
use windows::Win32::Graphics::Direct3D12::{
ID3D12Device, D3D12_COMPARISON_FUNC_NEVER, D3D12_FLOAT32_MAX, D3D12_SAMPLER_DESC,
D3D12_TEXTURE_ADDRESS_MODE,
@ -33,12 +33,12 @@ impl SamplerSet {
WrapMode::MirroredRepeat,
];
let mut heap = D3D12DescriptorHeap::new(device, 2 * wrap_modes.len())?;
let mut heap = unsafe { D3D12DescriptorHeap::new(device, 2 * wrap_modes.len())? };
for wrap_mode in wrap_modes {
for filter_mode in &[FilterMode::Linear, FilterMode::Nearest] {
unsafe {
let sampler = heap.alloc_slot()?;
let sampler = heap.allocate_descriptor()?;
device.CreateSampler(
&D3D12_SAMPLER_DESC {
Filter: (*filter_mode).into(),
@ -52,7 +52,7 @@ impl SamplerSet {
MinLOD: -D3D12_FLOAT32_MAX,
MaxLOD: D3D12_FLOAT32_MAX,
},
*sampler.deref().as_ref(),
*sampler.as_ref(),
);
samplers.insert((*wrap_mode, *filter_mode), sampler);
}

8
librashader-runtime-d3d12/src/texture.rs
View file

@ -1,6 +1,6 @@
use crate::descriptor_heap::{CpuStagingHeap, D3D12DescriptorHeapSlot, RenderTargetHeap};
use crate::descriptor_heap::{CpuStagingHeap, RenderTargetHeap};
use d3d12_descriptor_heap::D3D12DescriptorHeapSlot;
use librashader_common::{FilterMode, Size, WrapMode};
use std::ops::Deref;
use windows::Win32::Graphics::Direct3D12::{ID3D12Resource, D3D12_CPU_DESCRIPTOR_HANDLE};
use windows::Win32::Graphics::Dxgi::Common::DXGI_FORMAT;
@ -25,7 +25,7 @@ pub(crate) enum OutputDescriptor {
impl AsRef<D3D12_CPU_DESCRIPTOR_HANDLE> for InputDescriptor {
fn as_ref(&self) -> &D3D12_CPU_DESCRIPTOR_HANDLE {
match self {
InputDescriptor::Owned(h) => h.deref().as_ref(),
InputDescriptor::Owned(h) => h.as_ref(),
InputDescriptor::Raw(h) => h,
}
}
@ -34,7 +34,7 @@ impl AsRef<D3D12_CPU_DESCRIPTOR_HANDLE> for InputDescriptor {
impl AsRef<D3D12_CPU_DESCRIPTOR_HANDLE> for OutputDescriptor {
fn as_ref(&self) -> &D3D12_CPU_DESCRIPTOR_HANDLE {
match self {
OutputDescriptor::Owned(h) => h.deref().as_ref(),
OutputDescriptor::Owned(h) => h.as_ref(),
OutputDescriptor::Raw(h) => h,
}
}

176
librashader-runtime-d3d12/tests/hello_triangle/descriptor_heap.rs
View file

@ -1,30 +1,15 @@
use bitvec::bitvec;
use bitvec::boxed::BitBox;
use bitvec::order::Lsb0;
use std::marker::PhantomData;
use std::ops::Deref;
use std::sync::atomic::{AtomicUsize, Ordering};
use std::sync::Arc;
use d3d12_descriptor_heap::D3D12DescriptorHeapType;
use windows::Win32::Graphics::Direct3D12::{
ID3D12DescriptorHeap, ID3D12Device, D3D12_CPU_DESCRIPTOR_HANDLE, D3D12_DESCRIPTOR_HEAP_DESC,
D3D12_DESCRIPTOR_HEAP_FLAG_NONE, D3D12_DESCRIPTOR_HEAP_FLAG_SHADER_VISIBLE,
D3D12_DESCRIPTOR_HEAP_TYPE, D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV,
D3D12_GPU_DESCRIPTOR_HANDLE,
D3D12_DESCRIPTOR_HEAP_DESC, D3D12_DESCRIPTOR_HEAP_FLAG_NONE,
D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV,
};
pub trait D3D12HeapType {
fn get_desc(size: usize) -> D3D12_DESCRIPTOR_HEAP_DESC;
}
pub trait D3D12ShaderVisibleHeapType: D3D12HeapType {}
#[derive(Clone)]
pub struct CpuStagingHeap;
impl D3D12HeapType for CpuStagingHeap {
impl D3D12DescriptorHeapType for CpuStagingHeap {
// Lut texture heaps are CPU only and get bound to the descriptor heap of the shader.
fn get_desc(size: usize) -> D3D12_DESCRIPTOR_HEAP_DESC {
fn create_desc(size: usize) -> D3D12_DESCRIPTOR_HEAP_DESC {
D3D12_DESCRIPTOR_HEAP_DESC {
Type: D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV,
NumDescriptors: size as u32,
@ -33,154 +18,3 @@ impl D3D12HeapType for CpuStagingHeap {
}
}
}
pub type D3D12DescriptorHeapSlot<T> = Arc<D3D12DescriptorHeapSlotInner<T>>;
pub struct D3D12DescriptorHeapSlotInner<T> {
cpu_handle: D3D12_CPU_DESCRIPTOR_HANDLE,
gpu_handle: Option<D3D12_GPU_DESCRIPTOR_HANDLE>,
heap: Arc<D3D12DescriptorHeapInner>,
slot: usize,
_pd: PhantomData<T>,
}
impl<T> D3D12DescriptorHeapSlotInner<T> {
/// Get the index of the resource within the heap.
pub fn index(&self) -> usize {
self.slot
}
/// unsafe because type must match
pub unsafe fn copy_descriptor(&self, source: D3D12_CPU_DESCRIPTOR_HANDLE) {
unsafe {
let heap = &self.heap.deref();
heap.device
.CopyDescriptorsSimple(1, self.cpu_handle, source, heap.ty)
}
}
}
impl<T> AsRef<D3D12_CPU_DESCRIPTOR_HANDLE> for D3D12DescriptorHeapSlotInner<T> {
fn as_ref(&self) -> &D3D12_CPU_DESCRIPTOR_HANDLE {
&self.cpu_handle
}
}
impl<T: D3D12ShaderVisibleHeapType> AsRef<D3D12_GPU_DESCRIPTOR_HANDLE>
for D3D12DescriptorHeapSlotInner<T>
{
fn as_ref(&self) -> &D3D12_GPU_DESCRIPTOR_HANDLE {
// SAFETY: D3D12ShaderVisibleHeapType must have a GPU handle.
self.gpu_handle.as_ref().unwrap()
}
}
impl<T: D3D12ShaderVisibleHeapType> From<&D3D12DescriptorHeap<T>> for ID3D12DescriptorHeap {
fn from(value: &D3D12DescriptorHeap<T>) -> Self {
value.0.heap.clone()
}
}
#[derive(Debug)]
struct D3D12DescriptorHeapInner {
device: ID3D12Device,
heap: ID3D12DescriptorHeap,
ty: D3D12_DESCRIPTOR_HEAP_TYPE,
cpu_start: D3D12_CPU_DESCRIPTOR_HANDLE,
gpu_start: Option<D3D12_GPU_DESCRIPTOR_HANDLE>,
handle_size: usize,
start: AtomicUsize,
num_descriptors: usize,
map: BitBox<AtomicUsize>,
}
pub struct D3D12DescriptorHeap<T>(Arc<D3D12DescriptorHeapInner>, PhantomData<T>);
impl<T: D3D12HeapType> D3D12DescriptorHeap<T> {
pub fn new(
device: &ID3D12Device,
size: usize,
) -> Result<D3D12DescriptorHeap<T>, windows::core::Error> {
let desc = T::get_desc(size);
unsafe { D3D12DescriptorHeap::new_with_desc(device, desc) }
}
}
impl<T> D3D12DescriptorHeap<T> {
/// Gets a cloned handle to the inner heap
pub fn handle(&self) -> ID3D12DescriptorHeap {
self.0.heap.clone()
}
pub unsafe fn new_with_desc(
device: &ID3D12Device,
desc: D3D12_DESCRIPTOR_HEAP_DESC,
) -> Result<D3D12DescriptorHeap<T>, windows::core::Error> {
unsafe {
let heap: ID3D12DescriptorHeap = device.CreateDescriptorHeap(&desc)?;
let cpu_start = heap.GetCPUDescriptorHandleForHeapStart();
let gpu_start = if (desc.Flags & D3D12_DESCRIPTOR_HEAP_FLAG_SHADER_VISIBLE).0 != 0 {
Some(heap.GetGPUDescriptorHandleForHeapStart())
} else {
None
};
Ok(D3D12DescriptorHeap(
Arc::new(D3D12DescriptorHeapInner {
device: device.clone(),
heap,
ty: desc.Type,
cpu_start,
gpu_start,
handle_size: device.GetDescriptorHandleIncrementSize(desc.Type) as usize,
start: AtomicUsize::new(0),
num_descriptors: desc.NumDescriptors as usize,
map: bitvec![AtomicUsize, Lsb0; 0; desc.NumDescriptors as usize]
.into_boxed_bitslice(),
}),
PhantomData::default(),
))
}
}
pub fn alloc_slot(&mut self) -> D3D12DescriptorHeapSlot<T> {
let mut handle = D3D12_CPU_DESCRIPTOR_HANDLE { ptr: 0 };
let inner = &self.0;
let start = inner.start.load(Ordering::Acquire);
for i in start..inner.num_descriptors {
if !inner.map[i] {
inner.map.set_aliased(i, true);
handle.ptr = inner.cpu_start.ptr + (i * inner.handle_size);
inner.start.store(i + 1, Ordering::Release);
let gpu_handle = inner
.gpu_start
.map(|gpu_start| D3D12_GPU_DESCRIPTOR_HANDLE {
ptr: (handle.ptr as u64 - inner.cpu_start.ptr as u64) + gpu_start.ptr,
});
return Arc::new(D3D12DescriptorHeapSlotInner {
cpu_handle: handle,
slot: i,
heap: Arc::clone(&self.0),
gpu_handle,
_pd: Default::default(),
});
}
}
panic!("overflow")
}
}
impl<T> Drop for D3D12DescriptorHeapSlotInner<T> {
fn drop(&mut self) {
let inner = &self.heap;
inner.map.set_aliased(self.slot, false);
// inner.start > self.slot => inner.start = self.slot
inner.start.fetch_min(self.slot, Ordering::AcqRel);
}
}

18
librashader-runtime-d3d12/tests/hello_triangle/mod.rs
View file

@ -229,7 +229,8 @@ unsafe extern "system" fn debug_log(
pub mod d3d12_hello_triangle {
use super::*;
use crate::hello_triangle::descriptor_heap::{CpuStagingHeap, D3D12DescriptorHeap};
use crate::hello_triangle::descriptor_heap::CpuStagingHeap;
use d3d12_descriptor_heap::D3D12DescriptorHeap;
use librashader_common::{Size, Viewport};
use librashader_runtime_d3d12::{D3D12InputImage, D3D12OutputView, FilterChainD3D12};
use std::mem::ManuallyDrop;
@ -480,7 +481,7 @@ pub mod d3d12_hello_triangle {
fence,
fence_value,
fence_event,
frambuffer_heap: D3D12DescriptorHeap::new(&self.device, 1024).unwrap(),
frambuffer_heap: unsafe { D3D12DescriptorHeap::new(&self.device, 1024).unwrap() },
});
Ok(())
@ -496,7 +497,7 @@ pub mod d3d12_hello_triangle {
fn render(&mut self) {
if let Some(resources) = &mut self.resources {
let srv = resources.frambuffer_heap.alloc_slot();
let srv = resources.frambuffer_heap.allocate_descriptor().unwrap();
unsafe {
self.device.CreateShaderResourceView(
@ -512,17 +513,12 @@ pub mod d3d12_hello_triangle {
},
},
}),
*srv.deref().as_ref(),
*srv.as_ref(),
)
}
populate_command_list(
resources,
&mut self.filter,
self.framecount,
*srv.deref().as_ref(),
)
.unwrap();
populate_command_list(resources, &mut self.filter, self.framecount, *srv.as_ref())
.unwrap();
// Execute the command list.
let command_list: Option<ID3D12CommandList> = resources.command_list.cast().ok();