From 5a2cf2fea81bc7c2975c231d0caf9e39798234c2 Mon Sep 17 00:00:00 2001
From: Evgeny <evgeny.shcherbakov@amd.com>
Date: Fri, 23 Nov 2018 00:08:36 -0600
Subject: [PATCH] process/thread IDs in API records

[ROCm/roctracer commit: 65d8c3be396aa991cc1453e62a837cd755e17444]
---
 projects/roctracer/inc/ext/prof_protocol.h    | 18 ++++++++++++-----
 projects/roctracer/src/core/roctracer.cpp     | 13 +++++++-----
 .../test/MatrixTranspose/MatrixTranspose.cpp  | 20 +++++++++++++++----
 3 files changed, 37 insertions(+), 14 deletions(-)

diff --git a/projects/roctracer/inc/ext/prof_protocol.h b/projects/roctracer/inc/ext/prof_protocol.h
index 42501f7355..c31027e404 100644
--- a/projects/roctracer/inc/ext/prof_protocol.h
+++ b/projects/roctracer/inc/ext/prof_protocol.h
@@ -20,8 +20,8 @@ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
 THE SOFTWARE.
 */
 
-#ifndef INC_ROCTRACER_PROF_PROTOCOL_H_
-#define INC_ROCTRACER_PROF_PROTOCOL_H_
+#ifndef INC_EXT_PROF_PROTOCOL_H_
+#define INC_EXT_PROF_PROTOCOL_H_
 
 // Traced API domains
 typedef enum {
@@ -52,8 +52,16 @@ struct activity_record_t {
     activity_correlation_id_t correlation_id;      // activity correlation ID
     uint64_t begin_ns;                             // host begin timestamp
     uint64_t end_ns;                               // host end timestamp
-    int device_id;                                 // device id
-    uint64_t stream_id;                            // stream id
+    union {
+      struct {
+        int device_id;                             // device id
+        uint64_t queue_id;                         // queue id
+      };
+      struct {
+        uint32_t process_id;                       // device id
+        uint32_t thread_id;                        // thread id
+      };
+    };
     size_t bytes;                                  // data size bytes
 };
 
@@ -62,4 +70,4 @@ typedef activity_record_t* (*activity_sync_callback_t)(uint32_t cid, activity_re
 // Activity async calback type
 typedef void (*activity_async_callback_t)(uint32_t op, void* record, void* arg);
 
-#endif  // INC_ROCTRACER_PROF_PROTOCOL_H_
+#endif  // INC_EXT_PROF_PROTOCOL_H_
diff --git a/projects/roctracer/src/core/roctracer.cpp b/projects/roctracer/src/core/roctracer.cpp
index 1f4c887ff2..2c22e3d84a 100644
--- a/projects/roctracer/src/core/roctracer.cpp
+++ b/projects/roctracer/src/core/roctracer.cpp
@@ -29,6 +29,8 @@ THE SOFTWARE.
 #include <mutex>
 #include <string.h>
 #include <pthread.h>
+#include <unistd.h>
+#include <sys/syscall.h>
 
 #include "ext/hsa_rt_utils.hpp"
 #include "util/exception.h"
@@ -276,7 +278,7 @@ DESTRUCTOR_API void destructor() {
   util::Logger::Destroy();
 }
 
-roctracer_record_t* SyncActivityCallback(
+roctracer_record_t* HIP_SyncActivityCallback(
     uint32_t activity_id,
     roctracer_record_t* record,
     const void* callback_data,
@@ -303,7 +305,8 @@ roctracer_record_t* SyncActivityCallback(
     return record;
   } else {
     record->end_ns = timer.timestamp_ns();
-    Kalmar::CLAMP::GetActivityCoord(&(record->device_id), &(record->stream_id));
+    record->process_id = syscall(__NR_getpid);
+    record->thread_id = syscall(__NR_gettid);
     pool->Write(*record);
     // Clearing record in HCC
     Kalmar::CLAMP::SetActivityRecord(0);
@@ -311,7 +314,7 @@ roctracer_record_t* SyncActivityCallback(
   }
 }
 
-void AsyncActivityCallback(
+void HCC_AsyncActivityCallback(
     uint32_t op_id,
     void* record,
     void* arg)
@@ -460,12 +463,12 @@ PUBLIC_API roctracer_status_t roctracer_enable_activity(
       roctracer_enable_activity(ACTIVITY_DOMAIN_HIP_API, HIP_API_ID_ANY, pool);
       break;
     case ACTIVITY_DOMAIN_HCC_OPS: {
-      const bool err = Kalmar::CLAMP::SetActivityCallback(id, (void*)roctracer::AsyncActivityCallback, (void*)pool);
+      const bool err = Kalmar::CLAMP::SetActivityCallback(id, (void*)roctracer::HCC_AsyncActivityCallback, (void*)pool);
       if (err == true) HCC_EXC_RAISING(ROCTRACER_STATUS_HCC_OPS_ERR, "Kalmar::CLAMP::SetActivityCallback error");
       break;
     }
     case ACTIVITY_DOMAIN_HIP_API: {
-      const hipError_t hip_err = hipRegisterActivityCallback(id, (void*)roctracer::SyncActivityCallback, (void*)pool);
+      const hipError_t hip_err = hipRegisterActivityCallback(id, (void*)roctracer::HIP_SyncActivityCallback, (void*)pool);
       if (hip_err != hipSuccess) HIP_EXC_RAISING(ROCTRACER_STATUS_HIP_API_ERR, "hipRegisterActivityCallback error(" << hip_err << ")");
       break;
     }
diff --git a/projects/roctracer/test/MatrixTranspose/MatrixTranspose.cpp b/projects/roctracer/test/MatrixTranspose/MatrixTranspose.cpp
index 707aaa885a..7dadf38cc6 100644
--- a/projects/roctracer/test/MatrixTranspose/MatrixTranspose.cpp
+++ b/projects/roctracer/test/MatrixTranspose/MatrixTranspose.cpp
@@ -212,14 +212,26 @@ void activity_callback(const char* begin, const char* end, void* arg) {
   fprintf(stdout, "\tActivity records:\n"); fflush(stdout);
   while (record < end_record) {
     const char * name = roctracer_id_string(record->domain, record->activity_id, record->kind);
-    fprintf(stdout, "\t%s\tcorrelation_id(%lu) time_ns(%lu:%lu) device_id(%d) stream_id(%lu)",
+    fprintf(stdout, "\t%s\tcorrelation_id(%lu) time_ns(%lu:%lu)",
       name,
       record->correlation_id,
       record->begin_ns,
-      record->end_ns,
-      record->device_id,
-      record->stream_id
+      record->end_ns
     );
+    if (record->domain == ACTIVITY_DOMAIN_HIP_API) {
+      fprintf(stdout, " process_id(%u) thread_id(%u)",
+        record->process_id,
+        record->thread_id
+      );
+    } else if (record->domain == ACTIVITY_DOMAIN_HCC_OPS) {
+      fprintf(stdout, " device_id(%d) queue_id(%lu)",
+        record->device_id,
+        record->queue_id
+      );
+    } else {
+      fprintf(stderr, "Bad domain %d\n", record->domain);
+      abort();
+    }
     if (record->activity_id == hc::HSA_OP_ID_COPY) fprintf(stdout, " bytes(0x%zx)", record->bytes);
     fprintf(stdout, "\n");
     fflush(stdout);