[Bug middle-end/105001] New: If executing with non-nvptx offloading, but nvptx offloading compilation is enabled: FAIL: libgomp.c/pr104783.c execution test

tschwinge at gcc dot gnu.org gcc-bugzilla@gcc.gnu.org
Mon Mar 21 14:59:46 GMT 2022


https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105001

            Bug ID: 105001
           Summary: If executing with non-nvptx offloading, but nvptx
                    offloading compilation is enabled: FAIL:
                    libgomp.c/pr104783.c execution test
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Keywords: openmp
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: tschwinge at gcc dot gnu.org
                CC: ams at gcc dot gnu.org, jakub at gcc dot gnu.org, jules at gcc dot gnu.org,
                    vries at gcc dot gnu.org
  Target Milestone: ---

If executing with non-nvptx offloading (tested: GCN, Intel MIC (emulated)), but
nvptx offloading compilation is enabled, the PR104783 test case
'libgomp.c/pr104783.c' FAILs its execution test, SIGSEGV-like.

GCN:

    Memory access fault by GPU node-1 (Agent handle: 0x1a93b40) on address
(nil). Reason: Page not present or supervisor privilege.

Intel MIC (emulated):

    offload error: process on the device 0 unexpectedly exited with code 0

The problem goes away if compiling without nvptx offloading:
'-foffload=amdgcn-amdhsa', for example.

So, I suppose a problem with the middle end SIMT transformations?


Compiling manually, also I see:

    source-gcc/libgomp/testsuite/libgomp.c/pr104783.c: In function
‘main._omp_fn.0’:
    source-gcc/libgomp/testsuite/libgomp.c/pr104783.c:10:9: warning:
‘<anonymous>’ is used uninitialized [-Wuninitialized]
       10 | #pragma omp atomic update
          |         ^~~
    source-gcc/libgomp/testsuite/libgomp.c/pr104783.c:10:9: note: ‘<anonymous>’
was declared here
       10 | #pragma omp atomic update
          |         ^~~


More information about the Gcc-bugs mailing list