Return values larger than 2 registers using a return area pointer

LLVM and Cranelift disagree about how to return values that don't fit
in the registers designated for return values. LLVM will force the
entire return value to be passed by return area pointer, while
Cranelift will look at each IR level return value independently and
decide to pass it in a register or not, which would result in the
return value being passed partially in registers and partially through
a return area pointer.

While Cranelift may need to be fixed as the LLVM behavior is generally
more correct with respect to the surface language, forcing this
behavior in rustc itself makes it easier for other backends to conform
to the Rust ABI and for the C ABI rustc already handles this behavior
anyway.

In addition LLVM's decision to pass the return value in registers or
using a return area pointer depends on how exactly the return type is
lowered to an LLVM IR type. For example `Option<u128>` can be lowered
as `{ i128, i128 }` in which case the x86_64 backend would use a return
area pointer, or it could be passed as `{ i32, i128 }` in which case
the x86_64 backend would pass it in registers by taking advantage of an
LLVM ABI extension that allows using 3 registers for the x86_64 sysv
call conv rather than the officially specified 2 registers.

This adjustment is only necessary for the Rust ABI as for other ABI's
the calling convention implementations in rustc_target already ensure
any return value which doesn't fit in the available amount of return
registers is passed in the right way for the current target.

compiler/rustc_ty_utils/src/abi.rs

@@ -726,6 +726,49 @@ fn fn_abi_adjust_for_abi<'tcx>(
                 };
             }
 
+            if arg_idx.is_none() && arg.layout.size > Pointer(AddressSpace::DATA).size(cx) * 2 {
+                // Return values larger than 2 registers using a return area
+                // pointer. LLVM and Cranelift disagree about how to return
+                // values that don't fit in the registers designated for return
+                // values. LLVM will force the entire return value to be passed
+                // by return area pointer, while Cranelift will look at each IR level
+                // return value independently and decide to pass it in a
+                // register or not, which would result in the return value
+                // being passed partially in registers and partially through a
+                // return area pointer.
+                //
+                // While Cranelift may need to be fixed as the LLVM behavior is
+                // generally more correct with respect to the surface language,
+                // forcing this behavior in rustc itself makes it easier for
+                // other backends to conform to the Rust ABI and for the C ABI
+                // rustc already handles this behavior anyway.
+                //
+                // In addition LLVM's decision to pass the return value in
+                // registers or using a return area pointer depends on how
+                // exactly the return type is lowered to an LLVM IR type. For
+                // example `Option<u128>` can be lowered as `{ i128, i128 }`
+                // in which case the x86_64 backend would use a return area
+                // pointer, or it could be passed as `{ i32, i128 }` in which
+                // case the x86_64 backend would pass it in registers by taking
+                // advantage of an LLVM ABI extension that allows using 3
+                // registers for the x86_64 sysv call conv rather than the
+                // officially specified 2 registers.
+                //
+                // FIXME: Technically we should look at the amount of available
+                // return registers rather than guessing that there are 2
+                // registers for return values. In practice only a couple of
+                // architectures have less than 2 return registers. None of
+                // which supported by Cranelift.
+                //
+                // NOTE: This adjustment is only necessary for the Rust ABI as
+                // for other ABI's the calling convention implementations in
+                // rustc_target already ensure any return value which doesn't
+                // fit in the available amount of return registers is passed in
+                // the right way for the current target.
+                arg.make_indirect();
+                return;
+            }
+
             match arg.layout.abi {
                 Abi::Aggregate { .. } => {}
 

tests/codegen/tuple-layout-opt.rs

@@ -19,28 +19,28 @@ pub fn test_ScalarZstFirst(_: ScalarZstFirst) -> ScalarZstFirst {
 }
 
 type ScalarPairZstLast = (u8, u128, ());
-// CHECK: define {{(dso_local )?}}{ i128, i8 } @test_ScalarPairZstLast(i128 %_1.0, i8 %_1.1)
+// CHECK: define {{(dso_local )?}}void @test_ScalarPairZstLast(ptr sret([32 x i8]) align 16 %_0, i128 %_1.0, i8 %_1.1)
 #[no_mangle]
 pub fn test_ScalarPairZstLast(_: ScalarPairZstLast) -> ScalarPairZstLast {
     loop {}
 }
 
 type ScalarPairZstFirst = ((), u8, u128);
-// CHECK: define {{(dso_local )?}}{ i8, i128 } @test_ScalarPairZstFirst(i8 %_1.0, i128 %_1.1)
+// CHECK: define {{(dso_local )?}}void @test_ScalarPairZstFirst(ptr sret([32 x i8]) align 16 %_0, i8 %_1.0, i128 %_1.1)
 #[no_mangle]
 pub fn test_ScalarPairZstFirst(_: ScalarPairZstFirst) -> ScalarPairZstFirst {
     loop {}
 }
 
 type ScalarPairLotsOfZsts = ((), u8, (), u128, ());
-// CHECK: define {{(dso_local )?}}{ i128, i8 } @test_ScalarPairLotsOfZsts(i128 %_1.0, i8 %_1.1)
+// CHECK: define {{(dso_local )?}}void @test_ScalarPairLotsOfZsts(ptr sret([32 x i8]) align 16 %_0, i128 %_1.0, i8 %_1.1)
 #[no_mangle]
 pub fn test_ScalarPairLotsOfZsts(_: ScalarPairLotsOfZsts) -> ScalarPairLotsOfZsts {
     loop {}
 }
 
 type ScalarPairLottaNesting = (((), ((), u8, (), u128, ())), ());
-// CHECK: define {{(dso_local )?}}{ i128, i8 } @test_ScalarPairLottaNesting(i128 %_1.0, i8 %_1.1)
+// CHECK: define {{(dso_local )?}}void @test_ScalarPairLottaNesting(ptr sret([32 x i8]) align 16 %_0, i128 %_1.0, i8 %_1.1)
 #[no_mangle]
 pub fn test_ScalarPairLottaNesting(_: ScalarPairLottaNesting) -> ScalarPairLottaNesting {
     loop {}