fix: do not access UMD-managed SVM-allocation while in-use

It is illegal with UMD-managed migrations to access a SVM-allocation from CPU side while it is still in-use on GPU side. A CPU-side access fetches the data from GPU-side and then any subsequent GPU-side modification will not be observable on the CPU-side. With UMD-managed migrations there is no way to tell the GPU that it should fetch the allocation back from CPU-side before modifying it. Signed-off-by: Maciej Bielski <[email protected]>
oneapi-src · Jan 22, 2025 · a78e94c · a78e94c
1 parent d6f60a2
commit a78e94c
Showing 1 changed file with 0 additions and 1 deletion.
diff --git a/conformance_tests/core/test_copy/src/test_copy_events.cpp b/conformance_tests/core/test_copy/src/test_copy_events.cpp
@@ -225,7 +225,6 @@ void RunGivenMemoryCopiesWithDependenciesWhenExecutingCommandListTest(
   // Verify Copy Waits for Signal
   lzt::event_host_synchronize(test.hEvent, UINT64_MAX);
   EXPECT_EQ(ZE_RESULT_NOT_READY, zeEventQueryStatus(hEvent1));
-  EXPECT_NE(0, memcmp(src_buffer, dst_buffer, test.size));
 
   EXPECT_EQ(ZE_RESULT_SUCCESS, zeEventHostSignal(hEvent1));