rust_for_linux: -Zreg-struct-return commandline flag for X86 (#116973)

Author: azhogin

compiler/rustc_codegen_gcc/src/context.rs

@@ -544,7 +544,10 @@ impl<'gcc, 'tcx> HasWasmCAbiOpt for CodegenCx<'gcc, 'tcx> {
 
 impl<'gcc, 'tcx> HasX86AbiOpt for CodegenCx<'gcc, 'tcx> {
     fn x86_abi_opt(&self) -> X86Abi {
-        X86Abi { regparm: self.tcx.sess.opts.unstable_opts.regparm }
+        X86Abi {
+            regparm: self.tcx.sess.opts.unstable_opts.regparm,
+            reg_struct_return: self.tcx.sess.opts.unstable_opts.reg_struct_return,
+        }
     }
 }
 

compiler/rustc_interface/src/tests.rs

@@ -831,6 +831,7 @@ fn test_unstable_options_tracking_hash() {
     tracked!(precise_enum_drop_elaboration, false);
     tracked!(profile_sample_use, Some(PathBuf::from("abc")));
     tracked!(profiler_runtime, "abc".to_string());
+    tracked!(reg_struct_return, true);
     tracked!(regparm, Some(3));
     tracked!(relax_elf_relocations, Some(true));
     tracked!(remap_cwd_prefix, Some(PathBuf::from("abc")));

compiler/rustc_middle/src/ty/layout.rs

@@ -552,7 +552,10 @@ impl<'tcx> HasWasmCAbiOpt for TyCtxt<'tcx> {
 
 impl<'tcx> HasX86AbiOpt for TyCtxt<'tcx> {
     fn x86_abi_opt(&self) -> X86Abi {
-        X86Abi { regparm: self.sess.opts.unstable_opts.regparm }
+        X86Abi {
+            regparm: self.sess.opts.unstable_opts.regparm,
+            reg_struct_return: self.sess.opts.unstable_opts.reg_struct_return,
+        }
     }
 }
 

compiler/rustc_session/messages.ftl

@@ -135,5 +135,6 @@ session_unsupported_crate_type_for_target =
 
 session_unsupported_dwarf_version = requested DWARF version {$dwarf_version} is greater than 5
 
+session_unsupported_reg_struct_return_arch = `-Zreg-struct-return` is only supported on x86
 session_unsupported_regparm = `-Zregparm={$regparm}` is unsupported (valid values 0-3)
 session_unsupported_regparm_arch = `-Zregparm=N` is only supported on x86

compiler/rustc_session/src/errors.rs

@@ -489,6 +489,10 @@ pub(crate) struct UnsupportedRegparm {
 #[diag(session_unsupported_regparm_arch)]
 pub(crate) struct UnsupportedRegparmArch;
 
+#[derive(Diagnostic)]
+#[diag(session_unsupported_reg_struct_return_arch)]
+pub(crate) struct UnsupportedRegStructReturnArch;
+
 #[derive(Diagnostic)]
 #[diag(session_failed_to_create_profiler)]
 pub(crate) struct FailedToCreateProfiler {

compiler/rustc_session/src/options.rs

@@ -1985,6 +1985,9 @@ options! {
         "enable queries of the dependency graph for regression testing (default: no)"),
     randomize_layout: bool = (false, parse_bool, [TRACKED],
         "randomize the layout of types (default: no)"),
+    reg_struct_return: bool = (false, parse_bool, [TRACKED],
+        "On x86-32 targets, it overrides the default ABI to return small structs in registers.
+        It is UNSOUND to link together crates that use different values for this flag!"),
     regparm: Option<u32> = (None, parse_opt_number, [TRACKED],
         "On x86-32 targets, setting this to N causes the compiler to pass N arguments \
         in registers EAX, EDX, and ECX instead of on the stack for\

compiler/rustc_session/src/session.rs

@@ -1305,6 +1305,11 @@ fn validate_commandline_args_with_session_available(sess: &Session) {
             sess.dcx().emit_err(errors::UnsupportedRegparmArch);
         }
     }
+    if sess.opts.unstable_opts.reg_struct_return {
+        if sess.target.arch != "x86" {
+            sess.dcx().emit_err(errors::UnsupportedRegStructReturnArch);
+        }
+    }
 
     // The code model check applies to `thunk` and `thunk-extern`, but not `thunk-inline`, so it is
     // kept as a `match` to force a change if new ones are added, even if we currently only support

compiler/rustc_target/src/callconv/mod.rs

@@ -661,7 +661,9 @@ impl<'a, Ty> FnAbi<'a, Ty> {
                     }
                     _ => (x86::Flavor::General, None),
                 };
-                x86::compute_abi_info(cx, self, x86::X86Options { flavor, regparm });
+                let reg_struct_return = cx.x86_abi_opt().reg_struct_return;
+                let opts = x86::X86Options { flavor, regparm, reg_struct_return };
+                x86::compute_abi_info(cx, self, opts);
             }
             "x86_64" => match abi {
                 spec::abi::Abi::SysV64 { .. } => x86_64::compute_abi_info(cx, self),

compiler/rustc_target/src/callconv/x86.rs

@@ -14,6 +14,7 @@ pub(crate) enum Flavor {
 pub(crate) struct X86Options {
     pub flavor: Flavor,
     pub regparm: Option<u32>,
+    pub reg_struct_return: bool,
 }
 
 pub(crate) fn compute_abi_info<'a, Ty, C>(cx: &C, fn_abi: &mut FnAbi<'a, Ty>, opts: X86Options)
@@ -31,7 +32,7 @@ where
             // https://www.angelcode.com/dev/callconv/callconv.html
             // Clang's ABI handling is in lib/CodeGen/TargetInfo.cpp
             let t = cx.target_spec();
-            if t.abi_return_struct_as_int {
+            if t.abi_return_struct_as_int || opts.reg_struct_return {
                 // According to Clang, everyone but MSVC returns single-element
                 // float aggregates directly in a floating-point register.
                 if !t.is_like_msvc && fn_abi.ret.layout.is_single_fp_element(cx) {

compiler/rustc_target/src/spec/mod.rs

@@ -2108,6 +2108,8 @@ pub struct X86Abi {
     /// On x86-32 targets, the regparm N causes the compiler to pass arguments
     /// in registers EAX, EDX, and ECX instead of on the stack.
     pub regparm: Option<u32>,
+    /// Override the default ABI to return small structs in registers
+    pub reg_struct_return: bool,
 }
 
 pub trait HasX86AbiOpt {

src/doc/unstable-book/src/compiler-flags/reg-struct-return.md

@@ -0,0 +1,15 @@
+# `reg-struct-return`
+
+The tracking issue for this feature is: https://github.com/rust-lang/rust/issues/116973.
+
+------------------------
+
+Option -Zreg-struct-return causes the compiler to return small structs in registers
+instead of on the stack for extern "C"-like functions.
+It is UNSOUND to link together crates that use different values for this flag.
+It is only supported on `x86`.
+
+It is equivalent to [Clang]'s and [GCC]'s `-freg-struct-return`.
+
+[Clang]: https://clang.llvm.org/docs/ClangCommandLineReference.html#cmdoption-clang-freg-struct-return
+[GCC]: https://gcc.gnu.org/onlinedocs/gcc/Code-Gen-Options.html#index-freg-struct-return

tests/codegen/reg-struct-return.rs

@@ -0,0 +1,206 @@
+// Checks how `reg-struct-return` flag works with different calling conventions:
+// Return struct with 8/16/32/64 bit size will be converted into i8/i16/i32/i64
+// (like abi_return_struct_as_int target spec).
+// x86 only.
+
+//@ revisions: ENABLED DISABLED
+//@ add-core-stubs
+//@ compile-flags: --target i686-unknown-linux-gnu -O -C no-prepopulate-passes
+//@ [ENABLED] compile-flags: -Zreg-struct-return
+//@ needs-llvm-components: x86
+
+#![crate_type = "lib"]
+#![no_std]
+#![no_core]
+#![feature(no_core, lang_items)]
+
+extern crate minicore;
+use minicore::*;
+
+#[repr(C)]
+pub struct Foo {
+    x: u32,
+    y: u32,
+}
+
+#[repr(C)]
+pub struct Foo1 {
+    x: u32,
+}
+
+#[repr(C)]
+pub struct Foo2 {
+    x: bool,
+    y: bool,
+    z: i16,
+}
+
+#[repr(C)]
+pub struct Foo3 {
+    x: i16,
+    y: bool,
+    z: bool,
+}
+
+#[repr(C)]
+pub struct Foo4 {
+    x: char,
+    y: bool,
+    z: u8,
+}
+
+#[repr(C)]
+pub struct Foo5 {
+    x: u32,
+    y: u16,
+    z: u8,
+    a: bool,
+}
+
+#[repr(C)]
+pub struct FooOversize1 {
+    x: u32,
+    y: u32,
+    z: u32,
+}
+
+#[repr(C)]
+pub struct FooOversize2 {
+    f0: u16,
+    f1: u16,
+    f2: u16,
+    f3: u16,
+    f4: u16,
+}
+
+#[repr(C)]
+pub struct FooFloat1 {
+    x: f32,
+    y: f32,
+}
+
+#[repr(C)]
+pub struct FooFloat2 {
+    x: f64,
+}
+
+#[repr(C)]
+pub struct FooFloat3 {
+    x: f32,
+}
+
+pub mod tests {
+    use {
+        Foo, Foo1, Foo2, Foo3, Foo4, Foo5, FooFloat1, FooFloat2, FooFloat3, FooOversize1,
+        FooOversize2,
+    };
+
+    // ENABLED: i64 @f1()
+    // DISABLED: void @f1(ptr {{.*}}sret
+    #[no_mangle]
+    pub extern "fastcall" fn f1() -> Foo {
+        Foo { x: 1, y: 2 }
+    }
+
+    // CHECK: { i32, i32 } @f2()
+    #[no_mangle]
+    pub extern "Rust" fn f2() -> Foo {
+        Foo { x: 1, y: 2 }
+    }
+
+    // ENABLED: i64 @f3()
+    // DISABLED: void @f3(ptr {{.*}}sret
+    #[no_mangle]
+    pub extern "C" fn f3() -> Foo {
+        Foo { x: 1, y: 2 }
+    }
+
+    // ENABLED: i64 @f4()
+    // DISABLED: void @f4(ptr {{.*}}sret
+    #[no_mangle]
+    pub extern "cdecl" fn f4() -> Foo {
+        Foo { x: 1, y: 2 }
+    }
+
+    // ENABLED: i64 @f5()
+    // DISABLED: void @f5(ptr {{.*}}sret
+    #[no_mangle]
+    pub extern "stdcall" fn f5() -> Foo {
+        Foo { x: 1, y: 2 }
+    }
+
+    // ENABLED: i64 @f6()
+    // DISABLED: void @f6(ptr {{.*}}sret
+    #[no_mangle]
+    pub extern "thiscall" fn f6() -> Foo {
+        Foo { x: 1, y: 2 }
+    }
+
+    // ENABLED: i32 @f7()
+    // DISABLED: void @f7(ptr {{.*}}sret
+    #[no_mangle]
+    pub extern "C" fn f7() -> Foo1 {
+        Foo1 { x: 1 }
+    }
+
+    // ENABLED: i32 @f8()
+    // DISABLED: void @f8(ptr {{.*}}sret
+    #[no_mangle]
+    pub extern "C" fn f8() -> Foo2 {
+        Foo2 { x: true, y: false, z: 5 }
+    }
+
+    // ENABLED: i32 @f9()
+    // DISABLED: void @f9(ptr {{.*}}sret
+    #[no_mangle]
+    pub extern "C" fn f9() -> Foo3 {
+        Foo3 { x: 5, y: false, z: true }
+    }
+
+    // ENABLED: i64 @f10()
+    // DISABLED: void @f10(ptr {{.*}}sret
+    #[no_mangle]
+    pub extern "C" fn f10() -> Foo4 {
+        Foo4 { x: 'x', y: true, z: 170 }
+    }
+
+    // ENABLED: i64 @f11()
+    // DISABLED: void @f11(ptr {{.*}}sret
+    #[no_mangle]
+    pub extern "C" fn f11() -> Foo5 {
+        Foo5 { x: 1, y: 2, z: 3, a: true }
+    }
+
+    // CHECK: void @f12(ptr {{.*}}sret
+    #[no_mangle]
+    pub extern "C" fn f12() -> FooOversize1 {
+        FooOversize1 { x: 1, y: 2, z: 3 }
+    }
+
+    // CHECK: void @f13(ptr {{.*}}sret
+    #[no_mangle]
+    pub extern "C" fn f13() -> FooOversize2 {
+        FooOversize2 { f0: 1, f1: 2, f2: 3, f3: 4, f4: 5 }
+    }
+
+    // ENABLED: i64 @f14()
+    // DISABLED: void @f14(ptr {{.*}}sret
+    #[no_mangle]
+    pub extern "C" fn f14() -> FooFloat1 {
+        FooFloat1 { x: 1.0, y: 1.0 }
+    }
+
+    // ENABLED: double @f15()
+    // DISABLED: void @f15(ptr {{.*}}sret
+    #[no_mangle]
+    pub extern "C" fn f15() -> FooFloat2 {
+        FooFloat2 { x: 1.0 }
+    }
+
+    // ENABLED: float @f16()
+    // DISABLED: void @f16(ptr {{.*}}sret
+    #[no_mangle]
+    pub extern "C" fn f16() -> FooFloat3 {
+        FooFloat3 { x: 1.0 }
+    }
+}

tests/ui/invalid-compile-flags/reg-struct-return/requires-x86.aarch64.stderr

@@ -0,0 +1,4 @@
+error: `-Zreg-struct-return` is only supported on x86
+
+error: aborting due to 1 previous error
+

tests/ui/invalid-compile-flags/reg-struct-return/requires-x86.rs

@@ -0,0 +1,21 @@
+//@ revisions: x86 x86_64 aarch64
+
+//@ compile-flags: -Zreg-struct-return
+
+//@[x86] check-pass
+//@[x86] needs-llvm-components: x86
+//@[x86] compile-flags: --target i686-unknown-linux-gnu
+
+//@[x86_64] check-fail
+//@[x86_64] needs-llvm-components: x86
+//@[x86_64] compile-flags: --target x86_64-unknown-linux-gnu
+//@[x86_64] error-pattern: `-Zreg-struct-return` is only supported on x86
+
+//@[aarch64] check-fail
+//@[aarch64] needs-llvm-components: aarch64
+//@[aarch64] compile-flags: --target aarch64-unknown-linux-gnu
+//@[aarch64] error-pattern: `-Zreg-struct-return` is only supported on x86
+
+#![feature(no_core)]
+#![no_core]
+#![no_main]

tests/ui/invalid-compile-flags/reg-struct-return/requires-x86.x86_64.stderr

@@ -0,0 +1,4 @@
+error: `-Zreg-struct-return` is only supported on x86
+
+error: aborting due to 1 previous error
+