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

scripts/d4r_config.py

@@ -92,14 +92,14 @@ def main(argv):
         env["D4R_RR_ENABLE"] = "1" if rr_enabled else "0"
     rr_model = get("RayReconstruction", "Model")
     if rr_model is not None:
-        rr_presets = {"DEFAULT": 0, "D": 4, "E": 5}
+        rr_presets = {"DEFAULT": 0, "D": 4, "E": 5, "F": 6}   # F: DLSS 4.5, nvngx_dlssd.dll 310.9 or later
         rr_model = rr_model.upper()
         if rr_model in rr_presets:
             env["D4R_RR_PRESET"] = str(rr_presets[rr_model])
-        elif rr_model in ("0", "4", "5"):
+        elif rr_model in ("0", "4", "5", "6"):
             env["D4R_RR_PRESET"] = rr_model
         else:
-            fail(f"[RayReconstruction] Model {rr_model!r}: use auto, default, D or E")
+            fail(f"[RayReconstruction] Model {rr_model!r}: use auto, default, D, E or F")
 
     age = get("Latency", "FrameAge")
     if age is not None:

scripts/run_rr_engine_compare.sh

@@ -6,7 +6,7 @@ set -euo pipefail
 #
 # usage: run_rr_engine_compare.sh OUTPUT_DIR MODEL_DIR
 # environment: as scripts/run_rr_guide_compare.sh (D4R_PROTON_DIR, D4R_RUNTIME_DIR, D4R_VKD3D_DIR, D4R_ZLUDA_DIR,
-#   D4R_HARNESS_SIZE, D4R_HARNESS_FRAMES); D4R_RR_PRESET 5 = E (default; the model must match)
+#   D4R_HARNESS_SIZE, D4R_HARNESS_FRAMES); D4R_RR_PRESET 5 = E (default; the model must match), 6 = F with D4R_RR_DLL=<310.9 nvngx_dlssd.dll>
 ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
 OUTPUT_DIR="$(realpath -m "${1:?usage: run_rr_engine_compare.sh OUTPUT_DIR MODEL_DIR}")"
 MODEL_DIR="$(realpath "${2:?usage: run_rr_engine_compare.sh OUTPUT_DIR MODEL_DIR}")"
@@ -22,6 +22,8 @@ mkdir -p "$APP_DIR"
 cp -f "$ROOT/build/d3d12_dlss_harness.exe" "$ROOT/build/d4r_nvngx.dll" \
       "$RUNTIME_DIR/bin/nvngx_dlssd.dll" "$RUNTIME_DIR/bin/nvngx_dlss.dll" \
       "$VKD3D_DIR/d3d12.dll" "$VKD3D_DIR/d3d12core.dll" "$NVAPI_DIR/nvapi64.dll" "$APP_DIR/"
+# D4R_RR_DLL: another nvngx_dlssd.dll than the runtime's (preset F needs 310.9: D4R_RR_PRESET=6 with a compile_rr2.py model)
+[[ -z "${D4R_RR_DLL:-}" ]] || cp -f "$D4R_RR_DLL" "$APP_DIR/nvngx_dlssd.dll"
 
 export D4R_HARNESS_SKIP_BUILD=1 D4R_HARNESS_APP_DIR="$APP_DIR" D4R_HARNESS_NGX_DLL="$APP_DIR/d4r_nvngx.dll"
 export D4R_RUNTIME_DIR="$RUNTIME_DIR" D4R_ZLUDA_DIR="$ZLUDA_DIR"
@@ -32,9 +34,11 @@ export D4R_HARNESS_FEATURE_ID=13 D4R_HARNESS_SAVE_FRAMES=1
 export D4R_HARNESS_GUIDE_FORMATS="${D4R_HARNESS_GUIDE_FORMATS:-rgba16f,rgba16f,rgba8unorm,rgba8unorm}"
 export WINEDLLOVERRIDES="d3d12=n,b;d3d12core=n,b"
 
+# a preset F model also changes the shim's mapping of the hints (every one but 4 and 5 runs F then)
+F_DIR="$OUTPUT_DIR/no-preset-f-model"; grep -q "^preset 6" "$MODEL_DIR/rr.txt" && F_DIR="$MODEL_DIR"
 run_one() { # run_one LABEL ENGINE
   printf '\n== %s\n' "$1" >&2
-  D4R_ENGINE="$2" D4R_ENGINE_RR_SYNC=1 D4R_ENGINE_MODEL_RR_E_DIR="$MODEL_DIR" D4R_ENGINE_MODEL_RR_D_DIR="$MODEL_DIR" \
+  D4R_ENGINE="$2" D4R_ENGINE_RR_SYNC=1 D4R_ENGINE_MODEL_RR_E_DIR="$MODEL_DIR" D4R_ENGINE_MODEL_RR_D_DIR="$MODEL_DIR" D4R_ENGINE_MODEL_RR_F_DIR="$F_DIR" \
     "$ROOT/scripts/run_d3d12_dlss_harness_proton.sh" "$PROTON_DIR" "$OUTPUT_DIR/$1.raw" \
       "$FRAMES" "$IN_W" "$IN_H" "$OUT_W" "$OUT_H" > "$OUTPUT_DIR/$1.log" 2>&1 || true
   grep -E 'engine backend|engine frame|Ray Reconstruction|output read back|rror' "$OUTPUT_DIR/$1.log" | head -20 || true