diff options
Diffstat (limited to 'doc/ports/notes.dsux')
| -rw-r--r-- | doc/ports/notes.dsux | 282 |
1 files changed, 282 insertions, 0 deletions
diff --git a/doc/ports/notes.dsux b/doc/ports/notes.dsux new file mode 100644 index 00000000..0f132092 --- /dev/null +++ b/doc/ports/notes.dsux @@ -0,0 +1,282 @@ +Begin port to DECstation 3100. 23 July 1989. +Start with source only archive of VAX/Ultrix system. +[note - rsh/tar hung up mysteriously while loading files onto decstation, +[and had to be restarted]. +-------------------------- + +unix/as.mips + +unix/bin.mips + + Created AS and BIN directories for Dec/MIPS. (7/23) + +unix/as.mips/zsvjmp.s + Prepared a dummy version of zsvjmp for the moment. All this does is + set the output status value to zero. (prepared by using the compiler + to generate the assembler source for a small C program). (7/23) + +unix/boot/spp/xc.c + It appears that for some reason, the MIPS C compiler or linker will + not accept a file with a .e extension as the output (-o) file. + Had to modify XC to output a .E file temporarily, which is renamed + to .e after the link. (7/23) + +unix/mkpkg.sh +unix/os/mkpkg.sh +unix/shlib/mkpkg.sh +unix/boot/bootlib/mkpkg.sh +unix/boot/generic/mkpkg.sh +unix/boot/mkpkg/mkpkg.sh +unix/boot/rmbin/mkpkg.sh +unix/boot/rmfiles/mkpkg.sh +unix/boot/rtar/mkpkg.sh +unix/boot/spp/mkpkg.sh +unix/boot/spp/mkxc.sh +unix/boot/spp/rpp/mkpkg.sh +unix/boot/spp/xpp/mkpkg.sh +unix/boot/wtar/mkpkg.sh +unix/boot/xyacc/mkpkg.sh +unix/gdev/sgidev/mkpkg.sh +unix/bin.mips/fixnames + + Modified all the mkpkg.sh files to generate .E files. Added a + fixnames script to bin.mips to rename any .E files therein to .e, + and added code the the root mkpkg.sh to call fixnames at the + end of the bootstrap. (7/23) + +unix/hlib/cl.csh +unix/hlib/mkiraf.csh +unix/hlib/install +unix/hlib/libc/iraf.h + 1. Modified the iraf root pathnames in these files. + 2. In the install, also deleted the entries for edsym.e to avoid + an error message during the install. (7/23) + +unix/hlib/extern.pkg + Commented out the stsdas entries. (7/23) + +unix/hlib/mkpkg.inc +unix/hlib/irafuser.csh +unix/hlib/mkpkg.sf.MIPS + Set MACH to MIPS. Added a case for machine type MIPS. Added an + empty special file list. (7/23) + +unix/hlib/motd + Modified to say DECstation/IRAF. (7/23) + +unix/os/zxwhen.c + Added a #ifdef mips section with some hardware exception codes for + the MIPS. (7/23) + +-------------------- +Start bootstrap. (7/23) + +unix/hlib/irafuser.csh + Linking of RPP produced a "gp relocation out of range" error message. + It was necessary to compile with "-G 0" (no global variables indexed + by global pointer) in order to get this to go away. (7/23) + +unix/boot/bootlib/mkpkg.sh + Modified to mv libboot.a to $hbin (../../bin) rather than $hlib, + since the libboot.a in $hlib is a symbolic link. (7/23) + +-------------------- +Start sysgen. (7/23) + +unix/as/gkiprint.o + +unix/hlib/mkpkg.sf.MIPS + The file gkiprint.x would fail with a slew (like, 700) syntax and + other errors. There was nothing wrong with the code and the few + workarounds I tried had no effect. I finally managed to compile + the file by breaking it up into half a dozen pieces, compiling each + separately, and linking the resultant objects into a single large + object gkiprint.o. Since this bug will probably go away before long + I merely placed the manually generated object file in AS, and made + an entry in the special file list to use this object. (7/23) + +unix/as/imsetr.f + +unix/hlib/mkpkg.sf.MIPS + In this case the file imsetr.f would cause the first pass of f77 + to core dump (not an optimizer problem). I had to hack the fortran + code to workaround the bug, placing the hacked file in AS. The + offending statement was: + + memi(im+33) = and (memi(+33) , not(1 )) + + the following version is ok: + + i = not(1) + memi(im+33) = and (memi(+33) , i) + + The bug must have something to do with the bit intrinsics. (7/23) + +unix/as/fseti.f + # FIO +unix/as/fmfcache.f + # FMIO +unix/as/fmlfbrd.f + +unix/as/fmlfbwr.f + +unix/as/fmlfcopy.f + +unix/as/fmlfundel.f + +unix/hlib/mkpkg.sf.MIPS + Modified to work around the same compiler bug as above. The NOT of + any compile time integer constant used in an expression seems to + always cause a core dump. (7/23) + +unix/boot/spp/xc.c +unix/hlib/mkpkg.inc + 1. In the first sysgen, x_images.e failed to link with a "gp relocation + out of range" type problem. Recompiling the IMAGES package libpkg.a + with -G 0 fixed the problem. + 2. Passing this flag through XC, however, was not possible without + a code change to XC, as there was no way to pass two consecutive + arguments such as "-G 0". In order to support this XC was modified + to permit flags of the form "-/#flag", which are passed as "flag". + This is similar to "-/flag" which is passed as "-flag", except that + it allows flags with no "-" to be passed. Hokey, but backwards + compatible and it solves the problem. (7/23) + +----------------- +Started sysgen of NOAO packages. (7/23) + +noao/lib/database.o + +noao/lib/mkpkg.inc +noao/lib/mkpkg.sf.MIPS + The database.x file in DTOI would cause the compiler to become + terminally confused, as with gkiprint above. Manually constructed + the database.o object, placed the latter in noao$lib, and modified + the NOAO special file list to use this object rather than try to + compile the file. (7/24) + +unix/hlib/config.h +unix/hlib/libc/spp.h + Set the size of the ZDOJMP jump buffer to 84 (83 ints for the jmp_buf + buffer plus 1 for the address of the status variable). (7/24) + +unix/as.mips/zsvjmp.s + Wrote and installed the real zsvjmp.s. (7/24) + +------------------- +Sysgen/relink with new zsvjmp. +Everything appears to work now. (7/24) + +bin.mips +noao/bin.mips + Stripped the BINS to save disk space, as it appeared that the symbol + table information is unusually large on the decstation. (7/25) + + Normal Stripped + + core system bin 21.5 16.0 + noao bin 18.0 12.3 + ------- -------- + 39.5 28.3 + +unix/as/zsvjmp.s + Modified slightly to eliminate a delay slot. This failed to work + at first and I eventually discovered that the reason was that the + *assembler* was modifying my assembler code, reordering the + instructions and inserting nop's in delay slots, producing invalid + code! I had to add a "noreorder" directive to get it to leave the + code alone. (7/25) + +unix/os/zxwhen.c + As a first step in testing exception handling I tried the following + in the CL: + + cl> = 5 / 0 + + This caused the CL to core dump on the uncaught signal SIGTRAP. + On all other unix systems thus far, integer divide by zero has been + mapped to SIGFPE, but on the MIPS it is a trap. Modified zxwhen.c + to map SIGTRAP onto the VOS exception class X_ARITH. (7/25) + +unix/os/zzstrt.c + Added the following #ifdef mips code to zzstrt, to enable the overflow, + invalid, and divzero IEEE exceptions: + + set_fpc_csr (get_fpc_csr() | 07000); + + The {set|get}_fpc_csr() routines appear to be undocumented but I + found them anyhow by searching the symbol tables of the host libraries + in /usr/lib. (7/25) + + [LATER] Well, enabling the IEEE exceptions works, but somehow this + is making the software misbehave in a way I do not yet understand + and do not have time to look into right now. Completely unrelated + things like "stty vt100", "imheader dev$pix" fail mysteriously with + the IEEE trap enable bits set in the FPU context and status register. + This doesn't make any sense and we will have to do without the IEEE + traps for now. Possibly this is due to a bug in the OS which will + go away in the next release. + +unix/hlib/mkfloat.csh + The code in this script which lists the tar archive back with + "tar -tf", to verify that a good archive has been generated before + deleting the files, does not work on the DECstation. The problem + is that tar -tf produces different output here than on the other + systems; the directories are listed as files in addition to the files. + I had to change the + + tar -tf + to + tar -tf | grep -v '/$' + + to get things to work. (7/26) + + +unix/bin.mips +noao/bin.generic +bin.generic + Stripped all the executables (HSI, IRAF, NOAO). + Made IRAF and NOAO generic. (7/26) + + +Summary of various problems encountered. +------------------- + + fortran compiler core dumps + c compiler core dumps + cannot -o to a .e file + rsh|tar hung once + NO ADB!! + dbx dies occasionally with internal error: stack overflow + 2 user limit on logins - prevents even having two windows + open from Sun with a different rlogin session in each. + System is very fast single user; much slower if anything else + is running on the machine (memory too small?). + System crashed on first day of testing with 2 users busy doing things. + (there did not seem to be a system message file). + Some routines which use floating point produce different results on + the MIPS than on other systems. Examples thus far are the FITS + programs, and IMHISTOGRAM. No doubt more will be discovered. + IEEE exception handling is not working currently. + + +---------------- +V2.8BETA system frozen and archived. (7/26) + +---------------- +Begin upgrade to V2.9ALPHA, 6 DEC 1989. +Working with newly arrived IRAF DECstation 3100! + + Installed the V2.8ALPHA HSI. + Merged in all relevant V2.9 modifications to the HSI. + +unix/boot/spp/xc.c + Changed the syntax "-/#foo" -> "foo" to "-//foo" -> "foo" for + escaping command line arguments to XC. (1/6) + +---------------- +Started a bootstap. (1/6) + +mkpkg + Extended the mkpkg summary filter in the root mkpkg to filter out + the great number of "local variable iyy never used" warning messages + produced by the MIPS compiler. (1/6) + +pkg/plot/t_graph.x + This file contained a routine im_projection equivalent to a routine + of the same name in file improject.x, resulting in two externals of + the same name in the package library. (1/7) + +unix/boot/spp/xc.c + To avoid a circular reference problem in the DS3100 Fortran libraries, + which causes an undefined symbols "s_abort" problem when linking some + programs, added a second search of libF77 to the end of the XC host + library link list, after the -lm. (1/7) |