TLS support problem

[gcongiu]

_papi_hwi_my_thread is a TLS global variable. The variable, currently, is guarded in both threads.c & threads.h and declared only if the PAPI configure script can detect TLS support (either by default or as selected by the user through the --with-tls=<keyword> configure option). If TLS support is not detected the variable is undeclared and the build fails, as following reported by static analysis:

/spack/opt/spack/linux-rocky8-x86_64/gcc-9.5.0/llvm-16.0.6-gweege3utpobgkimb4fej76zih53yc6i/bin/../libexec/ccc-analyzer -fPIC -DPIC -shared -Wl,-soname -Wl,libpapi.so.7.0 -Xlinker "-rpath" -Xlinker "/usr/local/lib" -DPAPI_NO_MEMORY_MANAGEMENT -DSTATIC_PAPI_EVENTS_TABLE  -DUSE_PERFEVENT_RDPMC=1 -DPEINCLUDE=\"libpfm4/include/perfmon/perf_event.h\" -D_REENTRANT -D_GNU_SOURCE -DNO_TLS -Ilibpfm4/include -fvisibility=hidden -I. -g -Wextra  -Wall -DPAPI_NUM_COMP=3 -DOSLOCK=\"linux-lock.h\" -DOSCONTEXT=\"linux-context.h\" -O2 x86_cpuid_info.c papi_libpfm4_events.c papi.c papi_internal.c high-level/papi_hl.c extras.c sw_multiplex.c upper_PAPI_FWRAPPERS.c papi_fwrappers_.c papi_fwrappers__.c threads.c cpus.c linux-memory.c linux-timer.c linux-common.c  papi_preset.c papi_vector.c papi_memory.c components/perf_event/perf_event.c components/perf_event/pe_libpfm4_events.c components/perf_event_uncore/perf_event_uncore.c components/sysdetect/sysdetect.c components/sysdetect/nvidia_gpu.c components/sysdetect/amd_gpu.c components/sysdetect/cpu.c components/sysdetect/cpu_utils.c components/sysdetect/os_cpu_utils.c components/sysdetect/linux_cpu_utils.c components/sysdetect/x86_cpu_utils.c  -o libpapi.so.7.0.1.0 -Bdynamic -Llibpfm4/lib -lpfm 
[402](https://github.com/icl-utk-edu/papi/actions/runs/5519307326/jobs/10064398824?pr=45#step:3:403)
papi_internal.c: In function ‘_papi_hwi_set_papi_event_code’:
[403](https://github.com/icl-utk-edu/papi/actions/runs/5519307326/jobs/10064398824?pr=45#step:3:404)
papi_internal.c:124:3: error: ‘_papi_hwi_my_thread’ undeclared (first use in this function); did you mean ‘_papi_hwi_read’?
[404](https://github.com/icl-utk-edu/papi/actions/runs/5519307326/jobs/10064398824?pr=45#step:3:405)
   _papi_hwi_my_thread->tls_papi_event_code_changed = -1;
[405](https://github.com/icl-utk-edu/papi/actions/runs/5519307326/jobs/10064398824?pr=45#step:3:406)
   ^~~~~~~~~~~~~~~~~~~
[406](https://github.com/icl-utk-edu/papi/actions/runs/5519307326/jobs/10064398824?pr=45#step:3:407)
   _papi_hwi_read
[407](https://github.com/icl-utk-edu/papi/actions/runs/5519307326/jobs/10064398824?pr=45#step:3:408)
papi_internal.c:124:3: note: each undeclared identifier is reported only once for each function it appears in
[408](https://github.com/icl-utk-edu/papi/actions/runs/5519307326/jobs/10064398824?pr=45#step:3:409)
papi_internal.c: In function ‘_papi_hwi_get_papi_event_code’:
[409](https://github.com/icl-utk-edu/papi/actions/runs/5519307326/jobs/10064398824?pr=45#step:3:410)
papi_internal.c:138:9: error: ‘_papi_hwi_my_thread’ undeclared (first use in this function); did you mean ‘_papi_hwi_read’?
[410](https://github.com/icl-utk-edu/papi/actions/runs/5519307326/jobs/10064398824?pr=45#step:3:411)
  return _papi_hwi_my_thread->tls_papi_event_code;
[411](https://github.com/icl-utk-edu/papi/actions/runs/5519307326/jobs/10064398824?pr=45#step:3:412)
         ^~~~~~~~~~~~~~~~~~~
[412](https://github.com/icl-utk-edu/papi/actions/runs/5519307326/jobs/10064398824?pr=45#step:3:413)
         _papi_hwi_read
[413](https://github.com/icl-utk-edu/papi/actions/runs/5519307326/jobs/10064398824?pr=45#step:3:414)
papi_internal.c: In function ‘_papi_hwi_native_to_eventcode’:
[414](https://github.com/icl-utk-edu/papi/actions/runs/5519307326/jobs/10064398824?pr=45#step:3:415)
papi_internal.c:565:7: error: ‘_papi_hwi_my_thread’ undeclared (first use in this function); did you mean ‘_papi_hwi_read’?
[415](https://github.com/icl-utk-edu/papi/actions/runs/5519307326/jobs/10064398824?pr=45#step:3:416)
   if (_papi_hwi_my_thread->tls_papi_event_code_changed > 0) {
[416](https://github.com/icl-utk-edu/papi/actions/runs/5519307326/jobs/10064398824?pr=45#step:3:417)
       ^~~~~~~~~~~~~~~~~~~
[417](https://github.com/icl-utk-edu/papi/actions/runs/5519307326/jobs/10064398824?pr=45#step:3:418)
       _papi_hwi_read
[418](https://github.com/icl-utk-edu/papi/actions/runs/5519307326/jobs/10064398824?pr=45#step:3:419)
papi_internal.c: In function ‘_papi_hwi_get_papi_event_code’:
[419](https://github.com/icl-utk-edu/papi/actions/runs/5519307326/jobs/10064398824?pr=45#step:3:420)
papi_internal.c:139:1: warning: control reaches end of non-void function [-Wreturn-type]
[420](https://github.com/icl-utk-edu/papi/actions/runs/5519307326/jobs/10064398824?pr=45#step:3:421)
 }
[421](https://github.com/icl-utk-edu/papi/actions/runs/5519307326/jobs/10064398824?pr=45#step:3:422)
 ^

[gcongiu]

@jagode, threads.c contains the following code:

#ifdef HAVE_THREAD_LOCAL_STORAGE
#define THREAD_LOCAL_STORAGE_KEYWORD HAVE_THREAD_LOCAL_STORAGE
#else
#define THREAD_LOCAL_STORAGE_KEYWORD
#endif

#if defined(HAVE_THREAD_LOCAL_STORAGE)
extern THREAD_LOCAL_STORAGE_KEYWORD ThreadInfo_t *_papi_hwi_my_thread;
#endif

The code above has the following problem: if, for some reason, the PAPI configure script is unable to automatically detect the supported TLS and set HAVE_THREAD_LOCAL_STORAGE to the right keyword (e.g. __thread), PAPI fails to build. I don't think this is reasonable and we should have a fallback for when this happens. It almost looks like the code above had some logic to account for a fallback through the THREAD_LOCAL_STORAGE_KEYWORD, but that seems an half baked solution to me. In reality there is no real fallback.

[ashterenli]

I think I'm hitting this problem, when trying to build papi-7.0.1 with AMD aocc 4.1.0, which is based on clang-16.

I get:

checking for high performance thread local storage... no

and than

cc -fPIC -DPIC -shared -Wl,-soname -Wl,libpapi.so.7.0 -Xlinker "-rpath" -Xlinker "/localhome/shterenlikht/install/u_aocc/4.1.0/papi-7.0.1/lib" -DPAPI_NO_MEMORY_MANAGEMENT -DSTATIC_PAPI     _EVENTS_TABLE  -DUSE_PERFEVENT_RDPMC=1 -DPEINCLUDE=\"libpfm4/include/perfmon/perf_event.h\" -D_REENTRANT -D_GNU_SOURCE -DNO_TLS

Note -DNO_TLS

and finally:

2741 papi_internal.c:124:3: error: use of undeclared identifier '_papi_hwi_my_thread'; did you mean '_papi_hwi_read'?
2742                 _papi_hwi_my_thread->tls_papi_event_code_changed = -1;
2743                 ^~~~~~~~~~~~~~~~~~~
2744                 _papi_hwi_read
2745 ./papi_internal.h:466:5: note: '_papi_hwi_read' declared here
2746 int _papi_hwi_read( hwd_context_t * context, EventSetInfo_t * ESI,
2747     ^

[ashterenli]

Adding --with-tls=_Thread_local seems to work:

checking for high performance thread local storage... _Thread_local