Added kernel id to enqueue callback for kernel dispatch (#276)

Adds kernel id as parameter to rocprofiler_profile_counting_dispatch_callback_t.
Small cleanup of code in core.cpp.

samples/counter_collection/client.cpp

@@ -142,6 +142,7 @@ dispatch_callback(rocprofiler_queue_id_t /*queue_id*/,
                   const rocprofiler_agent_t* agent,
                   rocprofiler_correlation_id_t /*correlation_id*/,
                   const hsa_kernel_dispatch_packet_t* /*dispatch_packet*/,
+                  uint64_t /*kernel_id*/,
                   void* /*callback_data_args*/,
                   rocprofiler_profile_config_id_t* config)
 {

samples/counter_collection/print_functional_counters.cpp

@@ -111,6 +111,7 @@ dispatch_callback(rocprofiler_queue_id_t /*queue_id*/,
                   const rocprofiler_agent_t* agent,
                   rocprofiler_correlation_id_t /*correlation_id*/,
                   const hsa_kernel_dispatch_packet_t* /*dispatch_packet*/,
+                  uint64_t /*kernel_id*/,
                   void* /*callback_data_args*/,
                   rocprofiler_profile_config_id_t* config)
 {

source/include/rocprofiler-sdk/dispatch_profile.h

@@ -48,6 +48,7 @@ ROCPROFILER_EXTERN_C_INIT
  * @param [in] agent           Agent of this queue
  * @param [in] correlation_id  Correlation ID for this dispatch
  * @param [in] dispatch_packet Kernel dispatch packet about to be enqueued into HSA
+ * @param [in] kernel_id Kernel identifier
  * @param [in] callback_data_args Callback supplied via buffered_dispatch_profile_counting_service
  * @param [out] config         Profile config detailing the counters to collect for this kernel
  */
@@ -56,6 +57,7 @@ typedef void (*rocprofiler_profile_counting_dispatch_callback_t)(
     const rocprofiler_agent_t*          agent,
     rocprofiler_correlation_id_t        correlation_id,
     const hsa_kernel_dispatch_packet_t* dispatch_packet,
+    uint64_t                            kernel_id,
     void*                               callback_data_args,
     rocprofiler_profile_config_id_t*    config);
 

source/lib/rocprofiler-sdk/counters/core.cpp

@@ -132,6 +132,7 @@ queue_cb(const std::shared_ptr<counter_callback_info>& info,
          const hsa::Queue&                             queue,
          hsa::ClientID,
          const hsa::rocprofiler_packet&                                  pkt,
+         uint64_t                                                        kernel_id,
          const hsa::Queue::queue_info_session_t::external_corr_id_map_t& extern_corr_ids,
          const context::correlation_id*                                  correlation_id)
 {
@@ -154,6 +155,7 @@ queue_cb(const std::shared_ptr<counter_callback_info>& info,
                   queue.get_agent().get_rocp_agent(),
                   _corr_id_v,
                   &pkt.kernel_dispatch,
+                  kernel_id,
                   info->callback_args,
                   &req_profile);
     if(req_profile.handle == 0) return nullptr;
@@ -325,9 +327,10 @@ start_context(context::context* ctx)
                 [=](const hsa::Queue&                                               q,
                     hsa::ClientID                                                   c,
                     const hsa::rocprofiler_packet&                                  kern_pkt,
+                    uint64_t                                                        kernel_id,
                     const hsa::Queue::queue_info_session_t::external_corr_id_map_t& extern_corr_ids,
                     const context::correlation_id* correlation_id) {
-                    return queue_cb(cb, q, c, kern_pkt, extern_corr_ids, correlation_id);
+                    return queue_cb(cb, q, c, kern_pkt, kernel_id, extern_corr_ids, correlation_id);
                 },
                 // Completion CB
                 [=](const hsa::Queue&                       q,

source/lib/rocprofiler-sdk/hsa/queue.cpp

@@ -365,6 +365,7 @@ WriteInterceptor(const void* packets,
 
         // Copy kernel pkt, copy is to allow for signal to be modified
         rocprofiler_packet kernel_pkt = packets_arr[i];
+        uint64_t           kernel_id  = get_kernel_id(kernel_pkt.kernel_dispatch.kernel_object);
         queue.create_signal(HSA_AMD_SIGNAL_AMD_GPU_ONLY,
                             &kernel_pkt.ext_amd_aql_pm4.completion_signal);
 
@@ -378,8 +379,8 @@ WriteInterceptor(const void* packets,
         queue.signal_callback([&](const auto& map) {
             for(const auto& [client_id, cb_pair] : map)
             {
-                if(auto maybe_pkt =
-                       cb_pair.first(queue, client_id, kernel_pkt, extern_corr_ids, corr_id))
+                if(auto maybe_pkt = cb_pair.first(
+                       queue, client_id, kernel_pkt, kernel_id, extern_corr_ids, corr_id))
                 {
                     LOG_IF(FATAL, inst_pkt)
                         << "We do not support two injections into the HSA queue";
@@ -435,7 +436,6 @@ WriteInterceptor(const void* packets,
 
         LOG_IF(FATAL, packet_type != HSA_PACKET_TYPE_KERNEL_DISPATCH)
             << "get_kernel_id below might need to be updated";
-        uint64_t kernel_id = get_kernel_id(kernel_pkt.kernel_dispatch.kernel_object);
 
         // Enqueue the signal into the handler. Will call completed_cb when
         // signal completes.

source/lib/rocprofiler-sdk/hsa/queue.hpp

@@ -126,6 +126,7 @@ public:
         const Queue&,
         ClientID,
         const rocprofiler_packet&,
+        uint64_t,
         const queue_info_session_t::external_corr_id_map_t&,
         const context::correlation_id*)>;
     // Signals the completion of the kernel packet.