I think there may be some misunderstanding here.
Nobody thinks that Intel should go out of their way to help AMD, such as by making optimizations specific for AMD processors.
Rather, Intel licensed out an instruction set called x86, and there are standardized extensions to it that must be implemented exactly the same as Intel did or not at all.
All other compilers on Earth check if these extensions (such as SSE, AVX, AES-NI, etc.) are present from the CPUID flags, and Intel is also checking the CPUID flags, but they are checking for what brand of CPU it is rather than which extensions it supports.
If a compiler has a way to use AVX to run 16 computations at once (per thread) using AVX, for example, it does not need to make an AMD version.
It just needs to check if AVX is supported by the current CPU (not all Intel CPU's, nor all AMD CPU's, have this extension) and make use of that if available (in code that can be vectorized, the throughput increases 1600%, whereas micro-optimization is generally a gain far less than 10% or so.)
The brand name is not in any way related to which extensions the currently running CPU supports. Intel's compiler is still going out of its way to cripple competitors. This has nothing to do with going out of their way to accomodate AMD or not.
Imagine if when you tried to set a default browser, Microsoft made it so Windows would not let you unless the browser was from Microsoft, even if all browsers were standardized and it was easier to just check if the browser was supported on your computer.
Microsoft did this and got taken to court for it. They lost billions of dollars. Just for preventing competing browsers from bundling with Windows!
What Intel is doing here is far worse. No project run by volunteers with limited time, nor any commercial project with budgetary/tinme constrants, can afford to port everything they own to another compiler. Remember, many people start using Intel's compiler without any warning of this anticompetitive behaviour.
And no, unless you are talking about a personal pet/toy project, it is not trivial to port everything to a new compiler.
There are intrinsics (heavily promoted by Intel) which must all be rewritten.
There are cross-organization interactions of code that require everyone to use the same compiler for the duration of a project.
There are all sorts of legacy issues where you may end up with some binaries/libraries that you no longer have the source code to (or never had the source to,) and they may not have been certified to be used with code generated from other compilers. All throughout industry the costs of this could be infeasible, even where it is possible at all (which is mostly small projects.)