ayo
a617c818de
network natively on HIP with Vulkan->HIP shared memory, extend K kernels to
gfx11 (RDNA3), and add dependency/packaging/test tooling around them.
Windows: FSR RR backend, native HIP K runner, gfx11 K kernels, tooling
FSR Ray Regeneration (experimental)
- Add fsr_rr_backend.{h,cpp}, fsr_rr_passes.h and the fsr_rr/ prepare and
compose compute shaders. They run AMD's FidelityFX denoiser
(amd_fidelityfx_denoiser_dx12.dll, SDK 2.3, not bundled) on a private
vkd3d-proton device.
- Add the amdxc64 shim (tools/windows/amdxc64_shim) and two vkd3d-proton
patches for libcxx includes. AMD's amdxc64.dll is never loaded.
- d4r_nvngx_windows: RrFeature now selects between the native DLSS RR
backend and the FSR backend (D4R_RR_BACKEND=fsr). After the denoiser it
runs an internal Super Sampling stage (feature 1) to upscale. If SR is
unavailable and output != render size, the frame fails; it is never
returned noisy.
- d4r.ini / player_config: new [RayReconstruction] FsrDenoiser,
FsrStabilityBias (auto or >= 0) and FsrDetail (0-1), with validation, and
matching D4R_FSR_* environment variables.
- Add the FSR RR fixture, output-check and ngx test scripts, and a shader
compile script.
Native HIP K runner (Windows)
- CMake: add d4r_engine_k_hip (d4r-k-hip) plus separate volk/engine copies,
because VK_USE_PLATFORM_WIN32_KHR changes volk's entry points.
- Share Vulkan memory with HIP via Win32 opaque handles on Windows
(VK_KHR_external_memory_win32) and via fds on Linux. Handle types are now
intptr_t, and the platform choice is in D4R_EXT_MEMORY_* macros.
- HipNetwork: add debugRepeatHashes and debugNodeMicros, exposed as
D4R_HIP_REPEAT_CHECK, D4R_HIP_REPEAT_ALL and D4R_HIP_NODE_PROFILE in
replay.cpp.
- replay.cpp: display-resolution-motion frames (FrameFlagDisplayMotion) no
longer require near2rnd == out2rnd. Both values must still be in (0, 1].
K kernels on gfx11
- Allow position-token and enc1 tiling on __GFX11__ as well as gfx12.
- Add an opt-in ds_bpermute cross-half exchange
(D4R_WINDOWS_GFX11_DS_EXCHANGE, enabled for K on Windows gfx11). M keeps
its LDS path.
- Fix a block-barrier race in the position layer. Without POS_VSHARE a fast
wave could overwrite its hb rows while a slower wave still read them for a
later head, giving rare wrong results on gfx1101 (enc0 outputs differing
between identical launches).
Packaging and install
- install.bat, d4r.ini, WINDOWS_README: FSR option, and
install_dependencies.py to install pinned NVIDIA dependencies with staged,
hash-verified replacement and rollback.
- Update the release package, local package refresh and test scripts.
- Add the InteropBatchInputCopies option (default false). Batching is faster
on some synthetic runs, but player temporal exactness is unresolved.
Tooling and tests
- K model: capture_k_weights, build_k_model, compare_k_outputs,
benchmark_k_hip, build_capture_runtime, test_k_model and the
zluda-k-output-capture patch. Captured weights and models are private
local assets and must not enter release archives.
- Synthetic probes: test_m_ngx, test_sr_dynamic,
prepare_rr_dynamic_fixture and the d3d12/rr_ngx probe updates.
- Add tests/windows/test_install_dependencies.py.
Docs
- Update the windows-fsr-rr, k-performance, local-package, rr-interop and
rr-model-cache docs. Add windows-parity-handoff.md and
windows-parity-continuation-prompt.md.
52 lines
2.9 KiB
Python
52 lines
2.9 KiB
Python
"""Inspect preserved synthetic K half-output differences without running a GPU."""
|
|
import argparse
|
|
import json
|
|
from pathlib import Path
|
|
import sys
|
|
|
|
|
|
def main():
|
|
repo = Path(__file__).resolve().parents[2]
|
|
sys.path.insert(0, str(repo / '.tools/python/vendor'))
|
|
import numpy as np
|
|
parser = argparse.ArgumentParser(description=__doc__)
|
|
for name in ('reference', 'candidate', 'output'):
|
|
parser.add_argument('--' + name, required=True, type=lambda p: Path(p).resolve())
|
|
parser.add_argument('--width', required=True, type=int)
|
|
args = parser.parse_args()
|
|
if args.output.exists() or not args.output.is_relative_to((repo / 'build').resolve()) or args.width <= 0:
|
|
parser.error('use a fresh JSON output under build/ and a positive width')
|
|
names = sorted(p.name for p in args.reference.glob('output-*.rgba16f'))
|
|
if not names or names != sorted(p.name for p in args.candidate.glob('output-*.rgba16f')):
|
|
raise RuntimeError('Output sets differ or are empty')
|
|
rows = []
|
|
for name in names:
|
|
a = np.fromfile(args.reference / name, '<f2')
|
|
b = np.fromfile(args.candidate / name, '<f2')
|
|
if a.size != b.size or a.size % (args.width * 4):
|
|
raise RuntimeError('Invalid output dimensions')
|
|
indices = np.flatnonzero(a.view('<u2') != b.view('<u2'))
|
|
pixels = indices // 4
|
|
finite = np.isfinite(a) & np.isfinite(b)
|
|
max_difference = float(np.max(np.abs(a[finite].astype('f4') - b[finite].astype('f4')))) if finite.any() else None
|
|
rows.append(dict(frame=name, changedComponents=int(indices.size),
|
|
channels=[int(np.count_nonzero(indices % 4 == channel)) for channel in range(4)],
|
|
maxAbsoluteDifference=max_difference,
|
|
nonfiniteReference=int(np.count_nonzero(~np.isfinite(a))),
|
|
nonfiniteCandidate=int(np.count_nonzero(~np.isfinite(b))),
|
|
finite=bool(np.isfinite(a).all() and np.isfinite(b).all()),
|
|
boundsXY=[int((pixels % args.width).min()), int((pixels // args.width).min()),
|
|
int((pixels % args.width).max()), int((pixels // args.width).max())] if indices.size else None))
|
|
report = dict(exact=all(row['changedComponents'] == 0 for row in rows), rows=rows,
|
|
changedFrames=sum(row['changedComponents'] > 0 for row in rows),
|
|
changedComponents=sum(row['changedComponents'] for row in rows),
|
|
maximumAbsoluteDifference=max((row['maxAbsoluteDifference'] for row in rows if row['maxAbsoluteDifference'] is not None), default=None),
|
|
gameAccessed=False)
|
|
with args.output.open('x') as stream:
|
|
json.dump(report, stream, indent=2, allow_nan=False)
|
|
print(json.dumps({key: value for key, value in report.items() if key != 'rows'}, indent=2))
|
|
|
|
|
|
if __name__ == '__main__':
|
|
main()
|