[clang] [llvm] [RISCV][clang] Support XSfmm ABI attributes (PR #206260)

[Brandon Wu via cfe-commits]

================
@@ -8041,6 +8046,51 @@ static bool handleArmStateAttribute(Sema &S,
   return false;
 }
 
+static bool handleRISCVStateAttribute(Sema &S,
+                                      FunctionProtoType::ExtProtoInfo &EPI,
+                                      ParsedAttr &Attr,
+                                      FunctionType::RISCVStateValue State) {
+  if (!Attr.getNumArgs()) {
+    S.Diag(Attr.getLoc(), diag::err_missing_riscv_state) << Attr;
+    Attr.setInvalid();
+    return true;
+  }
+
+  for (unsigned I = 0; I < Attr.getNumArgs(); ++I) {
+    StringRef StateName;
+    SourceLocation LiteralLoc;
+    if (!S.checkStringLiteralArgumentAttr(Attr, I, StateName, &LiteralLoc))
+      return true;
+
+    unsigned Shift;
+    FunctionType::RISCVStateValue ExistingState;
+
+    // Determine which tile state this is and get its shift/mask
+    if (StateName == "xsfmm") {
+      Shift = FunctionType::RISCVXsfmmShift;
+      ExistingState = FunctionType::getRISCVXsfmmState(EPI.RISCVAttributes);
+    } else {
+      S.Diag(LiteralLoc, diag::err_unknown_riscv_state) << StateName;
+      Attr.setInvalid();
+      return true;
+    }
+
+    // __riscv_in, __riscv_out, __riscv_inout, __riscv_preserves, and
+    // __riscv_new are all mutually exclusive for the same state,
+    // so check if there are conflicting attributes.
----------------
4vtomat wrote:

Umm I think that might be confused or ambiguous to user if we try to make any call with `riscv_in` in `riscv_out` or other way around? Or is there any use case for this lol?


https://github.com/llvm/llvm-project/pull/206260