Fix the prefix for arm64 triple

[kongyi]

Summary

Triple.cpp still returns "arm64" as prefix for arm64 triple, causing Clang not being able to select the correct GCCBuiltin IR.

This patch changes the value to correct prefix "aarch64". Regression test will be added in the coming patch.

[rengolin]

Ouch, that looks like an oversight. Tim?

[t.p.northover]

It looks like it's only used by that part of Clang. We probably want a test though (even if it's in Clang rather than LLVM), since this is clearly causing problems for someone.

Also, are you planning to do the more thorough triple merge, Yi?

Cheers.

Tim.

[kongyi]

In D4516#8, @t.p.northover wrote:

It looks like it's only used by that part of Clang. We probably want a test though (even if it's in Clang rather than LLVM), since this is clearly causing problems for someone.

I have written a test case in Clang, however due to the limitation of Phabricator, I can't post it in the same post. I will take another look to see whether it is possible to test in LLVM.

Also, are you planning to do the more thorough triple merge, Yi?

Although the merge should be trivial, but I think it would be best for you guys to do it to avoid breaking your internal Apple clients.

[kongyi]

In D4516#9, @kongyi wrote:

In D4516#8, @t.p.northover wrote:

It looks like it's only used by that part of Clang. We probably want a test though (even if it's in Clang rather than LLVM), since this is clearly causing problems for someone.

I have written a test case in Clang, however due to the limitation of Phabricator, I can't post it in the same post.

Actually, it's part of D4520. The test makes sure that GCCBuiltin works when the triple supplied is ARM64, indicating LLVM returned the correct prefix.

I will take another look to see whether it is possible to test in LLVM.

I can't find a way to do that. It seems we can only test this indirectly from Clang.

[t.p.northover]

I have written a test case in Clang, however due to the limitation of Phabricator, I can't post it in the same post. I will take another look to see whether it is possible to test in LLVM.

No need to do that. I suspect it's very strongly linked to the C++
front-end so I'd be quite happy with a Clang test, but we should have
one.

Cheers.

Tim.

[kongyi]

In D4516#11, @t.p.northover wrote:

No need to do that. I suspect it's very strongly linked to the C++
front-end so I'd be quite happy with a Clang test, but we should have
one.

My plan to implement this is by D4521, where GCCBuiltin __builtin_arm_[dmb, dsb, isb] are tested to be correctly selected by Clang when arm64 triple is supplied. It's an indirect test, but that's the only test I can think of.

Do you have any suggestion? Otherwise ok to commit?

[t.p.northover]

Hi Yi,

My plan to implement this is by D4521, where GCCBuiltin __builtin_arm_[dmb, dsb, isb] are tested to be correctly selected by Clang when arm64 triple is supplied.

OK, I see. There aren't actually any AArch64 builtins yet so it can't be tested. That sounds reasonable. Go for it!

Tim.

[kongyi]

Closed by commit rL213240 (authored by @kongyi).

llvm/trunk/lib/Support/Triple.cpp

const char *Triple::getArchTypeName(ArchType Kind) {
   llvm_unreachable("Invalid ArchType!");
 }
 
 const char *Triple::getArchTypePrefix(ArchType Kind) {
   switch (Kind) {
   default:
     return nullptr;
 
+  case arm64:
+  case arm64_be:
   case aarch64:
   case aarch64_be:  return "aarch64";
 
   case arm:
   case armeb:
   case thumb:
   case thumbeb:     return "arm";
 
-  case arm64:       
-  case arm64_be:    return "arm64";
-
   case ppc64:
   case ppc64le:
   case ppc:         return "ppc";
 
   case mips:
   case mipsel:
   case mips64:
   case mips64el:    return "mips";