[clang] Mark __builtin_convertvector and __builtin_shufflevector as constexpr.

[c8ef]

Closes #107985.

LanguageExtensions.rst states that __builtin_shufflevector and __builtin_convertvector can be evaluated as constants, but this is not reflected in Butiltins.td. This patch aligns these two.

[llvmbot]

@llvm/pr-subscribers-clang

Author: None (c8ef)

Changes

Closes #107985.

---

Full diff: https://github.com/llvm/llvm-project/pull/112129.diff

1 Files Affected:

• (modified) clang/include/clang/Basic/Builtins.td (+2-2)

diff --git a/clang/include/clang/Basic/Builtins.td b/clang/include/clang/Basic/Builtins.td
index 7068473a0e12ac..362de6cfa9d0e1 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -1186,13 +1186,13 @@ def AllowRuntimeCheck : Builtin {
 
 def ShuffleVector : Builtin {
   let Spellings = ["__builtin_shufflevector"];
-  let Attributes = [NoThrow, Const, CustomTypeChecking];
+  let Attributes = [NoThrow, Const, CustomTypeChecking, Constexpr];
   let Prototype = "void(...)";
 }
 
 def ConvertVector : Builtin {
   let Spellings = ["__builtin_convertvector"];
-  let Attributes = [NoThrow, Const, CustomTypeChecking];
+  let Attributes = [NoThrow, Const, CustomTypeChecking, Constexpr];
   let Prototype = "void(...)";
 }
 

[RKSimon]

How are we already able to use these in constexpr if they are missing the attribute?

[c8ef]

I believe the difference lies in the AST construction. Most built-in functions generate a CallExpr AST node, which requires the constant evaluator to check if the function is built-in. However, in this case, Clang emits a ConvertVector node directly. Therefore, even if we don't mark it as constexpr, we can still evaluate it.

[c8ef]

And I think we need to align these two to make the feature test macro work.

[RKSimon]

Anyway you can test that?

[c8ef]

You can directly view it from the Clang AST on Godbolt. For this patch, I don't believe additional tests are necessary.

[philnik777]

You can directly view it from the Clang AST on Godbolt. For this patch, I don't believe additional tests are necessary.

Any bugfix should contain a regression test. Here you could simply check that __has_constexpr_builtin is true for these builtins, since that is the most obvious behaviour change.