From e611c5b27d2230ebdd5a291c2d44c03a1d151eae Mon Sep 17 00:00:00 2001
From: PancakeTAS <pancake@mgnet.work>
Date: Sun, 26 Apr 2026 01:10:19 +0200
Subject: [PATCH] feat(bindless): Split synchronization into another thread in
 debug tool

I think RenderDoc attaches a fence or something, so the dispatch never returns. And ontop of this it looks like timeline semaphores can be signaled back to a lower value, when shared across processes
---
 lsfg-vk-cli/src/tools/benchmark.cpp |  2 +-
 lsfg-vk-cli/src/tools/debug.cpp     | 19 +++++++++++++------
 2 files changed, 14 insertions(+), 7 deletions(-)

diff --git a/lsfg-vk-cli/src/tools/benchmark.cpp b/lsfg-vk-cli/src/tools/benchmark.cpp
index 88137e2..ebd826d 100644
--- a/lsfg-vk-cli/src/tools/benchmark.cpp
+++ b/lsfg-vk-cli/src/tools/benchmark.cpp
@@ -138,7 +138,7 @@ int benchmark::run(const Options& opts) {
 
                 auto success = sync.wait(vk, idx++);
                 if (!success)
-                    throw ls::error("failed to wait for frame");
+                    throw ls::error("Failed to wait for frame");
 
                 total_frames++;
                 generated_frames++;
diff --git a/lsfg-vk-cli/src/tools/debug.cpp b/lsfg-vk-cli/src/tools/debug.cpp
index b7b9c59..8c850f5 100644
--- a/lsfg-vk-cli/src/tools/debug.cpp
+++ b/lsfg-vk-cli/src/tools/debug.cpp
@@ -24,6 +24,7 @@
 #include <iostream>
 #include <optional>
 #include <string>
+#include <thread>
 #include <vector>
 
 #include <dlfcn.h>
@@ -185,6 +186,7 @@ int debug::run(const Options& opts) {
         // Render destination images
         const uint32_t total{static_cast<uint32_t>(opts.multiplier) - 1U};
 
+        size_t idx{1};
         for (size_t j = 0; j < paths.size(); j++) {
             uploadDDS(vk, source, paths.at(j).string(), j % 2);
 
@@ -192,19 +194,24 @@ int debug::run(const Options& opts) {
                 rdoc_api->StartFrameCapture(rdoc_device, nullptr);
             }
 
-            const size_t idx{(j + 1) * total * 2};
-            sync.signal(vk, idx - 1);
+            std::thread signal_thread{[&sync, &vk, &idx, total] {
+                for (size_t i = 0; i < total; i++) {
+                    sync.signal(vk, idx++);
+
+                    auto success = sync.wait(vk, idx++);
+                    if (!success)
+                        throw ls::error("Failed to wait for frame");
+                }
+            }};
 
             lsfgvk_ctx.dispatch(total);
 
-            auto success = sync.wait(vk, idx);
-            if (!success)
-                throw ls::error("Failed to wait for frame");
-
             if (rdoc_api) {
                 lsfgvk_ctx.idle();
                 rdoc_api->EndFrameCapture(rdoc_device, nullptr);
             }
+
+            signal_thread.join();
         }
 
         // Wait for idle