These are tarballs of native ARM EABI compilers made from standard GCC sources patched with the enclosed patches. They run on and generate code for armv4t architecture, and the flags -mcpu=ep9312 -mfpu=maverick -mfloat-abi=softfp make it emit Maverick code. Furthermore, -mieee generates code that works to full IEEE precision by disabling the Maverick instructions that drop tiny denormalised values. Floating point code runs at about half of the full speed, but still about 25% faster than softfloat. -fno-signed-zeros enables the "cfneg" instructions, which fail to convert 0 into -0. This gives a slight speed up (-mfast-math enables this among others). -mcirrus-di re-enables the buggy 64-bit integer instructions in the FPU. Left shifts by more that 31 places and right shift by more than 32 give wrong results and some other instruction sequences are buggy. The flag is included to aid debugging and for the fast and furious. They install everything under /usr/local and should not affect the main system compilers. Martin Guy , 20 March 2009