AMDGPU: mark {exp,log}10{,f,l} library functions as unavailable

[nhaehnle]

Summary

The SimplifyLibCalls part of InstCombine generates calls to those otherwise.

I wonder if at some point we shouldn't just call disableAllFunctions() and
then enable functions on a whitelist basis...

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=96495

[• tstellarAMD]

The whitelist idea sounds good, what is the advantage of a LibFunc being 'available' ?

[nhaehnle]

I think it's mostly used by SimplifyLibCalls and perhaps other parts of InstCombine to decide whether certain transformations are allowed.

Interestingly, NVPTX sets disableAllFunctions, but there is a comment saying that "having no standard library prevents e.g. many fastmath optimizations". And indeed, on my shader-db using disableAllFunctions() makes a difference. It might be limited to sqrt and abs functions.

I'll commit this change as-is for now, i.e. stick with blacklisting.

lib/Analysis/TargetLibraryInfo.cpp

assert(std::is_sorted(StandardNames.begin(), StandardNames.end(),
                         }) &&
          "TargetLibraryInfoImpl function names must be sorted");
 
   if (T.getArch() == Triple::r600 ||
       T.getArch() == Triple::amdgcn) {
     TLI.setUnavailable(LibFunc::ldexp);
     TLI.setUnavailable(LibFunc::ldexpf);
     TLI.setUnavailable(LibFunc::ldexpl);
+    TLI.setUnavailable(LibFunc::exp10);
+    TLI.setUnavailable(LibFunc::exp10f);
+    TLI.setUnavailable(LibFunc::exp10l);
+    TLI.setUnavailable(LibFunc::log10);
+    TLI.setUnavailable(LibFunc::log10f);
+    TLI.setUnavailable(LibFunc::log10l);
   }
 
   // There are no library implementations of mempcy and memset for AMD gpus and
   // these can be difficult to lower in the backend.
   if (T.getArch() == Triple::r600 ||
       T.getArch() == Triple::amdgcn) {
     TLI.setUnavailable(LibFunc::memcpy);
     TLI.setUnavailable(LibFunc::memset);