|In reply to Comment 115 (Don Cox):|
Don, that is not true. Running binaries through an abstract, intermediate state yields a penalty, even if it may be small.
It is true that the VP code is translated at load time. But the translated code cannot recieve the same amount of optimization as dedicated target-compilers can. Or are you trying to tell me that Tao's VP translator matches e.g. the Intel C++ compiler? Hardly.
What's more, for all that we know the Tao VP cannot utilize SIMD units like Altivec or MMX/SSE. You need optimized modules for that - there goes the write once.