engine/rr: native preset F (DLSS 4.5 Ray Reconstruction) and the reference's fused roundings

- engine/compile_rr2.py, runtime_rr2.inc, engine/rr/rr2_m.comp, exposure_f.comp,
  gen_rr2_enc0.py, rr2_pack.py, rr2_ref.py: the rr2_* model (nvngx_dlssd.dll 310.9's
  third network) as d4r's own shaders - exposure, enc0 gather/block split, enc1-5 and
  dec4-0 through one matrix shader, translated output passes; weights packed from the
  DLL, argument blocks from the capture
- ptx2rr: match the driver's fused roundings - opaque-zero OR around cvt.rn.f16.f32
  against v_fma_mix, half-away-from-zero formatted stores, UNORM 10:10:10:2 as RGBA32F
  of q/1023, OCML tanhF with the AMDGPU exp lowering, binary16 fma and fetch roundings;
  --probe for register bisection against ZLUDA
- dec0_m/swin_m/enc0_*: RR_MIX build option (D4R_RR_MIX=1), one v_fma_mix rounding per
  32-deep step, ~5% faster, not bit-exact; default build stays exact
- runtime_rr.inc: RModel/REngine dispatch to the D4RR0002 model; replay_rr: RGBA32F and
  A2B10G10R1010 texel sizes, roughnessPacked bit, larger debug staging
- shim: engine\rr-f model slot, 310.9 version probe so preset F is only claimed when the
  CUDA path has it; d4r.ini / d4r_config.py Model = F; run_rr_engine_compare.sh D4R_RR_DLL
- kernels/rr/swin_native.h + rr_swin_native_test.cpp: HIP port of encoder 1 for the
  FP8-product and rounding cost measurements
- engine/rr/README: preset F's block, options, capture-by-capture results, timings and
  the open camera-matrix difference

kernels/tools/extract_dlss_ptx.py

@@ -48,6 +48,23 @@ def lz4_block(src: bytes, size: int) -> bytes:
     return bytes(dst[:size])
 
 
+ZSTD_MAGIC = b'\x28\xb5\x2f\xfd'
+
+
+def zstd_frame(src: bytes) -> bytes:
+    """One zstd frame, through the zstandard module when it is installed and the zstd tool otherwise."""
+    try:
+        import zstandard
+        return zstandard.ZstdDecompressor().decompressobj().decompress(src)
+    except ImportError:
+        import subprocess
+        # the entry is padded: the tool decodes the frame and then complains about the bytes after it
+        out = subprocess.run(['zstd', '-dqc'], input=src, stdout=subprocess.PIPE, stderr=subprocess.DEVNULL).stdout
+        if not out:
+            raise RuntimeError('zstd could not decode a PTX entry')
+        return out
+
+
 def fatbins(data: bytes):
     """Yields (offset, bytes) of every fatbin container (magic, version 1) in file order."""
     magic = struct.pack('<I', FATBIN_MAGIC)
@@ -71,7 +88,10 @@ def ptx_entries(fatbin: bytes):
         decompressed = struct.unpack_from('<Q', fatbin, offset + 56)[0] if entry_header >= 64 else 0
         payload = fatbin[offset + entry_header:offset + entry_header + entry_size]
         if kind == 1:  # PTX
-            yield lz4_block(payload, decompressed) if (flags & 0x2000 and decompressed) else payload
+            if payload[:4] == ZSTD_MAGIC:   # nvngx_dlssd.dll 310.9 stores its rr2_* modules as zstd frames
+                yield zstd_frame(payload)
+            else:
+                yield lz4_block(payload, decompressed) if (flags & 0x2000 and decompressed) else payload
         offset += entry_header + entry_size