Jacob (=Jouk) Jansen
Mon, 20 Mar 2000 23:41:53 -0800
Hi, >There have been a bunch of seperate reports of memory problems under Dec >OSF/Tru64/Digital Unix. They all seem to boild down to the fact that the >VB->IM->OrFlag and VB->IM->AndFlag variables are being computed incorrectly in >gl_compute_orflag() in vbxform.c. > >I'm wondering if there is anyone out there who can easily reproduce any of >these problems can print out the values that are computed for OrFlag and >AndFlag, along with the array of values in VB->IM->Flag from which they are >computed. > >An even more adventurous soul might want to look at enabling some of the >printf's in gl_compute_orflag and seeing what might be going wrong. > >I'm also interested to see if this is a 64-bit issue that we only see under >Dec OSF, or if it is a compiler-optimization issue specific to that platform >and not really the code's fault -- so could people try disabling optimization >(-O0) and rerunning. > >The faults I'm describing are characterized by crashes in either > find_last_3f >or fixup_inputs > >I'm interested to hear if people are having crashes in these functions on >non-alpha systems I'm running Mesa on an alpha-system, with a different OS (VMS). Most of the Compaq-compiler have the same kernel. I never have seen crashes in the above routines. What compiler (DECC, gcc ,....) are used to get the error occuring? My general experience with Mesa on Alpha-VMS and DECC is that often crashes occur from initialization errors (parameters are may not be set to zero at the start of allocation, but Mesa asumes zero's or has typo's). Different optimaizations make a huge difference if this is the case. In the VMS-version I see one crash only at the moment. I use the latest alpha version of xlockmore (ftp://ftp.tux.org./pub/tux/bagleyd/xlockmore/) for testting. Running with xlock -mode random -install -modelist cage,rubik always results in a crash (tracebacks see below) However, if one of the other OpenGL modes is runned first, i.e. xlock -mode random -install -modelist allgl (click middle mousebutton to switch modes) also the rubik and cage modes appear nicely on the screen. In the xlockmore developing group we looked at the problem but could not find any problem, which seems not to occur on other systems than VMS-Alpha. Does have anybody any idea? Jouk polka-jj) xlock -nice 0 -inwindow -modelist allgl polka-jj) xlock -nice 0 -inwindow -modelist cage,rubik %SYSTEM-F-HPARITH, high performance arithmetic trap, Imask=00000000, Fmask=00000 001, summary=08, PC=0000000000AC7EB4, PS=0000001B -SYSTEM-F-FLTOVF, arithmetic trap, floating overflow at PC=0000000000AC7EB4, PS= 0000001B %TRACE-F-TRACEBACK, symbolic stack dump follows image module routine line rel PC abs PC LIBMESAGL SHADE shade_fast_rgba_two_sided_compacted 25911 0000000000010934 0000000000AC7EB4 LIBMESAGL PIPELINE gl_run_pipeline 20606 0000000000000B3C 0000000000ADCE5C LIBMESAGL VBXFORM gl_execute_cassette 21639 0000000000001B88 0000000000ACD0C8 LIBMESAGL VBXFORM gl_maybe_transform_vb 20756 00000000000000A4 0000000000ACB5E4 LIBMESAGL DLIST _mesa_EndList 25857 000000000000B828 00000000009E88B8 XLOCK CAGE draw_woodplank 29973 000000000000063C 00000000001D255C XLOCK CAGE draw_impossiblecage 29992 0000000000000F40 00000000001D2E60 XLOCK CAGE draw_cage 30167 0000000000000000 0000000000000000 XLOCK XLOCK justDisplay 32922 00000000000033B0 00000000001233B0 XLOCK XLOCK main 33928 0000000000005D78 0000000000125D78 XLOCK XLOCK __main 0 0000000000000070 0000000000120070 XLOCK 0 000000000022D5C0 000000000023D5C0 0 FFFFFFFF8403B3D4 FFFFFFFF8403B3D4 polka-jj) xlock -nice 0 -inwindow -modelist cage,rubik %SYSTEM-F-HPARITH, high performance arithmetic trap, Imask=00000000, Fmask=00000 001, summary=08, PC=0000000000AC7EC0, PS=0000001B -SYSTEM-F-FLTOVF, arithmetic trap, floating overflow at PC=0000000000AC7EC0, PS= 0000001B %TRACE-F-TRACEBACK, symbolic stack dump follows image module routine line rel PC abs PC LIBMESAGL SHADE shade_fast_rgba_two_sided_compacted 25913 0000000000010940 0000000000AC7EC0 LIBMESAGL PIPELINE gl_run_pipeline 20606 0000000000000B3C 0000000000ADCE5C LIBMESAGL VBXFORM gl_execute_cassette 21639 0000000000001B88 0000000000ACD0C8 LIBMESAGL VBXFORM gl_maybe_transform_vb 20756 00000000000000A4 0000000000ACB5E4 LIBMESAGL DLIST _mesa_EndList 25857 000000000000B828 00000000009E88B8 XLOCK RUBIK draw_cubit 28426 00000000000010D0 00000000001DBAA0 XLOCK RUBIK draw_cube 28601 0000000000003C5C 00000000001DE62C XLOCK RUBIK draw_rubik 29674 0000000000006900 00000000001E12D0 XLOCK XLOCK justDisplay 32922 00000000000033B0 00000000001233B0 XLOCK XLOCK main 33928 0000000000005D78 0000000000125D78 XLOCK XLOCK __main 0 0000000000000070 0000000000120070 XLOCK 0 000000000022D5C0 000000000023D5C0 0 FFFFFFFF8403B3D4 FFFFFFFF8403B3D4 Ceterum censeo tertium millennium post Christum natum anno MMI incepturum esse >------------------------------------------------------------------------------< Jouk Jansen [EMAIL PROTECTED] Technische Universiteit Delft tttttttttt uu uu ddddddd Nationaal centrum voor HREM tttttttttt uu uu dd dd Rotterdamseweg 137 tt uu uu dd dd 2628 AL Delft tt uu uu dd dd Nederland tt uu uu dd dd tel. 31-15-2781536 tt uuuuuuu ddddddd >------------------------------------------------------------------------------< _______________________________________________ Mesa-dev maillist - [EMAIL PROTECTED] http://lists.mesa3d.org/mailman/listinfo/mesa-dev