#2172 gcc-torture-execute-pr47337 fails for stm8 when using --max-allcos-per-node 15000

closed-works-for-me
nobody
None
STM8
7
2014-08-29
2013-05-25
No

When using --max-allocs-per-node 15000, the regression test gcc-torture-execute-pr47337 fails by timeout for stm8 as of revision #8675. To reproduce, add --max-allos-per-node 15000 to SDCCFLAGS in ports/stm8/spec.mk.

Philipp

Discussion

  • Philipp Klaus Krause

    It seems the call to baz() doesn't return.

    Philipp

     
  • Philipp Klaus Krause

    Something weird is going on here. If I add e = 0; anywhere (even at the beginning of testBug()) before the call to baz(), the test passes.

    Philipp

     
  • Philipp Klaus Krause

    I've looked into this further. But the generated .asm code still looks ok to me. At one time I even had a situation where the only difference in the generated asm code was that the one that passed had an subw x, #0x0001 in a palce where the failing version had a decw x (and the flags were not used). I have looked at the simulator, and both the subw and decw seem to be correct. Also many other tests that use subw, decw or both pass. I wonder if there is some weird interaction of code size with the test failing or whatever.

    For now I'll just disable this test, sine it starts to fail when doing semingly unrelated changes. If I find the time to do so later, I might also tra to put this test or something similar onr ela hardware.

    Philipp

     
  • Ben Shi

    Ben Shi - 2014-08-29
    • status: open --> closed-works-for-me
     
  • Ben Shi

    Ben Shi - 2014-08-29

    I can not reproduce it in rev9062. This test case passes both with --max-allcos-per-node 15000 specified and without.

    And there is a slight difference,
    with --max-allcos-per-node 15000
    gcc-torture-execute-pr47337 (f: 0, t: 0, c: 1, b: 2444, t: 17886)
    without
    gcc-torture-execute-pr47337 (f: 0, t: 0, c: 1, b: 2460, t: 18409)

     

Get latest updates about Open Source Projects, Conferences and News.

Sign up for the SourceForge newsletter:





No, thanks