[GHC] #7968: fft2 benchmark sometimes fails (nofib)

#7968: fft2 benchmark sometimes fails (nofib) ----------------------------------------+----------------------------------- Reporter: jstolarek | Owner: Type: bug | Status: new Priority: normal | Component: NoFib benchmark suite Version: 7.7 | Keywords: Os: Linux | Architecture: x86_64 (amd64) Failure: Incorrect result at runtime | Blockedby: Blocking: | Related: ----------------------------------------+----------------------------------- I noticed that sometimes `fft2` program in nofib benchmark suite returns values that are slightly different from expected (see error message at the end of this report). The most worrying for me is that it sometimes works properly and sometimes fails, as if the correctness of result depended on some random factor. {{{ ././fft2 512 < /dev/null expected stdout not matched by reality --- fft2.stdout7 2013-05-15 08:01:24.014049002 +0200 +++ /tmp/runtest10927.1 2013-06-04 18:06:08.401977002 +0200 @@ -1,3 +1,3 @@ -result1 = 2.59635799135966e-12 -result2 = 2.59635799135966e-12 -result3 = 4.8279900966008427e-8 +result1 = 2.6712796173433053e-12 +result2 = 2.6712796173433053e-12 +result3 = 4.827444399779779e-8 real 0m0.055s user 0m0.049s sys 0m0.005s ././fft2 512 < /dev/null expected stdout not matched by reality --- fft2.stdout7 2013-05-15 08:01:24.014049002 +0200 +++ /tmp/runtest10963.1 2013-06-04 18:06:08.507977002 +0200 @@ -1,3 +1,3 @@ -result1 = 2.59635799135966e-12 -result2 = 2.59635799135966e-12 -result3 = 4.8279900966008427e-8 +result1 = 2.6712796173433053e-12 +result2 = 2.6712796173433053e-12 +result3 = 4.827444399779779e-8 real 0m0.054s user 0m0.047s sys 0m0.005s ././fft2 512 < /dev/null expected stdout not matched by reality --- fft2.stdout7 2013-05-15 08:01:24.014049002 +0200 +++ /tmp/runtest10999.1 2013-06-04 18:06:08.602977002 +0200 @@ -1,3 +1,3 @@ -result1 = 2.59635799135966e-12 -result2 = 2.59635799135966e-12 -result3 = 4.8279900966008427e-8 +result1 = 2.6712796173433053e-12 +result2 = 2.6712796173433053e-12 +result3 = 4.827444399779779e-8 real 0m0.052s user 0m0.044s sys 0m0.007s ././fft2 512 < /dev/null expected stdout not matched by reality --- fft2.stdout7 2013-05-15 08:01:24.014049002 +0200 +++ /tmp/runtest11035.1 2013-06-04 18:06:08.712977002 +0200 @@ -1,3 +1,3 @@ -result1 = 2.59635799135966e-12 -result2 = 2.59635799135966e-12 -result3 = 4.8279900966008427e-8 +result1 = 2.6712796173433053e-12 +result2 = 2.6712796173433053e-12 +result3 = 4.827444399779779e-8 real 0m0.052s user 0m0.046s sys 0m0.006s ././fft2 512 < /dev/null expected stdout not matched by reality --- fft2.stdout7 2013-05-15 08:01:24.014049002 +0200 +++ /tmp/runtest11071.1 2013-06-04 18:06:08.816977002 +0200 @@ -1,3 +1,3 @@ -result1 = 2.59635799135966e-12 -result2 = 2.59635799135966e-12 -result3 = 4.8279900966008427e-8 +result1 = 2.6712796173433053e-12 +result2 = 2.6712796173433053e-12 +result3 = 4.827444399779779e-8 make[2]: *** [runtests] Błąd 1 Failed making all in fft2: 1 make[1]: *** [all] Błąd 1 Failed making all in spectral: 1 make: *** [all] Błąd 1 }}} -- Ticket URL: http://hackage.haskell.org/trac/ghc/ticket/7968 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler

#7968: fft2 benchmark sometimes fails (nofib) ----------------------------------------+----------------------------------- Reporter: jstolarek | Owner: Type: bug | Status: new Priority: normal | Component: NoFib benchmark suite Version: 7.7 | Keywords: Os: Linux | Architecture: x86_64 (amd64) Failure: Incorrect result at runtime | Blockedby: Blocking: | Related: ----------------------------------------+----------------------------------- Comment(by jstolarek): When I was reporting this bug I didn't notice that the values that are being matched are almost zero - no wonder that round-off errors might lead to slightly different results. It seems that whoever wrote this benchmark is well aware of this, because there are thirteen (sic!) different files with expected output values! This in itself shows that the expected results are not stable. Anyway, I tried to be smart and changed the expected values in fft2.stdout7 to what I get on my machine, but now mandel benchmark fails: {{{ real 0m3.537s user 0m3.504s sys 0m0.021s ././mandel < mandel.stdin expected stdout not matched by reality Binary files mandel.stdout-mingw and /tmp/runtest16086.1 differ }}} -- Ticket URL: http://hackage.haskell.org/trac/ghc/ticket/7968#comment:1 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler
participants (1)
-
GHC