Remove legacy roctracer/roctx services

CMakeLists.txt

@@ -60,30 +60,28 @@ add_caliper_option(WITH_FORTRAN           "Install Fortran interface" FALSE)
 add_caliper_option(WITH_PYTHON_BINDINGS   "Install Python bindings" FALSE)
 add_caliper_option(WITH_TOOLS             "Build Caliper tools" TRUE)
 
-add_caliper_option(WITH_NVTX      "Enable NVidia nvtx bindings for NVprof and NSight (requires CUDA)" FALSE)
-add_caliper_option(WITH_CUPTI     "Enable CUPTI service (CUDA performance analysis)" FALSE)
-add_caliper_option(WITH_PAPI      "Enable PAPI hardware counter service (requires papi)" FALSE)
-add_caliper_option(WITH_LIBPFM    "Enable libpfm (perf_event) sampling" FALSE)
-add_caliper_option(WITH_LIBDW     "Enable libdw support (for symbollookup service)" FALSE)
-add_caliper_option(WITH_LIBUNWIND "Enable libunwind support (for callpath service)" FALSE)
-add_caliper_option(WITH_MPI       "Enable MPI" FALSE)
-# add_caliper_option(WITH_MPIT      "Enable MPI-T" FALSE)
-add_caliper_option(WITH_OMPT      "Enable OMPT" FALSE)
-add_caliper_option(WITH_SAMPLER   "Enable Linux sampler (x86 and PPC Linux only)" FALSE)
-add_caliper_option(WITH_GOTCHA    "Enable GOTCHA wrapping" ${CALIPER_HAVE_LINUX})
-add_caliper_option(WITH_ROCTX     "Enable AMD RocTX support" FALSE)
-add_caliper_option(WITH_ROCTRACER "Enable AMD RocTracer support" FALSE)
-add_caliper_option(WITH_ROCM      "Enable AMD RocTracer and RocTX support" FALSE)
+add_caliper_option(WITH_NVTX        "Enable NVidia nvtx bindings for NVprof and NSight (requires CUDA)" FALSE)
+add_caliper_option(WITH_CUPTI       "Enable CUPTI service (CUDA performance analysis)" FALSE)
+add_caliper_option(WITH_PAPI        "Enable PAPI hardware counter service (requires papi)" FALSE)
+add_caliper_option(WITH_LIBPFM      "Enable libpfm (perf_event) sampling" FALSE)
+add_caliper_option(WITH_LIBDW       "Enable libdw support (for symbollookup service)" FALSE)
+add_caliper_option(WITH_LIBUNWIND   "Enable libunwind support (for callpath service)" FALSE)
+add_caliper_option(WITH_MPI         "Enable MPI" FALSE)
+# add_caliper_option(WITH_MPIT        "Enable MPI-T" FALSE)
+add_caliper_option(WITH_OMPT        "Enable OMPT" FALSE)
+add_caliper_option(WITH_SAMPLER     "Enable Linux sampler (x86 and PPC Linux only)" FALSE)
+add_caliper_option(WITH_GOTCHA      "Enable GOTCHA wrapping" ${CALIPER_HAVE_LINUX})
+add_caliper_option(WITH_ROCM        "Enable AMD ROCm profiling support (identical to WITH_ROCPROFILER)" FALSE)
 add_caliper_option(WITH_ROCPROFILER "Enable AMD rocprofiler-sdk support" FALSE)
-add_caliper_option(WITH_TAU       "Enable TAU service (TAU Performance System)" FALSE)
-add_caliper_option(WITH_VTUNE     "Enable Intel(R) VTune(tm) annotation bindings" FALSE)
-add_caliper_option(WITH_ADIAK     "Enable adiak support" FALSE)
-add_caliper_option(WITH_KOKKOS    "Enable Kokkos profiling support" TRUE)
-add_caliper_option(WITH_PCP       "Enable performance co-pilot support" FALSE)
-add_caliper_option(WITH_VARIORUM  "Enable Variorum support" FALSE)
-add_caliper_option(WITH_UMPIRE    "Enable Umpire statistics support" FALSE)
-add_caliper_option(WITH_CRAYPAT   "Enable CrayPAT region forwarding support" FALSE)
-add_caliper_option(WITH_LDMS      "Enable LDMS forwarder" FALSE)
+add_caliper_option(WITH_TAU         "Enable TAU service (TAU Performance System)" FALSE)
+add_caliper_option(WITH_VTUNE       "Enable Intel(R) VTune(tm) annotation bindings" FALSE)
+add_caliper_option(WITH_ADIAK       "Enable adiak support" FALSE)
+add_caliper_option(WITH_KOKKOS      "Enable Kokkos profiling support" TRUE)
+add_caliper_option(WITH_PCP         "Enable performance co-pilot support" FALSE)
+add_caliper_option(WITH_VARIORUM    "Enable Variorum support" FALSE)
+add_caliper_option(WITH_UMPIRE      "Enable Umpire statistics support" FALSE)
+add_caliper_option(WITH_CRAYPAT     "Enable CrayPAT region forwarding support" FALSE)
+add_caliper_option(WITH_LDMS        "Enable LDMS forwarder" FALSE)
 
 set(WITH_ARCH "" CACHE STRING "Enable features specific to the provided archspec CPU architecture name")
 if (NOT WITH_ARCH STREQUAL "")
@@ -338,36 +336,7 @@ if (WITH_MPI)
   endif()
 endif()
 
-if (WITH_ROCPROFILER AND (WITH_ROCTX OR WITH_ROCTRACER))
-  message(WARNING "ROCPROFILER and ROCTX/ROCTRACER support are both enabled.\n"
-    "This is not recommended, please select either ROCPROFILER or ROCTRACER/ROCTX")
-endif()
-
-if (WITH_ROCTX OR WITH_ROCM)
-  include(FindRocTX)
-  if (ROCTX_FOUND)
-    set(CALIPER_HAVE_ROCTX TRUE)
-    set(CALIPER_RocTX_CMAKE_MSG "Yes, using ${ROCTX_LIBRARIES}")
-    list(APPEND CALIPER_EXTERNAL_LIBS ${ROCTX_LIBRARIES})
-  else()
-    message(WARNING "RocTX support requested but ROCm was not found!\n"
-      "Set ROCM_PATH to installation path and re-run cmake.")
-  endif()
-endif()
-
-if (WITH_ROCTRACER OR WITH_ROCM)
-  include(FindRoctracer)
-  if (ROCTRACER_FOUND)
-    set(CALIPER_HAVE_ROCTRACER TRUE)
-    set(CALIPER_RocTracer_CMAKE_MSG "Yes, using ${ROCTRACER_LIBRARIES}")
-    list(APPEND CALIPER_EXTERNAL_LIBS ${ROCTRACER_LIBRARIES})
-  else()
-    message(WARNING "RocTracer support requested but ROCm was not found!\n"
-      "Set ROCM_PATH to installation path and re-run cmake.")
-  endif()
-endif()
-
-if (WITH_ROCPROFILER)
+if (WITH_ROCPROFILER OR WITH_ROCM)
   find_package(rocprofiler-sdk REQUIRED)
   find_package(rocprofiler-sdk-roctx REQUIRED)
   if (rocprofiler-sdk_FOUND)
@@ -641,8 +610,6 @@ set(CALIPER_MODULES
   Nvtx
   CUpti
   Kokkos
-  RocTracer
-  RocTX
   rocprofiler
   TAU
   VTune

src/caliper/controllers/ROCmActivityProfileController.cpp

@@ -42,16 +42,16 @@ class RocmActivityProfileController : public cali::ChannelController
         }
 
         std::string q_local =
-            " let act_count=first(sum#count,count) if rocm.activity"
-            ",dmin=scale(min#rocm.activity.duration,1e-9)"
+            " let "
+            " dmin=scale(min#rocm.activity.duration,1e-9)"
             ",davg=scale(avg#rocm.activity.duration,1e-9)"
             ",dmax=scale(max#rocm.activity.duration,1e-9)"
-            " select *,scale(sum#time.duration.ns,1e-9) as time"
+            " select *,scale(sum#rocm.host.duration,1e-9) as time"
             ",scale(sum#rocm.activity.duration,1e-9) as \"time (gpu)\""
             ",min(dmin) as \"min time/inst\""
             ",avg(davg) as \"avg time/inst\""
             ",max(dmax) as \"max time/inst\""
-            ",sum(act_count) as count"
+            ",sum(sum#rocm.activity.count) as instances"
             " group by path,rocm.kernel.name,rocm.activity,mpi.rank "
             " format ";
 
@@ -127,14 +127,13 @@ const char* controller_spec = R"json(
  "name"        : "rocm-activity-profile",
  "description" : "Record AMD ROCm activities and a write profile",
  "categories"  : [ "adiak", "metric", "output", "region", "event" ],
- "services"    : [ "aggregate", "roctracer", "event", "timer" ],
+ "services"    : [ "aggregate", "rocprofiler", "event" ],
  "config"      :
  {
-  "CALI_CHANNEL_FLUSH_ON_EXIT"        : "false",
-  "CALI_EVENT_ENABLE_SNAPSHOT_INFO"   : "false",
-  "CALI_ROCTRACER_TRACE_ACTIVITIES"   : "true",
-  "CALI_ROCTRACER_RECORD_KERNEL_NAMES": "true",
-  "CALI_ROCTRACER_SNAPSHOT_DURATION"  : "false"
+  "CALI_CHANNEL_FLUSH_ON_EXIT": "false",
+  "CALI_EVENT_ENABLE_SNAPSHOT_INFO": "false",
+  "CALI_ROCPROFILER_ENABLE_ACTIVITY_TRACING": "true",
+  "CALI_ROCPROFILER_ENABLE_SNAPSHOT_TIMESTAMPS": "true"
  },
  "defaults"    : { "node.order": "true", "memcpy": "true" },
  "options":

src/caliper/controllers/ROCmActivityReportController.cpp

@@ -35,17 +35,17 @@ class RocmActivityReportController : public cali::ChannelController
     {
         // Config for local aggregation
         std::string local_select =
-            " inclusive_scale(sum#time.duration.ns,1e-9) as \"Host Time\""
+            " inclusive_scale(sum#rocm.host.duration,1e-9) as \"Host Time\""
             ",inclusive_scale(sum#rocm.activity.duration,1e-9) as \"GPU Time\""
-            ",inclusive_ratio(sum#rocm.activity.duration,sum#time.duration.ns,100.0) as \"GPU %\"";
+            ",inclusive_ratio(sum#rocm.activity.duration,sum#rocm.host.duration,100.0) as \"GPU %\"";
 
         // Config for second aggregation step in MPI mode (cross-process aggregation)
         std::string cross_select =
-            " avg(iscale#sum#time.duration.ns) as \"Avg Host Time\""
-            ",max(iscale#sum#time.duration.ns) as \"Max Host Time\""
+            " avg(iscale#sum#rocm.host.duration) as \"Avg Host Time\""
+            ",max(iscale#sum#rocm.host.duration) as \"Max Host Time\""
             ",avg(iscale#sum#rocm.activity.duration) as \"Avg GPU Time\""
             ",max(iscale#sum#rocm.activity.duration) as \"Max GPU Time\""
-            ",ratio(iscale#sum#rocm.activity.duration,iscale#sum#time.duration.ns,100.0) as \"GPU %\"";
+            ",ratio(iscale#sum#rocm.activity.duration,iscale#sum#rocm.host.duration,100.0) as \"GPU %\"";
 
         std::string groupby = "path";
 
@@ -119,15 +119,16 @@ const char* controller_spec = R"json(
 {
  "name"        : "rocm-activity-report",
  "description" : "Record and print AMD ROCm activities (kernel executions, memcopies, etc.)",
- "categories"  : [ "output", "region", "treeformatter", "event" ],
- "services"    : [ "aggregate", "roctracer", "event", "timer" ],
+ "categories"  : [ "output", "region", "treeformatter", "event", "metric" ],
+ "services"    : [ "aggregate", "rocprofiler", "event" ],
  "config"      :
  {
-  "CALI_CHANNEL_FLUSH_ON_EXIT"       : "false",
-  "CALI_EVENT_ENABLE_SNAPSHOT_INFO"  : "false",
-  "CALI_ROCTRACER_TRACE_ACTIVITIES"  : "true"
+  "CALI_CHANNEL_FLUSH_ON_EXIT"      : "false",
+  "CALI_EVENT_ENABLE_SNAPSHOT_INFO" : "false",
+  "CALI_ROCPROFILER_ENABLE_ACTIVITY_TRACING"    : "true",
+  "CALI_ROCPROFILER_ENABLE_SNAPSHOT_TIMESTAMPS" : "true"
  },
- "defaults"    : { "order_as_visited": "true", "output.append": "true" },
+ "defaults": { "order_as_visited": "true", "output.append": "true" },
  "options":
  [
   {
@@ -137,7 +138,6 @@ const char* controller_spec = R"json(
   },{
    "name": "show_kernels",
    "type": "bool",
-   "config": { "CALI_ROCTRACER_RECORD_KERNEL_NAMES": "true" },
    "description": "Show kernel names"
   },{
    "name": "output.append",

src/caliper/controllers/controllers.cpp

@@ -701,16 +701,16 @@ const char* builtin_rocm_option_specs = R"json(
  "type": "bool",
  "description": "Profile HIP API functions",
  "category": "region",
- "services": [ "roctracer" ],
- "config": { "CALI_ROCTRACER_TRACE_ACTIVITIES": "false" }
+ "services": [ "rocprofiler" ],
+ "config": { "CALI_ROCPROFILER_ENABLE_API_CALLBACKS": "true" }
 },
 {
  "name": "rocm.gputime",
  "description": "Report GPU time in AMD ROCm activities",
  "type": "bool",
  "category": "metric",
- "services": [ "roctracer" ],
- "config": { "CALI_ROCTRACER_TRACE_ACTIVITIES": "true", "CALI_ROCTRACER_RECORD_KERNEL_NAMES": "false" },
+ "services": [ "rocprofiler" ],
+ "config": { "CALI_ROCPROFILER_ENABLE_ACTIVITY_TRACING": "true" },
  "query":
  {
   "local": "let t.gpu.r=first(sum#rocm.activity.duration,rocm.activity.duration) select inclusive_scale(t.gpu.r,1e-9) as \"GPU time (I)\" unit sec,scale(t.gpu.r,1e-9) as \"GPU time (E)\" unit sec",

src/services/CMakeLists.txt

@@ -95,12 +95,6 @@ endif()
 if (CALIPER_HAVE_PCP)
   add_subdirectory(pcp)
 endif()
-if (CALIPER_HAVE_ROCTX)
-  add_subdirectory(roctx)
-endif()
-if (CALIPER_HAVE_ROCTRACER)
-  add_subdirectory(roctracer)
-endif()
 if (CALIPER_HAVE_ROCPROFILER)
   add_subdirectory(rocprofiler)
 endif()

src/services/rocprofiler/CMakeLists.txt

@@ -10,5 +10,4 @@ target_link_libraries(caliper-rocprofiler PRIVATE
 add_service_objlib("caliper-rocprofiler")
 
 add_caliper_service("rocprofiler CALIPER_HAVE_ROCPROFILER")
-add_caliper_service("roctracer CALIPER_HAVE_ROCPROFILER")
 add_caliper_service("roctx CALIPER_HAVE_ROCPROFILER")