CLAUDE: gpuTrace(-Dtp.trace=1) on uncertain GpuQuad methods — JCuda/JNA oracle comparison

Add GpuQuad.gpuTrace(m) printing "[GPUTRACE] "+getClass().getSimpleName()+"."+m (off unless
-Dtp.trace=1). Instrument the un-overridden GPU methods (potential oracle gaps): getCltData,
presentCltData, eraseGpuCorrs, execCorr2D (bundled), readbackTasks, setFullFrameImages, getCorrTdData,
getCorrIndices, getCorrComboIndices, getExtra, getTextureIndices, getRBGA, execRBGA, execTextures.
Since GpuQuadJna extends GpuQuad, the trace prints "GpuQuad.X" under JCuda and "GpuQuadJna.X" if a JNA
run falls through to one (= coverage gap) -> reveals oracle's real GPU usage before any NPE.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

src/main/java/com/elphel/imagej/gpu/GpuQuad.java

@@ -490,6 +490,12 @@ public class GpuQuad{ // quad camera description
 		texture_stride_rgba = (int)(device_stride[0] / Sizeof.FLOAT);
 	}
 	
+	// Diagnostic trace (set -Dtp.trace=1). getClass().getSimpleName() prints "GpuQuad" under JCuda and
+	// "GpuQuadJna" if a JNA run falls through to an un-overridden base method -> reveals oracle's actual
+	// GPU-method usage (run JCuda first) and any JNA coverage gap, before an NPE. By Claude 2026-06-25.
+	public static boolean GPU_TRACE = "1".equals(System.getProperty("tp.trace"));
+	protected void gpuTrace(String m) { if (GPU_TRACE) System.out.println("[GPUTRACE] " + getClass().getSimpleName() + "." + m); }
+
 	// Backend selector (architecture B). Default JCUDA; set -Dtp.backend=jna to use the native
 	// (libtileproc.so via JNA) backend GpuQuadJna instead. JNA mode never initializes JCuda.
 	// Validate by running the same workflow both ways and diffing saved outputs.
@@ -995,6 +1001,7 @@ public class GpuQuad{ // quad camera description
 			int     num_sensors,
 			boolean use_aux    // while is it in class member? - just to be able to free
 			) {
+		gpuTrace("readbackTasks");
 		num_task_tiles = num_tasks;
 		TpTask [] tile_tasks = new TpTask[num_task_tiles];
 		int task_size = getTaskSize();
@@ -1060,6 +1067,7 @@ public class GpuQuad{ // quad camera description
 
 	public int [] getTextureIndices()
 	{
+		gpuTrace("getTextureIndices");
 		float [] ftexture_indices_len = new float[1];
 		cuMemcpyDtoH(Pointer.to(ftexture_indices_len), gpu_texture_indices_len,  1 * Sizeof.FLOAT);
 		int num_tiles =      Float.floatToIntBits(ftexture_indices_len[0]);
@@ -1234,6 +1242,7 @@ public class GpuQuad{ // quad camera description
 	*/
 	public float [][] getCltData( // only for color=0
 			boolean use_ref){
+		gpuTrace("getCltData");
 		CUdeviceptr [] gpu_sel_clt_h = 	use_ref ? gpu_clt_ref_h : gpu_clt_h;
 		int [] wh = use_ref ? gpu_clt_ref_wh : gpu_clt_wh;
 		int tilesX =  wh[0] / GPUTileProcessor.DTT_SIZE;
@@ -1257,6 +1266,7 @@ public class GpuQuad{ // quad camera description
 	}
 	
 	public float [][] presentCltData(boolean use_ref){
+		gpuTrace("presentCltData");
 		float [][] fclt = getCltData(use_ref);
 		float [][] pfclt = new float [fclt.length][fclt[0].length];
 		int [] wh = use_ref ? gpu_clt_ref_wh : gpu_clt_wh;
@@ -1473,6 +1483,7 @@ public class GpuQuad{ // quad camera description
 			int                       corr_mask,  // which correlation pairs to generate (maybe later - reduce size from 15x15)
 			//    			final int                 threadsMax,  // maximal number of threads to launch
 			final int                 debugLevel) {
+		gpuTrace("setFullFrameImages");
 		int tilesX =  img_width / GPUTileProcessor.DTT_SIZE;
 		int tilesY =  img_height / GPUTileProcessor.DTT_SIZE;
 		float [] target_disparities = new float [tilesX * tilesY];
@@ -2392,6 +2403,7 @@ public class GpuQuad{ // quad camera description
 			double [] scales,
 			double fat_zero,
 			int corr_radius) {
+		gpuTrace("execCorr2D");
 		if (this.gpuTileProcessor.GPU_CORRELATE2D_kernel == null)
 		{
 			IJ.showMessage("Error", "No GPU kernel: GPU_CORRELATE2D_kernel");
@@ -2692,6 +2704,7 @@ public class GpuQuad{ // quad camera description
 			double    min_agree,          // minimal number of channels to agree on a point (real number to work with fuzzy averages)
 			boolean   dust_remove,
 			int       keep_weights) {
+		gpuTrace("execRBGA");
 		if (GPUTileProcessor.USE_DS_DP) {
 			execRBGA_DP(
 					color_weights,  // double [] color_weights,
@@ -3261,6 +3274,7 @@ public class GpuQuad{ // quad camera description
 			boolean   calc_extra,
 			boolean   linescan_order
 			) {
+		gpuTrace("execTextures");
 		if (GPUTileProcessor.USE_DS_DP) {
 			execTextures_DP(
 					color_weights,
@@ -3983,6 +3997,7 @@ public class GpuQuad{ // quad camera description
 	}
 
 	public int [] getCorrIndices() {
+		gpuTrace("getCorrIndices");
 		int [] inum_corrs = new int[1];
 		cuMemcpyDtoH(Pointer.to(inum_corrs), gpu_num_corr_tiles,  1 * Sizeof.INT);
 		int num_corrs =      inum_corrs[0];
@@ -3994,6 +4009,7 @@ public class GpuQuad{ // quad camera description
 	
 	
 	public float [] getCorrTdData(){
+		gpuTrace("getCorrTdData");
 		int corr_size_td = 4 * GPUTileProcessor.DTT_SIZE * GPUTileProcessor.DTT_SIZE;
 		float [] cpu_corrs = new float [ num_corr_tiles * corr_size_td];
 		CUDA_MEMCPY2D copyD2H =   new CUDA_MEMCPY2D();
@@ -4040,6 +4056,7 @@ public class GpuQuad{ // quad camera description
 
 
 	public int [] getCorrComboIndices() {
+		gpuTrace("getCorrComboIndices");
 		//        	float [] fnum_corrs = new float[1];
 		//        	cuMemcpyDtoH(Pointer.to(fnum_corrs), gpu_num_corr_tiles,  1 * Sizeof.FLOAT);
 		//        	int num_corrs =      Float.floatToIntBits(fnum_corrs[0]);
@@ -4077,6 +4094,7 @@ public class GpuQuad{ // quad camera description
 	
 	
 	public float [][] getCorr2D(int corr_rad){
+		gpuTrace("eraseGpuCorrs");
 		int corr_size = (2 * corr_rad + 1) * (2 * corr_rad + 1);
 		float [] cpu_corrs = new float [ num_corr_tiles * corr_size];
 		CUDA_MEMCPY2D copyD2H =   new CUDA_MEMCPY2D();
@@ -4132,6 +4150,7 @@ public class GpuQuad{ // quad camera description
 	 * @return [ num_cams*(num_colors+1)][tilesX*tilesY] array for macro generation
 	 */
 	public float [][] getExtra(){
+		gpuTrace("getExtra");
 		int [] texture_indices = getTextureIndices();
 		int num_tile_extra = num_cams*(num_colors+1);
 		float [] diff_rgb_combo = new float[texture_indices.length * num_tile_extra];
@@ -4181,6 +4200,7 @@ public class GpuQuad{ // quad camera description
 		copy_rbga.srcXInBytes =     4 * Sizeof.FLOAT;
 
 		for (int ncol = 0; ncol<= num_colors; ncol++ ) {
+		gpuTrace("getRBGA");
 			copy_rbga.dstHost =     Pointer.to(rslt[ncol]);
 			copy_rbga.srcY =        4 + (woi.height +GPUTileProcessor.DTT_SIZE) * ncol;
 			cuMemcpy2D(copy_rbga); // run copy