cuda: fix bookkeeping events

argonne-lcf · Jun 13, 2024 · 6488701 · 6488701
1 parent 870abfd
commit 6488701
Showing 1 changed file with 48 additions and 7 deletions.
diff --git a/xprof/xprof.rb.in b/xprof/xprof.rb.in
@@ -492,16 +492,55 @@ end
 
 def enable_events_cuda(channel_name, tracing_mode: 'default', profiling: true)
   lttng_enable_bookkeeping = "lttng enable-event --userspace --session=#{lttng_session_uuid_bookkeeping} --channel=#{channel_name}"
+  required_events = [
+    "lttng_ust_cuda:cuGetProcAddress_v2_entry",
+    "lttng_ust_cuda:cuGetProcAddress_v2_exit",
+  ].join(',')
+    #"lttng_ust_cuda:cuGetExportTable_entry",
+    #"lttng_ust_cuda:cuGetExportTable_exit"
+    #"lttng_ust_cuda:cuGetProcAddress_v2_entry",
+    #"lttng_ust_cuda:cuGetProcAddress_v2_exit",
+    #"lttng_ust_cuda:cuDeviceGetCount_entry",
+    #"lttng_ust_cuda:cuDeviceGetCount_exit",
+    #"lttng_ust_cuda:cuDeviceGet_entry",
+    #"lttng_ust_cuda:cuDeviceGet_exit",
+    #"lttng_ust_cuda:cuDeviceGetName_entry",
+    #"lttng_ust_cuda:cuDeviceGetName_exit",
   bookkeeping_events = [
-    "lttng_ust_cuda:cuCtx*",
-    "lttng_ust_cuda:cuDevicePrimaryCtxRetain",
-    "lttng_ust_cuda:cuStreamCreate",
-    "lttng_ust_cuda:cuModuleGetFunction"
+    "lttng_ust_cuda:cuCtxCreate_entry",
+    "lttng_ust_cuda:cuCtxCreate_exit",
+    "lttng_ust_cuda:cuCtxCreate_v2_entry",
+    "lttng_ust_cuda:cuCtxCreate_v2_exit",
+    "lttng_ust_cuda:cuCtxCreate_v3_entry",
+    "lttng_ust_cuda:cuCtxCreate_v3_exit",
+    "lttng_ust_cuda:cuCtxDestroy_entry",
+    "lttng_ust_cuda:cuCtxDestroy_exit",
+    "lttng_ust_cuda:cuCtxDestroy_v2_entry",
+    "lttng_ust_cuda:cuCtxDestroy_v2_exit",
+    "lttng_ust_cuda:cuCtxSetCurrent_entry",
+    "lttng_ust_cuda:cuCtxSetCurrent_exit",
+    "lttng_ust_cuda:cuCtxPushCurrent_entry",
+    "lttng_ust_cuda:cuCtxPushCurrent_exit",
+    "lttng_ust_cuda:cuCtxPushCurrent_v2_entry",
+    "lttng_ust_cuda:cuCtxPushCurrent_v2_exit",
+    "lttng_ust_cuda:cuCtxPopCurrent_entry",
+    "lttng_ust_cuda:cuCtxPopCurrent_exit",
+    "lttng_ust_cuda:cuCtxPopCurrent_v2_entry",
+    "lttng_ust_cuda:cuCtxPopCurrent_v2_exit",
+    "lttng_ust_cuda:cuDevicePrimaryCtxRetain_entry",
+    "lttng_ust_cuda:cuDevicePrimaryCtxRetain_exit",
+    "lttng_ust_cuda:cuStreamCreate_entry",
+    "lttng_ust_cuda:cuStreamCreate_exit",
+    "lttng_ust_cuda:cuStreamCreateWithPriority_entry",
+    "lttng_ust_cuda:cuStreamCreateWithPriority_exit",
+    "lttng_ust_cuda:cuModuleGetFunction_entry",
+    "lttng_ust_cuda:cuModuleGetFunction_exit"
   ].join(',')
-  exec("#{lttng_enable_bookkeeping} #{bookkeeping_events}")
+  exec("#{lttng_enable_bookkeeping} #{bookkeeping_events},#{required_events}")
 
   lttng_enable = "lttng enable-event --userspace --session=#{lttng_session_uuid} --channel=#{channel_name}"
   exec("#{lttng_enable} lttng_ust_cuda:* -x #{bookkeeping_events}")
+  exec("#{lttng_enable} #{required_events}")
   exec("#{lttng_enable} lttng_ust_cuda_properties")
   exec("#{lttng_enable} lttng_ust_cuda_profiling:*") if profiling
 end
@@ -613,10 +652,12 @@ def on_node_processing(backends, syncd)
     opts << "--output #{thapi_trace_dir_tmp}"
     opts << "--backends #{backends.join(',')}"
     exec("#{BINDIR}/babeltrace_thapi #{opts.join(' ')} -- #{lttng_trace_dir_tmp}")
-    FileUtils.rm_f(lttng_trace_dir_tmp)
+    #FileUtils.rm_f(lttng_trace_dir_tmp)
+    LOGGER.info("tmpdir #{lttng_trace_dir_tmp}")
   else
     FileUtils.mkdir_p(File.dirname(thapi_trace_dir_tmp))
-    exec("mv #{lttng_trace_dir_tmp} #{thapi_trace_dir_tmp}")
+    exec("cp -rp #{lttng_trace_dir_tmp} #{thapi_trace_dir_tmp}")
+    LOGGER.info("tmpdir #{lttng_trace_dir_tmp}")
   end
 
   # Global Barrier