Auto merge of #103098 - Amanieu:asm-tied-fixed, r=bjorn3

asm: Match clang behavior for inlateout fixed register operands

We have 2 options for representing LLVM constraints for `inlateout` operands on a fixed register (e.g. `r0`): `={r0},0` or `={r0},{r0}`.

This PR changes the behavior to the latter, which matches the behavior of Clang since https://reviews.llvm.org/D87279.

compiler/rustc_codegen_llvm/src/asm.rs

@@ -130,15 +130,24 @@ impl<'ll, 'tcx> AsmBuilderMethods<'tcx> for Builder<'_, 'll, 'tcx> {
                     op_idx.insert(idx, constraints.len());
                     constraints.push(reg_to_llvm(reg, Some(&value.layout)));
                 }
-                InlineAsmOperandRef::InOut { reg, late: _, in_value, out_place: _ } => {
+                InlineAsmOperandRef::InOut { reg, late, in_value, out_place: _ } => {
                     let value = llvm_fixup_input(
                         self,
                         in_value.immediate(),
                         reg.reg_class(),
                         &in_value.layout,
                     );
                     inputs.push(value);
-                    constraints.push(format!("{}", op_idx[&idx]));
+
+                    // In the case of fixed registers, we have the choice of
+                    // either using a tied operand or duplicating the constraint.
+                    // We prefer the latter because it matches the behavior of
+                    // Clang.
+                    if late && matches!(reg, InlineAsmRegOrRegClass::Reg(_)) {
+                        constraints.push(format!("{}", reg_to_llvm(reg, Some(&in_value.layout))));
+                    } else {
+                        constraints.push(format!("{}", op_idx[&idx]));
+                    }
                 }
                 InlineAsmOperandRef::SymFn { instance } => {
                     inputs.push(self.cx.get_fn(instance));