home *** CD-ROM | disk | FTP | other *** search
-
- Known problems:
- - All demos/programs only work at 16-bit color or better
- - Bench (all versions):
- - 10x10 test doesn't display anything
- - if you move the GUI window, the rendering window remains in
- the upper-right corner, resulting in a black window.
- Bench (RGB mode):
- - turning lights off makes the scene go black
- - bilinear filtering is always turned on, can't select point samples
-
- Files:
- =====
- Batch.txt Contains commands for "benchd3d_bat" utility
- Bench.c The WTK Direct source code we used to create the tool
- gui.exe The GUI version of the benchmark program
- guivirge.exe The GUI version of the benchmark program for S3 Virge
- batch.exe The batch version of the benchmark program
- bat3dfx.exe The batch version of the benchmark program for 3Dfx
- Dome.tga The bitmap that is mapped onto the sky dome
- final.wrl The VRML1 file that contains the Martian scene
- Fly.pth The WTK path that controls flight over the terrain.
- Mfc30.dll Required for bench
- Msvcrt40.dll Required for bench
- Readme.txt this file
- RunBatch.bat A batch file showing how to run the batch version of
- the benchmark utility.
- RunSW.bat A batch file showing how to force the renderer to use
- the software (unaccelerated) even if D3D HW is used.
- RunSWRGB.bat A batch file showing how to force the renderer to use
- the software (unaccelerated) RGB mode (the default SW
- mode is RAMP mode).
- S8.tga The bitmap used for the Ripple test.
- Terrain.tga The bitmap used for the terrain.
-
- Install:
- =======
- - make sure your system is set to 16-bit color
- - run the batch files
-
- Other issues you should be aware of:
- ===================================
- - Display depth must be 16bit (65536 colors) or greater,
- however anything greater than 16bit does not improve quality
- (the RAMP render uses a finite number of possible colors),
- but does greatly decrease performance.
-
- - Because the Direct3D Immediate Mode nor the drivers support
- hardware texture memory management (unlike OpenGL), and
- WTKDirect does not attempt to do deal with this driver and hardware
- dependent issue, when running with hardware acceleration,
- you must have enough video memory available to hold all the
- textures the bench program attempts to load up at initialization.
- This seems to require setting display resolution to 800x600, or
- less on a 4MB card, and probably eliminates running this application
- on a 2MB card.
-
- - The Vertex Colors option should be turned off, since WTKDirect
- (more specifically the D3D RAMP software renderer) does not
- support vertex colors.
-
- - define the environment variable WTKSWRENDERERONLY to force
- the use of the software renderer. Otherwise, WTKDirect will attempt
- to use hardware acceleration if available. So, to force software
- rendering, put this line into your autoexec.bat,
- "set WTKSWRENDERERONLY=TRUE". Remove this line to allow hardware
- acceleration.
-
- - define the environment variable WTKUSESWRGB to force the use of the
- software RGB renderer. This is extremely slow, however does allow the
- use of bilinear filtering in software. Also, if you are running on an
- MMX platform, use this environment variable with the MMX capable
- D3D RGB drivers to get MMX support. "set WTKUSESWRGB=TRUE"
-
- - If the RAMP mode software renderer is being used (default software renderer),
- then the only texture filtering mode available is point (nearest neighbor).
- The benchmark program will let you change the filter settings,
- however, this will have no effect when running with the software
- RAMP renderer.
-
- - We have noticed some problems with various hardware drivers
- up to this point. Strange effects include failure to properly
- light textured polys, bizarre bilinear filtering, loss of zbuffering,
- and failure to deal with unlit scenes.
-
- - The executable benchd3dfx_batch.exe is for use with hardware using the
- 3Dfx Obsidian chipset. We have not seen a board which uses this chipset
- and supports GDI, therefore you will either need a second monitor or a
- vga pass-thru cable in order to view the rendering.
-
-
- BENCH.C: Sense8's Performance Analysis Tool
- ===========================================
- v1.1
- 9/12/96
-
- Welcome to Sense8's benchmarking program. This tool allows you to precisely
- measure many facets of performance. Key features:
-
- - Measure pixel fill rates for many types of polygons, "Overwrites"
- - Measure triangle/quad polygon rates for many types of polygons, "NxN
- Grid"
- - Measure a typical application, a fly-through of a 11k poly database,
- "Terrain"
- - Measure a CPU-intensive graphics application, "Ripple"
- - Each test can be extensively controlled to configure a particular
- measurement.
- - Bench.c can be compiled two different ways. One way is the GUI mode
- (this ships with the product) the other mode is a batch mode
- where all the tests can be predefined and you can control the window
- size by passing it in as a command-line argument. To use the batch
- mode, you will need to recompile bench.c with the "#define GUI" commented out.
-
- A description of the menu commands.
-
- Test Menu
- ----------
- Turn Prebuld On(Off)- Prebuilds all the geometry in the current scene. (Or
- deletes the previously prebuilt geometries.) When prebuilt is turned
- ON, you will not be able to see subsequent changes in the
- rendering modes, other than wireframe, until you turn it OFF. You must
- use prebuild on the triangle/quad/terrain tests to get the best
- performance. Prebuilding is a special WTK-optimization that creates
- triangle and quad strips from any arbitrary geometry.
- [Note: not supported on WTK Direct]
-
- Face Type- Choose between a planar geometry divided into triangles or quads.
- Choosing triangles will give you the highest "polygons/sec" because
- more triangles can be rendered per second than quads (due to the
- additional vertex a quad has).
-
- Measure (Stop measure)- Begins (Aborts) a measurement for the current model.
- Results are output to the DOS shell.
-
- Exit- Quit to the shell.
-
- Render
- -------
- Enable (Disable) Wireframe- Toggles the model between wireframe and solid
- viewing. The other rendering modes described below will not take
- effect unless wireframe is disabled. This toggles the
- WTRENDER_WIREFRAME rendering mode. Note: wireframe is rarely the
- fastest mode with either OpenGL or Direct3D.
-
- Turn Lighting On(Off)- Toggles the WTRENDER_LIGHTING rendering mode. This
- Turns off all lighting of the scene (fastest).
-
- Turn Smooth Shading On(Off)- Toggles the WTRENDER_SMOOTH rendering mode. This
- ignores all vertex normals. Shading resorts to flat-shading (fastest).
-
- Turn Texturing On(Off)- Toggles the WTRENDER_TEXTURED rendering mode. This is
- the major control for turning textures on/off. Textures will display
- with the appropriate attributes set by other modes.
-
- Turn Perspective Texturing On(Off)- Toggles the WTRENDER_PERSPECTIVE rendering
- mode. This turns forces all textures to be rendered with/without
- perspective correction. Note: 90% of hardware accelerators (with
- texture acceleration) draw only perspective correct textures.
- Toggling this will have no effect in these cases.
-
- Vertex
- -------
- Turn Normals On(Off)- Toggles the scene between vertices with normals and
- without. In this case the geometry is actually altered to remove/add
- vertex normals - though the effect is the same as turning smooth
- shading on/off.
-
- Turn Colors On(Off)- Toggles the scene between vertices with color and without.
- [Note: not supported on WTK Direct]
-
- Texture
- --------
- Point- Sets the texture filter to Point Sampling
- (Parameters: WTFILTER_NEAREST,WTFILTER_NEAREST).
-
- Bilinear- Sets the texture filter to Bilinear mode
- (Params: WTFILTER_LINEAR,WTFILTER_LINEAR).
- [Note: not supported on WTK Direct software RAMP mode, supported on
- WTK Direct software RGB and hardware modes ]
-
- Mipmap- Sets the texture file to Mipmapping mode
- (Params: WTFILTER_LINEAR,WTFILTER_LINEARMIPMAPLINEAR).
- [Note: not supported on WTK Direct]
-
- Model
- ------
- These are all the benchmarking tests that can be run.
-
- Polygon rate tests:
- X by X Grid- Builds a geometry that has X by X quads or twice as many triangles.
- Depending on the face type. This series of tests is used to examine
- polygon per second performance. We have many of these tests which allow
- various polygon size measurements. If you want to know the number of
- 50 pixel triangles, you could size your window to 800x800 and choose the
- 80x80 test. This should give you 50 pixel triangles or 100 pixel quads.
-
- This test is also affected by the multiplier you select. For example,
- you can select a 2X or 4X multiplier. This cause twice or four times
- as many polygons to be drawn on top of each other. This is sometimes
- necessary on high performance hardware to get the frame-rates reduced
- so that measurements are accurate (<10Hz to avoid quantization effects).
-
- Note: for best performance you must use the prebuild option. If you
- do use this option, then subsequent rendering changes will NOT be seen
- until you turn prebuild off.
-
- This measures the following results, where "[]" denotes options you
- can set independently:
- OUTPUT: Z-buffered, double-buffered, [lit], [Gouraud-shaded],
- [point-sampled/bilinear-filtered/mipmapped], [perspective correct],
- [N-sized], [M-color bits], [triangles/quads], [independent/strip]
- per second.
-
- Multiplier:
- 1X-4X This is a multiplier that is used only for the polygon rate tests.
- You can use this to draw X times as many polygons in order to reduce
- the overall frame-rate (which can be a problem on high-end systems).
- If you use this, polygons are drawn directly on top of each other
- which means they fail the Z-test and a pixel isn't actually drawn.
- However, 90% of the computations occurred to get to this point.
-
- Fill rate tests:
- X Overwrites Front->Back(Front<-Back)- Stacks X number of polygons in the
- viewing plane. This is a measurement of pixel fill rates. Prebuild
- will have no effect on this test. We provide two different modes
- of this; Front->Back draws the frontmost polygon first, Front<-Back
- draws just the opposite. This is necessary because some hardware
- has a preference for how this is done and large speed differences
- are possible if it isn't done correctly.
-
- This measures the following results, where "[]" denotes options you
- can set independently:
- OUTPUT: Z-buffered, double-buffered, [lit], [Gouraud-shaded],
- [point-sampled/bilinear-filtered/mipmapped], [perspective correct],
- [M-color bits], Megapixels per second.
-
- Terrain- This model is used to test how fast a scene fly-through will run on
- your system. This measurement is only a relative measurement to be
- compared against other systems. Prebuilding will have a large effect
- on this measurement because it comprises over 11k polygons, and
- about 6k are drawn per frame on average. It requires about 1.4 Mb of
- texture memory.
-
- This measures the following results, where "[]" denotes options you
- can set independently:
- OUTPUT: Z-buffered, double-buffered, [lit], [Gouraud-shaded],
- [point-sampled/bilinear-filtered/mipmapped], [perspective correct],
- [M-color bits], [independent/strip] Terrain frames per second.
-
- Ripple- Ripple is used to test the CPU power of your machine, and the effect
- of lots of vertex manipulations. This measurement is only a relative
- measurement to be compared against other systems. Prebuilding is
- disabled for this measurement.
-
- This measures the following results, where "[]" denotes options you
- can set independently:
- OUTPUT: Z-buffered, double-buffered, [lit], [Gouraud-shaded],
- [point-sampled/bilinear-filtered/mipmapped], [perspective correct],
- [M-color bits], Ripple frames per second.
-
-
- Batch File Description
- -----------------------
- If you use the batch version of the program, you can perform hands-off
- testing in an automated fashion and redirect the output to a file.
-
- bench_batch.exe [Window width height] -b[batchfile name]
-
- Example: bench_batch.exe 640 480 -bbatch.txt > 3DLabPermedia.txt
-
- We have supplied an example batch.txt file with a canned set of typical
- measurements (batch.txt).
-
- The following are menu items and their batch file equivalents.
-
- Menu Batch Command
- ----- --------------
- Test/Prebuld TP On/Off
- Test/FaceType TF Tri/Quad
- Test/Measure C
-
- Render/Wireframe RW On/Off
- Render/Lighting RL On/Off
- Render/Smooth Shading RS On/Off
- Render/Texturing RT On/Off
- Render/Perspective RP On/Off
-
- Vertex/Normals VN On/Off
- Vertex/Colors VC On/Off
-
- Texture/Point XP
- Texture/Bilinear XB
- Texture/Mipmap XM
-
- Model/Grid MG # (Any positive multiple of ten)
- Model/Multiplier MM # (Any positive integer)
- Model/Overwrites/Front->Back MO FB # (Any positive integer)
- Model/Overwrites/Front<-Back MO BF # (Any positive integer)
- Model/Terrain MT
- Model/Ripple MR
-
- Comment ; Comment goes here
-