From Derek Piasecki
With regard to Comment: FTC's compiler attack on Intel is thin.
The FTC's argument is not about the optimizations by compiler in the sense that computer scientists and compilers use the word. Specifically, it is not about the ordering, scheduling, and choice of instructions made at compile-time. Rather, it is about run-time choices made by the libraries that provide enhanced performance for vector operations. The purpose of these libraries is to provide an abstraction of the vector instructions (and higher-level operations that use them) so software engineers do not have to write the code for specific CPUs.
To support the fastest operation on a variety of CPUs with different instruction sets, the libraries have multiple code paths (implementations of the algorithms), one that uses basic x86 instructions and multiple others that use different sets or versions of the vector instructions (SSE, SSE2, etc.). The libraries decide at runtime which code path to use based on the processor's capabilities (the presence of support for SSE, SSE2 and other instructions). However, in addition to checking which instructions the CPU supports, the library also checks whether it is a "genuine Intel" CPU and if it is non-Intel it uses the code path that relies only on basic x86 instructions, even if the CPU supports the more advanced (faster) vector instructions. It is this deliberate and unnecessary ignoring of the capabilities of competitor's CPUs (choosing to use a slower code path when it can choose to use a faster one) that the FTC is arguing is anti-competitive.
I don't know what the FTC would say in general about libraries that refused to use fully-compatible performance features of competitors' processors, but the stated complaint was about not disclosing that fact.
Thanks to Derek Piasecki for this information.
One wonders what the arguments would have been if Intel's runtime libraries, on detecting a non-Intel CPU, simply stalled the software, which presumably would have been detected quickly. One could argue that, as it is, Intel has been doing other processor makers a favor by providing libraries that allow software to run on non-Intel processors at all!!! PDC