Current issues with 2.3 PPC build (Now with Graphics!)

[keybounce]

I'm having trouble with 2.3 building universal on a PPC.

Code: [Select]

stbmac:23 Michael$ svn info
Path: .
URL: https://ufoai.svn.sourceforge.net/svnroot/ufoai/ufoai/branches/ufoai_2.3
Repository Root: https://ufoai.svn.sourceforge.net/svnroot/ufoai
Repository UUID: 39b98707-f80d-0410-bc7e-c6b02f5409da
Revision: 30059
Node Kind: directory
Schedule: normal
Last Changed Author: tlh2000
Last Changed Rev: 30046
Last Changed Date: 2010-05-17 23:57:30 -0700 (Mon, 17 May 2010)

stbmac:23 Michael$ svn st
?      Makefile.in.orig
?      svn-diff.txt
?      port-readme.txt
?      svn-diff.nosignal.txt
?      svn-diff-2.3.txt
stbmac:23 Michael$ make models
 * [MOD] src/tools/ufomodel/ufomodel.c
lipo: can't create output file: debug-darwin-universal/tools/ufomodel/tools/ufomodel/ufomodel.o (No such file or directory)
make: *** [debug-darwin-universal/tools/ufomodel/tools/ufomodel/ufomodel.o] Error 1
stbmac:23 Michael$
stbmac:23 Michael$ make
 * [GAM] src/game/g_actor.c
 * [GAM] src/game/g_ai.c
 * [GAM] src/game/g_ai_lua.c
 * [GAM] src/game/g_client.c
 * [GAM] src/game/g_combat.c
 * [GAM] src/game/g_cmds.c
 * [GAM] src/game/g_edicts.c
^CInterrupted by signal 2
make: *** [debug-darwin-universal/game/game/g_edicts.o] Interrupt

stbmac:23 Michael$ make models
 * [MOD] src/tools/ufomodel/ufomodel.c
 * [MOD] src/shared/mathlib.c
 * [MOD] src/shared/byte.c
 * [MOD] src/shared/images.c
 * [MOD] src/shared/parse.c
 * [MOD] src/shared/shared.c
src/shared/shared.c: In function ‘Com_sprintf’:
src/shared/shared.c:473: warning: comparison is always false due to limited range of data type
src/shared/shared.c:475: warning: comparison is always true due to limited range of data type
src/shared/shared.c: In function ‘Com_sprintf’:
src/shared/shared.c:473: warning: comparison is always false due to limited range of data type
src/shared/shared.c:475: warning: comparison is always true due to limited range of data type
 * [MOD] src/common/files.c
 * [MOD] src/common/list.c
 * [MOD] src/common/mem.c
 * [MOD] src/common/unzip.c
 * [MOD] src/common/ioapi.c
 * [MOD] src/client/renderer/r_model_alias.c
 * [MOD] src/client/renderer/r_model_dpm.c
 * [MOD] src/client/renderer/r_model_md2.c
 * [MOD] src/client/renderer/r_model_md3.c
 * [MOD] src/client/renderer/r_model_obj.c
 * [MOD] src/ports/unix/unix_main.c
 * [MOD] ... linking -rdynamic (-lz  -ljpeg  -L/opt/local/lib -lpng12 -lz   -L/opt/local/lib -lSDLmain -lSDL -Wl,-framework,Cocoa -lSDL_image  -L/opt/local/lib -lSDLmain -lSDL -Wl,-framework,Cocoa)
./ufomodel -mdx -overwrite -v  -s 0.5 -f models/aircraft/craft_arm_plate/plating.md2
---- ufomodel 0.2 ----

---- filesystem initialization -----
Adding game dir: /usr/local/share/ufoai/base
Adding game dir: /Users/Michael/Documents/UFOAI-2.3-dev/base
Adding game dir: ./base
using ./base for writing
- model 'models/aircraft/craft_arm_plate/plating.md2'
Error: R_ModForName: unknown fileid for models/aircraft/craft_arm_plate/plating.md2
make: *** [base/models/aircraft/craft_arm_plate/plating.mdx] Error 1
stbmac:23 Michael$

First, I'm looking at 30046. Starting with an absolutely stock, clean copy.

'make models' will fail -- the directories for the .o files don't exist.
'make' will set them up. And then, 'make models' will compile.

But it doesn't run. It fails. And doesn't give any indication as to why.

Gdb begins ...

Edit 1:

Code: [Select]

Breakpoint 1, Com_Error (code=0, fmt=0x2455c "R_ModForName: unknown fileid for %s") at src/tools/ufomodel/ufomodel.c:178
178 va_start(argptr, fmt);
(gdb) bt
#0  Com_Error (code=0, fmt=0x2455c "R_ModForName: unknown fileid for %s") at src/tools/ufomodel/ufomodel.c:178
#1  0x00002590 in LoadModel (name=0x277bc "models/aircraft/craft_arm_plate/plating.md2") at src/tools/ufomodel/ufomodel.c:227
#2  0x00002814 in PrecalcNormalsAndTangents (filename=0x277bc "models/aircraft/craft_arm_plate/plating.md2") at src/tools/ufomodel/ufomodel.c:284
#3  0x00003ed8 in SDL_main (argc=8, argv=0x25095c0) at src/tools/ufomodel/ufomodel.c:678
#4  0x000236ec in -[SDLMain applicationDidFinishLaunching:] ()
#5  0x96d39688 in _nsnote_callback ()
#6  0x94ce243c in _CFXNotificationPostNotification ()
#7  0x96d36dc4 in -[NSNotificationCenter postNotificationName:object:userInfo:] ()
#8  0x93b64858 in -[NSApplication _postDidFinishNotification] ()
Not very deep in the code...

[keybounce]

Alright, do you see an obvious byteswap issue here?

stbmac:23 Michael$ egrep IDALIASHEADER src/*/*.h
src/common/qfiles.h:#define IDALIASHEADER   (('2'<<24)+('P'<<16)+('D'<<8)+'I')

Code: [Select]

(gdb) p buf
$8 = (byte *) 0x282be2c "IDP2\b"

(gdb) p LittleLong(*(unsigned *) buf)
$10 = 1229213746

(gdb) p  (('2'<<24)+('P'<<16)+('D'<<8)+'I')
$13 = 844121161


207 /* call the appropriate loader */
208 switch (LittleLong(*(unsigned *) buf)) {
209 case IDALIASHEADER:
210 /* MD2 header */
211 R_ModLoadAliasMD2Model(mod, buf, modfilelen, qfalse);
212 break;
(gdb)
213
214 case DPMHEADER:
215 R_ModLoadAliasDPMModel(mod, buf, modfilelen);
216 break;
217
218 case IDMD3HEADER:
219 /* MD3 header */
220 R_ModLoadAliasMD3Model(mod, buf, modfilelen);
221 break;
222
(gdb)
223 default:
224 if (!Q_strcasecmp(mod->name + strlen(mod->name) - 4, ".obj"))
225 R_LoadObjModel(mod, buf, modfilelen);
226 else
227 Com_Error(ERR_FATAL, "R_ModForName: unknown fileid for %s", mod->name);


[keybounce]

Definitely byteswap.

- model 'models/aircraft/craft_arm_plate/plating.md2'
Debug: IDAliasHeader is 32504449, swapped is 32504449
Debug: Buf is 49445032, swapped is 49445032
Error: R_ModForName: unknown fileid for models/aircraft/craft_arm_plate/plating.md2

LittleLong() is a no-op on PPC.
The bytes are backwards (those are hex numbers).

What's the best way to solve this? Is it:
1. Redefine LittleLong to do the swapping
2. Conditional #defines so that the value is different on PPC?
3. Something else?

[keybounce]

Ok, it looks like the answer is "Something else".

- LittleLong() is used everywhere.
- IDALIASHEADER is from the quake engine, and playing with it is likely to have trouble elsewhere.

Code: [Select]

stbmac:23 Michael$ egrep IDALIASHEADER src/*/*.c
src/server/sv_world.c: case IDALIASHEADER:
stbmac:23 Michael$ egrep IDALIASHEADER src/*/*/*.c
src/client/renderer/r_model.c: case IDALIASHEADER:
src/tools/ufomodel/ufomodel.c:Com_Printf("Debug: IDAliasHeader is %x, swapped is %x\n", IDALIASHEADER, LittleLong(IDALIASHEADER));
src/tools/ufomodel/ufomodel.c: case IDALIASHEADER:
src/tools/ufomodel/ufomodel.c: case IDALIASHEADER:
Or perhaps not -- it's used once in the server, once in the client rendering, and in the ufomodel code where I'm finding it broken.

The other option is to byteswap all the data being loaded in from the file before processing. And then, it looks like the file is in i386 format as "native", and the PPC is the only thing that's going to need swapping.

Which leads to the next question: How much is the data not being properly swapped in/out during generation and processing over all of the code? Is the code supposed to be byte order fixed, or is it supposed to be "Lets assume x86"?

[Mattn]

maybe SwapInit() isn't calles for ufomodel?

[keybounce]

stbmac:23 Michael$ find src -name '*.c' -print | xargs grep SwapInit
stbmac:23 Michael$

Can't find any call to SwapInit anywhere.

[keybounce]

Problem has been traced to the libsdl port in Macports.
Workaround testing in progress.

[keybounce]

Initial report, 30076, PPC mac.

First, I have one locally compiled map -- wilderness -- for testing. Most of the time I can start a skirmish. About 25% of the time, I get an error like this:

Code: [Select]

Change gametype to 'Skirmish mode'
------- Loading game.dylib -------
Reading symbols for shared libraries . done
LoadLibrary (./base/game.dylib)
==== InitGame ====
checksum for the map 'wilderness': 3167195999
ufo script checksum 229509907
Created AI player (team 7)
-------------------------------------
Connecting to localhost...
connection attempt from loopback connection
load material file: 'materials/wilderness.mat'
S_LoadSound: Unrecognized sound file type.
S_LoadSound: Could not find sound file: 'ambience/firesoft'
added light, ambient=0.476837
Starting the game...
Michael has joined team 1
music change to PsymongN5 (from van_theme)
(player 0) It's team 1's round
Michael has taken control over team 1.
********************
ERROR: Unknown model type in R_DrawEntities entity chain: 0
********************
Shutdown server: Server crashed.
==== ShutdownGame ====
Shutdown gametype 'Skirmish mode'
0   ufo                                 0x001c4a3c Sys_Backtrace + 36
1   ufo                                 0x001378fc Com_Error + 400
2   ufo                                 0x0017d414 R_DrawEntities + 608
3   ufo                                 0x00189224 R_RenderFrame + 548
4   ufo                                 0x00047b70 CL_ViewRender + 428
5   ufo                                 0x00019aec SCR_UpdateScreen + 384
6   ufo                                 0x00017d68 CL_Frame + 416
7   ufo                                 0x00139af8 Qcommon_Init + 2596
8   ufo                                 0x0013a1c4 Qcommon_Frame + 180
9   ufo                                 0x001c4e68 main + 356
10  ufo                                 0x00002874 start + 68
music change to van_theme (from PsymongN5)
music change to van_theme (from van_theme)
GDB does not trigger (I'm running it from gdb), so I can't actually investigate what's happened.

2. Screen resolutions are things like 1024x768, while in window mode. There is no subtraction for the two menu bars, so the bottom of the map does not display. This didn't happen in the past -- not only were there many more resolutions available, I could manually go into the config file and put 960x720 in. Not anymore.

3. I cannot specify GLSL shaders. Attempting gives me the error messages
Invalid value for cvar r_programs
Invalid value for cvar r_programs
Modified render related cvars
not using GLSL shaders
Invalid value for cvar r_programs
Invalid value for cvar r_programs
Modified render related cvars
not using GLSL shaders

4. Graphic display is messed up. The map ground is funky garbage. Again.

5. Default display settings are too dark. Again, macs tend to have calibrated displays because the OS includes a good calibration tool -- the default should be "use system default". Looking in the config file, there is a line for

set vid_ignoregamma "1" a

However, setting that to 1 (default was 0) did not actually work. Raising brightness seemed to be more important than gamma.

6. Attempting to start a campaign game brings up a world map where everything on the globe is too dark to see.

More information as I test more.

[keybounce]

Previously, I had built 26262 (trunk) and it seemed to be working OK. Now, I am unable to get 30076 (branch) to work correctly. My test map is wilderness.

26262 was able to display a map pretty much OK. There were occasional lighting issues, with no rhyme or reason to them -- things would appear dark. Sometimes it could be fixed by moving the mouse off window and back on. Sometimes by going down to level 1 and then back up. Never anything consistent. With wilderness specifically, in the upper right corner I can tell that there is a natural rock staircase, but there's also some strange stuff that I can't figure out by looking at it.

30076 is giving graphics that are mostly garbled, constantly too dark. But oddly, the water and wood dock staircase of Wilderness is almost perfect.

Here are some screen shots showing what I'm referring to.

(NB: going into a skirmish will sometimes give me a crash from the battle server, with no indication of why.)

Hardware: PPC mac mini, 9300 video card, 9200 driver. I don't think it has core image support. Will not let me select GLSL shaders.

[keybounce]

Sadly, I have to upload each screenshot separately.

Note that in 26262, I could specify 960 x 720 for the window size. At the time, I was using a 1024x768 monitor, and needed that. (Now I'm using a widescreen TV and the DVD drive ...)

Here's the water / dock corner, with an example of the old lighting issue.

[keybounce]

And here is the new (30076) build with the graphics.

Sadly, the water corner makes too big a file to upload here.

[keybounce]

Looking at 30076 on a different machine.

This one is an iBook g4, with a 9550. This supports GLSL shaders and core image.

Turning on GLSL shaders fixes the lighting nicely. But it raises the seconds per frame up to 25.

No shaders works ... well, more-or-less the same. I did find one camera position where the lighting flicker was very apparent. Here are two screen shots, with no motion, key pressing, or anything -- just an example of the back-and-forth lighting issue.

[keybounce]

EDIT: Make that a 9550 mobility. (Apparently, "mobility" is an important distinction.)

Finally, here is a look at the rear strip on the wilderness map. This back portion looks so different -- like a sharp knife -- that I'm hoping it gives someone who knows what's happening a forehead slap.

[Mattn]

if you have the time - please do some revision hoping to get the revision that reintroduced the lightmap artifacts on the ppc - did you try to disable glsl on the ppc, too? or only on the intel mac?

[keybounce]

I do not have an intel mac. Both machines are PPC.

I will do revision hoping tomorrow. The machine that could use GLSL shaders was too slow -- 25 seconds per frame -- to be usable.