In general, nothing special has to be done to use the DRI on different CPU architectures. There are, however, a few optimizations that are CPU-dependent. Mesa will determine at runtime which CPU-dependent optimizations should be used and enable them where appropriate.
The Pentium III SSE instructions are used in optimized vertex transformation functions in the Mesa-based DRI drivers. On Linux, SSE requires a recent kernel (such as 2.4.0-test11 or later) both at compile time and runtime.
AMD's 3DNow! instructions are used in optimized vertex transformation functions in the Mesa-based DRI drivers. 3DNow! is supported in most versions of Linux.
On newer Alpha processors a significant performance increase can be seen with the addition of the -mcpu= option to GCC. This option is dependent on the architecture of the processor. For example, -mcpu=ev6 will build specifically for the EV6 based AXP's, giving both byte and word alignment access to the DRI/Mesa drivers.
To enable this optimization edit your xc/config/host.def file and add the line:
#define DefaultGcc2AxpOpt -O2 -mcpu=ev6
Additional speed improvements to 3D rendering can be achieved by installing Compaq's Math Libraries (CPML) which can be obtained from http://www.support.compaq.com/alpha-tools/software/index.html
Once installed, you can add this line to your host.def to build with the CPML libraries:
#define UseCompaqMathLibrary YES
The host.def file is explained below.