1 files changed, 56 insertions, 0 deletions

diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc
index c858c50..1c0c6c6 100644
--- a/src/gpgpu-sim/gpu-sim.cc
+++ b/src/gpgpu-sim/gpu-sim.cc
@@ -1342,7 +1342,52 @@ void gpgpu_sim::print_stats(unsigned long long streamID) {
   }
 }
 
+// Static pointer + signal handler for MEMCO_DBG_DEADLOCK. When a
+// SIGSEGV / SIGABRT fires before the deadlock detector triggers, dump
+// per-warp state for the most recently active gpgpu_sim and re-raise.
+static gpgpu_sim *g_dbg_sim_for_handler = NULL;
+static volatile int g_dbg_sim_handler_in_progress = 0;
+static void memco_dbg_signal_handler(int sig) {
+  if (g_dbg_sim_handler_in_progress) {
+    // already dumping; if we got here again, give up and re-raise the
+    // default to avoid infinite recursion in handler.
+    signal(sig, SIG_DFL);
+    raise(sig);
+    return;
+  }
+  g_dbg_sim_handler_in_progress = 1;
+  fprintf(stdout,
+          "\n\n[DEADLOCK_DUMP] caught signal %d — dumping warp state "
+          "before crash\n", sig);
+  fflush(stdout);
+  if (g_dbg_sim_for_handler) g_dbg_sim_for_handler->dbg_dump_all_warps();
+  fflush(stdout);
+  signal(sig, SIG_DFL);
+  raise(sig);
+}
+
+void gpgpu_sim::dbg_dump_all_warps() {
+  for (unsigned i = 0; i < m_shader_config->n_simt_clusters; i++) {
+    if (m_cluster[i]->get_not_completed() > 0) {
+      m_cluster[i]->dump_warps_at_deadlock(stdout);
+    }
+  }
+}
+
 void gpgpu_sim::deadlock_check() {
+  // Lazy installer for the signal handler (only when MEMCO_DBG_DEADLOCK
+  // is set). Once installed, future signals are caught.
+  static bool sig_handler_installed = false;
+  if (!sig_handler_installed && getenv("MEMCO_DBG_DEADLOCK") != NULL) {
+    sig_handler_installed = true;
+    g_dbg_sim_for_handler = this;
+    signal(SIGSEGV, memco_dbg_signal_handler);
+    signal(SIGABRT, memco_dbg_signal_handler);
+  } else if (getenv("MEMCO_DBG_DEADLOCK") != NULL) {
+    // Keep the static pointer up to date in case multiple gpgpu_sim
+    // instances are created (rare).
+    g_dbg_sim_for_handler = this;
+  }
   if (m_config.gpu_deadlock_detect && gpu_deadlock) {
     fflush(stdout);
     printf(
@@ -1382,6 +1427,17 @@ void gpgpu_sim::deadlock_check() {
     printf(
         "\nRe-run the simulator in gdb and use debug routines in .gdbinit to "
         "debug this\n");
+    // MEMCO_DBG_DEADLOCK: dump per-warp state for any non-empty cluster
+    // before aborting. Helps localize which specific warp(s) are stuck
+    // and why (scoreboard / simt-table / ibuffer state).
+    if (getenv("MEMCO_DBG_DEADLOCK") != NULL) {
+      for (unsigned i = 0; i < m_shader_config->n_simt_clusters; i++) {
+        if (m_cluster[i]->get_not_completed() > 0) {
+          m_cluster[i]->dump_warps_at_deadlock(stdout);
+        }
+      }
+      fflush(stdout);
+    }
     fflush(stdout);
     abort();
   }