Mesa 6.5.3 Release Notes / April 27, 2007

Mesa 6.5.3 is a development release with many changes and new features. Mesa 7.0 is expected to follow shortly.

MD5 checksums

39f33ea64e34e2d5b20640b008b57649  MesaLib-6.5.3.tar.gz
46359457147c469745f24b5074a186f0  MesaLib-6.5.3.tar.bz2
a8946fa861634ce15971396f47992c41  MesaLib-6.5.3.zip
08e26948d57eaca74d02a530b2d8106e  MesaDemos-6.5.3.tar.gz
8af91773ab2653fe537499676b05f2e8  MesaDemos-6.5.3.tar.bz2
783f81b171bf89b0929abc894efd25a6  MesaDemos-6.5.3.zip
9467d415388fe1ad82991fb20704b812  MesaGLUT-6.5.3.tar.gz
360843e46b7ebb6909290b023f9b26fa  MesaGLUT-6.5.3.tar.bz2
7686065e5c15a30de08a1610860b6840  MesaGLUT-6.5.3.zip

Shared library numbering

Mesa 6.5.3 supports the OpenGL 2.0/2.1 API. However, the (unix) shared library version is still 1.5 (i.e. libGL.so.1.5.xxxxxx). Bumping the shared library version to 2.x would cause linking problems with existing OpenGL applications. Since OpenGL 2.x is backward compatible with OpenGL 1.x the shared library version number doesn’t have to be incremented (which would indicate an incompatible ABI).

Other OpenGL vendors name their OpenGL 2.x libraries libGL.so.1.0.xxxxx for the same reason.

New features

  • OpenGL 2.0 and 2.1 API support.

  • Entirely new Shading Language code generator. See the Shading Language page for more information.

  • Much faster software execution of vertex, fragment shaders.

  • New vertex buffer object (vbo) infrastructure

  • Updated glext.h file (version 39)

  • Updated glxext.h file (version 19)

  • GL_MAX_DRAWBUFFERS is now 4 (software rendering) so “multiple render targets” are really supported.

Bug fixes

  • Fog was errantly applied when a fragment shader was enabled (bug 9346)

  • glPush/PopClientAttrib didn’t handle VBO bindings correctly (bug 9445)

  • With 32-bit Z buffer, the fragment Z of lines and points was sometimes wrong.

  • GL_POST_CONVOLUTION_ALPHA_BIAS/SCALE was broken.

  • 1D convolution state could effect 2D image transfers

  • Overlapping glCopyPixels with negative Y zoom didn’t work (bug 10521)

  • Fixed a number of framebuffer/renderbuffer reference counting bugs

  • Fixed a few bugs in software-emulated alpha planes

  • Assorted minor bug fixes in glCopy/DrawPixels, glPixelZoom, etc.

  • Assorted DRI driver bug fixes.

  • Fixed a number of bugs that prevented “depth-peeling” rendering from working.

Internal code changes

  • Old array_cache module replaced by new vbo module. All geometry rendering is now cast in the form of vertex buffer objects.

  • Massive changes to the Shading Language compiler and related state.

  • Vertex/fragment shaders are compiled into GPU instructions and programs very similar to GL_ARB_vertex/fragment_program.

  • Vertex and fragment programs are executed with the same code now.

  • The SSE-optimized vertex program path has been removed since it didn’t support more than 12 temp registers, didn’t support branching/looping, etc.

To Do (someday) items

  • Switch to freeglut

  • Fix linux-glide target/driver.

  • Improved lambda and derivative calculation for frag progs.

Driver Status

Driver         Status
----------------------  ----------------------
DRI drivers     varies with the driver
XMesa/GLX (on Xlib) implements OpenGL 2.1
OSMesa (off-screen) implements OpenGL 2.1
Windows/Win32       implements OpenGL 2.1
Glide (3dfx Voodoo1/2)  implements OpenGL 1.3
SVGA            unsupported
Wind River UGL      unsupported
DJGPP           unsupported
GGI         unsupported
BeOS            unsupported
Allegro         unsupported
D3D         unsupported