From 315e8f3def4c3a502ef9966ad66f0e252f260f4e Mon Sep 17 00:00:00 2001
From: AI agent <ai@example.invalid>
Date: Mon, 6 Apr 2026 15:15:42 +0000
Subject: [PATCH] Add first integration queue round-trip test

---
 test/CMakeLists.txt                           |  2 +-
 .../queueBufferRoundTrip.cpp                  | 65 +++++++++++++++++++
 2 files changed, 66 insertions(+), 1 deletion(-)
 create mode 100644 test/integ/queueBufferRoundTrip/queueBufferRoundTrip.cpp

diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt
index 2994311b0..786c93c46 100644
--- a/test/CMakeLists.txt
+++ b/test/CMakeLists.txt
@@ -83,4 +83,4 @@ function(alpaka_add_test_target target_prefix folder)
 endfunction()
 
 alpaka_add_test_target(unit_test_ ${CMAKE_CURRENT_SOURCE_DIR}/unit/**)
-#alpaka_add_test_target(integ_test_ ${CMAKE_CURRENT_SOURCE_DIR}/integ/**)
+alpaka_add_test_target(integ_test_ ${CMAKE_CURRENT_SOURCE_DIR}/integ/**)
diff --git a/test/integ/queueBufferRoundTrip/queueBufferRoundTrip.cpp b/test/integ/queueBufferRoundTrip/queueBufferRoundTrip.cpp
new file mode 100644
index 000000000..2cded9e31
--- /dev/null
+++ b/test/integ/queueBufferRoundTrip/queueBufferRoundTrip.cpp
@@ -0,0 +1,65 @@
+/* Copyright 2026 OpenAI
+ * SPDX-License-Identifier: MPL-2.0
+ */
+
+#include <alpaka/alpaka.hpp>
+
+#include <catch2/catch_template_test_macros.hpp>
+#include <catch2/catch_test_macros.hpp>
+
+#include <iostream>
+
+using namespace alpaka;
+
+using TestApis = std::decay_t<decltype(onHost::allBackends(onHost::enabledApis, exec::enabledExecutors))>;
+
+struct AddOffsetKernel
+{
+    ALPAKA_FN_ACC void operator()(auto const& acc, alpaka::concepts::IMdSpan auto data, std::uint32_t offset) const
+    {
+        for(auto idx : onAcc::makeIdxMap(acc, onAcc::worker::threadsInGrid, onAcc::range::totalFrameSpecExtent))
+            data[idx.x()] += offset;
+    }
+};
+
+TEMPLATE_LIST_TEST_CASE("queue + memcpy + kernel round trip", "", TestApis)
+{
+    auto cfg = TestType::makeDict();
+    auto deviceSpec = cfg[object::deviceSpec];
+    auto exec = cfg[object::exec];
+
+    auto devSelector = onHost::makeDeviceSelector(deviceSpec);
+    if(!devSelector.isAvailable())
+    {
+        std::cout << "No device available for " << deviceSpec.getName() << std::endl;
+        return;
+    }
+
+    onHost::Device device = devSelector.makeDevice(0);
+    onHost::Queue queue = device.makeQueue(queueKind::blocking);
+
+    constexpr Vec extent = Vec{32u};
+    constexpr auto frameExtent = CVec<std::uint32_t, 8u>{};
+    constexpr std::uint32_t offset = 7u;
+
+    auto hostInput = onHost::alloc<std::uint32_t>(onHost::makeHostDevice(), extent);
+    auto deviceBuffer = onHost::alloc<std::uint32_t>(device, extent);
+    auto hostOutput = onHost::allocHostLike(deviceBuffer);
+
+    // This test intentionally keeps the data flow obvious:
+    // fill on host, copy to device, run one kernel, copy back, then validate the final values.
+    for(std::uint32_t i = 0; i < extent.x(); ++i)
+        hostInput[i] = i;
+    onHost::fill(queue, deviceBuffer, 99u);
+
+    onHost::memcpy(queue, deviceBuffer, hostInput);
+    queue.enqueue(exec, onHost::FrameSpec{extent / frameExtent, frameExtent}, KernelBundle{AddOffsetKernel{}, deviceBuffer, offset});
+    onHost::memcpy(queue, hostOutput, deviceBuffer);
+    onHost::wait(queue);
+
+    for(std::uint32_t i = 0; i < extent.x(); ++i)
+    {
+        CAPTURE(i);
+        CHECK(hostOutput[i] == i + offset);
+    }
+}