Expand Windows GPU support, add native FSR RR, and improve SR interop
Add a native D3D12 FSR Ray Regeneration route for gfx1200/gfx1201, using the game's device and queue with embedded HLSL guide preparation and output composition. Preserve the vkd3d compatibility route for RDNA3 and expose explicit route overrides and diagnostics. Improve Windows runtime compatibility: - Retry rejected application-ID initialization with a project identity. - Defer expensive RR backend creation until feature creation. - Recover from transient submission failures without disabling later frames. - Add shared-buffer export fallbacks through explicit heaps and a private device on the same adapter. - Safely evict old RR descriptor sets when input views change. - Treat requested RR presets as hints when a configured model is available. - Expand SR/RR input formats and handle diffuse/specular hit distances, packed normals, depth-stencil resources, and unused FSR guides. Reduce SR overhead by batching output downloads and format conversion under one completion wait. Optimize strict M enc2/tube attention with bounded FP32 accumulation and the original FP64 fallback for ambiguous rounding cases. Extend RDNA4 tooling with gfx1200/gfx1201 native FP8 RR model builds, capability checks, architecture-aware deployment, and experimental K/M kernel paths. Require matching model and runtime arithmetic families. Add optional embedded FSR shader optimizations keyed by shader hash and exact original SPIR-V contents. Preserve reflected metadata and record optimizer provenance, artifact hashes, and cache identity. Simplify Windows installation with automatic backend dependency downloads, download progress and ETA, backed-up absolute OptiScaler library paths, and recovery from interrupted RR model builds. Add build-kernels.bat and a kernels-only build matrix for all eleven registered RDNA targets, with code-object target validation and SHA256 manifests. Expand conversion probes, numerical stress tests, temporal benchmark tooling, installer/deployment tests, and setup documentation. Document synthetic regression and timing results, experimental build options, and remaining RDNA4 hardware validation requirements.
This commit is contained in:
ayo committed
1 parent
a617c818de
commit
ac4ee6c225
67 files changed
+3147
-326
No files matched your search
@@ -0,0 +1,63 @@
|
||||
#pragma once
|
||||
// RR guide formats shared by the native and FSR RR backends. Include after volk.h and rr_bridge_replay.h.
|
||||
#include <string>
|
||||
|
||||
namespace d4r::win::rr {
|
||||
// The Linux shim's RR guide formats (supported_input and aux_array_layout): every UNORM/SNORM/float
|
||||
// layout samples verbatim, as a D3D12 shader reads it; no CPU conversion or reinterpretation is involved.
|
||||
// Typeless resources use the float/UNORM view and sRGB is read as stored UNORM, like Linux.
|
||||
inline rr_bridge::Plane guide_format(DXGI_FORMAT format, size_t plane, const char* backend) {
|
||||
if (plane == Snapshot::Depth) {
|
||||
switch (format) {
|
||||
case DXGI_FORMAT_R32_FLOAT: case DXGI_FORMAT_R32_TYPELESS: case DXGI_FORMAT_D32_FLOAT:
|
||||
case DXGI_FORMAT_R32G8X24_TYPELESS: case DXGI_FORMAT_D32_FLOAT_S8X24_UINT: case DXGI_FORMAT_R32_FLOAT_X8X24_TYPELESS:
|
||||
return {nullptr, VK_FORMAT_R32_SFLOAT, 4};
|
||||
case DXGI_FORMAT_R16_FLOAT: return {nullptr, VK_FORMAT_R16_SFLOAT, 2};
|
||||
case DXGI_FORMAT_D16_UNORM: case DXGI_FORMAT_R16_UNORM: case DXGI_FORMAT_R16_TYPELESS: return {nullptr, VK_FORMAT_R16_UNORM, 2};
|
||||
default: break;
|
||||
}
|
||||
} else {
|
||||
switch (format) {
|
||||
case DXGI_FORMAT_R8_UNORM: case DXGI_FORMAT_R8_TYPELESS: return {nullptr, VK_FORMAT_R8_UNORM, 1};
|
||||
case DXGI_FORMAT_R8_SNORM: return {nullptr, VK_FORMAT_R8_SNORM, 1};
|
||||
case DXGI_FORMAT_R8G8_UNORM: case DXGI_FORMAT_R8G8_TYPELESS: return {nullptr, VK_FORMAT_R8G8_UNORM, 2};
|
||||
case DXGI_FORMAT_R8G8_SNORM: return {nullptr, VK_FORMAT_R8G8_SNORM, 2};
|
||||
case DXGI_FORMAT_R8G8B8A8_UNORM: case DXGI_FORMAT_R8G8B8A8_TYPELESS: case DXGI_FORMAT_R8G8B8A8_UNORM_SRGB:
|
||||
return {nullptr, VK_FORMAT_R8G8B8A8_UNORM, 4};
|
||||
case DXGI_FORMAT_R8G8B8A8_SNORM: return {nullptr, VK_FORMAT_R8G8B8A8_SNORM, 4};
|
||||
case DXGI_FORMAT_B8G8R8A8_UNORM: case DXGI_FORMAT_B8G8R8A8_TYPELESS: case DXGI_FORMAT_B8G8R8A8_UNORM_SRGB:
|
||||
return {nullptr, VK_FORMAT_B8G8R8A8_UNORM, 4};
|
||||
case DXGI_FORMAT_R16_UNORM: return {nullptr, VK_FORMAT_R16_UNORM, 2};
|
||||
case DXGI_FORMAT_R16_SNORM: return {nullptr, VK_FORMAT_R16_SNORM, 2};
|
||||
case DXGI_FORMAT_R16_FLOAT: case DXGI_FORMAT_R16_TYPELESS: return {nullptr, VK_FORMAT_R16_SFLOAT, 2};
|
||||
case DXGI_FORMAT_R16G16_UNORM: return {nullptr, VK_FORMAT_R16G16_UNORM, 4};
|
||||
case DXGI_FORMAT_R16G16_SNORM: return {nullptr, VK_FORMAT_R16G16_SNORM, 4};
|
||||
case DXGI_FORMAT_R16G16_FLOAT: case DXGI_FORMAT_R16G16_TYPELESS: return {nullptr, VK_FORMAT_R16G16_SFLOAT, 4};
|
||||
case DXGI_FORMAT_R16G16B16A16_UNORM: return {nullptr, VK_FORMAT_R16G16B16A16_UNORM, 8};
|
||||
case DXGI_FORMAT_R16G16B16A16_SNORM: return {nullptr, VK_FORMAT_R16G16B16A16_SNORM, 8};
|
||||
case DXGI_FORMAT_R16G16B16A16_FLOAT: case DXGI_FORMAT_R16G16B16A16_TYPELESS: return {nullptr, VK_FORMAT_R16G16B16A16_SFLOAT, 8};
|
||||
case DXGI_FORMAT_R32_FLOAT: case DXGI_FORMAT_R32_TYPELESS: return {nullptr, VK_FORMAT_R32_SFLOAT, 4};
|
||||
case DXGI_FORMAT_R32G32_FLOAT: case DXGI_FORMAT_R32G32_TYPELESS: return {nullptr, VK_FORMAT_R32G32_SFLOAT, 8};
|
||||
case DXGI_FORMAT_R32G32B32A32_FLOAT: case DXGI_FORMAT_R32G32B32A32_TYPELESS: return {nullptr, VK_FORMAT_R32G32B32A32_SFLOAT, 16};
|
||||
case DXGI_FORMAT_R10G10B10A2_UNORM: case DXGI_FORMAT_R10G10B10A2_TYPELESS: return {nullptr, VK_FORMAT_A2B10G10R10_UNORM_PACK32, 4};
|
||||
// Same bit layouts: R11G11B10 red in bits 0-10; R9G9B9E5 red in bits 0-8, exponent 27-31.
|
||||
case DXGI_FORMAT_R11G11B10_FLOAT: return {nullptr, VK_FORMAT_B10G11R11_UFLOAT_PACK32, 4};
|
||||
case DXGI_FORMAT_R9G9B9E5_SHAREDEXP: return {nullptr, VK_FORMAT_E5B9G9R9_UFLOAT_PACK32, 4};
|
||||
default: break;
|
||||
}
|
||||
}
|
||||
// Linux also accepts UINT/SINT guides by converting them to float on the CPU; this
|
||||
// sampled-image route has no conversion pass, so those still fail here, by name.
|
||||
throw std::runtime_error(std::string("Unsupported ") + backend + " sampled input format: plane " + std::to_string(plane) +
|
||||
" DXGI format " + std::to_string(unsigned(format)));
|
||||
}
|
||||
// UNORM guides store [0, 1]; normals in them are remapped to [-1, 1] before denoising.
|
||||
inline bool unorm_guide(VkFormat format) {
|
||||
switch (format) {
|
||||
case VK_FORMAT_R8G8B8A8_UNORM: case VK_FORMAT_B8G8R8A8_UNORM: case VK_FORMAT_A2B10G10R10_UNORM_PACK32:
|
||||
case VK_FORMAT_R16G16B16A16_UNORM: case VK_FORMAT_R8G8_UNORM: case VK_FORMAT_R16G16_UNORM:
|
||||
return true;
|
||||
default: return false;
|
||||
}
|
||||
}
|
||||
}
|
||||
Reference in new issue
Block a user