CLAUDE: JNA imclt ref_scene fix + one-shot ref/scene post-mortem renders (TEMP)

(1) GpuQuadJna.execImcltRbgAll now passes ref_scene -> tp_proc_exec_imclt(use_ref)
so renderFromTD(true) renders gpu_clt_ref, not gpu_clt (TpJna binding updated).

(2) TEMP post-mortem in CuasMotion.correlatePair (gate -Dtp.dbg.corrpair=1,
one-shot): after the inter-correlation, SAVE ref (gpu_clt_ref) + scene (gpu_clt)
CLT renders to the model dir via saveImagePlusInModelDirectory (persist past the
later crash; no window flood). DBG-REF blank/NaN => reference not loaded =>
explains the all-NaN CORR2D (inter corr needs both images). REMOVE after fix.

Needs the tile_processor_gpu imclt use_ref commit + libtileproc.so rebuild.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

src/main/java/com/elphel/imagej/cuas/CuasMotion.java

@@ -275,6 +275,7 @@ public class CuasMotion {
 	private double []               noise_map8 = null;
 	private int []                  noise_neibs = null;
 	private int                     debugLevel = 0;
+	private boolean                 corrpair_dbg_done = false; // TEMP post-mortem one-shot flag (By Claude 06/26/2026), remove with the -Dtp.dbg.corrpair probe
 	private GpuQuad                 gpuQuad = null;
 	private ImageDtt                image_dtt = null;
 	private int                     gpu_max_width;         
@@ -2904,8 +2905,31 @@ public class CuasMotion {
 			renderFromTD(false, "img"+dbg_suffix);
 		}
 		TDCorrTile [] tdCorrTiles = TDCorrTile.getFromGpu(gpuQuad);
+		// TEMP post-mortem (By Claude 06/26/2026): one-shot SAVE of reference + scene CLT renders (GPU->back)
+		// to localize the CORR2D-all-NaN divergence — is gpu_clt_ref actually populated? Both buffers are
+		// current here (ref from setRectilinearReferenceTD, scene from interRectilinearCorrTD). Saved (not
+		// shown) so they persist post-mortem; one-shot so no flood. Gate: -Dtp.dbg.corrpair=1. REMOVE after fix.
+		if (!corrpair_dbg_done && "1".equals(System.getProperty("tp.dbg.corrpair"))) {
+			corrpair_dbg_done = true;
+			String sfx = (dbg_suffix == null) ? "" : dbg_suffix;
+			saveTDRender(true,  "DBG-REF"+sfx);   // gpu_clt_ref render — blank/NaN => reference not loaded
+			saveTDRender(false, "DBG-IMG"+sfx);   // gpu_clt render (scene) — expected OK (matches SOURCE)
+		}
 		return tdCorrTiles;
 	}
+
+	// TEMP post-mortem helper (By Claude 06/26/2026): imclt the selected CLT buffer (ref/scene) back to an
+	// image and SAVE it to the model directory (persists after the run). REMOVE with the probe above.
+	private void saveTDRender(boolean use_reference, String title) {
+		gpuQuad.execImcltRbgAll(gpuQuad.getNumColors() <= 1, use_reference, null);
+		float [][] rbg = gpuQuad.getRBG(0); // single-camera rectilinear
+		ImagePlus imp = ShowDoubleFloatArrays.makeArrays(
+				rbg, gpuQuad.getImageWidth(), gpuQuad.getImageHeight(), title, null);
+		if (imp != null) {
+			parentCLT.saveImagePlusInModelDirectory(imp);
+			System.out.println("saveTDRender(): saved post-mortem render '"+title+"' (use_reference="+use_reference+")");
+		}
+	}
 	
 	public float [][] shiftAndRenderAccumulate(
 			CLTParameters       clt_parameters,

src/main/java/com/elphel/imagej/gpu/jna/GpuQuadJna.java

@@ -274,7 +274,7 @@ public class GpuQuadJna extends GpuQuad {
     // ---- inverse CLT -> RBG ----
     @Override public void execImcltRbgAll(boolean is_mono, boolean ref_scene, int[] wh) {
         ensureRbgCorr();
-        lib.tp_proc_exec_imclt(proc, 1); // apply_lpf = 1
+        lib.tp_proc_exec_imclt(proc, 1, ref_scene ? 1 : 0); // apply_lpf = 1; honor ref_scene (gpu_clt_ref). By Claude 06/26/2026.
     }
     @Override public float[][] getRBG(int ncam) {
         int out_width = getImageWidth(), out_height = getImageHeight();

src/main/java/com/elphel/imagej/gpu/jna/TpJna.java

@@ -83,7 +83,7 @@ public interface TpJna extends Library {
     /** Allocate imclt(RBG) + correlation buffers and store corr config. */
     int     tp_proc_setup_rbg_corr(Pointer proc, int numPairs, int s0,int s1,int s2,int s3,
                                    float cw0,float cw1,float cw2, int corrOutRad);
-    int     tp_proc_exec_imclt(Pointer proc, int applyLpf);
+    int     tp_proc_exec_imclt(Pointer proc, int applyLpf, int useRef);
     int     tp_proc_get_rbg(Pointer proc, int cam, float[] out);
     int     tp_proc_exec_corr2d(Pointer proc, double fatZero);
     /** de-pitch quad correlation; returns num_corr_combo. */