CLAUDE: tp_proc_get_tasks — DtoH readback of gpu_ftasks (for updateTasks)

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

jna/tp_jna.cpp

@@ -656,6 +656,9 @@ int tp_proc_set_center_image(TpProc* p, const float* d){ if(!p)return -1; cuCtxS
 int tp_proc_set_tasks(TpProc* p, const float* ftasks, int ntiles, int total_floats){ if(!p)return -1; cuCtxSetCurrent(p->mod->ctx);
   if(p->ftasks) cudaFree(p->ftasks); cudaMalloc((void**)&p->ftasks,(size_t)total_floats*sizeof(float));
   p->ntiles=ntiles; return cudaMemcpy(p->ftasks,ftasks,(size_t)total_floats*sizeof(float),cudaMemcpyHostToDevice)==cudaSuccess?0:-2; }
+// Read tasks back (DtoH) after calculate_tiles_offsets updated them in place (for GpuQuad.updateTasks).
+int tp_proc_get_tasks(TpProc* p, float* out, int total_floats){ if(!p||!p->ftasks)return -1; cuCtxSetCurrent(p->mod->ctx);
+  return cudaMemcpy(out,p->ftasks,(size_t)total_floats*sizeof(float),cudaMemcpyDeviceToHost)==cudaSuccess?0:-2; }
 
 // Upload to a named __constant__ symbol in the module (LPF filters: lpf_data / lpf_corr / lpf_rb_corr).
 // Mirrors GpuQuad.setLpfRbg/setLpfCorr (cuModuleGetGlobal + cuMemcpyHtoD), against the native module.