[Dri-devel] r200: TCL - ATI glory / (real) Benchmarks / GREAT work Keith!

SourceForge Headquarters 225 Broadway Suite 1600 San Diego, CA 92101 +1 (858) 454-5900

I am on the r200 branch since it's out. I've started with a somewhat 
overclocked Radeon 8500 QL (64 MB DDR-SDRAM) from one of my friends and own a 
very similar but not yet overclocked gfk, now.

Let's start with my system specs:

dual Athlon MP 1900+
MSI K7D Master-L (AMD 760 MPX)
1 GB DDR-SDRAM 266 CL2
Linux 2.4.19-jam0-pc (-AA stuff, SMP stuff, highmem DRM fix + page coloring)

I have to say it again:
It is a true SMP system so maybe some problems are SMP (locking) related.

After Keith released the first page flipping stuff gears was running at ~2970 
fps and Q3 show shining 130 fps @640x480 on a 1280x1024x24 desktop with the 
overclocked board. Gears went down to ~2890 fps and Q3 to 126 fps with the 
normal one.  Then he rearranged some ioctl stuff (last week) and gears was 
down to ~550/560 fps. UT and the Q3 ID logo and intro stuff stuttering very 
much since then. Only on my SMP system? Could it be texture upload? I see 
some bad numbers for some "texdown" modes.

As sone as he introduced the first TCL support on Friday (?) gears went up to 
~600 fps but stays there solid. Any answers?

So "gears" is a very usefull development tool to find immediately if some 
things are going "wrong". Not more and not less.

Now only numbers for my own board.
Latest r200 code from today.
1280x1024x24 desktop

Mesa/demos> ./glinfo
unknown chip id, assuming full r200 support
GL_VERSION: 1.2 Mesa 4.0.3
GL_EXTENSIONS: GL_ARB_imaging GL_ARB_multitexture GL_ARB_texture_env_add 
GL_ARB_texture_env_combine GL_ARB_texture_env_dot3 GL_ARB_transpose_matrix 
GL_EXT_abgr GL_EXT_bgra GL_EXT_blend_color GL_EXT_blend_logic_op 
GL_EXT_blend_minmax GL_EXT_blend_subtract GL_EXT_clip_volume_hint 
GL_EXT_convolution GL_EXT_compiled_vertex_array GL_EXT_histogram 
GL_EXT_packed_pixels GL_EXT_polygon_offset GL_EXT_rescale_normal 
GL_EXT_secondary_color GL_EXT_texture3D GL_EXT_texture_env_add 
GL_EXT_texture_env_combine GL_EXT_texture_env_dot3 
GL_EXT_texture_filter_anisotropic GL_EXT_texture_object 
GL_EXT_texture_lod_bias GL_EXT_vertex_array GL_IBM_rasterpos_clip 
GL_MESA_window_pos GL_NV_texgen_reflection GL_SGI_color_matrix 
GL_SGI_color_table
GL_RENDERER: Mesa DRI R200 20020611 AGP 4x x86/MMX/3DNow!/SSE TCL
GL_VENDOR: Tungsten Graphics, Inc.
GLU_VERSION: 1.3
GLU_EXTENSIONS: GLU_EXT_nurbs_tessellator GLU_EXT_object_space_tess
GLUT_API_VERSION: 3
GLUT_XLIB_IMPLEMENTATION: 15

Q3 (1.31, demo four, 640x480 window):
* stuttering during the ID Logo and intro (cinematics)
* 122 fps ???

UT (436):
* stuttering from time to time (the horizontal brown beam is very long then)
* 6.84 - 130 fps (the low during the stuttering)

Now some GREAT shining numbers.

gloss:
* cylinder is at 406-410 fps but was at ~445 (OC gfk) and ~420 without TCL
* teapot went from 84 fps (both cards) to 111 fps with TCL ;-)

VTK
* vtk sphere-bench.tcl-2.1 increased to much that my system is the 2end
   fastest ever tested in the world !!!
* when ATI ever could release Hyper-Z, TRUFORM, Scalable Level of Detail and
   maybe (S3) texture compression it could be the leader

Look here:
http://www.barre.nom.fr/vtk/bench.html

The VTK Simple Sphere Benchmark Results Database
http://rrvtk.free.fr/vtk/perfs/
http://rrvtk.free.fr/vtk/perfs/results.php3?famille=%25&flats=1&flat=1&strip=1&transp=1&wire=1&txt=1&txtp=1

Best results, summary :
  - i686 running Linux 2.4.19-jam0-pc
  - VTK 4.1.1 (rev: 1.1305, 2002/08/13 02:45:03)
  - OpenGL
  - Visual is 1280x1024, truecolor/truecolor/24
  - Tcl/Tk 8.3.3
  - window was 400 x 400
   1) 512x512 : 5981.7 kpolys/s : [stripper]
   2) 512x512 : 3633.6 kpolys/s : [small_sphere]
   3) 512x512 : 3624.2 kpolys/s : []
   4) 512x512 : 3620.4 kpolys/s : [transparency]
   5) 256x256 : 2911.0 kpolys/s : [texture, transparency]
   6) 512x512 : 2908.9 kpolys/s : [texture]
   7) 512x512 : 2603.1 kpolys/s : [wireframe]

TCL is so GOOD and worth it.
Please compare the detailed numbers (VTK 3.2.0) for TCL and without.

Without TCL:

------- BENCH COMBINATIONS -------

Benching for sphere resolutions : 32, 64, 128, 256, 512
Setting(s) : window is 400 x 400, sphere radius is 0.9
 Option(s) : [stripper]
    32x32  : 1062.8 kpolys/s
    64x64  : 1968.8 kpolys/s
   128x128 : 2406.7 kpolys/s
   256x256 : 2592.6 kpolys/s
   512x512 : 3062.0 kpolys/s

Benching for sphere resolutions : 32, 64, 128, 256, 512
Setting(s) : window is 400 x 400, sphere radius is 0.5
 Option(s) : [small_sphere]
    32x32  :  838.6 kpolys/s
    64x64  : 1174.6 kpolys/s
   128x128 : 1330.0 kpolys/s
   256x256 : 1373.2 kpolys/s
   512x512 : 1384.5 kpolys/s

Benching for sphere resolutions : 32, 64, 128, 256, 512
Setting(s) : window is 400 x 400, sphere radius is 0.9
    32x32  :  840.4 kpolys/s
    64x64  : 1189.2 kpolys/s
   128x128 : 1327.9 kpolys/s
   256x256 : 1371.1 kpolys/s
   512x512 : 1383.6 kpolys/s

Benching for sphere resolutions : 32, 64, 128, 256, 512
Setting(s) : window is 400 x 400, sphere radius is 0.9
 Option(s) : [transparency]
    32x32  :  838.8 kpolys/s
    64x64  : 1189.0 kpolys/s
   128x128 : 1329.0 kpolys/s
   256x256 : 1372.6 kpolys/s
   512x512 : 1384.5 kpolys/s

Benching for sphere resolutions : 32, 64, 128, 256, 512
Setting(s) : window is 400 x 400, sphere radius is 0.9
 Option(s) : [wireframe]
    32x32  :  664.7 kpolys/s
    64x64  : 1018.6 kpolys/s
   128x128 : 1122.1 kpolys/s
   256x256 : 1152.0 kpolys/s
   512x512 : 1159.2 kpolys/s

Benching for sphere resolutions : 32, 64, 128, 256, 512
Setting(s) : window is 400 x 400, sphere radius is 0.9
 Option(s) : [texture]
    32x32  :  732.2 kpolys/s
    64x64  : 1020.9 kpolys/s
   128x128 : 1136.8 kpolys/s
   256x256 : 1168.6 kpolys/s
   512x512 : 1176.7 kpolys/s

Benching for sphere resolutions : 32, 64, 128, 256, 512
Setting(s) : window is 400 x 400, sphere radius is 0.9
 Option(s) : [texture, transparency]
    32x32  :  728.6 kpolys/s
    64x64  : 1024.0 kpolys/s
   128x128 : 1133.9 kpolys/s
   256x256 : 1168.3 kpolys/s
   512x512 : 1176.9 kpolys/s

Best results, summary :
  - i686 running Linux 2.4.19-rc1-jam2-pc
  - VTK 3.2.0 (rev: 1.1265, 2002/07/20 00:00:27)
  - OpenGL
  - Visual is 1280x1024, truecolor/truecolor/24
  - Tcl/Tk 8.3.3
  - window was 400 x 400
   1) 512x512 : 3062.0 kpolys/s : [stripper]
   2) 512x512 : 1384.5 kpolys/s : [transparency]
   3) 512x512 : 1384.5 kpolys/s : [small_sphere]
   4) 512x512 : 1383.6 kpolys/s : []
   5) 512x512 : 1176.9 kpolys/s : [texture, transparency]
   6) 512x512 : 1176.7 kpolys/s : [texture]
   7) 512x512 : 1159.2 kpolys/s : [wireframe]

------- DONE (in 154 sec.) -------

With TCL:

------- BENCH COMBINATIONS -------

Benching for sphere resolutions : 32, 64, 128, 256, 512
Setting(s) : window is 400 x 400, sphere radius is 0.9
 Option(s) : [stripper]
    32x32  :  957.1 kpolys/s
    64x64  : 3887.3 kpolys/s
   128x128 : 5179.6 kpolys/s
   256x256 : 5396.0 kpolys/s
   512x512 : 5981.9 kpolys/s

Benching for sphere resolutions : 32, 64, 128, 256, 512
Setting(s) : window is 400 x 400, sphere radius is 0.5
 Option(s) : [small_sphere]
    32x32  : 1533.1 kpolys/s
    64x64  : 2900.1 kpolys/s
   128x128 : 3588.7 kpolys/s
   256x256 : 3705.7 kpolys/s
   512x512 : 3740.5 kpolys/s

Benching for sphere resolutions : 32, 64, 128, 256, 512
Setting(s) : window is 400 x 400, sphere radius is 0.9
    32x32  : 1016.8 kpolys/s
    64x64  : 2803.1 kpolys/s
   128x128 : 3578.8 kpolys/s
   256x256 : 3694.8 kpolys/s
   512x512 : 3729.0 kpolys/s

Benching for sphere resolutions : 32, 64, 128, 256, 512
Setting(s) : window is 400 x 400, sphere radius is 0.9
 Option(s) : [transparency]
    32x32  :  769.8 kpolys/s
    64x64  : 2885.2 kpolys/s
   128x128 : 3580.8 kpolys/s
   256x256 : 3693.6 kpolys/s
   512x512 : 3727.3 kpolys/s

Benching for sphere resolutions : 32, 64, 128, 256, 512
Setting(s) : window is 400 x 400, sphere radius is 0.9
 Option(s) : [wireframe]
    32x32  :  997.1 kpolys/s
    64x64  : 2119.2 kpolys/s
   128x128 : 2596.7 kpolys/s
   256x256 : 2632.3 kpolys/s
   512x512 : 2647.2 kpolys/s

Benching for sphere resolutions : 32, 64, 128, 256, 512
Setting(s) : window is 400 x 400, sphere radius is 0.9
 Option(s) : [texture]
    32x32  :  843.4 kpolys/s
    64x64  : 2121.6 kpolys/s
   128x128 : 2676.8 kpolys/s
   256x256 : 2853.6 kpolys/s
   512x512 : 2910.5 kpolys/s

Benching for sphere resolutions : 32, 64, 128, 256, 512
Setting(s) : window is 400 x 400, sphere radius is 0.9
 Option(s) : [texture, transparency]
    32x32  :  857.8 kpolys/s
    64x64  : 2206.8 kpolys/s
   128x128 : 2690.5 kpolys/s
   256x256 : 2861.3 kpolys/s
   512x512 : 2909.9 kpolys/s

Best results, summary :
  - i686 running Linux 2.4.19-jam0-pc
  - VTK 3.2.0 (rev: 1.1278, 2002/08/02 00:00:47)
  - OpenGL
  - Visual is 1280x1024, truecolor/truecolor/24
  - Tcl/Tk 8.3.3
  - window was 400 x 400
   1) 512x512 : 5981.9 kpolys/s : [stripper]
   2) 512x512 : 3740.5 kpolys/s : [small_sphere]
   3) 512x512 : 3729.0 kpolys/s : []
   4) 512x512 : 3727.3 kpolys/s : [transparency]
   5) 512x512 : 2910.5 kpolys/s : [texture]
   6) 512x512 : 2909.9 kpolys/s : [texture, transparency]
   7) 512x512 : 2647.2 kpolys/s : [wireframe]

------- DONE (in 67 sec.) -------

Now some bad news:

viewperf (6.1.2) do not running, yet.

Directory: /tmp/INSTALL/SOURCE/viewperf-6.1.2/AWadvs-04
unknown chip id, assuming full r200 support
viewperf: r200_vtxfmt.c:1061: r200VtxfmtUnbindContext: Assertion `vb.context 
== ctx' failed.
Abbruch

Directory: /opt/VTK/V4.0/VTK/bin
VTK/bin> ./TaskParallelism
unknown chip id, assuming full r200 support
r200_makeX86Normal3fv/197 CVAL 0 OFFSET 14 VAL 41968e20
r200_makeX86Normal3fv/198 CVAL 4 OFFSET 20 VAL 41968e24
r200_makeX86Normal3fv/199 CVAL 8 OFFSET 25 VAL 41968e28
r200_makeX86Normal3fv done
unknown chip id, assuming full r200 support
TaskParallelism: r200_vtxfmt.c:920: r200FlushVertices: Assertion `vb.context 
== ctx' failed.
Abbruch

Mesa/xdemos> ./wincopy
glXMakeContextCurrent failed in Redraw()
glXMakeContextCurrent failed in Redraw()
glXMakeContextCurrent failed in Redraw()
[-]

Context switching is broken?

Mesa/demos> startparsec

 PARSEC V0.99 build 0197.
 Copyright (c) 1996-2001 by Markus Hadwiger,
 Andreas Varga, Clemens Beer, Michael Woegerbauer,
 Alex Mastny, Stefan Poiss. All Rights Reserved.
---------------------------------------------------
 mailto:pa...@pa...    http://www.parsec.org
---------------------------------------------------

 LAN-TEST BUILD 0197.
 NON-COMMERCIAL DISTRIBUTION ONLY.
 SEE THE ACCOMPANYING README FOR DETAILS.

---------------------------------------------------

Using the OpenGL subsystem as rendering device.
Using XFree86-VidModeExtension 2.1.
Started sound server with pid 3881
IP address: 192.168.0.1
TCP/IP networking support found.
Network game will be available.
Sound system initialized.
Joystick code is disabled.
Booting game core.
Calculating colors...done.
Keyboard grab failed!
Joystick code is disabled.
X Error of failed request:  BadMatch (invalid parameter attributes)
  Major opcode of failed request:  42 (X_SetInputFocus)
  Serial number of failed request:  23
  Current serial number in output stream:  23
Executing console script (boot.con)...

>>>Keyboard grab failed!<<<

Should be enough for now.

Keith go one with your great work!
Kudos to Tungsten Graphics, too.
Last but not least the Weather Channel ;-)

-Dieter
-- 
Dieter Nützel
Graduate Student, Computer Science

University of Hamburg
Department of Computer Science
@home: Die...@ha...