Merge pull request intel#291 from hvdijk/assume-fma

Assume FMA support, bump x86-64 to x86-64-v3.

Author: hvdijk

llvm/lib/SYCLNativeCPUUtils/compiler_passes/compiler_pipeline/docs/utils.rst

@@ -1026,8 +1026,6 @@ The following builtins are replaced:
 * ``__mux_usefast`` - Whether to use faster, less accurate maths algorithms.
 * ``__mux_isembeddedprofile`` - Whether the mux target implements OpenCL
   embedded profile.
-* ``__mux_hasnativefma`` - Whether the mux target supports a fused multiply-add
-  instruction.
 
 Declarations matching each of these function names are searched for by
 ``ReplaceMuxMathDeclsPass``, and if found, a function body is created returning