Windows x86: Change i128 to return via the vector ABI

Clang and GCC both return `i128` in xmm0 on windows-msvc and
windows-gnu. Currently, Rust returns the type on the stack. Add a
calling convention adjustment so we also return scalar `i128`s using the
vector ABI, which makes our `i128` compatible with C.

In the future, Clang may change to return `i128` on the stack for its
`-msvc` targets (more at [1]). If this happens, the change here will
need to be adjusted to only affect MinGW.

Link: rust-lang/rust#134288

Author: tgross35

src/abi/mod.rs

@@ -122,7 +122,7 @@ impl<'tcx> FunctionCx<'_, '_, 'tcx> {
         &mut self,
         name: &str,
         params: Vec<AbiParam>,
-        returns: Vec<AbiParam>,
+        mut returns: Vec<AbiParam>,
         args: &[Value],
     ) -> Cow<'_, [Value]> {
         // Pass i128 arguments by-ref on Windows.
@@ -146,15 +146,19 @@ impl<'tcx> FunctionCx<'_, '_, 'tcx> {
             (params, args.into())
         };
 
-        // Return i128 using a return area pointer on Windows and s390x.
-        let adjust_ret_param =
-            if self.tcx.sess.target.is_like_windows || self.tcx.sess.target.arch == "s390x" {
-                returns.len() == 1 && returns[0].value_type == types::I128
-            } else {
-                false
-            };
+        let ret_single_i128 = returns.len() == 1 && returns[0].value_type == types::I128;
+        if ret_single_i128 && self.tcx.sess.target.is_like_windows {
+            // Return i128 using the vector ABI on Windows
+            returns[0].value_type = types::I64X2;
+
+            let ret = self.lib_call_unadjusted(name, params, returns, &args)[0];
 
-        if adjust_ret_param {
+            // FIXME(bytecodealliance/wasmtime#6104) use bitcast instead of store to get from i64x2 to i128
+            let ret_ptr = self.create_stack_slot(16, 16);
+            ret_ptr.store(self, ret, MemFlags::trusted());
+            Cow::Owned(vec![ret_ptr.load(self, types::I128, MemFlags::trusted())])
+        } else if ret_single_i128 && self.tcx.sess.target.arch == "s390x" {
+            // Return i128 using a return area pointer on s390x.
             let mut params = params;
             let mut args = args.to_vec();
 

src/cast.rs

@@ -96,25 +96,9 @@ pub(crate) fn clif_int_or_float_cast(
                 },
             );
 
-            if fx.tcx.sess.target.is_like_windows {
-                let ret = fx.lib_call(
-                    &name,
-                    vec![AbiParam::new(from_ty)],
-                    vec![AbiParam::new(types::I64X2)],
-                    &[from],
-                )[0];
-                // FIXME(bytecodealliance/wasmtime#6104) use bitcast instead of store to get from i64x2 to i128
-                let ret_ptr = fx.create_stack_slot(16, 16);
-                ret_ptr.store(fx, ret, MemFlags::trusted());
-                ret_ptr.load(fx, types::I128, MemFlags::trusted())
-            } else {
-                fx.lib_call(
-                    &name,
-                    vec![AbiParam::new(from_ty)],
-                    vec![AbiParam::new(types::I128)],
-                    &[from],
-                )[0]
-            }
+            fx.lib_call(&name, vec![AbiParam::new(from_ty)], vec![AbiParam::new(types::I128)], &[
+                from,
+            ])[0]
         } else if to_ty == types::I8 || to_ty == types::I16 {
             // FIXME implement fcvt_to_*int_sat.i8/i16
             let val = if to_signed {

src/codegen_i128.rs

@@ -33,28 +33,14 @@ pub(crate) fn maybe_codegen<'tcx>(
                 (BinOp::Rem, true) => "__modti3",
                 _ => unreachable!(),
             };
-            if fx.tcx.sess.target.is_like_windows {
-                let args = [lhs.load_scalar(fx), rhs.load_scalar(fx)];
-                let ret = fx.lib_call(
-                    name,
-                    vec![AbiParam::new(types::I128), AbiParam::new(types::I128)],
-                    vec![AbiParam::new(types::I64X2)],
-                    &args,
-                )[0];
-                // FIXME(bytecodealliance/wasmtime#6104) use bitcast instead of store to get from i64x2 to i128
-                let ret_place = CPlace::new_stack_slot(fx, lhs.layout());
-                ret_place.to_ptr().store(fx, ret, MemFlags::trusted());
-                Some(ret_place.to_cvalue(fx))
-            } else {
-                let args = [lhs.load_scalar(fx), rhs.load_scalar(fx)];
-                let ret_val = fx.lib_call(
-                    name,
-                    vec![AbiParam::new(types::I128), AbiParam::new(types::I128)],
-                    vec![AbiParam::new(types::I128)],
-                    &args,
-                )[0];
-                Some(CValue::by_val(ret_val, lhs.layout()))
-            }
+            let args = [lhs.load_scalar(fx), rhs.load_scalar(fx)];
+            let ret_val = fx.lib_call(
+                name,
+                vec![AbiParam::new(types::I128), AbiParam::new(types::I128)],
+                vec![AbiParam::new(types::I128)],
+                &args,
+            )[0];
+            Some(CValue::by_val(ret_val, lhs.layout()))
         }
         BinOp::Lt | BinOp::Le | BinOp::Eq | BinOp::Ge | BinOp::Gt | BinOp::Ne | BinOp::Cmp => None,
         BinOp::Shl | BinOp::ShlUnchecked | BinOp::Shr | BinOp::ShrUnchecked => None,