home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!mcsun!uknet!pavo.csi.cam.ac.uk!camcus!jp107
- From: jp107@cus.cam.ac.uk (Jon Peatfield)
- Newsgroups: comp.sys.hp
- Subject: X11R5 server SLOW, on 705 with GRX + xterm bug
- Message-ID: <JP107.92Sep15004240@grus.cus.cam.ac.uk>
- Date: 14 Sep 92 23:42:42 GMT
- Sender: news@infodev.cam.ac.uk (USENET news)
- Distribution: comp
- Organization: U of Cambridge, England
- Lines: 77
- Nntp-Posting-Host: grus.cus.cam.ac.uk
-
- I've just built the X11R5 (patchlevel 17) server on an HP705, using
- the HP server bits that were recently announced, and compiling with
- the unbundled cc (since I don't yet fully trust gcc-2.2.2 on the HP.)
- I have a few problems with it, like it core dumps when I quit by any
- method I've tried and won't reset the screen back to text mode... but
- my major concern is just how slow it is. Moving windows is so slow
- that I can see it drawing in the background -- even my aging
- SparcStation-1 (but mono screen -- bwtwo hardware in fact), looks fast
- by comparison. I've run some comparison tests with the x11perf
- comparing it to the X11R4 server that HP ship with the 705, and the
- difference is quite enlightening.
-
- With things to do with data structure access and the like, the R5
- server is slightly faster, but any access to the screen and it slows
- down by up to a factor of 64, the best (or worst if you prefer)
- example being for 500x500 stippled rectangle fill:
-
- HP-R4 MIT-R5 ratio Operation
- 243000.0 390000.0 ( 1.60) Dot
- 41800.0 48700.0 ( 1.17) 1x1 161x145 stippled rectangle
- 33600.0 8000.0 ( 0.24) 10x10 161x145 stippled rectangle
- 4950.0 138.0 ( 0.028) 100x100 161x145 stippled rectangle
- 364.0 5.8 ( 0.016) 500x500 161x145 stippled rectangle
- 38700.0 44100.0 ( 1.14) 1x1 161x145 tiled rectangle
- 25100.0 9530.0 ( 0.38) 10x10 161x145 tiled rectangle
- 1610.0 431.0 ( 0.27) 100x100 161x145 tiled rectangle
- 85.7 21.8 ( 0.25) 500x500 161x145 tiled rectangle
- 54100.0 88800.0 ( 1.64) X protocol NoOperation
- 439.0 682.0 ( 1.55) GetAtomName
- 436.0 648.0 ( 1.49) GetProperty
- 10700.0 12000.0 ( 1.12) Change graphics context
-
- Now I've seen things like this before when comparing accelerated
- Xservers against unaccelerated Xservers, so I'm starting to think that
- maybe the GRX has an accelerator, which the HP server makes use of,
- but the sample Ngle code doesn't. Can anyone confirm or deny this?
-
- Does anyone have enough info about the Ngle hardware (specifically the
- S9000_ID_A1439A, since that is what the GRX and CRX are), to write the
- ddx server code which will be able to make use of it? I'm prepared to
- do some hacking in the source, and once knew something about graphics
- systems, but havn't hacked the Xserver code before. Can anyone with
- more experience help me?
-
-
- Now for something slightly related.
-
- Has anyone else tried compiling the rest of the X11R5 distribution?
- If so you will probably have found that the clients/xterm won't
- compile. At least it doesn't on my systems. The problem is the use
- of _file in line 1059 of clients/xterm/main.c. Looking at the code it
- should all be inside the #ifdef DEBUG there anyway, but the use of
- _file doesn't work on HP-UX (8.07 anyway), since there is no _file
- field of the FILE struct. I altered this to use __fileL and __fileH,
- which is just as bad. *Why the hell is xterm so hacky?*
-
- Here is my diff:
- 1049a1050
- > #endif /* DEBUG */
- 1058,1061d1058
- < #ifdef hpux
- < stderr->__fileL = i %% 256 ;
- < stderr->__fileH = i / 256 ;
- < #else /* hpux */
- 1063d1059
- < #endif /* hpux */
- 1072d1067
- < #endif /* DEBUG */
-
- The xterm compiled this way seems to work, but I make no promises...
-
- -- Jon
- --
- Jon Peatfield, Computer Officer, the DAMTP, University of Cambridge
- Telephone: (+44 223) 33-7852 Mail: J.S.Peatfield@amtp.cam.ac.uk
-
- I'm an international arms dealer. You can be one too. Ask me how!
-