Improve stack memory usage at compile time

buildlib/tools/coverity.sh

@@ -104,21 +104,18 @@ run_coverity() {
  fi
  cov-analyze --jobs $parallel_jobs $COV_OPT --security --concurrency --dir $cov_build
  nerrors=$(cov-format-errors --dir $cov_build | awk '/Processing [0-9]+ errors?/ { print $2 }')
- rc=$(($rc+$nerrors))
 
  if [ $nerrors -gt 0 ]; then
- cov-format-errors --dir $cov_build --emacs-style
+ cov_error_log="$(echo -e "\n$(cov-format-errors --dir $cov_build --emacs-style)")"
  if [ -d "$WORKSPACE/$cov_build_id" ]; then
  rm -rf $WORKSPACE/$cov_build_id
  fi
  cp -ar $cov_build $WORKSPACE/$cov_build_id
- echo "not ok 1 Coverity Detected $nerrors failures"
  else
- echo "ok 1 Coverity found no issues"
  rm -rf $cov_build
  fi
  modules_for_coverity_unload
- return $rc
+ return $nerrors
 }
 
 parse_args $*
@@ -135,4 +132,9 @@ else
  cd "${ucx_build_dir}"
 fi
 
-run_coverity "$mode"
+run_coverity "$mode" || {
+ set +x
+ azure_log_error "Coverity found $nerrors issues: $cov_error_log"
+ azure_log_error "To debug, rerun with: $0 $mode"
+ exit 1
+}

config/m4/compiler.m4

@@ -449,6 +449,26 @@ AS_IF([test "x$enable_gcov" = xyes],
  [AC_LANG_SOURCE([[int main(int argc, char** argv){return 0;}]])])],
  [:])
 
+#
+# Enable stack usage check
+#
+# To suppress -Wframe-larger-than=8192 for specific functions, use:
+# #pragma GCC diagnostic push
+# #pragma GCC diagnostic ignored "-Wframe-larger-than="
+# void specific_function() {
+# /* Function code */
+# }
+# #pragma GCC diagnostic pop
+#
+AC_ARG_ENABLE([stack-usage-check],
+ AS_HELP_STRING([--enable-stack-usage-check], [Enable stack usage check with -Wframe-larger-than=8192]),
+ [],
+ [enable_stack_usage_check=yes])
+
+AS_IF([test "x$enable_stack_usage_check" = xyes],
+ [ADD_COMPILER_FLAG_IF_SUPPORTED([stack_usage_check], [-Wframe-larger-than=8192],
+ [AC_LANG_SOURCE([[int main(int argc, char** argv){return 0;}]])])],
+ [:])
 
 #
 # Check for C++ support

contrib/test_jenkins.sh

@@ -1119,7 +1119,19 @@ run_tests() {
  do_distributed_task 0 4 run_nt_buffer_transfer_tests
 }
 
+error_handler() {
+ set +x
+ local exit_code=$?
+ local test_name="$1"
+ local flag="$2"
+
+ azure_log_error "$test_name failed"
+ azure_log_error "To debug, rerun with: $flag $0"
+ exit $exit_code
+}
+
 run_test_proto_disable() {
+ trap 'error_handler "Proto V1 test" "PROTO_ENABLE=no"' ERR
  # build for devel tests and gtest
  build devel --enable-gtest
 
@@ -1130,11 +1142,13 @@ run_test_proto_disable() {
 }
 
 run_asan_check() {
+ trap 'error_handler "AddressSanitizer check" "ASAN_CHECK=yes"' ERR
  build devel --enable-gtest --enable-asan --without-valgrind
  run_gtest "default"
 }
 
 run_valgrind_check() {
+ trap 'error_handler "Valgrind check" "VALGRIND_CHECK=yes"' ERR
  if [[ $(uname -m) =~ "aarch" ]] || [[ $(uname -m) =~ "ppc" ]]; then
  echo "==== Skip valgrind tests on `uname -m` ===="
  return

src/tools/info/type_info.c

@@ -87,7 +87,7 @@ static void print_size(const char *name, size_t size)
  print_size(UCS_PP_QUOTE(_type) "." UCS_PP_QUOTE(_field), \
  ucs_field_sizeof(_type, _field))
 
-void print_type_info(const char * tl_name)
+static void print_ucs(const char *tl_name)
 {
  if (tl_name == NULL) {
  printf("UCS:\n");
@@ -131,8 +131,13 @@ void print_type_info(const char * tl_name)
  PRINT_SIZE(ucs_rcache_region_t);
  PRINT_SIZE(ucs_conn_match_elem_t);
  PRINT_SIZE(ucs_memory_info_t);
+ }
+}
 
- printf("\nUCT:\n");
+static void print_uct(const char *tl_name)
+{
+ if (tl_name == NULL) {
+ printf("UCT:\n");
  PRINT_SIZE(uct_am_handler_t);
  PRINT_SIZE(uct_base_iface_t);
  PRINT_SIZE(uct_completion_t);
@@ -152,13 +157,11 @@ void print_type_info(const char * tl_name)
  PRINT_SIZE(uct_rkey_bundle_t);
  PRINT_SIZE(uct_tcp_ep_t);
  PRINT_SIZE(uct_self_ep_t);
+ }
+}
 
-#ifdef HAVE_TL_UGNI
- PRINT_SIZE(uct_sockaddr_ugni_t);
- PRINT_SIZE(uct_sockaddr_smsg_ugni_t);
- PRINT_SIZE(uct_devaddr_ugni_t);
-#endif
-
+static void print_ib(const char *tl_name)
+{
 #if HAVE_IB
  printf("\nIB:\n");
  PRINT_SIZE(uct_ib_address_t);
@@ -170,13 +173,13 @@ void print_type_info(const char * tl_name)
  PRINT_SIZE(uct_ib_iface_recv_desc_t);
  PRINT_SIZE(uct_ib_recv_wr_t);
 #endif
- printf("\n");
- }
+}
 
+static void print_rc(const char *tl_name)
+{
 #if HAVE_TL_RC
  if (tl_name == NULL || !strcasecmp(tl_name, "rc_verbs") ||
- !strcasecmp(tl_name, "rc_mlx5"))
- {
+ !strcasecmp(tl_name, "rc_mlx5")) {
  printf("RC:\n");
  PRINT_SIZE(uct_rc_am_short_hdr_t);
  PRINT_SIZE(uct_rc_ep_t);
@@ -185,14 +188,12 @@ void print_type_info(const char * tl_name)
  PRINT_SIZE(uct_rc_iface_config_t);
  PRINT_SIZE(uct_rc_iface_send_op_t);
  PRINT_SIZE(uct_rc_iface_send_desc_t);
-
  PRINT_SIZE(uct_rc_iface_send_desc_t);
  if (tl_name == NULL || !strcasecmp(tl_name, "rc_verbs")) {
  PRINT_SIZE(uct_rc_verbs_ep_t);
  PRINT_SIZE(uct_rc_verbs_iface_config_t);
  PRINT_SIZE(uct_rc_verbs_iface_t);
  }
-
 #ifdef HAVE_MLX5_DV
  if (tl_name == NULL || !strcasecmp(tl_name, "rc_mlx5")) {
  PRINT_SIZE(uct_rc_mlx5_am_short_hdr_t);
@@ -205,7 +206,10 @@ void print_type_info(const char * tl_name)
  printf("\n");
  }
 #endif
+}
 
+static void print_dc(const char *tl_name)
+{
 #if HAVE_TL_DC
  if (tl_name == NULL || !strcasecmp(tl_name, "dc_mlx5"))
  {
@@ -217,11 +221,13 @@ void print_type_info(const char * tl_name)
  printf("\n");
  }
 #endif
+}
 
+static void print_ud(const char *tl_name)
+{
 #if HAVE_TL_UD
  if (tl_name == NULL || !strcasecmp(tl_name, "ud_verbs") ||
- !strcasecmp(tl_name, "ud_mlx5"))
- {
+ !strcasecmp(tl_name, "ud_mlx5")) {
  printf("UD:\n");
  PRINT_SIZE(uct_ud_ep_t);
  PRINT_SIZE(uct_ud_neth_t);
@@ -230,13 +236,11 @@ void print_type_info(const char * tl_name)
  PRINT_SIZE(uct_ud_ep_pending_op_t);
  PRINT_SIZE(uct_ud_send_skb_t);
  PRINT_SIZE(uct_ud_recv_skb_t);
-
  PRINT_SIZE(uct_rc_iface_send_desc_t);
  if (tl_name == NULL || !strcasecmp(tl_name, "ud_verbs")) {
  PRINT_SIZE(uct_ud_verbs_ep_t);
  PRINT_SIZE(uct_ud_verbs_iface_t);
  }
-
 #ifdef HAVE_MLX5_HW_UD
  if (tl_name == NULL || !strcasecmp(tl_name, "ud_mlx5")) {
  PRINT_SIZE(uct_ud_mlx5_ep_t);
@@ -246,20 +250,28 @@ void print_type_info(const char * tl_name)
  printf("\n");
  }
 #endif
+}
 
+static void print_ugni(const char *tl_name)
+{
 #ifdef HAVE_TL_UGNI
  if (tl_name == NULL || !strcasecmp(tl_name, "ugni")) {
  printf("UGNI:\n");
+ PRINT_SIZE(uct_sockaddr_ugni_t);
+ PRINT_SIZE(uct_sockaddr_smsg_ugni_t);
+ PRINT_SIZE(uct_devaddr_ugni_t);
  PRINT_SIZE(uct_ugni_device_t);
  PRINT_SIZE(uct_ugni_ep_t);
  PRINT_SIZE(uct_ugni_iface_t);
  PRINT_SIZE(uct_ugni_md_t);
  PRINT_SIZE(uct_ugni_compact_smsg_attr_t);
-
  printf("\n");
  }
 #endif
+}
 
+static void print_ucp()
+{
  printf("\nUCP:\n");
  PRINT_SIZE(ucp_context_t);
  PRINT_SIZE(ucp_worker_t);
@@ -291,5 +303,16 @@ void print_type_info(const char * tl_name)
  PRINT_SIZE(ucp_mem_t);
  PRINT_SIZE(ucp_rkey_t);
  PRINT_SIZE(ucp_wireup_msg_t);
+}
 
+void print_type_info(const char *tl_name)
+{
+ print_ucs(tl_name);
+ print_uct(tl_name);
+ print_ib(tl_name);
+ print_rc(tl_name);
+ print_dc(tl_name);
+ print_ud(tl_name);
+ print_ugni(tl_name);
+ print_ucp();
 }

src/tools/perf/lib/ucp_tests.cc

@@ -1033,35 +1033,69 @@ class ucp_perf_test_runner {
 #define TEST_CASE_ALL_AM(_perf, _case) \
  TEST_CASE(_perf, UCS_PP_TUPLE_0 _case, UCS_PP_TUPLE_1 _case, 0, 0)
 
-ucs_status_t ucp_perf_test_dispatch(ucx_perf_context_t *perf)
+static ucs_status_t dispatch_osd(ucx_perf_context_t *perf)
 {
  UCS_PP_FOREACH(TEST_CASE_ALL_OSD, perf,
- (UCX_PERF_CMD_PUT, UCX_PERF_TEST_TYPE_PINGPONG),
- (UCX_PERF_CMD_PUT, UCX_PERF_TEST_TYPE_PINGPONG_WAIT_MEM),
- (UCX_PERF_CMD_PUT, UCX_PERF_TEST_TYPE_STREAM_UNI),
- (UCX_PERF_CMD_GET, UCX_PERF_TEST_TYPE_STREAM_UNI),
- (UCX_PERF_CMD_ADD, UCX_PERF_TEST_TYPE_STREAM_UNI),
- (UCX_PERF_CMD_FADD, UCX_PERF_TEST_TYPE_STREAM_UNI),
- (UCX_PERF_CMD_SWAP, UCX_PERF_TEST_TYPE_STREAM_UNI),
- (UCX_PERF_CMD_CSWAP, UCX_PERF_TEST_TYPE_STREAM_UNI)
- );
+ (UCX_PERF_CMD_PUT, UCX_PERF_TEST_TYPE_PINGPONG),
+ (UCX_PERF_CMD_PUT, UCX_PERF_TEST_TYPE_PINGPONG_WAIT_MEM),
+ (UCX_PERF_CMD_PUT, UCX_PERF_TEST_TYPE_STREAM_UNI),
+ (UCX_PERF_CMD_GET, UCX_PERF_TEST_TYPE_STREAM_UNI),
+ (UCX_PERF_CMD_ADD, UCX_PERF_TEST_TYPE_STREAM_UNI),
+ (UCX_PERF_CMD_FADD, UCX_PERF_TEST_TYPE_STREAM_UNI),
+ (UCX_PERF_CMD_SWAP, UCX_PERF_TEST_TYPE_STREAM_UNI),
+ (UCX_PERF_CMD_CSWAP, UCX_PERF_TEST_TYPE_STREAM_UNI)
+ );
+ return UCS_ERR_INVALID_PARAM;
+}
 
+static ucs_status_t dispatch_tag(ucx_perf_context_t *perf)
+{
  UCS_PP_FOREACH(TEST_CASE_ALL_TAG, perf,
- (UCX_PERF_CMD_TAG, UCX_PERF_TEST_TYPE_PINGPONG),
- (UCX_PERF_CMD_TAG, UCX_PERF_TEST_TYPE_STREAM_UNI),
- (UCX_PERF_CMD_TAG_SYNC, UCX_PERF_TEST_TYPE_PINGPONG),
- (UCX_PERF_CMD_TAG_SYNC, UCX_PERF_TEST_TYPE_STREAM_UNI)
- );
+ (UCX_PERF_CMD_TAG, UCX_PERF_TEST_TYPE_PINGPONG),
+ (UCX_PERF_CMD_TAG, UCX_PERF_TEST_TYPE_STREAM_UNI),
+ (UCX_PERF_CMD_TAG_SYNC, UCX_PERF_TEST_TYPE_PINGPONG),
+ (UCX_PERF_CMD_TAG_SYNC, UCX_PERF_TEST_TYPE_STREAM_UNI)
+ );
+ return UCS_ERR_INVALID_PARAM;
+}
 
+static ucs_status_t dispatch_stream(ucx_perf_context_t *perf)
+{
  UCS_PP_FOREACH(TEST_CASE_ALL_STREAM, perf,
- (UCX_PERF_CMD_STREAM, UCX_PERF_TEST_TYPE_STREAM_UNI),
- (UCX_PERF_CMD_STREAM, UCX_PERF_TEST_TYPE_PINGPONG)
- );
+ (UCX_PERF_CMD_STREAM, UCX_PERF_TEST_TYPE_STREAM_UNI),
+ (UCX_PERF_CMD_STREAM, UCX_PERF_TEST_TYPE_PINGPONG)
+ );
+ return UCS_ERR_INVALID_PARAM;
+}
 
+static ucs_status_t dispatch_am(ucx_perf_context_t *perf)
+{
  UCS_PP_FOREACH(TEST_CASE_ALL_AM, perf,
- (UCX_PERF_CMD_AM, UCX_PERF_TEST_TYPE_PINGPONG),
- (UCX_PERF_CMD_AM, UCX_PERF_TEST_TYPE_STREAM_UNI)
- );
+ (UCX_PERF_CMD_AM, UCX_PERF_TEST_TYPE_PINGPONG),
+ (UCX_PERF_CMD_AM, UCX_PERF_TEST_TYPE_STREAM_UNI)
+ );
+ return UCS_ERR_INVALID_PARAM;
+}
+
+typedef ucs_status_t (*ucp_dispatch_func_t)(ucx_perf_context_t *perf);
+
+static ucp_dispatch_func_t dispatchers[] = {
+ dispatch_osd,
+ dispatch_tag,
+ dispatch_stream,
+ dispatch_am
+};
+
+ucs_status_t ucp_perf_test_dispatch(ucx_perf_context_t *perf) {
+ ucs_status_t status;
+ int num_dispatchers = sizeof(dispatchers) / sizeof(dispatchers[0]);
+
+ for (int i = 0; i < num_dispatchers; ++i) {
+ status = dispatchers[i](perf);
+ if (status != UCS_ERR_INVALID_PARAM) {
+ return status;
+ }
+ }
 
  ucs_error("Invalid test case: %d/%d/0x%x",
  perf->params.command, perf->params.test_type,

src/tools/perf/lib/uct_tests.cc

@@ -783,6 +783,8 @@ class uct_perf_test_runner {
  TEST_CASE_ALL_OSD(_perf, _case, UCT_PERF_DATA_LAYOUT_BCOPY) \
  TEST_CASE_ALL_OSD(_perf, _case, UCT_PERF_DATA_LAYOUT_ZCOPY)
 
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wframe-larger-than="
 ucs_status_t uct_perf_test_dispatch(ucx_perf_context_t *perf)
 {
  UCS_PP_FOREACH(TEST_CASE_ALL_DATA, perf,
@@ -801,3 +803,4 @@ ucs_status_t uct_perf_test_dispatch(ucx_perf_context_t *perf)
  ucs_error("Invalid test case");
  return UCS_ERR_INVALID_PARAM;
 }
+#pragma GCC diagnostic pop