Merge various benchmark improvements

* Remove ELL format from benchmarks for heavily imbalanced matrices
* Fix some Hybrid conversion issues in OpenMP
* Fix IterationControl usage in conversion benchmark
* Remove CudaExecutor temporaries from SpMV formats in benchmark
* Output exception messages into JSON

Related PR: #812

benchmark/blas/blas.cpp

@@ -472,6 +472,12 @@ void apply_blas(const char *operation_name, std::shared_ptr<gko::Executor> exec,
     } catch (const std::exception &e) {
         add_or_set_member(test_case["blas"][operation_name], "completed", false,
                           allocator);
+        if (FLAGS_keep_errors) {
+            rapidjson::Value msg_value;
+            msg_value.SetString(e.what(), allocator);
+            add_or_set_member(test_case["blas"][operation_name], "error",
+                              msg_value, allocator);
+        }
         std::cerr << "Error when processing test case " << test_case << "\n"
                   << "what(): " << e.what() << std::endl;
     }

benchmark/conversions/conversions.cpp

@@ -88,16 +88,22 @@ void convert_matrix(const gko::LinOp *matrix_from, const char *format_to,
             matrix_to->copy_from(matrix_from);
         }
         add_or_set_member(conversion_case[conversion_name], "time",
-                          timer->compute_average_time(), allocator);
+                          ic.compute_average_time(), allocator);
         add_or_set_member(conversion_case[conversion_name], "repetitions",
-                          timer->get_num_repetitions(), allocator);
+                          ic.get_num_repetitions(), allocator);
 
         // compute and write benchmark data
         add_or_set_member(conversion_case[conversion_name], "completed", true,
                           allocator);
     } catch (const std::exception &e) {
         add_or_set_member(test_case["conversions"][conversion_name],
                           "completed", false, allocator);
+        if (FLAGS_keep_errors) {
+            rapidjson::Value msg_value;
+            msg_value.SetString(e.what(), allocator);
+            add_or_set_member(test_case["conversions"][conversion_name],
+                              "error", msg_value, allocator);
+        }
         std::cerr << "Error when processing test case " << test_case << "\n"
                   << "what(): " << e.what() << std::endl;
     }
@@ -156,8 +162,7 @@ int main(int argc, char *argv[])
             try {
                 auto matrix_from =
                     share(formats::matrix_factory.at(format_from)(exec, data));
-                for (const auto &format : formats::matrix_factory) {
-                    const auto format_to = std::get<0>(format);
+                for (const auto &format_to : formats) {
                     if (format_from == format_to) {
                         continue;
                     }

benchmark/preconditioner/preconditioner.cpp

@@ -229,6 +229,12 @@ void run_preconditioner(const char *precond_name,
                           rapidjson::Value(rapidjson::kObjectType), allocator);
         add_or_set_member(test_case["preconditioner"][encoded_name.c_str()],
                           "completed", false, allocator);
+        if (FLAGS_keep_errors) {
+            rapidjson::Value msg_value;
+            msg_value.SetString(e.what(), allocator);
+            add_or_set_member(test_case["preconditioner"][encoded_name.c_str()],
+                              "error", msg_value, allocator);
+        }
         std::cerr << "Error when processing test case " << test_case << "\n"
                   << "what(): " << e.what() << std::endl;
     }

benchmark/run_all_benchmarks.sh

@@ -16,6 +16,16 @@ if [ ! "${EXECUTOR}" ]; then
     echo "EXECUTOR    environment variable not set - assuming \"${EXECUTOR}\"" 1>&2
 fi
 
+if [ ! "${REPETITIONS}" ]; then
+    REPETITIONS=10
+    echo "REPETITIONS    environment variable not set - assuming ${REPETITIONS}" 1>&2
+fi
+
+if [ ! "${SOLVER_REPETITIONS}" ]; then
+    SOLVER_REPETITIONS=1
+    echo "SOLVER_REPETITIONS    environment variable not set - assuming ${SOLVER_REPETITIONS}" 1>&2
+fi
+
 if [ ! "${SEGMENTS}" ]; then
     echo "SEGMENTS    environment variable not set - running entire suite" 1>&2
     SEGMENTS=1
@@ -35,6 +45,11 @@ if [ ! "${FORMATS}" ]; then
     FORMATS="csr,coo,ell,hybrid,sellp"
 fi
 
+if [ ! "${ELL_IMBALANCE_LIMIT}" ]; then
+    echo "ELL_IMBALANCE_LIMIT    environment variable not set - assuming 100" 1>&2
+    ELL_IMBALANCE_LIMIT=100
+fi
+
 if [ ! "${SOLVERS}" ]; then
     SOLVERS="bicgstab,cg,cgs,fcg,gmres,cb_gmres_reduce1,idr"
     echo "SOLVERS    environment variable not set - assuming \"${SOLVERS}\"" 1>&2
@@ -67,7 +82,7 @@ fi
 
 if [ ! "${SOLVERS_JACOBI_MAX_BS}" ]; then
     SOLVERS_JACOBI_MAX_BS="32"
-    "SOLVERS_JACOBI_MAX_BS environment variable not set - assuming \"${SOLVERS_JACOBI_MAX_BS}\"" 1>&2
+    echo "SOLVERS_JACOBI_MAX_BS environment variable not set - assuming \"${SOLVERS_JACOBI_MAX_BS}\"" 1>&2
 fi
 
 if [ ! "${BENCHMARK_PRECISION}" ]; then
@@ -202,6 +217,8 @@ run_conversion_benchmarks() {
     ./conversions/conversions${BENCH_SUFFIX} --backup="$1.bkp" --double_buffer="$1.bkp2" \
                 --executor="${EXECUTOR}" --formats="${FORMATS}" \
                 --device_id="${DEVICE_ID}" --gpu_timer=${GPU_TIMER} \
+                --repetitions="${REPETITIONS}" \
+                --ell_imbalance_limit="${ELL_IMBALANCE_LIMIT}" \
                 <"$1.imd" 2>&1 >"$1"
     keep_latest "$1" "$1.bkp" "$1.bkp2" "$1.imd"
 }
@@ -218,6 +235,8 @@ run_spmv_benchmarks() {
     ./spmv/spmv${BENCH_SUFFIX} --backup="$1.bkp" --double_buffer="$1.bkp2" \
                 --executor="${EXECUTOR}" --formats="${FORMATS}" \
                 --device_id="${DEVICE_ID}" --gpu_timer=${GPU_TIMER} \
+                --repetitions="${REPETITIONS}" \
+                --ell_imbalance_limit="${ELL_IMBALANCE_LIMIT}" \
                 <"$1.imd" 2>&1 >"$1"
     keep_latest "$1" "$1.bkp" "$1.bkp2" "$1.imd"
 }
@@ -239,6 +258,7 @@ run_solver_benchmarks() {
                     --gpu_timer=${GPU_TIMER} \
                     --jacobi_max_block_size=${SOLVERS_JACOBI_MAX_BS} --device_id="${DEVICE_ID}" \
                     --gmres_restart="${SOLVERS_GMRES_RESTART}" \
+                    --repetitions="${SOLVER_REPETITIONS}" \
                     <"$1.imd" 2>&1 >"$1"
     keep_latest "$1" "$1.bkp" "$1.bkp2" "$1.imd"
 }
@@ -265,6 +285,7 @@ run_preconditioner_benchmarks() {
                 --jacobi_max_block_size="${bsize}" \
                 --jacobi_storage="${prec}" \
                 --device_id="${DEVICE_ID}" --gpu_timer=${GPU_TIMER} \
+                --repetitions="${REPETITIONS}" \
                 <"$1.imd" 2>&1 >"$1"
             keep_latest "$1" "$1.bkp" "$1.bkp2" "$1.imd"
         done

benchmark/solver/solver.cpp

@@ -509,6 +509,12 @@ void solve_system(const std::string &solver_name,
     } catch (const std::exception &e) {
         add_or_set_member(test_case["solver"][precond_solver_name], "completed",
                           false, allocator);
+        if (FLAGS_keep_errors) {
+            rapidjson::Value msg_value;
+            msg_value.SetString(e.what(), allocator);
+            add_or_set_member(test_case["solver"][precond_solver_name], "error",
+                              msg_value, allocator);
+        }
         std::cerr << "Error when processing test case " << test_case << "\n"
                   << "what(): " << e.what() << std::endl;
     }

benchmark/spmv/spmv.cpp

@@ -130,7 +130,7 @@ void apply_spmv(const char *format_name, std::shared_ptr<gko::Executor> exec,
             for (auto _ : ic_tuning.run()) {
                 system_matrix->apply(lend(b), lend(x_clone));
             }
-            tuning_case["time"].PushBack(tuning_timer->compute_average_time(),
+            tuning_case["time"].PushBack(ic_tuning.compute_average_time(),
                                          allocator);
             tuning_case["values"].PushBack(val, allocator);
         }
@@ -154,6 +154,12 @@ void apply_spmv(const char *format_name, std::shared_ptr<gko::Executor> exec,
     } catch (const std::exception &e) {
         add_or_set_member(test_case["spmv"][format_name], "completed", false,
                           allocator);
+        if (FLAGS_keep_errors) {
+            rapidjson::Value msg_value;
+            msg_value.SetString(e.what(), allocator);
+            add_or_set_member(test_case["spmv"][format_name], "error",
+                              msg_value, allocator);
+        }
         std::cerr << "Error when processing test case " << test_case << "\n"
                   << "what(): " << e.what() << std::endl;
     }

benchmark/utils/formats.hpp

@@ -37,6 +37,7 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include <ginkgo/ginkgo.hpp>
 
 
+#include <algorithm>
 #include <map>
 #include <string>
 
@@ -152,6 +153,10 @@ std::string format_command =
 // the formats command-line argument
 DEFINE_string(formats, "coo", formats::format_command.c_str());
 
+DEFINE_int64(ell_imbalance_limit, 100,
+             "Maximal storage overhead above which ELL benchmarks will be "
+             "skipped. Negative values mean no limit.");
+
 
 namespace formats {
 
@@ -181,6 +186,51 @@ std::unique_ptr<MatrixType> read_matrix_from_data(
     return mat;
 }
 
+
+/**
+ * Creates a CSR strategy of the given type for the given executor if possible,
+ * falls back to csr::classical for executors without support for this strategy.
+ *
+ * @tparam Strategy  one of csr::automatical or csr::load_balance
+ */
+template <typename Strategy>
+std::shared_ptr<csr::strategy_type> create_gpu_strategy(
+    std::shared_ptr<const gko::Executor> exec)
+{
+    if (auto cuda = dynamic_cast<const gko::CudaExecutor *>(exec.get())) {
+        return std::make_shared<Strategy>(cuda->shared_from_this());
+    } else if (auto hip = dynamic_cast<const gko::HipExecutor *>(exec.get())) {
+        return std::make_shared<Strategy>(hip->shared_from_this());
+    } else {
+        return std::make_shared<csr::classical>();
+    }
+}
+
+
+/**
+ * Checks whether the given matrix data exceeds the ELL imbalance limit set by
+ * the --ell_imbalance_limit flag
+ *
+ * @throws gko::Error if the imbalance limit is exceeded
+ */
+void check_ell_admissibility(const gko::matrix_data<etype> &data)
+{
+    if (data.size[0] == 0 || FLAGS_ell_imbalance_limit < 0) {
+        return;
+    }
+    std::vector<gko::size_type> row_lengths(data.size[0]);
+    for (auto nz : data.nonzeros) {
+        row_lengths[nz.row]++;
+    }
+    auto max_len = *std::max_element(row_lengths.begin(), row_lengths.end());
+    auto avg_len = data.nonzeros.size() / std::max<double>(data.size[0], 1);
+    if (max_len / avg_len > FLAGS_ell_imbalance_limit) {
+        throw gko::Error(__FILE__, __LINE__,
+                         "Matrix exceeds ELL imbalance limit");
+    }
+}
+
+
 /**
  * Creates a Ginkgo matrix from the intermediate data representation format
  * gko::matrix_data with support for variable arguments.
@@ -201,15 +251,36 @@ const std::map<std::string, std::function<std::unique_ptr<gko::LinOp>(
                                 std::shared_ptr<const gko::Executor>,
                                 const gko::matrix_data<etype> &)>>
     matrix_factory{
-        {"csr", READ_MATRIX(csr, std::make_shared<csr::automatical>())},
-        {"csri", READ_MATRIX(csr, std::make_shared<csr::load_balance>())},
+        {"csr",
+         [](std::shared_ptr<const gko::Executor> exec,
+            const gko::matrix_data<etype> &data) -> std::unique_ptr<csr> {
+            auto mat =
+                csr::create(exec, create_gpu_strategy<csr::automatical>(exec));
+            mat->read(data);
+            return mat;
+         }},
+        {"csri",
+         [](std::shared_ptr<const gko::Executor> exec,
+            const gko::matrix_data<etype> &data) -> std::unique_ptr<csr> {
+             auto mat = csr::create(
+                 exec, create_gpu_strategy<csr::load_balance>(exec));
+             mat->read(data);
+             return mat;
+         }},
         {"csrm", READ_MATRIX(csr, std::make_shared<csr::merge_path>())},
         {"csrc", READ_MATRIX(csr, std::make_shared<csr::classical>())},
         {"coo", read_matrix_from_data<gko::matrix::Coo<etype>>},
-        {"ell", read_matrix_from_data<gko::matrix::Ell<etype>>},
+        {"ell", [](std::shared_ptr<const gko::Executor> exec,
+            const gko::matrix_data<etype> &data) {
+             check_ell_admissibility(data);
+             auto mat = gko::matrix::Ell<etype>::create(exec);
+             mat->read(data);
+             return mat;
+         }},
         {"ell-mixed",
          [](std::shared_ptr<const gko::Executor> exec,
             const gko::matrix_data<etype> &data) {
+             check_ell_admissibility(data);
              gko::matrix_data<gko::next_precision<etype>> conv_data;
              conv_data.size = data.size;
              conv_data.nonzeros.resize(data.nonzeros.size());
@@ -220,7 +291,8 @@ const std::map<std::string, std::function<std::unique_ptr<gko::LinOp>(
                  it->value = el.value;
                  ++it;
              }
-             auto mat = gko::matrix::Ell<gko::next_precision<etype>>::create(std::move(exec));
+             auto mat = gko::matrix::Ell<gko::next_precision<etype>>::create(
+                 std::move(exec));
              mat->read(conv_data);
              return mat;
          }},

benchmark/utils/general.hpp

@@ -84,6 +84,10 @@ DEFINE_string(double_buffer, "",
 DEFINE_bool(detailed, true,
             "If set, performs several runs to obtain more detailed results");
 
+DEFINE_bool(keep_errors, false,
+            "If set, writes exception messages during the execution into the "
+            "JSON output");
+
 DEFINE_bool(nested_names, false, "If set, separately logs nested operations");
 
 DEFINE_uint32(seed, 42, "Seed used for the random number generator");

omp/components/format_conversion.hpp

@@ -40,6 +40,9 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include <ginkgo/core/base/types.hpp>
 
 
+#include "core/components/prefix_sum.hpp"
+
+
 namespace gko {
 namespace kernels {
 namespace omp {
@@ -84,16 +87,25 @@ inline void convert_unsorted_idxs_to_ptrs(const IndexType *idxs,
 template <typename IndexType>
 inline void convert_sorted_idxs_to_ptrs(const IndexType *idxs,
                                         size_type num_nonzeros, IndexType *ptrs,
-                                        size_type length)
+                                        size_type num_rows)
 {
     ptrs[0] = 0;
-    ptrs[length - 1] = num_nonzeros;
 
-#pragma omp parallel for schedule( \
-    static, ceildiv(num_nonzeros, omp_get_max_threads()))
-    for (size_type i = 0; i < num_nonzeros - 1; i++) {
-        for (size_type j = idxs[i] + 1; j <= idxs[i + 1]; j++) {
-            ptrs[j] = i + 1;
+    if (num_nonzeros == 0) {
+#pragma omp parallel for
+        for (size_type row = 0; row < num_rows; row++) {
+            ptrs[row + 1] = 0;
+        }
+    } else {
+        // add virtual sentinel values 0 and num_rows to handle empty first and
+        // last rows
+#pragma omp parallel for
+        for (size_type i = 0; i <= num_nonzeros; i++) {
+            auto begin_row = i == 0 ? size_type{} : idxs[i - 1];
+            auto end_row = i == num_nonzeros ? num_rows : idxs[i];
+            for (auto row = begin_row; row < end_row; row++) {
+                ptrs[row + 1] = i;
+            }
         }
     }
 }

omp/matrix/coo_kernels.cpp

@@ -204,9 +204,9 @@ GKO_INSTANTIATE_FOR_EACH_VALUE_AND_INDEX_TYPE(
 template <typename IndexType>
 void convert_row_idxs_to_ptrs(std::shared_ptr<const OmpExecutor> exec,
                               const IndexType *idxs, size_type num_nonzeros,
-                              IndexType *ptrs, size_type length)
+                              IndexType *ptrs, size_type num_rows)
 {
-    convert_sorted_idxs_to_ptrs(idxs, num_nonzeros, ptrs, length);
+    convert_sorted_idxs_to_ptrs(idxs, num_nonzeros, ptrs, num_rows);
 }
 
 
@@ -222,8 +222,7 @@ void convert_to_csr(std::shared_ptr<const OmpExecutor> exec,
 
     const auto source_row_idxs = source->get_const_row_idxs();
 
-    convert_row_idxs_to_ptrs(exec, source_row_idxs, nnz, row_ptrs,
-                             num_rows + 1);
+    convert_row_idxs_to_ptrs(exec, source_row_idxs, nnz, row_ptrs, num_rows);
 }
 
 GKO_INSTANTIATE_FOR_EACH_VALUE_AND_INDEX_TYPE(