[OMPT] Use frames at different level when using clang version 5 or higher with debug flag

[sconvent]

Summary

Clang 5 or higher adds an intermediate function call in certain cases when compiling with debug flag. This revision updates the testcases to work correctly.

[Hahnfeld]

(note: please post full context information as described in https://llvm.org/docs/Phabricator.html#requesting-a-review-via-the-web-interface)

runtime/test/ompt/callback.h

74

I propose to always issue this warning: The test will also fail if the user specifies -DDEBUG without -g. In normal testing with lit, this won't do any harm because stderr is filtered away if the test is successful.

[Hahnfeld]

Just had a more thorough look at the tests which include all kinds of outlined functions. I previously suggested print_frame_in_task() because I thought Clang would only wrap the function for tasks. Maybe print_frame_from_outlined_fn() would be a better name after all?

runtime/test/ompt/parallel/nested.c

28–29

This should use the new macro

[sconvent]

Implemented all suggestions.

[Hahnfeld]

LG

runtime/test/ompt/callback.h

   printf("%" PRIu64 ": __builtin_frame_address(%d)=%p\n", ompt_get_thread_data()->value, level, __builtin_frame_address(level));\
 } while(0)
 
+// clang (version 5.0 and above) adds an intermediate function call with debug flag (-g)
+#if defined(TEST_NEED_PRINT_FRAME_IN_TASK)
+  #if defined(DEBUG) && defined(__clang__) && __clang_major__ >= 5
+    #define print_frame_in_task(level) print_frame(level+1)
+  #else
+    #define print_frame_in_task(level) print_frame(level)
+  #endif
+  #if !defined(DEBUG) && defined(__clang__) && __clang_major__ >= 5
+    #warning "Clang 5.0 and later add an additional wrapper function for tasks when compiling with debug information."
+    #warning "Please define -DDEBUG iff you manually pass in -g!"
+  #endif
+#endif
 
 // This macro helps to define a label at the current position that can be used
 // to get the current address in the code.

runtime/test/ompt/misc/control_tool.c

 // RUN: %libomp-compile-and-run | FileCheck %s
 // REQUIRES: ompt
 // XFAIL: gcc-4, gcc-5, gcc-6, gcc-7
+#define TEST_NEED_PRINT_FRAME_IN_TASK
 #include "callback.h"
 #include <omp.h>
 

 {
   #pragma omp parallel num_threads(1)
   {
-    print_frame(1);
+    print_frame_in_task(1);
     print_frame(0);
     omp_control_tool(omp_control_tool_flush, 1, NULL);
     print_current_address(0);

 
   // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
 
-  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: __builtin_frame_address(1)=[[EXIT_FRAME:0x[0-f]*]]
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: __builtin_frame_address({{.}})=[[EXIT_FRAME:0x[0-f]*]]
   // CHECK: {{^}}[[MASTER_ID]]: __builtin_frame_address(0)=[[REENTER_FRAME:0x[0-f]*]]
   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_control_tool: command=3, modifier=1, arg=[[NULL]], codeptr_ra=[[RETURN_ADDRESS:0x[0-f]*]], current_task_frame.exit=[[EXIT_FRAME]], current_task_frame.reenter=[[REENTER_FRAME]]
   // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]

runtime/test/ompt/parallel/nested.c

 // RUN: %libomp-compile-and-run | %sort-threads | FileCheck --check-prefix=THREADS %s
 // REQUIRES: ompt
 // XFAIL: gcc-4, gcc-5, gcc-6, gcc-7
+#define TEST_NEED_PRINT_FRAME_IN_TASK
 #include "callback.h"
 #include <omp.h>
 #include <unistd.h>

 
   #pragma omp parallel num_threads(4)
   {
-    print_frame(1);
+    print_frame_in_task(1);
     print_ids(0);
     print_ids(1);
     print_frame(0);

     
     #pragma omp parallel num_threads(4)
     {
+// clang (version 5.0 and above) adds an intermediate function call with debug flag (-g)
+#if defined(DEBUG) && defined(__clang__) && __clang_major__ >= 5
+      print_frame(2);
+#else
+      print_frame(1);
+#endif
       print_frame(1);
       print_ids(0);
       print_ids(1);

 
   // nested parallel masters
   // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[MASTER_ID]]: __builtin_frame_address(1)=[[EXIT:0x[0-f]+]]
+  // THREADS: {{^}}[[MASTER_ID]]: __builtin_frame_address({{.}})=[[EXIT:0x[0-f]+]]
   // THREADS: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
   // THREADS: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID:[0-9]+]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]
   // THREADS: {{^}}[[MASTER_ID]]: __builtin_frame_address(0)=[[REENTER:0x[0-f]+]]
   // THREADS: {{^}}[[MASTER_ID]]: ompt_event_parallel_begin: parent_task_id=[[IMPLICIT_TASK_ID]], parent_task_frame.exit=[[EXIT]], parent_task_frame.reenter=[[REENTER]], parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=4, codeptr_ra=[[NESTED_RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}, invoker=[[PARALLEL_INVOKER]]
   // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[MASTER_ID]]: __builtin_frame_address(1)=[[NESTED_EXIT:0x[0-f]+]]
+  // THREADS: {{^}}[[MASTER_ID]]: __builtin_frame_address({{.}})=[[NESTED_EXIT:0x[0-f]+]]
   // THREADS: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_EXIT]], reenter_frame=[[NULL]]
   // THREADS: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[REENTER]]
   // THREADS: {{^}}[[MASTER_ID]]: task level 2: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]

runtime/test/ompt/tasks/explicit_task.c

 // RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
 // REQUIRES: ompt
 // XFAIL: gcc-4, gcc-5, gcc-6, gcc-7
+#define TEST_NEED_PRINT_FRAME_IN_TASK
 #include "callback.h"
 #include <omp.h> 
 

   print_frame(0);
   #pragma omp parallel num_threads(2)
   {
-    print_frame(1);
+    print_frame_in_task(1);
     print_ids(0);
     print_ids(1);
     print_frame(0);

   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_parallel_begin: parent_task_id=[[PARENT_TASK_ID:[0-9]+]], parent_task_frame.exit=[[NULL]], parent_task_frame.reenter=[[MAIN_REENTER]], parallel_id=[[PARALLEL_ID:[0-9]+]], requested_team_size=2, codeptr_ra=0x{{[0-f]+}}, invoker=[[PARALLEL_INVOKER:[0-9]+]]
   // nested parallel masters
   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // CHECK: {{^}}[[MASTER_ID]]: __builtin_frame_address(1)=[[EXIT:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: __builtin_frame_address({{.}})=[[EXIT:0x[0-f]+]]
   // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
   // CHECK: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID:[0-9]+]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]
   // CHECK: {{^}}[[MASTER_ID]]: __builtin_frame_address(0)=[[REENTER:0x[0-f]+]]

   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
 
   // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // CHECK: {{^}}[[THREAD_ID]]: __builtin_frame_address(1)=[[EXIT:0x[0-f]+]]
+  // CHECK: {{^}}[[THREAD_ID]]: __builtin_frame_address({{.}})=[[EXIT:0x[0-f]+]]
   // CHECK: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
   // CHECK: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]
   // CHECK: {{^}}[[THREAD_ID]]: __builtin_frame_address(0)=[[REENTER:0x[0-f]+]]

runtime/test/ompt/tasks/serialized.c

 // RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
 // REQUIRES: ompt
 // XFAIL: gcc-4, gcc-5, gcc-6, gcc-7
+#define TEST_NEED_PRINT_FRAME_IN_TASK
 #include "callback.h"
 #include <omp.h>   
 #include <math.h>

   print_frame(0);
   #pragma omp parallel num_threads(2)
   {
-    print_frame(1);
+    print_frame_in_task(1);
     print_ids(0);
     print_ids(1);
     print_frame(0);

 
   // nested parallel masters
   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // CHECK: {{^}}[[MASTER_ID]]: __builtin_frame_address(1)=[[EXIT:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: __builtin_frame_address({{.}})=[[EXIT:0x[0-f]+]]
   // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
   // CHECK: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID:[0-9]+]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]
   // CHECK: {{^}}[[MASTER_ID]]: __builtin_frame_address(0)=[[REENTER:0x[0-f]+]]

   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
 
   // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // CHECK: {{^}}[[THREAD_ID]]: __builtin_frame_address(1)=[[EXIT:0x[0-f]+]]
+  // CHECK: {{^}}[[THREAD_ID]]: __builtin_frame_address({{.}})=[[EXIT:0x[0-f]+]]
   // CHECK: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
   // CHECK: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]
   // CHECK: {{^}}[[THREAD_ID]]: __builtin_frame_address(0)=[[REENTER:0x[0-f]+]]

runtime/test/ompt/tasks/task_in_joinbarrier.c

 // RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
 // REQUIRES: ompt
+#define TEST_NEED_PRINT_FRAME_IN_TASK
 #include "callback.h"
 #include <omp.h> 
 

   print_frame(0);
   #pragma omp parallel num_threads(2)
   {
-    print_frame(1);
+    print_frame_in_task(1);
     print_ids(0);
     print_ids(1);
     print_frame(0);

   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_parallel_begin: parent_task_id=[[PARENT_TASK_ID:[0-9]+]], parent_task_frame.exit=[[NULL]], parent_task_frame.reenter=[[MAIN_REENTER]], parallel_id=[[PARALLEL_ID:[0-9]+]], requested_team_size=2, codeptr_ra=0x{{[0-f]+}}, invoker=[[PARALLEL_INVOKER:[0-9]+]]
   // nested parallel masters
   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // CHECK: {{^}}[[MASTER_ID]]: __builtin_frame_address(1)=[[EXIT:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: __builtin_frame_address({{.}})=[[EXIT:0x[0-f]+]]
   // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
   // CHECK: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID:[0-9]+]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]
   // CHECK: {{^}}[[MASTER_ID]]: __builtin_frame_address(0)=[[REENTER:0x[0-f]+]]

   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
 
   // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // CHECK: {{^}}[[THREAD_ID]]: __builtin_frame_address(1)=[[EXIT:0x[0-f]+]]
+  // CHECK: {{^}}[[THREAD_ID]]: __builtin_frame_address({{.}})=[[EXIT:0x[0-f]+]]
   // CHECK: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
   // CHECK: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]
   // CHECK: {{^}}[[THREAD_ID]]: __builtin_frame_address(0)=[[REENTER:0x[0-f]+]]

runtime/test/ompt/tasks/untied_task.c

 // RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
 // REQUIRES: ompt
+#define TEST_NEED_PRINT_FRAME_IN_TASK
 #include "callback.h"
 #include <omp.h> 
 

   print_frame(0);
   #pragma omp parallel num_threads(2)
   {
-    print_frame(1);
+    print_frame_in_task(1);
     print_ids(0);
     print_ids(1);
     print_frame(0);

   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_parallel_begin: parent_task_id=[[PARENT_TASK_ID:[0-9]+]], parent_task_frame.exit=[[NULL]], parent_task_frame.reenter=[[MAIN_REENTER]], parallel_id=[[PARALLEL_ID:[0-9]+]], requested_team_size=2, codeptr_ra=0x{{[0-f]+}}, invoker=[[PARALLEL_INVOKER:[0-9]+]]
   // nested parallel masters
   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // CHECK: {{^}}[[MASTER_ID]]: __builtin_frame_address(1)=[[EXIT:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: __builtin_frame_address({{.}})=[[EXIT:0x[0-f]+]]
   // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
   // CHECK: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID:[0-9]+]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]
   // CHECK: {{^}}[[MASTER_ID]]: __builtin_frame_address(0)=[[REENTER:0x[0-f]+]]

   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
 
   // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // CHECK: {{^}}[[THREAD_ID]]: __builtin_frame_address(1)=[[EXIT:0x[0-f]+]]
+  // CHECK: {{^}}[[THREAD_ID]]: __builtin_frame_address({{.}})=[[EXIT:0x[0-f]+]]
   // CHECK: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
   // CHECK: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]
   // CHECK: {{^}}[[THREAD_ID]]: __builtin_frame_address(0)=[[REENTER:0x[0-f]+]]