adding dispatch_packet field to rocprofiler_callback_data_t

Change-Id: I3cbe256130fa23dd24b094f27286d566ed339230


[ROCm/rocprofiler commit: b8fc3818de]

projects/rocprofiler/inc/rocprofiler.h

@@ -67,7 +67,7 @@
 #include <stdint.h>
 
 #define ROCPROFILER_VERSION_MAJOR 1
-#define ROCPROFILER_VERSION_MINOR 0
+#define ROCPROFILER_VERSION_MINOR 1
 
 #ifdef __cplusplus
 extern "C" {
@@ -237,7 +237,7 @@ typedef struct {
   uint32_t agent_index;                                // GPU index
   const hsa_queue_t* queue;                            // HSA queue
   uint64_t queue_index;                                // Index in the queue
-  uint64_t kernel_object;                              // Kernel object handle
+  const hsa_kernel_dispatch_packet_t* packet;          // HSA dispatch packet
   const char* kernel_name;                             // Kernel name
   const rocprofiler_dispatch_record_t* record;         // Dispatch record
 } rocprofiler_callback_data_t;

projects/rocprofiler/src/core/intercept_queue.h

@@ -101,7 +101,7 @@ class InterceptQueue {
                                             obj->agent_info_->dev_index,
                                             obj->queue_,
                                             user_que_idx,
-                                            dispatch_packet->kernel_object,
+                                            dispatch_packet,
                                             kernel_name,
                                             record};
         hsa_status_t status = dispatch_callback_(&data, callback_data_, &group);

projects/rocprofiler/test/tool/metrics.xml

@@ -17,7 +17,7 @@
   # FETCH_SIZE, kilobytes
   # The total kilobytes fetched from the video memory. This is measured with all extra fetches and any cache or memory effects taken into account.
   <metric name="FETCH_SIZE" expr=(TCC_MC_RDREQ_sum*32)/1024 ></metric>
-  # WRITE_SIZE
+  # WRITE_SIZE, kilobytes
   # The total kilobytes written to the video memory. This is measured with all extra fetches and any cache or memory effects taken into account.
   <metric name="WRITE_SIZE" expr=(TCC_MC_WRREQ_sum*32)/1024 ></metric>
 </gfx8>
@@ -41,7 +41,7 @@
   # FETCH_SIZE, kilobytes
   # The total kilobytes fetched from the video memory. This is measured with all extra fetches and any cache or memory effects taken into account.
   <metric name="FETCH_SIZE" expr=(TCC_EA_RDREQ_32B_sum*32+(TCC_EA_RDREQ_sum-TCC_EA_RDREQ_32B_sum)*64)/1024 ></metric>
-  # WRITE_SIZE
+  # WRITE_SIZE, kilobytes
   # The total kilobytes written to the video memory. This is measured with all extra fetches and any cache or memory effects taken into account.
   <metric name="WRITE_SIZE" expr=((TCC_EA_WRREQ_sum-TCC_EA_WRREQ_64B_sum)*32+TCC_EA_WRREQ_64B_sum*64)/1024 ></metric>
 </gfx9>