Mesa 5.1 release notes
December 17, 2003
PLEASE READ!!!!
Introduction
------------
Mesa uses an even/odd version number scheme like the Linux kernel.
Even-numbered versions (such as 5.0) designate stable releases.
Odd-numbered versions (such as 5.1) designate new developmental releases.
Bug fixes
---------
See the VERSIONS file for a list of bugs fixed in this release.
New Features in Mesa 5.1
------------------------
GL_ARB_vertex_program / GL_ARB_fragment_program
Michal Krol and Karl Rasche implemented these extensions. Thanks!
Be aware that there may be some rough edges and lurking bugs.
GL_ATI_texture_env_combine3 extension
This adds a few new texture combine modes.
Contributed by Ian Romanick.
GL_SGI_texture_color_table
Adds a color table lookup to the RGBA texture path. There's a separate
color table for each texture unit.
Contributed by Eric Plante.
GL_NV_fragment_program
NVIDIA's fragment-level programming feature.
Possible lurking bugs:
- the DDX and DDY commands aren't fully tested
- there may be bugs in the parser
- the TEX and TXP instructions both do perspective correction
- the pack/unpack instructions may not be correct
GL_EXT_depth_bounds_test
This extension adds a scissor-like test for the Z axis. It's used to
optimize stencil-volume shadow algorithms.
GL_NV_light_max_exponent
Lifts the 128 limit for max light exponent.
GL_EXT_texture_rectangle
Identical to GL_NV_texture_rectangle
GL_ARB_occlusion_query
Useful for visibility-based culling.
GL_ARB_texture_non_power_of_two
Removes the restriction that texture dimensions must be powers of two.
GL_ARB_vertex_buffer_object
Allows server-side vertex arrays, optimized host/card data transfers, etc.
GL_ARB_point_sprite
ARB-approved version of GL_NV_point_sprite. Basically allows textures
to be applied to points.
GL_IBM_multimode_draw_arrays
Allows multiple vertex arrays to be drawn with one call, including arrays
of different types of primitives.
GL_SUN_multi_draw_arrays
An alias for GL_EXT_multi_draw_arrays, standard in OpenGL 1.4.
Faster glDrawPixels / glCopyPixels in X11 driver
If your X screen is 32bpp, glDrawPixels to the front color buffer will
be accelerated (via XPutImage()) if the image format is GL_BGRA and the
type is GL_UNSIGNED_BYTE. No raster operations, such as depth test,
blend, fog, etc. can be enabled.
If your X screen is 16bpp, glDrawPixels to the front color buffer will
be accelerated (via XPutImage()) if the image format is GL_RGB and the
type is GL_UNSIGNED_SHORT_5_6_5. No raster operations, such as depth
test, blend, fog, etc. can be enabled.
glCopyPixels() calls for the front color buffer will be accelerated
(via XCopyArea()) if no raster operations, such as depth test, blend,
fog, pixel zoom, etc. are enabled.
The speed-up over typical software rendering is a factor of 10 for
glDrawPixels and 100 for glCopyPixels.
With the addition of GL_ARB_occlusion_query, GL_ARB_vertex_buffer_object,
GL_ARB_texture_non_power_of_two and GL_EXT_shadow_funcs, Mesa 5.1 supports
all the new features of OpenGL 1.5. Mesa 6.0 (the next stable release)
will advertise GL_VERSION = "1.5".
Vertex/Fragment program debugger
--------------------------------
GL_MESA_program_debug is an experimental extension to support
interactive debugging of vertex and fragment programs. See the
docs/MESA_program_debug.spec file for details.
The bulk of the vertex/fragment program debugger is implemented
outside of Mesa. The GL_MESA_program_debug extension just has minimal
hooks for stopping running programs and inspecting programs.
The progs/tests/debugger.c (only in CVS) program is an example of how
the extension can be used. Presently, the debugger code and demo code
is in the same file. Eventually the debugger code should be moved
into a reusable module.
As it is now, the demo lets you set breakpoings in vertex/fragment
programs, single step, and print intermediate register values. It's
basically just a proof of concept.
Directory tree reorganization
-----------------------------
The directory structure for Mesa has been overhauled to improve its layout.
All source code for Mesa, GLU, GLUT, etc is now under the src/ directory
in appropriate subdirectories.
The Mesa source code and drivers has been reorganized under src/mesa/.
All demonstration programs and tests are now in subdirectories under progs/.
Build System Changes
--------------------
The GNU automake/autoconf support has been removed. As it was, it seldom
worked on anything but Linux. The Mesa developers aren't big fans of
automake/autoconf/libtool and didn't have the time to maintain it.
If someone wants to contribute new automake/autoconf support (and is
willing to maintain it), it may be re-incorporated into Mesa, subject
to some requirements.
The "old style" makefile system has been updated:
1. Make-config has been trimmed down to fewer, modern configurations.
2. Most of the bin/mklib.* scripts have been rolled into a new "mklib"
script that works on all sorts of systems. There are probably some
bugs in it, but it's been tested on Linux, SunOS 5.8 and IRIX 6.5.
Improvements/contributes are greatly appreciated.
3. The Makefile.X11 files have been cleaned up in various ways
Source File Changes
-------------------
The mmath.[ch] files are obsolete. Their contents have been moved
into the imports.[ch] and macros.[ch] files.
The files related to vertex and fragment programming have changed.
Old files:
vpexec.[ch]
vpparse.[ch]
vpstate.[ch]
New files:
program.[ch] - generic ARB/NV program code
arbprogram.[ch] - ARB program API functions
arbfragparse.[ch] - ARB fragment program parsing
arbvertparse.[ch] - ARB vertex program parsing
arbparse.[ch] - ARB vertex/fragment parsing
arbparse_syn.h - vertex/fragment program syntax
nvprogram.[ch] - NV program API functions
nvvertprog.h - NV vertex program definitions
nvfragprog.h - NV fragment program definitions
nvvertparse.[ch] - NV vertex program parser
nvfragparse.[ch] - NV fragment program parser
nvvertexec.[ch] - NV vertex program execution
swrast/s_nvfragprog.[ch] - NV fragment program execution
The files related to per-vertex handling have changed.
Old files:
tnl/t_eval_api.c - old per-vertex code
tnl/t_imm_alloc.c - old per-vertex code
tnl/t_imm_api.c - old per-vertex code
tnl/t_imm_debug.c - old per-vertex code
tnl/t_imm_dlist.c - old per-vertex code
tnl/t_imm_elt.c - old per-vertex code
tnl/t_imm_eval.c - old per-vertex code
tnl/t_imm_exec.c - old per-vertex code
tnl/t_imm_fixup.c - old per-vertex code
tnl/t_vtx_sse.c - old per-vertex code
tnl/t_vtx_x86.c - old per-vertex code
New files:
tnl/t_save_api.c - new per-vertex code
tnl/t_save_loopback.c - new per-vertex code
tnl/t_save_playback.c - new per-vertex code
tnl/t_vtx_eval.c - old per-vertex code
Other new files:
bufferobj.[ch] - GL_ARB_vertex_buffer_object functions
version.h - defines the Mesa version info
Other removed files:
swrast/s_histogram.[ch] - moved into src/histogram.c
Other Changes
-------------
The ctx->Driver.CreateTexture function has been removed - it wasn't used.
New device driver hook functions:
NewTextureObject - used to allocate struct gl_texture_objects
NewTextureImage - used to allocate struct gl_texture_images
New ctx->Texture._EnabledCoordUnits field:
With the addition of GL_NV_fragment_program we may need to interpolate
various sets of texture coordinates even when the corresponding texture
unit is not enabled. That is, glEnable(GL_TEXTURE_xD) may never get
called but we still may have to interpolate texture coordinates across
triangles so that the fragment program will get them.
This new field indicates which sets of texture coordinates are needed.
If a bit is set in the ctx->Texture._EnabledUnits bitmask is set, the
same bit MUST be set in ctx->Texture._EnabledCoordUnits.
The ctx->_TriangleCaps field is deprecated.
Instead of testing the DD_* bits in _TriangleCaps, you should instead
directly test the relevant state variables, or use one of the helper
functions like NEED_SECONDARY_COLOR() at the bottom of context.h
While testing _TriangleCaps bits was fast, it was kludgey, and setting
the bits in the first place could be error prone.
New vertex processing code.
The code behind glBegin, glEnd, glVertex, glNormal, etc. has been
totally rewritten. It's a cleaner implementation now and should use
less memory. (Keith)
To Do
-----
Add screen-awareness to fakeglx.c
Device Driver Status
--------------------
A number of Mesa's software drivers haven't been actively maintained for
some time. We rely on volunteers to maintain many of these drivers.
Here's the current status of all included drivers:
Driver Status
---------------------- ---------------------
XMesa (Xlib) implements OpenGL 1.4
OSMesa (off-screen) implements OpenGL 1.4
FX (3dfx Voodoo1/2) implements OpenGL 1.3
SVGA implements OpenGL 1.3
Wind River UGL implements OpenGL 1.3
Windows/Win32 implements OpenGL 1.4
DJGPP implements OpenGL 1.4
GGI implements OpenGL 1.3
BeOS implements OpenGL 1.4
Allegro needs updating
D3D needs updating
Note: supporting OpenGL 1.4 (vs. 1.3 or 1.2) usually only requires that the
driver call the _mesa_enable_1_4_extensions() function.
----------------------------------------------------------------------