add timeout for the mhp-bench and add jira issue IDs`

Author: haichangsi

CMakeLists.txt

@@ -223,7 +223,7 @@ function(add_mhp_ctest_impl)
   if(DEFINED AMC_TIMEOUT)
     set_tests_properties(${AMC_TEST_NAME} PROPERTIES TIMEOUT ${AMC_TIMEOUT})
   else()
-    set_tests_properties(${AMC_TEST_NAME} PROPERTIES TIMEOUT 600)
+    set_tests_properties(${AMC_TEST_NAME} PROPERTIES TIMEOUT 60)
   endif()
 
   if(AMC_TESTLABEL)

benchmarks/gbench/common/distributed_vector.cpp

@@ -163,18 +163,20 @@ void check_reduce(T actual) {
   }
 }
 
-// template <class... BackendT> static void Reduce_DR(benchmark::State &state) {
-//   T actual{};
-//   xhp::distributed_vector<T, BackendT...> src(default_vector_size, fill);
-//   Stats stats(state, sizeof(T) * src.size(), 0);
-//   for (auto _ : state) {
-//     for (std::size_t i = 0; i < default_repetitions; i++) {
-//       stats.rep();
-//       actual = xhp::reduce(src);
-//     }
-//   }
-//   check_reduce(actual);
-// }
+template <class... BackendT>
+[[maybe_unused]] static void Reduce_DR(benchmark::State &state) {
+  T actual{};
+  xhp::distributed_vector<T, BackendT...> src(default_vector_size, fill);
+  Stats stats(state, sizeof(T) * src.size(), 0);
+  for (auto _ : state) {
+    for (std::size_t i = 0; i < default_repetitions; i++) {
+      stats.rep();
+      actual = xhp::reduce(src);
+    }
+  }
+  check_reduce(actual);
+}
+// disabled due to DRA-134
 // DR_BENCHMARK(Reduce_DR);
 
 #ifdef DRISHMEM

benchmarks/gbench/mhp/CMakeLists.txt

@@ -22,6 +22,7 @@ add_executable(
   mpi.cpp)
 # cmake-format: on
 
+# disabled with SYCL due to DRA-135
 if(NOT ENABLE_SYCL)
   target_sources(mhp-bench PRIVATE mdspan.cpp)
 endif()
@@ -66,12 +67,12 @@ if(CMAKE_BUILD_TYPE STREQUAL "Release" AND NOT MPI_IMPL STREQUAL "openmpi")
   # intermittent fails with: ONEAPI_DEVICE_SELECTOR=opencl:cpu mpirun -n 1
   # ./mhp-bench --vector-size 30000 --rows 100 --columns 100 --check
   add_mhp_ctest(
-    NAME mhp-bench TARGS --vector-size 30000 --rows 100 --columns 100 --check
-    --benchmark_filter=-FFT3D.*)
+    NAME mhp-bench TIMEOUT 200 TARGS --vector-size 30000 --rows 100 --columns
+    100 --check --benchmark_filter=-FFT3D.*)
   if(ENABLE_SYCL)
     add_mhp_ctest(
-      NAME mhp-bench SYCL TARGS --vector-size 30000 --rows 100 --columns 100
-      --check --benchmark_filter=-.*DPL.*)
+      NAME mhp-bench TIMEOUT 200 SYCL TARGS --vector-size 30000 --rows 100
+      --columns 100 --check --benchmark_filter=-.*DPL.*)
   endif()
 endif()
 

benchmarks/gbench/mhp/shallow_water.cpp

@@ -1286,7 +1286,8 @@ int main(int argc, char *argv[]) {
 
 static void ShallowWater_DR(benchmark::State &state) {
 
-  int n = 1400;
+  // changed from 1400 to reduce CI time
+  int n = 140;
   std::size_t nread, nwrite, nflop;
   ShallowWater::calculate_complexity(n, n, nread, nwrite, nflop);
   Stats stats(state, nread, nwrite, nflop);

benchmarks/gbench/mhp/stencil_2d.cpp

@@ -465,50 +465,49 @@ DR_BENCHMARK(Stencil2D_Reference);
 // Distributed vector of floats. Granularity ensures segments contain
 // whole rows. Explicitly process segments SPMD-style with SYCL
 //
-// static void Stencil2D_SegmentedSYCL_DR(benchmark::State &state) {
-//   auto s = default_shape();
-//   auto rows = s[0];
-//   auto cols = s[1];
-
-//   if (rows == 0) {
-//     return;
-//   }
-
-//   auto dist = dr::mhp::distribution().halo(cols).granularity(cols);
-//   dr::mhp::distributed_vector<T> a(rows * cols, init_val, dist);
-//   dr::mhp::distributed_vector<T> b(rows * cols, init_val, dist);
-//   Stats stats(state, sizeof(T) * a.size(), sizeof(T) * b.size());
-
-//   // fails on devcloud
-//   // Checker checker;
-//   auto in =
-//       dr::mhp::local_segment(rng::subrange(a.begin() + cols, a.end() -
-//       cols));
-//   auto out =
-//       dr::mhp::local_segment(rng::subrange(b.begin() + cols, b.end() -
-//       cols));
-//   auto size = rng::size(in);
-//   assert(size % cols == 0);
-//   auto row_slice = size / cols;
-
-//   auto q = dr::mhp::sycl_queue();
-//   sycl::range global(row_slice, cols - 2);
-
-//   for (auto _ : state) {
-//     for (std::size_t s = 0; s < stencil_steps; s++) {
-//       stats.rep();
-//       auto op = [=](auto it) {
-//         stencil_1darray_op(in, out, cols, it[0], it[1] + 1);
-//       };
-//       dr::mhp::halo(stencil_steps % 2 ? b : a).exchange();
-//       q.parallel_for(sycl::range(row_slice, cols - 2), op).wait();
-//       std::swap(in, out);
-//     }
-//     // fails on devcloud
-//     // checker.check(stencil_steps % 2 ? b : a);
-//   }
-// }
+[[maybe_unused]] static void
+Stencil2D_SegmentedSYCL_DR(benchmark::State &state) {
+  auto s = default_shape();
+  auto rows = s[0];
+  auto cols = s[1];
+
+  if (rows == 0) {
+    return;
+  }
+
+  auto dist = dr::mhp::distribution().halo(cols).granularity(cols);
+  dr::mhp::distributed_vector<T> a(rows * cols, init_val, dist);
+  dr::mhp::distributed_vector<T> b(rows * cols, init_val, dist);
+  Stats stats(state, sizeof(T) * a.size(), sizeof(T) * b.size());
+
+  // fails on devcloud
+  // Checker checker;
+  auto in =
+      dr::mhp::local_segment(rng::subrange(a.begin() + cols, a.end() - cols));
+  auto out =
+      dr::mhp::local_segment(rng::subrange(b.begin() + cols, b.end() - cols));
+  auto size = rng::size(in);
+  assert(size % cols == 0);
+  auto row_slice = size / cols;
+
+  auto q = dr::mhp::sycl_queue();
+  sycl::range global(row_slice, cols - 2);
 
+  for (auto _ : state) {
+    for (std::size_t s = 0; s < stencil_steps; s++) {
+      stats.rep();
+      auto op = [=](auto it) {
+        stencil_1darray_op(in, out, cols, it[0], it[1] + 1);
+      };
+      dr::mhp::halo(stencil_steps % 2 ? b : a).exchange();
+      q.parallel_for(sycl::range(row_slice, cols - 2), op).wait();
+      std::swap(in, out);
+    }
+    // fails on devcloud
+    // checker.check(stencil_steps % 2 ? b : a);
+  }
+}
+// disabled due to DRA-136
 // DR_BENCHMARK(Stencil2D_SegmentedSYCL_DR);
 
 #endif // SYCL_LANGUAGE_VERSION