Merge branch 'fft-admit' into rc175: FFT indexer construction waits for GPU work beside it before failing

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EizBKhTqrAYago9KJG3dA3

common/CUDAWrapper.cpp

@@ -28,27 +28,50 @@ std::string get_gpu_description() {
 }
 
 namespace {
-    std::mutex gpu_work_beside_mutex;
-    std::condition_variable gpu_work_beside_done;
-    int gpu_work_beside = 0;
+    // How many of one kind of GPU work are running, for a thread to wait until none is.
+    struct GPUWorkCount {
+        std::mutex m;
+        std::condition_variable none;
+        int n = 0;
+        void Add() {
+            std::lock_guard lock(m);
+            ++n;
+        }
+        void Remove() {
+            {
+                std::lock_guard lock(m);
+                --n;
+            }
+            none.notify_all();
+        }
+    };
+    GPUWorkCount gpu_work_beside, gpu_image_passes;
 }
 
 GPUWorkBeside::GPUWorkBeside() {
-    std::lock_guard lock(gpu_work_beside_mutex);
-    ++gpu_work_beside;
+    gpu_work_beside.Add();
 }
 
 GPUWorkBeside::~GPUWorkBeside() {
-    {
-        std::lock_guard lock(gpu_work_beside_mutex);
-        --gpu_work_beside;
-    }
-    gpu_work_beside_done.notify_all();
+    gpu_work_beside.Remove();
 }
 
 bool wait_for_gpu_work_beside(std::chrono::seconds timeout) {
-    std::unique_lock lock(gpu_work_beside_mutex);
-    return gpu_work_beside_done.wait_for(lock, timeout, [] { return gpu_work_beside == 0; });
+    std::unique_lock lock(gpu_work_beside.m);
+    return gpu_work_beside.none.wait_for(lock, timeout, [] { return gpu_work_beside.n == 0; });
+}
+
+GPUImagePass::GPUImagePass() {
+    gpu_image_passes.Add();
+}
+
+GPUImagePass::~GPUImagePass() {
+    gpu_image_passes.Remove();
+}
+
+void wait_for_gpu_image_passes() {
+    std::unique_lock lock(gpu_image_passes.m);
+    gpu_image_passes.none.wait(lock, [] { return gpu_image_passes.n == 0; });
 }
 
 #ifndef JFJOCH_USE_CUDA