reverted to ab4bc29 of btx_zeinterval_callbacks

Author: myrepo1

backends/ze/btx_zeinterval_callbacks.hpp

@@ -41,31 +41,44 @@ typedef std::tuple<hostname_t, process_id_t, zes_device_handle_t, zes_pwr_handle
 typedef std::tuple<hostname_t, process_id_t, zes_device_handle_t, zes_freq_handle_t> hpdfreq_t;
 typedef std::tuple<hostname_t, process_id_t, zes_device_handle_t, zes_engine_handle_t> hpdeng_t;
 
-typedef std::tuple<hostname_t, process_id_t, zes_device_handle_t, zes_mem_handle_t, uint32_t> hpdmsd_t;
-typedef std::tuple<hostname_t, process_id_t, zes_device_handle_t, zes_fabric_port_handle_t, uint32_t> hpdfsd_t;
-typedef std::tuple<hostname_t, process_id_t, zes_device_handle_t, zes_engine_handle_t, uint32_t> hpdesd_t;
-typedef std::tuple<hostname_t, process_id_t, zes_device_handle_t, zes_pwr_handle_t, uint32_t> hpdpwrd_t;
+typedef std::tuple<hostname_t, process_id_t, zes_device_handle_t, zes_mem_handle_t, uint32_t>
+    hpdmsd_t;
+typedef std::
+    tuple<hostname_t, process_id_t, zes_device_handle_t, zes_fabric_port_handle_t, uint32_t>
+        hpdfsd_t;
+typedef std::tuple<hostname_t, process_id_t, zes_device_handle_t, zes_engine_handle_t, uint32_t>
+    hpdesd_t;
+typedef std::tuple<hostname_t, process_id_t, zes_device_handle_t, zes_pwr_handle_t, uint32_t>
+    hpdpwrd_t;
 
 using btx_kernel_group_size_t = std::tuple<uint32_t, uint32_t, uint32_t>;
 using btx_kernel_desct_t =
     std::tuple<std::string /*ze_kernel_desc_t*/, ze_kernel_properties_t, btx_kernel_group_size_t>;
 
-enum class btx_event_t { TRAFFIC, KERNEL, OTHER };
+// SIGNAL = zeCommandListAppendSignalEvent. Ring entry is created so state
+// stays consistent, but filtered out of the device tally (no GPU work).
+enum class btx_event_t { TRAFFIC, KERNEL, SIGNAL, OTHER };
 using btx_additional_info_traffic_t = std::tuple<int64_t /*ts*/, size_t /*size*/>;
 using btx_additional_info_kernel_t = std::string /*metadata*/;
 using btx_additional_info =
     std::variant<std::monostate, btx_additional_info_traffic_t, btx_additional_info_kernel_t>;
 
-using btx_launch_desc_t =
-    std::tuple<ze_command_list_handle_t, std::string /*name*/, int64_t /*ts min */,
-               btx_event_t /* type of enum */, btx_additional_info /* additional data */>;
-
-using btx_event_desct_t =
-    std::tuple<thread_id_t, ze_command_queue_desc_t, ze_command_list_handle_t,
-               bool /*hCommandListIsImmediate*/, ze_device_handle_t, std::string /*name*/,
-               int64_t /*ts born min*/, clock_lttng_device_t /* clock sync pair */,
-               btx_event_t /* type of enum */,
-               btx_additional_info /* pointer to additional data */>;
+using btx_launch_desc_t = std::tuple<ze_command_list_handle_t,
+                                     std::string /*name*/,
+                                     int64_t /*ts min */,
+                                     btx_event_t /* type of enum */,
+                                     btx_additional_info /* additional data */>;
+
+using btx_event_desct_t = std::tuple<thread_id_t,
+                                     ze_command_queue_desc_t,
+                                     ze_command_list_handle_t,
+                                     bool /*hCommandListIsImmediate*/,
+                                     ze_device_handle_t,
+                                     std::string /*name*/,
+                                     int64_t /*ts born min*/,
+                                     clock_lttng_device_t /* clock sync pair */,
+                                     btx_event_t /* type of enum */,
+                                     btx_additional_info /* pointer to additional data */>;
 
 using btx_command_list_desc_t =
     std::tuple<ze_command_queue_desc_t, ze_device_handle_t, bool /*hCommandListIsImmediate*/>;
@@ -82,7 +95,18 @@ struct data_s {
   std::unordered_map<hp_command_queue_t, ze_command_queue_desc_t> commandQueueToDesc;
 
   std::unordered_map<hpt_t, btx_launch_desc_t> threadToLastLaunchInfo;
-  std::unordered_map<hp_event_t, btx_event_desct_t> eventToBtxDesct;
+
+  /* Per-event metadata ring. An hEvent can be the signal event of N
+   * Appends in one build phase, and the cl can be resubmitted M times,
+   * yielding M*N result events. We store the N Appends as a vector and
+   * advance `cursor` per result, wrapping at the end. A new push that
+   * arrives after the cursor advanced indicates a new build phase —
+   * we clear and start over so the ring tracks only the current phase. */
+  struct event_ring_t {
+    std::vector<btx_event_desct_t> entries;
+    size_t cursor = 0;
+  };
+  std::unordered_map<hp_event_t, event_ring_t> eventToBtxDesct;
   // Require for non IMM
   std::unordered_map<hp_command_list_t, std::unordered_set<ze_event_handle_t>> commandListToEvents;