diff --git a/SYCL/Regression/host_unified_memory.cpp b/SYCL/Regression/host_unified_memory.cpp
new file mode 100644
index 0000000000..586149b9c2
--- /dev/null
+++ b/SYCL/Regression/host_unified_memory.cpp
@@ -0,0 +1,48 @@
+// REQUIRES: gpu && linux
+// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
+// RUN: %GPU_RUN_PLACEHOLDER SYCL_HOST_UNIFIED_MEMORY=1 %t.out
+
+#include <CL/sycl.hpp>
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+
+using namespace cl::sycl;
+
+static buffer<char, 1> *inBufP = nullptr;
+
+int main(int argc, char *argv[]) {
+  queue Q;
+  auto BE =
+      (bool)(Q.get_device()
+                 .template get_info<sycl::info::device::opencl_c_version>()
+                 .empty())
+          ? "L0"
+          : "OpenCL";
+  device dev = Q.get_device();
+  size_t max_compute_units = dev.get_info<info::device::max_compute_units>();
+  printf("Device: %s max_compute_units %zu, Backend: %s\n",
+         dev.get_info<info::device::name>().c_str(), max_compute_units, BE);
+
+  size_t size = 32;
+  void *p;
+  posix_memalign(&p, 8, size);
+  property::buffer::use_host_ptr prop_use_host_ptr;
+  inBufP = new buffer<char, 1>((char *)p, size, prop_use_host_ptr);
+
+  int iters = 3;
+  for (int r = 0; r < iters; r++) {
+    printf("Start iter %d\n", r);
+    { host_accessor InhostAcc{*inBufP, write_only}; }
+    Q.submit([&](handler &cgh) {
+      accessor inAcc{*inBufP, cgh, read_write};
+      cgh.single_task<class X5>([=]() { (void)inAcc; });
+    });
+    Q.wait();
+    printf("End iter\n");
+  }
+  delete inBufP;
+  free(p);
+
+  return 0;
+}